* [PATCH 0/7] x86/Linux Target Description Changes
@ 2024-02-01 15:28 Andrew Burgess
2024-02-01 15:28 ` [PATCH 1/7] gdbserver: convert have_ptrace_getregset to a tribool Andrew Burgess
` (7 more replies)
0 siblings, 8 replies; 116+ messages in thread
From: Andrew Burgess @ 2024-02-01 15:28 UTC (permalink / raw)
To: gdb-patches; +Cc: Andrew Burgess
This series started when I ran the test script
gdb.server/connect-with-no-symbol-file.exp on an x86-64/Linux target
(with the unix board file) and noticed that two core files were left
behind.
The test is a gdbserver test (even when the unix board is used), and
the two core files are generated by the inferior as gdbserver detaches
from them.
I tracked the problem down to some issues with how gdbserver creates
its target descriptions.
Instead of just fixing gdbserver I think it would be good to share
more of the target description creation code between GDB and
gdbserver, the issue that gdbserver has is already fixed on the GDB
side.
By patch #5 enough code is shared that the problem mentioned above is
fixed.
Patches #6 and #7 try to share more code between GDB and gdbserver as
I think this is a good thing.
Thanks,
Andrew
---
Andrew Burgess (7):
gdbserver: convert have_ptrace_getregset to a tribool
gdb/x86: move reading of cs and ds state into gdb/nat directory
gdbserver/x86: move no-xml code earlier in x86_linux_read_description
gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition
gdb/gdbserver: share some code relating to target description creation
gdbserver: update target description creation for x86/linux
gdb/gdbserver: share x86/linux tdesc caching
gdb/Makefile.in | 1 +
gdb/amd64-linux-tdep.c | 33 +--
gdb/amd64-linux-tdep.h | 6 -
gdb/configure.nat | 4 +-
gdb/i386-linux-tdep.c | 32 +--
gdb/i386-linux-tdep.h | 23 --
gdb/nat/x86-linux-tdesc.c | 405 +++++++++++++++++++++++++++++++++++
gdb/nat/x86-linux-tdesc.h | 115 ++++++++++
gdb/nat/x86-linux.c | 47 ++++
gdb/nat/x86-linux.h | 48 +++++
gdb/x86-linux-nat.c | 123 ++---------
gdbserver/Makefile.in | 4 +
gdbserver/configure.srv | 4 +
gdbserver/linux-amd64-ipa.cc | 45 +---
gdbserver/linux-arm-low.cc | 6 +-
gdbserver/linux-i386-ipa.cc | 25 +--
gdbserver/linux-low.cc | 2 +-
gdbserver/linux-low.h | 2 +-
gdbserver/linux-x86-low.cc | 185 +++++-----------
gdbserver/linux-x86-tdesc.cc | 141 +-----------
gdbserver/linux-x86-tdesc.h | 56 -----
21 files changed, 737 insertions(+), 570 deletions(-)
create mode 100644 gdb/nat/x86-linux-tdesc.c
create mode 100644 gdb/nat/x86-linux-tdesc.h
delete mode 100644 gdbserver/linux-x86-tdesc.h
base-commit: 05d1b4b4ad7d74a64cc71c53d621241fc393fcb6
--
2.25.4
^ permalink raw reply [flat|nested] 116+ messages in thread* [PATCH 1/7] gdbserver: convert have_ptrace_getregset to a tribool 2024-02-01 15:28 [PATCH 0/7] x86/Linux Target Description Changes Andrew Burgess @ 2024-02-01 15:28 ` Andrew Burgess 2024-02-01 15:28 ` [PATCH 2/7] gdb/x86: move reading of cs and ds state into gdb/nat directory Andrew Burgess ` (6 subsequent siblings) 7 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-02-01 15:28 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess Convert the have_ptrace_getregset global within gdbserver to a tribool. This brings the flag into alignment with the corresponding flag in GDB. The gdbserver have_ptrace_getregset variable is already used as a tribool, it just doesn't have the tribool type. In a future commit I plan to share more code between GDB and gdbserver, and having this variable be the same type in both code bases will make the sharing much easier. There should be no user visible changes after this commit. --- gdbserver/linux-arm-low.cc | 6 +++--- gdbserver/linux-low.cc | 2 +- gdbserver/linux-low.h | 2 +- gdbserver/linux-x86-low.cc | 10 +++++----- 4 files changed, 10 insertions(+), 10 deletions(-) diff --git a/gdbserver/linux-arm-low.cc b/gdbserver/linux-arm-low.cc index 396ec88081b..b4f0e071c92 100644 --- a/gdbserver/linux-arm-low.cc +++ b/gdbserver/linux-arm-low.cc @@ -1007,9 +1007,9 @@ arm_target::low_arch_setup () /* Check if PTRACE_GETREGSET works. */ if (ptrace (PTRACE_GETREGSET, tid, NT_PRSTATUS, &iov) == 0) - have_ptrace_getregset = 1; + have_ptrace_getregset = TRIBOOL_TRUE; else - have_ptrace_getregset = 0; + have_ptrace_getregset = TRIBOOL_FALSE; } bool @@ -1122,7 +1122,7 @@ arm_target::get_regs_info () { const struct target_desc *tdesc = current_process ()->tdesc; - if (have_ptrace_getregset == 1 + if (have_ptrace_getregset == TRIBOOL_TRUE && (is_aarch32_linux_description (tdesc) || arm_linux_get_tdesc_fp_type (tdesc) == ARM_FP_TYPE_VFPV3)) return ®s_info_aarch32; diff --git a/gdbserver/linux-low.cc b/gdbserver/linux-low.cc index 444eebc6bbe..2949619f503 100644 --- a/gdbserver/linux-low.cc +++ b/gdbserver/linux-low.cc @@ -135,7 +135,7 @@ typedef struct #endif /* Does the current host support PTRACE_GETREGSET? */ -int have_ptrace_getregset = -1; +enum tribool have_ptrace_getregset = TRIBOOL_UNKNOWN; /* Return TRUE if THREAD is the leader thread of the process. */ diff --git a/gdbserver/linux-low.h b/gdbserver/linux-low.h index d34d2738238..eaf87527338 100644 --- a/gdbserver/linux-low.h +++ b/gdbserver/linux-low.h @@ -951,7 +951,7 @@ void thread_db_notice_clone (struct thread_info *parent_thr, ptid_t child_ptid); bool thread_db_thread_handle (ptid_t ptid, gdb_byte **handle, int *handle_len); -extern int have_ptrace_getregset; +extern enum tribool have_ptrace_getregset; /* Search for the value with type MATCH in the auxv vector, with entries of length WORDSIZE bytes, of process with pid PID. If found, store the diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 3af0a009052..0a3bd2c8670 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -899,7 +899,7 @@ x86_linux_read_description (void) if (ptrace (PTRACE_GETFPXREGS, tid, 0, (long) &fpxregs) < 0) { have_ptrace_getfpxregs = 0; - have_ptrace_getregset = 0; + have_ptrace_getregset = TRIBOOL_FALSE; return i386_linux_read_description (X86_XSTATE_X87); } else @@ -918,7 +918,7 @@ x86_linux_read_description (void) return tdesc_i386_linux_no_xml.get (); } - if (have_ptrace_getregset == -1) + if (have_ptrace_getregset == TRIBOOL_UNKNOWN) { uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; struct iovec iov; @@ -929,10 +929,10 @@ x86_linux_read_description (void) /* Check if PTRACE_GETREGSET works. */ if (ptrace (PTRACE_GETREGSET, tid, (unsigned int) NT_X86_XSTATE, (long) &iov) < 0) - have_ptrace_getregset = 0; + have_ptrace_getregset = TRIBOOL_FALSE; else { - have_ptrace_getregset = 1; + have_ptrace_getregset = TRIBOOL_TRUE; /* Get XCR0 from XSAVE extended state. */ xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET @@ -951,7 +951,7 @@ x86_linux_read_description (void) } /* Check the native XCR0 only if PTRACE_GETREGSET is available. */ - xcr0_features = (have_ptrace_getregset + xcr0_features = (have_ptrace_getregset == TRIBOOL_TRUE && (xcr0 & X86_XSTATE_ALL_MASK)); if (xcr0_features) -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCH 2/7] gdb/x86: move reading of cs and ds state into gdb/nat directory 2024-02-01 15:28 [PATCH 0/7] x86/Linux Target Description Changes Andrew Burgess 2024-02-01 15:28 ` [PATCH 1/7] gdbserver: convert have_ptrace_getregset to a tribool Andrew Burgess @ 2024-02-01 15:28 ` Andrew Burgess 2024-02-01 15:28 ` [PATCH 3/7] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess ` (5 subsequent siblings) 7 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-02-01 15:28 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess This patch is part of a series that has the aim of making the code that, for x86, reads the target description for a native process shared between GDB and gdbserver. Within GDB part of this process involves reading the cs and ds state from the 'struct user_regs_struct' using a ptrace call. This isn't done by gdbserver, which is part of the motivation for this whole series; the approach gdbserver takes is inferior to the approach GDB takes. This commit moves the reading of cs and ds, which is used to figure out if a thread is 32-bit or 64-bit (or in x32 mode), into the gdb/nat directory so that the code could be shared with gdbserver, but at this point I'm not actually using the code in gdbserver, that will come later. As such there should be no user visible changes after this commit, GDB continues to do things as it did before (reading cs/ds), while gdbserver continues to use its own approach (which doesn't require reading cs/ds). --- gdb/nat/x86-linux.c | 47 +++++++++++++++++++++++++++++++++++++++++++++ gdb/nat/x86-linux.h | 28 +++++++++++++++++++++++++++ gdb/x86-linux-nat.c | 42 +++++----------------------------------- 3 files changed, 80 insertions(+), 37 deletions(-) diff --git a/gdb/nat/x86-linux.c b/gdb/nat/x86-linux.c index e61f4d749ba..4242a1baafb 100644 --- a/gdb/nat/x86-linux.c +++ b/gdb/nat/x86-linux.c @@ -20,6 +20,8 @@ #include "gdbsupport/common-defs.h" #include "x86-linux.h" #include "x86-linux-dregs.h" +#include "nat/gdb_ptrace.h" +#include <sys/user.h> /* Per-thread arch-specific data we want to keep. */ @@ -80,3 +82,48 @@ x86_linux_prepare_to_resume (struct lwp_info *lwp) { x86_linux_update_debug_registers (lwp); } + +#ifdef __x86_64__ +/* Value of CS segment register: + 64bit process: 0x33 + 32bit process: 0x23 */ +#define AMD64_LINUX_USER64_CS 0x33 + +/* Value of DS segment register: + LP64 process: 0x0 + X32 process: 0x2b */ +#define AMD64_LINUX_X32_DS 0x2b +#endif + +/* See nat/x86-linux.h. */ + +x86_linux_arch_size +x86_linux_ptrace_get_arch_size (int tid) +{ +#ifdef __x86_64__ + unsigned long cs; + unsigned long ds; + + /* Get CS register. */ + errno = 0; + cs = ptrace (PTRACE_PEEKUSER, tid, + offsetof (struct user_regs_struct, cs), 0); + if (errno != 0) + perror_with_name (_("Couldn't get CS register")); + + bool is_64bit = cs == AMD64_LINUX_USER64_CS; + + /* Get DS register. */ + errno = 0; + ds = ptrace (PTRACE_PEEKUSER, tid, + offsetof (struct user_regs_struct, ds), 0); + if (errno != 0) + perror_with_name (_("Couldn't get DS register")); + + bool is_x32 = ds == AMD64_LINUX_X32_DS; + + return x86_linux_arch_size (is_64bit, is_x32); +#else + return x86_linux_arch_size (false, false); +#endif +} diff --git a/gdb/nat/x86-linux.h b/gdb/nat/x86-linux.h index 822882173f9..15153ea277e 100644 --- a/gdb/nat/x86-linux.h +++ b/gdb/nat/x86-linux.h @@ -47,4 +47,32 @@ extern void x86_linux_delete_thread (struct arch_lwp_info *arch_lwp); extern void x86_linux_prepare_to_resume (struct lwp_info *lwp); +/* Return value from x86_linux_ptrace_get_arch_size function. Indicates if + a thread is 32-bit, 64-bit, or x32. */ + +struct x86_linux_arch_size +{ + explicit x86_linux_arch_size (bool is_64bit, bool is_x32) + : m_is_64bit (is_64bit), + m_is_x32 (is_x32) + { + /* Nothing. */ + } + + bool is_64bit () const + { return m_is_64bit; } + + bool is_x32 () const + { return m_is_x32; } + +private: + bool m_is_64bit = false; + bool m_is_x32 = false; +}; + +/* Use ptrace calls to figure out if thread TID is 32-bit, 64-bit, or + 64-bit running in x32 mode. */ + +extern x86_linux_arch_size x86_linux_ptrace_get_arch_size (int tid); + #endif /* NAT_X86_LINUX_H */ diff --git a/gdb/x86-linux-nat.c b/gdb/x86-linux-nat.c index b93ffca38db..b39d05c401f 100644 --- a/gdb/x86-linux-nat.c +++ b/gdb/x86-linux-nat.c @@ -91,18 +91,6 @@ x86_linux_nat_target::post_startup_inferior (ptid_t ptid) linux_nat_target::post_startup_inferior (ptid); } -#ifdef __x86_64__ -/* Value of CS segment register: - 64bit process: 0x33 - 32bit process: 0x23 */ -#define AMD64_LINUX_USER64_CS 0x33 - -/* Value of DS segment register: - LP64 process: 0x0 - X32 process: 0x2b */ -#define AMD64_LINUX_X32_DS 0x2b -#endif - /* Get Linux/x86 target description from running target. */ const struct target_desc * @@ -122,31 +110,11 @@ x86_linux_nat_target::read_description () tid = inferior_ptid.pid (); #ifdef __x86_64__ - { - unsigned long cs; - unsigned long ds; - - /* Get CS register. */ - errno = 0; - cs = ptrace (PTRACE_PEEKUSER, tid, - offsetof (struct user_regs_struct, cs), 0); - if (errno != 0) - perror_with_name (_("Couldn't get CS register")); - - is_64bit = cs == AMD64_LINUX_USER64_CS; - - /* Get DS register. */ - errno = 0; - ds = ptrace (PTRACE_PEEKUSER, tid, - offsetof (struct user_regs_struct, ds), 0); - if (errno != 0) - perror_with_name (_("Couldn't get DS register")); - - is_x32 = ds == AMD64_LINUX_X32_DS; - - if (sizeof (void *) == 4 && is_64bit && !is_x32) - error (_("Can't debug 64-bit process with 32-bit GDB")); - } + + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); + is_64bit = arch_size.is_64bit (); + is_x32 = arch_size.is_x32 (); + #elif HAVE_PTRACE_GETFPXREGS if (have_ptrace_getfpxregs == -1) { -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCH 3/7] gdbserver/x86: move no-xml code earlier in x86_linux_read_description 2024-02-01 15:28 [PATCH 0/7] x86/Linux Target Description Changes Andrew Burgess 2024-02-01 15:28 ` [PATCH 1/7] gdbserver: convert have_ptrace_getregset to a tribool Andrew Burgess 2024-02-01 15:28 ` [PATCH 2/7] gdb/x86: move reading of cs and ds state into gdb/nat directory Andrew Burgess @ 2024-02-01 15:28 ` Andrew Burgess 2024-02-01 15:28 ` [PATCH 4/7] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess ` (4 subsequent siblings) 7 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-02-01 15:28 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess This commit is part of a series that aims to share more of the x86 target description reading/generation code between GDB and gdbserver. There are a huge number of similarities between the code in gdbserver's x86_linux_read_description function and GDB's x86_linux_nat_target::read_description function, and it is this similarity that I plan, in a later commit, to share between GDB and gdbserver. However, one thing that is different in x86_linux_read_description is the code inside the '!use_xml' block. This is the code that handles the case where gdbserver is not allowed to send an XML target description back to GDB. In this case gdbserver uses some predefined, fixed, target descriptions. First, it's worth noting that I suspect this code is not tested any more. I couldn't find anything in the testsuite that tries to disable XML target description support. And the idea of having a single "fixed" target description really doesn't work well when we think about all the various x86 extensions that exist. Part of me would like to rip out the no-xml support in gdbserver (at least for x86), and if a GDB connects that doesn't support XML target descriptions, gdbserver can just give an error and drop the connection. GDB has supported XML target descriptions for 16 years now, I think it would be reasonable for our shipped gdbserver to drop support for the old way of doing things. Anyway.... this commit doesn't do that. What I did notice was that, over time, the '!use_xml' block appears to have "drifted" within the x86_linux_read_description function; it's now not the first check we do. Instead we make some ptrace calls and return a target description generated based on the result of these ptrace calls. Surely it only makes sense to generate variable target descriptions if we can send these back to GDB? So in this commit I propose to move the '!use_xml' block earlier in the x86_linux_read_description function. The benefit of this is that this leaves the later half of x86_linux_read_description much more similar to the GDB function x86_linux_nat_target::read_description and sets us up for potentially sharing code between GDB and gdbserver in a later commit. --- gdbserver/linux-x86-low.cc | 27 ++++++++++++++++----------- 1 file changed, 16 insertions(+), 11 deletions(-) diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 0a3bd2c8670..ec80bfe905c 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -891,6 +891,22 @@ x86_linux_read_description (void) #endif } + /* If we are not allowed to send an XML target description then we need + to use the hard-wired target descriptions. This corresponds to GDB's + default machine for x86. + + This check needs to occur before any returns statements that might + generate some alternative target descriptions. */ + if (!use_xml) + { +#ifdef __x86_64__ + if (machine == EM_X86_64) + return tdesc_amd64_linux_no_xml.get (); + else +#endif + return tdesc_i386_linux_no_xml.get (); + } + #if !defined __x86_64__ && defined HAVE_PTRACE_GETFPXREGS if (machine == EM_386 && have_ptrace_getfpxregs == -1) { @@ -907,17 +923,6 @@ x86_linux_read_description (void) } #endif - if (!use_xml) - { - /* Don't use XML. */ -#ifdef __x86_64__ - if (machine == EM_X86_64) - return tdesc_amd64_linux_no_xml.get (); - else -#endif - return tdesc_i386_linux_no_xml.get (); - } - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) { uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCH 4/7] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition 2024-02-01 15:28 [PATCH 0/7] x86/Linux Target Description Changes Andrew Burgess ` (2 preceding siblings ...) 2024-02-01 15:28 ` [PATCH 3/7] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess @ 2024-02-01 15:28 ` Andrew Burgess 2024-02-01 15:28 ` [PATCH 5/7] gdb/gdbserver: share some code relating to target description creation Andrew Burgess ` (3 subsequent siblings) 7 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-02-01 15:28 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess Share the definition of I386_LINUX_XSAVE_XCR0_OFFSET between GDB and gdbserver. This commit is part of a series that aims to share more of the x86 target description creation code between GDB and gdbserver. The I386_LINUX_XSAVE_XCR0_OFFSET #define is used as part of the target description creation, and I noticed that this constant is defined separately for GDB and gdbserver. This commit moves the definition into gdb/nat/x86-linux.h, which allows the #define to be shared. There should be no user visible changes after this commit. --- gdb/amd64-linux-tdep.c | 1 + gdb/i386-linux-tdep.c | 1 + gdb/i386-linux-tdep.h | 20 -------------------- gdb/nat/x86-linux.h | 20 ++++++++++++++++++++ gdbserver/linux-x86-low.cc | 22 +--------------------- 5 files changed, 23 insertions(+), 41 deletions(-) diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c index fe7f5d4a259..051c3f74eda 100644 --- a/gdb/amd64-linux-tdep.c +++ b/gdb/amd64-linux-tdep.c @@ -42,6 +42,7 @@ #include "arch/amd64.h" #include "target-descriptions.h" #include "expop.h" +#include "nat/x86-linux.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_AMD64 "syscalls/amd64-linux.xml" diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index 27aea18c436..43cb2961e28 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -40,6 +40,7 @@ #include "i387-tdep.h" #include "gdbsupport/x86-xstate.h" +#include "nat/x86-linux.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_I386 "syscalls/i386-linux.xml" diff --git a/gdb/i386-linux-tdep.h b/gdb/i386-linux-tdep.h index 5891747572b..07593c6a8ec 100644 --- a/gdb/i386-linux-tdep.h +++ b/gdb/i386-linux-tdep.h @@ -58,26 +58,6 @@ extern void i386_linux_report_signal_info (struct gdbarch *gdbarch, /* Return the target description according to XCR0. */ extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); -/* Format of XSAVE extended state is: - struct - { - fxsave_bytes[0..463] - sw_usable_bytes[464..511] - xstate_hdr_bytes[512..575] - extended state regions (AVX, MPX, AVX512, PKRU, etc.) - }; - - Same memory layout will be used for the coredump NT_X86_XSTATE - representing the XSAVE extended state registers. - - The first 8 bytes of the sw_usable_bytes[464..467] is the OS enabled - extended state mask, which is the same as the extended control register - 0 (the XFEATURE_ENABLED_MASK register), XCR0. We can use this mask - together with the mask saved in the xstate_hdr_bytes to determine what - states the processor/OS supports and what state, used or initialized, - the process/thread is in. */ -#define I386_LINUX_XSAVE_XCR0_OFFSET 464 - extern int i386_linux_gregset_reg_offset[]; /* Return x86 siginfo type. */ diff --git a/gdb/nat/x86-linux.h b/gdb/nat/x86-linux.h index 15153ea277e..855a8d14f91 100644 --- a/gdb/nat/x86-linux.h +++ b/gdb/nat/x86-linux.h @@ -22,6 +22,26 @@ #include "nat/linux-nat.h" +/* Format of XSAVE extended state is: + struct + { + fxsave_bytes[0..463] + sw_usable_bytes[464..511] + xstate_hdr_bytes[512..575] + extended state regions (AVX, MPX, AVX512, PKRU, etc.) + }; + + Same memory layout will be used for the coredump NT_X86_XSTATE + representing the XSAVE extended state registers. + + The first 8 bytes of the sw_usable_bytes[464..467] is the OS enabled + extended state mask, which is the same as the extended control register + 0 (the XFEATURE_ENABLED_MASK register), XCR0. We can use this mask + together with the mask saved in the xstate_hdr_bytes to determine what + states the processor/OS supports and what state, used or initialized, + the process/thread is in. */ +#define I386_LINUX_XSAVE_XCR0_OFFSET 464 + /* Set whether our local mirror of LWP's debug registers has been changed since the values were last written to the thread. Nonzero indicates that a change has been made, zero indicates no change. */ diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index ec80bfe905c..7bade8b3cbf 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -27,6 +27,7 @@ #include "gdbsupport/x86-xstate.h" #include "nat/x86-xstate.h" #include "nat/gdb_ptrace.h" +#include "nat/x86-linux.h" #ifdef __x86_64__ #include "nat/amd64-linux-siginfo.h" @@ -832,27 +833,6 @@ x86_target::low_siginfo_fixup (siginfo_t *ptrace, gdb_byte *inf, int direction) \f static int use_xml; -/* Format of XSAVE extended state is: - struct - { - fxsave_bytes[0..463] - sw_usable_bytes[464..511] - xstate_hdr_bytes[512..575] - avx_bytes[576..831] - future_state etc - }; - - Same memory layout will be used for the coredump NT_X86_XSTATE - representing the XSAVE extended state registers. - - The first 8 bytes of the sw_usable_bytes[464..467] is the OS enabled - extended state mask, which is the same as the extended control register - 0 (the XFEATURE_ENABLED_MASK register), XCR0. We can use this mask - together with the mask saved in the xstate_hdr_bytes to determine what - states the processor/OS supports and what state, used or initialized, - the process/thread is in. */ -#define I386_LINUX_XSAVE_XCR0_OFFSET 464 - /* Does the current host support the GETFPXREGS request? The header file may or may not define it, and even if it is defined, the kernel will return EIO if it's running on a pre-SSE processor. */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCH 5/7] gdb/gdbserver: share some code relating to target description creation 2024-02-01 15:28 [PATCH 0/7] x86/Linux Target Description Changes Andrew Burgess ` (3 preceding siblings ...) 2024-02-01 15:28 ` [PATCH 4/7] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess @ 2024-02-01 15:28 ` Andrew Burgess 2024-02-01 15:28 ` [PATCH 6/7] gdbserver: update target description creation for x86/linux Andrew Burgess ` (2 subsequent siblings) 7 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-02-01 15:28 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess This commit is part of a series to share more of the x86 target description creation code between GDB and gdbserver. Unlike previous commits which were mostly refactoring, this commit is the first that makes a real change, though that change should mostly be for gdbserver; I've largely adopted the "GDB" way of doing things for gdbserver, and this fixes a real gdbserver bug. On a x86-64 Linux target, running the test: gdb.server/connect-with-no-symbol-file.exp results in two core files being created. Both of these core files are from the inferior process, created after gdbserver has detached. In this test a gdbserver process is started and then, after gdbserver has started, but before GDB attaches, we either delete the inferior executable, or change its permissions so it can't be read. Only after doing this do we attempt to connect with GDB. As GDB connects to gdbserver, gdbserver attempts to figure out the target description so that it can send the description to GDB, this involves a call to x86_linux_read_description. In x86_linux_read_description one of the first things we do is try to figure out if the process is 32-bit or 64-bit. To do this we look up the executable via the thread-id, and then attempt to read the architecture size from the executable. This isn't going to work if the executable has been deleted, or is no longer readable. And so, as we can't read the executable, we default to an i386 target and use an i386 target description. A consequence of using an i386 target description is that addresses are assumed to be 32-bits. Here's an example session that shows the problems this causes. This is run on an x86-64 machine, and the test binary (xx.x) is a standard 64-bit x86-64 binary: shell_1$ gdbserver --once localhost :54321 /tmp/xx.x shell_2$ gdb -q (gdb) set sysroot (gdb) shell chmod 000 /tmp/xx.x (gdb) target remote :54321 Remote debugging using :54321 warning: /tmp/xx.x: Permission denied. 0xf7fd3110 in ?? () (gdb) show architecture The target architecture is set to "auto" (currently "i386"). (gdb) p/x $pc $1 = 0xf7fd3110 (gdb) info proc mappings process 2412639 Mapped address spaces: Start Addr End Addr Size Offset Perms objfile 0x400000 0x401000 0x1000 0x0 r--p /tmp/xx.x 0x401000 0x402000 0x1000 0x1000 r-xp /tmp/xx.x 0x402000 0x403000 0x1000 0x2000 r--p /tmp/xx.x 0x403000 0x405000 0x2000 0x2000 rw-p /tmp/xx.x 0xf7fcb000 0xf7fcf000 0x4000 0x0 r--p [vvar] 0xf7fcf000 0xf7fd1000 0x2000 0x0 r-xp [vdso] 0xf7fd1000 0xf7fd3000 0x2000 0x0 r--p /usr/lib64/ld-2.30.so 0xf7fd3000 0xf7ff3000 0x20000 0x2000 r-xp /usr/lib64/ld-2.30.so 0xf7ff3000 0xf7ffb000 0x8000 0x22000 r--p /usr/lib64/ld-2.30.so 0xf7ffc000 0xf7ffe000 0x2000 0x2a000 rw-p /usr/lib64/ld-2.30.so 0xf7ffe000 0xf7fff000 0x1000 0x0 rw-p 0xfffda000 0xfffff000 0x25000 0x0 rw-p [stack] 0xff600000 0xff601000 0x1000 0x0 r-xp [vsyscall] (gdb) info inferiors Num Description Connection Executable * 1 process 2412639 1 (remote :54321) (gdb) shell cat /proc/2412639/maps 00400000-00401000 r--p 00000000 fd:03 45907133 /tmp/xx.x 00401000-00402000 r-xp 00001000 fd:03 45907133 /tmp/xx.x 00402000-00403000 r--p 00002000 fd:03 45907133 /tmp/xx.x 00403000-00405000 rw-p 00002000 fd:03 45907133 /tmp/xx.x 7ffff7fcb000-7ffff7fcf000 r--p 00000000 00:00 0 [vvar] 7ffff7fcf000-7ffff7fd1000 r-xp 00000000 00:00 0 [vdso] 7ffff7fd1000-7ffff7fd3000 r--p 00000000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7fd3000-7ffff7ff3000 r-xp 00002000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7ff3000-7ffff7ffb000 r--p 00022000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7ffc000-7ffff7ffe000 rw-p 0002a000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7ffe000-7ffff7fff000 rw-p 00000000 00:00 0 7ffffffda000-7ffffffff000 rw-p 00000000 00:00 0 [stack] ffffffffff600000-ffffffffff601000 r-xp 00000000 00:00 0 [vsyscall] (gdb) Notice the difference between the mappings reported via GDB and those reported directly from the kernel via /proc/PID/maps, the addresses of every mapping is clamped to 32-bits for GDB, while the kernel reports real 64-bit addresses. Notice also that the $pc value is a 32-bit value. It appears to be within one of the mappings reported by GDB, but is outside any of the mappings reported from the kernel. And this is where the problem arises. When gdbserver detaches from the inferior we pass the inferior the address from which it should resume. Due to the 32/64 bit confusion we tell the inferior to resume from the 32-bit $pc value, which is not within any valid mapping, and so, as soon as the inferior resumes, it segfaults. If we look at how GDB (not gdbserver) figures out its target description then we see an interesting difference. GDB doesn't try to read the executable. Instead GDB uses ptrace to query the thread's state, and uses this to figure out the if the thread is 32 or 64 bit. If we update gdbserver to do it the "GDB" way then the above problem is resolved, gdbserver now sees the process as 64-bit, and when we detach from the inferior we give it the correct 64-bit address, and the inferior no longer segfaults. Now, I could just update the gdbserver code, but better, I think, to share one copy of the code between GDB and gdbserver in gdb/nat/. That is what this commit does. The cores of x86_linux_read_description from gdbserver and x86_linux_nat_target::read_description from GDB are moved into a new file gdb/nat/x86-linux-tdesc.c and combined into a single function x86_linux_tdesc_for_tid which is called from each location. This new function does things the GDB way, the only changes are to allow for the sharing; we now have a callback function to call the first time that the xcr0 state is read, this allows for GDB and gdbserver to perform their own initialisation as needed, and additionally, the new function takes a pointer for where to cache the xcr0 value, this isn't needed for this commit, but will be useful in a later commit where gdbserver will want to read this cached xcr0 value. Another thing to note about this commit is how the functions i386_linux_read_description and amd64_linux_read_description are handled. For now I've left these function as implemented separately in GDB and gdbserver. I've moved the declarations of these functions into gdb/nat/x86-linux-tdesc.h, but the implementations are left as separate. A later commit in this series will make these functions shared too, but doing this is not trivial, so I've left that for a separate commit. Merging the declarations as I've done here ensures that everyone implements the function to the same API, and once these functions are shared (in a later commit) we'll want a shared declaration anyway. --- gdb/Makefile.in | 1 + gdb/amd64-linux-tdep.c | 1 + gdb/amd64-linux-tdep.h | 6 -- gdb/configure.nat | 4 +- gdb/i386-linux-tdep.c | 1 + gdb/i386-linux-tdep.h | 3 - gdb/nat/x86-linux-tdesc.c | 118 ++++++++++++++++++++++++++++ gdb/nat/x86-linux-tdesc.h | 75 ++++++++++++++++++ gdb/x86-linux-nat.c | 91 ++++------------------ gdbserver/configure.srv | 2 + gdbserver/linux-amd64-ipa.cc | 1 + gdbserver/linux-i386-ipa.cc | 1 + gdbserver/linux-x86-low.cc | 144 +++++++++++------------------------ gdbserver/linux-x86-tdesc.cc | 1 + gdbserver/linux-x86-tdesc.h | 7 -- 15 files changed, 261 insertions(+), 195 deletions(-) create mode 100644 gdb/nat/x86-linux-tdesc.c create mode 100644 gdb/nat/x86-linux-tdesc.h diff --git a/gdb/Makefile.in b/gdb/Makefile.in index 0e0f19c40c9..76bed92d84e 100644 --- a/gdb/Makefile.in +++ b/gdb/Makefile.in @@ -1586,6 +1586,7 @@ HFILES_NO_SRCDIR = \ nat/x86-gcc-cpuid.h \ nat/x86-linux.h \ nat/x86-linux-dregs.h \ + nat/x86-linux-tdesc.h \ python/py-event.h \ python/py-events.h \ python/py-stopevent.h \ diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c index 051c3f74eda..87cefaa92e9 100644 --- a/gdb/amd64-linux-tdep.c +++ b/gdb/amd64-linux-tdep.c @@ -43,6 +43,7 @@ #include "target-descriptions.h" #include "expop.h" #include "nat/x86-linux.h" +#include "nat/x86-linux-tdesc.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_AMD64 "syscalls/amd64-linux.xml" diff --git a/gdb/amd64-linux-tdep.h b/gdb/amd64-linux-tdep.h index 2003dcda78f..0ec49e7fe03 100644 --- a/gdb/amd64-linux-tdep.h +++ b/gdb/amd64-linux-tdep.h @@ -43,12 +43,6 @@ extern struct target_desc *tdesc_x32_linux; extern struct target_desc *tdesc_x32_avx_linux; extern struct target_desc *tdesc_x32_avx_avx512_linux; -/* Return the right amd64-linux target descriptions according to - XCR0_FEATURES_BIT and IS_X32. */ - -const target_desc *amd64_linux_read_description (uint64_t xcr0_features_bit, - bool is_x32); - /* Enum that defines the syscall identifiers for amd64 linux. Used for process record/replay, these will be translated into a gdb-canonical set of syscall ids in linux-record.c. */ diff --git a/gdb/configure.nat b/gdb/configure.nat index 8b98511cef7..4bcc0696027 100644 --- a/gdb/configure.nat +++ b/gdb/configure.nat @@ -256,7 +256,7 @@ case ${gdb_host} in NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ nat/x86-xstate.o \ i386-linux-nat.o x86-linux-nat.o nat/linux-btrace.o \ - nat/x86-linux.o nat/x86-linux-dregs.o" + nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o" ;; ia64) # Host: Intel IA-64 running GNU/Linux @@ -322,7 +322,7 @@ case ${gdb_host} in NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ nat/x86-xstate.o amd64-nat.o amd64-linux-nat.o x86-linux-nat.o \ nat/linux-btrace.o \ - nat/x86-linux.o nat/x86-linux-dregs.o \ + nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o \ nat/amd64-linux-siginfo.o" ;; sparc) diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index 43cb2961e28..895573f8b4f 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -41,6 +41,7 @@ #include "i387-tdep.h" #include "gdbsupport/x86-xstate.h" #include "nat/x86-linux.h" +#include "nat/x86-linux-tdesc.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_I386 "syscalls/i386-linux.xml" diff --git a/gdb/i386-linux-tdep.h b/gdb/i386-linux-tdep.h index 07593c6a8ec..e8691cd778e 100644 --- a/gdb/i386-linux-tdep.h +++ b/gdb/i386-linux-tdep.h @@ -55,9 +55,6 @@ extern void i386_linux_report_signal_info (struct gdbarch *gdbarch, struct ui_out *uiout, enum gdb_signal siggnal); -/* Return the target description according to XCR0. */ -extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); - extern int i386_linux_gregset_reg_offset[]; /* Return x86 siginfo type. */ diff --git a/gdb/nat/x86-linux-tdesc.c b/gdb/nat/x86-linux-tdesc.c new file mode 100644 index 00000000000..78d22f55d67 --- /dev/null +++ b/gdb/nat/x86-linux-tdesc.c @@ -0,0 +1,118 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#include "gdbsupport/common-defs.h" +#include "nat/x86-linux-tdesc.h" +#ifdef __x86_64__ +#include "arch/amd64.h" +#endif +#include "arch/i386.h" + +#include "gdbsupport/common-defs.h" +#include "nat/x86-linux.h" +#include "nat/x86-linux-dregs.h" +#include "nat/gdb_ptrace.h" +#include "nat/x86-xstate.h" +#include "nat/x86-linux-tdesc.h" + +#include <cstdint> +#include <cstdlib> +#include <linux/uio.h> +#include <elf.h> +#include <sys/user.h> +#include <sys/user.h> + +/* See nat/x86-linux-tdesc.h. */ + +const target_desc * +x86_linux_tdesc_for_tid (int tid, enum tribool *have_ptrace_getregset, + gdb::function_view<void (uint64_t)> xcr0_init_cb, + const char *error_msg, uint64_t *xcr0_storage) +{ +#ifdef __x86_64__ + + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); + bool is_64bit = arch_size.is_64bit (); + bool is_x32 = arch_size.is_x32 (); + + if (sizeof (void *) == 4 && is_64bit && !is_x32) + error ("%s", error_msg); + +#elif HAVE_PTRACE_GETFPXREGS + if (have_ptrace_getfpxregs == -1) + { + elf_fpxregset_t fpxregs; + + if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) + { + have_ptrace_getfpxregs = 0; + *have_ptrace_getregset = TRIBOOL_FALSE; + return i386_linux_read_description (X86_XSTATE_X87_MASK); + } + } +#endif + + if (*have_ptrace_getregset == TRIBOOL_UNKNOWN) + { + uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; + struct iovec iov; + + iov.iov_base = xstateregs; + iov.iov_len = sizeof (xstateregs); + + /* Check if PTRACE_GETREGSET works. */ + if (ptrace (PTRACE_GETREGSET, tid, + (unsigned int) NT_X86_XSTATE, &iov) < 0) + { + *have_ptrace_getregset = TRIBOOL_FALSE; + *xcr0_storage = 0; + } + else + { + *have_ptrace_getregset = TRIBOOL_TRUE; + + /* Get XCR0 from XSAVE extended state. */ + *xcr0_storage = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET + / sizeof (uint64_t))]; + + xcr0_init_cb (*xcr0_storage); + } + } + + /* Check the native XCR0 only if PTRACE_GETREGSET is available. If + PTRACE_GETREGSET is not available then set xcr0_features_bits to + zero so that the "no-features" descriptions are returned by the + switches below. */ + uint64_t xcr0_features_bits; + if (*have_ptrace_getregset == TRIBOOL_TRUE) + xcr0_features_bits = *xcr0_storage & X86_XSTATE_ALL_MASK; + else + xcr0_features_bits = 0; + +#ifdef __x86_64__ + if (is_64bit) + { + return amd64_linux_read_description (xcr0_features_bits, is_x32); + } + else +#endif + return i386_linux_read_description (xcr0_features_bits); + + gdb_assert_not_reached ("failed to return tdesc"); +} diff --git a/gdb/nat/x86-linux-tdesc.h b/gdb/nat/x86-linux-tdesc.h new file mode 100644 index 00000000000..3727a8bf95e --- /dev/null +++ b/gdb/nat/x86-linux-tdesc.h @@ -0,0 +1,75 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#ifndef NAT_X86_LINUX_TDESC_H +#define NAT_X86_LINUX_TDESC_H + +#include "gdbsupport/function-view.h" + +struct target_desc; + +/* Return the target description for Linux thread TID. + + When *HAVE_PTRACE_GETREGSET is TRIBOOL_UNKNOWN then the current value of + xcr0 is read using ptrace calls and stored into *XCR0_STORAGE. Then + XCR0_INIT_CB is called with the value of *XCR0_STORAGE and + *HAVE_PTRACE_GETREGSET is set to TRIBOOL_TRUE. + + If the attempt to read xcr0 using ptrace fails then *XCR0_STORAGE is set + to zero and *HAVE_PTRACE_GETREGSET is set to TRIBOOL_FALSE. + + The storage pointed to by XCR0_STORAGE must exist until the program + terminates, this storage is used to cache the xcr0 value. As such + XCR0_INIT_CB will only be called once if xcr0 is successfully read using + ptrace, or not at all if the ptrace call fails. + + This function returns a target description based on the extracted xcr0 + value along with other characteristics of the thread identified by TID. + + This function can return nullptr if we encounter a machine configuration + for which a target_desc cannot be created. Ideally this would not be + the case, we should be able to create a target description for every + possible machine configuration. See amd64_linux_read_description and + i386_linux_read_description for cases when nullptr might be + returned. + + ERROR_MSG is using in an error() call if we try to create a target + description for a 64-bit process but this is a 32-bit build of GDB. */ + +extern const target_desc * +x86_linux_tdesc_for_tid (int tid, enum tribool *have_ptrace_getregset, + gdb::function_view<void (uint64_t)> xcr0_init_cb, + const char *error_msg, uint64_t *xcr0_storage); + +#ifdef __x86_64__ + +/* Return the right amd64-linux target descriptions according to + XCR0_FEATURES_BIT and IS_X32. This is implemented separately in both + GDB and gdbserver. */ + +extern const target_desc *amd64_linux_read_description + (uint64_t xcr0_features_bit, bool is_x32); + +#endif + +/* Return the target description according to XCR0. This is implemented + separately in both GDB and gdbserver. */ +extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); + +#endif /* NAT_X86_LINUX_TDESC_H */ diff --git a/gdb/x86-linux-nat.c b/gdb/x86-linux-nat.c index b39d05c401f..872e27e739f 100644 --- a/gdb/x86-linux-nat.c +++ b/gdb/x86-linux-nat.c @@ -42,6 +42,7 @@ #include "nat/x86-linux.h" #include "nat/x86-linux-dregs.h" #include "nat/linux-ptrace.h" +#include "nat/x86-linux-tdesc.h" /* linux_nat_target::low_new_fork implementation. */ @@ -96,90 +97,26 @@ x86_linux_nat_target::post_startup_inferior (ptid_t ptid) const struct target_desc * x86_linux_nat_target::read_description () { - int tid; - int is_64bit = 0; -#ifdef __x86_64__ - int is_x32; -#endif - static uint64_t xcr0; - uint64_t xcr0_features_bits; + static uint64_t xcr0_storage; if (inferior_ptid == null_ptid) return this->beneath ()->read_description (); - tid = inferior_ptid.pid (); - -#ifdef __x86_64__ - - x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); - is_64bit = arch_size.is_64bit (); - is_x32 = arch_size.is_x32 (); - -#elif HAVE_PTRACE_GETFPXREGS - if (have_ptrace_getfpxregs == -1) - { - elf_fpxregset_t fpxregs; - - if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) - { - have_ptrace_getfpxregs = 0; - have_ptrace_getregset = TRIBOOL_FALSE; - return i386_linux_read_description (X86_XSTATE_X87_MASK); - } - } -#endif - - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) - { - uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; - struct iovec iov; - - iov.iov_base = xstateregs; - iov.iov_len = sizeof (xstateregs); - - /* Check if PTRACE_GETREGSET works. */ - if (ptrace (PTRACE_GETREGSET, tid, - (unsigned int) NT_X86_XSTATE, &iov) < 0) - have_ptrace_getregset = TRIBOOL_FALSE; - else - { - have_ptrace_getregset = TRIBOOL_TRUE; - - /* Get XCR0 from XSAVE extended state. */ - xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET - / sizeof (uint64_t))]; - - m_xsave_layout = x86_fetch_xsave_layout (xcr0, x86_xsave_length ()); - } - } - - /* Check the native XCR0 only if PTRACE_GETREGSET is available. If - PTRACE_GETREGSET is not available then set xcr0_features_bits to - zero so that the "no-features" descriptions are returned by the - switches below. */ - if (have_ptrace_getregset == TRIBOOL_TRUE) - xcr0_features_bits = xcr0 & X86_XSTATE_ALL_MASK; - else - xcr0_features_bits = 0; - - if (is_64bit) - { -#ifdef __x86_64__ - return amd64_linux_read_description (xcr0_features_bits, is_x32); -#endif - } - else - { - const struct target_desc * tdesc - = i386_linux_read_description (xcr0_features_bits); + int tid = inferior_ptid.pid (); - if (tdesc == NULL) - tdesc = i386_linux_read_description (X86_XSTATE_SSE_MASK); + const char *error_msg + = _("Can't debug 64-bit process with 32-bit GDB"); - return tdesc; - } + /* Callback that is triggered the first time x86_linux_tdesc_for_tid + reads the xcr0 register. Setup other bits of state */ + auto cb = [&] (uint64_t xcr0) + { + this->m_xsave_layout + = x86_fetch_xsave_layout (xcr0, x86_xsave_length ()); + }; - gdb_assert_not_reached ("failed to return tdesc"); + return x86_linux_tdesc_for_tid (tid, &have_ptrace_getregset, cb, + error_msg, &xcr0_storage); } \f diff --git a/gdbserver/configure.srv b/gdbserver/configure.srv index 9e861a75088..7a2702d78bf 100644 --- a/gdbserver/configure.srv +++ b/gdbserver/configure.srv @@ -109,6 +109,7 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/linux-btrace.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" + srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" srv_linux_usrregs=yes srv_linux_regsets=yes srv_linux_thread_db=yes @@ -371,6 +372,7 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/linux-btrace.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" + srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" srv_tgtobj="${srv_tgtobj} nat/amd64-linux-siginfo.o" srv_linux_usrregs=yes # This is for i386 progs. srv_linux_regsets=yes diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index 54e4c9812bb..f97b0d6a1d9 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -23,6 +23,7 @@ #include "tracepoint.h" #include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" +#include "nat/x86-linux-tdesc.h" /* Defined in auto-generated file amd64-linux.c. */ void init_registers_amd64_linux (void); diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc index 2e4646f8c03..459b8055b5c 100644 --- a/gdbserver/linux-i386-ipa.cc +++ b/gdbserver/linux-i386-ipa.cc @@ -23,6 +23,7 @@ #include "tracepoint.h" #include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" +#include "nat/x86-linux-tdesc.h" /* GDB register numbers. */ diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 7bade8b3cbf..9bf369f8a34 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -48,6 +48,7 @@ #include "nat/x86-linux.h" #include "nat/x86-linux-dregs.h" #include "linux-x86-tdesc.h" +#include "nat/x86-linux-tdesc.h" #ifdef __x86_64__ static target_desc_up tdesc_amd64_linux_no_xml; @@ -844,32 +845,20 @@ int have_ptrace_getfpxregs = #endif ; +/* Cached xcr0 value. This is initialised the first time + x86_linux_read_description is called. */ + +static uint64_t xcr0_storage; + /* Get Linux/x86 target description from running target. */ static const struct target_desc * x86_linux_read_description (void) { - unsigned int machine; - int is_elf64; - int xcr0_features; - int tid; - static uint64_t xcr0; - static int xsave_len; - struct regset_info *regset; - - tid = lwpid_of (current_thread); + int tid = lwpid_of (current_thread); - is_elf64 = linux_pid_exe_is_elf_64_file (tid, &machine); - - if (sizeof (void *) == 4) - { - if (is_elf64 > 0) - error (_("Can't debug 64-bit process with 32-bit GDBserver")); -#ifndef __x86_64__ - else if (machine == EM_X86_64) - error (_("Can't debug x86-64 process with 32-bit GDBserver")); -#endif - } + const char *error_msg + = _("Can't debug 64-bit process with 32-bit GDBserver"); /* If we are not allowed to send an XML target description then we need to use the hard-wired target descriptions. This corresponds to GDB's @@ -879,99 +868,54 @@ x86_linux_read_description (void) generate some alternative target descriptions. */ if (!use_xml) { + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); + bool is_64bit = arch_size.is_64bit (); + bool is_x32 = arch_size.is_x32 (); + + if (sizeof (void *) == 4 && is_64bit && !is_x32) + error ("%s", error_msg); + #ifdef __x86_64__ - if (machine == EM_X86_64) + if (is_64bit && !is_x32) return tdesc_amd64_linux_no_xml.get (); else #endif return tdesc_i386_linux_no_xml.get (); } -#if !defined __x86_64__ && defined HAVE_PTRACE_GETFPXREGS - if (machine == EM_386 && have_ptrace_getfpxregs == -1) - { - elf_fpxregset_t fpxregs; - - if (ptrace (PTRACE_GETFPXREGS, tid, 0, (long) &fpxregs) < 0) - { - have_ptrace_getfpxregs = 0; - have_ptrace_getregset = TRIBOOL_FALSE; - return i386_linux_read_description (X86_XSTATE_X87); - } - else - have_ptrace_getfpxregs = 1; - } -#endif - - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) - { - uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; - struct iovec iov; - - iov.iov_base = xstateregs; - iov.iov_len = sizeof (xstateregs); - - /* Check if PTRACE_GETREGSET works. */ - if (ptrace (PTRACE_GETREGSET, tid, - (unsigned int) NT_X86_XSTATE, (long) &iov) < 0) - have_ptrace_getregset = TRIBOOL_FALSE; - else - { - have_ptrace_getregset = TRIBOOL_TRUE; - - /* Get XCR0 from XSAVE extended state. */ - xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET - / sizeof (uint64_t))]; - - xsave_len = x86_xsave_length (); - - /* Use PTRACE_GETREGSET if it is available. */ - for (regset = x86_regsets; - regset->fill_function != NULL; regset++) - if (regset->get_request == PTRACE_GETREGSET) - regset->size = xsave_len; - else if (regset->type != GENERAL_REGS) - regset->size = 0; - } - } - - /* Check the native XCR0 only if PTRACE_GETREGSET is available. */ - xcr0_features = (have_ptrace_getregset == TRIBOOL_TRUE - && (xcr0 & X86_XSTATE_ALL_MASK)); - - if (xcr0_features) - i387_set_xsave_mask (xcr0, xsave_len); + /* Callback that is triggered the first time x86_linux_tdesc_for_tid + reads the xcr0 register. Setup other bits of state */ + auto cb = [] (uint64_t xcr0) + { + i387_set_xsave_mask (xcr0, x86_xsave_length ()); + }; - if (machine == EM_X86_64) - { -#ifdef __x86_64__ - const target_desc *tdesc = NULL; + /* If have_ptrace_getregset is changed to true by calling + x86_linux_tdesc_for_tid then we will perform some additional + initialisation. */ + bool have_ptrace_getregset_is_unknown + = have_ptrace_getregset == TRIBOOL_UNKNOWN; - if (xcr0_features) - { - tdesc = amd64_linux_read_description (xcr0 & X86_XSTATE_ALL_MASK, - !is_elf64); - } + const target_desc *tdesc + = x86_linux_tdesc_for_tid (tid, &have_ptrace_getregset, cb, error_msg, + &xcr0_storage); - if (tdesc == NULL) - tdesc = amd64_linux_read_description (X86_XSTATE_SSE_MASK, !is_elf64); - return tdesc; -#endif - } - else + if (have_ptrace_getregset_is_unknown + && have_ptrace_getregset == TRIBOOL_TRUE) { - const target_desc *tdesc = NULL; - - if (xcr0_features) - tdesc = i386_linux_read_description (xcr0 & X86_XSTATE_ALL_MASK); - - if (tdesc == NULL) - tdesc = i386_linux_read_description (X86_XSTATE_SSE); - - return tdesc; + int xsave_len = x86_xsave_length (); + + /* Use PTRACE_GETREGSET if it is available. */ + for (regset_info *regset = x86_regsets; + regset->fill_function != nullptr; + regset++) + if (regset->get_request == PTRACE_GETREGSET) + regset->size = xsave_len; + else if (regset->type != GENERAL_REGS) + regset->size = 0; } - gdb_assert_not_reached ("failed to return tdesc"); + return tdesc; } /* Update all the target description of all processes; a new GDB diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc index 626207fc477..9fd64d8574b 100644 --- a/gdbserver/linux-x86-tdesc.cc +++ b/gdbserver/linux-x86-tdesc.cc @@ -26,6 +26,7 @@ #include "arch/amd64.h" #endif #include "x86-tdesc.h" +#include "nat/x86-linux-tdesc.h" /* Return the right x86_linux_tdesc index for a given XCR0. Return X86_TDESC_LAST if can't find a match. */ diff --git a/gdbserver/linux-x86-tdesc.h b/gdbserver/linux-x86-tdesc.h index f9561b129ae..576aaf5e165 100644 --- a/gdbserver/linux-x86-tdesc.h +++ b/gdbserver/linux-x86-tdesc.h @@ -46,11 +46,4 @@ int amd64_get_ipa_tdesc_idx (const struct target_desc *tdesc); const struct target_desc *i386_get_ipa_tdesc (int idx); -#ifdef __x86_64__ -const struct target_desc *amd64_linux_read_description (uint64_t xcr0, - bool is_x32); -#endif - -const struct target_desc *i386_linux_read_description (uint64_t xcr0); - #endif /* GDBSERVER_LINUX_X86_TDESC_H */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCH 6/7] gdbserver: update target description creation for x86/linux 2024-02-01 15:28 [PATCH 0/7] x86/Linux Target Description Changes Andrew Burgess ` (4 preceding siblings ...) 2024-02-01 15:28 ` [PATCH 5/7] gdb/gdbserver: share some code relating to target description creation Andrew Burgess @ 2024-02-01 15:28 ` Andrew Burgess 2024-02-01 15:28 ` [PATCH 7/7] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess 2024-03-05 17:00 ` [PATCHv2 0/7] x86/Linux Target Description Changes Andrew Burgess 7 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-02-01 15:28 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess This commit is part of a series which aims to share more of the target description creation between GDB and gdbserver for x86/Linux. After some refactoring, the previous commit actually started to share some code, we added the shared x86_linux_tdesc_for_tid function into nat/x86-linux-tdesc.c. However, this function still relies on amd64_linux_read_description and i386_linux_read_description which are implemented separately for both gdbserver and GDB. Given that at their core, all these functions to is: 1. take an xcr0 value as input, 2. mask out some feature bits, 3. look for a cached pre-generated target description and return it if found, 4. if no cached target description is found then call either amd64_create_target_description or i386_create_target_description to create a new target description, which is then added to the cache. Return the newly created target description. The inner functions amd64_create_target_description and i386_create_target_description are already shared between GDB and gdbserver (in the gdb/arch/ directory), so the only thing that the *_read_description functions really do is add the caching layer, and it feels like this really could be shared. However, we have a small problem. On the GDB side we create target descriptions using a different set of cpu features than on the gdbserver side! This means that for the exact same target, we might get a different target description when using native GDB vs using gdbserver. This surely feels like a mistake, I would expect to get the same target description on each. The table below shows the number of possible different target descriptions that we can create on the GDB side vs on the gdbserver side for each target type: | GDB | gdbserver ------|-----|---------- i386 | 64 | 7 amd64 | 32 | 7 x32 | 16 | 7 So in theory, all I want to do is move the GDB version of *_read_description into the nat/ directory and have gdbserver use that, then both GDB and gdbserver would be able to create any of the possible target descriptions. Unfortunately it's a little more complex than that due to the in process agent (IPA). When the IPA is in use, gdbserver sends a target description index to the IPA, and the IPA uses this to find the correct target description to use. ** START OF AN ASIDE ** Back in the day I suspect this approach made perfect sense. However since this commit: commit a8806230241d201f808d856eaae4d44088117b0c Date: Thu Dec 7 17:07:01 2017 +0000 Initialize target description early in IPA I think passing the index is now more trouble than its worth. We used to pass the index, and then use that index to lookup which target description to instantiate and use. However, the above commit fixed an issue where we can't call malloc() within (certain parts of) the IPA (apparently), so instead we now pre-compute _every_ possible target description within the IPA. The index is now only used to lookup which of the (many) pre-computed target descriptions to use. It would (I think) have been easier all around if the IPA just self-inspected, figured out its own xcr0 value, and used that to create the one target description that is required. So long as the xcr0 to target description code is shared (at compile time) with gdbserver, then we can be sure that the IPA will derive the same target description as gdbserver, and we would avoid all this index passing business, which has made this commit so very, very painful. ** END OF AN ASIDE ** Currently then for x86/linux, gdbserver sends a number between 0 and 7 to the IPA, and the IPA uses this to create a target description. However, I am proposing that gdbserver should now create one of (up to) 64 different target descriptions for i386, so this 0 to 7 index isn't going to be good enough any more (amd64 and x32 have slightly fewer possible target descriptions, but still more than 8, so the problem is the same). For a while I wondered if I was going to have to try and find some backward compatible solution for this mess. But after seeing how lightly the IPA is actually documented, I wonder if it is not the case that there is a tight coupling between a version of gdbserver and a version of the IPA? At least I'm hoping so. In this commit I have thrown out the old IPA target description index numbering scheme, and switched to a completely new numbering scheme. Instead of the index that is passed being arbitrary, the index is instead calculated from the set of cpu features that are present on the target. Within the IPA we can then reverse this logic to recreate the xcr0 value based on the index, and from the xcr0 value we can create the correct target description. With the gdbserver to IPA numbering scheme issue resolved I have then update the gdbserver versions of amd64_linux_read_description and i386_linux_read_description so that they create target descriptions using the same set of cpu features as GDB itself. After this gdbserver should now always come up with the same target description as GDB does on any x86/Linux target. This commit does not introduce any new code sharing between GDB and gdbserver as previous commits in this series does. Instead this commit is all about bringing GDB and gdbserver into alignment functionally so that the next commit can merge the GDB and gdbserver versions of these functions. --- gdbserver/linux-amd64-ipa.cc | 43 +---- gdbserver/linux-i386-ipa.cc | 23 +-- gdbserver/linux-x86-low.cc | 15 +- gdbserver/linux-x86-tdesc.cc | 315 +++++++++++++++++++++++++---------- gdbserver/linux-x86-tdesc.h | 49 +++--- 5 files changed, 277 insertions(+), 168 deletions(-) diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index f97b0d6a1d9..42de7d2da16 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -169,47 +169,21 @@ supply_static_tracepoint_registers (struct regcache *regcache, #endif /* HAVE_UST */ -#if !defined __ILP32__ -/* Map the tdesc index to xcr0 mask. */ -static uint64_t idx2mask[X86_TDESC_LAST] = { - X86_XSTATE_X87_MASK, - X86_XSTATE_SSE_MASK, - X86_XSTATE_AVX_MASK, - X86_XSTATE_MPX_MASK, - X86_XSTATE_AVX_MPX_MASK, - X86_XSTATE_AVX_AVX512_MASK, - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, -}; -#endif - /* Return target_desc to use for IPA, given the tdesc index passed by gdbserver. */ const struct target_desc * get_ipa_tdesc (int idx) { - if (idx >= X86_TDESC_LAST) - { - internal_error ("unknown ipa tdesc index: %d", idx); - } + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); #if defined __ILP32__ - switch (idx) - { - case X86_TDESC_SSE: - return amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); - case X86_TDESC_AVX: - return amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); - case X86_TDESC_AVX_AVX512: - return amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, true); - default: - break; - } + bool is_x32 = true; #else - return amd64_linux_read_description (idx2mask[idx], false); + bool is_x32 = false; #endif - internal_error ("unknown ipa tdesc index: %d", idx); + return amd64_linux_read_description (xcr0, is_x32); } /* Allocate buffer for the jump pads. The branch instruction has a @@ -277,11 +251,10 @@ void initialize_low_tracepoint (void) { #if defined __ILP32__ - amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); - amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); - amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, true); + for (auto i = 0; i < x86_linux_x32_ipa_tdesc_count (); i++) + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), true); #else - for (auto i = 0; i < X86_TDESC_LAST; i++) - amd64_linux_read_description (idx2mask[i], false); + for (auto i = 0; i < x86_linux_amd64_ipa_tdesc_count (); i++) + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), false); #endif } diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc index 459b8055b5c..246bcb9813b 100644 --- a/gdbserver/linux-i386-ipa.cc +++ b/gdbserver/linux-i386-ipa.cc @@ -246,28 +246,15 @@ initialize_fast_tracepoint_trampoline_buffer (void) } } -/* Map the tdesc index to xcr0 mask. */ -static uint64_t idx2mask[X86_TDESC_LAST] = { - X86_XSTATE_X87_MASK, - X86_XSTATE_SSE_MASK, - X86_XSTATE_AVX_MASK, - X86_XSTATE_MPX_MASK, - X86_XSTATE_AVX_MPX_MASK, - X86_XSTATE_AVX_AVX512_MASK, - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, -}; - /* Return target_desc to use for IPA, given the tdesc index passed by gdbserver. */ const struct target_desc * get_ipa_tdesc (int idx) { - if (idx >= X86_TDESC_LAST) - { - internal_error ("unknown ipa tdesc index: %d", idx); - } - return i386_linux_read_description (idx2mask[idx]); + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); + + return i386_linux_read_description (xcr0); } /* Allocate buffer for the jump pads. On i386, we can reach an arbitrary @@ -289,6 +276,6 @@ void initialize_low_tracepoint (void) { initialize_fast_tracepoint_trampoline_buffer (); - for (auto i = 0; i < X86_TDESC_LAST; i++) - i386_linux_read_description (idx2mask[i]); + for (auto i = 0; i < x86_linux_i386_ipa_tdesc_count (); i++) + i386_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i)); } diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 9bf369f8a34..6dbefdb2f26 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -2892,14 +2892,17 @@ x86_target::get_ipa_tdesc_idx () struct regcache *regcache = get_thread_regcache (current_thread, 0); const struct target_desc *tdesc = regcache->tdesc; + if (!use_xml) + { + if (tdesc == tdesc_i386_linux_no_xml.get () #ifdef __x86_64__ - return amd64_get_ipa_tdesc_idx (tdesc); -#endif - - if (tdesc == tdesc_i386_linux_no_xml.get ()) - return X86_TDESC_SSE; + || tdesc == tdesc_amd64_linux_no_xml.get () +#endif /* __x86_64__ */ + ) + return x86_linux_xcr0_to_tdesc_idx (X86_XSTATE_SSE_MASK); + } - return i386_get_ipa_tdesc_idx (tdesc); + return x86_linux_xcr0_to_tdesc_idx (xcr0_storage); } /* The linux target ops object. */ diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc index 9fd64d8574b..87cf368a336 100644 --- a/gdbserver/linux-x86-tdesc.cc +++ b/gdbserver/linux-x86-tdesc.cc @@ -28,96 +28,277 @@ #include "x86-tdesc.h" #include "nat/x86-linux-tdesc.h" -/* Return the right x86_linux_tdesc index for a given XCR0. Return - X86_TDESC_LAST if can't find a match. */ +/* A structure used to describe a single cpu feature that might, or might + not, be checked for when creating a target description for one of i386, + amd64, or x32. */ -static enum x86_linux_tdesc -xcr0_to_tdesc_idx (uint64_t xcr0, bool is_x32) +struct x86_tdesc_feature { + /* The cpu feature mask. This is a mask against an xcr0 value. */ + uint64_t feature; + + /* Is this feature checked when creating an i386 target description. */ + bool is_i386; + + /* Is this feature checked when creating an amd64 target description. */ + bool is_amd64; + + /* Is this feature checked when creating an x32 target description. */ + bool is_x32; +}; + +/* A constant table that describes all of the cpu features that are + checked when building a target description for i386, amd64, or x32. */ + +static constexpr x86_tdesc_feature x86_linux_all_tdesc_features[] = { + /* Feature, i386, amd64, x32. */ + { X86_XSTATE_PKRU, true, true, true }, + { X86_XSTATE_AVX512, true, true, true }, + { X86_XSTATE_AVX, true, true, true }, + { X86_XSTATE_MPX, true, true, false }, + { X86_XSTATE_SSE, true, false, false }, + { X86_XSTATE_X87, true, false, false } +}; + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an i386 target description. */ + +static constexpr uint64_t +x86_linux_i386_tdesc_feature_mask () { - if (xcr0 & X86_XSTATE_PKRU) - { - if (is_x32) - { - /* No x32 MPX and PKU, fall back to avx_avx512. */ - return X86_TDESC_AVX_AVX512; - } - else - return X86_TDESC_AVX_MPX_AVX512_PKU; - } - else if (xcr0 & X86_XSTATE_AVX512) - return X86_TDESC_AVX_AVX512; - else if ((xcr0 & X86_XSTATE_AVX_MPX_MASK) == X86_XSTATE_AVX_MPX_MASK) + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_i386) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an amd64 target description. */ + +static constexpr uint64_t +x86_linux_amd64_tdesc_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_amd64) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an x32 target description. */ + +static constexpr uint64_t +x86_linux_x32_tdesc_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_x32) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an i386 target description. */ + +static constexpr int +x86_linux_i386_tdesc_count () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_i386) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an amd64 target description. */ + +static constexpr int +x86_linux_amd64_tdesc_count () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_amd64) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an x32 target description. */ + +static constexpr int +x86_linux_x32_tdesc_count () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_x32) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +#ifdef IN_PROCESS_AGENT + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_amd64_ipa_tdesc_count () +{ + return x86_linux_amd64_tdesc_count (); +} + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_x32_ipa_tdesc_count () +{ + return x86_linux_x32_tdesc_count (); +} + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_i386_ipa_tdesc_count () +{ + return x86_linux_i386_tdesc_count (); +} + +#endif /* IN_PROCESS_AGENT */ + +/* Convert an xcr0 value into an integer. The integer will be passed to + the in-process-agent where it will then be passed to + x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ + +int +x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0) +{ + /* The following table shows which features are checked for when creating + the target descriptions (see nat/x86-linux-tdesc.c), the feature order + represents the bit order within the generated index number. + + i386 | x87 sse mpx avx avx512 pkru + amd64 | mpx avx avx512 pkru + i32 | avx avx512 pkru + + The features are ordered so that for each mode (i386, amd64, i32) the + generated index will form a continuous range. */ + + int idx = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) { - if (is_x32) /* No MPX on x32. */ - return X86_TDESC_AVX; - else - return X86_TDESC_AVX_MPX; + if ((xcr0 & x86_linux_all_tdesc_features[i].feature) != 0) + idx |= (1 << i); } - else if (xcr0 & X86_XSTATE_MPX) + + return idx; +} + + +#ifdef IN_PROCESS_AGENT + +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +uint64_t +x86_linux_tdesc_idx_to_xcr0 (int idx) +{ + uint64_t xcr0 = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) { - if (is_x32) /* No MPX on x32. */ - return X86_TDESC_AVX; - else - return X86_TDESC_MPX; + if ((idx & (1 << i)) != 0) + xcr0 |= x86_linux_all_tdesc_features[i].feature; } - else if (xcr0 & X86_XSTATE_AVX) - return X86_TDESC_AVX; - else if (xcr0 & X86_XSTATE_SSE) - return X86_TDESC_SSE; - else if (xcr0 & X86_XSTATE_X87) - return X86_TDESC_MMX; - else - return X86_TDESC_LAST; + + return xcr0; } +#endif /* IN_PROCESS_AGENT */ + #if defined __i386__ || !defined IN_PROCESS_AGENT -static struct target_desc *i386_tdescs[X86_TDESC_LAST] = { }; +/* A cache of all possible i386 target descriptions. */ -/* Return the target description according to XCR0. */ +static struct target_desc *i386_tdescs[x86_linux_i386_tdesc_count ()] = { }; + +/* See nat/x86-linux-tdesc.h. */ const struct target_desc * i386_linux_read_description (uint64_t xcr0) { - enum x86_linux_tdesc idx = xcr0_to_tdesc_idx (xcr0, false); + xcr0 &= x86_linux_i386_tdesc_feature_mask (); + int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - if (idx == X86_TDESC_LAST) - return NULL; + gdb_assert (idx >= 0 && idx < x86_linux_i386_tdesc_count ()); - struct target_desc **tdesc = &i386_tdescs[idx]; + target_desc **tdesc = &i386_tdescs[idx]; - if (*tdesc == NULL) + if (*tdesc == nullptr) { *tdesc = i386_create_target_description (xcr0, true, false); init_target_desc (*tdesc, i386_expedite_regs); } - return *tdesc;; + return *tdesc; } #endif #ifdef __x86_64__ -static target_desc *amd64_tdescs[X86_TDESC_LAST] = { }; -static target_desc *x32_tdescs[X86_TDESC_LAST] = { }; +/* A cache of all possible amd64 target descriptions. */ + +static target_desc *amd64_tdescs[x86_linux_amd64_tdesc_count ()] = { }; + +/* A cache of all possible x32 target descriptions. */ + +static target_desc *x32_tdescs[x86_linux_x32_tdesc_count ()] = { }; + +/* See nat/x86-linux-tdesc.h. */ const struct target_desc * amd64_linux_read_description (uint64_t xcr0, bool is_x32) { - enum x86_linux_tdesc idx = xcr0_to_tdesc_idx (xcr0, is_x32); + if (is_x32) + xcr0 &= x86_linux_x32_tdesc_feature_mask (); + else + xcr0 &= x86_linux_amd64_tdesc_feature_mask (); + + int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - if (idx == X86_TDESC_LAST) - return NULL; + if (is_x32) + gdb_assert (idx >= 0 && idx < x86_linux_x32_tdesc_count ()); + else + gdb_assert (idx >= 0 && idx < x86_linux_amd64_tdesc_count ()); - struct target_desc **tdesc = NULL; + target_desc **tdesc = nullptr; if (is_x32) tdesc = &x32_tdescs[idx]; else tdesc = &amd64_tdescs[idx]; - if (*tdesc == NULL) + if (*tdesc == nullptr) { *tdesc = amd64_create_target_description (xcr0, is_x32, true, true); @@ -127,39 +308,3 @@ amd64_linux_read_description (uint64_t xcr0, bool is_x32) } #endif - -#ifndef IN_PROCESS_AGENT - -int -i386_get_ipa_tdesc_idx (const struct target_desc *tdesc) -{ - for (int i = 0; i < X86_TDESC_LAST; i++) - { - if (tdesc == i386_tdescs[i]) - return i; - } - - /* If none tdesc is found, return the one with minimum features. */ - return X86_TDESC_MMX; -} - -#if defined __x86_64__ -int -amd64_get_ipa_tdesc_idx (const struct target_desc *tdesc) -{ - for (int i = 0; i < X86_TDESC_LAST; i++) - { - if (tdesc == amd64_tdescs[i]) - return i; - } - for (int i = 0; i < X86_TDESC_LAST; i++) - { - if (tdesc == x32_tdescs[i]) - return i; - } - - return X86_TDESC_SSE; -} - -#endif -#endif diff --git a/gdbserver/linux-x86-tdesc.h b/gdbserver/linux-x86-tdesc.h index 576aaf5e165..70456e4be44 100644 --- a/gdbserver/linux-x86-tdesc.h +++ b/gdbserver/linux-x86-tdesc.h @@ -21,29 +21,30 @@ #ifndef GDBSERVER_LINUX_X86_TDESC_H #define GDBSERVER_LINUX_X86_TDESC_H -/* Note: since IPA obviously knows what ABI it's running on (i386 vs x86_64 - vs x32), it's sufficient to pass only the register set here. This, - together with the ABI known at IPA compile time, maps to a tdesc. */ - -enum x86_linux_tdesc { - X86_TDESC_MMX = 0, - X86_TDESC_SSE = 1, - X86_TDESC_AVX = 2, - X86_TDESC_MPX = 3, - X86_TDESC_AVX_MPX = 4, - X86_TDESC_AVX_AVX512 = 5, - X86_TDESC_AVX_MPX_AVX512_PKU = 6, - X86_TDESC_LAST = 7, -}; - -#if defined __i386__ || !defined IN_PROCESS_AGENT -int i386_get_ipa_tdesc_idx (const struct target_desc *tdesc); -#endif - -#if defined __x86_64__ && !defined IN_PROCESS_AGENT -int amd64_get_ipa_tdesc_idx (const struct target_desc *tdesc); -#endif - -const struct target_desc *i386_get_ipa_tdesc (int idx); +/* Convert an xcr0 value into an integer. The integer will be passed to + the in-process-agent where it will then be passed to + x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ + +extern int x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0); + +#ifdef IN_PROCESS_AGENT + +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +extern uint64_t x86_linux_tdesc_idx_to_xcr0 (int idx); + +/* Within the in-process-agent we need to pre-initialise all of the target + descriptions, to do this we need to know how many target descriptions + there are for each different target type. These functions return the + target description count for the relevant target. */ + +extern int x86_linux_amd64_ipa_tdesc_count (); +extern int x86_linux_x32_ipa_tdesc_count (); +extern int x86_linux_i386_ipa_tdesc_count (); + + +#endif /* IN_PROCESS_AGENT */ #endif /* GDBSERVER_LINUX_X86_TDESC_H */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCH 7/7] gdb/gdbserver: share x86/linux tdesc caching 2024-02-01 15:28 [PATCH 0/7] x86/Linux Target Description Changes Andrew Burgess ` (5 preceding siblings ...) 2024-02-01 15:28 ` [PATCH 6/7] gdbserver: update target description creation for x86/linux Andrew Burgess @ 2024-02-01 15:28 ` Andrew Burgess 2024-03-05 17:00 ` [PATCHv2 0/7] x86/Linux Target Description Changes Andrew Burgess 7 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-02-01 15:28 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess This commit builds on the previous series of commits to share the target description caching code between GDB and gdbserver for x86/Linux targets. The objective of this commit is to move the four functions (2 each of) i386_linux_read_description and amd64_linux_read_description into gdb/nat/x86-linux-tdesc.c and combine them so we have just a single copy of each. Then both GDB and gdbserver will link against these shared functions. It is worth reading the description of the previous commit to see why this merging is not as simple as it seems: on the gdbserver side we actually have two users of these functions, gdbserver itself, and the in process agent (IPA). However, the previous commit streamlined the gdbserver code, and so now it is simple to move the two functions along with all their support functions from the gdbserver directory into the gdb/nat/ directory, and then GDB is fine to call these functions. One small curiosity with this patch is the function x86_linux_post_init_tdesc. On the gdbserver side the two functions amd64_linux_read_description and i386_linux_read_description have some functionality that is not present on the GDB side, that is some additional configuration that is performed as each target description is created to setup the expedited registers. To support this I've added the function x86_linux_post_init_tdesc. This function is called from the two *_linux_read_description functions, but is implemented separately for GDB and gdbserver. This does mean adding back some non-shared code when this whole series has been about sharing code, but now the only non-shared bit is the single line that is actually different between GDB and gdbserver, all the rest, which is identical, is now shared. I did need to add a new rule to the gdbserver Makefile, this is to allow the nat/x86-linux-tdesc.c file to be compiled for the IPA. --- gdb/amd64-linux-tdep.c | 31 ---- gdb/i386-linux-tdep.c | 30 +--- gdb/nat/x86-linux-tdesc.c | 287 +++++++++++++++++++++++++++++++++++ gdb/nat/x86-linux-tdesc.h | 56 ++++++- gdbserver/Makefile.in | 4 + gdbserver/configure.srv | 2 + gdbserver/linux-amd64-ipa.cc | 1 - gdbserver/linux-i386-ipa.cc | 1 - gdbserver/linux-x86-low.cc | 1 - gdbserver/linux-x86-tdesc.cc | 285 +--------------------------------- gdbserver/linux-x86-tdesc.h | 50 ------ 11 files changed, 355 insertions(+), 393 deletions(-) delete mode 100644 gdbserver/linux-x86-tdesc.h diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c index 87cefaa92e9..d489a54abfc 100644 --- a/gdb/amd64-linux-tdep.c +++ b/gdb/amd64-linux-tdep.c @@ -1579,37 +1579,6 @@ amd64_linux_record_signal (struct gdbarch *gdbarch, return 0; } -const target_desc * -amd64_linux_read_description (uint64_t xcr0_features_bit, bool is_x32) -{ - static target_desc *amd64_linux_tdescs \ - [2/*AVX*/][2/*MPX*/][2/*AVX512*/][2/*PKRU*/] = {}; - static target_desc *x32_linux_tdescs \ - [2/*AVX*/][2/*AVX512*/][2/*PKRU*/] = {}; - - target_desc **tdesc; - - if (is_x32) - { - tdesc = &x32_linux_tdescs[(xcr0_features_bit & X86_XSTATE_AVX) ? 1 : 0 ] - [(xcr0_features_bit & X86_XSTATE_AVX512) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_PKRU) ? 1 : 0]; - } - else - { - tdesc = &amd64_linux_tdescs[(xcr0_features_bit & X86_XSTATE_AVX) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_MPX) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_AVX512) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_PKRU) ? 1 : 0]; - } - - if (*tdesc == NULL) - *tdesc = amd64_create_target_description (xcr0_features_bit, is_x32, - true, true); - - return *tdesc; -} - /* Get Linux/x86 target description from core dump. */ static const struct target_desc * diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index 895573f8b4f..0ddd8c665d6 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -680,29 +680,12 @@ i386_linux_core_read_x86_xsave_layout (struct gdbarch *gdbarch, return i386_linux_core_read_xsave_info (core_bfd, layout) != 0; } -/* See i386-linux-tdep.h. */ +/* See nat/x86-linux-tdesc.h. */ -const struct target_desc * -i386_linux_read_description (uint64_t xcr0) +void +x86_linux_post_init_tdesc (target_desc *tdesc, bool is_64bit) { - if (xcr0 == 0) - return NULL; - - static struct target_desc *i386_linux_tdescs \ - [2/*X87*/][2/*SSE*/][2/*AVX*/][2/*MPX*/][2/*AVX512*/][2/*PKRU*/] = {}; - struct target_desc **tdesc; - - tdesc = &i386_linux_tdescs[(xcr0 & X86_XSTATE_X87) ? 1 : 0] - [(xcr0 & X86_XSTATE_SSE) ? 1 : 0] - [(xcr0 & X86_XSTATE_AVX) ? 1 : 0] - [(xcr0 & X86_XSTATE_MPX) ? 1 : 0] - [(xcr0 & X86_XSTATE_AVX512) ? 1 : 0] - [(xcr0 & X86_XSTATE_PKRU) ? 1 : 0]; - - if (*tdesc == NULL) - *tdesc = i386_create_target_description (xcr0, true, false); - - return *tdesc; + /* Nothing. */ } /* Get Linux/x86 target description from core dump. */ @@ -715,7 +698,10 @@ i386_linux_core_read_description (struct gdbarch *gdbarch, /* Linux/i386. */ x86_xsave_layout layout; uint64_t xcr0 = i386_linux_core_read_xsave_info (abfd, layout); - const struct target_desc *tdesc = i386_linux_read_description (xcr0); + + const struct target_desc *tdesc; + if (xcr0 != 0) + tdesc = i386_linux_read_description (xcr0); if (tdesc != NULL) return tdesc; diff --git a/gdb/nat/x86-linux-tdesc.c b/gdb/nat/x86-linux-tdesc.c index 78d22f55d67..297d1873680 100644 --- a/gdb/nat/x86-linux-tdesc.c +++ b/gdb/nat/x86-linux-tdesc.c @@ -38,6 +38,8 @@ #include <sys/user.h> #include <sys/user.h> +#ifndef IN_PROCESS_AGENT + /* See nat/x86-linux-tdesc.h. */ const target_desc * @@ -116,3 +118,288 @@ x86_linux_tdesc_for_tid (int tid, enum tribool *have_ptrace_getregset, gdb_assert_not_reached ("failed to return tdesc"); } + +#endif /* !IN_PROCESS_AGENT */ + + + +/* A structure used to describe a single cpu feature that might, or might + not, be checked for when creating a target description for one of i386, + amd64, or x32. */ + +struct x86_tdesc_feature { + /* The cpu feature mask. This is a mask against an xcr0 value. */ + uint64_t feature; + + /* Is this feature checked when creating an i386 target description. */ + bool is_i386; + + /* Is this feature checked when creating an amd64 target description. */ + bool is_amd64; + + /* Is this feature checked when creating an x32 target description. */ + bool is_x32; +}; + +/* A constant table that describes all of the cpu features that are + checked when building a target description for i386, amd64, or x32. */ + +static constexpr x86_tdesc_feature x86_linux_all_tdesc_features[] = { + /* Feature, i386, amd64, x32. */ + { X86_XSTATE_PKRU, true, true, true }, + { X86_XSTATE_AVX512, true, true, true }, + { X86_XSTATE_AVX, true, true, true }, + { X86_XSTATE_MPX, true, true, false }, + { X86_XSTATE_SSE, true, false, false }, + { X86_XSTATE_X87, true, false, false } +}; + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an i386 target description. */ + +static constexpr uint64_t +x86_linux_i386_tdesc_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_i386) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an amd64 target description. */ + +static constexpr uint64_t +x86_linux_amd64_tdesc_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_amd64) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an x32 target description. */ + +static constexpr uint64_t +x86_linux_x32_tdesc_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_x32) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an i386 target description. */ + +static constexpr int +x86_linux_i386_tdesc_count () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_i386) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an amd64 target description. */ + +static constexpr int +x86_linux_amd64_tdesc_count () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_amd64) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an x32 target description. */ + +static constexpr int +x86_linux_x32_tdesc_count () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_x32) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +#ifdef IN_PROCESS_AGENT + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_amd64_ipa_tdesc_count () +{ + return x86_linux_amd64_tdesc_count (); +} + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_x32_ipa_tdesc_count () +{ + return x86_linux_x32_tdesc_count (); +} + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_i386_ipa_tdesc_count () +{ + return x86_linux_i386_tdesc_count (); +} + +#endif /* IN_PROCESS_AGENT */ + +/* Convert an xcr0 value into an integer. The integer will be passed to + the in-process-agent where it will then be passed to + x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ + +int +x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0) +{ + /* The following table shows which features are checked for when creating + the target descriptions (see nat/x86-linux-tdesc.c), the feature order + represents the bit order within the generated index number. + + i386 | x87 sse mpx avx avx512 pkru + amd64 | mpx avx avx512 pkru + i32 | avx avx512 pkru + + The features are ordered so that for each mode (i386, amd64, i32) the + generated index will form a continuous range. */ + + int idx = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) + { + if ((xcr0 & x86_linux_all_tdesc_features[i].feature) != 0) + idx |= (1 << i); + } + + return idx; +} + + +#ifdef IN_PROCESS_AGENT + +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +uint64_t +x86_linux_tdesc_idx_to_xcr0 (int idx) +{ + uint64_t xcr0 = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) + { + if ((idx & (1 << i)) != 0) + xcr0 |= x86_linux_all_tdesc_features[i].feature; + } + + return xcr0; +} + +#endif /* IN_PROCESS_AGENT */ + +#if defined __i386__ || !defined IN_PROCESS_AGENT + +/* A cache of all possible i386 target descriptions. */ + +static struct target_desc *i386_tdescs[x86_linux_i386_tdesc_count ()] = { }; + +/* See nat/x86-linux-tdesc.h. */ + +const struct target_desc * +i386_linux_read_description (uint64_t xcr0) +{ + xcr0 &= x86_linux_i386_tdesc_feature_mask (); + int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); + + gdb_assert (idx >= 0 && idx < x86_linux_i386_tdesc_count ()); + + target_desc **tdesc = &i386_tdescs[idx]; + + if (*tdesc == nullptr) + { + *tdesc = i386_create_target_description (xcr0, true, false); + + x86_linux_post_init_tdesc (*tdesc, false); + } + + return *tdesc; +} +#endif + +#ifdef __x86_64__ + +/* A cache of all possible amd64 target descriptions. */ + +static target_desc *amd64_tdescs[x86_linux_amd64_tdesc_count ()] = { }; + +/* A cache of all possible x32 target descriptions. */ + +static target_desc *x32_tdescs[x86_linux_x32_tdesc_count ()] = { }; + +/* See nat/x86-linux-tdesc.h. */ + +const struct target_desc * +amd64_linux_read_description (uint64_t xcr0, bool is_x32) +{ + if (is_x32) + xcr0 &= x86_linux_x32_tdesc_feature_mask (); + else + xcr0 &= x86_linux_amd64_tdesc_feature_mask (); + + int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); + + if (is_x32) + gdb_assert (idx >= 0 && idx < x86_linux_x32_tdesc_count ()); + else + gdb_assert (idx >= 0 && idx < x86_linux_amd64_tdesc_count ()); + + target_desc **tdesc = nullptr; + + if (is_x32) + tdesc = &x32_tdescs[idx]; + else + tdesc = &amd64_tdescs[idx]; + + if (*tdesc == nullptr) + { + *tdesc = amd64_create_target_description (xcr0, is_x32, true, true); + + x86_linux_post_init_tdesc (*tdesc, true); + } + return *tdesc; +} + +#endif diff --git a/gdb/nat/x86-linux-tdesc.h b/gdb/nat/x86-linux-tdesc.h index 3727a8bf95e..648fe0ed0b2 100644 --- a/gdb/nat/x86-linux-tdesc.h +++ b/gdb/nat/x86-linux-tdesc.h @@ -24,6 +24,8 @@ struct target_desc; +#ifndef IN_PROCESS_AGENT + /* Return the target description for Linux thread TID. When *HAVE_PTRACE_GETREGSET is TRIBOOL_UNKNOWN then the current value of @@ -57,19 +59,57 @@ x86_linux_tdesc_for_tid (int tid, enum tribool *have_ptrace_getregset, gdb::function_view<void (uint64_t)> xcr0_init_cb, const char *error_msg, uint64_t *xcr0_storage); +#endif /* !IN_PROCESS_AGENT */ + #ifdef __x86_64__ -/* Return the right amd64-linux target descriptions according to - XCR0_FEATURES_BIT and IS_X32. This is implemented separately in both - GDB and gdbserver. */ +/* Return the AMD64 target descriptions corresponding to XCR0 and IS_X32. */ -extern const target_desc *amd64_linux_read_description - (uint64_t xcr0_features_bit, bool is_x32); +extern const target_desc *amd64_linux_read_description (uint64_t xcr0, + bool is_x32); -#endif +#endif /* __x86_64__ */ + +/* Return the i386 target description corresponding to XCR0. */ -/* Return the target description according to XCR0. This is implemented - separately in both GDB and gdbserver. */ extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); +/* This function is called from amd64_linux_read_description and + i386_linux_read_description after a new target description has been + created, TDESC is the new target description, IS_64BIT will be true + when called from amd64_linux_read_description, otherwise IS_64BIT will + be false. If the *_linux_read_description functions found a cached + target description then this function will not be called. + + Both GDB and gdbserver have their own implementations of this + function. */ + +extern void x86_linux_post_init_tdesc (target_desc *tdesc, bool is_64bit); + +/* Convert an xcr0 value into an integer. The integer will be passed to + the in-process-agent where it will then be passed to + x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ + +extern int x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0); + + +#ifdef IN_PROCESS_AGENT + +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +extern uint64_t x86_linux_tdesc_idx_to_xcr0 (int idx); + +/* Within the in-process-agent we need to pre-initialise all of the target + descriptions, to do this we need to know how many target descriptions + there are for each different target type. These functions return the + target description count for the relevant target. */ + +extern int x86_linux_amd64_ipa_tdesc_count (); +extern int x86_linux_x32_ipa_tdesc_count (); +extern int x86_linux_i386_ipa_tdesc_count (); + +#endif /* IN_PROCESS_AGENT */ + #endif /* NAT_X86_LINUX_TDESC_H */ diff --git a/gdbserver/Makefile.in b/gdbserver/Makefile.in index d12f8746611..c97649378e8 100644 --- a/gdbserver/Makefile.in +++ b/gdbserver/Makefile.in @@ -517,6 +517,10 @@ gdbsupport/%-ipa.o: ../gdbsupport/%.cc $(IPAGENT_COMPILE) $< $(POSTCOMPILE) +nat/%-ipa.o: ../gdb/nat/%.c + $(IPAGENT_COMPILE) $< + $(POSTCOMPILE) + %-ipa.o: ../gdb/%.c $(IPAGENT_COMPILE) -x c++ $< $(POSTCOMPILE) diff --git a/gdbserver/configure.srv b/gdbserver/configure.srv index 7a2702d78bf..36a457f21ed 100644 --- a/gdbserver/configure.srv +++ b/gdbserver/configure.srv @@ -116,6 +116,7 @@ case "${gdbserver_host}" in srv_linux_btrace=yes ipa_obj="linux-i386-ipa.o linux-x86-tdesc-ipa.o" ipa_obj="${ipa_obj} arch/i386-ipa.o" + ipa_obj="${ipa_obj} nat/x86-linux-tdesc-ipa.o" ;; i[34567]86-*-mingw*) srv_regobj="" srv_tgtobj="x86-low.o nat/x86-dregs.o win32-low.o" @@ -380,6 +381,7 @@ case "${gdbserver_host}" in srv_linux_btrace=yes ipa_obj="linux-amd64-ipa.o linux-x86-tdesc-ipa.o" ipa_obj="${ipa_obj} arch/amd64-ipa.o" + ipa_obj="${ipa_obj} nat/x86-linux-tdesc-ipa.o" ;; x86_64-*-mingw*) srv_regobj="" srv_tgtobj="x86-low.o nat/x86-dregs.o" diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index 42de7d2da16..afa6aab5bab 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -21,7 +21,6 @@ #include "server.h" #include <sys/mman.h> #include "tracepoint.h" -#include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" #include "nat/x86-linux-tdesc.h" diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc index 246bcb9813b..f6d26f784af 100644 --- a/gdbserver/linux-i386-ipa.cc +++ b/gdbserver/linux-i386-ipa.cc @@ -21,7 +21,6 @@ #include "server.h" #include <sys/mman.h> #include "tracepoint.h" -#include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" #include "nat/x86-linux-tdesc.h" diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 6dbefdb2f26..7134d20dd7f 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -47,7 +47,6 @@ #include "nat/linux-nat.h" #include "nat/x86-linux.h" #include "nat/x86-linux-dregs.h" -#include "linux-x86-tdesc.h" #include "nat/x86-linux-tdesc.h" #ifdef __x86_64__ diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc index 87cf368a336..39d1d51b8c0 100644 --- a/gdbserver/linux-x86-tdesc.cc +++ b/gdbserver/linux-x86-tdesc.cc @@ -19,292 +19,19 @@ #include "server.h" #include "tdesc.h" -#include "linux-x86-tdesc.h" -#include "arch/i386.h" -#include "gdbsupport/x86-xstate.h" -#ifdef __x86_64__ -#include "arch/amd64.h" -#endif #include "x86-tdesc.h" #include "nat/x86-linux-tdesc.h" -/* A structure used to describe a single cpu feature that might, or might - not, be checked for when creating a target description for one of i386, - amd64, or x32. */ - -struct x86_tdesc_feature { - /* The cpu feature mask. This is a mask against an xcr0 value. */ - uint64_t feature; - - /* Is this feature checked when creating an i386 target description. */ - bool is_i386; - - /* Is this feature checked when creating an amd64 target description. */ - bool is_amd64; - - /* Is this feature checked when creating an x32 target description. */ - bool is_x32; -}; - -/* A constant table that describes all of the cpu features that are - checked when building a target description for i386, amd64, or x32. */ - -static constexpr x86_tdesc_feature x86_linux_all_tdesc_features[] = { - /* Feature, i386, amd64, x32. */ - { X86_XSTATE_PKRU, true, true, true }, - { X86_XSTATE_AVX512, true, true, true }, - { X86_XSTATE_AVX, true, true, true }, - { X86_XSTATE_MPX, true, true, false }, - { X86_XSTATE_SSE, true, false, false }, - { X86_XSTATE_X87, true, false, false } -}; - -/* Return a compile time constant which is a mask of all the cpu features - that are checked for when building an i386 target description. */ - -static constexpr uint64_t -x86_linux_i386_tdesc_feature_mask () -{ - uint64_t mask = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_i386) - mask |= entry.feature; - - return mask; -} - -/* Return a compile time constant which is a mask of all the cpu features - that are checked for when building an amd64 target description. */ - -static constexpr uint64_t -x86_linux_amd64_tdesc_feature_mask () -{ - uint64_t mask = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_amd64) - mask |= entry.feature; - - return mask; -} - -/* Return a compile time constant which is a mask of all the cpu features - that are checked for when building an x32 target description. */ - -static constexpr uint64_t -x86_linux_x32_tdesc_feature_mask () -{ - uint64_t mask = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_x32) - mask |= entry.feature; - - return mask; -} - -/* Return a compile time constant which is a count of the number of cpu - features that are checked for when building an i386 target description. */ - -static constexpr int -x86_linux_i386_tdesc_count () -{ - uint64_t count = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_i386) - ++count; - - gdb_assert (count > 0); - - return (1 << count); -} - -/* Return a compile time constant which is a count of the number of cpu - features that are checked for when building an amd64 target description. */ - -static constexpr int -x86_linux_amd64_tdesc_count () -{ - uint64_t count = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_amd64) - ++count; - - gdb_assert (count > 0); - - return (1 << count); -} - -/* Return a compile time constant which is a count of the number of cpu - features that are checked for when building an x32 target description. */ - -static constexpr int -x86_linux_x32_tdesc_count () -{ - uint64_t count = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_x32) - ++count; - - gdb_assert (count > 0); - - return (1 << count); -} - -#ifdef IN_PROCESS_AGENT - -/* See linux-x86-tdesc.h. */ - -int -x86_linux_amd64_ipa_tdesc_count () -{ - return x86_linux_amd64_tdesc_count (); -} - -/* See linux-x86-tdesc.h. */ - -int -x86_linux_x32_ipa_tdesc_count () -{ - return x86_linux_x32_tdesc_count (); -} - -/* See linux-x86-tdesc.h. */ - -int -x86_linux_i386_ipa_tdesc_count () -{ - return x86_linux_i386_tdesc_count (); -} - -#endif /* IN_PROCESS_AGENT */ - -/* Convert an xcr0 value into an integer. The integer will be passed to - the in-process-agent where it will then be passed to - x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ - -int -x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0) -{ - /* The following table shows which features are checked for when creating - the target descriptions (see nat/x86-linux-tdesc.c), the feature order - represents the bit order within the generated index number. - - i386 | x87 sse mpx avx avx512 pkru - amd64 | mpx avx avx512 pkru - i32 | avx avx512 pkru - - The features are ordered so that for each mode (i386, amd64, i32) the - generated index will form a continuous range. */ - - int idx = 0; - - for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) - { - if ((xcr0 & x86_linux_all_tdesc_features[i].feature) != 0) - idx |= (1 << i); - } - - return idx; -} - - -#ifdef IN_PROCESS_AGENT - -/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) - into an xcr0 value which can then be used to create a target - description. */ - -uint64_t -x86_linux_tdesc_idx_to_xcr0 (int idx) -{ - uint64_t xcr0 = 0; - - for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) - { - if ((idx & (1 << i)) != 0) - xcr0 |= x86_linux_all_tdesc_features[i].feature; - } - - return xcr0; -} - -#endif /* IN_PROCESS_AGENT */ - -#if defined __i386__ || !defined IN_PROCESS_AGENT - -/* A cache of all possible i386 target descriptions. */ - -static struct target_desc *i386_tdescs[x86_linux_i386_tdesc_count ()] = { }; - /* See nat/x86-linux-tdesc.h. */ -const struct target_desc * -i386_linux_read_description (uint64_t xcr0) +void +x86_linux_post_init_tdesc (target_desc *tdesc, bool is_64bit) { - xcr0 &= x86_linux_i386_tdesc_feature_mask (); - int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - - gdb_assert (idx >= 0 && idx < x86_linux_i386_tdesc_count ()); - - target_desc **tdesc = &i386_tdescs[idx]; - - if (*tdesc == nullptr) - { - *tdesc = i386_create_target_description (xcr0, true, false); - - init_target_desc (*tdesc, i386_expedite_regs); - } - - return *tdesc; -} -#endif - #ifdef __x86_64__ - -/* A cache of all possible amd64 target descriptions. */ - -static target_desc *amd64_tdescs[x86_linux_amd64_tdesc_count ()] = { }; - -/* A cache of all possible x32 target descriptions. */ - -static target_desc *x32_tdescs[x86_linux_x32_tdesc_count ()] = { }; - -/* See nat/x86-linux-tdesc.h. */ - -const struct target_desc * -amd64_linux_read_description (uint64_t xcr0, bool is_x32) -{ - if (is_x32) - xcr0 &= x86_linux_x32_tdesc_feature_mask (); + if (is_64bit) + init_target_desc (tdesc, amd64_expedite_regs); else - xcr0 &= x86_linux_amd64_tdesc_feature_mask (); - - int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - - if (is_x32) - gdb_assert (idx >= 0 && idx < x86_linux_x32_tdesc_count ()); - else - gdb_assert (idx >= 0 && idx < x86_linux_amd64_tdesc_count ()); - - target_desc **tdesc = nullptr; - - if (is_x32) - tdesc = &x32_tdescs[idx]; - else - tdesc = &amd64_tdescs[idx]; - - if (*tdesc == nullptr) - { - *tdesc = amd64_create_target_description (xcr0, is_x32, true, true); - - init_target_desc (*tdesc, amd64_expedite_regs); - } - return *tdesc; +#endif + init_target_desc (tdesc, i386_expedite_regs); } -#endif diff --git a/gdbserver/linux-x86-tdesc.h b/gdbserver/linux-x86-tdesc.h deleted file mode 100644 index 70456e4be44..00000000000 --- a/gdbserver/linux-x86-tdesc.h +++ /dev/null @@ -1,50 +0,0 @@ -/* Low level support for x86 (i386 and x86-64), shared between gdbserver - and IPA. - - Copyright (C) 2016-2024 Free Software Foundation, Inc. - - This file is part of GDB. - - This program is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published by - the Free Software Foundation; either version 3 of the License, or - (at your option) any later version. - - This program is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - GNU General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see <http://www.gnu.org/licenses/>. */ - -#ifndef GDBSERVER_LINUX_X86_TDESC_H -#define GDBSERVER_LINUX_X86_TDESC_H - -/* Convert an xcr0 value into an integer. The integer will be passed to - the in-process-agent where it will then be passed to - x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ - -extern int x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0); - -#ifdef IN_PROCESS_AGENT - -/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) - into an xcr0 value which can then be used to create a target - description. */ - -extern uint64_t x86_linux_tdesc_idx_to_xcr0 (int idx); - -/* Within the in-process-agent we need to pre-initialise all of the target - descriptions, to do this we need to know how many target descriptions - there are for each different target type. These functions return the - target description count for the relevant target. */ - -extern int x86_linux_amd64_ipa_tdesc_count (); -extern int x86_linux_x32_ipa_tdesc_count (); -extern int x86_linux_i386_ipa_tdesc_count (); - - -#endif /* IN_PROCESS_AGENT */ - -#endif /* GDBSERVER_LINUX_X86_TDESC_H */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv2 0/7] x86/Linux Target Description Changes 2024-02-01 15:28 [PATCH 0/7] x86/Linux Target Description Changes Andrew Burgess ` (6 preceding siblings ...) 2024-02-01 15:28 ` [PATCH 7/7] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess @ 2024-03-05 17:00 ` Andrew Burgess 2024-03-05 17:00 ` [PATCHv2 1/7] gdbserver: convert have_ptrace_getregset to a tribool Andrew Burgess ` (8 more replies) 7 siblings, 9 replies; 116+ messages in thread From: Andrew Burgess @ 2024-03-05 17:00 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess In v2: - Rebase to current upstream/master, no merge conflicts, - Retested. --- Andrew Burgess (7): gdbserver: convert have_ptrace_getregset to a tribool gdb/x86: move reading of cs and ds state into gdb/nat directory gdbserver/x86: move no-xml code earlier in x86_linux_read_description gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition gdb/gdbserver: share some code relating to target description creation gdbserver: update target description creation for x86/linux gdb/gdbserver: share x86/linux tdesc caching gdb/Makefile.in | 1 + gdb/amd64-linux-tdep.c | 33 +-- gdb/amd64-linux-tdep.h | 6 - gdb/configure.nat | 4 +- gdb/i386-linux-tdep.c | 32 +-- gdb/i386-linux-tdep.h | 23 -- gdb/nat/x86-linux-tdesc.c | 405 +++++++++++++++++++++++++++++++++++ gdb/nat/x86-linux-tdesc.h | 115 ++++++++++ gdb/nat/x86-linux.c | 47 ++++ gdb/nat/x86-linux.h | 48 +++++ gdb/x86-linux-nat.c | 123 ++--------- gdbserver/Makefile.in | 4 + gdbserver/configure.srv | 4 + gdbserver/linux-amd64-ipa.cc | 45 +--- gdbserver/linux-arm-low.cc | 6 +- gdbserver/linux-i386-ipa.cc | 25 +-- gdbserver/linux-low.cc | 2 +- gdbserver/linux-low.h | 2 +- gdbserver/linux-x86-low.cc | 185 +++++----------- gdbserver/linux-x86-tdesc.cc | 141 +----------- gdbserver/linux-x86-tdesc.h | 56 ----- 21 files changed, 737 insertions(+), 570 deletions(-) create mode 100644 gdb/nat/x86-linux-tdesc.c create mode 100644 gdb/nat/x86-linux-tdesc.h delete mode 100644 gdbserver/linux-x86-tdesc.h base-commit: f08311ceb1ba4e19eab7070e676416337455a074 -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv2 1/7] gdbserver: convert have_ptrace_getregset to a tribool 2024-03-05 17:00 ` [PATCHv2 0/7] x86/Linux Target Description Changes Andrew Burgess @ 2024-03-05 17:00 ` Andrew Burgess 2024-03-05 17:00 ` [PATCHv2 2/7] gdb/x86: move reading of cs and ds state into gdb/nat directory Andrew Burgess ` (7 subsequent siblings) 8 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-03-05 17:00 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess Convert the have_ptrace_getregset global within gdbserver to a tribool. This brings the flag into alignment with the corresponding flag in GDB. The gdbserver have_ptrace_getregset variable is already used as a tribool, it just doesn't have the tribool type. In a future commit I plan to share more code between GDB and gdbserver, and having this variable be the same type in both code bases will make the sharing much easier. There should be no user visible changes after this commit. --- gdbserver/linux-arm-low.cc | 6 +++--- gdbserver/linux-low.cc | 2 +- gdbserver/linux-low.h | 2 +- gdbserver/linux-x86-low.cc | 10 +++++----- 4 files changed, 10 insertions(+), 10 deletions(-) diff --git a/gdbserver/linux-arm-low.cc b/gdbserver/linux-arm-low.cc index 396ec88081b..b4f0e071c92 100644 --- a/gdbserver/linux-arm-low.cc +++ b/gdbserver/linux-arm-low.cc @@ -1007,9 +1007,9 @@ arm_target::low_arch_setup () /* Check if PTRACE_GETREGSET works. */ if (ptrace (PTRACE_GETREGSET, tid, NT_PRSTATUS, &iov) == 0) - have_ptrace_getregset = 1; + have_ptrace_getregset = TRIBOOL_TRUE; else - have_ptrace_getregset = 0; + have_ptrace_getregset = TRIBOOL_FALSE; } bool @@ -1122,7 +1122,7 @@ arm_target::get_regs_info () { const struct target_desc *tdesc = current_process ()->tdesc; - if (have_ptrace_getregset == 1 + if (have_ptrace_getregset == TRIBOOL_TRUE && (is_aarch32_linux_description (tdesc) || arm_linux_get_tdesc_fp_type (tdesc) == ARM_FP_TYPE_VFPV3)) return ®s_info_aarch32; diff --git a/gdbserver/linux-low.cc b/gdbserver/linux-low.cc index 9d5a6242803..dfa26d451b0 100644 --- a/gdbserver/linux-low.cc +++ b/gdbserver/linux-low.cc @@ -135,7 +135,7 @@ typedef struct #endif /* Does the current host support PTRACE_GETREGSET? */ -int have_ptrace_getregset = -1; +enum tribool have_ptrace_getregset = TRIBOOL_UNKNOWN; /* Return TRUE if THREAD is the leader thread of the process. */ diff --git a/gdbserver/linux-low.h b/gdbserver/linux-low.h index d34d2738238..eaf87527338 100644 --- a/gdbserver/linux-low.h +++ b/gdbserver/linux-low.h @@ -951,7 +951,7 @@ void thread_db_notice_clone (struct thread_info *parent_thr, ptid_t child_ptid); bool thread_db_thread_handle (ptid_t ptid, gdb_byte **handle, int *handle_len); -extern int have_ptrace_getregset; +extern enum tribool have_ptrace_getregset; /* Search for the value with type MATCH in the auxv vector, with entries of length WORDSIZE bytes, of process with pid PID. If found, store the diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 3af0a009052..0a3bd2c8670 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -899,7 +899,7 @@ x86_linux_read_description (void) if (ptrace (PTRACE_GETFPXREGS, tid, 0, (long) &fpxregs) < 0) { have_ptrace_getfpxregs = 0; - have_ptrace_getregset = 0; + have_ptrace_getregset = TRIBOOL_FALSE; return i386_linux_read_description (X86_XSTATE_X87); } else @@ -918,7 +918,7 @@ x86_linux_read_description (void) return tdesc_i386_linux_no_xml.get (); } - if (have_ptrace_getregset == -1) + if (have_ptrace_getregset == TRIBOOL_UNKNOWN) { uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; struct iovec iov; @@ -929,10 +929,10 @@ x86_linux_read_description (void) /* Check if PTRACE_GETREGSET works. */ if (ptrace (PTRACE_GETREGSET, tid, (unsigned int) NT_X86_XSTATE, (long) &iov) < 0) - have_ptrace_getregset = 0; + have_ptrace_getregset = TRIBOOL_FALSE; else { - have_ptrace_getregset = 1; + have_ptrace_getregset = TRIBOOL_TRUE; /* Get XCR0 from XSAVE extended state. */ xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET @@ -951,7 +951,7 @@ x86_linux_read_description (void) } /* Check the native XCR0 only if PTRACE_GETREGSET is available. */ - xcr0_features = (have_ptrace_getregset + xcr0_features = (have_ptrace_getregset == TRIBOOL_TRUE && (xcr0 & X86_XSTATE_ALL_MASK)); if (xcr0_features) -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv2 2/7] gdb/x86: move reading of cs and ds state into gdb/nat directory 2024-03-05 17:00 ` [PATCHv2 0/7] x86/Linux Target Description Changes Andrew Burgess 2024-03-05 17:00 ` [PATCHv2 1/7] gdbserver: convert have_ptrace_getregset to a tribool Andrew Burgess @ 2024-03-05 17:00 ` Andrew Burgess 2024-03-05 17:00 ` [PATCHv2 3/7] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess ` (6 subsequent siblings) 8 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-03-05 17:00 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess This patch is part of a series that has the aim of making the code that, for x86, reads the target description for a native process shared between GDB and gdbserver. Within GDB part of this process involves reading the cs and ds state from the 'struct user_regs_struct' using a ptrace call. This isn't done by gdbserver, which is part of the motivation for this whole series; the approach gdbserver takes is inferior to the approach GDB takes. This commit moves the reading of cs and ds, which is used to figure out if a thread is 32-bit or 64-bit (or in x32 mode), into the gdb/nat directory so that the code could be shared with gdbserver, but at this point I'm not actually using the code in gdbserver, that will come later. As such there should be no user visible changes after this commit, GDB continues to do things as it did before (reading cs/ds), while gdbserver continues to use its own approach (which doesn't require reading cs/ds). --- gdb/nat/x86-linux.c | 47 +++++++++++++++++++++++++++++++++++++++++++++ gdb/nat/x86-linux.h | 28 +++++++++++++++++++++++++++ gdb/x86-linux-nat.c | 42 +++++----------------------------------- 3 files changed, 80 insertions(+), 37 deletions(-) diff --git a/gdb/nat/x86-linux.c b/gdb/nat/x86-linux.c index e61f4d749ba..4242a1baafb 100644 --- a/gdb/nat/x86-linux.c +++ b/gdb/nat/x86-linux.c @@ -20,6 +20,8 @@ #include "gdbsupport/common-defs.h" #include "x86-linux.h" #include "x86-linux-dregs.h" +#include "nat/gdb_ptrace.h" +#include <sys/user.h> /* Per-thread arch-specific data we want to keep. */ @@ -80,3 +82,48 @@ x86_linux_prepare_to_resume (struct lwp_info *lwp) { x86_linux_update_debug_registers (lwp); } + +#ifdef __x86_64__ +/* Value of CS segment register: + 64bit process: 0x33 + 32bit process: 0x23 */ +#define AMD64_LINUX_USER64_CS 0x33 + +/* Value of DS segment register: + LP64 process: 0x0 + X32 process: 0x2b */ +#define AMD64_LINUX_X32_DS 0x2b +#endif + +/* See nat/x86-linux.h. */ + +x86_linux_arch_size +x86_linux_ptrace_get_arch_size (int tid) +{ +#ifdef __x86_64__ + unsigned long cs; + unsigned long ds; + + /* Get CS register. */ + errno = 0; + cs = ptrace (PTRACE_PEEKUSER, tid, + offsetof (struct user_regs_struct, cs), 0); + if (errno != 0) + perror_with_name (_("Couldn't get CS register")); + + bool is_64bit = cs == AMD64_LINUX_USER64_CS; + + /* Get DS register. */ + errno = 0; + ds = ptrace (PTRACE_PEEKUSER, tid, + offsetof (struct user_regs_struct, ds), 0); + if (errno != 0) + perror_with_name (_("Couldn't get DS register")); + + bool is_x32 = ds == AMD64_LINUX_X32_DS; + + return x86_linux_arch_size (is_64bit, is_x32); +#else + return x86_linux_arch_size (false, false); +#endif +} diff --git a/gdb/nat/x86-linux.h b/gdb/nat/x86-linux.h index 822882173f9..15153ea277e 100644 --- a/gdb/nat/x86-linux.h +++ b/gdb/nat/x86-linux.h @@ -47,4 +47,32 @@ extern void x86_linux_delete_thread (struct arch_lwp_info *arch_lwp); extern void x86_linux_prepare_to_resume (struct lwp_info *lwp); +/* Return value from x86_linux_ptrace_get_arch_size function. Indicates if + a thread is 32-bit, 64-bit, or x32. */ + +struct x86_linux_arch_size +{ + explicit x86_linux_arch_size (bool is_64bit, bool is_x32) + : m_is_64bit (is_64bit), + m_is_x32 (is_x32) + { + /* Nothing. */ + } + + bool is_64bit () const + { return m_is_64bit; } + + bool is_x32 () const + { return m_is_x32; } + +private: + bool m_is_64bit = false; + bool m_is_x32 = false; +}; + +/* Use ptrace calls to figure out if thread TID is 32-bit, 64-bit, or + 64-bit running in x32 mode. */ + +extern x86_linux_arch_size x86_linux_ptrace_get_arch_size (int tid); + #endif /* NAT_X86_LINUX_H */ diff --git a/gdb/x86-linux-nat.c b/gdb/x86-linux-nat.c index b93ffca38db..b39d05c401f 100644 --- a/gdb/x86-linux-nat.c +++ b/gdb/x86-linux-nat.c @@ -91,18 +91,6 @@ x86_linux_nat_target::post_startup_inferior (ptid_t ptid) linux_nat_target::post_startup_inferior (ptid); } -#ifdef __x86_64__ -/* Value of CS segment register: - 64bit process: 0x33 - 32bit process: 0x23 */ -#define AMD64_LINUX_USER64_CS 0x33 - -/* Value of DS segment register: - LP64 process: 0x0 - X32 process: 0x2b */ -#define AMD64_LINUX_X32_DS 0x2b -#endif - /* Get Linux/x86 target description from running target. */ const struct target_desc * @@ -122,31 +110,11 @@ x86_linux_nat_target::read_description () tid = inferior_ptid.pid (); #ifdef __x86_64__ - { - unsigned long cs; - unsigned long ds; - - /* Get CS register. */ - errno = 0; - cs = ptrace (PTRACE_PEEKUSER, tid, - offsetof (struct user_regs_struct, cs), 0); - if (errno != 0) - perror_with_name (_("Couldn't get CS register")); - - is_64bit = cs == AMD64_LINUX_USER64_CS; - - /* Get DS register. */ - errno = 0; - ds = ptrace (PTRACE_PEEKUSER, tid, - offsetof (struct user_regs_struct, ds), 0); - if (errno != 0) - perror_with_name (_("Couldn't get DS register")); - - is_x32 = ds == AMD64_LINUX_X32_DS; - - if (sizeof (void *) == 4 && is_64bit && !is_x32) - error (_("Can't debug 64-bit process with 32-bit GDB")); - } + + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); + is_64bit = arch_size.is_64bit (); + is_x32 = arch_size.is_x32 (); + #elif HAVE_PTRACE_GETFPXREGS if (have_ptrace_getfpxregs == -1) { -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv2 3/7] gdbserver/x86: move no-xml code earlier in x86_linux_read_description 2024-03-05 17:00 ` [PATCHv2 0/7] x86/Linux Target Description Changes Andrew Burgess 2024-03-05 17:00 ` [PATCHv2 1/7] gdbserver: convert have_ptrace_getregset to a tribool Andrew Burgess 2024-03-05 17:00 ` [PATCHv2 2/7] gdb/x86: move reading of cs and ds state into gdb/nat directory Andrew Burgess @ 2024-03-05 17:00 ` Andrew Burgess 2024-03-05 17:00 ` [PATCHv2 4/7] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess ` (5 subsequent siblings) 8 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-03-05 17:00 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess This commit is part of a series that aims to share more of the x86 target description reading/generation code between GDB and gdbserver. There are a huge number of similarities between the code in gdbserver's x86_linux_read_description function and GDB's x86_linux_nat_target::read_description function, and it is this similarity that I plan, in a later commit, to share between GDB and gdbserver. However, one thing that is different in x86_linux_read_description is the code inside the '!use_xml' block. This is the code that handles the case where gdbserver is not allowed to send an XML target description back to GDB. In this case gdbserver uses some predefined, fixed, target descriptions. First, it's worth noting that I suspect this code is not tested any more. I couldn't find anything in the testsuite that tries to disable XML target description support. And the idea of having a single "fixed" target description really doesn't work well when we think about all the various x86 extensions that exist. Part of me would like to rip out the no-xml support in gdbserver (at least for x86), and if a GDB connects that doesn't support XML target descriptions, gdbserver can just give an error and drop the connection. GDB has supported XML target descriptions for 16 years now, I think it would be reasonable for our shipped gdbserver to drop support for the old way of doing things. Anyway.... this commit doesn't do that. What I did notice was that, over time, the '!use_xml' block appears to have "drifted" within the x86_linux_read_description function; it's now not the first check we do. Instead we make some ptrace calls and return a target description generated based on the result of these ptrace calls. Surely it only makes sense to generate variable target descriptions if we can send these back to GDB? So in this commit I propose to move the '!use_xml' block earlier in the x86_linux_read_description function. The benefit of this is that this leaves the later half of x86_linux_read_description much more similar to the GDB function x86_linux_nat_target::read_description and sets us up for potentially sharing code between GDB and gdbserver in a later commit. --- gdbserver/linux-x86-low.cc | 27 ++++++++++++++++----------- 1 file changed, 16 insertions(+), 11 deletions(-) diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 0a3bd2c8670..ec80bfe905c 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -891,6 +891,22 @@ x86_linux_read_description (void) #endif } + /* If we are not allowed to send an XML target description then we need + to use the hard-wired target descriptions. This corresponds to GDB's + default machine for x86. + + This check needs to occur before any returns statements that might + generate some alternative target descriptions. */ + if (!use_xml) + { +#ifdef __x86_64__ + if (machine == EM_X86_64) + return tdesc_amd64_linux_no_xml.get (); + else +#endif + return tdesc_i386_linux_no_xml.get (); + } + #if !defined __x86_64__ && defined HAVE_PTRACE_GETFPXREGS if (machine == EM_386 && have_ptrace_getfpxregs == -1) { @@ -907,17 +923,6 @@ x86_linux_read_description (void) } #endif - if (!use_xml) - { - /* Don't use XML. */ -#ifdef __x86_64__ - if (machine == EM_X86_64) - return tdesc_amd64_linux_no_xml.get (); - else -#endif - return tdesc_i386_linux_no_xml.get (); - } - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) { uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv2 4/7] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition 2024-03-05 17:00 ` [PATCHv2 0/7] x86/Linux Target Description Changes Andrew Burgess ` (2 preceding siblings ...) 2024-03-05 17:00 ` [PATCHv2 3/7] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess @ 2024-03-05 17:00 ` Andrew Burgess 2024-03-05 17:00 ` [PATCHv2 5/7] gdb/gdbserver: share some code relating to target description creation Andrew Burgess ` (4 subsequent siblings) 8 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-03-05 17:00 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess Share the definition of I386_LINUX_XSAVE_XCR0_OFFSET between GDB and gdbserver. This commit is part of a series that aims to share more of the x86 target description creation code between GDB and gdbserver. The I386_LINUX_XSAVE_XCR0_OFFSET #define is used as part of the target description creation, and I noticed that this constant is defined separately for GDB and gdbserver. This commit moves the definition into gdb/nat/x86-linux.h, which allows the #define to be shared. There should be no user visible changes after this commit. --- gdb/amd64-linux-tdep.c | 1 + gdb/i386-linux-tdep.c | 1 + gdb/i386-linux-tdep.h | 20 -------------------- gdb/nat/x86-linux.h | 20 ++++++++++++++++++++ gdbserver/linux-x86-low.cc | 22 +--------------------- 5 files changed, 23 insertions(+), 41 deletions(-) diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c index 1deb13b4e9a..a512ec5dd02 100644 --- a/gdb/amd64-linux-tdep.c +++ b/gdb/amd64-linux-tdep.c @@ -42,6 +42,7 @@ #include "arch/amd64.h" #include "target-descriptions.h" #include "expop.h" +#include "nat/x86-linux.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_AMD64 "syscalls/amd64-linux.xml" diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index a2f937690c2..8dd7203b6c3 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -40,6 +40,7 @@ #include "i387-tdep.h" #include "gdbsupport/x86-xstate.h" +#include "nat/x86-linux.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_I386 "syscalls/i386-linux.xml" diff --git a/gdb/i386-linux-tdep.h b/gdb/i386-linux-tdep.h index 5891747572b..07593c6a8ec 100644 --- a/gdb/i386-linux-tdep.h +++ b/gdb/i386-linux-tdep.h @@ -58,26 +58,6 @@ extern void i386_linux_report_signal_info (struct gdbarch *gdbarch, /* Return the target description according to XCR0. */ extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); -/* Format of XSAVE extended state is: - struct - { - fxsave_bytes[0..463] - sw_usable_bytes[464..511] - xstate_hdr_bytes[512..575] - extended state regions (AVX, MPX, AVX512, PKRU, etc.) - }; - - Same memory layout will be used for the coredump NT_X86_XSTATE - representing the XSAVE extended state registers. - - The first 8 bytes of the sw_usable_bytes[464..467] is the OS enabled - extended state mask, which is the same as the extended control register - 0 (the XFEATURE_ENABLED_MASK register), XCR0. We can use this mask - together with the mask saved in the xstate_hdr_bytes to determine what - states the processor/OS supports and what state, used or initialized, - the process/thread is in. */ -#define I386_LINUX_XSAVE_XCR0_OFFSET 464 - extern int i386_linux_gregset_reg_offset[]; /* Return x86 siginfo type. */ diff --git a/gdb/nat/x86-linux.h b/gdb/nat/x86-linux.h index 15153ea277e..855a8d14f91 100644 --- a/gdb/nat/x86-linux.h +++ b/gdb/nat/x86-linux.h @@ -22,6 +22,26 @@ #include "nat/linux-nat.h" +/* Format of XSAVE extended state is: + struct + { + fxsave_bytes[0..463] + sw_usable_bytes[464..511] + xstate_hdr_bytes[512..575] + extended state regions (AVX, MPX, AVX512, PKRU, etc.) + }; + + Same memory layout will be used for the coredump NT_X86_XSTATE + representing the XSAVE extended state registers. + + The first 8 bytes of the sw_usable_bytes[464..467] is the OS enabled + extended state mask, which is the same as the extended control register + 0 (the XFEATURE_ENABLED_MASK register), XCR0. We can use this mask + together with the mask saved in the xstate_hdr_bytes to determine what + states the processor/OS supports and what state, used or initialized, + the process/thread is in. */ +#define I386_LINUX_XSAVE_XCR0_OFFSET 464 + /* Set whether our local mirror of LWP's debug registers has been changed since the values were last written to the thread. Nonzero indicates that a change has been made, zero indicates no change. */ diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index ec80bfe905c..7bade8b3cbf 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -27,6 +27,7 @@ #include "gdbsupport/x86-xstate.h" #include "nat/x86-xstate.h" #include "nat/gdb_ptrace.h" +#include "nat/x86-linux.h" #ifdef __x86_64__ #include "nat/amd64-linux-siginfo.h" @@ -832,27 +833,6 @@ x86_target::low_siginfo_fixup (siginfo_t *ptrace, gdb_byte *inf, int direction) \f static int use_xml; -/* Format of XSAVE extended state is: - struct - { - fxsave_bytes[0..463] - sw_usable_bytes[464..511] - xstate_hdr_bytes[512..575] - avx_bytes[576..831] - future_state etc - }; - - Same memory layout will be used for the coredump NT_X86_XSTATE - representing the XSAVE extended state registers. - - The first 8 bytes of the sw_usable_bytes[464..467] is the OS enabled - extended state mask, which is the same as the extended control register - 0 (the XFEATURE_ENABLED_MASK register), XCR0. We can use this mask - together with the mask saved in the xstate_hdr_bytes to determine what - states the processor/OS supports and what state, used or initialized, - the process/thread is in. */ -#define I386_LINUX_XSAVE_XCR0_OFFSET 464 - /* Does the current host support the GETFPXREGS request? The header file may or may not define it, and even if it is defined, the kernel will return EIO if it's running on a pre-SSE processor. */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv2 5/7] gdb/gdbserver: share some code relating to target description creation 2024-03-05 17:00 ` [PATCHv2 0/7] x86/Linux Target Description Changes Andrew Burgess ` (3 preceding siblings ...) 2024-03-05 17:00 ` [PATCHv2 4/7] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess @ 2024-03-05 17:00 ` Andrew Burgess 2024-03-05 17:00 ` [PATCHv2 6/7] gdbserver: update target description creation for x86/linux Andrew Burgess ` (3 subsequent siblings) 8 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-03-05 17:00 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess This commit is part of a series to share more of the x86 target description creation code between GDB and gdbserver. Unlike previous commits which were mostly refactoring, this commit is the first that makes a real change, though that change should mostly be for gdbserver; I've largely adopted the "GDB" way of doing things for gdbserver, and this fixes a real gdbserver bug. On a x86-64 Linux target, running the test: gdb.server/connect-with-no-symbol-file.exp results in two core files being created. Both of these core files are from the inferior process, created after gdbserver has detached. In this test a gdbserver process is started and then, after gdbserver has started, but before GDB attaches, we either delete the inferior executable, or change its permissions so it can't be read. Only after doing this do we attempt to connect with GDB. As GDB connects to gdbserver, gdbserver attempts to figure out the target description so that it can send the description to GDB, this involves a call to x86_linux_read_description. In x86_linux_read_description one of the first things we do is try to figure out if the process is 32-bit or 64-bit. To do this we look up the executable via the thread-id, and then attempt to read the architecture size from the executable. This isn't going to work if the executable has been deleted, or is no longer readable. And so, as we can't read the executable, we default to an i386 target and use an i386 target description. A consequence of using an i386 target description is that addresses are assumed to be 32-bits. Here's an example session that shows the problems this causes. This is run on an x86-64 machine, and the test binary (xx.x) is a standard 64-bit x86-64 binary: shell_1$ gdbserver --once localhost :54321 /tmp/xx.x shell_2$ gdb -q (gdb) set sysroot (gdb) shell chmod 000 /tmp/xx.x (gdb) target remote :54321 Remote debugging using :54321 warning: /tmp/xx.x: Permission denied. 0xf7fd3110 in ?? () (gdb) show architecture The target architecture is set to "auto" (currently "i386"). (gdb) p/x $pc $1 = 0xf7fd3110 (gdb) info proc mappings process 2412639 Mapped address spaces: Start Addr End Addr Size Offset Perms objfile 0x400000 0x401000 0x1000 0x0 r--p /tmp/xx.x 0x401000 0x402000 0x1000 0x1000 r-xp /tmp/xx.x 0x402000 0x403000 0x1000 0x2000 r--p /tmp/xx.x 0x403000 0x405000 0x2000 0x2000 rw-p /tmp/xx.x 0xf7fcb000 0xf7fcf000 0x4000 0x0 r--p [vvar] 0xf7fcf000 0xf7fd1000 0x2000 0x0 r-xp [vdso] 0xf7fd1000 0xf7fd3000 0x2000 0x0 r--p /usr/lib64/ld-2.30.so 0xf7fd3000 0xf7ff3000 0x20000 0x2000 r-xp /usr/lib64/ld-2.30.so 0xf7ff3000 0xf7ffb000 0x8000 0x22000 r--p /usr/lib64/ld-2.30.so 0xf7ffc000 0xf7ffe000 0x2000 0x2a000 rw-p /usr/lib64/ld-2.30.so 0xf7ffe000 0xf7fff000 0x1000 0x0 rw-p 0xfffda000 0xfffff000 0x25000 0x0 rw-p [stack] 0xff600000 0xff601000 0x1000 0x0 r-xp [vsyscall] (gdb) info inferiors Num Description Connection Executable * 1 process 2412639 1 (remote :54321) (gdb) shell cat /proc/2412639/maps 00400000-00401000 r--p 00000000 fd:03 45907133 /tmp/xx.x 00401000-00402000 r-xp 00001000 fd:03 45907133 /tmp/xx.x 00402000-00403000 r--p 00002000 fd:03 45907133 /tmp/xx.x 00403000-00405000 rw-p 00002000 fd:03 45907133 /tmp/xx.x 7ffff7fcb000-7ffff7fcf000 r--p 00000000 00:00 0 [vvar] 7ffff7fcf000-7ffff7fd1000 r-xp 00000000 00:00 0 [vdso] 7ffff7fd1000-7ffff7fd3000 r--p 00000000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7fd3000-7ffff7ff3000 r-xp 00002000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7ff3000-7ffff7ffb000 r--p 00022000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7ffc000-7ffff7ffe000 rw-p 0002a000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7ffe000-7ffff7fff000 rw-p 00000000 00:00 0 7ffffffda000-7ffffffff000 rw-p 00000000 00:00 0 [stack] ffffffffff600000-ffffffffff601000 r-xp 00000000 00:00 0 [vsyscall] (gdb) Notice the difference between the mappings reported via GDB and those reported directly from the kernel via /proc/PID/maps, the addresses of every mapping is clamped to 32-bits for GDB, while the kernel reports real 64-bit addresses. Notice also that the $pc value is a 32-bit value. It appears to be within one of the mappings reported by GDB, but is outside any of the mappings reported from the kernel. And this is where the problem arises. When gdbserver detaches from the inferior we pass the inferior the address from which it should resume. Due to the 32/64 bit confusion we tell the inferior to resume from the 32-bit $pc value, which is not within any valid mapping, and so, as soon as the inferior resumes, it segfaults. If we look at how GDB (not gdbserver) figures out its target description then we see an interesting difference. GDB doesn't try to read the executable. Instead GDB uses ptrace to query the thread's state, and uses this to figure out the if the thread is 32 or 64 bit. If we update gdbserver to do it the "GDB" way then the above problem is resolved, gdbserver now sees the process as 64-bit, and when we detach from the inferior we give it the correct 64-bit address, and the inferior no longer segfaults. Now, I could just update the gdbserver code, but better, I think, to share one copy of the code between GDB and gdbserver in gdb/nat/. That is what this commit does. The cores of x86_linux_read_description from gdbserver and x86_linux_nat_target::read_description from GDB are moved into a new file gdb/nat/x86-linux-tdesc.c and combined into a single function x86_linux_tdesc_for_tid which is called from each location. This new function does things the GDB way, the only changes are to allow for the sharing; we now have a callback function to call the first time that the xcr0 state is read, this allows for GDB and gdbserver to perform their own initialisation as needed, and additionally, the new function takes a pointer for where to cache the xcr0 value, this isn't needed for this commit, but will be useful in a later commit where gdbserver will want to read this cached xcr0 value. Another thing to note about this commit is how the functions i386_linux_read_description and amd64_linux_read_description are handled. For now I've left these function as implemented separately in GDB and gdbserver. I've moved the declarations of these functions into gdb/nat/x86-linux-tdesc.h, but the implementations are left as separate. A later commit in this series will make these functions shared too, but doing this is not trivial, so I've left that for a separate commit. Merging the declarations as I've done here ensures that everyone implements the function to the same API, and once these functions are shared (in a later commit) we'll want a shared declaration anyway. --- gdb/Makefile.in | 1 + gdb/amd64-linux-tdep.c | 1 + gdb/amd64-linux-tdep.h | 6 -- gdb/configure.nat | 4 +- gdb/i386-linux-tdep.c | 1 + gdb/i386-linux-tdep.h | 3 - gdb/nat/x86-linux-tdesc.c | 118 ++++++++++++++++++++++++++++ gdb/nat/x86-linux-tdesc.h | 75 ++++++++++++++++++ gdb/x86-linux-nat.c | 91 ++++------------------ gdbserver/configure.srv | 2 + gdbserver/linux-amd64-ipa.cc | 1 + gdbserver/linux-i386-ipa.cc | 1 + gdbserver/linux-x86-low.cc | 144 +++++++++++------------------------ gdbserver/linux-x86-tdesc.cc | 1 + gdbserver/linux-x86-tdesc.h | 7 -- 15 files changed, 261 insertions(+), 195 deletions(-) create mode 100644 gdb/nat/x86-linux-tdesc.c create mode 100644 gdb/nat/x86-linux-tdesc.h diff --git a/gdb/Makefile.in b/gdb/Makefile.in index 0e0f19c40c9..76bed92d84e 100644 --- a/gdb/Makefile.in +++ b/gdb/Makefile.in @@ -1586,6 +1586,7 @@ HFILES_NO_SRCDIR = \ nat/x86-gcc-cpuid.h \ nat/x86-linux.h \ nat/x86-linux-dregs.h \ + nat/x86-linux-tdesc.h \ python/py-event.h \ python/py-events.h \ python/py-stopevent.h \ diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c index a512ec5dd02..7e0900dc6f9 100644 --- a/gdb/amd64-linux-tdep.c +++ b/gdb/amd64-linux-tdep.c @@ -43,6 +43,7 @@ #include "target-descriptions.h" #include "expop.h" #include "nat/x86-linux.h" +#include "nat/x86-linux-tdesc.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_AMD64 "syscalls/amd64-linux.xml" diff --git a/gdb/amd64-linux-tdep.h b/gdb/amd64-linux-tdep.h index 2003dcda78f..0ec49e7fe03 100644 --- a/gdb/amd64-linux-tdep.h +++ b/gdb/amd64-linux-tdep.h @@ -43,12 +43,6 @@ extern struct target_desc *tdesc_x32_linux; extern struct target_desc *tdesc_x32_avx_linux; extern struct target_desc *tdesc_x32_avx_avx512_linux; -/* Return the right amd64-linux target descriptions according to - XCR0_FEATURES_BIT and IS_X32. */ - -const target_desc *amd64_linux_read_description (uint64_t xcr0_features_bit, - bool is_x32); - /* Enum that defines the syscall identifiers for amd64 linux. Used for process record/replay, these will be translated into a gdb-canonical set of syscall ids in linux-record.c. */ diff --git a/gdb/configure.nat b/gdb/configure.nat index 8b98511cef7..4bcc0696027 100644 --- a/gdb/configure.nat +++ b/gdb/configure.nat @@ -256,7 +256,7 @@ case ${gdb_host} in NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ nat/x86-xstate.o \ i386-linux-nat.o x86-linux-nat.o nat/linux-btrace.o \ - nat/x86-linux.o nat/x86-linux-dregs.o" + nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o" ;; ia64) # Host: Intel IA-64 running GNU/Linux @@ -322,7 +322,7 @@ case ${gdb_host} in NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ nat/x86-xstate.o amd64-nat.o amd64-linux-nat.o x86-linux-nat.o \ nat/linux-btrace.o \ - nat/x86-linux.o nat/x86-linux-dregs.o \ + nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o \ nat/amd64-linux-siginfo.o" ;; sparc) diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index 8dd7203b6c3..f5f7a36bf36 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -41,6 +41,7 @@ #include "i387-tdep.h" #include "gdbsupport/x86-xstate.h" #include "nat/x86-linux.h" +#include "nat/x86-linux-tdesc.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_I386 "syscalls/i386-linux.xml" diff --git a/gdb/i386-linux-tdep.h b/gdb/i386-linux-tdep.h index 07593c6a8ec..e8691cd778e 100644 --- a/gdb/i386-linux-tdep.h +++ b/gdb/i386-linux-tdep.h @@ -55,9 +55,6 @@ extern void i386_linux_report_signal_info (struct gdbarch *gdbarch, struct ui_out *uiout, enum gdb_signal siggnal); -/* Return the target description according to XCR0. */ -extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); - extern int i386_linux_gregset_reg_offset[]; /* Return x86 siginfo type. */ diff --git a/gdb/nat/x86-linux-tdesc.c b/gdb/nat/x86-linux-tdesc.c new file mode 100644 index 00000000000..78d22f55d67 --- /dev/null +++ b/gdb/nat/x86-linux-tdesc.c @@ -0,0 +1,118 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#include "gdbsupport/common-defs.h" +#include "nat/x86-linux-tdesc.h" +#ifdef __x86_64__ +#include "arch/amd64.h" +#endif +#include "arch/i386.h" + +#include "gdbsupport/common-defs.h" +#include "nat/x86-linux.h" +#include "nat/x86-linux-dregs.h" +#include "nat/gdb_ptrace.h" +#include "nat/x86-xstate.h" +#include "nat/x86-linux-tdesc.h" + +#include <cstdint> +#include <cstdlib> +#include <linux/uio.h> +#include <elf.h> +#include <sys/user.h> +#include <sys/user.h> + +/* See nat/x86-linux-tdesc.h. */ + +const target_desc * +x86_linux_tdesc_for_tid (int tid, enum tribool *have_ptrace_getregset, + gdb::function_view<void (uint64_t)> xcr0_init_cb, + const char *error_msg, uint64_t *xcr0_storage) +{ +#ifdef __x86_64__ + + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); + bool is_64bit = arch_size.is_64bit (); + bool is_x32 = arch_size.is_x32 (); + + if (sizeof (void *) == 4 && is_64bit && !is_x32) + error ("%s", error_msg); + +#elif HAVE_PTRACE_GETFPXREGS + if (have_ptrace_getfpxregs == -1) + { + elf_fpxregset_t fpxregs; + + if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) + { + have_ptrace_getfpxregs = 0; + *have_ptrace_getregset = TRIBOOL_FALSE; + return i386_linux_read_description (X86_XSTATE_X87_MASK); + } + } +#endif + + if (*have_ptrace_getregset == TRIBOOL_UNKNOWN) + { + uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; + struct iovec iov; + + iov.iov_base = xstateregs; + iov.iov_len = sizeof (xstateregs); + + /* Check if PTRACE_GETREGSET works. */ + if (ptrace (PTRACE_GETREGSET, tid, + (unsigned int) NT_X86_XSTATE, &iov) < 0) + { + *have_ptrace_getregset = TRIBOOL_FALSE; + *xcr0_storage = 0; + } + else + { + *have_ptrace_getregset = TRIBOOL_TRUE; + + /* Get XCR0 from XSAVE extended state. */ + *xcr0_storage = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET + / sizeof (uint64_t))]; + + xcr0_init_cb (*xcr0_storage); + } + } + + /* Check the native XCR0 only if PTRACE_GETREGSET is available. If + PTRACE_GETREGSET is not available then set xcr0_features_bits to + zero so that the "no-features" descriptions are returned by the + switches below. */ + uint64_t xcr0_features_bits; + if (*have_ptrace_getregset == TRIBOOL_TRUE) + xcr0_features_bits = *xcr0_storage & X86_XSTATE_ALL_MASK; + else + xcr0_features_bits = 0; + +#ifdef __x86_64__ + if (is_64bit) + { + return amd64_linux_read_description (xcr0_features_bits, is_x32); + } + else +#endif + return i386_linux_read_description (xcr0_features_bits); + + gdb_assert_not_reached ("failed to return tdesc"); +} diff --git a/gdb/nat/x86-linux-tdesc.h b/gdb/nat/x86-linux-tdesc.h new file mode 100644 index 00000000000..3727a8bf95e --- /dev/null +++ b/gdb/nat/x86-linux-tdesc.h @@ -0,0 +1,75 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#ifndef NAT_X86_LINUX_TDESC_H +#define NAT_X86_LINUX_TDESC_H + +#include "gdbsupport/function-view.h" + +struct target_desc; + +/* Return the target description for Linux thread TID. + + When *HAVE_PTRACE_GETREGSET is TRIBOOL_UNKNOWN then the current value of + xcr0 is read using ptrace calls and stored into *XCR0_STORAGE. Then + XCR0_INIT_CB is called with the value of *XCR0_STORAGE and + *HAVE_PTRACE_GETREGSET is set to TRIBOOL_TRUE. + + If the attempt to read xcr0 using ptrace fails then *XCR0_STORAGE is set + to zero and *HAVE_PTRACE_GETREGSET is set to TRIBOOL_FALSE. + + The storage pointed to by XCR0_STORAGE must exist until the program + terminates, this storage is used to cache the xcr0 value. As such + XCR0_INIT_CB will only be called once if xcr0 is successfully read using + ptrace, or not at all if the ptrace call fails. + + This function returns a target description based on the extracted xcr0 + value along with other characteristics of the thread identified by TID. + + This function can return nullptr if we encounter a machine configuration + for which a target_desc cannot be created. Ideally this would not be + the case, we should be able to create a target description for every + possible machine configuration. See amd64_linux_read_description and + i386_linux_read_description for cases when nullptr might be + returned. + + ERROR_MSG is using in an error() call if we try to create a target + description for a 64-bit process but this is a 32-bit build of GDB. */ + +extern const target_desc * +x86_linux_tdesc_for_tid (int tid, enum tribool *have_ptrace_getregset, + gdb::function_view<void (uint64_t)> xcr0_init_cb, + const char *error_msg, uint64_t *xcr0_storage); + +#ifdef __x86_64__ + +/* Return the right amd64-linux target descriptions according to + XCR0_FEATURES_BIT and IS_X32. This is implemented separately in both + GDB and gdbserver. */ + +extern const target_desc *amd64_linux_read_description + (uint64_t xcr0_features_bit, bool is_x32); + +#endif + +/* Return the target description according to XCR0. This is implemented + separately in both GDB and gdbserver. */ +extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); + +#endif /* NAT_X86_LINUX_TDESC_H */ diff --git a/gdb/x86-linux-nat.c b/gdb/x86-linux-nat.c index b39d05c401f..872e27e739f 100644 --- a/gdb/x86-linux-nat.c +++ b/gdb/x86-linux-nat.c @@ -42,6 +42,7 @@ #include "nat/x86-linux.h" #include "nat/x86-linux-dregs.h" #include "nat/linux-ptrace.h" +#include "nat/x86-linux-tdesc.h" /* linux_nat_target::low_new_fork implementation. */ @@ -96,90 +97,26 @@ x86_linux_nat_target::post_startup_inferior (ptid_t ptid) const struct target_desc * x86_linux_nat_target::read_description () { - int tid; - int is_64bit = 0; -#ifdef __x86_64__ - int is_x32; -#endif - static uint64_t xcr0; - uint64_t xcr0_features_bits; + static uint64_t xcr0_storage; if (inferior_ptid == null_ptid) return this->beneath ()->read_description (); - tid = inferior_ptid.pid (); - -#ifdef __x86_64__ - - x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); - is_64bit = arch_size.is_64bit (); - is_x32 = arch_size.is_x32 (); - -#elif HAVE_PTRACE_GETFPXREGS - if (have_ptrace_getfpxregs == -1) - { - elf_fpxregset_t fpxregs; - - if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) - { - have_ptrace_getfpxregs = 0; - have_ptrace_getregset = TRIBOOL_FALSE; - return i386_linux_read_description (X86_XSTATE_X87_MASK); - } - } -#endif - - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) - { - uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; - struct iovec iov; - - iov.iov_base = xstateregs; - iov.iov_len = sizeof (xstateregs); - - /* Check if PTRACE_GETREGSET works. */ - if (ptrace (PTRACE_GETREGSET, tid, - (unsigned int) NT_X86_XSTATE, &iov) < 0) - have_ptrace_getregset = TRIBOOL_FALSE; - else - { - have_ptrace_getregset = TRIBOOL_TRUE; - - /* Get XCR0 from XSAVE extended state. */ - xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET - / sizeof (uint64_t))]; - - m_xsave_layout = x86_fetch_xsave_layout (xcr0, x86_xsave_length ()); - } - } - - /* Check the native XCR0 only if PTRACE_GETREGSET is available. If - PTRACE_GETREGSET is not available then set xcr0_features_bits to - zero so that the "no-features" descriptions are returned by the - switches below. */ - if (have_ptrace_getregset == TRIBOOL_TRUE) - xcr0_features_bits = xcr0 & X86_XSTATE_ALL_MASK; - else - xcr0_features_bits = 0; - - if (is_64bit) - { -#ifdef __x86_64__ - return amd64_linux_read_description (xcr0_features_bits, is_x32); -#endif - } - else - { - const struct target_desc * tdesc - = i386_linux_read_description (xcr0_features_bits); + int tid = inferior_ptid.pid (); - if (tdesc == NULL) - tdesc = i386_linux_read_description (X86_XSTATE_SSE_MASK); + const char *error_msg + = _("Can't debug 64-bit process with 32-bit GDB"); - return tdesc; - } + /* Callback that is triggered the first time x86_linux_tdesc_for_tid + reads the xcr0 register. Setup other bits of state */ + auto cb = [&] (uint64_t xcr0) + { + this->m_xsave_layout + = x86_fetch_xsave_layout (xcr0, x86_xsave_length ()); + }; - gdb_assert_not_reached ("failed to return tdesc"); + return x86_linux_tdesc_for_tid (tid, &have_ptrace_getregset, cb, + error_msg, &xcr0_storage); } \f diff --git a/gdbserver/configure.srv b/gdbserver/configure.srv index 9e861a75088..7a2702d78bf 100644 --- a/gdbserver/configure.srv +++ b/gdbserver/configure.srv @@ -109,6 +109,7 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/linux-btrace.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" + srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" srv_linux_usrregs=yes srv_linux_regsets=yes srv_linux_thread_db=yes @@ -371,6 +372,7 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/linux-btrace.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" + srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" srv_tgtobj="${srv_tgtobj} nat/amd64-linux-siginfo.o" srv_linux_usrregs=yes # This is for i386 progs. srv_linux_regsets=yes diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index 54e4c9812bb..f97b0d6a1d9 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -23,6 +23,7 @@ #include "tracepoint.h" #include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" +#include "nat/x86-linux-tdesc.h" /* Defined in auto-generated file amd64-linux.c. */ void init_registers_amd64_linux (void); diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc index 2e4646f8c03..459b8055b5c 100644 --- a/gdbserver/linux-i386-ipa.cc +++ b/gdbserver/linux-i386-ipa.cc @@ -23,6 +23,7 @@ #include "tracepoint.h" #include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" +#include "nat/x86-linux-tdesc.h" /* GDB register numbers. */ diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 7bade8b3cbf..9bf369f8a34 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -48,6 +48,7 @@ #include "nat/x86-linux.h" #include "nat/x86-linux-dregs.h" #include "linux-x86-tdesc.h" +#include "nat/x86-linux-tdesc.h" #ifdef __x86_64__ static target_desc_up tdesc_amd64_linux_no_xml; @@ -844,32 +845,20 @@ int have_ptrace_getfpxregs = #endif ; +/* Cached xcr0 value. This is initialised the first time + x86_linux_read_description is called. */ + +static uint64_t xcr0_storage; + /* Get Linux/x86 target description from running target. */ static const struct target_desc * x86_linux_read_description (void) { - unsigned int machine; - int is_elf64; - int xcr0_features; - int tid; - static uint64_t xcr0; - static int xsave_len; - struct regset_info *regset; - - tid = lwpid_of (current_thread); + int tid = lwpid_of (current_thread); - is_elf64 = linux_pid_exe_is_elf_64_file (tid, &machine); - - if (sizeof (void *) == 4) - { - if (is_elf64 > 0) - error (_("Can't debug 64-bit process with 32-bit GDBserver")); -#ifndef __x86_64__ - else if (machine == EM_X86_64) - error (_("Can't debug x86-64 process with 32-bit GDBserver")); -#endif - } + const char *error_msg + = _("Can't debug 64-bit process with 32-bit GDBserver"); /* If we are not allowed to send an XML target description then we need to use the hard-wired target descriptions. This corresponds to GDB's @@ -879,99 +868,54 @@ x86_linux_read_description (void) generate some alternative target descriptions. */ if (!use_xml) { + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); + bool is_64bit = arch_size.is_64bit (); + bool is_x32 = arch_size.is_x32 (); + + if (sizeof (void *) == 4 && is_64bit && !is_x32) + error ("%s", error_msg); + #ifdef __x86_64__ - if (machine == EM_X86_64) + if (is_64bit && !is_x32) return tdesc_amd64_linux_no_xml.get (); else #endif return tdesc_i386_linux_no_xml.get (); } -#if !defined __x86_64__ && defined HAVE_PTRACE_GETFPXREGS - if (machine == EM_386 && have_ptrace_getfpxregs == -1) - { - elf_fpxregset_t fpxregs; - - if (ptrace (PTRACE_GETFPXREGS, tid, 0, (long) &fpxregs) < 0) - { - have_ptrace_getfpxregs = 0; - have_ptrace_getregset = TRIBOOL_FALSE; - return i386_linux_read_description (X86_XSTATE_X87); - } - else - have_ptrace_getfpxregs = 1; - } -#endif - - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) - { - uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; - struct iovec iov; - - iov.iov_base = xstateregs; - iov.iov_len = sizeof (xstateregs); - - /* Check if PTRACE_GETREGSET works. */ - if (ptrace (PTRACE_GETREGSET, tid, - (unsigned int) NT_X86_XSTATE, (long) &iov) < 0) - have_ptrace_getregset = TRIBOOL_FALSE; - else - { - have_ptrace_getregset = TRIBOOL_TRUE; - - /* Get XCR0 from XSAVE extended state. */ - xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET - / sizeof (uint64_t))]; - - xsave_len = x86_xsave_length (); - - /* Use PTRACE_GETREGSET if it is available. */ - for (regset = x86_regsets; - regset->fill_function != NULL; regset++) - if (regset->get_request == PTRACE_GETREGSET) - regset->size = xsave_len; - else if (regset->type != GENERAL_REGS) - regset->size = 0; - } - } - - /* Check the native XCR0 only if PTRACE_GETREGSET is available. */ - xcr0_features = (have_ptrace_getregset == TRIBOOL_TRUE - && (xcr0 & X86_XSTATE_ALL_MASK)); - - if (xcr0_features) - i387_set_xsave_mask (xcr0, xsave_len); + /* Callback that is triggered the first time x86_linux_tdesc_for_tid + reads the xcr0 register. Setup other bits of state */ + auto cb = [] (uint64_t xcr0) + { + i387_set_xsave_mask (xcr0, x86_xsave_length ()); + }; - if (machine == EM_X86_64) - { -#ifdef __x86_64__ - const target_desc *tdesc = NULL; + /* If have_ptrace_getregset is changed to true by calling + x86_linux_tdesc_for_tid then we will perform some additional + initialisation. */ + bool have_ptrace_getregset_is_unknown + = have_ptrace_getregset == TRIBOOL_UNKNOWN; - if (xcr0_features) - { - tdesc = amd64_linux_read_description (xcr0 & X86_XSTATE_ALL_MASK, - !is_elf64); - } + const target_desc *tdesc + = x86_linux_tdesc_for_tid (tid, &have_ptrace_getregset, cb, error_msg, + &xcr0_storage); - if (tdesc == NULL) - tdesc = amd64_linux_read_description (X86_XSTATE_SSE_MASK, !is_elf64); - return tdesc; -#endif - } - else + if (have_ptrace_getregset_is_unknown + && have_ptrace_getregset == TRIBOOL_TRUE) { - const target_desc *tdesc = NULL; - - if (xcr0_features) - tdesc = i386_linux_read_description (xcr0 & X86_XSTATE_ALL_MASK); - - if (tdesc == NULL) - tdesc = i386_linux_read_description (X86_XSTATE_SSE); - - return tdesc; + int xsave_len = x86_xsave_length (); + + /* Use PTRACE_GETREGSET if it is available. */ + for (regset_info *regset = x86_regsets; + regset->fill_function != nullptr; + regset++) + if (regset->get_request == PTRACE_GETREGSET) + regset->size = xsave_len; + else if (regset->type != GENERAL_REGS) + regset->size = 0; } - gdb_assert_not_reached ("failed to return tdesc"); + return tdesc; } /* Update all the target description of all processes; a new GDB diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc index 626207fc477..9fd64d8574b 100644 --- a/gdbserver/linux-x86-tdesc.cc +++ b/gdbserver/linux-x86-tdesc.cc @@ -26,6 +26,7 @@ #include "arch/amd64.h" #endif #include "x86-tdesc.h" +#include "nat/x86-linux-tdesc.h" /* Return the right x86_linux_tdesc index for a given XCR0. Return X86_TDESC_LAST if can't find a match. */ diff --git a/gdbserver/linux-x86-tdesc.h b/gdbserver/linux-x86-tdesc.h index f9561b129ae..576aaf5e165 100644 --- a/gdbserver/linux-x86-tdesc.h +++ b/gdbserver/linux-x86-tdesc.h @@ -46,11 +46,4 @@ int amd64_get_ipa_tdesc_idx (const struct target_desc *tdesc); const struct target_desc *i386_get_ipa_tdesc (int idx); -#ifdef __x86_64__ -const struct target_desc *amd64_linux_read_description (uint64_t xcr0, - bool is_x32); -#endif - -const struct target_desc *i386_linux_read_description (uint64_t xcr0); - #endif /* GDBSERVER_LINUX_X86_TDESC_H */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv2 6/7] gdbserver: update target description creation for x86/linux 2024-03-05 17:00 ` [PATCHv2 0/7] x86/Linux Target Description Changes Andrew Burgess ` (4 preceding siblings ...) 2024-03-05 17:00 ` [PATCHv2 5/7] gdb/gdbserver: share some code relating to target description creation Andrew Burgess @ 2024-03-05 17:00 ` Andrew Burgess 2024-03-19 16:01 ` John Baldwin 2024-03-05 17:00 ` [PATCHv2 7/7] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess ` (2 subsequent siblings) 8 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-03-05 17:00 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess This commit is part of a series which aims to share more of the target description creation between GDB and gdbserver for x86/Linux. After some refactoring, the previous commit actually started to share some code, we added the shared x86_linux_tdesc_for_tid function into nat/x86-linux-tdesc.c. However, this function still relies on amd64_linux_read_description and i386_linux_read_description which are implemented separately for both gdbserver and GDB. Given that at their core, all these functions to is: 1. take an xcr0 value as input, 2. mask out some feature bits, 3. look for a cached pre-generated target description and return it if found, 4. if no cached target description is found then call either amd64_create_target_description or i386_create_target_description to create a new target description, which is then added to the cache. Return the newly created target description. The inner functions amd64_create_target_description and i386_create_target_description are already shared between GDB and gdbserver (in the gdb/arch/ directory), so the only thing that the *_read_description functions really do is add the caching layer, and it feels like this really could be shared. However, we have a small problem. On the GDB side we create target descriptions using a different set of cpu features than on the gdbserver side! This means that for the exact same target, we might get a different target description when using native GDB vs using gdbserver. This surely feels like a mistake, I would expect to get the same target description on each. The table below shows the number of possible different target descriptions that we can create on the GDB side vs on the gdbserver side for each target type: | GDB | gdbserver ------|-----|---------- i386 | 64 | 7 amd64 | 32 | 7 x32 | 16 | 7 So in theory, all I want to do is move the GDB version of *_read_description into the nat/ directory and have gdbserver use that, then both GDB and gdbserver would be able to create any of the possible target descriptions. Unfortunately it's a little more complex than that due to the in process agent (IPA). When the IPA is in use, gdbserver sends a target description index to the IPA, and the IPA uses this to find the correct target description to use. ** START OF AN ASIDE ** Back in the day I suspect this approach made perfect sense. However since this commit: commit a8806230241d201f808d856eaae4d44088117b0c Date: Thu Dec 7 17:07:01 2017 +0000 Initialize target description early in IPA I think passing the index is now more trouble than its worth. We used to pass the index, and then use that index to lookup which target description to instantiate and use. However, the above commit fixed an issue where we can't call malloc() within (certain parts of) the IPA (apparently), so instead we now pre-compute _every_ possible target description within the IPA. The index is now only used to lookup which of the (many) pre-computed target descriptions to use. It would (I think) have been easier all around if the IPA just self-inspected, figured out its own xcr0 value, and used that to create the one target description that is required. So long as the xcr0 to target description code is shared (at compile time) with gdbserver, then we can be sure that the IPA will derive the same target description as gdbserver, and we would avoid all this index passing business, which has made this commit so very, very painful. ** END OF AN ASIDE ** Currently then for x86/linux, gdbserver sends a number between 0 and 7 to the IPA, and the IPA uses this to create a target description. However, I am proposing that gdbserver should now create one of (up to) 64 different target descriptions for i386, so this 0 to 7 index isn't going to be good enough any more (amd64 and x32 have slightly fewer possible target descriptions, but still more than 8, so the problem is the same). For a while I wondered if I was going to have to try and find some backward compatible solution for this mess. But after seeing how lightly the IPA is actually documented, I wonder if it is not the case that there is a tight coupling between a version of gdbserver and a version of the IPA? At least I'm hoping so. In this commit I have thrown out the old IPA target description index numbering scheme, and switched to a completely new numbering scheme. Instead of the index that is passed being arbitrary, the index is instead calculated from the set of cpu features that are present on the target. Within the IPA we can then reverse this logic to recreate the xcr0 value based on the index, and from the xcr0 value we can create the correct target description. With the gdbserver to IPA numbering scheme issue resolved I have then update the gdbserver versions of amd64_linux_read_description and i386_linux_read_description so that they create target descriptions using the same set of cpu features as GDB itself. After this gdbserver should now always come up with the same target description as GDB does on any x86/Linux target. This commit does not introduce any new code sharing between GDB and gdbserver as previous commits in this series does. Instead this commit is all about bringing GDB and gdbserver into alignment functionally so that the next commit can merge the GDB and gdbserver versions of these functions. --- gdbserver/linux-amd64-ipa.cc | 43 +---- gdbserver/linux-i386-ipa.cc | 23 +-- gdbserver/linux-x86-low.cc | 15 +- gdbserver/linux-x86-tdesc.cc | 315 +++++++++++++++++++++++++---------- gdbserver/linux-x86-tdesc.h | 49 +++--- 5 files changed, 277 insertions(+), 168 deletions(-) diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index f97b0d6a1d9..42de7d2da16 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -169,47 +169,21 @@ supply_static_tracepoint_registers (struct regcache *regcache, #endif /* HAVE_UST */ -#if !defined __ILP32__ -/* Map the tdesc index to xcr0 mask. */ -static uint64_t idx2mask[X86_TDESC_LAST] = { - X86_XSTATE_X87_MASK, - X86_XSTATE_SSE_MASK, - X86_XSTATE_AVX_MASK, - X86_XSTATE_MPX_MASK, - X86_XSTATE_AVX_MPX_MASK, - X86_XSTATE_AVX_AVX512_MASK, - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, -}; -#endif - /* Return target_desc to use for IPA, given the tdesc index passed by gdbserver. */ const struct target_desc * get_ipa_tdesc (int idx) { - if (idx >= X86_TDESC_LAST) - { - internal_error ("unknown ipa tdesc index: %d", idx); - } + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); #if defined __ILP32__ - switch (idx) - { - case X86_TDESC_SSE: - return amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); - case X86_TDESC_AVX: - return amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); - case X86_TDESC_AVX_AVX512: - return amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, true); - default: - break; - } + bool is_x32 = true; #else - return amd64_linux_read_description (idx2mask[idx], false); + bool is_x32 = false; #endif - internal_error ("unknown ipa tdesc index: %d", idx); + return amd64_linux_read_description (xcr0, is_x32); } /* Allocate buffer for the jump pads. The branch instruction has a @@ -277,11 +251,10 @@ void initialize_low_tracepoint (void) { #if defined __ILP32__ - amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); - amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); - amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, true); + for (auto i = 0; i < x86_linux_x32_ipa_tdesc_count (); i++) + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), true); #else - for (auto i = 0; i < X86_TDESC_LAST; i++) - amd64_linux_read_description (idx2mask[i], false); + for (auto i = 0; i < x86_linux_amd64_ipa_tdesc_count (); i++) + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), false); #endif } diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc index 459b8055b5c..246bcb9813b 100644 --- a/gdbserver/linux-i386-ipa.cc +++ b/gdbserver/linux-i386-ipa.cc @@ -246,28 +246,15 @@ initialize_fast_tracepoint_trampoline_buffer (void) } } -/* Map the tdesc index to xcr0 mask. */ -static uint64_t idx2mask[X86_TDESC_LAST] = { - X86_XSTATE_X87_MASK, - X86_XSTATE_SSE_MASK, - X86_XSTATE_AVX_MASK, - X86_XSTATE_MPX_MASK, - X86_XSTATE_AVX_MPX_MASK, - X86_XSTATE_AVX_AVX512_MASK, - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, -}; - /* Return target_desc to use for IPA, given the tdesc index passed by gdbserver. */ const struct target_desc * get_ipa_tdesc (int idx) { - if (idx >= X86_TDESC_LAST) - { - internal_error ("unknown ipa tdesc index: %d", idx); - } - return i386_linux_read_description (idx2mask[idx]); + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); + + return i386_linux_read_description (xcr0); } /* Allocate buffer for the jump pads. On i386, we can reach an arbitrary @@ -289,6 +276,6 @@ void initialize_low_tracepoint (void) { initialize_fast_tracepoint_trampoline_buffer (); - for (auto i = 0; i < X86_TDESC_LAST; i++) - i386_linux_read_description (idx2mask[i]); + for (auto i = 0; i < x86_linux_i386_ipa_tdesc_count (); i++) + i386_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i)); } diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 9bf369f8a34..6dbefdb2f26 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -2892,14 +2892,17 @@ x86_target::get_ipa_tdesc_idx () struct regcache *regcache = get_thread_regcache (current_thread, 0); const struct target_desc *tdesc = regcache->tdesc; + if (!use_xml) + { + if (tdesc == tdesc_i386_linux_no_xml.get () #ifdef __x86_64__ - return amd64_get_ipa_tdesc_idx (tdesc); -#endif - - if (tdesc == tdesc_i386_linux_no_xml.get ()) - return X86_TDESC_SSE; + || tdesc == tdesc_amd64_linux_no_xml.get () +#endif /* __x86_64__ */ + ) + return x86_linux_xcr0_to_tdesc_idx (X86_XSTATE_SSE_MASK); + } - return i386_get_ipa_tdesc_idx (tdesc); + return x86_linux_xcr0_to_tdesc_idx (xcr0_storage); } /* The linux target ops object. */ diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc index 9fd64d8574b..87cf368a336 100644 --- a/gdbserver/linux-x86-tdesc.cc +++ b/gdbserver/linux-x86-tdesc.cc @@ -28,96 +28,277 @@ #include "x86-tdesc.h" #include "nat/x86-linux-tdesc.h" -/* Return the right x86_linux_tdesc index for a given XCR0. Return - X86_TDESC_LAST if can't find a match. */ +/* A structure used to describe a single cpu feature that might, or might + not, be checked for when creating a target description for one of i386, + amd64, or x32. */ -static enum x86_linux_tdesc -xcr0_to_tdesc_idx (uint64_t xcr0, bool is_x32) +struct x86_tdesc_feature { + /* The cpu feature mask. This is a mask against an xcr0 value. */ + uint64_t feature; + + /* Is this feature checked when creating an i386 target description. */ + bool is_i386; + + /* Is this feature checked when creating an amd64 target description. */ + bool is_amd64; + + /* Is this feature checked when creating an x32 target description. */ + bool is_x32; +}; + +/* A constant table that describes all of the cpu features that are + checked when building a target description for i386, amd64, or x32. */ + +static constexpr x86_tdesc_feature x86_linux_all_tdesc_features[] = { + /* Feature, i386, amd64, x32. */ + { X86_XSTATE_PKRU, true, true, true }, + { X86_XSTATE_AVX512, true, true, true }, + { X86_XSTATE_AVX, true, true, true }, + { X86_XSTATE_MPX, true, true, false }, + { X86_XSTATE_SSE, true, false, false }, + { X86_XSTATE_X87, true, false, false } +}; + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an i386 target description. */ + +static constexpr uint64_t +x86_linux_i386_tdesc_feature_mask () { - if (xcr0 & X86_XSTATE_PKRU) - { - if (is_x32) - { - /* No x32 MPX and PKU, fall back to avx_avx512. */ - return X86_TDESC_AVX_AVX512; - } - else - return X86_TDESC_AVX_MPX_AVX512_PKU; - } - else if (xcr0 & X86_XSTATE_AVX512) - return X86_TDESC_AVX_AVX512; - else if ((xcr0 & X86_XSTATE_AVX_MPX_MASK) == X86_XSTATE_AVX_MPX_MASK) + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_i386) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an amd64 target description. */ + +static constexpr uint64_t +x86_linux_amd64_tdesc_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_amd64) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an x32 target description. */ + +static constexpr uint64_t +x86_linux_x32_tdesc_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_x32) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an i386 target description. */ + +static constexpr int +x86_linux_i386_tdesc_count () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_i386) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an amd64 target description. */ + +static constexpr int +x86_linux_amd64_tdesc_count () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_amd64) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an x32 target description. */ + +static constexpr int +x86_linux_x32_tdesc_count () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_x32) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +#ifdef IN_PROCESS_AGENT + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_amd64_ipa_tdesc_count () +{ + return x86_linux_amd64_tdesc_count (); +} + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_x32_ipa_tdesc_count () +{ + return x86_linux_x32_tdesc_count (); +} + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_i386_ipa_tdesc_count () +{ + return x86_linux_i386_tdesc_count (); +} + +#endif /* IN_PROCESS_AGENT */ + +/* Convert an xcr0 value into an integer. The integer will be passed to + the in-process-agent where it will then be passed to + x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ + +int +x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0) +{ + /* The following table shows which features are checked for when creating + the target descriptions (see nat/x86-linux-tdesc.c), the feature order + represents the bit order within the generated index number. + + i386 | x87 sse mpx avx avx512 pkru + amd64 | mpx avx avx512 pkru + i32 | avx avx512 pkru + + The features are ordered so that for each mode (i386, amd64, i32) the + generated index will form a continuous range. */ + + int idx = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) { - if (is_x32) /* No MPX on x32. */ - return X86_TDESC_AVX; - else - return X86_TDESC_AVX_MPX; + if ((xcr0 & x86_linux_all_tdesc_features[i].feature) != 0) + idx |= (1 << i); } - else if (xcr0 & X86_XSTATE_MPX) + + return idx; +} + + +#ifdef IN_PROCESS_AGENT + +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +uint64_t +x86_linux_tdesc_idx_to_xcr0 (int idx) +{ + uint64_t xcr0 = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) { - if (is_x32) /* No MPX on x32. */ - return X86_TDESC_AVX; - else - return X86_TDESC_MPX; + if ((idx & (1 << i)) != 0) + xcr0 |= x86_linux_all_tdesc_features[i].feature; } - else if (xcr0 & X86_XSTATE_AVX) - return X86_TDESC_AVX; - else if (xcr0 & X86_XSTATE_SSE) - return X86_TDESC_SSE; - else if (xcr0 & X86_XSTATE_X87) - return X86_TDESC_MMX; - else - return X86_TDESC_LAST; + + return xcr0; } +#endif /* IN_PROCESS_AGENT */ + #if defined __i386__ || !defined IN_PROCESS_AGENT -static struct target_desc *i386_tdescs[X86_TDESC_LAST] = { }; +/* A cache of all possible i386 target descriptions. */ -/* Return the target description according to XCR0. */ +static struct target_desc *i386_tdescs[x86_linux_i386_tdesc_count ()] = { }; + +/* See nat/x86-linux-tdesc.h. */ const struct target_desc * i386_linux_read_description (uint64_t xcr0) { - enum x86_linux_tdesc idx = xcr0_to_tdesc_idx (xcr0, false); + xcr0 &= x86_linux_i386_tdesc_feature_mask (); + int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - if (idx == X86_TDESC_LAST) - return NULL; + gdb_assert (idx >= 0 && idx < x86_linux_i386_tdesc_count ()); - struct target_desc **tdesc = &i386_tdescs[idx]; + target_desc **tdesc = &i386_tdescs[idx]; - if (*tdesc == NULL) + if (*tdesc == nullptr) { *tdesc = i386_create_target_description (xcr0, true, false); init_target_desc (*tdesc, i386_expedite_regs); } - return *tdesc;; + return *tdesc; } #endif #ifdef __x86_64__ -static target_desc *amd64_tdescs[X86_TDESC_LAST] = { }; -static target_desc *x32_tdescs[X86_TDESC_LAST] = { }; +/* A cache of all possible amd64 target descriptions. */ + +static target_desc *amd64_tdescs[x86_linux_amd64_tdesc_count ()] = { }; + +/* A cache of all possible x32 target descriptions. */ + +static target_desc *x32_tdescs[x86_linux_x32_tdesc_count ()] = { }; + +/* See nat/x86-linux-tdesc.h. */ const struct target_desc * amd64_linux_read_description (uint64_t xcr0, bool is_x32) { - enum x86_linux_tdesc idx = xcr0_to_tdesc_idx (xcr0, is_x32); + if (is_x32) + xcr0 &= x86_linux_x32_tdesc_feature_mask (); + else + xcr0 &= x86_linux_amd64_tdesc_feature_mask (); + + int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - if (idx == X86_TDESC_LAST) - return NULL; + if (is_x32) + gdb_assert (idx >= 0 && idx < x86_linux_x32_tdesc_count ()); + else + gdb_assert (idx >= 0 && idx < x86_linux_amd64_tdesc_count ()); - struct target_desc **tdesc = NULL; + target_desc **tdesc = nullptr; if (is_x32) tdesc = &x32_tdescs[idx]; else tdesc = &amd64_tdescs[idx]; - if (*tdesc == NULL) + if (*tdesc == nullptr) { *tdesc = amd64_create_target_description (xcr0, is_x32, true, true); @@ -127,39 +308,3 @@ amd64_linux_read_description (uint64_t xcr0, bool is_x32) } #endif - -#ifndef IN_PROCESS_AGENT - -int -i386_get_ipa_tdesc_idx (const struct target_desc *tdesc) -{ - for (int i = 0; i < X86_TDESC_LAST; i++) - { - if (tdesc == i386_tdescs[i]) - return i; - } - - /* If none tdesc is found, return the one with minimum features. */ - return X86_TDESC_MMX; -} - -#if defined __x86_64__ -int -amd64_get_ipa_tdesc_idx (const struct target_desc *tdesc) -{ - for (int i = 0; i < X86_TDESC_LAST; i++) - { - if (tdesc == amd64_tdescs[i]) - return i; - } - for (int i = 0; i < X86_TDESC_LAST; i++) - { - if (tdesc == x32_tdescs[i]) - return i; - } - - return X86_TDESC_SSE; -} - -#endif -#endif diff --git a/gdbserver/linux-x86-tdesc.h b/gdbserver/linux-x86-tdesc.h index 576aaf5e165..70456e4be44 100644 --- a/gdbserver/linux-x86-tdesc.h +++ b/gdbserver/linux-x86-tdesc.h @@ -21,29 +21,30 @@ #ifndef GDBSERVER_LINUX_X86_TDESC_H #define GDBSERVER_LINUX_X86_TDESC_H -/* Note: since IPA obviously knows what ABI it's running on (i386 vs x86_64 - vs x32), it's sufficient to pass only the register set here. This, - together with the ABI known at IPA compile time, maps to a tdesc. */ - -enum x86_linux_tdesc { - X86_TDESC_MMX = 0, - X86_TDESC_SSE = 1, - X86_TDESC_AVX = 2, - X86_TDESC_MPX = 3, - X86_TDESC_AVX_MPX = 4, - X86_TDESC_AVX_AVX512 = 5, - X86_TDESC_AVX_MPX_AVX512_PKU = 6, - X86_TDESC_LAST = 7, -}; - -#if defined __i386__ || !defined IN_PROCESS_AGENT -int i386_get_ipa_tdesc_idx (const struct target_desc *tdesc); -#endif - -#if defined __x86_64__ && !defined IN_PROCESS_AGENT -int amd64_get_ipa_tdesc_idx (const struct target_desc *tdesc); -#endif - -const struct target_desc *i386_get_ipa_tdesc (int idx); +/* Convert an xcr0 value into an integer. The integer will be passed to + the in-process-agent where it will then be passed to + x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ + +extern int x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0); + +#ifdef IN_PROCESS_AGENT + +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +extern uint64_t x86_linux_tdesc_idx_to_xcr0 (int idx); + +/* Within the in-process-agent we need to pre-initialise all of the target + descriptions, to do this we need to know how many target descriptions + there are for each different target type. These functions return the + target description count for the relevant target. */ + +extern int x86_linux_amd64_ipa_tdesc_count (); +extern int x86_linux_x32_ipa_tdesc_count (); +extern int x86_linux_i386_ipa_tdesc_count (); + + +#endif /* IN_PROCESS_AGENT */ #endif /* GDBSERVER_LINUX_X86_TDESC_H */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv2 6/7] gdbserver: update target description creation for x86/linux 2024-03-05 17:00 ` [PATCHv2 6/7] gdbserver: update target description creation for x86/linux Andrew Burgess @ 2024-03-19 16:01 ` John Baldwin 2024-03-19 18:34 ` Andrew Burgess 0 siblings, 1 reply; 116+ messages in thread From: John Baldwin @ 2024-03-19 16:01 UTC (permalink / raw) To: Andrew Burgess, gdb-patches On 3/5/24 9:00 AM, Andrew Burgess wrote: > This commit is part of a series which aims to share more of the target > description creation between GDB and gdbserver for x86/Linux. > > After some refactoring, the previous commit actually started to share > some code, we added the shared x86_linux_tdesc_for_tid function into > nat/x86-linux-tdesc.c. However, this function still relies on > amd64_linux_read_description and i386_linux_read_description which are > implemented separately for both gdbserver and GDB. Given that at > their core, all these functions to is: > > 1. take an xcr0 value as input, > 2. mask out some feature bits, > 3. look for a cached pre-generated target description and return it > if found, > 4. if no cached target description is found then call either > amd64_create_target_description or > i386_create_target_description to create a new target > description, which is then added to the cache. Return the newly > created target description. > > The inner functions amd64_create_target_description and > i386_create_target_description are already shared between GDB and > gdbserver (in the gdb/arch/ directory), so the only thing that > the *_read_description functions really do is add the caching layer, > and it feels like this really could be shared. > > However, we have a small problem. > > On the GDB side we create target descriptions using a different set of > cpu features than on the gdbserver side! This means that for the > exact same target, we might get a different target description when > using native GDB vs using gdbserver. This surely feels like a > mistake, I would expect to get the same target description on each. > > The table below shows the number of possible different target > descriptions that we can create on the GDB side vs on the gdbserver > side for each target type: > > | GDB | gdbserver > ------|-----|---------- > i386 | 64 | 7 > amd64 | 32 | 7 > x32 | 16 | 7 > > So in theory, all I want to do is move the GDB version > of *_read_description into the nat/ directory and have gdbserver use > that, then both GDB and gdbserver would be able to create any of the > possible target descriptions. > > Unfortunately it's a little more complex than that due to the in > process agent (IPA). > > When the IPA is in use, gdbserver sends a target description index to > the IPA, and the IPA uses this to find the correct target description > to use. > > ** START OF AN ASIDE ** > > Back in the day I suspect this approach made perfect sense. However > since this commit: > > commit a8806230241d201f808d856eaae4d44088117b0c > Date: Thu Dec 7 17:07:01 2017 +0000 > > Initialize target description early in IPA > > I think passing the index is now more trouble than its worth. > > We used to pass the index, and then use that index to lookup which > target description to instantiate and use. However, the above commit > fixed an issue where we can't call malloc() within (certain parts of) > the IPA (apparently), so instead we now pre-compute _every_ possible > target description within the IPA. The index is now only used to > lookup which of the (many) pre-computed target descriptions to use. > > It would (I think) have been easier all around if the IPA just > self-inspected, figured out its own xcr0 value, and used that to > create the one target description that is required. So long as the > xcr0 to target description code is shared (at compile time) with > gdbserver, then we can be sure that the IPA will derive the same > target description as gdbserver, and we would avoid all this index > passing business, which has made this commit so very, very painful. > > ** END OF AN ASIDE ** > > Currently then for x86/linux, gdbserver sends a number between 0 and 7 > to the IPA, and the IPA uses this to create a target description. > > However, I am proposing that gdbserver should now create one of (up > to) 64 different target descriptions for i386, so this 0 to 7 index > isn't going to be good enough any more (amd64 and x32 have slightly > fewer possible target descriptions, but still more than 8, so the > problem is the same). > > For a while I wondered if I was going to have to try and find some > backward compatible solution for this mess. But after seeing how > lightly the IPA is actually documented, I wonder if it is not the case > that there is a tight coupling between a version of gdbserver and a > version of the IPA? At least I'm hoping so. > > In this commit I have thrown out the old IPA target description index > numbering scheme, and switched to a completely new numbering scheme. > Instead of the index that is passed being arbitrary, the index is > instead calculated from the set of cpu features that are present on > the target. Within the IPA we can then reverse this logic to recreate > the xcr0 value based on the index, and from the xcr0 value we can > create the correct target description. > > With the gdbserver to IPA numbering scheme issue resolved I have then > update the gdbserver versions of amd64_linux_read_description and > i386_linux_read_description so that they create target descriptions > using the same set of cpu features as GDB itself. > > After this gdbserver should now always come up with the same target > description as GDB does on any x86/Linux target. > > This commit does not introduce any new code sharing between GDB and > gdbserver as previous commits in this series does. Instead this > commit is all about bringing GDB and gdbserver into alignment > functionally so that the next commit can merge the GDB and gdbserver > versions of these functions. It does seem like the IPA case should just compute the tdesc it needs if I understand you correctly. BTW, on other arches like aarch64 and I believe risc-v, a flat array is not used for the tdesc cache. Now the cache is a std::unordered_map<> and there is a 'struct foo_features' that has an operator== and std::hash<> specialization. It really would be nice if x86 was the same, but that would require abandoning a notion of a portable "index" shared between gdbserver and the IPA. -- John Baldwin ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv2 6/7] gdbserver: update target description creation for x86/linux 2024-03-19 16:01 ` John Baldwin @ 2024-03-19 18:34 ` Andrew Burgess 2024-03-21 17:28 ` John Baldwin 0 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-03-19 18:34 UTC (permalink / raw) To: John Baldwin, gdb-patches John Baldwin <jhb@FreeBSD.org> writes: > On 3/5/24 9:00 AM, Andrew Burgess wrote: >> This commit is part of a series which aims to share more of the target >> description creation between GDB and gdbserver for x86/Linux. >> >> After some refactoring, the previous commit actually started to share >> some code, we added the shared x86_linux_tdesc_for_tid function into >> nat/x86-linux-tdesc.c. However, this function still relies on >> amd64_linux_read_description and i386_linux_read_description which are >> implemented separately for both gdbserver and GDB. Given that at >> their core, all these functions to is: >> >> 1. take an xcr0 value as input, >> 2. mask out some feature bits, >> 3. look for a cached pre-generated target description and return it >> if found, >> 4. if no cached target description is found then call either >> amd64_create_target_description or >> i386_create_target_description to create a new target >> description, which is then added to the cache. Return the newly >> created target description. >> >> The inner functions amd64_create_target_description and >> i386_create_target_description are already shared between GDB and >> gdbserver (in the gdb/arch/ directory), so the only thing that >> the *_read_description functions really do is add the caching layer, >> and it feels like this really could be shared. >> >> However, we have a small problem. >> >> On the GDB side we create target descriptions using a different set of >> cpu features than on the gdbserver side! This means that for the >> exact same target, we might get a different target description when >> using native GDB vs using gdbserver. This surely feels like a >> mistake, I would expect to get the same target description on each. >> >> The table below shows the number of possible different target >> descriptions that we can create on the GDB side vs on the gdbserver >> side for each target type: >> >> | GDB | gdbserver >> ------|-----|---------- >> i386 | 64 | 7 >> amd64 | 32 | 7 >> x32 | 16 | 7 >> >> So in theory, all I want to do is move the GDB version >> of *_read_description into the nat/ directory and have gdbserver use >> that, then both GDB and gdbserver would be able to create any of the >> possible target descriptions. >> >> Unfortunately it's a little more complex than that due to the in >> process agent (IPA). >> >> When the IPA is in use, gdbserver sends a target description index to >> the IPA, and the IPA uses this to find the correct target description >> to use. >> >> ** START OF AN ASIDE ** >> >> Back in the day I suspect this approach made perfect sense. However >> since this commit: >> >> commit a8806230241d201f808d856eaae4d44088117b0c >> Date: Thu Dec 7 17:07:01 2017 +0000 >> >> Initialize target description early in IPA >> >> I think passing the index is now more trouble than its worth. >> >> We used to pass the index, and then use that index to lookup which >> target description to instantiate and use. However, the above commit >> fixed an issue where we can't call malloc() within (certain parts of) >> the IPA (apparently), so instead we now pre-compute _every_ possible >> target description within the IPA. The index is now only used to >> lookup which of the (many) pre-computed target descriptions to use. >> >> It would (I think) have been easier all around if the IPA just >> self-inspected, figured out its own xcr0 value, and used that to >> create the one target description that is required. So long as the >> xcr0 to target description code is shared (at compile time) with >> gdbserver, then we can be sure that the IPA will derive the same >> target description as gdbserver, and we would avoid all this index >> passing business, which has made this commit so very, very painful. >> >> ** END OF AN ASIDE ** >> >> Currently then for x86/linux, gdbserver sends a number between 0 and 7 >> to the IPA, and the IPA uses this to create a target description. >> >> However, I am proposing that gdbserver should now create one of (up >> to) 64 different target descriptions for i386, so this 0 to 7 index >> isn't going to be good enough any more (amd64 and x32 have slightly >> fewer possible target descriptions, but still more than 8, so the >> problem is the same). >> >> For a while I wondered if I was going to have to try and find some >> backward compatible solution for this mess. But after seeing how >> lightly the IPA is actually documented, I wonder if it is not the case >> that there is a tight coupling between a version of gdbserver and a >> version of the IPA? At least I'm hoping so. >> >> In this commit I have thrown out the old IPA target description index >> numbering scheme, and switched to a completely new numbering scheme. >> Instead of the index that is passed being arbitrary, the index is >> instead calculated from the set of cpu features that are present on >> the target. Within the IPA we can then reverse this logic to recreate >> the xcr0 value based on the index, and from the xcr0 value we can >> create the correct target description. >> >> With the gdbserver to IPA numbering scheme issue resolved I have then >> update the gdbserver versions of amd64_linux_read_description and >> i386_linux_read_description so that they create target descriptions >> using the same set of cpu features as GDB itself. >> >> After this gdbserver should now always come up with the same target >> description as GDB does on any x86/Linux target. >> >> This commit does not introduce any new code sharing between GDB and >> gdbserver as previous commits in this series does. Instead this >> commit is all about bringing GDB and gdbserver into alignment >> functionally so that the next commit can merge the GDB and gdbserver >> versions of these functions. > > It does seem like the IPA case should just compute the tdesc it needs > if I understand you correctly. Yes, totally. And it's not like it would even add much code to the IPA, I believe it already pulls in all of the target description creation code, the only thing that is "saved" is the code that probes the hardware to figure out which description we should use. This whole area seems like it has not aged well :-/ > > BTW, on other arches like aarch64 and I believe risc-v, a flat array > is not used for the tdesc cache. Now the cache is a std::unordered_map<> > and there is a 'struct foo_features' that has an operator== and > std::hash<> specialization. It really would be nice if x86 was the same, > but that would require abandoning a notion of a portable "index" shared > between gdbserver and the IPA. Indeed. I did the worked on the feature/hash/lookup for RISC-V so I had this in mind when looking at this code. Originally I had planned to switch to a map style cache, but when I found the IPA/index passing code I lost my nerve and tried to find a solution that was inline with what we had currently. I don't really know about the IPA enough to make claims about what is/isn't a reasonable change in this area. Thanks, Andrew ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv2 6/7] gdbserver: update target description creation for x86/linux 2024-03-19 18:34 ` Andrew Burgess @ 2024-03-21 17:28 ` John Baldwin 2024-03-26 10:01 ` Luis Machado 0 siblings, 1 reply; 116+ messages in thread From: John Baldwin @ 2024-03-21 17:28 UTC (permalink / raw) To: Andrew Burgess, gdb-patches On 3/19/24 11:34 AM, Andrew Burgess wrote: > John Baldwin <jhb@FreeBSD.org> writes: > >> On 3/5/24 9:00 AM, Andrew Burgess wrote: >>> This commit is part of a series which aims to share more of the target >>> description creation between GDB and gdbserver for x86/Linux. >>> >>> After some refactoring, the previous commit actually started to share >>> some code, we added the shared x86_linux_tdesc_for_tid function into >>> nat/x86-linux-tdesc.c. However, this function still relies on >>> amd64_linux_read_description and i386_linux_read_description which are >>> implemented separately for both gdbserver and GDB. Given that at >>> their core, all these functions to is: >>> >>> 1. take an xcr0 value as input, >>> 2. mask out some feature bits, >>> 3. look for a cached pre-generated target description and return it >>> if found, >>> 4. if no cached target description is found then call either >>> amd64_create_target_description or >>> i386_create_target_description to create a new target >>> description, which is then added to the cache. Return the newly >>> created target description. >>> >>> The inner functions amd64_create_target_description and >>> i386_create_target_description are already shared between GDB and >>> gdbserver (in the gdb/arch/ directory), so the only thing that >>> the *_read_description functions really do is add the caching layer, >>> and it feels like this really could be shared. >>> >>> However, we have a small problem. >>> >>> On the GDB side we create target descriptions using a different set of >>> cpu features than on the gdbserver side! This means that for the >>> exact same target, we might get a different target description when >>> using native GDB vs using gdbserver. This surely feels like a >>> mistake, I would expect to get the same target description on each. >>> >>> The table below shows the number of possible different target >>> descriptions that we can create on the GDB side vs on the gdbserver >>> side for each target type: >>> >>> | GDB | gdbserver >>> ------|-----|---------- >>> i386 | 64 | 7 >>> amd64 | 32 | 7 >>> x32 | 16 | 7 >>> >>> So in theory, all I want to do is move the GDB version >>> of *_read_description into the nat/ directory and have gdbserver use >>> that, then both GDB and gdbserver would be able to create any of the >>> possible target descriptions. >>> >>> Unfortunately it's a little more complex than that due to the in >>> process agent (IPA). >>> >>> When the IPA is in use, gdbserver sends a target description index to >>> the IPA, and the IPA uses this to find the correct target description >>> to use. >>> >>> ** START OF AN ASIDE ** >>> >>> Back in the day I suspect this approach made perfect sense. However >>> since this commit: >>> >>> commit a8806230241d201f808d856eaae4d44088117b0c >>> Date: Thu Dec 7 17:07:01 2017 +0000 >>> >>> Initialize target description early in IPA >>> >>> I think passing the index is now more trouble than its worth. >>> >>> We used to pass the index, and then use that index to lookup which >>> target description to instantiate and use. However, the above commit >>> fixed an issue where we can't call malloc() within (certain parts of) >>> the IPA (apparently), so instead we now pre-compute _every_ possible >>> target description within the IPA. The index is now only used to >>> lookup which of the (many) pre-computed target descriptions to use. >>> >>> It would (I think) have been easier all around if the IPA just >>> self-inspected, figured out its own xcr0 value, and used that to >>> create the one target description that is required. So long as the >>> xcr0 to target description code is shared (at compile time) with >>> gdbserver, then we can be sure that the IPA will derive the same >>> target description as gdbserver, and we would avoid all this index >>> passing business, which has made this commit so very, very painful. >>> >>> ** END OF AN ASIDE ** >>> >>> Currently then for x86/linux, gdbserver sends a number between 0 and 7 >>> to the IPA, and the IPA uses this to create a target description. >>> >>> However, I am proposing that gdbserver should now create one of (up >>> to) 64 different target descriptions for i386, so this 0 to 7 index >>> isn't going to be good enough any more (amd64 and x32 have slightly >>> fewer possible target descriptions, but still more than 8, so the >>> problem is the same). >>> >>> For a while I wondered if I was going to have to try and find some >>> backward compatible solution for this mess. But after seeing how >>> lightly the IPA is actually documented, I wonder if it is not the case >>> that there is a tight coupling between a version of gdbserver and a >>> version of the IPA? At least I'm hoping so. >>> >>> In this commit I have thrown out the old IPA target description index >>> numbering scheme, and switched to a completely new numbering scheme. >>> Instead of the index that is passed being arbitrary, the index is >>> instead calculated from the set of cpu features that are present on >>> the target. Within the IPA we can then reverse this logic to recreate >>> the xcr0 value based on the index, and from the xcr0 value we can >>> create the correct target description. >>> >>> With the gdbserver to IPA numbering scheme issue resolved I have then >>> update the gdbserver versions of amd64_linux_read_description and >>> i386_linux_read_description so that they create target descriptions >>> using the same set of cpu features as GDB itself. >>> >>> After this gdbserver should now always come up with the same target >>> description as GDB does on any x86/Linux target. >>> >>> This commit does not introduce any new code sharing between GDB and >>> gdbserver as previous commits in this series does. Instead this >>> commit is all about bringing GDB and gdbserver into alignment >>> functionally so that the next commit can merge the GDB and gdbserver >>> versions of these functions. >> >> It does seem like the IPA case should just compute the tdesc it needs >> if I understand you correctly. > > Yes, totally. And it's not like it would even add much code to the IPA, > I believe it already pulls in all of the target description creation > code, the only thing that is "saved" is the code that probes the > hardware to figure out which description we should use. > > This whole area seems like it has not aged well :-/ > >> >> BTW, on other arches like aarch64 and I believe risc-v, a flat array >> is not used for the tdesc cache. Now the cache is a std::unordered_map<> >> and there is a 'struct foo_features' that has an operator== and >> std::hash<> specialization. It really would be nice if x86 was the same, >> but that would require abandoning a notion of a portable "index" shared >> between gdbserver and the IPA. > > Indeed. I did the worked on the feature/hash/lookup for RISC-V so I had > this in mind when looking at this code. Originally I had planned to > switch to a map style cache, but when I found the IPA/index passing code > I lost my nerve and tried to find a solution that was inline with what > we had currently. > > I don't really know about the IPA enough to make claims about what > is/isn't a reasonable change in this area. I have zero knowledge about the IPA myself. OTOH, if no one else steps up with a strong opinion here, I would suggest that you go ahead and fix the IPA as you described so it is more self-contained and then we can use a saner approach for dealing with tdesc caching for x86. -- John Baldwin ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv2 6/7] gdbserver: update target description creation for x86/linux 2024-03-21 17:28 ` John Baldwin @ 2024-03-26 10:01 ` Luis Machado 2024-03-26 15:31 ` Tom Tromey 0 siblings, 1 reply; 116+ messages in thread From: Luis Machado @ 2024-03-26 10:01 UTC (permalink / raw) To: John Baldwin, Andrew Burgess, gdb-patches On 3/21/24 17:28, John Baldwin wrote: > On 3/19/24 11:34 AM, Andrew Burgess wrote: >> John Baldwin <jhb@FreeBSD.org> writes: >> >>> On 3/5/24 9:00 AM, Andrew Burgess wrote: >>>> This commit is part of a series which aims to share more of the target >>>> description creation between GDB and gdbserver for x86/Linux. >>>> >>>> After some refactoring, the previous commit actually started to share >>>> some code, we added the shared x86_linux_tdesc_for_tid function into >>>> nat/x86-linux-tdesc.c. However, this function still relies on >>>> amd64_linux_read_description and i386_linux_read_description which are >>>> implemented separately for both gdbserver and GDB. Given that at >>>> their core, all these functions to is: >>>> >>>> 1. take an xcr0 value as input, >>>> 2. mask out some feature bits, >>>> 3. look for a cached pre-generated target description and return it >>>> if found, >>>> 4. if no cached target description is found then call either >>>> amd64_create_target_description or >>>> i386_create_target_description to create a new target >>>> description, which is then added to the cache. Return the newly >>>> created target description. >>>> >>>> The inner functions amd64_create_target_description and >>>> i386_create_target_description are already shared between GDB and >>>> gdbserver (in the gdb/arch/ directory), so the only thing that >>>> the *_read_description functions really do is add the caching layer, >>>> and it feels like this really could be shared. >>>> >>>> However, we have a small problem. >>>> >>>> On the GDB side we create target descriptions using a different set of >>>> cpu features than on the gdbserver side! This means that for the >>>> exact same target, we might get a different target description when >>>> using native GDB vs using gdbserver. This surely feels like a >>>> mistake, I would expect to get the same target description on each. >>>> >>>> The table below shows the number of possible different target >>>> descriptions that we can create on the GDB side vs on the gdbserver >>>> side for each target type: >>>> >>>> | GDB | gdbserver >>>> ------|-----|---------- >>>> i386 | 64 | 7 >>>> amd64 | 32 | 7 >>>> x32 | 16 | 7 >>>> >>>> So in theory, all I want to do is move the GDB version >>>> of *_read_description into the nat/ directory and have gdbserver use >>>> that, then both GDB and gdbserver would be able to create any of the >>>> possible target descriptions. >>>> >>>> Unfortunately it's a little more complex than that due to the in >>>> process agent (IPA). >>>> >>>> When the IPA is in use, gdbserver sends a target description index to >>>> the IPA, and the IPA uses this to find the correct target description >>>> to use. >>>> >>>> ** START OF AN ASIDE ** >>>> >>>> Back in the day I suspect this approach made perfect sense. However >>>> since this commit: >>>> >>>> commit a8806230241d201f808d856eaae4d44088117b0c >>>> Date: Thu Dec 7 17:07:01 2017 +0000 >>>> >>>> Initialize target description early in IPA >>>> >>>> I think passing the index is now more trouble than its worth. >>>> >>>> We used to pass the index, and then use that index to lookup which >>>> target description to instantiate and use. However, the above commit >>>> fixed an issue where we can't call malloc() within (certain parts of) >>>> the IPA (apparently), so instead we now pre-compute _every_ possible >>>> target description within the IPA. The index is now only used to >>>> lookup which of the (many) pre-computed target descriptions to use. >>>> >>>> It would (I think) have been easier all around if the IPA just >>>> self-inspected, figured out its own xcr0 value, and used that to >>>> create the one target description that is required. So long as the >>>> xcr0 to target description code is shared (at compile time) with >>>> gdbserver, then we can be sure that the IPA will derive the same >>>> target description as gdbserver, and we would avoid all this index >>>> passing business, which has made this commit so very, very painful. >>>> >>>> ** END OF AN ASIDE ** >>>> >>>> Currently then for x86/linux, gdbserver sends a number between 0 and 7 >>>> to the IPA, and the IPA uses this to create a target description. >>>> >>>> However, I am proposing that gdbserver should now create one of (up >>>> to) 64 different target descriptions for i386, so this 0 to 7 index >>>> isn't going to be good enough any more (amd64 and x32 have slightly >>>> fewer possible target descriptions, but still more than 8, so the >>>> problem is the same). >>>> >>>> For a while I wondered if I was going to have to try and find some >>>> backward compatible solution for this mess. But after seeing how >>>> lightly the IPA is actually documented, I wonder if it is not the case >>>> that there is a tight coupling between a version of gdbserver and a >>>> version of the IPA? At least I'm hoping so. >>>> >>>> In this commit I have thrown out the old IPA target description index >>>> numbering scheme, and switched to a completely new numbering scheme. >>>> Instead of the index that is passed being arbitrary, the index is >>>> instead calculated from the set of cpu features that are present on >>>> the target. Within the IPA we can then reverse this logic to recreate >>>> the xcr0 value based on the index, and from the xcr0 value we can >>>> create the correct target description. >>>> >>>> With the gdbserver to IPA numbering scheme issue resolved I have then >>>> update the gdbserver versions of amd64_linux_read_description and >>>> i386_linux_read_description so that they create target descriptions >>>> using the same set of cpu features as GDB itself. >>>> >>>> After this gdbserver should now always come up with the same target >>>> description as GDB does on any x86/Linux target. >>>> >>>> This commit does not introduce any new code sharing between GDB and >>>> gdbserver as previous commits in this series does. Instead this >>>> commit is all about bringing GDB and gdbserver into alignment >>>> functionally so that the next commit can merge the GDB and gdbserver >>>> versions of these functions. >>> >>> It does seem like the IPA case should just compute the tdesc it needs >>> if I understand you correctly. >> >> Yes, totally. And it's not like it would even add much code to the IPA, >> I believe it already pulls in all of the target description creation >> code, the only thing that is "saved" is the code that probes the >> hardware to figure out which description we should use. >> >> This whole area seems like it has not aged well :-/ >> >>> >>> BTW, on other arches like aarch64 and I believe risc-v, a flat array >>> is not used for the tdesc cache. Now the cache is a std::unordered_map<> >>> and there is a 'struct foo_features' that has an operator== and >>> std::hash<> specialization. It really would be nice if x86 was the same, >>> but that would require abandoning a notion of a portable "index" shared >>> between gdbserver and the IPA. >> >> Indeed. I did the worked on the feature/hash/lookup for RISC-V so I had >> this in mind when looking at this code. Originally I had planned to >> switch to a map style cache, but when I found the IPA/index passing code >> I lost my nerve and tried to find a solution that was inline with what >> we had currently. >> >> I don't really know about the IPA enough to make claims about what >> is/isn't a reasonable change in this area. > > I have zero knowledge about the IPA myself. OTOH, if no one else steps > up with a strong opinion here, I would suggest that you go ahead and > fix the IPA as you described so it is more self-contained and then we can > use a saner approach for dealing with tdesc caching for x86. > Should we start thinking of deprecating the IPA if no one shows any interest in keeping it alive? It is there to provide support for (fast/static) tracepoints, and I haven't seen anyone using it in the open for years now. The cost of maintenance may not justify keeping it alive. ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv2 6/7] gdbserver: update target description creation for x86/linux 2024-03-26 10:01 ` Luis Machado @ 2024-03-26 15:31 ` Tom Tromey 0 siblings, 0 replies; 116+ messages in thread From: Tom Tromey @ 2024-03-26 15:31 UTC (permalink / raw) To: Luis Machado; +Cc: John Baldwin, Andrew Burgess, gdb-patches >>>>> "Luis" == Luis Machado <luis.machado@arm.com> writes: Luis> Should we start thinking of deprecating the IPA if no one shows any interest in Luis> keeping it alive? It is there to provide support for (fast/static) tracepoints, Luis> and I haven't seen anyone using it in the open for years now. Luis> The cost of maintenance may not justify keeping it alive. This makes sense to me as well. thanks, Tom ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv2 7/7] gdb/gdbserver: share x86/linux tdesc caching 2024-03-05 17:00 ` [PATCHv2 0/7] x86/Linux Target Description Changes Andrew Burgess ` (5 preceding siblings ...) 2024-03-05 17:00 ` [PATCHv2 6/7] gdbserver: update target description creation for x86/linux Andrew Burgess @ 2024-03-05 17:00 ` Andrew Burgess 2024-03-19 16:05 ` [PATCHv2 0/7] x86/Linux Target Description Changes John Baldwin 2024-03-23 16:35 ` [PATCHv3 0/8] " Andrew Burgess 8 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-03-05 17:00 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess This commit builds on the previous series of commits to share the target description caching code between GDB and gdbserver for x86/Linux targets. The objective of this commit is to move the four functions (2 each of) i386_linux_read_description and amd64_linux_read_description into gdb/nat/x86-linux-tdesc.c and combine them so we have just a single copy of each. Then both GDB and gdbserver will link against these shared functions. It is worth reading the description of the previous commit to see why this merging is not as simple as it seems: on the gdbserver side we actually have two users of these functions, gdbserver itself, and the in process agent (IPA). However, the previous commit streamlined the gdbserver code, and so now it is simple to move the two functions along with all their support functions from the gdbserver directory into the gdb/nat/ directory, and then GDB is fine to call these functions. One small curiosity with this patch is the function x86_linux_post_init_tdesc. On the gdbserver side the two functions amd64_linux_read_description and i386_linux_read_description have some functionality that is not present on the GDB side, that is some additional configuration that is performed as each target description is created to setup the expedited registers. To support this I've added the function x86_linux_post_init_tdesc. This function is called from the two *_linux_read_description functions, but is implemented separately for GDB and gdbserver. This does mean adding back some non-shared code when this whole series has been about sharing code, but now the only non-shared bit is the single line that is actually different between GDB and gdbserver, all the rest, which is identical, is now shared. I did need to add a new rule to the gdbserver Makefile, this is to allow the nat/x86-linux-tdesc.c file to be compiled for the IPA. --- gdb/amd64-linux-tdep.c | 31 ---- gdb/i386-linux-tdep.c | 30 +--- gdb/nat/x86-linux-tdesc.c | 287 +++++++++++++++++++++++++++++++++++ gdb/nat/x86-linux-tdesc.h | 56 ++++++- gdbserver/Makefile.in | 4 + gdbserver/configure.srv | 2 + gdbserver/linux-amd64-ipa.cc | 1 - gdbserver/linux-i386-ipa.cc | 1 - gdbserver/linux-x86-low.cc | 1 - gdbserver/linux-x86-tdesc.cc | 285 +--------------------------------- gdbserver/linux-x86-tdesc.h | 50 ------ 11 files changed, 355 insertions(+), 393 deletions(-) delete mode 100644 gdbserver/linux-x86-tdesc.h diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c index 7e0900dc6f9..e149034de21 100644 --- a/gdb/amd64-linux-tdep.c +++ b/gdb/amd64-linux-tdep.c @@ -1579,37 +1579,6 @@ amd64_linux_record_signal (struct gdbarch *gdbarch, return 0; } -const target_desc * -amd64_linux_read_description (uint64_t xcr0_features_bit, bool is_x32) -{ - static target_desc *amd64_linux_tdescs \ - [2/*AVX*/][2/*MPX*/][2/*AVX512*/][2/*PKRU*/] = {}; - static target_desc *x32_linux_tdescs \ - [2/*AVX*/][2/*AVX512*/][2/*PKRU*/] = {}; - - target_desc **tdesc; - - if (is_x32) - { - tdesc = &x32_linux_tdescs[(xcr0_features_bit & X86_XSTATE_AVX) ? 1 : 0 ] - [(xcr0_features_bit & X86_XSTATE_AVX512) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_PKRU) ? 1 : 0]; - } - else - { - tdesc = &amd64_linux_tdescs[(xcr0_features_bit & X86_XSTATE_AVX) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_MPX) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_AVX512) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_PKRU) ? 1 : 0]; - } - - if (*tdesc == NULL) - *tdesc = amd64_create_target_description (xcr0_features_bit, is_x32, - true, true); - - return *tdesc; -} - /* Get Linux/x86 target description from core dump. */ static const struct target_desc * diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index f5f7a36bf36..ed1442236e8 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -681,29 +681,12 @@ i386_linux_core_read_x86_xsave_layout (struct gdbarch *gdbarch, layout) != 0; } -/* See i386-linux-tdep.h. */ +/* See nat/x86-linux-tdesc.h. */ -const struct target_desc * -i386_linux_read_description (uint64_t xcr0) +void +x86_linux_post_init_tdesc (target_desc *tdesc, bool is_64bit) { - if (xcr0 == 0) - return NULL; - - static struct target_desc *i386_linux_tdescs \ - [2/*X87*/][2/*SSE*/][2/*AVX*/][2/*MPX*/][2/*AVX512*/][2/*PKRU*/] = {}; - struct target_desc **tdesc; - - tdesc = &i386_linux_tdescs[(xcr0 & X86_XSTATE_X87) ? 1 : 0] - [(xcr0 & X86_XSTATE_SSE) ? 1 : 0] - [(xcr0 & X86_XSTATE_AVX) ? 1 : 0] - [(xcr0 & X86_XSTATE_MPX) ? 1 : 0] - [(xcr0 & X86_XSTATE_AVX512) ? 1 : 0] - [(xcr0 & X86_XSTATE_PKRU) ? 1 : 0]; - - if (*tdesc == NULL) - *tdesc = i386_create_target_description (xcr0, true, false); - - return *tdesc; + /* Nothing. */ } /* Get Linux/x86 target description from core dump. */ @@ -716,7 +699,10 @@ i386_linux_core_read_description (struct gdbarch *gdbarch, /* Linux/i386. */ x86_xsave_layout layout; uint64_t xcr0 = i386_linux_core_read_xsave_info (abfd, layout); - const struct target_desc *tdesc = i386_linux_read_description (xcr0); + + const struct target_desc *tdesc; + if (xcr0 != 0) + tdesc = i386_linux_read_description (xcr0); if (tdesc != NULL) return tdesc; diff --git a/gdb/nat/x86-linux-tdesc.c b/gdb/nat/x86-linux-tdesc.c index 78d22f55d67..297d1873680 100644 --- a/gdb/nat/x86-linux-tdesc.c +++ b/gdb/nat/x86-linux-tdesc.c @@ -38,6 +38,8 @@ #include <sys/user.h> #include <sys/user.h> +#ifndef IN_PROCESS_AGENT + /* See nat/x86-linux-tdesc.h. */ const target_desc * @@ -116,3 +118,288 @@ x86_linux_tdesc_for_tid (int tid, enum tribool *have_ptrace_getregset, gdb_assert_not_reached ("failed to return tdesc"); } + +#endif /* !IN_PROCESS_AGENT */ + + + +/* A structure used to describe a single cpu feature that might, or might + not, be checked for when creating a target description for one of i386, + amd64, or x32. */ + +struct x86_tdesc_feature { + /* The cpu feature mask. This is a mask against an xcr0 value. */ + uint64_t feature; + + /* Is this feature checked when creating an i386 target description. */ + bool is_i386; + + /* Is this feature checked when creating an amd64 target description. */ + bool is_amd64; + + /* Is this feature checked when creating an x32 target description. */ + bool is_x32; +}; + +/* A constant table that describes all of the cpu features that are + checked when building a target description for i386, amd64, or x32. */ + +static constexpr x86_tdesc_feature x86_linux_all_tdesc_features[] = { + /* Feature, i386, amd64, x32. */ + { X86_XSTATE_PKRU, true, true, true }, + { X86_XSTATE_AVX512, true, true, true }, + { X86_XSTATE_AVX, true, true, true }, + { X86_XSTATE_MPX, true, true, false }, + { X86_XSTATE_SSE, true, false, false }, + { X86_XSTATE_X87, true, false, false } +}; + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an i386 target description. */ + +static constexpr uint64_t +x86_linux_i386_tdesc_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_i386) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an amd64 target description. */ + +static constexpr uint64_t +x86_linux_amd64_tdesc_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_amd64) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an x32 target description. */ + +static constexpr uint64_t +x86_linux_x32_tdesc_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_x32) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an i386 target description. */ + +static constexpr int +x86_linux_i386_tdesc_count () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_i386) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an amd64 target description. */ + +static constexpr int +x86_linux_amd64_tdesc_count () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_amd64) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an x32 target description. */ + +static constexpr int +x86_linux_x32_tdesc_count () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_x32) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +#ifdef IN_PROCESS_AGENT + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_amd64_ipa_tdesc_count () +{ + return x86_linux_amd64_tdesc_count (); +} + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_x32_ipa_tdesc_count () +{ + return x86_linux_x32_tdesc_count (); +} + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_i386_ipa_tdesc_count () +{ + return x86_linux_i386_tdesc_count (); +} + +#endif /* IN_PROCESS_AGENT */ + +/* Convert an xcr0 value into an integer. The integer will be passed to + the in-process-agent where it will then be passed to + x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ + +int +x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0) +{ + /* The following table shows which features are checked for when creating + the target descriptions (see nat/x86-linux-tdesc.c), the feature order + represents the bit order within the generated index number. + + i386 | x87 sse mpx avx avx512 pkru + amd64 | mpx avx avx512 pkru + i32 | avx avx512 pkru + + The features are ordered so that for each mode (i386, amd64, i32) the + generated index will form a continuous range. */ + + int idx = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) + { + if ((xcr0 & x86_linux_all_tdesc_features[i].feature) != 0) + idx |= (1 << i); + } + + return idx; +} + + +#ifdef IN_PROCESS_AGENT + +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +uint64_t +x86_linux_tdesc_idx_to_xcr0 (int idx) +{ + uint64_t xcr0 = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) + { + if ((idx & (1 << i)) != 0) + xcr0 |= x86_linux_all_tdesc_features[i].feature; + } + + return xcr0; +} + +#endif /* IN_PROCESS_AGENT */ + +#if defined __i386__ || !defined IN_PROCESS_AGENT + +/* A cache of all possible i386 target descriptions. */ + +static struct target_desc *i386_tdescs[x86_linux_i386_tdesc_count ()] = { }; + +/* See nat/x86-linux-tdesc.h. */ + +const struct target_desc * +i386_linux_read_description (uint64_t xcr0) +{ + xcr0 &= x86_linux_i386_tdesc_feature_mask (); + int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); + + gdb_assert (idx >= 0 && idx < x86_linux_i386_tdesc_count ()); + + target_desc **tdesc = &i386_tdescs[idx]; + + if (*tdesc == nullptr) + { + *tdesc = i386_create_target_description (xcr0, true, false); + + x86_linux_post_init_tdesc (*tdesc, false); + } + + return *tdesc; +} +#endif + +#ifdef __x86_64__ + +/* A cache of all possible amd64 target descriptions. */ + +static target_desc *amd64_tdescs[x86_linux_amd64_tdesc_count ()] = { }; + +/* A cache of all possible x32 target descriptions. */ + +static target_desc *x32_tdescs[x86_linux_x32_tdesc_count ()] = { }; + +/* See nat/x86-linux-tdesc.h. */ + +const struct target_desc * +amd64_linux_read_description (uint64_t xcr0, bool is_x32) +{ + if (is_x32) + xcr0 &= x86_linux_x32_tdesc_feature_mask (); + else + xcr0 &= x86_linux_amd64_tdesc_feature_mask (); + + int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); + + if (is_x32) + gdb_assert (idx >= 0 && idx < x86_linux_x32_tdesc_count ()); + else + gdb_assert (idx >= 0 && idx < x86_linux_amd64_tdesc_count ()); + + target_desc **tdesc = nullptr; + + if (is_x32) + tdesc = &x32_tdescs[idx]; + else + tdesc = &amd64_tdescs[idx]; + + if (*tdesc == nullptr) + { + *tdesc = amd64_create_target_description (xcr0, is_x32, true, true); + + x86_linux_post_init_tdesc (*tdesc, true); + } + return *tdesc; +} + +#endif diff --git a/gdb/nat/x86-linux-tdesc.h b/gdb/nat/x86-linux-tdesc.h index 3727a8bf95e..648fe0ed0b2 100644 --- a/gdb/nat/x86-linux-tdesc.h +++ b/gdb/nat/x86-linux-tdesc.h @@ -24,6 +24,8 @@ struct target_desc; +#ifndef IN_PROCESS_AGENT + /* Return the target description for Linux thread TID. When *HAVE_PTRACE_GETREGSET is TRIBOOL_UNKNOWN then the current value of @@ -57,19 +59,57 @@ x86_linux_tdesc_for_tid (int tid, enum tribool *have_ptrace_getregset, gdb::function_view<void (uint64_t)> xcr0_init_cb, const char *error_msg, uint64_t *xcr0_storage); +#endif /* !IN_PROCESS_AGENT */ + #ifdef __x86_64__ -/* Return the right amd64-linux target descriptions according to - XCR0_FEATURES_BIT and IS_X32. This is implemented separately in both - GDB and gdbserver. */ +/* Return the AMD64 target descriptions corresponding to XCR0 and IS_X32. */ -extern const target_desc *amd64_linux_read_description - (uint64_t xcr0_features_bit, bool is_x32); +extern const target_desc *amd64_linux_read_description (uint64_t xcr0, + bool is_x32); -#endif +#endif /* __x86_64__ */ + +/* Return the i386 target description corresponding to XCR0. */ -/* Return the target description according to XCR0. This is implemented - separately in both GDB and gdbserver. */ extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); +/* This function is called from amd64_linux_read_description and + i386_linux_read_description after a new target description has been + created, TDESC is the new target description, IS_64BIT will be true + when called from amd64_linux_read_description, otherwise IS_64BIT will + be false. If the *_linux_read_description functions found a cached + target description then this function will not be called. + + Both GDB and gdbserver have their own implementations of this + function. */ + +extern void x86_linux_post_init_tdesc (target_desc *tdesc, bool is_64bit); + +/* Convert an xcr0 value into an integer. The integer will be passed to + the in-process-agent where it will then be passed to + x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ + +extern int x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0); + + +#ifdef IN_PROCESS_AGENT + +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +extern uint64_t x86_linux_tdesc_idx_to_xcr0 (int idx); + +/* Within the in-process-agent we need to pre-initialise all of the target + descriptions, to do this we need to know how many target descriptions + there are for each different target type. These functions return the + target description count for the relevant target. */ + +extern int x86_linux_amd64_ipa_tdesc_count (); +extern int x86_linux_x32_ipa_tdesc_count (); +extern int x86_linux_i386_ipa_tdesc_count (); + +#endif /* IN_PROCESS_AGENT */ + #endif /* NAT_X86_LINUX_TDESC_H */ diff --git a/gdbserver/Makefile.in b/gdbserver/Makefile.in index d12f8746611..c97649378e8 100644 --- a/gdbserver/Makefile.in +++ b/gdbserver/Makefile.in @@ -517,6 +517,10 @@ gdbsupport/%-ipa.o: ../gdbsupport/%.cc $(IPAGENT_COMPILE) $< $(POSTCOMPILE) +nat/%-ipa.o: ../gdb/nat/%.c + $(IPAGENT_COMPILE) $< + $(POSTCOMPILE) + %-ipa.o: ../gdb/%.c $(IPAGENT_COMPILE) -x c++ $< $(POSTCOMPILE) diff --git a/gdbserver/configure.srv b/gdbserver/configure.srv index 7a2702d78bf..36a457f21ed 100644 --- a/gdbserver/configure.srv +++ b/gdbserver/configure.srv @@ -116,6 +116,7 @@ case "${gdbserver_host}" in srv_linux_btrace=yes ipa_obj="linux-i386-ipa.o linux-x86-tdesc-ipa.o" ipa_obj="${ipa_obj} arch/i386-ipa.o" + ipa_obj="${ipa_obj} nat/x86-linux-tdesc-ipa.o" ;; i[34567]86-*-mingw*) srv_regobj="" srv_tgtobj="x86-low.o nat/x86-dregs.o win32-low.o" @@ -380,6 +381,7 @@ case "${gdbserver_host}" in srv_linux_btrace=yes ipa_obj="linux-amd64-ipa.o linux-x86-tdesc-ipa.o" ipa_obj="${ipa_obj} arch/amd64-ipa.o" + ipa_obj="${ipa_obj} nat/x86-linux-tdesc-ipa.o" ;; x86_64-*-mingw*) srv_regobj="" srv_tgtobj="x86-low.o nat/x86-dregs.o" diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index 42de7d2da16..afa6aab5bab 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -21,7 +21,6 @@ #include "server.h" #include <sys/mman.h> #include "tracepoint.h" -#include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" #include "nat/x86-linux-tdesc.h" diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc index 246bcb9813b..f6d26f784af 100644 --- a/gdbserver/linux-i386-ipa.cc +++ b/gdbserver/linux-i386-ipa.cc @@ -21,7 +21,6 @@ #include "server.h" #include <sys/mman.h> #include "tracepoint.h" -#include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" #include "nat/x86-linux-tdesc.h" diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 6dbefdb2f26..7134d20dd7f 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -47,7 +47,6 @@ #include "nat/linux-nat.h" #include "nat/x86-linux.h" #include "nat/x86-linux-dregs.h" -#include "linux-x86-tdesc.h" #include "nat/x86-linux-tdesc.h" #ifdef __x86_64__ diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc index 87cf368a336..39d1d51b8c0 100644 --- a/gdbserver/linux-x86-tdesc.cc +++ b/gdbserver/linux-x86-tdesc.cc @@ -19,292 +19,19 @@ #include "server.h" #include "tdesc.h" -#include "linux-x86-tdesc.h" -#include "arch/i386.h" -#include "gdbsupport/x86-xstate.h" -#ifdef __x86_64__ -#include "arch/amd64.h" -#endif #include "x86-tdesc.h" #include "nat/x86-linux-tdesc.h" -/* A structure used to describe a single cpu feature that might, or might - not, be checked for when creating a target description for one of i386, - amd64, or x32. */ - -struct x86_tdesc_feature { - /* The cpu feature mask. This is a mask against an xcr0 value. */ - uint64_t feature; - - /* Is this feature checked when creating an i386 target description. */ - bool is_i386; - - /* Is this feature checked when creating an amd64 target description. */ - bool is_amd64; - - /* Is this feature checked when creating an x32 target description. */ - bool is_x32; -}; - -/* A constant table that describes all of the cpu features that are - checked when building a target description for i386, amd64, or x32. */ - -static constexpr x86_tdesc_feature x86_linux_all_tdesc_features[] = { - /* Feature, i386, amd64, x32. */ - { X86_XSTATE_PKRU, true, true, true }, - { X86_XSTATE_AVX512, true, true, true }, - { X86_XSTATE_AVX, true, true, true }, - { X86_XSTATE_MPX, true, true, false }, - { X86_XSTATE_SSE, true, false, false }, - { X86_XSTATE_X87, true, false, false } -}; - -/* Return a compile time constant which is a mask of all the cpu features - that are checked for when building an i386 target description. */ - -static constexpr uint64_t -x86_linux_i386_tdesc_feature_mask () -{ - uint64_t mask = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_i386) - mask |= entry.feature; - - return mask; -} - -/* Return a compile time constant which is a mask of all the cpu features - that are checked for when building an amd64 target description. */ - -static constexpr uint64_t -x86_linux_amd64_tdesc_feature_mask () -{ - uint64_t mask = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_amd64) - mask |= entry.feature; - - return mask; -} - -/* Return a compile time constant which is a mask of all the cpu features - that are checked for when building an x32 target description. */ - -static constexpr uint64_t -x86_linux_x32_tdesc_feature_mask () -{ - uint64_t mask = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_x32) - mask |= entry.feature; - - return mask; -} - -/* Return a compile time constant which is a count of the number of cpu - features that are checked for when building an i386 target description. */ - -static constexpr int -x86_linux_i386_tdesc_count () -{ - uint64_t count = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_i386) - ++count; - - gdb_assert (count > 0); - - return (1 << count); -} - -/* Return a compile time constant which is a count of the number of cpu - features that are checked for when building an amd64 target description. */ - -static constexpr int -x86_linux_amd64_tdesc_count () -{ - uint64_t count = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_amd64) - ++count; - - gdb_assert (count > 0); - - return (1 << count); -} - -/* Return a compile time constant which is a count of the number of cpu - features that are checked for when building an x32 target description. */ - -static constexpr int -x86_linux_x32_tdesc_count () -{ - uint64_t count = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_x32) - ++count; - - gdb_assert (count > 0); - - return (1 << count); -} - -#ifdef IN_PROCESS_AGENT - -/* See linux-x86-tdesc.h. */ - -int -x86_linux_amd64_ipa_tdesc_count () -{ - return x86_linux_amd64_tdesc_count (); -} - -/* See linux-x86-tdesc.h. */ - -int -x86_linux_x32_ipa_tdesc_count () -{ - return x86_linux_x32_tdesc_count (); -} - -/* See linux-x86-tdesc.h. */ - -int -x86_linux_i386_ipa_tdesc_count () -{ - return x86_linux_i386_tdesc_count (); -} - -#endif /* IN_PROCESS_AGENT */ - -/* Convert an xcr0 value into an integer. The integer will be passed to - the in-process-agent where it will then be passed to - x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ - -int -x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0) -{ - /* The following table shows which features are checked for when creating - the target descriptions (see nat/x86-linux-tdesc.c), the feature order - represents the bit order within the generated index number. - - i386 | x87 sse mpx avx avx512 pkru - amd64 | mpx avx avx512 pkru - i32 | avx avx512 pkru - - The features are ordered so that for each mode (i386, amd64, i32) the - generated index will form a continuous range. */ - - int idx = 0; - - for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) - { - if ((xcr0 & x86_linux_all_tdesc_features[i].feature) != 0) - idx |= (1 << i); - } - - return idx; -} - - -#ifdef IN_PROCESS_AGENT - -/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) - into an xcr0 value which can then be used to create a target - description. */ - -uint64_t -x86_linux_tdesc_idx_to_xcr0 (int idx) -{ - uint64_t xcr0 = 0; - - for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) - { - if ((idx & (1 << i)) != 0) - xcr0 |= x86_linux_all_tdesc_features[i].feature; - } - - return xcr0; -} - -#endif /* IN_PROCESS_AGENT */ - -#if defined __i386__ || !defined IN_PROCESS_AGENT - -/* A cache of all possible i386 target descriptions. */ - -static struct target_desc *i386_tdescs[x86_linux_i386_tdesc_count ()] = { }; - /* See nat/x86-linux-tdesc.h. */ -const struct target_desc * -i386_linux_read_description (uint64_t xcr0) +void +x86_linux_post_init_tdesc (target_desc *tdesc, bool is_64bit) { - xcr0 &= x86_linux_i386_tdesc_feature_mask (); - int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - - gdb_assert (idx >= 0 && idx < x86_linux_i386_tdesc_count ()); - - target_desc **tdesc = &i386_tdescs[idx]; - - if (*tdesc == nullptr) - { - *tdesc = i386_create_target_description (xcr0, true, false); - - init_target_desc (*tdesc, i386_expedite_regs); - } - - return *tdesc; -} -#endif - #ifdef __x86_64__ - -/* A cache of all possible amd64 target descriptions. */ - -static target_desc *amd64_tdescs[x86_linux_amd64_tdesc_count ()] = { }; - -/* A cache of all possible x32 target descriptions. */ - -static target_desc *x32_tdescs[x86_linux_x32_tdesc_count ()] = { }; - -/* See nat/x86-linux-tdesc.h. */ - -const struct target_desc * -amd64_linux_read_description (uint64_t xcr0, bool is_x32) -{ - if (is_x32) - xcr0 &= x86_linux_x32_tdesc_feature_mask (); + if (is_64bit) + init_target_desc (tdesc, amd64_expedite_regs); else - xcr0 &= x86_linux_amd64_tdesc_feature_mask (); - - int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - - if (is_x32) - gdb_assert (idx >= 0 && idx < x86_linux_x32_tdesc_count ()); - else - gdb_assert (idx >= 0 && idx < x86_linux_amd64_tdesc_count ()); - - target_desc **tdesc = nullptr; - - if (is_x32) - tdesc = &x32_tdescs[idx]; - else - tdesc = &amd64_tdescs[idx]; - - if (*tdesc == nullptr) - { - *tdesc = amd64_create_target_description (xcr0, is_x32, true, true); - - init_target_desc (*tdesc, amd64_expedite_regs); - } - return *tdesc; +#endif + init_target_desc (tdesc, i386_expedite_regs); } -#endif diff --git a/gdbserver/linux-x86-tdesc.h b/gdbserver/linux-x86-tdesc.h deleted file mode 100644 index 70456e4be44..00000000000 --- a/gdbserver/linux-x86-tdesc.h +++ /dev/null @@ -1,50 +0,0 @@ -/* Low level support for x86 (i386 and x86-64), shared between gdbserver - and IPA. - - Copyright (C) 2016-2024 Free Software Foundation, Inc. - - This file is part of GDB. - - This program is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published by - the Free Software Foundation; either version 3 of the License, or - (at your option) any later version. - - This program is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - GNU General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see <http://www.gnu.org/licenses/>. */ - -#ifndef GDBSERVER_LINUX_X86_TDESC_H -#define GDBSERVER_LINUX_X86_TDESC_H - -/* Convert an xcr0 value into an integer. The integer will be passed to - the in-process-agent where it will then be passed to - x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ - -extern int x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0); - -#ifdef IN_PROCESS_AGENT - -/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) - into an xcr0 value which can then be used to create a target - description. */ - -extern uint64_t x86_linux_tdesc_idx_to_xcr0 (int idx); - -/* Within the in-process-agent we need to pre-initialise all of the target - descriptions, to do this we need to know how many target descriptions - there are for each different target type. These functions return the - target description count for the relevant target. */ - -extern int x86_linux_amd64_ipa_tdesc_count (); -extern int x86_linux_x32_ipa_tdesc_count (); -extern int x86_linux_i386_ipa_tdesc_count (); - - -#endif /* IN_PROCESS_AGENT */ - -#endif /* GDBSERVER_LINUX_X86_TDESC_H */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv2 0/7] x86/Linux Target Description Changes 2024-03-05 17:00 ` [PATCHv2 0/7] x86/Linux Target Description Changes Andrew Burgess ` (6 preceding siblings ...) 2024-03-05 17:00 ` [PATCHv2 7/7] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess @ 2024-03-19 16:05 ` John Baldwin 2024-03-23 16:35 ` [PATCHv3 0/8] " Andrew Burgess 8 siblings, 0 replies; 116+ messages in thread From: John Baldwin @ 2024-03-19 16:05 UTC (permalink / raw) To: Andrew Burgess, gdb-patches On 3/5/24 9:00 AM, Andrew Burgess wrote: > In v2: > > - Rebase to current upstream/master, no merge conflicts, > > - Retested. This series all seems fine to me. I did have one comment about what I think would be ideal for x86 tdesc caching in the future in reply to one of the patches. Approved-By: John Baldwin <jhb@FreeBSD.org> -- John Baldwin ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv3 0/8] x86/Linux Target Description Changes 2024-03-05 17:00 ` [PATCHv2 0/7] x86/Linux Target Description Changes Andrew Burgess ` (7 preceding siblings ...) 2024-03-19 16:05 ` [PATCHv2 0/7] x86/Linux Target Description Changes John Baldwin @ 2024-03-23 16:35 ` Andrew Burgess 2024-03-23 16:35 ` [PATCHv3 1/8] gdbserver: convert have_ptrace_getregset to a tribool Andrew Burgess ` (9 more replies) 8 siblings, 10 replies; 116+ messages in thread From: Andrew Burgess @ 2024-03-23 16:35 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, hjl.tools In v3: - Rebased. Nasty merge conflict with 4bb20a6244b7091 which I think I've resolved, but am unable to test. Reposting so the author of that other commit can validate. - Initial testing looks good. Full tests are still running. In v2: - Rebase to current upstream/master, no merge conflicts, - Retested. --- Andrew Burgess (8): gdbserver: convert have_ptrace_getregset to a tribool gdb/x86: move reading of cs and ds state into gdb/nat directory gdbserver/x86: move no-xml code earlier in x86_linux_read_description gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition gdb/gdbserver: share some code relating to target description creation gdb/arch: assert that X86_XSTATE_MPX is not set for x32 gdbserver: update target description creation for x86/linux gdb/gdbserver: share x86/linux tdesc caching gdb/Makefile.in | 1 + gdb/amd64-linux-tdep.c | 33 +-- gdb/amd64-linux-tdep.h | 6 - gdb/arch/amd64.c | 8 +- gdb/configure.nat | 4 +- gdb/i386-linux-tdep.c | 32 +-- gdb/i386-linux-tdep.h | 23 -- gdb/nat/x86-linux-tdesc.c | 411 +++++++++++++++++++++++++++++++++++ gdb/nat/x86-linux-tdesc.h | 115 ++++++++++ gdb/nat/x86-linux.c | 47 ++++ gdb/nat/x86-linux.h | 48 ++++ gdb/x86-linux-nat.c | 123 ++--------- gdbserver/Makefile.in | 4 + gdbserver/configure.srv | 4 + gdbserver/linux-amd64-ipa.cc | 45 +--- gdbserver/linux-arm-low.cc | 6 +- gdbserver/linux-i386-ipa.cc | 25 +-- gdbserver/linux-low.cc | 2 +- gdbserver/linux-low.h | 2 +- gdbserver/linux-x86-low.cc | 189 +++++----------- gdbserver/linux-x86-tdesc.cc | 141 +----------- gdbserver/linux-x86-tdesc.h | 56 ----- 22 files changed, 749 insertions(+), 576 deletions(-) create mode 100644 gdb/nat/x86-linux-tdesc.c create mode 100644 gdb/nat/x86-linux-tdesc.h delete mode 100644 gdbserver/linux-x86-tdesc.h base-commit: e9315f148d56b3f4c7cfeef469633e85933d412c -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv3 1/8] gdbserver: convert have_ptrace_getregset to a tribool 2024-03-23 16:35 ` [PATCHv3 0/8] " Andrew Burgess @ 2024-03-23 16:35 ` Andrew Burgess 2024-03-23 16:35 ` [PATCHv3 2/8] gdb/x86: move reading of cs and ds state into gdb/nat directory Andrew Burgess ` (8 subsequent siblings) 9 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-03-23 16:35 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, hjl.tools, John Baldwin Convert the have_ptrace_getregset global within gdbserver to a tribool. This brings the flag into alignment with the corresponding flag in GDB. The gdbserver have_ptrace_getregset variable is already used as a tribool, it just doesn't have the tribool type. In a future commit I plan to share more code between GDB and gdbserver, and having this variable be the same type in both code bases will make the sharing much easier. There should be no user visible changes after this commit. Approved-By: John Baldwin <jhb@FreeBSD.org> --- gdbserver/linux-arm-low.cc | 6 +++--- gdbserver/linux-low.cc | 2 +- gdbserver/linux-low.h | 2 +- gdbserver/linux-x86-low.cc | 10 +++++----- 4 files changed, 10 insertions(+), 10 deletions(-) diff --git a/gdbserver/linux-arm-low.cc b/gdbserver/linux-arm-low.cc index 396ec88081b..b4f0e071c92 100644 --- a/gdbserver/linux-arm-low.cc +++ b/gdbserver/linux-arm-low.cc @@ -1007,9 +1007,9 @@ arm_target::low_arch_setup () /* Check if PTRACE_GETREGSET works. */ if (ptrace (PTRACE_GETREGSET, tid, NT_PRSTATUS, &iov) == 0) - have_ptrace_getregset = 1; + have_ptrace_getregset = TRIBOOL_TRUE; else - have_ptrace_getregset = 0; + have_ptrace_getregset = TRIBOOL_FALSE; } bool @@ -1122,7 +1122,7 @@ arm_target::get_regs_info () { const struct target_desc *tdesc = current_process ()->tdesc; - if (have_ptrace_getregset == 1 + if (have_ptrace_getregset == TRIBOOL_TRUE && (is_aarch32_linux_description (tdesc) || arm_linux_get_tdesc_fp_type (tdesc) == ARM_FP_TYPE_VFPV3)) return ®s_info_aarch32; diff --git a/gdbserver/linux-low.cc b/gdbserver/linux-low.cc index 9d5a6242803..dfa26d451b0 100644 --- a/gdbserver/linux-low.cc +++ b/gdbserver/linux-low.cc @@ -135,7 +135,7 @@ typedef struct #endif /* Does the current host support PTRACE_GETREGSET? */ -int have_ptrace_getregset = -1; +enum tribool have_ptrace_getregset = TRIBOOL_UNKNOWN; /* Return TRUE if THREAD is the leader thread of the process. */ diff --git a/gdbserver/linux-low.h b/gdbserver/linux-low.h index d34d2738238..eaf87527338 100644 --- a/gdbserver/linux-low.h +++ b/gdbserver/linux-low.h @@ -951,7 +951,7 @@ void thread_db_notice_clone (struct thread_info *parent_thr, ptid_t child_ptid); bool thread_db_thread_handle (ptid_t ptid, gdb_byte **handle, int *handle_len); -extern int have_ptrace_getregset; +extern enum tribool have_ptrace_getregset; /* Search for the value with type MATCH in the auxv vector, with entries of length WORDSIZE bytes, of process with pid PID. If found, store the diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 933d1fb012a..04202e355bb 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -899,7 +899,7 @@ x86_linux_read_description (void) if (ptrace (PTRACE_GETFPXREGS, tid, 0, (long) &fpxregs) < 0) { have_ptrace_getfpxregs = 0; - have_ptrace_getregset = 0; + have_ptrace_getregset = TRIBOOL_FALSE; return i386_linux_read_description (X86_XSTATE_X87); } else @@ -918,7 +918,7 @@ x86_linux_read_description (void) return tdesc_i386_linux_no_xml.get (); } - if (have_ptrace_getregset == -1) + if (have_ptrace_getregset == TRIBOOL_UNKNOWN) { uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; struct iovec iov; @@ -929,10 +929,10 @@ x86_linux_read_description (void) /* Check if PTRACE_GETREGSET works. */ if (ptrace (PTRACE_GETREGSET, tid, (unsigned int) NT_X86_XSTATE, (long) &iov) < 0) - have_ptrace_getregset = 0; + have_ptrace_getregset = TRIBOOL_FALSE; else { - have_ptrace_getregset = 1; + have_ptrace_getregset = TRIBOOL_TRUE; /* Get XCR0 from XSAVE extended state. */ xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET @@ -955,7 +955,7 @@ x86_linux_read_description (void) } /* Check the native XCR0 only if PTRACE_GETREGSET is available. */ - xcr0_features = (have_ptrace_getregset + xcr0_features = (have_ptrace_getregset == TRIBOOL_TRUE && (xcr0 & X86_XSTATE_ALL_MASK)); if (xcr0_features) -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv3 2/8] gdb/x86: move reading of cs and ds state into gdb/nat directory 2024-03-23 16:35 ` [PATCHv3 0/8] " Andrew Burgess 2024-03-23 16:35 ` [PATCHv3 1/8] gdbserver: convert have_ptrace_getregset to a tribool Andrew Burgess @ 2024-03-23 16:35 ` Andrew Burgess 2024-03-23 16:35 ` [PATCHv3 3/8] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess ` (7 subsequent siblings) 9 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-03-23 16:35 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, hjl.tools, John Baldwin This patch is part of a series that has the aim of making the code that, for x86, reads the target description for a native process shared between GDB and gdbserver. Within GDB part of this process involves reading the cs and ds state from the 'struct user_regs_struct' using a ptrace call. This isn't done by gdbserver, which is part of the motivation for this whole series; the approach gdbserver takes is inferior to the approach GDB takes. This commit moves the reading of cs and ds, which is used to figure out if a thread is 32-bit or 64-bit (or in x32 mode), into the gdb/nat directory so that the code could be shared with gdbserver, but at this point I'm not actually using the code in gdbserver, that will come later. As such there should be no user visible changes after this commit, GDB continues to do things as it did before (reading cs/ds), while gdbserver continues to use its own approach (which doesn't require reading cs/ds). Approved-By: John Baldwin <jhb@FreeBSD.org> --- gdb/nat/x86-linux.c | 47 +++++++++++++++++++++++++++++++++++++++++++++ gdb/nat/x86-linux.h | 28 +++++++++++++++++++++++++++ gdb/x86-linux-nat.c | 42 +++++----------------------------------- 3 files changed, 80 insertions(+), 37 deletions(-) diff --git a/gdb/nat/x86-linux.c b/gdb/nat/x86-linux.c index e61f4d749ba..4242a1baafb 100644 --- a/gdb/nat/x86-linux.c +++ b/gdb/nat/x86-linux.c @@ -20,6 +20,8 @@ #include "gdbsupport/common-defs.h" #include "x86-linux.h" #include "x86-linux-dregs.h" +#include "nat/gdb_ptrace.h" +#include <sys/user.h> /* Per-thread arch-specific data we want to keep. */ @@ -80,3 +82,48 @@ x86_linux_prepare_to_resume (struct lwp_info *lwp) { x86_linux_update_debug_registers (lwp); } + +#ifdef __x86_64__ +/* Value of CS segment register: + 64bit process: 0x33 + 32bit process: 0x23 */ +#define AMD64_LINUX_USER64_CS 0x33 + +/* Value of DS segment register: + LP64 process: 0x0 + X32 process: 0x2b */ +#define AMD64_LINUX_X32_DS 0x2b +#endif + +/* See nat/x86-linux.h. */ + +x86_linux_arch_size +x86_linux_ptrace_get_arch_size (int tid) +{ +#ifdef __x86_64__ + unsigned long cs; + unsigned long ds; + + /* Get CS register. */ + errno = 0; + cs = ptrace (PTRACE_PEEKUSER, tid, + offsetof (struct user_regs_struct, cs), 0); + if (errno != 0) + perror_with_name (_("Couldn't get CS register")); + + bool is_64bit = cs == AMD64_LINUX_USER64_CS; + + /* Get DS register. */ + errno = 0; + ds = ptrace (PTRACE_PEEKUSER, tid, + offsetof (struct user_regs_struct, ds), 0); + if (errno != 0) + perror_with_name (_("Couldn't get DS register")); + + bool is_x32 = ds == AMD64_LINUX_X32_DS; + + return x86_linux_arch_size (is_64bit, is_x32); +#else + return x86_linux_arch_size (false, false); +#endif +} diff --git a/gdb/nat/x86-linux.h b/gdb/nat/x86-linux.h index 822882173f9..15153ea277e 100644 --- a/gdb/nat/x86-linux.h +++ b/gdb/nat/x86-linux.h @@ -47,4 +47,32 @@ extern void x86_linux_delete_thread (struct arch_lwp_info *arch_lwp); extern void x86_linux_prepare_to_resume (struct lwp_info *lwp); +/* Return value from x86_linux_ptrace_get_arch_size function. Indicates if + a thread is 32-bit, 64-bit, or x32. */ + +struct x86_linux_arch_size +{ + explicit x86_linux_arch_size (bool is_64bit, bool is_x32) + : m_is_64bit (is_64bit), + m_is_x32 (is_x32) + { + /* Nothing. */ + } + + bool is_64bit () const + { return m_is_64bit; } + + bool is_x32 () const + { return m_is_x32; } + +private: + bool m_is_64bit = false; + bool m_is_x32 = false; +}; + +/* Use ptrace calls to figure out if thread TID is 32-bit, 64-bit, or + 64-bit running in x32 mode. */ + +extern x86_linux_arch_size x86_linux_ptrace_get_arch_size (int tid); + #endif /* NAT_X86_LINUX_H */ diff --git a/gdb/x86-linux-nat.c b/gdb/x86-linux-nat.c index b93ffca38db..b39d05c401f 100644 --- a/gdb/x86-linux-nat.c +++ b/gdb/x86-linux-nat.c @@ -91,18 +91,6 @@ x86_linux_nat_target::post_startup_inferior (ptid_t ptid) linux_nat_target::post_startup_inferior (ptid); } -#ifdef __x86_64__ -/* Value of CS segment register: - 64bit process: 0x33 - 32bit process: 0x23 */ -#define AMD64_LINUX_USER64_CS 0x33 - -/* Value of DS segment register: - LP64 process: 0x0 - X32 process: 0x2b */ -#define AMD64_LINUX_X32_DS 0x2b -#endif - /* Get Linux/x86 target description from running target. */ const struct target_desc * @@ -122,31 +110,11 @@ x86_linux_nat_target::read_description () tid = inferior_ptid.pid (); #ifdef __x86_64__ - { - unsigned long cs; - unsigned long ds; - - /* Get CS register. */ - errno = 0; - cs = ptrace (PTRACE_PEEKUSER, tid, - offsetof (struct user_regs_struct, cs), 0); - if (errno != 0) - perror_with_name (_("Couldn't get CS register")); - - is_64bit = cs == AMD64_LINUX_USER64_CS; - - /* Get DS register. */ - errno = 0; - ds = ptrace (PTRACE_PEEKUSER, tid, - offsetof (struct user_regs_struct, ds), 0); - if (errno != 0) - perror_with_name (_("Couldn't get DS register")); - - is_x32 = ds == AMD64_LINUX_X32_DS; - - if (sizeof (void *) == 4 && is_64bit && !is_x32) - error (_("Can't debug 64-bit process with 32-bit GDB")); - } + + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); + is_64bit = arch_size.is_64bit (); + is_x32 = arch_size.is_x32 (); + #elif HAVE_PTRACE_GETFPXREGS if (have_ptrace_getfpxregs == -1) { -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv3 3/8] gdbserver/x86: move no-xml code earlier in x86_linux_read_description 2024-03-23 16:35 ` [PATCHv3 0/8] " Andrew Burgess 2024-03-23 16:35 ` [PATCHv3 1/8] gdbserver: convert have_ptrace_getregset to a tribool Andrew Burgess 2024-03-23 16:35 ` [PATCHv3 2/8] gdb/x86: move reading of cs and ds state into gdb/nat directory Andrew Burgess @ 2024-03-23 16:35 ` Andrew Burgess 2024-03-23 16:35 ` [PATCHv3 4/8] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess ` (6 subsequent siblings) 9 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-03-23 16:35 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, hjl.tools, John Baldwin This commit is part of a series that aims to share more of the x86 target description reading/generation code between GDB and gdbserver. There are a huge number of similarities between the code in gdbserver's x86_linux_read_description function and GDB's x86_linux_nat_target::read_description function, and it is this similarity that I plan, in a later commit, to share between GDB and gdbserver. However, one thing that is different in x86_linux_read_description is the code inside the '!use_xml' block. This is the code that handles the case where gdbserver is not allowed to send an XML target description back to GDB. In this case gdbserver uses some predefined, fixed, target descriptions. First, it's worth noting that I suspect this code is not tested any more. I couldn't find anything in the testsuite that tries to disable XML target description support. And the idea of having a single "fixed" target description really doesn't work well when we think about all the various x86 extensions that exist. Part of me would like to rip out the no-xml support in gdbserver (at least for x86), and if a GDB connects that doesn't support XML target descriptions, gdbserver can just give an error and drop the connection. GDB has supported XML target descriptions for 16 years now, I think it would be reasonable for our shipped gdbserver to drop support for the old way of doing things. Anyway.... this commit doesn't do that. What I did notice was that, over time, the '!use_xml' block appears to have "drifted" within the x86_linux_read_description function; it's now not the first check we do. Instead we make some ptrace calls and return a target description generated based on the result of these ptrace calls. Surely it only makes sense to generate variable target descriptions if we can send these back to GDB? So in this commit I propose to move the '!use_xml' block earlier in the x86_linux_read_description function. The benefit of this is that this leaves the later half of x86_linux_read_description much more similar to the GDB function x86_linux_nat_target::read_description and sets us up for potentially sharing code between GDB and gdbserver in a later commit. Approved-By: John Baldwin <jhb@FreeBSD.org> --- gdbserver/linux-x86-low.cc | 27 ++++++++++++++++----------- 1 file changed, 16 insertions(+), 11 deletions(-) diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 04202e355bb..872c3fc69c7 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -891,6 +891,22 @@ x86_linux_read_description (void) #endif } + /* If we are not allowed to send an XML target description then we need + to use the hard-wired target descriptions. This corresponds to GDB's + default machine for x86. + + This check needs to occur before any returns statements that might + generate some alternative target descriptions. */ + if (!use_xml) + { +#ifdef __x86_64__ + if (machine == EM_X86_64) + return tdesc_amd64_linux_no_xml.get (); + else +#endif + return tdesc_i386_linux_no_xml.get (); + } + #if !defined __x86_64__ && defined HAVE_PTRACE_GETFPXREGS if (machine == EM_386 && have_ptrace_getfpxregs == -1) { @@ -907,17 +923,6 @@ x86_linux_read_description (void) } #endif - if (!use_xml) - { - /* Don't use XML. */ -#ifdef __x86_64__ - if (machine == EM_X86_64) - return tdesc_amd64_linux_no_xml.get (); - else -#endif - return tdesc_i386_linux_no_xml.get (); - } - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) { uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv3 4/8] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition 2024-03-23 16:35 ` [PATCHv3 0/8] " Andrew Burgess ` (2 preceding siblings ...) 2024-03-23 16:35 ` [PATCHv3 3/8] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess @ 2024-03-23 16:35 ` Andrew Burgess 2024-03-23 16:35 ` [PATCHv3 5/8] gdb/gdbserver: share some code relating to target description creation Andrew Burgess ` (5 subsequent siblings) 9 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-03-23 16:35 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, hjl.tools, John Baldwin Share the definition of I386_LINUX_XSAVE_XCR0_OFFSET between GDB and gdbserver. This commit is part of a series that aims to share more of the x86 target description creation code between GDB and gdbserver. The I386_LINUX_XSAVE_XCR0_OFFSET #define is used as part of the target description creation, and I noticed that this constant is defined separately for GDB and gdbserver. This commit moves the definition into gdb/nat/x86-linux.h, which allows the #define to be shared. There should be no user visible changes after this commit. Approved-By: John Baldwin <jhb@FreeBSD.org> --- gdb/amd64-linux-tdep.c | 1 + gdb/i386-linux-tdep.c | 1 + gdb/i386-linux-tdep.h | 20 -------------------- gdb/nat/x86-linux.h | 20 ++++++++++++++++++++ gdbserver/linux-x86-low.cc | 22 +--------------------- 5 files changed, 23 insertions(+), 41 deletions(-) diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c index 1deb13b4e9a..a512ec5dd02 100644 --- a/gdb/amd64-linux-tdep.c +++ b/gdb/amd64-linux-tdep.c @@ -42,6 +42,7 @@ #include "arch/amd64.h" #include "target-descriptions.h" #include "expop.h" +#include "nat/x86-linux.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_AMD64 "syscalls/amd64-linux.xml" diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index a2f937690c2..8dd7203b6c3 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -40,6 +40,7 @@ #include "i387-tdep.h" #include "gdbsupport/x86-xstate.h" +#include "nat/x86-linux.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_I386 "syscalls/i386-linux.xml" diff --git a/gdb/i386-linux-tdep.h b/gdb/i386-linux-tdep.h index 5891747572b..07593c6a8ec 100644 --- a/gdb/i386-linux-tdep.h +++ b/gdb/i386-linux-tdep.h @@ -58,26 +58,6 @@ extern void i386_linux_report_signal_info (struct gdbarch *gdbarch, /* Return the target description according to XCR0. */ extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); -/* Format of XSAVE extended state is: - struct - { - fxsave_bytes[0..463] - sw_usable_bytes[464..511] - xstate_hdr_bytes[512..575] - extended state regions (AVX, MPX, AVX512, PKRU, etc.) - }; - - Same memory layout will be used for the coredump NT_X86_XSTATE - representing the XSAVE extended state registers. - - The first 8 bytes of the sw_usable_bytes[464..467] is the OS enabled - extended state mask, which is the same as the extended control register - 0 (the XFEATURE_ENABLED_MASK register), XCR0. We can use this mask - together with the mask saved in the xstate_hdr_bytes to determine what - states the processor/OS supports and what state, used or initialized, - the process/thread is in. */ -#define I386_LINUX_XSAVE_XCR0_OFFSET 464 - extern int i386_linux_gregset_reg_offset[]; /* Return x86 siginfo type. */ diff --git a/gdb/nat/x86-linux.h b/gdb/nat/x86-linux.h index 15153ea277e..855a8d14f91 100644 --- a/gdb/nat/x86-linux.h +++ b/gdb/nat/x86-linux.h @@ -22,6 +22,26 @@ #include "nat/linux-nat.h" +/* Format of XSAVE extended state is: + struct + { + fxsave_bytes[0..463] + sw_usable_bytes[464..511] + xstate_hdr_bytes[512..575] + extended state regions (AVX, MPX, AVX512, PKRU, etc.) + }; + + Same memory layout will be used for the coredump NT_X86_XSTATE + representing the XSAVE extended state registers. + + The first 8 bytes of the sw_usable_bytes[464..467] is the OS enabled + extended state mask, which is the same as the extended control register + 0 (the XFEATURE_ENABLED_MASK register), XCR0. We can use this mask + together with the mask saved in the xstate_hdr_bytes to determine what + states the processor/OS supports and what state, used or initialized, + the process/thread is in. */ +#define I386_LINUX_XSAVE_XCR0_OFFSET 464 + /* Set whether our local mirror of LWP's debug registers has been changed since the values were last written to the thread. Nonzero indicates that a change has been made, zero indicates no change. */ diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 872c3fc69c7..30d876efc5d 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -27,6 +27,7 @@ #include "gdbsupport/x86-xstate.h" #include "nat/x86-xstate.h" #include "nat/gdb_ptrace.h" +#include "nat/x86-linux.h" #ifdef __x86_64__ #include "nat/amd64-linux-siginfo.h" @@ -832,27 +833,6 @@ x86_target::low_siginfo_fixup (siginfo_t *ptrace, gdb_byte *inf, int direction) \f static int use_xml; -/* Format of XSAVE extended state is: - struct - { - fxsave_bytes[0..463] - sw_usable_bytes[464..511] - xstate_hdr_bytes[512..575] - avx_bytes[576..831] - future_state etc - }; - - Same memory layout will be used for the coredump NT_X86_XSTATE - representing the XSAVE extended state registers. - - The first 8 bytes of the sw_usable_bytes[464..467] is the OS enabled - extended state mask, which is the same as the extended control register - 0 (the XFEATURE_ENABLED_MASK register), XCR0. We can use this mask - together with the mask saved in the xstate_hdr_bytes to determine what - states the processor/OS supports and what state, used or initialized, - the process/thread is in. */ -#define I386_LINUX_XSAVE_XCR0_OFFSET 464 - /* Does the current host support the GETFPXREGS request? The header file may or may not define it, and even if it is defined, the kernel will return EIO if it's running on a pre-SSE processor. */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv3 5/8] gdb/gdbserver: share some code relating to target description creation 2024-03-23 16:35 ` [PATCHv3 0/8] " Andrew Burgess ` (3 preceding siblings ...) 2024-03-23 16:35 ` [PATCHv3 4/8] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess @ 2024-03-23 16:35 ` Andrew Burgess 2024-03-23 16:35 ` [PATCHv3 6/8] gdb/arch: assert that X86_XSTATE_MPX is not set for x32 Andrew Burgess ` (4 subsequent siblings) 9 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-03-23 16:35 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, hjl.tools, John Baldwin This commit is part of a series to share more of the x86 target description creation code between GDB and gdbserver. Unlike previous commits which were mostly refactoring, this commit is the first that makes a real change, though that change should mostly be for gdbserver; I've largely adopted the "GDB" way of doing things for gdbserver, and this fixes a real gdbserver bug. On a x86-64 Linux target, running the test: gdb.server/connect-with-no-symbol-file.exp results in two core files being created. Both of these core files are from the inferior process, created after gdbserver has detached. In this test a gdbserver process is started and then, after gdbserver has started, but before GDB attaches, we either delete the inferior executable, or change its permissions so it can't be read. Only after doing this do we attempt to connect with GDB. As GDB connects to gdbserver, gdbserver attempts to figure out the target description so that it can send the description to GDB, this involves a call to x86_linux_read_description. In x86_linux_read_description one of the first things we do is try to figure out if the process is 32-bit or 64-bit. To do this we look up the executable via the thread-id, and then attempt to read the architecture size from the executable. This isn't going to work if the executable has been deleted, or is no longer readable. And so, as we can't read the executable, we default to an i386 target and use an i386 target description. A consequence of using an i386 target description is that addresses are assumed to be 32-bits. Here's an example session that shows the problems this causes. This is run on an x86-64 machine, and the test binary (xx.x) is a standard 64-bit x86-64 binary: shell_1$ gdbserver --once localhost :54321 /tmp/xx.x shell_2$ gdb -q (gdb) set sysroot (gdb) shell chmod 000 /tmp/xx.x (gdb) target remote :54321 Remote debugging using :54321 warning: /tmp/xx.x: Permission denied. 0xf7fd3110 in ?? () (gdb) show architecture The target architecture is set to "auto" (currently "i386"). (gdb) p/x $pc $1 = 0xf7fd3110 (gdb) info proc mappings process 2412639 Mapped address spaces: Start Addr End Addr Size Offset Perms objfile 0x400000 0x401000 0x1000 0x0 r--p /tmp/xx.x 0x401000 0x402000 0x1000 0x1000 r-xp /tmp/xx.x 0x402000 0x403000 0x1000 0x2000 r--p /tmp/xx.x 0x403000 0x405000 0x2000 0x2000 rw-p /tmp/xx.x 0xf7fcb000 0xf7fcf000 0x4000 0x0 r--p [vvar] 0xf7fcf000 0xf7fd1000 0x2000 0x0 r-xp [vdso] 0xf7fd1000 0xf7fd3000 0x2000 0x0 r--p /usr/lib64/ld-2.30.so 0xf7fd3000 0xf7ff3000 0x20000 0x2000 r-xp /usr/lib64/ld-2.30.so 0xf7ff3000 0xf7ffb000 0x8000 0x22000 r--p /usr/lib64/ld-2.30.so 0xf7ffc000 0xf7ffe000 0x2000 0x2a000 rw-p /usr/lib64/ld-2.30.so 0xf7ffe000 0xf7fff000 0x1000 0x0 rw-p 0xfffda000 0xfffff000 0x25000 0x0 rw-p [stack] 0xff600000 0xff601000 0x1000 0x0 r-xp [vsyscall] (gdb) info inferiors Num Description Connection Executable * 1 process 2412639 1 (remote :54321) (gdb) shell cat /proc/2412639/maps 00400000-00401000 r--p 00000000 fd:03 45907133 /tmp/xx.x 00401000-00402000 r-xp 00001000 fd:03 45907133 /tmp/xx.x 00402000-00403000 r--p 00002000 fd:03 45907133 /tmp/xx.x 00403000-00405000 rw-p 00002000 fd:03 45907133 /tmp/xx.x 7ffff7fcb000-7ffff7fcf000 r--p 00000000 00:00 0 [vvar] 7ffff7fcf000-7ffff7fd1000 r-xp 00000000 00:00 0 [vdso] 7ffff7fd1000-7ffff7fd3000 r--p 00000000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7fd3000-7ffff7ff3000 r-xp 00002000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7ff3000-7ffff7ffb000 r--p 00022000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7ffc000-7ffff7ffe000 rw-p 0002a000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7ffe000-7ffff7fff000 rw-p 00000000 00:00 0 7ffffffda000-7ffffffff000 rw-p 00000000 00:00 0 [stack] ffffffffff600000-ffffffffff601000 r-xp 00000000 00:00 0 [vsyscall] (gdb) Notice the difference between the mappings reported via GDB and those reported directly from the kernel via /proc/PID/maps, the addresses of every mapping is clamped to 32-bits for GDB, while the kernel reports real 64-bit addresses. Notice also that the $pc value is a 32-bit value. It appears to be within one of the mappings reported by GDB, but is outside any of the mappings reported from the kernel. And this is where the problem arises. When gdbserver detaches from the inferior we pass the inferior the address from which it should resume. Due to the 32/64 bit confusion we tell the inferior to resume from the 32-bit $pc value, which is not within any valid mapping, and so, as soon as the inferior resumes, it segfaults. If we look at how GDB (not gdbserver) figures out its target description then we see an interesting difference. GDB doesn't try to read the executable. Instead GDB uses ptrace to query the thread's state, and uses this to figure out the if the thread is 32 or 64 bit. If we update gdbserver to do it the "GDB" way then the above problem is resolved, gdbserver now sees the process as 64-bit, and when we detach from the inferior we give it the correct 64-bit address, and the inferior no longer segfaults. Now, I could just update the gdbserver code, but better, I think, to share one copy of the code between GDB and gdbserver in gdb/nat/. That is what this commit does. The cores of x86_linux_read_description from gdbserver and x86_linux_nat_target::read_description from GDB are moved into a new file gdb/nat/x86-linux-tdesc.c and combined into a single function x86_linux_tdesc_for_tid which is called from each location. This new function does things the GDB way, the only changes are to allow for the sharing; we now have a callback function to call the first time that the xcr0 state is read, this allows for GDB and gdbserver to perform their own initialisation as needed, and additionally, the new function takes a pointer for where to cache the xcr0 value, this isn't needed for this commit, but will be useful in a later commit where gdbserver will want to read this cached xcr0 value. Another thing to note about this commit is how the functions i386_linux_read_description and amd64_linux_read_description are handled. For now I've left these function as implemented separately in GDB and gdbserver. I've moved the declarations of these functions into gdb/nat/x86-linux-tdesc.h, but the implementations are left as separate. A later commit in this series will make these functions shared too, but doing this is not trivial, so I've left that for a separate commit. Merging the declarations as I've done here ensures that everyone implements the function to the same API, and once these functions are shared (in a later commit) we'll want a shared declaration anyway. Approved-By: John Baldwin <jhb@FreeBSD.org> --- gdb/Makefile.in | 1 + gdb/amd64-linux-tdep.c | 1 + gdb/amd64-linux-tdep.h | 6 -- gdb/configure.nat | 4 +- gdb/i386-linux-tdep.c | 1 + gdb/i386-linux-tdep.h | 3 - gdb/nat/x86-linux-tdesc.c | 124 +++++++++++++++++++++++++++++ gdb/nat/x86-linux-tdesc.h | 75 ++++++++++++++++++ gdb/x86-linux-nat.c | 91 ++++----------------- gdbserver/configure.srv | 2 + gdbserver/linux-amd64-ipa.cc | 1 + gdbserver/linux-i386-ipa.cc | 1 + gdbserver/linux-x86-low.cc | 148 +++++++++++------------------------ gdbserver/linux-x86-tdesc.cc | 1 + gdbserver/linux-x86-tdesc.h | 7 -- 15 files changed, 267 insertions(+), 199 deletions(-) create mode 100644 gdb/nat/x86-linux-tdesc.c create mode 100644 gdb/nat/x86-linux-tdesc.h diff --git a/gdb/Makefile.in b/gdb/Makefile.in index 331620375ae..38f4d5fde98 100644 --- a/gdb/Makefile.in +++ b/gdb/Makefile.in @@ -1585,6 +1585,7 @@ HFILES_NO_SRCDIR = \ nat/x86-gcc-cpuid.h \ nat/x86-linux.h \ nat/x86-linux-dregs.h \ + nat/x86-linux-tdesc.h \ python/py-event.h \ python/py-events.h \ python/py-stopevent.h \ diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c index a512ec5dd02..7e0900dc6f9 100644 --- a/gdb/amd64-linux-tdep.c +++ b/gdb/amd64-linux-tdep.c @@ -43,6 +43,7 @@ #include "target-descriptions.h" #include "expop.h" #include "nat/x86-linux.h" +#include "nat/x86-linux-tdesc.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_AMD64 "syscalls/amd64-linux.xml" diff --git a/gdb/amd64-linux-tdep.h b/gdb/amd64-linux-tdep.h index 2003dcda78f..0ec49e7fe03 100644 --- a/gdb/amd64-linux-tdep.h +++ b/gdb/amd64-linux-tdep.h @@ -43,12 +43,6 @@ extern struct target_desc *tdesc_x32_linux; extern struct target_desc *tdesc_x32_avx_linux; extern struct target_desc *tdesc_x32_avx_avx512_linux; -/* Return the right amd64-linux target descriptions according to - XCR0_FEATURES_BIT and IS_X32. */ - -const target_desc *amd64_linux_read_description (uint64_t xcr0_features_bit, - bool is_x32); - /* Enum that defines the syscall identifiers for amd64 linux. Used for process record/replay, these will be translated into a gdb-canonical set of syscall ids in linux-record.c. */ diff --git a/gdb/configure.nat b/gdb/configure.nat index 8b98511cef7..4bcc0696027 100644 --- a/gdb/configure.nat +++ b/gdb/configure.nat @@ -256,7 +256,7 @@ case ${gdb_host} in NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ nat/x86-xstate.o \ i386-linux-nat.o x86-linux-nat.o nat/linux-btrace.o \ - nat/x86-linux.o nat/x86-linux-dregs.o" + nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o" ;; ia64) # Host: Intel IA-64 running GNU/Linux @@ -322,7 +322,7 @@ case ${gdb_host} in NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ nat/x86-xstate.o amd64-nat.o amd64-linux-nat.o x86-linux-nat.o \ nat/linux-btrace.o \ - nat/x86-linux.o nat/x86-linux-dregs.o \ + nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o \ nat/amd64-linux-siginfo.o" ;; sparc) diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index 8dd7203b6c3..f5f7a36bf36 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -41,6 +41,7 @@ #include "i387-tdep.h" #include "gdbsupport/x86-xstate.h" #include "nat/x86-linux.h" +#include "nat/x86-linux-tdesc.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_I386 "syscalls/i386-linux.xml" diff --git a/gdb/i386-linux-tdep.h b/gdb/i386-linux-tdep.h index 07593c6a8ec..e8691cd778e 100644 --- a/gdb/i386-linux-tdep.h +++ b/gdb/i386-linux-tdep.h @@ -55,9 +55,6 @@ extern void i386_linux_report_signal_info (struct gdbarch *gdbarch, struct ui_out *uiout, enum gdb_signal siggnal); -/* Return the target description according to XCR0. */ -extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); - extern int i386_linux_gregset_reg_offset[]; /* Return x86 siginfo type. */ diff --git a/gdb/nat/x86-linux-tdesc.c b/gdb/nat/x86-linux-tdesc.c new file mode 100644 index 00000000000..be7014d2b5f --- /dev/null +++ b/gdb/nat/x86-linux-tdesc.c @@ -0,0 +1,124 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#include "gdbsupport/common-defs.h" +#include "nat/x86-linux-tdesc.h" +#ifdef __x86_64__ +#include "arch/amd64.h" +#endif +#include "arch/i386.h" + +#include "gdbsupport/common-defs.h" +#include "nat/x86-linux.h" +#include "nat/x86-linux-dregs.h" +#include "nat/gdb_ptrace.h" +#include "nat/x86-xstate.h" +#include "nat/x86-linux-tdesc.h" + +#include <cstdint> +#include <cstdlib> +#include <linux/uio.h> +#include <elf.h> +#include <sys/user.h> +#include <sys/user.h> + +/* See nat/x86-linux-tdesc.h. */ + +const target_desc * +x86_linux_tdesc_for_tid (int tid, enum tribool *have_ptrace_getregset, + gdb::function_view<void (uint64_t)> xcr0_init_cb, + const char *error_msg, uint64_t *xcr0_storage) +{ +#ifdef __x86_64__ + + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); + bool is_64bit = arch_size.is_64bit (); + bool is_x32 = arch_size.is_x32 (); + + if (sizeof (void *) == 4 && is_64bit && !is_x32) + error ("%s", error_msg); + +#elif HAVE_PTRACE_GETFPXREGS + if (have_ptrace_getfpxregs == -1) + { + elf_fpxregset_t fpxregs; + + if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) + { + have_ptrace_getfpxregs = 0; + *have_ptrace_getregset = TRIBOOL_FALSE; + return i386_linux_read_description (X86_XSTATE_X87_MASK); + } + } +#endif + + if (*have_ptrace_getregset == TRIBOOL_UNKNOWN) + { + uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; + struct iovec iov; + + iov.iov_base = xstateregs; + iov.iov_len = sizeof (xstateregs); + + /* Check if PTRACE_GETREGSET works. */ + if (ptrace (PTRACE_GETREGSET, tid, + (unsigned int) NT_X86_XSTATE, &iov) < 0) + { + *have_ptrace_getregset = TRIBOOL_FALSE; + *xcr0_storage = 0; + } + else + { + *have_ptrace_getregset = TRIBOOL_TRUE; + + /* Get XCR0 from XSAVE extended state. */ + *xcr0_storage = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET + / sizeof (uint64_t))]; + +#ifdef __x86_64__ + /* No MPX on x32. */ + if (is_64bit && is_x32) + *xcr0_storage &= ~X86_XSTATE_MPX; +#endif /* __x86_64__ */ + + xcr0_init_cb (*xcr0_storage); + } + } + + /* Check the native XCR0 only if PTRACE_GETREGSET is available. If + PTRACE_GETREGSET is not available then set xcr0_features_bits to + zero so that the "no-features" descriptions are returned by the + switches below. */ + uint64_t xcr0_features_bits; + if (*have_ptrace_getregset == TRIBOOL_TRUE) + xcr0_features_bits = *xcr0_storage & X86_XSTATE_ALL_MASK; + else + xcr0_features_bits = 0; + +#ifdef __x86_64__ + if (is_64bit) + { + return amd64_linux_read_description (xcr0_features_bits, is_x32); + } + else +#endif + return i386_linux_read_description (xcr0_features_bits); + + gdb_assert_not_reached ("failed to return tdesc"); +} diff --git a/gdb/nat/x86-linux-tdesc.h b/gdb/nat/x86-linux-tdesc.h new file mode 100644 index 00000000000..3727a8bf95e --- /dev/null +++ b/gdb/nat/x86-linux-tdesc.h @@ -0,0 +1,75 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#ifndef NAT_X86_LINUX_TDESC_H +#define NAT_X86_LINUX_TDESC_H + +#include "gdbsupport/function-view.h" + +struct target_desc; + +/* Return the target description for Linux thread TID. + + When *HAVE_PTRACE_GETREGSET is TRIBOOL_UNKNOWN then the current value of + xcr0 is read using ptrace calls and stored into *XCR0_STORAGE. Then + XCR0_INIT_CB is called with the value of *XCR0_STORAGE and + *HAVE_PTRACE_GETREGSET is set to TRIBOOL_TRUE. + + If the attempt to read xcr0 using ptrace fails then *XCR0_STORAGE is set + to zero and *HAVE_PTRACE_GETREGSET is set to TRIBOOL_FALSE. + + The storage pointed to by XCR0_STORAGE must exist until the program + terminates, this storage is used to cache the xcr0 value. As such + XCR0_INIT_CB will only be called once if xcr0 is successfully read using + ptrace, or not at all if the ptrace call fails. + + This function returns a target description based on the extracted xcr0 + value along with other characteristics of the thread identified by TID. + + This function can return nullptr if we encounter a machine configuration + for which a target_desc cannot be created. Ideally this would not be + the case, we should be able to create a target description for every + possible machine configuration. See amd64_linux_read_description and + i386_linux_read_description for cases when nullptr might be + returned. + + ERROR_MSG is using in an error() call if we try to create a target + description for a 64-bit process but this is a 32-bit build of GDB. */ + +extern const target_desc * +x86_linux_tdesc_for_tid (int tid, enum tribool *have_ptrace_getregset, + gdb::function_view<void (uint64_t)> xcr0_init_cb, + const char *error_msg, uint64_t *xcr0_storage); + +#ifdef __x86_64__ + +/* Return the right amd64-linux target descriptions according to + XCR0_FEATURES_BIT and IS_X32. This is implemented separately in both + GDB and gdbserver. */ + +extern const target_desc *amd64_linux_read_description + (uint64_t xcr0_features_bit, bool is_x32); + +#endif + +/* Return the target description according to XCR0. This is implemented + separately in both GDB and gdbserver. */ +extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); + +#endif /* NAT_X86_LINUX_TDESC_H */ diff --git a/gdb/x86-linux-nat.c b/gdb/x86-linux-nat.c index b39d05c401f..872e27e739f 100644 --- a/gdb/x86-linux-nat.c +++ b/gdb/x86-linux-nat.c @@ -42,6 +42,7 @@ #include "nat/x86-linux.h" #include "nat/x86-linux-dregs.h" #include "nat/linux-ptrace.h" +#include "nat/x86-linux-tdesc.h" /* linux_nat_target::low_new_fork implementation. */ @@ -96,90 +97,26 @@ x86_linux_nat_target::post_startup_inferior (ptid_t ptid) const struct target_desc * x86_linux_nat_target::read_description () { - int tid; - int is_64bit = 0; -#ifdef __x86_64__ - int is_x32; -#endif - static uint64_t xcr0; - uint64_t xcr0_features_bits; + static uint64_t xcr0_storage; if (inferior_ptid == null_ptid) return this->beneath ()->read_description (); - tid = inferior_ptid.pid (); - -#ifdef __x86_64__ - - x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); - is_64bit = arch_size.is_64bit (); - is_x32 = arch_size.is_x32 (); - -#elif HAVE_PTRACE_GETFPXREGS - if (have_ptrace_getfpxregs == -1) - { - elf_fpxregset_t fpxregs; - - if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) - { - have_ptrace_getfpxregs = 0; - have_ptrace_getregset = TRIBOOL_FALSE; - return i386_linux_read_description (X86_XSTATE_X87_MASK); - } - } -#endif - - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) - { - uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; - struct iovec iov; - - iov.iov_base = xstateregs; - iov.iov_len = sizeof (xstateregs); - - /* Check if PTRACE_GETREGSET works. */ - if (ptrace (PTRACE_GETREGSET, tid, - (unsigned int) NT_X86_XSTATE, &iov) < 0) - have_ptrace_getregset = TRIBOOL_FALSE; - else - { - have_ptrace_getregset = TRIBOOL_TRUE; - - /* Get XCR0 from XSAVE extended state. */ - xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET - / sizeof (uint64_t))]; - - m_xsave_layout = x86_fetch_xsave_layout (xcr0, x86_xsave_length ()); - } - } - - /* Check the native XCR0 only if PTRACE_GETREGSET is available. If - PTRACE_GETREGSET is not available then set xcr0_features_bits to - zero so that the "no-features" descriptions are returned by the - switches below. */ - if (have_ptrace_getregset == TRIBOOL_TRUE) - xcr0_features_bits = xcr0 & X86_XSTATE_ALL_MASK; - else - xcr0_features_bits = 0; - - if (is_64bit) - { -#ifdef __x86_64__ - return amd64_linux_read_description (xcr0_features_bits, is_x32); -#endif - } - else - { - const struct target_desc * tdesc - = i386_linux_read_description (xcr0_features_bits); + int tid = inferior_ptid.pid (); - if (tdesc == NULL) - tdesc = i386_linux_read_description (X86_XSTATE_SSE_MASK); + const char *error_msg + = _("Can't debug 64-bit process with 32-bit GDB"); - return tdesc; - } + /* Callback that is triggered the first time x86_linux_tdesc_for_tid + reads the xcr0 register. Setup other bits of state */ + auto cb = [&] (uint64_t xcr0) + { + this->m_xsave_layout + = x86_fetch_xsave_layout (xcr0, x86_xsave_length ()); + }; - gdb_assert_not_reached ("failed to return tdesc"); + return x86_linux_tdesc_for_tid (tid, &have_ptrace_getregset, cb, + error_msg, &xcr0_storage); } \f diff --git a/gdbserver/configure.srv b/gdbserver/configure.srv index 9e861a75088..7a2702d78bf 100644 --- a/gdbserver/configure.srv +++ b/gdbserver/configure.srv @@ -109,6 +109,7 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/linux-btrace.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" + srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" srv_linux_usrregs=yes srv_linux_regsets=yes srv_linux_thread_db=yes @@ -371,6 +372,7 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/linux-btrace.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" + srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" srv_tgtobj="${srv_tgtobj} nat/amd64-linux-siginfo.o" srv_linux_usrregs=yes # This is for i386 progs. srv_linux_regsets=yes diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index 54e4c9812bb..f97b0d6a1d9 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -23,6 +23,7 @@ #include "tracepoint.h" #include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" +#include "nat/x86-linux-tdesc.h" /* Defined in auto-generated file amd64-linux.c. */ void init_registers_amd64_linux (void); diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc index 2e4646f8c03..459b8055b5c 100644 --- a/gdbserver/linux-i386-ipa.cc +++ b/gdbserver/linux-i386-ipa.cc @@ -23,6 +23,7 @@ #include "tracepoint.h" #include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" +#include "nat/x86-linux-tdesc.h" /* GDB register numbers. */ diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 30d876efc5d..9bf369f8a34 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -48,6 +48,7 @@ #include "nat/x86-linux.h" #include "nat/x86-linux-dregs.h" #include "linux-x86-tdesc.h" +#include "nat/x86-linux-tdesc.h" #ifdef __x86_64__ static target_desc_up tdesc_amd64_linux_no_xml; @@ -844,32 +845,20 @@ int have_ptrace_getfpxregs = #endif ; +/* Cached xcr0 value. This is initialised the first time + x86_linux_read_description is called. */ + +static uint64_t xcr0_storage; + /* Get Linux/x86 target description from running target. */ static const struct target_desc * x86_linux_read_description (void) { - unsigned int machine; - int is_elf64; - int xcr0_features; - int tid; - static uint64_t xcr0; - static int xsave_len; - struct regset_info *regset; - - tid = lwpid_of (current_thread); - - is_elf64 = linux_pid_exe_is_elf_64_file (tid, &machine); + int tid = lwpid_of (current_thread); - if (sizeof (void *) == 4) - { - if (is_elf64 > 0) - error (_("Can't debug 64-bit process with 32-bit GDBserver")); -#ifndef __x86_64__ - else if (machine == EM_X86_64) - error (_("Can't debug x86-64 process with 32-bit GDBserver")); -#endif - } + const char *error_msg + = _("Can't debug 64-bit process with 32-bit GDBserver"); /* If we are not allowed to send an XML target description then we need to use the hard-wired target descriptions. This corresponds to GDB's @@ -879,103 +868,54 @@ x86_linux_read_description (void) generate some alternative target descriptions. */ if (!use_xml) { + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); + bool is_64bit = arch_size.is_64bit (); + bool is_x32 = arch_size.is_x32 (); + + if (sizeof (void *) == 4 && is_64bit && !is_x32) + error ("%s", error_msg); + #ifdef __x86_64__ - if (machine == EM_X86_64) + if (is_64bit && !is_x32) return tdesc_amd64_linux_no_xml.get (); else #endif return tdesc_i386_linux_no_xml.get (); } -#if !defined __x86_64__ && defined HAVE_PTRACE_GETFPXREGS - if (machine == EM_386 && have_ptrace_getfpxregs == -1) - { - elf_fpxregset_t fpxregs; - - if (ptrace (PTRACE_GETFPXREGS, tid, 0, (long) &fpxregs) < 0) - { - have_ptrace_getfpxregs = 0; - have_ptrace_getregset = TRIBOOL_FALSE; - return i386_linux_read_description (X86_XSTATE_X87); - } - else - have_ptrace_getfpxregs = 1; - } -#endif - - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) - { - uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; - struct iovec iov; - - iov.iov_base = xstateregs; - iov.iov_len = sizeof (xstateregs); - - /* Check if PTRACE_GETREGSET works. */ - if (ptrace (PTRACE_GETREGSET, tid, - (unsigned int) NT_X86_XSTATE, (long) &iov) < 0) - have_ptrace_getregset = TRIBOOL_FALSE; - else - { - have_ptrace_getregset = TRIBOOL_TRUE; - - /* Get XCR0 from XSAVE extended state. */ - xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET - / sizeof (uint64_t))]; - - /* No MPX on x32. */ - if (machine == EM_X86_64 && !is_elf64) - xcr0 &= ~X86_XSTATE_MPX; - - xsave_len = x86_xsave_length (); - - /* Use PTRACE_GETREGSET if it is available. */ - for (regset = x86_regsets; - regset->fill_function != NULL; regset++) - if (regset->get_request == PTRACE_GETREGSET) - regset->size = xsave_len; - else if (regset->type != GENERAL_REGS) - regset->size = 0; - } - } - - /* Check the native XCR0 only if PTRACE_GETREGSET is available. */ - xcr0_features = (have_ptrace_getregset == TRIBOOL_TRUE - && (xcr0 & X86_XSTATE_ALL_MASK)); - - if (xcr0_features) - i387_set_xsave_mask (xcr0, xsave_len); + /* Callback that is triggered the first time x86_linux_tdesc_for_tid + reads the xcr0 register. Setup other bits of state */ + auto cb = [] (uint64_t xcr0) + { + i387_set_xsave_mask (xcr0, x86_xsave_length ()); + }; - if (machine == EM_X86_64) - { -#ifdef __x86_64__ - const target_desc *tdesc = NULL; + /* If have_ptrace_getregset is changed to true by calling + x86_linux_tdesc_for_tid then we will perform some additional + initialisation. */ + bool have_ptrace_getregset_is_unknown + = have_ptrace_getregset == TRIBOOL_UNKNOWN; - if (xcr0_features) - { - tdesc = amd64_linux_read_description (xcr0 & X86_XSTATE_ALL_MASK, - !is_elf64); - } + const target_desc *tdesc + = x86_linux_tdesc_for_tid (tid, &have_ptrace_getregset, cb, error_msg, + &xcr0_storage); - if (tdesc == NULL) - tdesc = amd64_linux_read_description (X86_XSTATE_SSE_MASK, !is_elf64); - return tdesc; -#endif - } - else + if (have_ptrace_getregset_is_unknown + && have_ptrace_getregset == TRIBOOL_TRUE) { - const target_desc *tdesc = NULL; - - if (xcr0_features) - tdesc = i386_linux_read_description (xcr0 & X86_XSTATE_ALL_MASK); - - if (tdesc == NULL) - tdesc = i386_linux_read_description (X86_XSTATE_SSE); - - return tdesc; + int xsave_len = x86_xsave_length (); + + /* Use PTRACE_GETREGSET if it is available. */ + for (regset_info *regset = x86_regsets; + regset->fill_function != nullptr; + regset++) + if (regset->get_request == PTRACE_GETREGSET) + regset->size = xsave_len; + else if (regset->type != GENERAL_REGS) + regset->size = 0; } - gdb_assert_not_reached ("failed to return tdesc"); + return tdesc; } /* Update all the target description of all processes; a new GDB diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc index 626207fc477..9fd64d8574b 100644 --- a/gdbserver/linux-x86-tdesc.cc +++ b/gdbserver/linux-x86-tdesc.cc @@ -26,6 +26,7 @@ #include "arch/amd64.h" #endif #include "x86-tdesc.h" +#include "nat/x86-linux-tdesc.h" /* Return the right x86_linux_tdesc index for a given XCR0. Return X86_TDESC_LAST if can't find a match. */ diff --git a/gdbserver/linux-x86-tdesc.h b/gdbserver/linux-x86-tdesc.h index f9561b129ae..576aaf5e165 100644 --- a/gdbserver/linux-x86-tdesc.h +++ b/gdbserver/linux-x86-tdesc.h @@ -46,11 +46,4 @@ int amd64_get_ipa_tdesc_idx (const struct target_desc *tdesc); const struct target_desc *i386_get_ipa_tdesc (int idx); -#ifdef __x86_64__ -const struct target_desc *amd64_linux_read_description (uint64_t xcr0, - bool is_x32); -#endif - -const struct target_desc *i386_linux_read_description (uint64_t xcr0); - #endif /* GDBSERVER_LINUX_X86_TDESC_H */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv3 6/8] gdb/arch: assert that X86_XSTATE_MPX is not set for x32 2024-03-23 16:35 ` [PATCHv3 0/8] " Andrew Burgess ` (4 preceding siblings ...) 2024-03-23 16:35 ` [PATCHv3 5/8] gdb/gdbserver: share some code relating to target description creation Andrew Burgess @ 2024-03-23 16:35 ` Andrew Burgess 2024-03-23 16:35 ` [PATCHv3 7/8] gdbserver: update target description creation for x86/linux Andrew Burgess ` (3 subsequent siblings) 9 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-03-23 16:35 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, hjl.tools While trying to merge this commit: commit 4bb20a6244b7091a9a7a2ae35dfbd7e8db27550a Date: Wed Mar 20 04:13:18 2024 -0700 gdbserver: Clear X86_XSTATE_MPX bits in xcr0 on x32 With this patch series of mine: https://inbox.sourceware.org/gdb-patches/cover.1706801009.git.aburgess@redhat.com I worried that there could be other paths that could result in an xcr0 value that has X86_XSTATE_MPX set in x32 mode. As everyone eventually calls amd64_create_target_description to build their target description, I figured we could assert in here that if X86_XSTATE_MPX is set then we should not be an x32 target, this should uncover any other bugs in this area. I'm not currently able to build/run any x32 binaries, so I have no way to test this. Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=31511 --- gdb/arch/amd64.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-) diff --git a/gdb/arch/amd64.c b/gdb/arch/amd64.c index b8ff4270079..f182e686dda 100644 --- a/gdb/arch/amd64.c +++ b/gdb/arch/amd64.c @@ -66,8 +66,12 @@ amd64_create_target_description (uint64_t xcr0, bool is_x32, bool is_linux, if (xcr0 & X86_XSTATE_AVX) regnum = create_feature_i386_64bit_avx (tdesc.get (), regnum); - if ((xcr0 & X86_XSTATE_MPX) && !is_x32) - regnum = create_feature_i386_64bit_mpx (tdesc.get (), regnum); + if (xcr0 & X86_XSTATE_MPX) + { + /* MPX is not available on x32. */ + gdb_assert (!is_x32); + regnum = create_feature_i386_64bit_mpx (tdesc.get (), regnum); + } if (xcr0 & X86_XSTATE_AVX512) regnum = create_feature_i386_64bit_avx512 (tdesc.get (), regnum); -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv3 7/8] gdbserver: update target description creation for x86/linux 2024-03-23 16:35 ` [PATCHv3 0/8] " Andrew Burgess ` (5 preceding siblings ...) 2024-03-23 16:35 ` [PATCHv3 6/8] gdb/arch: assert that X86_XSTATE_MPX is not set for x32 Andrew Burgess @ 2024-03-23 16:35 ` Andrew Burgess 2024-03-23 16:35 ` [PATCHv3 8/8] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess ` (2 subsequent siblings) 9 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-03-23 16:35 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, hjl.tools, John Baldwin This commit is part of a series which aims to share more of the target description creation between GDB and gdbserver for x86/Linux. After some refactoring, the previous commit actually started to share some code, we added the shared x86_linux_tdesc_for_tid function into nat/x86-linux-tdesc.c. However, this function still relies on amd64_linux_read_description and i386_linux_read_description which are implemented separately for both gdbserver and GDB. Given that at their core, all these functions to is: 1. take an xcr0 value as input, 2. mask out some feature bits, 3. look for a cached pre-generated target description and return it if found, 4. if no cached target description is found then call either amd64_create_target_description or i386_create_target_description to create a new target description, which is then added to the cache. Return the newly created target description. The inner functions amd64_create_target_description and i386_create_target_description are already shared between GDB and gdbserver (in the gdb/arch/ directory), so the only thing that the *_read_description functions really do is add the caching layer, and it feels like this really could be shared. However, we have a small problem. On the GDB side we create target descriptions using a different set of cpu features than on the gdbserver side! This means that for the exact same target, we might get a different target description when using native GDB vs using gdbserver. This surely feels like a mistake, I would expect to get the same target description on each. The table below shows the number of possible different target descriptions that we can create on the GDB side vs on the gdbserver side for each target type: | GDB | gdbserver ------|-----|---------- i386 | 64 | 7 amd64 | 32 | 7 x32 | 16 | 7 So in theory, all I want to do is move the GDB version of *_read_description into the nat/ directory and have gdbserver use that, then both GDB and gdbserver would be able to create any of the possible target descriptions. Unfortunately it's a little more complex than that due to the in process agent (IPA). When the IPA is in use, gdbserver sends a target description index to the IPA, and the IPA uses this to find the correct target description to use. ** START OF AN ASIDE ** Back in the day I suspect this approach made perfect sense. However since this commit: commit a8806230241d201f808d856eaae4d44088117b0c Date: Thu Dec 7 17:07:01 2017 +0000 Initialize target description early in IPA I think passing the index is now more trouble than its worth. We used to pass the index, and then use that index to lookup which target description to instantiate and use. However, the above commit fixed an issue where we can't call malloc() within (certain parts of) the IPA (apparently), so instead we now pre-compute _every_ possible target description within the IPA. The index is now only used to lookup which of the (many) pre-computed target descriptions to use. It would (I think) have been easier all around if the IPA just self-inspected, figured out its own xcr0 value, and used that to create the one target description that is required. So long as the xcr0 to target description code is shared (at compile time) with gdbserver, then we can be sure that the IPA will derive the same target description as gdbserver, and we would avoid all this index passing business, which has made this commit so very, very painful. ** END OF AN ASIDE ** Currently then for x86/linux, gdbserver sends a number between 0 and 7 to the IPA, and the IPA uses this to create a target description. However, I am proposing that gdbserver should now create one of (up to) 64 different target descriptions for i386, so this 0 to 7 index isn't going to be good enough any more (amd64 and x32 have slightly fewer possible target descriptions, but still more than 8, so the problem is the same). For a while I wondered if I was going to have to try and find some backward compatible solution for this mess. But after seeing how lightly the IPA is actually documented, I wonder if it is not the case that there is a tight coupling between a version of gdbserver and a version of the IPA? At least I'm hoping so. In this commit I have thrown out the old IPA target description index numbering scheme, and switched to a completely new numbering scheme. Instead of the index that is passed being arbitrary, the index is instead calculated from the set of cpu features that are present on the target. Within the IPA we can then reverse this logic to recreate the xcr0 value based on the index, and from the xcr0 value we can create the correct target description. With the gdbserver to IPA numbering scheme issue resolved I have then update the gdbserver versions of amd64_linux_read_description and i386_linux_read_description so that they create target descriptions using the same set of cpu features as GDB itself. After this gdbserver should now always come up with the same target description as GDB does on any x86/Linux target. This commit does not introduce any new code sharing between GDB and gdbserver as previous commits in this series does. Instead this commit is all about bringing GDB and gdbserver into alignment functionally so that the next commit can merge the GDB and gdbserver versions of these functions. Approved-By: John Baldwin <jhb@FreeBSD.org> --- gdbserver/linux-amd64-ipa.cc | 43 +---- gdbserver/linux-i386-ipa.cc | 23 +-- gdbserver/linux-x86-low.cc | 15 +- gdbserver/linux-x86-tdesc.cc | 315 +++++++++++++++++++++++++---------- gdbserver/linux-x86-tdesc.h | 49 +++--- 5 files changed, 277 insertions(+), 168 deletions(-) diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index f97b0d6a1d9..42de7d2da16 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -169,47 +169,21 @@ supply_static_tracepoint_registers (struct regcache *regcache, #endif /* HAVE_UST */ -#if !defined __ILP32__ -/* Map the tdesc index to xcr0 mask. */ -static uint64_t idx2mask[X86_TDESC_LAST] = { - X86_XSTATE_X87_MASK, - X86_XSTATE_SSE_MASK, - X86_XSTATE_AVX_MASK, - X86_XSTATE_MPX_MASK, - X86_XSTATE_AVX_MPX_MASK, - X86_XSTATE_AVX_AVX512_MASK, - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, -}; -#endif - /* Return target_desc to use for IPA, given the tdesc index passed by gdbserver. */ const struct target_desc * get_ipa_tdesc (int idx) { - if (idx >= X86_TDESC_LAST) - { - internal_error ("unknown ipa tdesc index: %d", idx); - } + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); #if defined __ILP32__ - switch (idx) - { - case X86_TDESC_SSE: - return amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); - case X86_TDESC_AVX: - return amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); - case X86_TDESC_AVX_AVX512: - return amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, true); - default: - break; - } + bool is_x32 = true; #else - return amd64_linux_read_description (idx2mask[idx], false); + bool is_x32 = false; #endif - internal_error ("unknown ipa tdesc index: %d", idx); + return amd64_linux_read_description (xcr0, is_x32); } /* Allocate buffer for the jump pads. The branch instruction has a @@ -277,11 +251,10 @@ void initialize_low_tracepoint (void) { #if defined __ILP32__ - amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); - amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); - amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, true); + for (auto i = 0; i < x86_linux_x32_ipa_tdesc_count (); i++) + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), true); #else - for (auto i = 0; i < X86_TDESC_LAST; i++) - amd64_linux_read_description (idx2mask[i], false); + for (auto i = 0; i < x86_linux_amd64_ipa_tdesc_count (); i++) + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), false); #endif } diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc index 459b8055b5c..246bcb9813b 100644 --- a/gdbserver/linux-i386-ipa.cc +++ b/gdbserver/linux-i386-ipa.cc @@ -246,28 +246,15 @@ initialize_fast_tracepoint_trampoline_buffer (void) } } -/* Map the tdesc index to xcr0 mask. */ -static uint64_t idx2mask[X86_TDESC_LAST] = { - X86_XSTATE_X87_MASK, - X86_XSTATE_SSE_MASK, - X86_XSTATE_AVX_MASK, - X86_XSTATE_MPX_MASK, - X86_XSTATE_AVX_MPX_MASK, - X86_XSTATE_AVX_AVX512_MASK, - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, -}; - /* Return target_desc to use for IPA, given the tdesc index passed by gdbserver. */ const struct target_desc * get_ipa_tdesc (int idx) { - if (idx >= X86_TDESC_LAST) - { - internal_error ("unknown ipa tdesc index: %d", idx); - } - return i386_linux_read_description (idx2mask[idx]); + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); + + return i386_linux_read_description (xcr0); } /* Allocate buffer for the jump pads. On i386, we can reach an arbitrary @@ -289,6 +276,6 @@ void initialize_low_tracepoint (void) { initialize_fast_tracepoint_trampoline_buffer (); - for (auto i = 0; i < X86_TDESC_LAST; i++) - i386_linux_read_description (idx2mask[i]); + for (auto i = 0; i < x86_linux_i386_ipa_tdesc_count (); i++) + i386_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i)); } diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 9bf369f8a34..6dbefdb2f26 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -2892,14 +2892,17 @@ x86_target::get_ipa_tdesc_idx () struct regcache *regcache = get_thread_regcache (current_thread, 0); const struct target_desc *tdesc = regcache->tdesc; + if (!use_xml) + { + if (tdesc == tdesc_i386_linux_no_xml.get () #ifdef __x86_64__ - return amd64_get_ipa_tdesc_idx (tdesc); -#endif - - if (tdesc == tdesc_i386_linux_no_xml.get ()) - return X86_TDESC_SSE; + || tdesc == tdesc_amd64_linux_no_xml.get () +#endif /* __x86_64__ */ + ) + return x86_linux_xcr0_to_tdesc_idx (X86_XSTATE_SSE_MASK); + } - return i386_get_ipa_tdesc_idx (tdesc); + return x86_linux_xcr0_to_tdesc_idx (xcr0_storage); } /* The linux target ops object. */ diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc index 9fd64d8574b..87cf368a336 100644 --- a/gdbserver/linux-x86-tdesc.cc +++ b/gdbserver/linux-x86-tdesc.cc @@ -28,96 +28,277 @@ #include "x86-tdesc.h" #include "nat/x86-linux-tdesc.h" -/* Return the right x86_linux_tdesc index for a given XCR0. Return - X86_TDESC_LAST if can't find a match. */ +/* A structure used to describe a single cpu feature that might, or might + not, be checked for when creating a target description for one of i386, + amd64, or x32. */ -static enum x86_linux_tdesc -xcr0_to_tdesc_idx (uint64_t xcr0, bool is_x32) +struct x86_tdesc_feature { + /* The cpu feature mask. This is a mask against an xcr0 value. */ + uint64_t feature; + + /* Is this feature checked when creating an i386 target description. */ + bool is_i386; + + /* Is this feature checked when creating an amd64 target description. */ + bool is_amd64; + + /* Is this feature checked when creating an x32 target description. */ + bool is_x32; +}; + +/* A constant table that describes all of the cpu features that are + checked when building a target description for i386, amd64, or x32. */ + +static constexpr x86_tdesc_feature x86_linux_all_tdesc_features[] = { + /* Feature, i386, amd64, x32. */ + { X86_XSTATE_PKRU, true, true, true }, + { X86_XSTATE_AVX512, true, true, true }, + { X86_XSTATE_AVX, true, true, true }, + { X86_XSTATE_MPX, true, true, false }, + { X86_XSTATE_SSE, true, false, false }, + { X86_XSTATE_X87, true, false, false } +}; + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an i386 target description. */ + +static constexpr uint64_t +x86_linux_i386_tdesc_feature_mask () { - if (xcr0 & X86_XSTATE_PKRU) - { - if (is_x32) - { - /* No x32 MPX and PKU, fall back to avx_avx512. */ - return X86_TDESC_AVX_AVX512; - } - else - return X86_TDESC_AVX_MPX_AVX512_PKU; - } - else if (xcr0 & X86_XSTATE_AVX512) - return X86_TDESC_AVX_AVX512; - else if ((xcr0 & X86_XSTATE_AVX_MPX_MASK) == X86_XSTATE_AVX_MPX_MASK) + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_i386) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an amd64 target description. */ + +static constexpr uint64_t +x86_linux_amd64_tdesc_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_amd64) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an x32 target description. */ + +static constexpr uint64_t +x86_linux_x32_tdesc_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_x32) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an i386 target description. */ + +static constexpr int +x86_linux_i386_tdesc_count () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_i386) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an amd64 target description. */ + +static constexpr int +x86_linux_amd64_tdesc_count () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_amd64) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an x32 target description. */ + +static constexpr int +x86_linux_x32_tdesc_count () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_x32) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +#ifdef IN_PROCESS_AGENT + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_amd64_ipa_tdesc_count () +{ + return x86_linux_amd64_tdesc_count (); +} + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_x32_ipa_tdesc_count () +{ + return x86_linux_x32_tdesc_count (); +} + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_i386_ipa_tdesc_count () +{ + return x86_linux_i386_tdesc_count (); +} + +#endif /* IN_PROCESS_AGENT */ + +/* Convert an xcr0 value into an integer. The integer will be passed to + the in-process-agent where it will then be passed to + x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ + +int +x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0) +{ + /* The following table shows which features are checked for when creating + the target descriptions (see nat/x86-linux-tdesc.c), the feature order + represents the bit order within the generated index number. + + i386 | x87 sse mpx avx avx512 pkru + amd64 | mpx avx avx512 pkru + i32 | avx avx512 pkru + + The features are ordered so that for each mode (i386, amd64, i32) the + generated index will form a continuous range. */ + + int idx = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) { - if (is_x32) /* No MPX on x32. */ - return X86_TDESC_AVX; - else - return X86_TDESC_AVX_MPX; + if ((xcr0 & x86_linux_all_tdesc_features[i].feature) != 0) + idx |= (1 << i); } - else if (xcr0 & X86_XSTATE_MPX) + + return idx; +} + + +#ifdef IN_PROCESS_AGENT + +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +uint64_t +x86_linux_tdesc_idx_to_xcr0 (int idx) +{ + uint64_t xcr0 = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) { - if (is_x32) /* No MPX on x32. */ - return X86_TDESC_AVX; - else - return X86_TDESC_MPX; + if ((idx & (1 << i)) != 0) + xcr0 |= x86_linux_all_tdesc_features[i].feature; } - else if (xcr0 & X86_XSTATE_AVX) - return X86_TDESC_AVX; - else if (xcr0 & X86_XSTATE_SSE) - return X86_TDESC_SSE; - else if (xcr0 & X86_XSTATE_X87) - return X86_TDESC_MMX; - else - return X86_TDESC_LAST; + + return xcr0; } +#endif /* IN_PROCESS_AGENT */ + #if defined __i386__ || !defined IN_PROCESS_AGENT -static struct target_desc *i386_tdescs[X86_TDESC_LAST] = { }; +/* A cache of all possible i386 target descriptions. */ -/* Return the target description according to XCR0. */ +static struct target_desc *i386_tdescs[x86_linux_i386_tdesc_count ()] = { }; + +/* See nat/x86-linux-tdesc.h. */ const struct target_desc * i386_linux_read_description (uint64_t xcr0) { - enum x86_linux_tdesc idx = xcr0_to_tdesc_idx (xcr0, false); + xcr0 &= x86_linux_i386_tdesc_feature_mask (); + int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - if (idx == X86_TDESC_LAST) - return NULL; + gdb_assert (idx >= 0 && idx < x86_linux_i386_tdesc_count ()); - struct target_desc **tdesc = &i386_tdescs[idx]; + target_desc **tdesc = &i386_tdescs[idx]; - if (*tdesc == NULL) + if (*tdesc == nullptr) { *tdesc = i386_create_target_description (xcr0, true, false); init_target_desc (*tdesc, i386_expedite_regs); } - return *tdesc;; + return *tdesc; } #endif #ifdef __x86_64__ -static target_desc *amd64_tdescs[X86_TDESC_LAST] = { }; -static target_desc *x32_tdescs[X86_TDESC_LAST] = { }; +/* A cache of all possible amd64 target descriptions. */ + +static target_desc *amd64_tdescs[x86_linux_amd64_tdesc_count ()] = { }; + +/* A cache of all possible x32 target descriptions. */ + +static target_desc *x32_tdescs[x86_linux_x32_tdesc_count ()] = { }; + +/* See nat/x86-linux-tdesc.h. */ const struct target_desc * amd64_linux_read_description (uint64_t xcr0, bool is_x32) { - enum x86_linux_tdesc idx = xcr0_to_tdesc_idx (xcr0, is_x32); + if (is_x32) + xcr0 &= x86_linux_x32_tdesc_feature_mask (); + else + xcr0 &= x86_linux_amd64_tdesc_feature_mask (); + + int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - if (idx == X86_TDESC_LAST) - return NULL; + if (is_x32) + gdb_assert (idx >= 0 && idx < x86_linux_x32_tdesc_count ()); + else + gdb_assert (idx >= 0 && idx < x86_linux_amd64_tdesc_count ()); - struct target_desc **tdesc = NULL; + target_desc **tdesc = nullptr; if (is_x32) tdesc = &x32_tdescs[idx]; else tdesc = &amd64_tdescs[idx]; - if (*tdesc == NULL) + if (*tdesc == nullptr) { *tdesc = amd64_create_target_description (xcr0, is_x32, true, true); @@ -127,39 +308,3 @@ amd64_linux_read_description (uint64_t xcr0, bool is_x32) } #endif - -#ifndef IN_PROCESS_AGENT - -int -i386_get_ipa_tdesc_idx (const struct target_desc *tdesc) -{ - for (int i = 0; i < X86_TDESC_LAST; i++) - { - if (tdesc == i386_tdescs[i]) - return i; - } - - /* If none tdesc is found, return the one with minimum features. */ - return X86_TDESC_MMX; -} - -#if defined __x86_64__ -int -amd64_get_ipa_tdesc_idx (const struct target_desc *tdesc) -{ - for (int i = 0; i < X86_TDESC_LAST; i++) - { - if (tdesc == amd64_tdescs[i]) - return i; - } - for (int i = 0; i < X86_TDESC_LAST; i++) - { - if (tdesc == x32_tdescs[i]) - return i; - } - - return X86_TDESC_SSE; -} - -#endif -#endif diff --git a/gdbserver/linux-x86-tdesc.h b/gdbserver/linux-x86-tdesc.h index 576aaf5e165..70456e4be44 100644 --- a/gdbserver/linux-x86-tdesc.h +++ b/gdbserver/linux-x86-tdesc.h @@ -21,29 +21,30 @@ #ifndef GDBSERVER_LINUX_X86_TDESC_H #define GDBSERVER_LINUX_X86_TDESC_H -/* Note: since IPA obviously knows what ABI it's running on (i386 vs x86_64 - vs x32), it's sufficient to pass only the register set here. This, - together with the ABI known at IPA compile time, maps to a tdesc. */ - -enum x86_linux_tdesc { - X86_TDESC_MMX = 0, - X86_TDESC_SSE = 1, - X86_TDESC_AVX = 2, - X86_TDESC_MPX = 3, - X86_TDESC_AVX_MPX = 4, - X86_TDESC_AVX_AVX512 = 5, - X86_TDESC_AVX_MPX_AVX512_PKU = 6, - X86_TDESC_LAST = 7, -}; - -#if defined __i386__ || !defined IN_PROCESS_AGENT -int i386_get_ipa_tdesc_idx (const struct target_desc *tdesc); -#endif - -#if defined __x86_64__ && !defined IN_PROCESS_AGENT -int amd64_get_ipa_tdesc_idx (const struct target_desc *tdesc); -#endif - -const struct target_desc *i386_get_ipa_tdesc (int idx); +/* Convert an xcr0 value into an integer. The integer will be passed to + the in-process-agent where it will then be passed to + x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ + +extern int x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0); + +#ifdef IN_PROCESS_AGENT + +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +extern uint64_t x86_linux_tdesc_idx_to_xcr0 (int idx); + +/* Within the in-process-agent we need to pre-initialise all of the target + descriptions, to do this we need to know how many target descriptions + there are for each different target type. These functions return the + target description count for the relevant target. */ + +extern int x86_linux_amd64_ipa_tdesc_count (); +extern int x86_linux_x32_ipa_tdesc_count (); +extern int x86_linux_i386_ipa_tdesc_count (); + + +#endif /* IN_PROCESS_AGENT */ #endif /* GDBSERVER_LINUX_X86_TDESC_H */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv3 8/8] gdb/gdbserver: share x86/linux tdesc caching 2024-03-23 16:35 ` [PATCHv3 0/8] " Andrew Burgess ` (6 preceding siblings ...) 2024-03-23 16:35 ` [PATCHv3 7/8] gdbserver: update target description creation for x86/linux Andrew Burgess @ 2024-03-23 16:35 ` Andrew Burgess 2024-03-26 12:17 ` Andrew Burgess 2024-03-25 17:20 ` [PATCHv3 0/8] x86/Linux Target Description Changes Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 00/10] " Andrew Burgess 9 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-03-23 16:35 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, hjl.tools, John Baldwin This commit builds on the previous series of commits to share the target description caching code between GDB and gdbserver for x86/Linux targets. The objective of this commit is to move the four functions (2 each of) i386_linux_read_description and amd64_linux_read_description into gdb/nat/x86-linux-tdesc.c and combine them so we have just a single copy of each. Then both GDB and gdbserver will link against these shared functions. It is worth reading the description of the previous commit to see why this merging is not as simple as it seems: on the gdbserver side we actually have two users of these functions, gdbserver itself, and the in process agent (IPA). However, the previous commit streamlined the gdbserver code, and so now it is simple to move the two functions along with all their support functions from the gdbserver directory into the gdb/nat/ directory, and then GDB is fine to call these functions. One small curiosity with this patch is the function x86_linux_post_init_tdesc. On the gdbserver side the two functions amd64_linux_read_description and i386_linux_read_description have some functionality that is not present on the GDB side, that is some additional configuration that is performed as each target description is created to setup the expedited registers. To support this I've added the function x86_linux_post_init_tdesc. This function is called from the two *_linux_read_description functions, but is implemented separately for GDB and gdbserver. This does mean adding back some non-shared code when this whole series has been about sharing code, but now the only non-shared bit is the single line that is actually different between GDB and gdbserver, all the rest, which is identical, is now shared. I did need to add a new rule to the gdbserver Makefile, this is to allow the nat/x86-linux-tdesc.c file to be compiled for the IPA. Approved-By: John Baldwin <jhb@FreeBSD.org> --- gdb/amd64-linux-tdep.c | 31 ---- gdb/i386-linux-tdep.c | 30 +--- gdb/nat/x86-linux-tdesc.c | 287 +++++++++++++++++++++++++++++++++++ gdb/nat/x86-linux-tdesc.h | 56 ++++++- gdbserver/Makefile.in | 4 + gdbserver/configure.srv | 2 + gdbserver/linux-amd64-ipa.cc | 1 - gdbserver/linux-i386-ipa.cc | 1 - gdbserver/linux-x86-low.cc | 1 - gdbserver/linux-x86-tdesc.cc | 285 +--------------------------------- gdbserver/linux-x86-tdesc.h | 50 ------ 11 files changed, 355 insertions(+), 393 deletions(-) delete mode 100644 gdbserver/linux-x86-tdesc.h diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c index 7e0900dc6f9..e149034de21 100644 --- a/gdb/amd64-linux-tdep.c +++ b/gdb/amd64-linux-tdep.c @@ -1579,37 +1579,6 @@ amd64_linux_record_signal (struct gdbarch *gdbarch, return 0; } -const target_desc * -amd64_linux_read_description (uint64_t xcr0_features_bit, bool is_x32) -{ - static target_desc *amd64_linux_tdescs \ - [2/*AVX*/][2/*MPX*/][2/*AVX512*/][2/*PKRU*/] = {}; - static target_desc *x32_linux_tdescs \ - [2/*AVX*/][2/*AVX512*/][2/*PKRU*/] = {}; - - target_desc **tdesc; - - if (is_x32) - { - tdesc = &x32_linux_tdescs[(xcr0_features_bit & X86_XSTATE_AVX) ? 1 : 0 ] - [(xcr0_features_bit & X86_XSTATE_AVX512) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_PKRU) ? 1 : 0]; - } - else - { - tdesc = &amd64_linux_tdescs[(xcr0_features_bit & X86_XSTATE_AVX) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_MPX) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_AVX512) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_PKRU) ? 1 : 0]; - } - - if (*tdesc == NULL) - *tdesc = amd64_create_target_description (xcr0_features_bit, is_x32, - true, true); - - return *tdesc; -} - /* Get Linux/x86 target description from core dump. */ static const struct target_desc * diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index f5f7a36bf36..ed1442236e8 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -681,29 +681,12 @@ i386_linux_core_read_x86_xsave_layout (struct gdbarch *gdbarch, layout) != 0; } -/* See i386-linux-tdep.h. */ +/* See nat/x86-linux-tdesc.h. */ -const struct target_desc * -i386_linux_read_description (uint64_t xcr0) +void +x86_linux_post_init_tdesc (target_desc *tdesc, bool is_64bit) { - if (xcr0 == 0) - return NULL; - - static struct target_desc *i386_linux_tdescs \ - [2/*X87*/][2/*SSE*/][2/*AVX*/][2/*MPX*/][2/*AVX512*/][2/*PKRU*/] = {}; - struct target_desc **tdesc; - - tdesc = &i386_linux_tdescs[(xcr0 & X86_XSTATE_X87) ? 1 : 0] - [(xcr0 & X86_XSTATE_SSE) ? 1 : 0] - [(xcr0 & X86_XSTATE_AVX) ? 1 : 0] - [(xcr0 & X86_XSTATE_MPX) ? 1 : 0] - [(xcr0 & X86_XSTATE_AVX512) ? 1 : 0] - [(xcr0 & X86_XSTATE_PKRU) ? 1 : 0]; - - if (*tdesc == NULL) - *tdesc = i386_create_target_description (xcr0, true, false); - - return *tdesc; + /* Nothing. */ } /* Get Linux/x86 target description from core dump. */ @@ -716,7 +699,10 @@ i386_linux_core_read_description (struct gdbarch *gdbarch, /* Linux/i386. */ x86_xsave_layout layout; uint64_t xcr0 = i386_linux_core_read_xsave_info (abfd, layout); - const struct target_desc *tdesc = i386_linux_read_description (xcr0); + + const struct target_desc *tdesc; + if (xcr0 != 0) + tdesc = i386_linux_read_description (xcr0); if (tdesc != NULL) return tdesc; diff --git a/gdb/nat/x86-linux-tdesc.c b/gdb/nat/x86-linux-tdesc.c index be7014d2b5f..c438dfae84f 100644 --- a/gdb/nat/x86-linux-tdesc.c +++ b/gdb/nat/x86-linux-tdesc.c @@ -38,6 +38,8 @@ #include <sys/user.h> #include <sys/user.h> +#ifndef IN_PROCESS_AGENT + /* See nat/x86-linux-tdesc.h. */ const target_desc * @@ -122,3 +124,288 @@ x86_linux_tdesc_for_tid (int tid, enum tribool *have_ptrace_getregset, gdb_assert_not_reached ("failed to return tdesc"); } + +#endif /* !IN_PROCESS_AGENT */ + + + +/* A structure used to describe a single cpu feature that might, or might + not, be checked for when creating a target description for one of i386, + amd64, or x32. */ + +struct x86_tdesc_feature { + /* The cpu feature mask. This is a mask against an xcr0 value. */ + uint64_t feature; + + /* Is this feature checked when creating an i386 target description. */ + bool is_i386; + + /* Is this feature checked when creating an amd64 target description. */ + bool is_amd64; + + /* Is this feature checked when creating an x32 target description. */ + bool is_x32; +}; + +/* A constant table that describes all of the cpu features that are + checked when building a target description for i386, amd64, or x32. */ + +static constexpr x86_tdesc_feature x86_linux_all_tdesc_features[] = { + /* Feature, i386, amd64, x32. */ + { X86_XSTATE_PKRU, true, true, true }, + { X86_XSTATE_AVX512, true, true, true }, + { X86_XSTATE_AVX, true, true, true }, + { X86_XSTATE_MPX, true, true, false }, + { X86_XSTATE_SSE, true, false, false }, + { X86_XSTATE_X87, true, false, false } +}; + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an i386 target description. */ + +static constexpr uint64_t +x86_linux_i386_tdesc_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_i386) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an amd64 target description. */ + +static constexpr uint64_t +x86_linux_amd64_tdesc_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_amd64) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an x32 target description. */ + +static constexpr uint64_t +x86_linux_x32_tdesc_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_x32) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an i386 target description. */ + +static constexpr int +x86_linux_i386_tdesc_count () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_i386) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an amd64 target description. */ + +static constexpr int +x86_linux_amd64_tdesc_count () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_amd64) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an x32 target description. */ + +static constexpr int +x86_linux_x32_tdesc_count () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_x32) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +#ifdef IN_PROCESS_AGENT + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_amd64_ipa_tdesc_count () +{ + return x86_linux_amd64_tdesc_count (); +} + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_x32_ipa_tdesc_count () +{ + return x86_linux_x32_tdesc_count (); +} + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_i386_ipa_tdesc_count () +{ + return x86_linux_i386_tdesc_count (); +} + +#endif /* IN_PROCESS_AGENT */ + +/* Convert an xcr0 value into an integer. The integer will be passed to + the in-process-agent where it will then be passed to + x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ + +int +x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0) +{ + /* The following table shows which features are checked for when creating + the target descriptions (see nat/x86-linux-tdesc.c), the feature order + represents the bit order within the generated index number. + + i386 | x87 sse mpx avx avx512 pkru + amd64 | mpx avx avx512 pkru + i32 | avx avx512 pkru + + The features are ordered so that for each mode (i386, amd64, i32) the + generated index will form a continuous range. */ + + int idx = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) + { + if ((xcr0 & x86_linux_all_tdesc_features[i].feature) != 0) + idx |= (1 << i); + } + + return idx; +} + + +#ifdef IN_PROCESS_AGENT + +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +uint64_t +x86_linux_tdesc_idx_to_xcr0 (int idx) +{ + uint64_t xcr0 = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) + { + if ((idx & (1 << i)) != 0) + xcr0 |= x86_linux_all_tdesc_features[i].feature; + } + + return xcr0; +} + +#endif /* IN_PROCESS_AGENT */ + +#if defined __i386__ || !defined IN_PROCESS_AGENT + +/* A cache of all possible i386 target descriptions. */ + +static struct target_desc *i386_tdescs[x86_linux_i386_tdesc_count ()] = { }; + +/* See nat/x86-linux-tdesc.h. */ + +const struct target_desc * +i386_linux_read_description (uint64_t xcr0) +{ + xcr0 &= x86_linux_i386_tdesc_feature_mask (); + int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); + + gdb_assert (idx >= 0 && idx < x86_linux_i386_tdesc_count ()); + + target_desc **tdesc = &i386_tdescs[idx]; + + if (*tdesc == nullptr) + { + *tdesc = i386_create_target_description (xcr0, true, false); + + x86_linux_post_init_tdesc (*tdesc, false); + } + + return *tdesc; +} +#endif + +#ifdef __x86_64__ + +/* A cache of all possible amd64 target descriptions. */ + +static target_desc *amd64_tdescs[x86_linux_amd64_tdesc_count ()] = { }; + +/* A cache of all possible x32 target descriptions. */ + +static target_desc *x32_tdescs[x86_linux_x32_tdesc_count ()] = { }; + +/* See nat/x86-linux-tdesc.h. */ + +const struct target_desc * +amd64_linux_read_description (uint64_t xcr0, bool is_x32) +{ + if (is_x32) + xcr0 &= x86_linux_x32_tdesc_feature_mask (); + else + xcr0 &= x86_linux_amd64_tdesc_feature_mask (); + + int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); + + if (is_x32) + gdb_assert (idx >= 0 && idx < x86_linux_x32_tdesc_count ()); + else + gdb_assert (idx >= 0 && idx < x86_linux_amd64_tdesc_count ()); + + target_desc **tdesc = nullptr; + + if (is_x32) + tdesc = &x32_tdescs[idx]; + else + tdesc = &amd64_tdescs[idx]; + + if (*tdesc == nullptr) + { + *tdesc = amd64_create_target_description (xcr0, is_x32, true, true); + + x86_linux_post_init_tdesc (*tdesc, true); + } + return *tdesc; +} + +#endif diff --git a/gdb/nat/x86-linux-tdesc.h b/gdb/nat/x86-linux-tdesc.h index 3727a8bf95e..648fe0ed0b2 100644 --- a/gdb/nat/x86-linux-tdesc.h +++ b/gdb/nat/x86-linux-tdesc.h @@ -24,6 +24,8 @@ struct target_desc; +#ifndef IN_PROCESS_AGENT + /* Return the target description for Linux thread TID. When *HAVE_PTRACE_GETREGSET is TRIBOOL_UNKNOWN then the current value of @@ -57,19 +59,57 @@ x86_linux_tdesc_for_tid (int tid, enum tribool *have_ptrace_getregset, gdb::function_view<void (uint64_t)> xcr0_init_cb, const char *error_msg, uint64_t *xcr0_storage); +#endif /* !IN_PROCESS_AGENT */ + #ifdef __x86_64__ -/* Return the right amd64-linux target descriptions according to - XCR0_FEATURES_BIT and IS_X32. This is implemented separately in both - GDB and gdbserver. */ +/* Return the AMD64 target descriptions corresponding to XCR0 and IS_X32. */ -extern const target_desc *amd64_linux_read_description - (uint64_t xcr0_features_bit, bool is_x32); +extern const target_desc *amd64_linux_read_description (uint64_t xcr0, + bool is_x32); -#endif +#endif /* __x86_64__ */ + +/* Return the i386 target description corresponding to XCR0. */ -/* Return the target description according to XCR0. This is implemented - separately in both GDB and gdbserver. */ extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); +/* This function is called from amd64_linux_read_description and + i386_linux_read_description after a new target description has been + created, TDESC is the new target description, IS_64BIT will be true + when called from amd64_linux_read_description, otherwise IS_64BIT will + be false. If the *_linux_read_description functions found a cached + target description then this function will not be called. + + Both GDB and gdbserver have their own implementations of this + function. */ + +extern void x86_linux_post_init_tdesc (target_desc *tdesc, bool is_64bit); + +/* Convert an xcr0 value into an integer. The integer will be passed to + the in-process-agent where it will then be passed to + x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ + +extern int x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0); + + +#ifdef IN_PROCESS_AGENT + +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +extern uint64_t x86_linux_tdesc_idx_to_xcr0 (int idx); + +/* Within the in-process-agent we need to pre-initialise all of the target + descriptions, to do this we need to know how many target descriptions + there are for each different target type. These functions return the + target description count for the relevant target. */ + +extern int x86_linux_amd64_ipa_tdesc_count (); +extern int x86_linux_x32_ipa_tdesc_count (); +extern int x86_linux_i386_ipa_tdesc_count (); + +#endif /* IN_PROCESS_AGENT */ + #endif /* NAT_X86_LINUX_TDESC_H */ diff --git a/gdbserver/Makefile.in b/gdbserver/Makefile.in index c7120895a26..bd6f68e7f2c 100644 --- a/gdbserver/Makefile.in +++ b/gdbserver/Makefile.in @@ -519,6 +519,10 @@ gdbsupport/%-ipa.o: ../gdbsupport/%.cc $(IPAGENT_COMPILE) $< $(POSTCOMPILE) +nat/%-ipa.o: ../gdb/nat/%.c + $(IPAGENT_COMPILE) $< + $(POSTCOMPILE) + %-ipa.o: ../gdb/%.c $(IPAGENT_COMPILE) -x c++ $< $(POSTCOMPILE) diff --git a/gdbserver/configure.srv b/gdbserver/configure.srv index 7a2702d78bf..36a457f21ed 100644 --- a/gdbserver/configure.srv +++ b/gdbserver/configure.srv @@ -116,6 +116,7 @@ case "${gdbserver_host}" in srv_linux_btrace=yes ipa_obj="linux-i386-ipa.o linux-x86-tdesc-ipa.o" ipa_obj="${ipa_obj} arch/i386-ipa.o" + ipa_obj="${ipa_obj} nat/x86-linux-tdesc-ipa.o" ;; i[34567]86-*-mingw*) srv_regobj="" srv_tgtobj="x86-low.o nat/x86-dregs.o win32-low.o" @@ -380,6 +381,7 @@ case "${gdbserver_host}" in srv_linux_btrace=yes ipa_obj="linux-amd64-ipa.o linux-x86-tdesc-ipa.o" ipa_obj="${ipa_obj} arch/amd64-ipa.o" + ipa_obj="${ipa_obj} nat/x86-linux-tdesc-ipa.o" ;; x86_64-*-mingw*) srv_regobj="" srv_tgtobj="x86-low.o nat/x86-dregs.o" diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index 42de7d2da16..afa6aab5bab 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -21,7 +21,6 @@ #include "server.h" #include <sys/mman.h> #include "tracepoint.h" -#include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" #include "nat/x86-linux-tdesc.h" diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc index 246bcb9813b..f6d26f784af 100644 --- a/gdbserver/linux-i386-ipa.cc +++ b/gdbserver/linux-i386-ipa.cc @@ -21,7 +21,6 @@ #include "server.h" #include <sys/mman.h> #include "tracepoint.h" -#include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" #include "nat/x86-linux-tdesc.h" diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 6dbefdb2f26..7134d20dd7f 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -47,7 +47,6 @@ #include "nat/linux-nat.h" #include "nat/x86-linux.h" #include "nat/x86-linux-dregs.h" -#include "linux-x86-tdesc.h" #include "nat/x86-linux-tdesc.h" #ifdef __x86_64__ diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc index 87cf368a336..39d1d51b8c0 100644 --- a/gdbserver/linux-x86-tdesc.cc +++ b/gdbserver/linux-x86-tdesc.cc @@ -19,292 +19,19 @@ #include "server.h" #include "tdesc.h" -#include "linux-x86-tdesc.h" -#include "arch/i386.h" -#include "gdbsupport/x86-xstate.h" -#ifdef __x86_64__ -#include "arch/amd64.h" -#endif #include "x86-tdesc.h" #include "nat/x86-linux-tdesc.h" -/* A structure used to describe a single cpu feature that might, or might - not, be checked for when creating a target description for one of i386, - amd64, or x32. */ - -struct x86_tdesc_feature { - /* The cpu feature mask. This is a mask against an xcr0 value. */ - uint64_t feature; - - /* Is this feature checked when creating an i386 target description. */ - bool is_i386; - - /* Is this feature checked when creating an amd64 target description. */ - bool is_amd64; - - /* Is this feature checked when creating an x32 target description. */ - bool is_x32; -}; - -/* A constant table that describes all of the cpu features that are - checked when building a target description for i386, amd64, or x32. */ - -static constexpr x86_tdesc_feature x86_linux_all_tdesc_features[] = { - /* Feature, i386, amd64, x32. */ - { X86_XSTATE_PKRU, true, true, true }, - { X86_XSTATE_AVX512, true, true, true }, - { X86_XSTATE_AVX, true, true, true }, - { X86_XSTATE_MPX, true, true, false }, - { X86_XSTATE_SSE, true, false, false }, - { X86_XSTATE_X87, true, false, false } -}; - -/* Return a compile time constant which is a mask of all the cpu features - that are checked for when building an i386 target description. */ - -static constexpr uint64_t -x86_linux_i386_tdesc_feature_mask () -{ - uint64_t mask = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_i386) - mask |= entry.feature; - - return mask; -} - -/* Return a compile time constant which is a mask of all the cpu features - that are checked for when building an amd64 target description. */ - -static constexpr uint64_t -x86_linux_amd64_tdesc_feature_mask () -{ - uint64_t mask = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_amd64) - mask |= entry.feature; - - return mask; -} - -/* Return a compile time constant which is a mask of all the cpu features - that are checked for when building an x32 target description. */ - -static constexpr uint64_t -x86_linux_x32_tdesc_feature_mask () -{ - uint64_t mask = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_x32) - mask |= entry.feature; - - return mask; -} - -/* Return a compile time constant which is a count of the number of cpu - features that are checked for when building an i386 target description. */ - -static constexpr int -x86_linux_i386_tdesc_count () -{ - uint64_t count = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_i386) - ++count; - - gdb_assert (count > 0); - - return (1 << count); -} - -/* Return a compile time constant which is a count of the number of cpu - features that are checked for when building an amd64 target description. */ - -static constexpr int -x86_linux_amd64_tdesc_count () -{ - uint64_t count = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_amd64) - ++count; - - gdb_assert (count > 0); - - return (1 << count); -} - -/* Return a compile time constant which is a count of the number of cpu - features that are checked for when building an x32 target description. */ - -static constexpr int -x86_linux_x32_tdesc_count () -{ - uint64_t count = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_x32) - ++count; - - gdb_assert (count > 0); - - return (1 << count); -} - -#ifdef IN_PROCESS_AGENT - -/* See linux-x86-tdesc.h. */ - -int -x86_linux_amd64_ipa_tdesc_count () -{ - return x86_linux_amd64_tdesc_count (); -} - -/* See linux-x86-tdesc.h. */ - -int -x86_linux_x32_ipa_tdesc_count () -{ - return x86_linux_x32_tdesc_count (); -} - -/* See linux-x86-tdesc.h. */ - -int -x86_linux_i386_ipa_tdesc_count () -{ - return x86_linux_i386_tdesc_count (); -} - -#endif /* IN_PROCESS_AGENT */ - -/* Convert an xcr0 value into an integer. The integer will be passed to - the in-process-agent where it will then be passed to - x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ - -int -x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0) -{ - /* The following table shows which features are checked for when creating - the target descriptions (see nat/x86-linux-tdesc.c), the feature order - represents the bit order within the generated index number. - - i386 | x87 sse mpx avx avx512 pkru - amd64 | mpx avx avx512 pkru - i32 | avx avx512 pkru - - The features are ordered so that for each mode (i386, amd64, i32) the - generated index will form a continuous range. */ - - int idx = 0; - - for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) - { - if ((xcr0 & x86_linux_all_tdesc_features[i].feature) != 0) - idx |= (1 << i); - } - - return idx; -} - - -#ifdef IN_PROCESS_AGENT - -/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) - into an xcr0 value which can then be used to create a target - description. */ - -uint64_t -x86_linux_tdesc_idx_to_xcr0 (int idx) -{ - uint64_t xcr0 = 0; - - for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) - { - if ((idx & (1 << i)) != 0) - xcr0 |= x86_linux_all_tdesc_features[i].feature; - } - - return xcr0; -} - -#endif /* IN_PROCESS_AGENT */ - -#if defined __i386__ || !defined IN_PROCESS_AGENT - -/* A cache of all possible i386 target descriptions. */ - -static struct target_desc *i386_tdescs[x86_linux_i386_tdesc_count ()] = { }; - /* See nat/x86-linux-tdesc.h. */ -const struct target_desc * -i386_linux_read_description (uint64_t xcr0) +void +x86_linux_post_init_tdesc (target_desc *tdesc, bool is_64bit) { - xcr0 &= x86_linux_i386_tdesc_feature_mask (); - int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - - gdb_assert (idx >= 0 && idx < x86_linux_i386_tdesc_count ()); - - target_desc **tdesc = &i386_tdescs[idx]; - - if (*tdesc == nullptr) - { - *tdesc = i386_create_target_description (xcr0, true, false); - - init_target_desc (*tdesc, i386_expedite_regs); - } - - return *tdesc; -} -#endif - #ifdef __x86_64__ - -/* A cache of all possible amd64 target descriptions. */ - -static target_desc *amd64_tdescs[x86_linux_amd64_tdesc_count ()] = { }; - -/* A cache of all possible x32 target descriptions. */ - -static target_desc *x32_tdescs[x86_linux_x32_tdesc_count ()] = { }; - -/* See nat/x86-linux-tdesc.h. */ - -const struct target_desc * -amd64_linux_read_description (uint64_t xcr0, bool is_x32) -{ - if (is_x32) - xcr0 &= x86_linux_x32_tdesc_feature_mask (); + if (is_64bit) + init_target_desc (tdesc, amd64_expedite_regs); else - xcr0 &= x86_linux_amd64_tdesc_feature_mask (); - - int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - - if (is_x32) - gdb_assert (idx >= 0 && idx < x86_linux_x32_tdesc_count ()); - else - gdb_assert (idx >= 0 && idx < x86_linux_amd64_tdesc_count ()); - - target_desc **tdesc = nullptr; - - if (is_x32) - tdesc = &x32_tdescs[idx]; - else - tdesc = &amd64_tdescs[idx]; - - if (*tdesc == nullptr) - { - *tdesc = amd64_create_target_description (xcr0, is_x32, true, true); - - init_target_desc (*tdesc, amd64_expedite_regs); - } - return *tdesc; +#endif + init_target_desc (tdesc, i386_expedite_regs); } -#endif diff --git a/gdbserver/linux-x86-tdesc.h b/gdbserver/linux-x86-tdesc.h deleted file mode 100644 index 70456e4be44..00000000000 --- a/gdbserver/linux-x86-tdesc.h +++ /dev/null @@ -1,50 +0,0 @@ -/* Low level support for x86 (i386 and x86-64), shared between gdbserver - and IPA. - - Copyright (C) 2016-2024 Free Software Foundation, Inc. - - This file is part of GDB. - - This program is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published by - the Free Software Foundation; either version 3 of the License, or - (at your option) any later version. - - This program is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - GNU General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see <http://www.gnu.org/licenses/>. */ - -#ifndef GDBSERVER_LINUX_X86_TDESC_H -#define GDBSERVER_LINUX_X86_TDESC_H - -/* Convert an xcr0 value into an integer. The integer will be passed to - the in-process-agent where it will then be passed to - x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ - -extern int x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0); - -#ifdef IN_PROCESS_AGENT - -/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) - into an xcr0 value which can then be used to create a target - description. */ - -extern uint64_t x86_linux_tdesc_idx_to_xcr0 (int idx); - -/* Within the in-process-agent we need to pre-initialise all of the target - descriptions, to do this we need to know how many target descriptions - there are for each different target type. These functions return the - target description count for the relevant target. */ - -extern int x86_linux_amd64_ipa_tdesc_count (); -extern int x86_linux_x32_ipa_tdesc_count (); -extern int x86_linux_i386_ipa_tdesc_count (); - - -#endif /* IN_PROCESS_AGENT */ - -#endif /* GDBSERVER_LINUX_X86_TDESC_H */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv3 8/8] gdb/gdbserver: share x86/linux tdesc caching 2024-03-23 16:35 ` [PATCHv3 8/8] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess @ 2024-03-26 12:17 ` Andrew Burgess 0 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-03-26 12:17 UTC (permalink / raw) To: gdb-patches; +Cc: hjl.tools, John Baldwin Simon pointed out some build issues encountered when building with Clang. While test building using Clang an additional issue was highlighted, which I've fixed by pushing the patch below. Sorry for anyone who ran into this issue. Thanks, Andrew --- commit 24df37a10f8773ad5db07dc000f694d6405e3a36 Author: Andrew Burgess <aburgess@redhat.com> Date: Tue Mar 26 12:05:07 2024 +0000 gdb: fix possible uninitialised variable use After this commit: commit 198ff6ff819c240545f9fc68b39636fd376d4ba9 Date: Tue Jan 30 15:37:23 2024 +0000 gdb/gdbserver: share x86/linux tdesc caching a possible use of an uninitialised variable was introduced, the 'tdesc' variable in i386_linux_core_read_description might be read without being written too if 'xcr0' was 0. This is fixed in this commit. I've updated the function to follow the same pattern as amd64_linux_core_read_description, if xcr0 is 0 then we select a default xcr0 value and use that to select a tdesc. diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index ed1442236e8..2e84384306d 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -700,17 +700,15 @@ i386_linux_core_read_description (struct gdbarch *gdbarch, x86_xsave_layout layout; uint64_t xcr0 = i386_linux_core_read_xsave_info (abfd, layout); - const struct target_desc *tdesc; - if (xcr0 != 0) - tdesc = i386_linux_read_description (xcr0); - - if (tdesc != NULL) - return tdesc; + if (xcr0 == 0) + { + if (bfd_get_section_by_name (abfd, ".reg-xfp") != nullptr) + xcr0 = X86_XSTATE_SSE_MASK; + else + xcr0 = X86_XSTATE_X87_MASK; + } - if (bfd_get_section_by_name (abfd, ".reg-xfp") != NULL) - return i386_linux_read_description (X86_XSTATE_SSE_MASK); - else - return i386_linux_read_description (X86_XSTATE_X87_MASK); + return i386_linux_read_description (xcr0); } /* Similar to i386_supply_fpregset, but use XSAVE extended state. */ ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv3 0/8] x86/Linux Target Description Changes 2024-03-23 16:35 ` [PATCHv3 0/8] " Andrew Burgess ` (7 preceding siblings ...) 2024-03-23 16:35 ` [PATCHv3 8/8] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess @ 2024-03-25 17:20 ` Andrew Burgess 2024-03-25 18:26 ` Simon Marchi 2024-03-26 16:36 ` Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 00/10] " Andrew Burgess 9 siblings, 2 replies; 116+ messages in thread From: Andrew Burgess @ 2024-03-25 17:20 UTC (permalink / raw) To: gdb-patches; +Cc: hjl.tools Andrew Burgess <aburgess@redhat.com> writes: > In v3: > > - Rebased. Nasty merge conflict with 4bb20a6244b7091 which I think > I've resolved, but am unable to test. Reposting so the author of > that other commit can validate. > > - Initial testing looks good. Full tests are still running. Testing completed with no issues. H.J. Lu confirmed that this versions didn't break the x32 behaviour. I've gone ahead and pushed these patches. If anything crops up then do let me know. Thanks, Andrew ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv3 0/8] x86/Linux Target Description Changes 2024-03-25 17:20 ` [PATCHv3 0/8] x86/Linux Target Description Changes Andrew Burgess @ 2024-03-25 18:26 ` Simon Marchi 2024-03-26 12:15 ` Andrew Burgess 2024-03-26 16:36 ` Andrew Burgess 1 sibling, 1 reply; 116+ messages in thread From: Simon Marchi @ 2024-03-25 18:26 UTC (permalink / raw) To: Andrew Burgess, gdb-patches; +Cc: hjl.tools On 3/25/24 13:20, Andrew Burgess wrote: > Andrew Burgess <aburgess@redhat.com> writes: > >> In v3: >> >> - Rebased. Nasty merge conflict with 4bb20a6244b7091 which I think >> I've resolved, but am unable to test. Reposting so the author of >> that other commit can validate. >> >> - Initial testing looks good. Full tests are still running. > > Testing completed with no issues. H.J. Lu confirmed that this versions > didn't break the x32 behaviour. I've gone ahead and pushed these > patches. > > If anything crops up then do let me know. > > Thanks, > Andrew > Hi Andrew, I think your series introduces some build failures with Clang. One is easy, it's a missing `-x c++` in gdbserver/Makefile.in. The other is: CXX nat/x86-linux-tdesc-ipa.o /home/smarchi/src/binutils-gdb/gdbserver/../gdb/nat/x86-linux-tdesc.c:167:1: error: unused function 'x86_linux_i386_tdesc_feature_mask' [-Werror,-Wunused-function] 167 | x86_linux_i386_tdesc_feature_mask () | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ It's possible that this function just needs to be moved in the same "#ifdef" as where it's used, but since I didn't follow your work closely, I prefer to let you fix it, in case I'm missing something. Thanks, Simon ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv3 0/8] x86/Linux Target Description Changes 2024-03-25 18:26 ` Simon Marchi @ 2024-03-26 12:15 ` Andrew Burgess 2024-03-26 13:51 ` H.J. Lu 0 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-03-26 12:15 UTC (permalink / raw) To: Simon Marchi, gdb-patches; +Cc: hjl.tools Simon Marchi <simark@simark.ca> writes: > On 3/25/24 13:20, Andrew Burgess wrote: >> Andrew Burgess <aburgess@redhat.com> writes: >> >>> In v3: >>> >>> - Rebased. Nasty merge conflict with 4bb20a6244b7091 which I think >>> I've resolved, but am unable to test. Reposting so the author of >>> that other commit can validate. >>> >>> - Initial testing looks good. Full tests are still running. >> >> Testing completed with no issues. H.J. Lu confirmed that this versions >> didn't break the x32 behaviour. I've gone ahead and pushed these >> patches. >> >> If anything crops up then do let me know. >> >> Thanks, >> Andrew >> > > Hi Andrew, > > I think your series introduces some build failures with Clang. One is > easy, it's a missing `-x c++` in gdbserver/Makefile.in. Thanks for fixing this one. > > The other is: > > CXX nat/x86-linux-tdesc-ipa.o > /home/smarchi/src/binutils-gdb/gdbserver/../gdb/nat/x86-linux-tdesc.c:167:1: error: unused function 'x86_linux_i386_tdesc_feature_mask' [-Werror,-Wunused-function] > 167 | x86_linux_i386_tdesc_feature_mask () > | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ > > It's possible that this function just needs to be moved in the same > "#ifdef" as where it's used, but since I didn't follow your work > closely, I prefer to let you fix it, in case I'm missing something. Sorry for the breakage. I pushed the patch below to resolve this issue. Thanks, Andrew -- commit f4c19f89ef43dbce8065532c808e1aeb05d08994 Author: Andrew Burgess <aburgess@redhat.com> Date: Tue Mar 26 12:09:27 2024 +0000 gdb/gdbserver: fix some defined but unused function warnings This commit: commit 198ff6ff819c240545f9fc68b39636fd376d4ba9 Date: Tue Jan 30 15:37:23 2024 +0000 gdb/gdbserver: share x86/linux tdesc caching added some functions which are always defined, but their use is guarded within various #ifdef blocks. As a result we were seeing errors about defined, but unused, functions. I've fixed this problem in this commit by wrapping the function definitions within #ifdef blocks. I'm a little worried that there might be too many #ifdef blocks within this file, however, I'm going to commit this fix for now as this will fix the build, then I'll think about if there's a better way to split this file so we might avoid some of these #ifdef blocks. diff --git a/gdb/nat/x86-linux-tdesc.c b/gdb/nat/x86-linux-tdesc.c index c438dfae84f..8a02f77fa6a 100644 --- a/gdb/nat/x86-linux-tdesc.c +++ b/gdb/nat/x86-linux-tdesc.c @@ -160,6 +160,8 @@ static constexpr x86_tdesc_feature x86_linux_all_tdesc_features[] = { { X86_XSTATE_X87, true, false, false } }; +#if defined __i386__ || !defined IN_PROCESS_AGENT + /* Return a compile time constant which is a mask of all the cpu features that are checked for when building an i386 target description. */ @@ -175,6 +177,10 @@ x86_linux_i386_tdesc_feature_mask () return mask; } +#endif /* __i386__ || !IN_PROCESS_AGENT */ + +#ifdef __x86_64__ + /* Return a compile time constant which is a mask of all the cpu features that are checked for when building an amd64 target description. */ @@ -205,6 +211,8 @@ x86_linux_x32_tdesc_feature_mask () return mask; } +#endif /* __x86_64__ */ + /* Return a compile time constant which is a count of the number of cpu features that are checked for when building an i386 target description. */ @@ -222,6 +230,8 @@ x86_linux_i386_tdesc_count () return (1 << count); } +#if defined __x86_64__ || defined IN_PROCESS_AGENT + /* Return a compile time constant which is a count of the number of cpu features that are checked for when building an amd64 target description. */ @@ -256,6 +266,8 @@ x86_linux_x32_tdesc_count () return (1 << count); } +#endif /* __x86_64__ || IN_PROCESS_AGENT */ + #ifdef IN_PROCESS_AGENT /* See linux-x86-tdesc.h. */ ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv3 0/8] x86/Linux Target Description Changes 2024-03-26 12:15 ` Andrew Burgess @ 2024-03-26 13:51 ` H.J. Lu 2024-03-26 14:16 ` H.J. Lu 0 siblings, 1 reply; 116+ messages in thread From: H.J. Lu @ 2024-03-26 13:51 UTC (permalink / raw) To: Andrew Burgess; +Cc: Simon Marchi, gdb-patches On Tue, Mar 26, 2024 at 5:16 AM Andrew Burgess <aburgess@redhat.com> wrote: > > Simon Marchi <simark@simark.ca> writes: > > > On 3/25/24 13:20, Andrew Burgess wrote: > >> Andrew Burgess <aburgess@redhat.com> writes: > >> > >>> In v3: > >>> > >>> - Rebased. Nasty merge conflict with 4bb20a6244b7091 which I think > >>> I've resolved, but am unable to test. Reposting so the author of > >>> that other commit can validate. > >>> > >>> - Initial testing looks good. Full tests are still running. > >> > >> Testing completed with no issues. H.J. Lu confirmed that this versions > >> didn't break the x32 behaviour. I've gone ahead and pushed these > >> patches. > >> > >> If anything crops up then do let me know. > >> > >> Thanks, > >> Andrew > >> > > > > Hi Andrew, > > > > I think your series introduces some build failures with Clang. One is > > easy, it's a missing `-x c++` in gdbserver/Makefile.in. > > Thanks for fixing this one. > > > > > The other is: > > > > CXX nat/x86-linux-tdesc-ipa.o > > /home/smarchi/src/binutils-gdb/gdbserver/../gdb/nat/x86-linux-tdesc.c:167:1: error: unused function 'x86_linux_i386_tdesc_feature_mask' [-Werror,-Wunused-function] > > 167 | x86_linux_i386_tdesc_feature_mask () > > | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ > > > > It's possible that this function just needs to be moved in the same > > "#ifdef" as where it's used, but since I didn't follow your work > > closely, I prefer to let you fix it, in case I'm missing something. > > Sorry for the breakage. > > I pushed the patch below to resolve this issue. > gdbserver from your old patch: https://patchwork.sourceware.org/project/gdb/list/?series=32189 works on x32. But the current master gives me: /export/gnu/import/git/sources/gdb/gdbserver/regcache.cc:273: A problem internal to GDBserver has been detected. Unknown register bnd0raw requested again. -- H.J. ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv3 0/8] x86/Linux Target Description Changes 2024-03-26 13:51 ` H.J. Lu @ 2024-03-26 14:16 ` H.J. Lu 0 siblings, 0 replies; 116+ messages in thread From: H.J. Lu @ 2024-03-26 14:16 UTC (permalink / raw) To: Andrew Burgess; +Cc: Simon Marchi, gdb-patches On Tue, Mar 26, 2024 at 6:51 AM H.J. Lu <hjl.tools@gmail.com> wrote: > > On Tue, Mar 26, 2024 at 5:16 AM Andrew Burgess <aburgess@redhat.com> wrote: > > > > Simon Marchi <simark@simark.ca> writes: > > > > > On 3/25/24 13:20, Andrew Burgess wrote: > > >> Andrew Burgess <aburgess@redhat.com> writes: > > >> > > >>> In v3: > > >>> > > >>> - Rebased. Nasty merge conflict with 4bb20a6244b7091 which I think > > >>> I've resolved, but am unable to test. Reposting so the author of > > >>> that other commit can validate. > > >>> > > >>> - Initial testing looks good. Full tests are still running. > > >> > > >> Testing completed with no issues. H.J. Lu confirmed that this versions > > >> didn't break the x32 behaviour. I've gone ahead and pushed these > > >> patches. > > >> > > >> If anything crops up then do let me know. > > >> > > >> Thanks, > > >> Andrew > > >> > > > > > > Hi Andrew, > > > > > > I think your series introduces some build failures with Clang. One is > > > easy, it's a missing `-x c++` in gdbserver/Makefile.in. > > > > Thanks for fixing this one. > > > > > > > > The other is: > > > > > > CXX nat/x86-linux-tdesc-ipa.o > > > /home/smarchi/src/binutils-gdb/gdbserver/../gdb/nat/x86-linux-tdesc.c:167:1: error: unused function 'x86_linux_i386_tdesc_feature_mask' [-Werror,-Wunused-function] > > > 167 | x86_linux_i386_tdesc_feature_mask () > > > | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ > > > > > > It's possible that this function just needs to be moved in the same > > > "#ifdef" as where it's used, but since I didn't follow your work > > > closely, I prefer to let you fix it, in case I'm missing something. > > > > Sorry for the breakage. > > > > I pushed the patch below to resolve this issue. > > > > gdbserver from your old patch: > > https://patchwork.sourceware.org/project/gdb/list/?series=32189 > > works on x32. But the current master gives me: > > /export/gnu/import/git/sources/gdb/gdbserver/regcache.cc:273: A > problem internal to GDBserver has been detected. > Unknown register bnd0raw requested > > again. Never mind. I was using the wrong source. x32 works. -- H.J. ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv3 0/8] x86/Linux Target Description Changes 2024-03-25 17:20 ` [PATCHv3 0/8] x86/Linux Target Description Changes Andrew Burgess 2024-03-25 18:26 ` Simon Marchi @ 2024-03-26 16:36 ` Andrew Burgess 2024-03-26 19:03 ` Andrew Burgess 1 sibling, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-03-26 16:36 UTC (permalink / raw) To: gdb-patches; +Cc: hjl.tools Andrew Burgess <aburgess@redhat.com> writes: > Andrew Burgess <aburgess@redhat.com> writes: > >> In v3: >> >> - Rebased. Nasty merge conflict with 4bb20a6244b7091 which I think >> I've resolved, but am unable to test. Reposting so the author of >> that other commit can validate. >> >> - Initial testing looks good. Full tests are still running. > > Testing completed with no issues. H.J. Lu confirmed that this versions > didn't break the x32 behaviour. I've gone ahead and pushed these > patches. > > If anything crops up then do let me know. I'm aware that this series broke pretty much anything that was not x86 based when configured with --enable-targets=all. The problem is that, as part of this commit, I moved generic tdep (arch specific) code into the nat/ directory as part of an effort to share the code between GDB and gdbserver. This was incorrect. I'm open to reverting this series, however, I'm currently working on a fix, splitting the nat/ code between nat/ and arch/. I'm hoping to have a fix today. If that doesn't look possible then I'll likely revert this series before the end of my day. Apologies for the (serious) breakage. Thanks, Andrew ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv3 0/8] x86/Linux Target Description Changes 2024-03-26 16:36 ` Andrew Burgess @ 2024-03-26 19:03 ` Andrew Burgess 0 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-03-26 19:03 UTC (permalink / raw) To: gdb-patches; +Cc: hjl.tools Andrew Burgess <aburgess@redhat.com> writes: > Andrew Burgess <aburgess@redhat.com> writes: > >> Andrew Burgess <aburgess@redhat.com> writes: >> >>> In v3: >>> >>> - Rebased. Nasty merge conflict with 4bb20a6244b7091 which I think >>> I've resolved, but am unable to test. Reposting so the author of >>> that other commit can validate. >>> >>> - Initial testing looks good. Full tests are still running. >> >> Testing completed with no issues. H.J. Lu confirmed that this versions >> didn't break the x32 behaviour. I've gone ahead and pushed these >> patches. >> >> If anything crops up then do let me know. > > I'm aware that this series broke pretty much anything that was not x86 > based when configured with --enable-targets=all. > > The problem is that, as part of this commit, I moved generic tdep (arch > specific) code into the nat/ directory as part of an effort to share the > code between GDB and gdbserver. This was incorrect. > > I'm open to reverting this series, however, I'm currently working on a > fix, splitting the nat/ code between nat/ and arch/. I'm hoping to have > a fix today. If that doesn't look possible then I'll likely revert this > series before the end of my day. > > Apologies for the (serious) breakage. I have now reverted all of the commits from this series, as well as the small fixes which I pushed earlier today. I've also reverts the fix which Simon pushed that was for code from this series. Splitting the code between nat/ and arch/ mostly worked, but I was still running into some edge cases which I couldn't get working, so I could either: + push a partial fix, which still left some things broken, or + revert everything, take a step back and re-work the series. I think reverting and coming at this fresh tomorrow is the better choice. I apologise for all the churn. The post-revert branch seemed to build fine, so hopefully I've not left anything in a broken state. Let me know if I've left anything broken, I'll check back in a couple of hours to see if there is more fixing needed. Thanks, Andrew ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv4 00/10] x86/Linux Target Description Changes 2024-03-23 16:35 ` [PATCHv3 0/8] " Andrew Burgess ` (8 preceding siblings ...) 2024-03-25 17:20 ` [PATCHv3 0/8] x86/Linux Target Description Changes Andrew Burgess @ 2024-04-05 12:33 ` Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 01/10] gdbserver/ipa/x86: remove unneeded declarations Andrew Burgess ` (12 more replies) 9 siblings, 13 replies; 116+ messages in thread From: Andrew Burgess @ 2024-04-05 12:33 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess In v4: - I tried merging V3, but it turned out I broke pretty much everything that wasn't x86 based when configured with --enable-targets=all, - The problem was a failure to correctly split the shared code between the gdb/arch/ and gdb/nat/ directories, as a consequence, code which is needed on a non x86 based host to support x86 based targets wasn't available to the compilation, and the build failed, - In V4 I've gone through every patch and resplit the code in a way which I now believe is correct, I've done the following tests: + On a non x86 host I've built GDB to support only the current host as a target, to support all targets, and to support x86-64 and i386 linux targets, + On an i386 virtual machine I built GDB only for the host as a target, and for all targets. I regression tested the all targets build for unix, native-gdbserver, and native-extended-gdbserver, + On an x86-64 machine I've built GDB for only the current host as a target, and for all targets. I regression tested the all targets build for unix, native-gdbserver, and native-extended-gdbserver. - Only patches 6, 8, and 10 require significant review. All of the other patches are pretty trivial (though reviews always welcome). - I think there's more improvements that can be made to the x86 target description creation/lookup/caching. This series only changes the Linux lookup, and we still cache i386/amd64/x32 separately. In the future I think we can merge all x86 target description caching into a single data structure, this would be for all OS variants and all ABI variants. Though making that "grand unification" will certainly require some of the code in this series to change, I think the bulk of it will remain, and trying to do everything in one series is just going to result in an even larger series. I'd prefer to get these first patches merged, then come back to build on this work once this is merged and we know there's no problems with it. In v3: - Rebased. Nasty merge conflict with 4bb20a6244b7091 which I think I've resolved, but am unable to test. Reposting so the author of that other commit can validate. - Initial testing looks good. Full tests are still running. In v2: - Rebase to current upstream/master, no merge conflicts, - Retested. --- Andrew Burgess (10): gdbserver/ipa/x86: remove unneeded declarations gdbserver: convert have_ptrace_getregset to a tribool gdb/x86: move reading of cs and ds state into gdb/nat directory gdbserver/x86: move no-xml code earlier in x86_linux_read_description gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition gdb/gdbserver: share some code relating to target description creation gdb/arch: assert that X86_XSTATE_MPX is not set for x32 gdbserver: update target description creation for x86/linux gdb: move xcr0 == 0 check into i386_linux_core_read_description gdb/gdbserver: share x86/linux tdesc caching gdb/Makefile.in | 8 + gdb/amd64-linux-tdep.c | 32 +--- gdb/amd64-linux-tdep.h | 6 - gdb/arch/amd64-linux-tdesc.c | 61 +++++++ gdb/arch/amd64-linux-tdesc.h | 30 ++++ gdb/arch/amd64.c | 8 +- gdb/arch/i386-linux-tdesc.c | 51 ++++++ gdb/arch/i386-linux-tdesc.h | 29 ++++ gdb/arch/x86-linux-tdesc-features.c | 247 ++++++++++++++++++++++++++++ gdb/arch/x86-linux-tdesc-features.h | 62 +++++++ gdb/arch/x86-linux-tdesc.h | 37 +++++ gdb/configure.nat | 8 +- gdb/configure.tgt | 11 +- gdb/i386-linux-tdep.c | 42 ++--- gdb/i386-linux-tdep.h | 23 --- gdb/nat/x86-linux-tdesc.c | 120 ++++++++++++++ gdb/nat/x86-linux-tdesc.h | 60 +++++++ gdb/nat/x86-linux.c | 47 ++++++ gdb/nat/x86-linux.h | 28 ++++ gdb/x86-linux-nat.c | 123 ++------------ gdbserver/configure.srv | 11 ++ gdbserver/linux-amd64-ipa.cc | 50 ++---- gdbserver/linux-arm-low.cc | 6 +- gdbserver/linux-i386-ipa.cc | 26 +-- gdbserver/linux-low.cc | 2 +- gdbserver/linux-low.h | 2 +- gdbserver/linux-x86-low.cc | 192 +++++++-------------- gdbserver/linux-x86-tdesc.cc | 142 +--------------- gdbserver/linux-x86-tdesc.h | 56 ------- gdbsupport/x86-xstate.h | 20 +++ 30 files changed, 949 insertions(+), 591 deletions(-) create mode 100644 gdb/arch/amd64-linux-tdesc.c create mode 100644 gdb/arch/amd64-linux-tdesc.h create mode 100644 gdb/arch/i386-linux-tdesc.c create mode 100644 gdb/arch/i386-linux-tdesc.h create mode 100644 gdb/arch/x86-linux-tdesc-features.c create mode 100644 gdb/arch/x86-linux-tdesc-features.h create mode 100644 gdb/arch/x86-linux-tdesc.h create mode 100644 gdb/nat/x86-linux-tdesc.c create mode 100644 gdb/nat/x86-linux-tdesc.h delete mode 100644 gdbserver/linux-x86-tdesc.h base-commit: 16810e455feb26ef826a3ed876d6d7e6d24818b0 -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv4 01/10] gdbserver/ipa/x86: remove unneeded declarations 2024-04-05 12:33 ` [PATCHv4 00/10] " Andrew Burgess @ 2024-04-05 12:33 ` Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 02/10] gdbserver: convert have_ptrace_getregset to a tribool Andrew Burgess ` (11 subsequent siblings) 12 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-04-05 12:33 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess Spotted some declarations in gdbserver/linux-amd64-ipa.cc that are no longer needed. These are: 1. 'init_registers_amd64_linux' - the comment claims this function is auto generated, but I don't believe that this is still the case. Also the function is not used in this file, 2. 'tdesc_amd64_linux' - this variable doesn't seem to exist any more, I suspect this was renamed to 'tdesc_amd64_linux_no_xml', but neither are used in this file, so lets remove the declaration. The amd64 in-process-agent still builds fine after this commit. There should be no user visible changes after this commit. --- gdbserver/linux-amd64-ipa.cc | 4 ---- 1 file changed, 4 deletions(-) diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index 13c8a5bb360..a6346750f49 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -23,10 +23,6 @@ #include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" -/* Defined in auto-generated file amd64-linux.c. */ -void init_registers_amd64_linux (void); -extern const struct target_desc *tdesc_amd64_linux; - /* fast tracepoints collect registers. */ #define FT_CR_RIP 0 -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv4 02/10] gdbserver: convert have_ptrace_getregset to a tribool 2024-04-05 12:33 ` [PATCHv4 00/10] " Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 01/10] gdbserver/ipa/x86: remove unneeded declarations Andrew Burgess @ 2024-04-05 12:33 ` Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 03/10] gdb/x86: move reading of cs and ds state into gdb/nat directory Andrew Burgess ` (10 subsequent siblings) 12 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-04-05 12:33 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, John Baldwin Convert the have_ptrace_getregset global within gdbserver to a tribool. This brings the flag into alignment with the corresponding flag in GDB. The gdbserver have_ptrace_getregset variable is already used as a tribool, it just doesn't have the tribool type. In a future commit I plan to share more code between GDB and gdbserver, and having this variable be the same type in both code bases will make the sharing much easier. There should be no user visible changes after this commit. Approved-By: John Baldwin <jhb@FreeBSD.org> --- gdbserver/linux-arm-low.cc | 6 +++--- gdbserver/linux-low.cc | 2 +- gdbserver/linux-low.h | 2 +- gdbserver/linux-x86-low.cc | 10 +++++----- 4 files changed, 10 insertions(+), 10 deletions(-) diff --git a/gdbserver/linux-arm-low.cc b/gdbserver/linux-arm-low.cc index 17b64c09aff..eec4649b235 100644 --- a/gdbserver/linux-arm-low.cc +++ b/gdbserver/linux-arm-low.cc @@ -1006,9 +1006,9 @@ arm_target::low_arch_setup () /* Check if PTRACE_GETREGSET works. */ if (ptrace (PTRACE_GETREGSET, tid, NT_PRSTATUS, &iov) == 0) - have_ptrace_getregset = 1; + have_ptrace_getregset = TRIBOOL_TRUE; else - have_ptrace_getregset = 0; + have_ptrace_getregset = TRIBOOL_FALSE; } bool @@ -1121,7 +1121,7 @@ arm_target::get_regs_info () { const struct target_desc *tdesc = current_process ()->tdesc; - if (have_ptrace_getregset == 1 + if (have_ptrace_getregset == TRIBOOL_TRUE && (is_aarch32_linux_description (tdesc) || arm_linux_get_tdesc_fp_type (tdesc) == ARM_FP_TYPE_VFPV3)) return ®s_info_aarch32; diff --git a/gdbserver/linux-low.cc b/gdbserver/linux-low.cc index d8842f0972d..5ba61178132 100644 --- a/gdbserver/linux-low.cc +++ b/gdbserver/linux-low.cc @@ -134,7 +134,7 @@ typedef struct #endif /* Does the current host support PTRACE_GETREGSET? */ -int have_ptrace_getregset = -1; +enum tribool have_ptrace_getregset = TRIBOOL_UNKNOWN; /* Return TRUE if THREAD is the leader thread of the process. */ diff --git a/gdbserver/linux-low.h b/gdbserver/linux-low.h index d34d2738238..eaf87527338 100644 --- a/gdbserver/linux-low.h +++ b/gdbserver/linux-low.h @@ -951,7 +951,7 @@ void thread_db_notice_clone (struct thread_info *parent_thr, ptid_t child_ptid); bool thread_db_thread_handle (ptid_t ptid, gdb_byte **handle, int *handle_len); -extern int have_ptrace_getregset; +extern enum tribool have_ptrace_getregset; /* Search for the value with type MATCH in the auxv vector, with entries of length WORDSIZE bytes, of process with pid PID. If found, store the diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 2532603451a..2603fb2ac5d 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -898,7 +898,7 @@ x86_linux_read_description (void) if (ptrace (PTRACE_GETFPXREGS, tid, 0, (long) &fpxregs) < 0) { have_ptrace_getfpxregs = 0; - have_ptrace_getregset = 0; + have_ptrace_getregset = TRIBOOL_FALSE; return i386_linux_read_description (X86_XSTATE_X87); } else @@ -917,7 +917,7 @@ x86_linux_read_description (void) return tdesc_i386_linux_no_xml.get (); } - if (have_ptrace_getregset == -1) + if (have_ptrace_getregset == TRIBOOL_UNKNOWN) { uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; struct iovec iov; @@ -928,10 +928,10 @@ x86_linux_read_description (void) /* Check if PTRACE_GETREGSET works. */ if (ptrace (PTRACE_GETREGSET, tid, (unsigned int) NT_X86_XSTATE, (long) &iov) < 0) - have_ptrace_getregset = 0; + have_ptrace_getregset = TRIBOOL_FALSE; else { - have_ptrace_getregset = 1; + have_ptrace_getregset = TRIBOOL_TRUE; /* Get XCR0 from XSAVE extended state. */ xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET @@ -954,7 +954,7 @@ x86_linux_read_description (void) } /* Check the native XCR0 only if PTRACE_GETREGSET is available. */ - xcr0_features = (have_ptrace_getregset + xcr0_features = (have_ptrace_getregset == TRIBOOL_TRUE && (xcr0 & X86_XSTATE_ALL_MASK)); if (xcr0_features) -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv4 03/10] gdb/x86: move reading of cs and ds state into gdb/nat directory 2024-04-05 12:33 ` [PATCHv4 00/10] " Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 01/10] gdbserver/ipa/x86: remove unneeded declarations Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 02/10] gdbserver: convert have_ptrace_getregset to a tribool Andrew Burgess @ 2024-04-05 12:33 ` Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 04/10] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess ` (9 subsequent siblings) 12 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-04-05 12:33 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, John Baldwin This patch is part of a series that has the aim sharing the x86 Linux target description creation code between GDB and gdbserver. Within GDB part of this process involves reading the cs and ds state from the 'struct user_regs_struct' using a ptrace call. This isn't done by gdbserver, which is part of the motivation for this whole series; the approach gdbserver takes is inferior to the approach GDB takes (gdbserver relies on reading the file being debugged, and extracting similar information from the file headers). This commit moves the reading of cs and ds, which is used to figure out if a thread is 32-bit or 64-bit (or in x32 mode), into the gdb/nat directory so that the code can be shared with gdbserver, but at this point I'm not actually using the code in gdbserver, that will come later. As such there should be no user visible changes after this commit, GDB continues to do things as it did before (reading cs/ds), while gdbserver continues to use its own approach (which doesn't require reading cs/ds). Approved-By: John Baldwin <jhb@FreeBSD.org> --- gdb/nat/x86-linux.c | 47 +++++++++++++++++++++++++++++++++++++++++++++ gdb/nat/x86-linux.h | 28 +++++++++++++++++++++++++++ gdb/x86-linux-nat.c | 42 +++++----------------------------------- 3 files changed, 80 insertions(+), 37 deletions(-) diff --git a/gdb/nat/x86-linux.c b/gdb/nat/x86-linux.c index 7a21c8f2c26..ad3ed3c2289 100644 --- a/gdb/nat/x86-linux.c +++ b/gdb/nat/x86-linux.c @@ -19,6 +19,8 @@ #include "x86-linux.h" #include "x86-linux-dregs.h" +#include "nat/gdb_ptrace.h" +#include <sys/user.h> /* Per-thread arch-specific data we want to keep. */ @@ -79,3 +81,48 @@ x86_linux_prepare_to_resume (struct lwp_info *lwp) { x86_linux_update_debug_registers (lwp); } + +#ifdef __x86_64__ +/* Value of CS segment register: + 64bit process: 0x33 + 32bit process: 0x23 */ +#define AMD64_LINUX_USER64_CS 0x33 + +/* Value of DS segment register: + LP64 process: 0x0 + X32 process: 0x2b */ +#define AMD64_LINUX_X32_DS 0x2b +#endif + +/* See nat/x86-linux.h. */ + +x86_linux_arch_size +x86_linux_ptrace_get_arch_size (int tid) +{ +#ifdef __x86_64__ + unsigned long cs; + unsigned long ds; + + /* Get CS register. */ + errno = 0; + cs = ptrace (PTRACE_PEEKUSER, tid, + offsetof (struct user_regs_struct, cs), 0); + if (errno != 0) + perror_with_name (_("Couldn't get CS register")); + + bool is_64bit = cs == AMD64_LINUX_USER64_CS; + + /* Get DS register. */ + errno = 0; + ds = ptrace (PTRACE_PEEKUSER, tid, + offsetof (struct user_regs_struct, ds), 0); + if (errno != 0) + perror_with_name (_("Couldn't get DS register")); + + bool is_x32 = ds == AMD64_LINUX_X32_DS; + + return x86_linux_arch_size (is_64bit, is_x32); +#else + return x86_linux_arch_size (false, false); +#endif +} diff --git a/gdb/nat/x86-linux.h b/gdb/nat/x86-linux.h index 822882173f9..15153ea277e 100644 --- a/gdb/nat/x86-linux.h +++ b/gdb/nat/x86-linux.h @@ -47,4 +47,32 @@ extern void x86_linux_delete_thread (struct arch_lwp_info *arch_lwp); extern void x86_linux_prepare_to_resume (struct lwp_info *lwp); +/* Return value from x86_linux_ptrace_get_arch_size function. Indicates if + a thread is 32-bit, 64-bit, or x32. */ + +struct x86_linux_arch_size +{ + explicit x86_linux_arch_size (bool is_64bit, bool is_x32) + : m_is_64bit (is_64bit), + m_is_x32 (is_x32) + { + /* Nothing. */ + } + + bool is_64bit () const + { return m_is_64bit; } + + bool is_x32 () const + { return m_is_x32; } + +private: + bool m_is_64bit = false; + bool m_is_x32 = false; +}; + +/* Use ptrace calls to figure out if thread TID is 32-bit, 64-bit, or + 64-bit running in x32 mode. */ + +extern x86_linux_arch_size x86_linux_ptrace_get_arch_size (int tid); + #endif /* NAT_X86_LINUX_H */ diff --git a/gdb/x86-linux-nat.c b/gdb/x86-linux-nat.c index a3d8ffb60f1..f91db492d05 100644 --- a/gdb/x86-linux-nat.c +++ b/gdb/x86-linux-nat.c @@ -90,18 +90,6 @@ x86_linux_nat_target::post_startup_inferior (ptid_t ptid) linux_nat_target::post_startup_inferior (ptid); } -#ifdef __x86_64__ -/* Value of CS segment register: - 64bit process: 0x33 - 32bit process: 0x23 */ -#define AMD64_LINUX_USER64_CS 0x33 - -/* Value of DS segment register: - LP64 process: 0x0 - X32 process: 0x2b */ -#define AMD64_LINUX_X32_DS 0x2b -#endif - /* Get Linux/x86 target description from running target. */ const struct target_desc * @@ -121,31 +109,11 @@ x86_linux_nat_target::read_description () tid = inferior_ptid.pid (); #ifdef __x86_64__ - { - unsigned long cs; - unsigned long ds; - - /* Get CS register. */ - errno = 0; - cs = ptrace (PTRACE_PEEKUSER, tid, - offsetof (struct user_regs_struct, cs), 0); - if (errno != 0) - perror_with_name (_("Couldn't get CS register")); - - is_64bit = cs == AMD64_LINUX_USER64_CS; - - /* Get DS register. */ - errno = 0; - ds = ptrace (PTRACE_PEEKUSER, tid, - offsetof (struct user_regs_struct, ds), 0); - if (errno != 0) - perror_with_name (_("Couldn't get DS register")); - - is_x32 = ds == AMD64_LINUX_X32_DS; - - if (sizeof (void *) == 4 && is_64bit && !is_x32) - error (_("Can't debug 64-bit process with 32-bit GDB")); - } + + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); + is_64bit = arch_size.is_64bit (); + is_x32 = arch_size.is_x32 (); + #elif HAVE_PTRACE_GETFPXREGS if (have_ptrace_getfpxregs == -1) { -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv4 04/10] gdbserver/x86: move no-xml code earlier in x86_linux_read_description 2024-04-05 12:33 ` [PATCHv4 00/10] " Andrew Burgess ` (2 preceding siblings ...) 2024-04-05 12:33 ` [PATCHv4 03/10] gdb/x86: move reading of cs and ds state into gdb/nat directory Andrew Burgess @ 2024-04-05 12:33 ` Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 05/10] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess ` (8 subsequent siblings) 12 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-04-05 12:33 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, John Baldwin This commit is part of a series that aims to share more of the x86 target description reading/generation code between GDB and gdbserver. There are a huge number of similarities between the code in gdbserver's x86_linux_read_description function and GDB's x86_linux_nat_target::read_description function, and it is this similarity that I plan, in a later commit, to share between GDB and gdbserver. However, one thing that is different in x86_linux_read_description is the code inside the '!use_xml' block. This is the code that handles the case where gdbserver is not allowed to send an XML target description back to GDB. In this case gdbserver uses some predefined, fixed, target descriptions. First, it's worth noting that I suspect this code is not tested any more. I couldn't find anything in the testsuite that tries to disable XML target description support. And the idea of having a single "fixed" target description really doesn't work well when we think about all the various x86 extensions that exist. Part of me would like to rip out the no-xml support in gdbserver (at least for x86), and if a GDB connects that doesn't support XML target descriptions, gdbserver can just give an error and drop the connection. GDB has supported XML target descriptions for 16 years now, I think it would be reasonable for our shipped gdbserver to drop support for the old way of doing things. Anyway.... this commit doesn't do that. What I did notice was that, over time, the '!use_xml' block appears to have "drifted" within the x86_linux_read_description function; it's now not the first check we do. Instead we make some ptrace calls and return a target description generated based on the result of these ptrace calls. Surely it only makes sense to generate variable target descriptions if we can send these back to GDB? So in this commit I propose to move the '!use_xml' block earlier in the x86_linux_read_description function. The benefit of this is that this leaves the later half of x86_linux_read_description much more similar to the GDB function x86_linux_nat_target::read_description and sets us up for potentially sharing code between GDB and gdbserver in a later commit. Approved-By: John Baldwin <jhb@FreeBSD.org> --- gdbserver/linux-x86-low.cc | 27 ++++++++++++++++----------- 1 file changed, 16 insertions(+), 11 deletions(-) diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 2603fb2ac5d..eeeddcb9429 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -890,6 +890,22 @@ x86_linux_read_description (void) #endif } + /* If we are not allowed to send an XML target description then we need + to use the hard-wired target descriptions. This corresponds to GDB's + default machine for x86. + + This check needs to occur before any returns statements that might + generate some alternative target descriptions. */ + if (!use_xml) + { +#ifdef __x86_64__ + if (machine == EM_X86_64) + return tdesc_amd64_linux_no_xml.get (); + else +#endif + return tdesc_i386_linux_no_xml.get (); + } + #if !defined __x86_64__ && defined HAVE_PTRACE_GETFPXREGS if (machine == EM_386 && have_ptrace_getfpxregs == -1) { @@ -906,17 +922,6 @@ x86_linux_read_description (void) } #endif - if (!use_xml) - { - /* Don't use XML. */ -#ifdef __x86_64__ - if (machine == EM_X86_64) - return tdesc_amd64_linux_no_xml.get (); - else -#endif - return tdesc_i386_linux_no_xml.get (); - } - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) { uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv4 05/10] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition 2024-04-05 12:33 ` [PATCHv4 00/10] " Andrew Burgess ` (3 preceding siblings ...) 2024-04-05 12:33 ` [PATCHv4 04/10] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess @ 2024-04-05 12:33 ` Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 06/10] gdb/gdbserver: share some code relating to target description creation Andrew Burgess ` (7 subsequent siblings) 12 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-04-05 12:33 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess Share the definition of I386_LINUX_XSAVE_XCR0_OFFSET between GDB and gdbserver. This commit is part of a series that aims to share more of the x86 target description creation code between GDB and gdbserver. The I386_LINUX_XSAVE_XCR0_OFFSET #define is used as part of the target description creation, and I noticed that this constant is defined separately for GDB and gdbserver. This commit moves the definition into gdbsupport/x86-xstate.h, which allows the #define to be shared. There should be no user visible changes after this commit. --- gdb/i386-linux-tdep.h | 20 -------------------- gdbserver/linux-x86-low.cc | 21 --------------------- gdbsupport/x86-xstate.h | 20 ++++++++++++++++++++ 3 files changed, 20 insertions(+), 41 deletions(-) diff --git a/gdb/i386-linux-tdep.h b/gdb/i386-linux-tdep.h index 5891747572b..07593c6a8ec 100644 --- a/gdb/i386-linux-tdep.h +++ b/gdb/i386-linux-tdep.h @@ -58,26 +58,6 @@ extern void i386_linux_report_signal_info (struct gdbarch *gdbarch, /* Return the target description according to XCR0. */ extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); -/* Format of XSAVE extended state is: - struct - { - fxsave_bytes[0..463] - sw_usable_bytes[464..511] - xstate_hdr_bytes[512..575] - extended state regions (AVX, MPX, AVX512, PKRU, etc.) - }; - - Same memory layout will be used for the coredump NT_X86_XSTATE - representing the XSAVE extended state registers. - - The first 8 bytes of the sw_usable_bytes[464..467] is the OS enabled - extended state mask, which is the same as the extended control register - 0 (the XFEATURE_ENABLED_MASK register), XCR0. We can use this mask - together with the mask saved in the xstate_hdr_bytes to determine what - states the processor/OS supports and what state, used or initialized, - the process/thread is in. */ -#define I386_LINUX_XSAVE_XCR0_OFFSET 464 - extern int i386_linux_gregset_reg_offset[]; /* Return x86 siginfo type. */ diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index eeeddcb9429..e8ef3667eb4 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -831,27 +831,6 @@ x86_target::low_siginfo_fixup (siginfo_t *ptrace, gdb_byte *inf, int direction) \f static int use_xml; -/* Format of XSAVE extended state is: - struct - { - fxsave_bytes[0..463] - sw_usable_bytes[464..511] - xstate_hdr_bytes[512..575] - avx_bytes[576..831] - future_state etc - }; - - Same memory layout will be used for the coredump NT_X86_XSTATE - representing the XSAVE extended state registers. - - The first 8 bytes of the sw_usable_bytes[464..467] is the OS enabled - extended state mask, which is the same as the extended control register - 0 (the XFEATURE_ENABLED_MASK register), XCR0. We can use this mask - together with the mask saved in the xstate_hdr_bytes to determine what - states the processor/OS supports and what state, used or initialized, - the process/thread is in. */ -#define I386_LINUX_XSAVE_XCR0_OFFSET 464 - /* Does the current host support the GETFPXREGS request? The header file may or may not define it, and even if it is defined, the kernel will return EIO if it's running on a pre-SSE processor. */ diff --git a/gdbsupport/x86-xstate.h b/gdbsupport/x86-xstate.h index 89c1143fbe1..11b37544aa3 100644 --- a/gdbsupport/x86-xstate.h +++ b/gdbsupport/x86-xstate.h @@ -120,4 +120,24 @@ constexpr bool operator!= (const x86_xsave_layout &lhs, #define I387_MXCSR_INIT_VAL 0x1f80 +/* Format of XSAVE extended state is: + struct + { + fxsave_bytes[0..463] + sw_usable_bytes[464..511] + xstate_hdr_bytes[512..575] + extended state regions (AVX, MPX, AVX512, PKRU, etc.) + }; + + Same memory layout will be used for the coredump NT_X86_XSTATE + representing the XSAVE extended state registers. + + The first 8 bytes of the sw_usable_bytes[464..467] is the OS enabled + extended state mask, which is the same as the extended control register + 0 (the XFEATURE_ENABLED_MASK register), XCR0. We can use this mask + together with the mask saved in the xstate_hdr_bytes to determine what + states the processor/OS supports and what state, used or initialized, + the process/thread is in. */ +#define I386_LINUX_XSAVE_XCR0_OFFSET 464 + #endif /* COMMON_X86_XSTATE_H */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv4 06/10] gdb/gdbserver: share some code relating to target description creation 2024-04-05 12:33 ` [PATCHv4 00/10] " Andrew Burgess ` (4 preceding siblings ...) 2024-04-05 12:33 ` [PATCHv4 05/10] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess @ 2024-04-05 12:33 ` Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 07/10] gdb/arch: assert that X86_XSTATE_MPX is not set for x32 Andrew Burgess ` (6 subsequent siblings) 12 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-04-05 12:33 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess This commit is part of a series to share more of the x86 target description creation code between GDB and gdbserver. Unlike previous commits which were mostly refactoring, this commit is the first that makes a real change, though that change should mostly be for gdbserver; I've largely adopted the "GDB" way of doing things for gdbserver, and this fixes a real gdbserver bug. On a x86-64 Linux target, running the test: gdb.server/connect-with-no-symbol-file.exp results in two core files being created. Both of these core files are from the inferior process, created after gdbserver has detached. In this test a gdbserver process is started and then, after gdbserver has started, but before GDB attaches, we either delete the inferior executable, or change its permissions so it can't be read. Only after doing this do we attempt to connect with GDB. As GDB connects to gdbserver, gdbserver attempts to figure out the target description so that it can send the description to GDB, this involves a call to x86_linux_read_description. In x86_linux_read_description one of the first things we do is try to figure out if the process is 32-bit or 64-bit. To do this we look up the executable via the thread-id, and then attempt to read the architecture size from the executable. This isn't going to work if the executable has been deleted, or is no longer readable. And so, as we can't read the executable, we default to an i386 target and use an i386 target description. A consequence of using an i386 target description is that addresses are assumed to be 32-bits. Here's an example session that shows the problems this causes. This is run on an x86-64 machine, and the test binary (xx.x) is a standard 64-bit x86-64 binary: shell_1$ gdbserver --once localhost :54321 /tmp/xx.x shell_2$ gdb -q (gdb) set sysroot (gdb) shell chmod 000 /tmp/xx.x (gdb) target remote :54321 Remote debugging using :54321 warning: /tmp/xx.x: Permission denied. 0xf7fd3110 in ?? () (gdb) show architecture The target architecture is set to "auto" (currently "i386"). (gdb) p/x $pc $1 = 0xf7fd3110 (gdb) info proc mappings process 2412639 Mapped address spaces: Start Addr End Addr Size Offset Perms objfile 0x400000 0x401000 0x1000 0x0 r--p /tmp/xx.x 0x401000 0x402000 0x1000 0x1000 r-xp /tmp/xx.x 0x402000 0x403000 0x1000 0x2000 r--p /tmp/xx.x 0x403000 0x405000 0x2000 0x2000 rw-p /tmp/xx.x 0xf7fcb000 0xf7fcf000 0x4000 0x0 r--p [vvar] 0xf7fcf000 0xf7fd1000 0x2000 0x0 r-xp [vdso] 0xf7fd1000 0xf7fd3000 0x2000 0x0 r--p /usr/lib64/ld-2.30.so 0xf7fd3000 0xf7ff3000 0x20000 0x2000 r-xp /usr/lib64/ld-2.30.so 0xf7ff3000 0xf7ffb000 0x8000 0x22000 r--p /usr/lib64/ld-2.30.so 0xf7ffc000 0xf7ffe000 0x2000 0x2a000 rw-p /usr/lib64/ld-2.30.so 0xf7ffe000 0xf7fff000 0x1000 0x0 rw-p 0xfffda000 0xfffff000 0x25000 0x0 rw-p [stack] 0xff600000 0xff601000 0x1000 0x0 r-xp [vsyscall] (gdb) info inferiors Num Description Connection Executable * 1 process 2412639 1 (remote :54321) (gdb) shell cat /proc/2412639/maps 00400000-00401000 r--p 00000000 fd:03 45907133 /tmp/xx.x 00401000-00402000 r-xp 00001000 fd:03 45907133 /tmp/xx.x 00402000-00403000 r--p 00002000 fd:03 45907133 /tmp/xx.x 00403000-00405000 rw-p 00002000 fd:03 45907133 /tmp/xx.x 7ffff7fcb000-7ffff7fcf000 r--p 00000000 00:00 0 [vvar] 7ffff7fcf000-7ffff7fd1000 r-xp 00000000 00:00 0 [vdso] 7ffff7fd1000-7ffff7fd3000 r--p 00000000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7fd3000-7ffff7ff3000 r-xp 00002000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7ff3000-7ffff7ffb000 r--p 00022000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7ffc000-7ffff7ffe000 rw-p 0002a000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7ffe000-7ffff7fff000 rw-p 00000000 00:00 0 7ffffffda000-7ffffffff000 rw-p 00000000 00:00 0 [stack] ffffffffff600000-ffffffffff601000 r-xp 00000000 00:00 0 [vsyscall] (gdb) Notice the difference between the mappings reported via GDB and those reported directly from the kernel via /proc/PID/maps, the addresses of every mapping is clamped to 32-bits for GDB, while the kernel reports real 64-bit addresses. Notice also that the $pc value is a 32-bit value. It appears to be within one of the mappings reported by GDB, but is outside any of the mappings reported from the kernel. And this is where the problem arises. When gdbserver detaches from the inferior we pass the inferior the address from which it should resume. Due to the 32/64 bit confusion we tell the inferior to resume from the 32-bit $pc value, which is not within any valid mapping, and so, as soon as the inferior resumes, it segfaults. If we look at how GDB (not gdbserver) figures out its target description then we see an interesting difference. GDB doesn't try to read the executable. Instead GDB uses ptrace to query the thread's state, and uses this to figure out the if the thread is 32 or 64 bit. If we update gdbserver to do it the "GDB" way then the above problem is resolved, gdbserver now sees the process as 64-bit, and when we detach from the inferior we give it the correct 64-bit address, and the inferior no longer segfaults. Now, I could just update the gdbserver code, but better, I think, to share one copy of the code between GDB and gdbserver in gdb/nat/. That is what this commit does. The cores of x86_linux_read_description from gdbserver and x86_linux_nat_target::read_description from GDB are moved into a new file gdb/nat/x86-linux-tdesc.c and combined into a single function x86_linux_tdesc_for_tid which is called from each location. This new function does things the GDB way, the only changes are to allow for the sharing; we now have a callback function to call the first time that the xcr0 state is read, this allows for GDB and gdbserver to perform their own initialisation as needed, and additionally, the new function takes a pointer for where to cache the xcr0 value, this isn't needed for this commit, but will be useful in a later commit where gdbserver will want to read this cached xcr0 value. Another thing to note about this commit is how the functions i386_linux_read_description and amd64_linux_read_description are handled. For now I've left these function as implemented separately in GDB and gdbserver. I've moved the declarations of these functions into gdb/arch/{i386,amd64}-linux-tdesc.h, but the implementations are left where they are. A later commit in this series will make these functions shared too, but doing this is not trivial, so I've left that for a separate commit. Merging the declarations as I've done here ensures that everyone implements the function to the same API, and once these functions are shared (in a later commit) we'll want a shared declaration anyway. --- gdb/Makefile.in | 3 + gdb/amd64-linux-tdep.c | 1 + gdb/amd64-linux-tdep.h | 6 -- gdb/arch/amd64-linux-tdesc.h | 30 +++++++ gdb/arch/i386-linux-tdesc.h | 29 +++++++ gdb/configure.nat | 4 +- gdb/i386-linux-tdep.c | 1 + gdb/i386-linux-tdep.h | 3 - gdb/nat/x86-linux-tdesc.c | 120 ++++++++++++++++++++++++++++ gdb/nat/x86-linux-tdesc.h | 60 ++++++++++++++ gdb/x86-linux-nat.c | 91 ++++----------------- gdbserver/configure.srv | 2 + gdbserver/linux-amd64-ipa.cc | 1 + gdbserver/linux-i386-ipa.cc | 1 + gdbserver/linux-x86-low.cc | 151 +++++++++++------------------------ gdbserver/linux-x86-tdesc.cc | 2 + gdbserver/linux-x86-tdesc.h | 7 -- 17 files changed, 313 insertions(+), 199 deletions(-) create mode 100644 gdb/arch/amd64-linux-tdesc.h create mode 100644 gdb/arch/i386-linux-tdesc.h create mode 100644 gdb/nat/x86-linux-tdesc.c create mode 100644 gdb/nat/x86-linux-tdesc.h diff --git a/gdb/Makefile.in b/gdb/Makefile.in index a9f641c0659..f560ba023bf 100644 --- a/gdb/Makefile.in +++ b/gdb/Makefile.in @@ -1550,8 +1550,10 @@ HFILES_NO_SRCDIR = \ arch/aarch64-insn.h \ arch/aarch64-mte-linux.h \ arch/aarch64-scalable-linux.h \ + arch/amd64-linux-tdesc.h \ arch/arc.h \ arch/arm.h \ + arch/i386-linux-tdesc.h \ arch/i386.h \ arch/loongarch.h \ arch/ppc-linux-common.h \ @@ -1607,6 +1609,7 @@ HFILES_NO_SRCDIR = \ nat/x86-gcc-cpuid.h \ nat/x86-linux.h \ nat/x86-linux-dregs.h \ + nat/x86-linux-tdesc.h \ python/py-event.h \ python/py-events.h \ python/py-stopevent.h \ diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c index 9d560ac4fbf..bcb9868e79e 100644 --- a/gdb/amd64-linux-tdep.c +++ b/gdb/amd64-linux-tdep.c @@ -41,6 +41,7 @@ #include "arch/amd64.h" #include "target-descriptions.h" #include "expop.h" +#include "arch/amd64-linux-tdesc.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_AMD64 "syscalls/amd64-linux.xml" diff --git a/gdb/amd64-linux-tdep.h b/gdb/amd64-linux-tdep.h index 2003dcda78f..0ec49e7fe03 100644 --- a/gdb/amd64-linux-tdep.h +++ b/gdb/amd64-linux-tdep.h @@ -43,12 +43,6 @@ extern struct target_desc *tdesc_x32_linux; extern struct target_desc *tdesc_x32_avx_linux; extern struct target_desc *tdesc_x32_avx_avx512_linux; -/* Return the right amd64-linux target descriptions according to - XCR0_FEATURES_BIT and IS_X32. */ - -const target_desc *amd64_linux_read_description (uint64_t xcr0_features_bit, - bool is_x32); - /* Enum that defines the syscall identifiers for amd64 linux. Used for process record/replay, these will be translated into a gdb-canonical set of syscall ids in linux-record.c. */ diff --git a/gdb/arch/amd64-linux-tdesc.h b/gdb/arch/amd64-linux-tdesc.h new file mode 100644 index 00000000000..db425b60df6 --- /dev/null +++ b/gdb/arch/amd64-linux-tdesc.h @@ -0,0 +1,30 @@ +/* Target description related code for GNU/Linux x86-64. + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#ifndef ARCH_AMD64_LINUX_TDESC_H +#define ARCH_AMD64_LINUX_TDESC_H + +struct target_desc; + +/* Return the AMD64 target descriptions corresponding to XCR0 and IS_X32. */ + +extern const target_desc *amd64_linux_read_description (uint64_t xcr0, + bool is_x32); + +#endif /* ARCH_AMD64_LINUX_TDESC_H */ diff --git a/gdb/arch/i386-linux-tdesc.h b/gdb/arch/i386-linux-tdesc.h new file mode 100644 index 00000000000..0b736337a75 --- /dev/null +++ b/gdb/arch/i386-linux-tdesc.h @@ -0,0 +1,29 @@ +/* Target description related code for GNU/Linux i386. + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#ifndef ARCH_I386_LINUX_TDESC_H +#define ARCH_I386_LINUX_TDESC_H + +struct target_desc; + +/* Return the i386 target description corresponding to XCR0. */ + +extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); + +#endif /* ARCH_I386_LINUX_TDESC_H */ diff --git a/gdb/configure.nat b/gdb/configure.nat index 8b98511cef7..4bcc0696027 100644 --- a/gdb/configure.nat +++ b/gdb/configure.nat @@ -256,7 +256,7 @@ case ${gdb_host} in NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ nat/x86-xstate.o \ i386-linux-nat.o x86-linux-nat.o nat/linux-btrace.o \ - nat/x86-linux.o nat/x86-linux-dregs.o" + nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o" ;; ia64) # Host: Intel IA-64 running GNU/Linux @@ -322,7 +322,7 @@ case ${gdb_host} in NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ nat/x86-xstate.o amd64-nat.o amd64-linux-nat.o x86-linux-nat.o \ nat/linux-btrace.o \ - nat/x86-linux.o nat/x86-linux-dregs.o \ + nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o \ nat/amd64-linux-siginfo.o" ;; sparc) diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index 44730f204db..78ebc99d3df 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -39,6 +39,7 @@ #include "i387-tdep.h" #include "gdbsupport/x86-xstate.h" +#include "arch/i386-linux-tdesc.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_I386 "syscalls/i386-linux.xml" diff --git a/gdb/i386-linux-tdep.h b/gdb/i386-linux-tdep.h index 07593c6a8ec..e8691cd778e 100644 --- a/gdb/i386-linux-tdep.h +++ b/gdb/i386-linux-tdep.h @@ -55,9 +55,6 @@ extern void i386_linux_report_signal_info (struct gdbarch *gdbarch, struct ui_out *uiout, enum gdb_signal siggnal); -/* Return the target description according to XCR0. */ -extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); - extern int i386_linux_gregset_reg_offset[]; /* Return x86 siginfo type. */ diff --git a/gdb/nat/x86-linux-tdesc.c b/gdb/nat/x86-linux-tdesc.c new file mode 100644 index 00000000000..783a64857fe --- /dev/null +++ b/gdb/nat/x86-linux-tdesc.c @@ -0,0 +1,120 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#include "nat/x86-linux-tdesc.h" +#ifdef __x86_64__ +#include "arch/amd64.h" +#include "arch/amd64-linux-tdesc.h" +#endif +#include "arch/i386.h" +#include "arch/i386-linux-tdesc.h" + +#include "nat/x86-linux.h" +#include "nat/gdb_ptrace.h" +#include "nat/x86-xstate.h" + +#include <sys/uio.h> +#include <elf.h> + +#ifndef IN_PROCESS_AGENT + +/* See nat/x86-linux-tdesc.h. */ + +const target_desc * +x86_linux_tdesc_for_tid (int tid, enum tribool *have_ptrace_getregset, + gdb::function_view<void (uint64_t)> xcr0_init_cb, + const char *error_msg, uint64_t *xcr0_storage) +{ +#ifdef __x86_64__ + + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); + bool is_64bit = arch_size.is_64bit (); + bool is_x32 = arch_size.is_x32 (); + + if (sizeof (void *) == 4 && is_64bit && !is_x32) + error ("%s", error_msg); + +#elif HAVE_PTRACE_GETFPXREGS + if (have_ptrace_getfpxregs == -1) + { + elf_fpxregset_t fpxregs; + + if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) + { + have_ptrace_getfpxregs = 0; + *have_ptrace_getregset = TRIBOOL_FALSE; + return i386_linux_read_description (X86_XSTATE_X87_MASK); + } + } +#endif + + if (*have_ptrace_getregset == TRIBOOL_UNKNOWN) + { + uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; + struct iovec iov; + + iov.iov_base = xstateregs; + iov.iov_len = sizeof (xstateregs); + + /* Check if PTRACE_GETREGSET works. */ + if (ptrace (PTRACE_GETREGSET, tid, + (unsigned int) NT_X86_XSTATE, &iov) < 0) + { + *have_ptrace_getregset = TRIBOOL_FALSE; + *xcr0_storage = 0; + } + else + { + *have_ptrace_getregset = TRIBOOL_TRUE; + + /* Get XCR0 from XSAVE extended state. */ + *xcr0_storage = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET + / sizeof (uint64_t))]; + +#ifdef __x86_64__ + /* No MPX on x32. */ + if (is_64bit && is_x32) + *xcr0_storage &= ~X86_XSTATE_MPX; +#endif /* __x86_64__ */ + + xcr0_init_cb (*xcr0_storage); + } + } + + /* Check the native XCR0 only if PTRACE_GETREGSET is available. If + PTRACE_GETREGSET is not available then set xcr0_features_bits to + zero so that the "no-features" descriptions are returned by the + switches below. */ + uint64_t xcr0_features_bits; + if (*have_ptrace_getregset == TRIBOOL_TRUE) + xcr0_features_bits = *xcr0_storage & X86_XSTATE_ALL_MASK; + else + xcr0_features_bits = 0; + +#ifdef __x86_64__ + if (is_64bit) + { + return amd64_linux_read_description (xcr0_features_bits, is_x32); + } + else +#endif + return i386_linux_read_description (xcr0_features_bits); +} + +#endif /* !IN_PROCESS_AGENT */ diff --git a/gdb/nat/x86-linux-tdesc.h b/gdb/nat/x86-linux-tdesc.h new file mode 100644 index 00000000000..8c481609876 --- /dev/null +++ b/gdb/nat/x86-linux-tdesc.h @@ -0,0 +1,60 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#ifndef NAT_X86_LINUX_TDESC_H +#define NAT_X86_LINUX_TDESC_H + +#include "gdbsupport/function-view.h" + +struct target_desc; + +/* Return the target description for Linux thread TID. + + When *HAVE_PTRACE_GETREGSET is TRIBOOL_UNKNOWN then the current value of + xcr0 is read using ptrace calls and stored into *XCR0_STORAGE. Then + XCR0_INIT_CB is called with the value of *XCR0_STORAGE and + *HAVE_PTRACE_GETREGSET is set to TRIBOOL_TRUE. + + If the attempt to read xcr0 using ptrace fails then *XCR0_STORAGE is set + to zero and *HAVE_PTRACE_GETREGSET is set to TRIBOOL_FALSE. + + The storage pointed to by XCR0_STORAGE must exist until the program + terminates, this storage is used to cache the xcr0 value. As such + XCR0_INIT_CB will only be called once if xcr0 is successfully read using + ptrace, or not at all if the ptrace call fails. + + This function returns a target description based on the extracted xcr0 + value along with other characteristics of the thread identified by TID. + + This function can return nullptr if we encounter a machine configuration + for which a target_desc cannot be created. Ideally this would not be + the case, we should be able to create a target description for every + possible machine configuration. See amd64_linux_read_description and + i386_linux_read_description for cases when nullptr might be + returned. + + ERROR_MSG is using in an error() call if we try to create a target + description for a 64-bit process but this is a 32-bit build of GDB. */ + +extern const target_desc * +x86_linux_tdesc_for_tid (int tid, enum tribool *have_ptrace_getregset, + gdb::function_view<void (uint64_t)> xcr0_init_cb, + const char *error_msg, uint64_t *xcr0_storage); + +#endif /* NAT_X86_LINUX_TDESC_H */ diff --git a/gdb/x86-linux-nat.c b/gdb/x86-linux-nat.c index f91db492d05..57424fe7c3a 100644 --- a/gdb/x86-linux-nat.c +++ b/gdb/x86-linux-nat.c @@ -41,6 +41,7 @@ #include "nat/x86-linux.h" #include "nat/x86-linux-dregs.h" #include "nat/linux-ptrace.h" +#include "nat/x86-linux-tdesc.h" /* linux_nat_target::low_new_fork implementation. */ @@ -95,90 +96,26 @@ x86_linux_nat_target::post_startup_inferior (ptid_t ptid) const struct target_desc * x86_linux_nat_target::read_description () { - int tid; - int is_64bit = 0; -#ifdef __x86_64__ - int is_x32; -#endif - static uint64_t xcr0; - uint64_t xcr0_features_bits; + static uint64_t xcr0_storage; if (inferior_ptid == null_ptid) return this->beneath ()->read_description (); - tid = inferior_ptid.pid (); - -#ifdef __x86_64__ - - x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); - is_64bit = arch_size.is_64bit (); - is_x32 = arch_size.is_x32 (); - -#elif HAVE_PTRACE_GETFPXREGS - if (have_ptrace_getfpxregs == -1) - { - elf_fpxregset_t fpxregs; - - if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) - { - have_ptrace_getfpxregs = 0; - have_ptrace_getregset = TRIBOOL_FALSE; - return i386_linux_read_description (X86_XSTATE_X87_MASK); - } - } -#endif - - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) - { - uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; - struct iovec iov; - - iov.iov_base = xstateregs; - iov.iov_len = sizeof (xstateregs); - - /* Check if PTRACE_GETREGSET works. */ - if (ptrace (PTRACE_GETREGSET, tid, - (unsigned int) NT_X86_XSTATE, &iov) < 0) - have_ptrace_getregset = TRIBOOL_FALSE; - else - { - have_ptrace_getregset = TRIBOOL_TRUE; - - /* Get XCR0 from XSAVE extended state. */ - xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET - / sizeof (uint64_t))]; - - m_xsave_layout = x86_fetch_xsave_layout (xcr0, x86_xsave_length ()); - } - } - - /* Check the native XCR0 only if PTRACE_GETREGSET is available. If - PTRACE_GETREGSET is not available then set xcr0_features_bits to - zero so that the "no-features" descriptions are returned by the - switches below. */ - if (have_ptrace_getregset == TRIBOOL_TRUE) - xcr0_features_bits = xcr0 & X86_XSTATE_ALL_MASK; - else - xcr0_features_bits = 0; - - if (is_64bit) - { -#ifdef __x86_64__ - return amd64_linux_read_description (xcr0_features_bits, is_x32); -#endif - } - else - { - const struct target_desc * tdesc - = i386_linux_read_description (xcr0_features_bits); + int tid = inferior_ptid.pid (); - if (tdesc == NULL) - tdesc = i386_linux_read_description (X86_XSTATE_SSE_MASK); + const char *error_msg + = _("Can't debug 64-bit process with 32-bit GDB"); - return tdesc; - } + /* Callback that is triggered the first time x86_linux_tdesc_for_tid + reads the xcr0 register. Setup other bits of state. */ + auto cb = [&] (uint64_t xcr0) + { + this->m_xsave_layout + = x86_fetch_xsave_layout (xcr0, x86_xsave_length ()); + }; - gdb_assert_not_reached ("failed to return tdesc"); + return x86_linux_tdesc_for_tid (tid, &have_ptrace_getregset, cb, + error_msg, &xcr0_storage); } \f diff --git a/gdbserver/configure.srv b/gdbserver/configure.srv index 9e861a75088..7a2702d78bf 100644 --- a/gdbserver/configure.srv +++ b/gdbserver/configure.srv @@ -109,6 +109,7 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/linux-btrace.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" + srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" srv_linux_usrregs=yes srv_linux_regsets=yes srv_linux_thread_db=yes @@ -371,6 +372,7 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/linux-btrace.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" + srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" srv_tgtobj="${srv_tgtobj} nat/amd64-linux-siginfo.o" srv_linux_usrregs=yes # This is for i386 progs. srv_linux_regsets=yes diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index a6346750f49..df5e6aca081 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -22,6 +22,7 @@ #include "tracepoint.h" #include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" +#include "arch/amd64-linux-tdesc.h" /* fast tracepoints collect registers. */ diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc index 8f14e0937d4..aa346fc9bc3 100644 --- a/gdbserver/linux-i386-ipa.cc +++ b/gdbserver/linux-i386-ipa.cc @@ -22,6 +22,7 @@ #include "tracepoint.h" #include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" +#include "arch/i386-linux-tdesc.h" /* GDB register numbers. */ diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index e8ef3667eb4..37c15aa688b 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -29,8 +29,11 @@ #ifdef __x86_64__ #include "nat/amd64-linux-siginfo.h" +#include "arch/amd64-linux-tdesc.h" #endif +#include "arch/i386-linux-tdesc.h" + #include "gdb_proc_service.h" /* Don't include elf/common.h if linux/elf.h got included by gdb_proc_service.h. */ @@ -46,6 +49,7 @@ #include "nat/x86-linux.h" #include "nat/x86-linux-dregs.h" #include "linux-x86-tdesc.h" +#include "nat/x86-linux-tdesc.h" #ifdef __x86_64__ static target_desc_up tdesc_amd64_linux_no_xml; @@ -842,32 +846,20 @@ int have_ptrace_getfpxregs = #endif ; +/* Cached xcr0 value. This is initialised the first time + x86_linux_read_description is called. */ + +static uint64_t xcr0_storage; + /* Get Linux/x86 target description from running target. */ static const struct target_desc * x86_linux_read_description (void) { - unsigned int machine; - int is_elf64; - int xcr0_features; - int tid; - static uint64_t xcr0; - static int xsave_len; - struct regset_info *regset; - - tid = lwpid_of (current_thread); - - is_elf64 = linux_pid_exe_is_elf_64_file (tid, &machine); + int tid = lwpid_of (current_thread); - if (sizeof (void *) == 4) - { - if (is_elf64 > 0) - error (_("Can't debug 64-bit process with 32-bit GDBserver")); -#ifndef __x86_64__ - else if (machine == EM_X86_64) - error (_("Can't debug x86-64 process with 32-bit GDBserver")); -#endif - } + const char *error_msg + = _("Can't debug 64-bit process with 32-bit GDBserver"); /* If we are not allowed to send an XML target description then we need to use the hard-wired target descriptions. This corresponds to GDB's @@ -877,103 +869,54 @@ x86_linux_read_description (void) generate some alternative target descriptions. */ if (!use_xml) { + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); + bool is_64bit = arch_size.is_64bit (); + bool is_x32 = arch_size.is_x32 (); + + if (sizeof (void *) == 4 && is_64bit && !is_x32) + error ("%s", error_msg); + #ifdef __x86_64__ - if (machine == EM_X86_64) + if (is_64bit && !is_x32) return tdesc_amd64_linux_no_xml.get (); else #endif return tdesc_i386_linux_no_xml.get (); } -#if !defined __x86_64__ && defined HAVE_PTRACE_GETFPXREGS - if (machine == EM_386 && have_ptrace_getfpxregs == -1) - { - elf_fpxregset_t fpxregs; - - if (ptrace (PTRACE_GETFPXREGS, tid, 0, (long) &fpxregs) < 0) - { - have_ptrace_getfpxregs = 0; - have_ptrace_getregset = TRIBOOL_FALSE; - return i386_linux_read_description (X86_XSTATE_X87); - } - else - have_ptrace_getfpxregs = 1; - } -#endif - - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) - { - uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; - struct iovec iov; - - iov.iov_base = xstateregs; - iov.iov_len = sizeof (xstateregs); - - /* Check if PTRACE_GETREGSET works. */ - if (ptrace (PTRACE_GETREGSET, tid, - (unsigned int) NT_X86_XSTATE, (long) &iov) < 0) - have_ptrace_getregset = TRIBOOL_FALSE; - else - { - have_ptrace_getregset = TRIBOOL_TRUE; - - /* Get XCR0 from XSAVE extended state. */ - xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET - / sizeof (uint64_t))]; - - /* No MPX on x32. */ - if (machine == EM_X86_64 && !is_elf64) - xcr0 &= ~X86_XSTATE_MPX; - - xsave_len = x86_xsave_length (); - - /* Use PTRACE_GETREGSET if it is available. */ - for (regset = x86_regsets; - regset->fill_function != NULL; regset++) - if (regset->get_request == PTRACE_GETREGSET) - regset->size = xsave_len; - else if (regset->type != GENERAL_REGS) - regset->size = 0; - } - } - - /* Check the native XCR0 only if PTRACE_GETREGSET is available. */ - xcr0_features = (have_ptrace_getregset == TRIBOOL_TRUE - && (xcr0 & X86_XSTATE_ALL_MASK)); - - if (xcr0_features) - i387_set_xsave_mask (xcr0, xsave_len); + /* Callback that is triggered the first time x86_linux_tdesc_for_tid + reads the xcr0 register. Setup other bits of state */ + auto cb = [] (uint64_t xcr0) + { + i387_set_xsave_mask (xcr0, x86_xsave_length ()); + }; - if (machine == EM_X86_64) - { -#ifdef __x86_64__ - const target_desc *tdesc = NULL; + /* If have_ptrace_getregset is changed to true by calling + x86_linux_tdesc_for_tid then we will perform some additional + initialisation. */ + bool have_ptrace_getregset_is_unknown + = have_ptrace_getregset == TRIBOOL_UNKNOWN; - if (xcr0_features) - { - tdesc = amd64_linux_read_description (xcr0 & X86_XSTATE_ALL_MASK, - !is_elf64); - } + const target_desc *tdesc + = x86_linux_tdesc_for_tid (tid, &have_ptrace_getregset, cb, error_msg, + &xcr0_storage); - if (tdesc == NULL) - tdesc = amd64_linux_read_description (X86_XSTATE_SSE_MASK, !is_elf64); - return tdesc; -#endif - } - else + if (have_ptrace_getregset_is_unknown + && have_ptrace_getregset == TRIBOOL_TRUE) { - const target_desc *tdesc = NULL; - - if (xcr0_features) - tdesc = i386_linux_read_description (xcr0 & X86_XSTATE_ALL_MASK); - - if (tdesc == NULL) - tdesc = i386_linux_read_description (X86_XSTATE_SSE); - - return tdesc; + int xsave_len = x86_xsave_length (); + + /* Use PTRACE_GETREGSET if it is available. */ + for (regset_info *regset = x86_regsets; + regset->fill_function != nullptr; + regset++) + if (regset->get_request == PTRACE_GETREGSET) + regset->size = xsave_len; + else if (regset->type != GENERAL_REGS) + regset->size = 0; } - gdb_assert_not_reached ("failed to return tdesc"); + return tdesc; } /* Update all the target description of all processes; a new GDB diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc index cd3b5d80e37..af3735aa895 100644 --- a/gdbserver/linux-x86-tdesc.cc +++ b/gdbserver/linux-x86-tdesc.cc @@ -23,8 +23,10 @@ #include "gdbsupport/x86-xstate.h" #ifdef __x86_64__ #include "arch/amd64.h" +#include "arch/amd64-linux-tdesc.h" #endif #include "x86-tdesc.h" +#include "arch/i386-linux-tdesc.h" /* Return the right x86_linux_tdesc index for a given XCR0. Return X86_TDESC_LAST if can't find a match. */ diff --git a/gdbserver/linux-x86-tdesc.h b/gdbserver/linux-x86-tdesc.h index f9561b129ae..576aaf5e165 100644 --- a/gdbserver/linux-x86-tdesc.h +++ b/gdbserver/linux-x86-tdesc.h @@ -46,11 +46,4 @@ int amd64_get_ipa_tdesc_idx (const struct target_desc *tdesc); const struct target_desc *i386_get_ipa_tdesc (int idx); -#ifdef __x86_64__ -const struct target_desc *amd64_linux_read_description (uint64_t xcr0, - bool is_x32); -#endif - -const struct target_desc *i386_linux_read_description (uint64_t xcr0); - #endif /* GDBSERVER_LINUX_X86_TDESC_H */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv4 07/10] gdb/arch: assert that X86_XSTATE_MPX is not set for x32 2024-04-05 12:33 ` [PATCHv4 00/10] " Andrew Burgess ` (5 preceding siblings ...) 2024-04-05 12:33 ` [PATCHv4 06/10] gdb/gdbserver: share some code relating to target description creation Andrew Burgess @ 2024-04-05 12:33 ` Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 08/10] gdbserver: update target description creation for x86/linux Andrew Burgess ` (5 subsequent siblings) 12 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-04-05 12:33 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess While rebasing this series past this commit: commit 4bb20a6244b7091a9a7a2ae35dfbd7e8db27550a Date: Wed Mar 20 04:13:18 2024 -0700 gdbserver: Clear X86_XSTATE_MPX bits in xcr0 on x32 I worried that there could be other paths that might result in an xcr0 value which has X86_XSTATE_MPX set in x32 mode. As everyone eventually calls amd64_create_target_description to build their target description, I figured we could assert in here that if X86_XSTATE_MPX is set then we should not be an x32 target, this will uncover any other bugs in this area. I'm not currently able to build/run any x32 binaries, so I have no way to test this, but the author of commit 4bb20a6244b7091 did test this series with that assert in place and didn't see any problems. Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=31511 --- gdb/arch/amd64.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-) diff --git a/gdb/arch/amd64.c b/gdb/arch/amd64.c index cb9683c6931..94d55d72a4e 100644 --- a/gdb/arch/amd64.c +++ b/gdb/arch/amd64.c @@ -65,8 +65,12 @@ amd64_create_target_description (uint64_t xcr0, bool is_x32, bool is_linux, if (xcr0 & X86_XSTATE_AVX) regnum = create_feature_i386_64bit_avx (tdesc.get (), regnum); - if ((xcr0 & X86_XSTATE_MPX) && !is_x32) - regnum = create_feature_i386_64bit_mpx (tdesc.get (), regnum); + if (xcr0 & X86_XSTATE_MPX) + { + /* MPX is not available on x32. */ + gdb_assert (!is_x32); + regnum = create_feature_i386_64bit_mpx (tdesc.get (), regnum); + } if (xcr0 & X86_XSTATE_AVX512) regnum = create_feature_i386_64bit_avx512 (tdesc.get (), regnum); -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv4 08/10] gdbserver: update target description creation for x86/linux 2024-04-05 12:33 ` [PATCHv4 00/10] " Andrew Burgess ` (6 preceding siblings ...) 2024-04-05 12:33 ` [PATCHv4 07/10] gdb/arch: assert that X86_XSTATE_MPX is not set for x32 Andrew Burgess @ 2024-04-05 12:33 ` Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 09/10] gdb: move xcr0 == 0 check into i386_linux_core_read_description Andrew Burgess ` (4 subsequent siblings) 12 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-04-05 12:33 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, John Baldwin This commit is part of a series which aims to share more of the target description creation between GDB and gdbserver for x86/Linux. After some refactoring earlier in this series the shared x86_linux_tdesc_for_tid function was added into nat/x86-linux-tdesc.c. However, this function still relies on amd64_linux_read_description and i386_linux_read_description which are implemented separately for both gdbserver and GDB. Given that at their core, all these functions do is: 1. take an xcr0 value as input, 2. mask out some feature bits, 3. look for a cached pre-generated target description and return it if found, 4. if no cached target description is found then call either amd64_create_target_description or i386_create_target_description to create a new target description, which is then added to the cache. Return the newly created target description. The inner functions amd64_create_target_description and i386_create_target_description are already shared between GDB and gdbserver (in the gdb/arch/ directory), so the only thing that the *_read_description functions really do is add the caching layer, and it feels like this really could be shared. However, we have a small problem. On the GDB side we create target descriptions using a different set of cpu features than on the gdbserver side! This means that for the exact same target, we might get a different target description when using native GDB vs using gdbserver. This surely feels like a mistake, I would expect to get the same target description on each. The table below shows the number of possible different target descriptions that we can create on the GDB side vs on the gdbserver side for each target type: | GDB | gdbserver ------|-----|---------- i386 | 64 | 7 amd64 | 32 | 7 x32 | 16 | 7 So in theory, all I want to do is move the GDB version of *_read_description into the arch/ directory and have gdbserver use that, then both GDB and gdbserver would be able to create any of the possible target descriptions. Unfortunately it's a little more complex than that due to the in process agent (IPA). When the IPA is in use, gdbserver sends a target description index to the IPA, and the IPA uses this to find the correct target description to use. ** START OF AN ASIDE ** Back in the day I suspect this approach made perfect sense. However since this commit: commit a8806230241d201f808d856eaae4d44088117b0c Date: Thu Dec 7 17:07:01 2017 +0000 Initialize target description early in IPA I think passing the index is now more trouble than its worth. We used to pass the index, and then use that index to lookup which target description to instantiate and use. However, the above commit fixed an issue where we can't call malloc() within (certain parts of) the IPA (apparently), so instead we now pre-compute _every_ possible target description within the IPA. The index is now only used to lookup which of the (many) pre-computed target descriptions to use. It would (I think) have been easier all around if the IPA just self-inspected, figured out its own xcr0 value, and used that to create the one target description that is required. So long as the xcr0 to target description code is shared (at compile time) with gdbserver, then we can be sure that the IPA will derive the same target description as gdbserver, and we would avoid all this index passing business, which has made this commit so very, very painful. ** END OF AN ASIDE ** Currently then for x86/linux, gdbserver sends a number between 0 and 7 to the IPA, and the IPA uses this to create a target description. However, I am proposing that gdbserver should now create one of (up to) 64 different target descriptions for i386, so this 0 to 7 index isn't going to be good enough any more (amd64 and x32 have slightly fewer possible target descriptions, but still more than 8, so the problem is the same). For a while I wondered if I was going to have to try and find some backward compatible solution for this mess. But after seeing how lightly the IPA is actually documented, I wonder if it is not the case that there is a tight coupling between a version of gdbserver and a version of the IPA? At least I'm hoping so. In this commit I have thrown out the old IPA target description index numbering scheme, and switched to a completely new numbering scheme. Instead of the index that is passed being arbitrary, the index is instead calculated from the set of cpu features that are present on the target. Within the IPA we can then reverse this logic to recreate the xcr0 value based on the index, and from the xcr0 value we can choose the correct target description. With the gdbserver to IPA numbering scheme issue resolved I have then update the gdbserver versions of amd64_linux_read_description and i386_linux_read_description so that they create target descriptions using the same set of cpu features as GDB itself. After this gdbserver should now always come up with the same target description as GDB does on any x86/Linux target. This commit does not introduce any new code sharing between GDB and gdbserver as previous commits in this series have done. Instead this commit is all about bringing GDB and gdbserver into alignment functionally so that the next commit(s) can merge the GDB and gdbserver versions of these functions. Approved-By: John Baldwin <jhb@FreeBSD.org> --- gdbserver/linux-amd64-ipa.cc | 43 +---- gdbserver/linux-i386-ipa.cc | 23 +-- gdbserver/linux-x86-low.cc | 15 +- gdbserver/linux-x86-tdesc.cc | 316 +++++++++++++++++++++++++---------- gdbserver/linux-x86-tdesc.h | 49 +++--- 5 files changed, 278 insertions(+), 168 deletions(-) diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index df5e6aca081..0c80812cc6f 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -164,47 +164,21 @@ supply_static_tracepoint_registers (struct regcache *regcache, #endif /* HAVE_UST */ -#if !defined __ILP32__ -/* Map the tdesc index to xcr0 mask. */ -static uint64_t idx2mask[X86_TDESC_LAST] = { - X86_XSTATE_X87_MASK, - X86_XSTATE_SSE_MASK, - X86_XSTATE_AVX_MASK, - X86_XSTATE_MPX_MASK, - X86_XSTATE_AVX_MPX_MASK, - X86_XSTATE_AVX_AVX512_MASK, - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, -}; -#endif - /* Return target_desc to use for IPA, given the tdesc index passed by gdbserver. */ const struct target_desc * get_ipa_tdesc (int idx) { - if (idx >= X86_TDESC_LAST) - { - internal_error ("unknown ipa tdesc index: %d", idx); - } + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); #if defined __ILP32__ - switch (idx) - { - case X86_TDESC_SSE: - return amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); - case X86_TDESC_AVX: - return amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); - case X86_TDESC_AVX_AVX512: - return amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, true); - default: - break; - } + bool is_x32 = true; #else - return amd64_linux_read_description (idx2mask[idx], false); + bool is_x32 = false; #endif - internal_error ("unknown ipa tdesc index: %d", idx); + return amd64_linux_read_description (xcr0, is_x32); } /* Allocate buffer for the jump pads. The branch instruction has a @@ -272,11 +246,10 @@ void initialize_low_tracepoint (void) { #if defined __ILP32__ - amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); - amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); - amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, true); + for (auto i = 0; i < x86_linux_x32_tdesc_count (); i++) + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), true); #else - for (auto i = 0; i < X86_TDESC_LAST; i++) - amd64_linux_read_description (idx2mask[i], false); + for (auto i = 0; i < x86_linux_amd64_tdesc_count (); i++) + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), false); #endif } diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc index aa346fc9bc3..c1c3152fb04 100644 --- a/gdbserver/linux-i386-ipa.cc +++ b/gdbserver/linux-i386-ipa.cc @@ -245,28 +245,15 @@ initialize_fast_tracepoint_trampoline_buffer (void) } } -/* Map the tdesc index to xcr0 mask. */ -static uint64_t idx2mask[X86_TDESC_LAST] = { - X86_XSTATE_X87_MASK, - X86_XSTATE_SSE_MASK, - X86_XSTATE_AVX_MASK, - X86_XSTATE_MPX_MASK, - X86_XSTATE_AVX_MPX_MASK, - X86_XSTATE_AVX_AVX512_MASK, - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, -}; - /* Return target_desc to use for IPA, given the tdesc index passed by gdbserver. */ const struct target_desc * get_ipa_tdesc (int idx) { - if (idx >= X86_TDESC_LAST) - { - internal_error ("unknown ipa tdesc index: %d", idx); - } - return i386_linux_read_description (idx2mask[idx]); + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); + + return i386_linux_read_description (xcr0); } /* Allocate buffer for the jump pads. On i386, we can reach an arbitrary @@ -288,6 +275,6 @@ void initialize_low_tracepoint (void) { initialize_fast_tracepoint_trampoline_buffer (); - for (auto i = 0; i < X86_TDESC_LAST; i++) - i386_linux_read_description (idx2mask[i]); + for (auto i = 0; i < x86_linux_i386_tdesc_count (); i++) + i386_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i)); } diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 37c15aa688b..4f473730efb 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -2893,14 +2893,17 @@ x86_target::get_ipa_tdesc_idx () struct regcache *regcache = get_thread_regcache (current_thread, 0); const struct target_desc *tdesc = regcache->tdesc; + if (!use_xml) + { + if (tdesc == tdesc_i386_linux_no_xml.get () #ifdef __x86_64__ - return amd64_get_ipa_tdesc_idx (tdesc); -#endif - - if (tdesc == tdesc_i386_linux_no_xml.get ()) - return X86_TDESC_SSE; + || tdesc == tdesc_amd64_linux_no_xml.get () +#endif /* __x86_64__ */ + ) + return x86_linux_xcr0_to_tdesc_idx (X86_XSTATE_SSE_MASK); + } - return i386_get_ipa_tdesc_idx (tdesc); + return x86_linux_xcr0_to_tdesc_idx (xcr0_storage); } /* The linux target ops object. */ diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc index af3735aa895..5e12526bf17 100644 --- a/gdbserver/linux-x86-tdesc.cc +++ b/gdbserver/linux-x86-tdesc.cc @@ -28,96 +28,278 @@ #include "x86-tdesc.h" #include "arch/i386-linux-tdesc.h" -/* Return the right x86_linux_tdesc index for a given XCR0. Return - X86_TDESC_LAST if can't find a match. */ +/* A structure used to describe a single cpu feature that might, or might + not, be checked for when creating a target description for one of i386, + amd64, or x32. */ -static enum x86_linux_tdesc -xcr0_to_tdesc_idx (uint64_t xcr0, bool is_x32) +struct x86_tdesc_feature { + /* The cpu feature mask. This is a mask against an xcr0 value. */ + uint64_t feature; + + /* Is this feature checked when creating an i386 target description. */ + bool is_i386; + + /* Is this feature checked when creating an amd64 target description. */ + bool is_amd64; + + /* Is this feature checked when creating an x32 target description. */ + bool is_x32; +}; + +/* A constant table that describes all of the cpu features that are + checked when building a target description for i386, amd64, or x32. */ + +static constexpr x86_tdesc_feature x86_linux_all_tdesc_features[] = { + /* Feature, i386, amd64, x32. */ + { X86_XSTATE_PKRU, true, true, true }, + { X86_XSTATE_AVX512, true, true, true }, + { X86_XSTATE_AVX, true, true, true }, + { X86_XSTATE_MPX, true, true, false }, + { X86_XSTATE_SSE, true, false, false }, + { X86_XSTATE_X87, true, false, false } +}; + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an i386 target description. */ + +static constexpr uint64_t +x86_linux_i386_tdesc_feature_mask () { - if (xcr0 & X86_XSTATE_PKRU) - { - if (is_x32) - { - /* No x32 MPX and PKU, fall back to avx_avx512. */ - return X86_TDESC_AVX_AVX512; - } - else - return X86_TDESC_AVX_MPX_AVX512_PKU; - } - else if (xcr0 & X86_XSTATE_AVX512) - return X86_TDESC_AVX_AVX512; - else if ((xcr0 & X86_XSTATE_AVX_MPX_MASK) == X86_XSTATE_AVX_MPX_MASK) + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_i386) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an amd64 target description. */ + +static constexpr uint64_t +x86_linux_amd64_tdesc_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_amd64) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an x32 target description. */ + +static constexpr uint64_t +x86_linux_x32_tdesc_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_x32) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an i386 target description. */ + +static constexpr int +x86_linux_i386_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_i386) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an amd64 target description. */ + +static constexpr int +x86_linux_amd64_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_amd64) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an x32 target description. */ + +static constexpr int +x86_linux_x32_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_x32) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +#ifdef IN_PROCESS_AGENT + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_amd64_tdesc_count () +{ + return x86_linux_amd64_tdesc_count_1 (); +} + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_x32_tdesc_count () +{ + return x86_linux_x32_tdesc_count_1 (); +} + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_i386_tdesc_count () +{ + return x86_linux_i386_tdesc_count_1 (); +} + +#endif /* IN_PROCESS_AGENT */ + +/* Convert an xcr0 value into an integer. The integer will be passed to + the in-process-agent where it will then be passed to + x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ + +int +x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0) +{ + /* The following table shows which features are checked for when creating + the target descriptions (see nat/x86-linux-tdesc.c), the feature order + represents the bit order within the generated index number. + + i386 | x87 sse mpx avx avx512 pkru + amd64 | mpx avx avx512 pkru + i32 | avx avx512 pkru + + The features are ordered so that for each mode (i386, amd64, i32) the + generated index will form a continuous range. */ + + int idx = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) { - if (is_x32) /* No MPX on x32. */ - return X86_TDESC_AVX; - else - return X86_TDESC_AVX_MPX; + if ((xcr0 & x86_linux_all_tdesc_features[i].feature) + == x86_linux_all_tdesc_features[i].feature) + idx |= (1 << i); } - else if (xcr0 & X86_XSTATE_MPX) + + return idx; +} + + +#ifdef IN_PROCESS_AGENT + +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +uint64_t +x86_linux_tdesc_idx_to_xcr0 (int idx) +{ + uint64_t xcr0 = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) { - if (is_x32) /* No MPX on x32. */ - return X86_TDESC_AVX; - else - return X86_TDESC_MPX; + if ((idx & (1 << i)) != 0) + xcr0 |= x86_linux_all_tdesc_features[i].feature; } - else if (xcr0 & X86_XSTATE_AVX) - return X86_TDESC_AVX; - else if (xcr0 & X86_XSTATE_SSE) - return X86_TDESC_SSE; - else if (xcr0 & X86_XSTATE_X87) - return X86_TDESC_MMX; - else - return X86_TDESC_LAST; + + return xcr0; } +#endif /* IN_PROCESS_AGENT */ + #if defined __i386__ || !defined IN_PROCESS_AGENT -static struct target_desc *i386_tdescs[X86_TDESC_LAST] = { }; +/* A cache of all possible i386 target descriptions. */ -/* Return the target description according to XCR0. */ +static struct target_desc *i386_tdescs[x86_linux_i386_tdesc_count_1 ()] = { }; + +/* See nat/x86-linux-tdesc.h. */ const struct target_desc * i386_linux_read_description (uint64_t xcr0) { - enum x86_linux_tdesc idx = xcr0_to_tdesc_idx (xcr0, false); + xcr0 &= x86_linux_i386_tdesc_feature_mask (); + int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - if (idx == X86_TDESC_LAST) - return NULL; + gdb_assert (idx >= 0 && idx < x86_linux_i386_tdesc_count_1 ()); - struct target_desc **tdesc = &i386_tdescs[idx]; + target_desc **tdesc = &i386_tdescs[idx]; - if (*tdesc == NULL) + if (*tdesc == nullptr) { *tdesc = i386_create_target_description (xcr0, true, false); init_target_desc (*tdesc, i386_expedite_regs); } - return *tdesc;; + return *tdesc; } #endif #ifdef __x86_64__ -static target_desc *amd64_tdescs[X86_TDESC_LAST] = { }; -static target_desc *x32_tdescs[X86_TDESC_LAST] = { }; +/* A cache of all possible amd64 target descriptions. */ + +static target_desc *amd64_tdescs[x86_linux_amd64_tdesc_count_1 ()] = { }; + +/* A cache of all possible x32 target descriptions. */ + +static target_desc *x32_tdescs[x86_linux_x32_tdesc_count_1 ()] = { }; + +/* See nat/x86-linux-tdesc.h. */ const struct target_desc * amd64_linux_read_description (uint64_t xcr0, bool is_x32) { - enum x86_linux_tdesc idx = xcr0_to_tdesc_idx (xcr0, is_x32); + if (is_x32) + xcr0 &= x86_linux_x32_tdesc_feature_mask (); + else + xcr0 &= x86_linux_amd64_tdesc_feature_mask (); + + int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - if (idx == X86_TDESC_LAST) - return NULL; + if (is_x32) + gdb_assert (idx >= 0 && idx < x86_linux_x32_tdesc_count_1 ()); + else + gdb_assert (idx >= 0 && idx < x86_linux_amd64_tdesc_count_1 ()); - struct target_desc **tdesc = NULL; + target_desc **tdesc = nullptr; if (is_x32) tdesc = &x32_tdescs[idx]; else tdesc = &amd64_tdescs[idx]; - if (*tdesc == NULL) + if (*tdesc == nullptr) { *tdesc = amd64_create_target_description (xcr0, is_x32, true, true); @@ -127,39 +309,3 @@ amd64_linux_read_description (uint64_t xcr0, bool is_x32) } #endif - -#ifndef IN_PROCESS_AGENT - -int -i386_get_ipa_tdesc_idx (const struct target_desc *tdesc) -{ - for (int i = 0; i < X86_TDESC_LAST; i++) - { - if (tdesc == i386_tdescs[i]) - return i; - } - - /* If none tdesc is found, return the one with minimum features. */ - return X86_TDESC_MMX; -} - -#if defined __x86_64__ -int -amd64_get_ipa_tdesc_idx (const struct target_desc *tdesc) -{ - for (int i = 0; i < X86_TDESC_LAST; i++) - { - if (tdesc == amd64_tdescs[i]) - return i; - } - for (int i = 0; i < X86_TDESC_LAST; i++) - { - if (tdesc == x32_tdescs[i]) - return i; - } - - return X86_TDESC_SSE; -} - -#endif -#endif diff --git a/gdbserver/linux-x86-tdesc.h b/gdbserver/linux-x86-tdesc.h index 576aaf5e165..3d6e0e51833 100644 --- a/gdbserver/linux-x86-tdesc.h +++ b/gdbserver/linux-x86-tdesc.h @@ -21,29 +21,30 @@ #ifndef GDBSERVER_LINUX_X86_TDESC_H #define GDBSERVER_LINUX_X86_TDESC_H -/* Note: since IPA obviously knows what ABI it's running on (i386 vs x86_64 - vs x32), it's sufficient to pass only the register set here. This, - together with the ABI known at IPA compile time, maps to a tdesc. */ - -enum x86_linux_tdesc { - X86_TDESC_MMX = 0, - X86_TDESC_SSE = 1, - X86_TDESC_AVX = 2, - X86_TDESC_MPX = 3, - X86_TDESC_AVX_MPX = 4, - X86_TDESC_AVX_AVX512 = 5, - X86_TDESC_AVX_MPX_AVX512_PKU = 6, - X86_TDESC_LAST = 7, -}; - -#if defined __i386__ || !defined IN_PROCESS_AGENT -int i386_get_ipa_tdesc_idx (const struct target_desc *tdesc); -#endif - -#if defined __x86_64__ && !defined IN_PROCESS_AGENT -int amd64_get_ipa_tdesc_idx (const struct target_desc *tdesc); -#endif - -const struct target_desc *i386_get_ipa_tdesc (int idx); +/* Convert an xcr0 value into an integer. The integer will be passed to + the in-process-agent where it will then be passed to + x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ + +extern int x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0); + +#ifdef IN_PROCESS_AGENT + +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +extern uint64_t x86_linux_tdesc_idx_to_xcr0 (int idx); + +/* Within the in-process-agent we need to pre-initialise all of the target + descriptions, to do this we need to know how many target descriptions + there are for each different target type. These functions return the + target description count for the relevant target. */ + +extern int x86_linux_amd64_tdesc_count (); +extern int x86_linux_x32_tdesc_count (); +extern int x86_linux_i386_tdesc_count (); + + +#endif /* IN_PROCESS_AGENT */ #endif /* GDBSERVER_LINUX_X86_TDESC_H */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv4 09/10] gdb: move xcr0 == 0 check into i386_linux_core_read_description 2024-04-05 12:33 ` [PATCHv4 00/10] " Andrew Burgess ` (7 preceding siblings ...) 2024-04-05 12:33 ` [PATCHv4 08/10] gdbserver: update target description creation for x86/linux Andrew Burgess @ 2024-04-05 12:33 ` Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 10/10] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess ` (3 subsequent siblings) 12 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-04-05 12:33 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess Currently, in i386_linux_core_read_description, if GDB fails to extract an xcr0 value from the core file, then we will have a default zero value for the xcr0 variable, we still call the i386_linux_read_description function, which checks for this zero value and returns nullptr. Back in i386_linux_core_read_description we spot the nullptr return value from i386_linux_read_description and call i386_linux_read_description again, but this time passing a default value for xcr0. In the next commit I plan to rework i386_linux_read_description, and in so doing I will remove the check for xcr0 == 0, this is inline with how the amd64 code is written. However, this means that the 'xcr0 == 0' check needs to move up the stack to i386_linux_core_read_description, again, this brings the i386 code into line with the amd64 code. This is just a refactor in preparation for the next commit, there should be no user visible changes after this commit. --- gdb/i386-linux-tdep.c | 15 ++++++++------- 1 file changed, 8 insertions(+), 7 deletions(-) diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index 78ebc99d3df..c796f87780b 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -714,15 +714,16 @@ i386_linux_core_read_description (struct gdbarch *gdbarch, /* Linux/i386. */ x86_xsave_layout layout; uint64_t xcr0 = i386_linux_core_read_xsave_info (abfd, layout); - const struct target_desc *tdesc = i386_linux_read_description (xcr0); - if (tdesc != NULL) - return tdesc; + if (xcr0 == 0) + { + if (bfd_get_section_by_name (abfd, ".reg-xfp") != nullptr) + xcr0 = X86_XSTATE_SSE_MASK; + else + xcr0 = X86_XSTATE_X87_MASK; + } - if (bfd_get_section_by_name (abfd, ".reg-xfp") != NULL) - return i386_linux_read_description (X86_XSTATE_SSE_MASK); - else - return i386_linux_read_description (X86_XSTATE_X87_MASK); + return i386_linux_read_description (xcr0); } /* Similar to i386_supply_fpregset, but use XSAVE extended state. */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv4 10/10] gdb/gdbserver: share x86/linux tdesc caching 2024-04-05 12:33 ` [PATCHv4 00/10] " Andrew Burgess ` (8 preceding siblings ...) 2024-04-05 12:33 ` [PATCHv4 09/10] gdb: move xcr0 == 0 check into i386_linux_core_read_description Andrew Burgess @ 2024-04-05 12:33 ` Andrew Burgess 2024-04-09 18:37 ` [PATCHv4 00/10] x86/Linux Target Description Changes John Baldwin ` (2 subsequent siblings) 12 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-04-05 12:33 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess This commit builds on the previous series of commits to share the target description caching code between GDB and gdbserver for x86/Linux targets. The objective of this commit is to move the four functions (2 each of) i386_linux_read_description and amd64_linux_read_description into the gdb/arch/ directory and combine them so we have just a single copy of each. Then GDB, gdbserver, and the in-process-agent (IPA) will link against these shared functions. It is worth reading the description of the previous commit(s) to see why this merging is not as simple as it seems, but in summary, gdbserver used to generate a different set of possible target descriptions than GDB. The previous commit(s) fixed this, so now it should be simpler to share the functions. One curiosity with this patch is the function x86_linux_post_init_tdesc. On the gdbserver side the two functions amd64_linux_read_description and i386_linux_read_description have some functionality that is not present on the GDB side, that is some additional configuration that is performed as each target description is created to setup the expedited registers. To support this I've added the function x86_linux_post_init_tdesc. This function is called from the two *_linux_read_description functions, but is implemented separately for GDB and gdbserver. An alternative approach that avoids adding x86_linux_post_init_tdesc would be to have x86_linux_tdesc_for_tid return a non-const target description, in x86_target::low_arch_setup we could then inspect the target description to figure out if it is 64-bit or not, and modify the target description as needed. In the end I figured that adding the x86_linux_post_init_tdesc function was good enough, so went with that solution. The contents of gdbserver/linux-x86-low.cc have moved to gdb/arch/x86-linux-tdesc-features.c, and gdbserver/linux-x86-tdesc.h has moved to gdb/arch/x86-linux-tdesc-features.h, this change leads to some updates in the #includes in the gdbserver/ directory. For testing I've done the following: - Built on x86-64 GNU/Linux for all targets, and just for the native target, - Build on i386 GNU/Linux for all targets, and just for the native target, - Build on a 64-bit, non-x86 GNU/Linux for all targets, just for the native target, and for targets x86_64-*-linux and i386-*-linux. This did flush out a bunch of build issues where I'd failed to add a required file in a configure.* file, but I think everything should now be good. --- gdb/Makefile.in | 5 + gdb/amd64-linux-tdep.c | 31 -- gdb/arch/amd64-linux-tdesc.c | 61 ++++ gdb/arch/i386-linux-tdesc.c | 51 +++ gdb/arch/x86-linux-tdesc-features.c | 247 +++++++++++++++ .../arch/x86-linux-tdesc-features.h | 50 +-- gdb/arch/x86-linux-tdesc.h | 37 +++ gdb/configure.nat | 6 +- gdb/configure.tgt | 11 +- gdb/i386-linux-tdep.c | 26 +- gdbserver/configure.srv | 9 + gdbserver/linux-amd64-ipa.cc | 2 +- gdbserver/linux-i386-ipa.cc | 2 +- gdbserver/linux-x86-low.cc | 2 +- gdbserver/linux-x86-tdesc.cc | 292 +----------------- 15 files changed, 469 insertions(+), 363 deletions(-) create mode 100644 gdb/arch/amd64-linux-tdesc.c create mode 100644 gdb/arch/i386-linux-tdesc.c create mode 100644 gdb/arch/x86-linux-tdesc-features.c rename gdbserver/linux-x86-tdesc.h => gdb/arch/x86-linux-tdesc-features.h (52%) create mode 100644 gdb/arch/x86-linux-tdesc.h diff --git a/gdb/Makefile.in b/gdb/Makefile.in index f560ba023bf..9dc84cc8f0c 100644 --- a/gdb/Makefile.in +++ b/gdb/Makefile.in @@ -748,6 +748,7 @@ ALL_64_TARGET_OBS = \ arch/aarch64-insn.o \ arch/aarch64-mte-linux.o \ arch/aarch64-scalable-linux.o \ + arch/amd64-linux-tdesc.o \ arch/amd64.o \ arch/riscv.o \ bpf-tdep.o \ @@ -788,9 +789,11 @@ ALL_TARGET_OBS = \ arch/arm.o \ arch/arm-get-next-pcs.o \ arch/arm-linux.o \ + arch/i386-linux-tdesc.o \ arch/i386.o \ arch/loongarch.o \ arch/ppc-linux-common.o \ + arch/x86-linux-tdesc-features.o \ arm-bsd-tdep.o \ arm-fbsd-tdep.o \ arm-linux-tdep.o \ @@ -1559,6 +1562,8 @@ HFILES_NO_SRCDIR = \ arch/ppc-linux-common.h \ arch/ppc-linux-tdesc.h \ arch/riscv.h \ + arch/x86-linux-tdesc-features.h \ + arch/x86-linux-tdesc.h \ cli/cli-cmds.h \ cli/cli-decode.h \ cli/cli-script.h \ diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c index bcb9868e79e..c707745cd9a 100644 --- a/gdb/amd64-linux-tdep.c +++ b/gdb/amd64-linux-tdep.c @@ -1577,37 +1577,6 @@ amd64_linux_record_signal (struct gdbarch *gdbarch, return 0; } -const target_desc * -amd64_linux_read_description (uint64_t xcr0_features_bit, bool is_x32) -{ - static target_desc *amd64_linux_tdescs \ - [2/*AVX*/][2/*MPX*/][2/*AVX512*/][2/*PKRU*/] = {}; - static target_desc *x32_linux_tdescs \ - [2/*AVX*/][2/*AVX512*/][2/*PKRU*/] = {}; - - target_desc **tdesc; - - if (is_x32) - { - tdesc = &x32_linux_tdescs[(xcr0_features_bit & X86_XSTATE_AVX) ? 1 : 0 ] - [(xcr0_features_bit & X86_XSTATE_AVX512) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_PKRU) ? 1 : 0]; - } - else - { - tdesc = &amd64_linux_tdescs[(xcr0_features_bit & X86_XSTATE_AVX) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_MPX) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_AVX512) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_PKRU) ? 1 : 0]; - } - - if (*tdesc == NULL) - *tdesc = amd64_create_target_description (xcr0_features_bit, is_x32, - true, true); - - return *tdesc; -} - /* Get Linux/x86 target description from core dump. */ static const struct target_desc * diff --git a/gdb/arch/amd64-linux-tdesc.c b/gdb/arch/amd64-linux-tdesc.c new file mode 100644 index 00000000000..63b5ddfcece --- /dev/null +++ b/gdb/arch/amd64-linux-tdesc.c @@ -0,0 +1,61 @@ +/* Target description related code for GNU/Linux x86-64. + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#include "arch/x86-linux-tdesc.h" +#include "arch/amd64-linux-tdesc.h" +#include "arch/amd64.h" +#include "arch/x86-linux-tdesc-features.h" + + +/* See arch/amd64-linux-tdesc.h. */ + +const struct target_desc * +amd64_linux_read_description (uint64_t xcr0, bool is_x32) +{ + /* The type used for the amd64 and x32 target description caches. */ + using tdesc_cache_type = std::unordered_map<uint64_t, const target_desc_up>; + + /* Caches for the previously seen amd64 and x32 target descriptions, + indexed by the xcr0 value that created the target description. These + need to be static within this function to ensure they are initialised + before first use. */ + static tdesc_cache_type amd64_tdesc_cache, x32_tdesc_cache; + + tdesc_cache_type &tdesc_cache = is_x32 ? x32_tdesc_cache : amd64_tdesc_cache; + + xcr0 &= is_x32 + ? x86_linux_x32_tdesc_feature_mask () + : x86_linux_amd64_tdesc_feature_mask (); + + const auto it = tdesc_cache.find (xcr0); + if (it != tdesc_cache.end ()) + return it->second.get (); + + /* Create the previously unseen target description. */ + target_desc_up tdesc (amd64_create_target_description (xcr0, is_x32, + true, true)); + x86_linux_post_init_tdesc (tdesc.get (), true); + + /* Add to the cache, and return a pointer borrowed from the + target_desc_up. This is safe as the cache (and the pointers contained + within it) are not deleted until GDB exits. */ + target_desc *ptr = tdesc.get (); + tdesc_cache.emplace (xcr0, std::move (tdesc)); + return ptr; +} diff --git a/gdb/arch/i386-linux-tdesc.c b/gdb/arch/i386-linux-tdesc.c new file mode 100644 index 00000000000..6d1f1793457 --- /dev/null +++ b/gdb/arch/i386-linux-tdesc.c @@ -0,0 +1,51 @@ +/* Target description related code for GNU/Linux i386. + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#include "arch/x86-linux-tdesc.h" +#include "arch/i386-linux-tdesc.h" +#include "arch/i386.h" +#include "arch/x86-linux-tdesc-features.h" + +/* See arch/i386-linux-tdesc.h. */ + +const target_desc * +i386_linux_read_description (uint64_t xcr0) +{ + /* Cache of previously seen i386 target descriptions, indexed by the xcr0 + value that created the target description. This needs to be static + within this function to ensure it is initialised before first use. */ + static std::unordered_map<uint64_t, const target_desc_up> i386_tdesc_cache; + + xcr0 &= x86_linux_i386_tdesc_feature_mask (); + + const auto it = i386_tdesc_cache.find (xcr0); + if (it != i386_tdesc_cache.end ()) + return it->second.get (); + + /* Create the previously unseen target description. */ + target_desc_up tdesc (i386_create_target_description (xcr0, true, false)); + x86_linux_post_init_tdesc (tdesc.get (), false); + + /* Add to the cache, and return a pointer borrowed from the + target_desc_up. This is safe as the cache (and the pointers contained + within it) are not deleted until GDB exits. */ + target_desc *ptr = tdesc.get (); + i386_tdesc_cache.emplace (xcr0, std::move (tdesc)); + return ptr; +} diff --git a/gdb/arch/x86-linux-tdesc-features.c b/gdb/arch/x86-linux-tdesc-features.c new file mode 100644 index 00000000000..91abd9b1928 --- /dev/null +++ b/gdb/arch/x86-linux-tdesc-features.c @@ -0,0 +1,247 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#include "arch/x86-linux-tdesc-features.h" + +/* A structure used to describe a single cpu feature that might, or might + not, be checked for when creating a target description for one of i386, + amd64, or x32. */ + +struct x86_tdesc_feature { + /* The cpu feature mask. This is a mask against an xcr0 value. */ + uint64_t feature; + + /* Is this feature checked when creating an i386 target description. */ + bool is_i386; + + /* Is this feature checked when creating an amd64 target description. */ + bool is_amd64; + + /* Is this feature checked when creating an x32 target description. */ + bool is_x32; +}; + +/* A constant table that describes all of the cpu features that are + checked when building a target description for i386, amd64, or x32. */ + +static constexpr x86_tdesc_feature x86_linux_all_tdesc_features[] = { + /* Feature, i386, amd64, x32. */ + { X86_XSTATE_PKRU, true, true, true }, + { X86_XSTATE_AVX512, true, true, true }, + { X86_XSTATE_AVX, true, true, true }, + { X86_XSTATE_MPX, true, true, false }, + { X86_XSTATE_SSE, true, false, false }, + { X86_XSTATE_X87, true, false, false } +}; + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an i386 target description. */ + +static constexpr uint64_t +x86_linux_i386_tdesc_feature_mask_1 () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_i386) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an amd64 target description. */ + +static constexpr uint64_t +x86_linux_amd64_tdesc_feature_mask_1 () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_amd64) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an x32 target description. */ + +static constexpr uint64_t +x86_linux_x32_tdesc_feature_mask_1 () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_x32) + mask |= entry.feature; + + return mask; +} + +/* See arch/x86-linux-tdesc-features.h. */ + +uint64_t +x86_linux_amd64_tdesc_feature_mask () +{ + return x86_linux_amd64_tdesc_feature_mask_1 (); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +uint64_t +x86_linux_x32_tdesc_feature_mask () +{ + return x86_linux_x32_tdesc_feature_mask_1 (); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +uint64_t +x86_linux_i386_tdesc_feature_mask () +{ + return x86_linux_i386_tdesc_feature_mask_1 (); +} + +#ifdef GDBSERVER + +/* See arch/x86-linux-tdesc-features.h. */ + +int +x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0) +{ + /* The following table shows which features are checked for when creating + the target descriptions (see nat/x86-linux-tdesc.c), the feature order + represents the bit order within the generated index number. + + i386 | x87 sse mpx avx avx512 pkru + amd64 | mpx avx avx512 pkru + i32 | avx avx512 pkru + + The features are ordered so that for each mode (i386, amd64, i32) the + generated index will form a continuous range. */ + + int idx = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) + { + if ((xcr0 & x86_linux_all_tdesc_features[i].feature) + == x86_linux_all_tdesc_features[i].feature) + idx |= (1 << i); + } + + return idx; +} + +#endif /* GDBSERVER */ + +#ifdef IN_PROCESS_AGENT + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an i386 target description. */ + +static constexpr int +x86_linux_i386_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_i386) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an amd64 target description. */ + +static constexpr int +x86_linux_amd64_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_amd64) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an x32 target description. */ + +static constexpr int +x86_linux_x32_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_x32) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +int +x86_linux_amd64_tdesc_count () +{ + return x86_linux_amd64_tdesc_count_1 (); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +int +x86_linux_x32_tdesc_count () +{ + return x86_linux_x32_tdesc_count_1 (); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +int +x86_linux_i386_tdesc_count () +{ + return x86_linux_i386_tdesc_count_1 (); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +uint64_t +x86_linux_tdesc_idx_to_xcr0 (int idx) +{ + uint64_t xcr0 = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) + { + if ((idx & (1 << i)) != 0) + xcr0 |= x86_linux_all_tdesc_features[i].feature; + } + + return xcr0; +} + +#endif /* IN_PROCESS_AGENT */ diff --git a/gdbserver/linux-x86-tdesc.h b/gdb/arch/x86-linux-tdesc-features.h similarity index 52% rename from gdbserver/linux-x86-tdesc.h rename to gdb/arch/x86-linux-tdesc-features.h index 3d6e0e51833..0d3db587174 100644 --- a/gdbserver/linux-x86-tdesc.h +++ b/gdb/arch/x86-linux-tdesc-features.h @@ -1,7 +1,6 @@ -/* Low level support for x86 (i386 and x86-64), shared between gdbserver - and IPA. +/* Target description related code for GNU/Linux x86 (i386 and x86-64). - Copyright (C) 2016-2024 Free Software Foundation, Inc. + Copyright (C) 2024 Free Software Foundation, Inc. This file is part of GDB. @@ -18,33 +17,46 @@ You should have received a copy of the GNU General Public License along with this program. If not, see <http://www.gnu.org/licenses/>. */ -#ifndef GDBSERVER_LINUX_X86_TDESC_H -#define GDBSERVER_LINUX_X86_TDESC_H +#ifndef ARCH_X86_LINUX_TDESC_FEATURES_H +#define ARCH_X86_LINUX_TDESC_FEATURES_H -/* Convert an xcr0 value into an integer. The integer will be passed to - the in-process-agent where it will then be passed to - x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ +#include "gdbsupport/x86-xstate.h" +#include "gdbsupport/gdb_assert.h" -extern int x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0); +/* Return a mask of X86_STATE_* feature flags. The returned mask indicates + the set of features which are checked for when creating the target + description for each of amd64, x32, and i386. */ -#ifdef IN_PROCESS_AGENT +extern uint64_t x86_linux_amd64_tdesc_feature_mask (); +extern uint64_t x86_linux_x32_tdesc_feature_mask (); +extern uint64_t x86_linux_i386_tdesc_feature_mask (); -/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) - into an xcr0 value which can then be used to create a target - description. */ +#ifdef GDBSERVER -extern uint64_t x86_linux_tdesc_idx_to_xcr0 (int idx); +/* Convert an xcr0 value into an integer. The integer will be passed from + gdbserver to the in-process-agent where it will then be passed through + x86_linux_tdesc_idx_to_xcr0 to get back the original xcr0 value. */ + +extern int x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0); -/* Within the in-process-agent we need to pre-initialise all of the target - descriptions, to do this we need to know how many target descriptions - there are for each different target type. These functions return the - target description count for the relevant target. */ +#endif /* GDBSERVER */ + +#ifdef IN_PROCESS_AGENT + +/* Return the maximum possible number of target descriptions for each of + amd64, x32, and i386. These are used by the in-process-agent to + generate every possible target description. */ extern int x86_linux_amd64_tdesc_count (); extern int x86_linux_x32_tdesc_count (); extern int x86_linux_i386_tdesc_count (); +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +extern uint64_t x86_linux_tdesc_idx_to_xcr0 (int idx); #endif /* IN_PROCESS_AGENT */ -#endif /* GDBSERVER_LINUX_X86_TDESC_H */ +#endif /* ARCH_X86_LINUX_TDESC_FEATURES_H */ diff --git a/gdb/arch/x86-linux-tdesc.h b/gdb/arch/x86-linux-tdesc.h new file mode 100644 index 00000000000..152592fcf76 --- /dev/null +++ b/gdb/arch/x86-linux-tdesc.h @@ -0,0 +1,37 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#ifndef ARCH_X86_LINUX_TDESC_H +#define ARCH_X86_LINUX_TDESC_H + +struct target_desc; + +/* This function is called from amd64_linux_read_description and + i386_linux_read_description after a new target description has been + created, TDESC is the new target description, IS_64BIT will be true + when called from amd64_linux_read_description, otherwise IS_64BIT will + be false. If the *_linux_read_description functions found a cached + target description then this function will not be called. + + Both GDB and gdbserver have their own implementations of this + function. */ + +extern void x86_linux_post_init_tdesc (target_desc *tdesc, bool is_64bit); + +#endif /* ARCH_X86_LINUX_TDESC_H */ diff --git a/gdb/configure.nat b/gdb/configure.nat index 4bcc0696027..24e1824b01c 100644 --- a/gdb/configure.nat +++ b/gdb/configure.nat @@ -256,7 +256,8 @@ case ${gdb_host} in NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ nat/x86-xstate.o \ i386-linux-nat.o x86-linux-nat.o nat/linux-btrace.o \ - nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o" + nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o \ + arch/i386-linux-tdesc.o arch/x86-linux-tdesc-features.o" ;; ia64) # Host: Intel IA-64 running GNU/Linux @@ -323,7 +324,8 @@ case ${gdb_host} in nat/x86-xstate.o amd64-nat.o amd64-linux-nat.o x86-linux-nat.o \ nat/linux-btrace.o \ nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o \ - nat/amd64-linux-siginfo.o" + nat/amd64-linux-siginfo.o arch/x86-linux-tdesc-features.o \ + arch/i386-linux-tdesc.o arch/amd64-linux-tdesc.o" ;; sparc) # Host: GNU/Linux UltraSPARC diff --git a/gdb/configure.tgt b/gdb/configure.tgt index 47a674201f9..8326c458eb1 100644 --- a/gdb/configure.tgt +++ b/gdb/configure.tgt @@ -320,10 +320,13 @@ i[34567]86-*-linux*) gdb_target_obs="i386-linux-tdep.o \ glibc-tdep.o \ solib-svr4.o symfile-mem.o \ - linux-tdep.o linux-record.o" + linux-tdep.o linux-record.o \ + arch/i386-linux-tdesc.o \ + arch/x86-linux-tdesc-features.o" if test "x$enable_64_bit_bfd" = "xyes"; then # Target: GNU/Linux x86-64 - gdb_target_obs="amd64-linux-tdep.o ${gdb_target_obs}" + gdb_target_obs="amd64-linux-tdep.o \ + arch/amd64-linux-tdesc.o ${gdb_target_obs}" fi ;; i[34567]86-*-gnu*) @@ -718,7 +721,9 @@ x86_64-*-linux*) # Target: GNU/Linux x86-64 gdb_target_obs="amd64-linux-tdep.o ${i386_tobjs} \ i386-linux-tdep.o glibc-tdep.o \ - solib-svr4.o symfile-mem.o linux-tdep.o linux-record.o" + solib-svr4.o symfile-mem.o linux-tdep.o linux-record.o \ + arch/i386-linux-tdesc.o arch/amd64-linux-tdesc.o \ + arch/x86-linux-tdesc-features.o" ;; x86_64-*-freebsd* | x86_64-*-kfreebsd*-gnu) # Target: FreeBSD/amd64 diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index c796f87780b..d4d820b9f98 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -40,6 +40,7 @@ #include "i387-tdep.h" #include "gdbsupport/x86-xstate.h" #include "arch/i386-linux-tdesc.h" +#include "arch/x86-linux-tdesc.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_I386 "syscalls/i386-linux.xml" @@ -679,29 +680,12 @@ i386_linux_core_read_x86_xsave_layout (struct gdbarch *gdbarch, layout) != 0; } -/* See i386-linux-tdep.h. */ +/* See arch/x86-linux-tdesc.h. */ -const struct target_desc * -i386_linux_read_description (uint64_t xcr0) +void +x86_linux_post_init_tdesc (target_desc *tdesc, bool is_64bit) { - if (xcr0 == 0) - return NULL; - - static struct target_desc *i386_linux_tdescs \ - [2/*X87*/][2/*SSE*/][2/*AVX*/][2/*MPX*/][2/*AVX512*/][2/*PKRU*/] = {}; - struct target_desc **tdesc; - - tdesc = &i386_linux_tdescs[(xcr0 & X86_XSTATE_X87) ? 1 : 0] - [(xcr0 & X86_XSTATE_SSE) ? 1 : 0] - [(xcr0 & X86_XSTATE_AVX) ? 1 : 0] - [(xcr0 & X86_XSTATE_MPX) ? 1 : 0] - [(xcr0 & X86_XSTATE_AVX512) ? 1 : 0] - [(xcr0 & X86_XSTATE_PKRU) ? 1 : 0]; - - if (*tdesc == NULL) - *tdesc = i386_create_target_description (xcr0, true, false); - - return *tdesc; + /* Nothing. */ } /* Get Linux/x86 target description from core dump. */ diff --git a/gdbserver/configure.srv b/gdbserver/configure.srv index 7a2702d78bf..e17b5cf280c 100644 --- a/gdbserver/configure.srv +++ b/gdbserver/configure.srv @@ -110,12 +110,16 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" + srv_tgtobj="${srv_tgtobj} arch/x86-linux-tdesc-features.o" + srv_tgtobj="${srv_tgtobj} arch/i386-linux-tdesc.o" srv_linux_usrregs=yes srv_linux_regsets=yes srv_linux_thread_db=yes srv_linux_btrace=yes ipa_obj="linux-i386-ipa.o linux-x86-tdesc-ipa.o" ipa_obj="${ipa_obj} arch/i386-ipa.o" + ipa_obj="${ipa_obj} arch/x86-linux-tdesc-features-ipa.o" + ipa_obj="${ipa_obj} arch/i386-linux-tdesc-ipa.o" ;; i[34567]86-*-mingw*) srv_regobj="" srv_tgtobj="x86-low.o nat/x86-dregs.o win32-low.o" @@ -374,12 +378,17 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" srv_tgtobj="${srv_tgtobj} nat/amd64-linux-siginfo.o" + srv_tgtobj="${srv_tgtobj} arch/x86-linux-tdesc-features.o" + srv_tgtobj="${srv_tgtobj} arch/amd64-linux-tdesc.o" + srv_tgtobj="${srv_tgtobj} arch/i386-linux-tdesc.o" srv_linux_usrregs=yes # This is for i386 progs. srv_linux_regsets=yes srv_linux_thread_db=yes srv_linux_btrace=yes ipa_obj="linux-amd64-ipa.o linux-x86-tdesc-ipa.o" ipa_obj="${ipa_obj} arch/amd64-ipa.o" + ipa_obj="${ipa_obj} arch/x86-linux-tdesc-features-ipa.o" + ipa_obj="${ipa_obj} arch/amd64-linux-tdesc-ipa.o" ;; x86_64-*-mingw*) srv_regobj="" srv_tgtobj="x86-low.o nat/x86-dregs.o" diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index 0c80812cc6f..89857f208b8 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -20,9 +20,9 @@ #include <sys/mman.h> #include "tracepoint.h" -#include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" #include "arch/amd64-linux-tdesc.h" +#include "arch/x86-linux-tdesc-features.h" /* fast tracepoints collect registers. */ diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc index c1c3152fb04..8100c9f9840 100644 --- a/gdbserver/linux-i386-ipa.cc +++ b/gdbserver/linux-i386-ipa.cc @@ -20,9 +20,9 @@ #include <sys/mman.h> #include "tracepoint.h" -#include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" #include "arch/i386-linux-tdesc.h" +#include "arch/x86-linux-tdesc-features.h" /* GDB register numbers. */ diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 4f473730efb..63607cae390 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -33,6 +33,7 @@ #endif #include "arch/i386-linux-tdesc.h" +#include "arch/x86-linux-tdesc-features.h" #include "gdb_proc_service.h" /* Don't include elf/common.h if linux/elf.h got included by @@ -48,7 +49,6 @@ #include "nat/linux-nat.h" #include "nat/x86-linux.h" #include "nat/x86-linux-dregs.h" -#include "linux-x86-tdesc.h" #include "nat/x86-linux-tdesc.h" #ifdef __x86_64__ diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc index 5e12526bf17..13c80762605 100644 --- a/gdbserver/linux-x86-tdesc.cc +++ b/gdbserver/linux-x86-tdesc.cc @@ -17,295 +17,19 @@ You should have received a copy of the GNU General Public License along with this program. If not, see <http://www.gnu.org/licenses/>. */ +#include "arch/x86-linux-tdesc.h" #include "tdesc.h" -#include "linux-x86-tdesc.h" -#include "arch/i386.h" -#include "gdbsupport/x86-xstate.h" -#ifdef __x86_64__ -#include "arch/amd64.h" -#include "arch/amd64-linux-tdesc.h" -#endif #include "x86-tdesc.h" -#include "arch/i386-linux-tdesc.h" - -/* A structure used to describe a single cpu feature that might, or might - not, be checked for when creating a target description for one of i386, - amd64, or x32. */ - -struct x86_tdesc_feature { - /* The cpu feature mask. This is a mask against an xcr0 value. */ - uint64_t feature; - - /* Is this feature checked when creating an i386 target description. */ - bool is_i386; - - /* Is this feature checked when creating an amd64 target description. */ - bool is_amd64; - - /* Is this feature checked when creating an x32 target description. */ - bool is_x32; -}; - -/* A constant table that describes all of the cpu features that are - checked when building a target description for i386, amd64, or x32. */ - -static constexpr x86_tdesc_feature x86_linux_all_tdesc_features[] = { - /* Feature, i386, amd64, x32. */ - { X86_XSTATE_PKRU, true, true, true }, - { X86_XSTATE_AVX512, true, true, true }, - { X86_XSTATE_AVX, true, true, true }, - { X86_XSTATE_MPX, true, true, false }, - { X86_XSTATE_SSE, true, false, false }, - { X86_XSTATE_X87, true, false, false } -}; - -/* Return a compile time constant which is a mask of all the cpu features - that are checked for when building an i386 target description. */ - -static constexpr uint64_t -x86_linux_i386_tdesc_feature_mask () -{ - uint64_t mask = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_i386) - mask |= entry.feature; - - return mask; -} - -/* Return a compile time constant which is a mask of all the cpu features - that are checked for when building an amd64 target description. */ - -static constexpr uint64_t -x86_linux_amd64_tdesc_feature_mask () -{ - uint64_t mask = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_amd64) - mask |= entry.feature; - - return mask; -} -/* Return a compile time constant which is a mask of all the cpu features - that are checked for when building an x32 target description. */ +/* See arch/x86-linux-tdesc.h. */ -static constexpr uint64_t -x86_linux_x32_tdesc_feature_mask () +void +x86_linux_post_init_tdesc (target_desc *tdesc, bool is_64bit) { - uint64_t mask = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_x32) - mask |= entry.feature; - - return mask; -} - -/* Return a compile time constant which is a count of the number of cpu - features that are checked for when building an i386 target description. */ - -static constexpr int -x86_linux_i386_tdesc_count_1 () -{ - uint64_t count = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_i386) - ++count; - - gdb_assert (count > 0); - - return (1 << count); -} - -/* Return a compile time constant which is a count of the number of cpu - features that are checked for when building an amd64 target description. */ - -static constexpr int -x86_linux_amd64_tdesc_count_1 () -{ - uint64_t count = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_amd64) - ++count; - - gdb_assert (count > 0); - - return (1 << count); -} - -/* Return a compile time constant which is a count of the number of cpu - features that are checked for when building an x32 target description. */ - -static constexpr int -x86_linux_x32_tdesc_count_1 () -{ - uint64_t count = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_x32) - ++count; - - gdb_assert (count > 0); - - return (1 << count); -} - -#ifdef IN_PROCESS_AGENT - -/* See linux-x86-tdesc.h. */ - -int -x86_linux_amd64_tdesc_count () -{ - return x86_linux_amd64_tdesc_count_1 (); -} - -/* See linux-x86-tdesc.h. */ - -int -x86_linux_x32_tdesc_count () -{ - return x86_linux_x32_tdesc_count_1 (); -} - -/* See linux-x86-tdesc.h. */ - -int -x86_linux_i386_tdesc_count () -{ - return x86_linux_i386_tdesc_count_1 (); -} - -#endif /* IN_PROCESS_AGENT */ - -/* Convert an xcr0 value into an integer. The integer will be passed to - the in-process-agent where it will then be passed to - x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ - -int -x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0) -{ - /* The following table shows which features are checked for when creating - the target descriptions (see nat/x86-linux-tdesc.c), the feature order - represents the bit order within the generated index number. - - i386 | x87 sse mpx avx avx512 pkru - amd64 | mpx avx avx512 pkru - i32 | avx avx512 pkru - - The features are ordered so that for each mode (i386, amd64, i32) the - generated index will form a continuous range. */ - - int idx = 0; - - for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) - { - if ((xcr0 & x86_linux_all_tdesc_features[i].feature) - == x86_linux_all_tdesc_features[i].feature) - idx |= (1 << i); - } - - return idx; -} - - -#ifdef IN_PROCESS_AGENT - -/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) - into an xcr0 value which can then be used to create a target - description. */ - -uint64_t -x86_linux_tdesc_idx_to_xcr0 (int idx) -{ - uint64_t xcr0 = 0; - - for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) - { - if ((idx & (1 << i)) != 0) - xcr0 |= x86_linux_all_tdesc_features[i].feature; - } - - return xcr0; -} - -#endif /* IN_PROCESS_AGENT */ - -#if defined __i386__ || !defined IN_PROCESS_AGENT - -/* A cache of all possible i386 target descriptions. */ - -static struct target_desc *i386_tdescs[x86_linux_i386_tdesc_count_1 ()] = { }; - -/* See nat/x86-linux-tdesc.h. */ - -const struct target_desc * -i386_linux_read_description (uint64_t xcr0) -{ - xcr0 &= x86_linux_i386_tdesc_feature_mask (); - int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - - gdb_assert (idx >= 0 && idx < x86_linux_i386_tdesc_count_1 ()); - - target_desc **tdesc = &i386_tdescs[idx]; - - if (*tdesc == nullptr) - { - *tdesc = i386_create_target_description (xcr0, true, false); - - init_target_desc (*tdesc, i386_expedite_regs); - } - - return *tdesc; -} -#endif - #ifdef __x86_64__ - -/* A cache of all possible amd64 target descriptions. */ - -static target_desc *amd64_tdescs[x86_linux_amd64_tdesc_count_1 ()] = { }; - -/* A cache of all possible x32 target descriptions. */ - -static target_desc *x32_tdescs[x86_linux_x32_tdesc_count_1 ()] = { }; - -/* See nat/x86-linux-tdesc.h. */ - -const struct target_desc * -amd64_linux_read_description (uint64_t xcr0, bool is_x32) -{ - if (is_x32) - xcr0 &= x86_linux_x32_tdesc_feature_mask (); + if (is_64bit) + init_target_desc (tdesc, amd64_expedite_regs); else - xcr0 &= x86_linux_amd64_tdesc_feature_mask (); - - int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - - if (is_x32) - gdb_assert (idx >= 0 && idx < x86_linux_x32_tdesc_count_1 ()); - else - gdb_assert (idx >= 0 && idx < x86_linux_amd64_tdesc_count_1 ()); - - target_desc **tdesc = nullptr; - - if (is_x32) - tdesc = &x32_tdescs[idx]; - else - tdesc = &amd64_tdescs[idx]; - - if (*tdesc == nullptr) - { - *tdesc = amd64_create_target_description (xcr0, is_x32, true, true); - - init_target_desc (*tdesc, amd64_expedite_regs); - } - return *tdesc; -} - #endif + init_target_desc (tdesc, i386_expedite_regs); +} -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv4 00/10] x86/Linux Target Description Changes 2024-04-05 12:33 ` [PATCHv4 00/10] " Andrew Burgess ` (9 preceding siblings ...) 2024-04-05 12:33 ` [PATCHv4 10/10] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess @ 2024-04-09 18:37 ` John Baldwin 2024-04-25 13:35 ` Willgerodt, Felix 2024-04-26 15:01 ` [PATCHv5 00/11] " Andrew Burgess 12 siblings, 0 replies; 116+ messages in thread From: John Baldwin @ 2024-04-09 18:37 UTC (permalink / raw) To: Andrew Burgess, gdb-patches On 4/5/24 8:33 AM, Andrew Burgess wrote: > In v4: > > - I tried merging V3, but it turned out I broke pretty much > everything that wasn't x86 based when configured with > --enable-targets=all, > > - The problem was a failure to correctly split the shared code > between the gdb/arch/ and gdb/nat/ directories, as a consequence, > code which is needed on a non x86 based host to support x86 based > targets wasn't available to the compilation, and the build failed, > > - In V4 I've gone through every patch and resplit the code in a way > which I now believe is correct, I've done the following tests: > > + On a non x86 host I've built GDB to support only the current > host as a target, to support all targets, and to support x86-64 > and i386 linux targets, > > + On an i386 virtual machine I built GDB only for the host as a > target, and for all targets. I regression tested the all targets > build for unix, native-gdbserver, and native-extended-gdbserver, > > + On an x86-64 machine I've built GDB for only the current host as > a target, and for all targets. I regression tested the all targets > build for unix, native-gdbserver, and native-extended-gdbserver. > > - Only patches 6, 8, and 10 require significant review. All of the > other patches are pretty trivial (though reviews always welcome). > > - I think there's more improvements that can be made to the x86 > target description creation/lookup/caching. This series only > changes the Linux lookup, and we still cache i386/amd64/x32 > separately. > > In the future I think we can merge all x86 target description > caching into a single data structure, this would be for all OS > variants and all ABI variants. > > Though making that "grand unification" will certainly require some > of the code in this series to change, I think the bulk of it will > remain, and trying to do everything in one series is just going to > result in an even larger series. I'd prefer to get these first > patches merged, then come back to build on this work once this is > merged and we know there's no problems with it. > > In v3: > > - Rebased. Nasty merge conflict with 4bb20a6244b7091 which I think > I've resolved, but am unable to test. Reposting so the author of > that other commit can validate. > > - Initial testing looks good. Full tests are still running. > > In v2: > > - Rebase to current upstream/master, no merge conflicts, > > - Retested. I re-read this series and it does look ok to me still, but I thought V3 was ok as well. :-P Your testing matrix described above for V4 seems reasonable for catching what you ran into with V3. Approved-By: John Baldwin <jhb@FreeBSD.org> -- John Baldwin ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv4 00/10] x86/Linux Target Description Changes 2024-04-05 12:33 ` [PATCHv4 00/10] " Andrew Burgess ` (10 preceding siblings ...) 2024-04-09 18:37 ` [PATCHv4 00/10] x86/Linux Target Description Changes John Baldwin @ 2024-04-25 13:35 ` Willgerodt, Felix 2024-04-25 16:06 ` Andrew Burgess 2024-04-26 15:01 ` [PATCHv5 00/11] " Andrew Burgess 12 siblings, 1 reply; 116+ messages in thread From: Willgerodt, Felix @ 2024-04-25 13:35 UTC (permalink / raw) To: Andrew Burgess, gdb-patches > -----Original Message----- > From: Andrew Burgess <aburgess@redhat.com> > Sent: Freitag, 5. April 2024 14:33 > To: gdb-patches@sourceware.org > Cc: Andrew Burgess <aburgess@redhat.com> > Subject: [PATCHv4 00/10] x86/Linux Target Description Changes > > In v4: > > - I tried merging V3, but it turned out I broke pretty much > everything that wasn't x86 based when configured with > --enable-targets=all, > > - The problem was a failure to correctly split the shared code > between the gdb/arch/ and gdb/nat/ directories, as a consequence, > code which is needed on a non x86 based host to support x86 based > targets wasn't available to the compilation, and the build failed, > > - In V4 I've gone through every patch and resplit the code in a way > which I now believe is correct, I've done the following tests: > > + On a non x86 host I've built GDB to support only the current > host as a target, to support all targets, and to support x86-64 > and i386 linux targets, > > + On an i386 virtual machine I built GDB only for the host as a > target, and for all targets. I regression tested the all targets > build for unix, native-gdbserver, and native-extended-gdbserver, > > + On an x86-64 machine I've built GDB for only the current host as > a target, and for all targets. I regression tested the all targets > build for unix, native-gdbserver, and native-extended-gdbserver. > > - Only patches 6, 8, and 10 require significant review. All of the > other patches are pretty trivial (though reviews always welcome). > > - I think there's more improvements that can be made to the x86 > target description creation/lookup/caching. This series only > changes the Linux lookup, and we still cache i386/amd64/x32 > separately. > > In the future I think we can merge all x86 target description > caching into a single data structure, this would be for all OS > variants and all ABI variants. > > Though making that "grand unification" will certainly require some > of the code in this series to change, I think the bulk of it will > remain, and trying to do everything in one series is just going to > result in an even larger series. I'd prefer to get these first > patches merged, then come back to build on this work once this is > merged and we know there's no problems with it. > > In v3: > > - Rebased. Nasty merge conflict with 4bb20a6244b7091 which I think > I've resolved, but am unable to test. Reposting so the author of > that other commit can validate. > > - Initial testing looks good. Full tests are still running. > > In v2: > > - Rebase to current upstream/master, no merge conflicts, > > - Retested. > > --- > > Andrew Burgess (10): > gdbserver/ipa/x86: remove unneeded declarations > gdbserver: convert have_ptrace_getregset to a tribool > gdb/x86: move reading of cs and ds state into gdb/nat directory > gdbserver/x86: move no-xml code earlier in x86_linux_read_description > gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition > gdb/gdbserver: share some code relating to target description creation > gdb/arch: assert that X86_XSTATE_MPX is not set for x32 > gdbserver: update target description creation for x86/linux > gdb: move xcr0 == 0 check into i386_linux_core_read_description > gdb/gdbserver: share x86/linux tdesc caching > > gdb/Makefile.in | 8 + > gdb/amd64-linux-tdep.c | 32 +--- > gdb/amd64-linux-tdep.h | 6 - > gdb/arch/amd64-linux-tdesc.c | 61 +++++++ > gdb/arch/amd64-linux-tdesc.h | 30 ++++ > gdb/arch/amd64.c | 8 +- > gdb/arch/i386-linux-tdesc.c | 51 ++++++ > gdb/arch/i386-linux-tdesc.h | 29 ++++ > gdb/arch/x86-linux-tdesc-features.c | 247 ++++++++++++++++++++++++++++ > gdb/arch/x86-linux-tdesc-features.h | 62 +++++++ > gdb/arch/x86-linux-tdesc.h | 37 +++++ > gdb/configure.nat | 8 +- > gdb/configure.tgt | 11 +- > gdb/i386-linux-tdep.c | 42 ++--- > gdb/i386-linux-tdep.h | 23 --- > gdb/nat/x86-linux-tdesc.c | 120 ++++++++++++++ > gdb/nat/x86-linux-tdesc.h | 60 +++++++ > gdb/nat/x86-linux.c | 47 ++++++ > gdb/nat/x86-linux.h | 28 ++++ > gdb/x86-linux-nat.c | 123 ++------------ > gdbserver/configure.srv | 11 ++ > gdbserver/linux-amd64-ipa.cc | 50 ++---- > gdbserver/linux-arm-low.cc | 6 +- > gdbserver/linux-i386-ipa.cc | 26 +-- > gdbserver/linux-low.cc | 2 +- > gdbserver/linux-low.h | 2 +- > gdbserver/linux-x86-low.cc | 192 +++++++-------------- > gdbserver/linux-x86-tdesc.cc | 142 +--------------- > gdbserver/linux-x86-tdesc.h | 56 ------- > gdbsupport/x86-xstate.h | 20 +++ > 30 files changed, 949 insertions(+), 591 deletions(-) > create mode 100644 gdb/arch/amd64-linux-tdesc.c > create mode 100644 gdb/arch/amd64-linux-tdesc.h > create mode 100644 gdb/arch/i386-linux-tdesc.c > create mode 100644 gdb/arch/i386-linux-tdesc.h > create mode 100644 gdb/arch/x86-linux-tdesc-features.c > create mode 100644 gdb/arch/x86-linux-tdesc-features.h > create mode 100644 gdb/arch/x86-linux-tdesc.h > create mode 100644 gdb/nat/x86-linux-tdesc.c > create mode 100644 gdb/nat/x86-linux-tdesc.h > delete mode 100644 gdbserver/linux-x86-tdesc.h > > > base-commit: 16810e455feb26ef826a3ed876d6d7e6d24818b0 > -- > 2.25.4 Hi Andrew, Thanks for doing this, I like what this does. I am still in the middle of reviewing this. Though in first testing I saw that I can't build a 32-bit gdbserver on Ubuntu 22.04. I am building like this: export CFLAGS="-g -O0 -m32" export CXXFLAGS="${CFLAGS} -m32" export LDFLAGS="-m32" $src/gdb/configure \ --build="x86_64-linux-gnu" \ --host="i686-linux-gnu" \ --disable-gdb make -s -j$(nproc) all-gdbserver Not sure if there is an easier way of doing this, but this works for master. With your series I see this: CXX nat/x86-linux-tdesc.o /gdb/gdbserver/../gdb/nat/x86-linux-tdesc.c: In function ?const target_desc* x86_linux_tdesc_for_tid(int, tribool*, gdb::function_view<void(long long unsigned int)>, const char*, uint64_t*)?: /gdb/gdbserver/../gdb/nat/x86-linux-tdesc.c:54:7: error: ?have_ptrace_getfpxregs? was not declared in this scope; did you mean ?have_ptrace_getregset?? 54 | if (have_ptrace_getfpxregs == -1) | ^~~~~~~~~~~~~~~~~~~~~~ | have_ptrace_getregset /gdb/gdbserver/../gdb/nat/x86-linux-tdesc.c:56:7: error: ?elf_fpxregset_t? was not declared in this scope 56 | elf_fpxregset_t fpxregs; | ^~~~~~~~~~~~~~~ In file included from /gdb/gdbserver/../gdb/nat/x86-linux-tdesc.c:29: /gdb/gdbserver/../gdb/nat/x86-linux-tdesc.c:58:53: error: ?fpxregs? was not declared in this scope 58 | if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) | ^~~~~~~ /gdb/gdbserver/../gdb/nat/gdb_ptrace.h:142:57: note: in definition of macro ?ptrace? 142 | ptrace ((PTRACE_TYPE_ARG1) request, pid, addr, data) | ^~~~ make[1]: *** [Makefile:577: nat/x86-linux-tdesc.o] Error 1 make: *** [Makefile:13074: install-gdbserver] Error 2 I didn't investigate further yet, sorry. Regards, Felix Intel Deutschland GmbH Registered Address: Am Campeon 10, 85579 Neubiberg, Germany Tel: +49 89 99 8853-0, www.intel.de <http://www.intel.de> Managing Directors: Christin Eisenschmid, Sharon Heck, Tiffany Doon Silva Chairperson of the Supervisory Board: Nicole Lau Registered Office: Munich Commercial Register: Amtsgericht Muenchen HRB 186928 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv4 00/10] x86/Linux Target Description Changes 2024-04-25 13:35 ` Willgerodt, Felix @ 2024-04-25 16:06 ` Andrew Burgess 0 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-04-25 16:06 UTC (permalink / raw) To: Willgerodt, Felix, gdb-patches "Willgerodt, Felix" <felix.willgerodt@intel.com> writes: >> -----Original Message----- >> From: Andrew Burgess <aburgess@redhat.com> >> Sent: Freitag, 5. April 2024 14:33 >> To: gdb-patches@sourceware.org >> Cc: Andrew Burgess <aburgess@redhat.com> >> Subject: [PATCHv4 00/10] x86/Linux Target Description Changes >> >> In v4: >> >> - I tried merging V3, but it turned out I broke pretty much >> everything that wasn't x86 based when configured with >> --enable-targets=all, >> >> - The problem was a failure to correctly split the shared code >> between the gdb/arch/ and gdb/nat/ directories, as a consequence, >> code which is needed on a non x86 based host to support x86 based >> targets wasn't available to the compilation, and the build failed, >> >> - In V4 I've gone through every patch and resplit the code in a way >> which I now believe is correct, I've done the following tests: >> >> + On a non x86 host I've built GDB to support only the current >> host as a target, to support all targets, and to support x86-64 >> and i386 linux targets, >> >> + On an i386 virtual machine I built GDB only for the host as a >> target, and for all targets. I regression tested the all targets >> build for unix, native-gdbserver, and native-extended-gdbserver, >> >> + On an x86-64 machine I've built GDB for only the current host as >> a target, and for all targets. I regression tested the all targets >> build for unix, native-gdbserver, and native-extended-gdbserver. >> >> - Only patches 6, 8, and 10 require significant review. All of the >> other patches are pretty trivial (though reviews always welcome). >> >> - I think there's more improvements that can be made to the x86 >> target description creation/lookup/caching. This series only >> changes the Linux lookup, and we still cache i386/amd64/x32 >> separately. >> >> In the future I think we can merge all x86 target description >> caching into a single data structure, this would be for all OS >> variants and all ABI variants. >> >> Though making that "grand unification" will certainly require some >> of the code in this series to change, I think the bulk of it will >> remain, and trying to do everything in one series is just going to >> result in an even larger series. I'd prefer to get these first >> patches merged, then come back to build on this work once this is >> merged and we know there's no problems with it. >> >> In v3: >> >> - Rebased. Nasty merge conflict with 4bb20a6244b7091 which I think >> I've resolved, but am unable to test. Reposting so the author of >> that other commit can validate. >> >> - Initial testing looks good. Full tests are still running. >> >> In v2: >> >> - Rebase to current upstream/master, no merge conflicts, >> >> - Retested. >> >> --- >> >> Andrew Burgess (10): >> gdbserver/ipa/x86: remove unneeded declarations >> gdbserver: convert have_ptrace_getregset to a tribool >> gdb/x86: move reading of cs and ds state into gdb/nat directory >> gdbserver/x86: move no-xml code earlier in x86_linux_read_description >> gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition >> gdb/gdbserver: share some code relating to target description creation >> gdb/arch: assert that X86_XSTATE_MPX is not set for x32 >> gdbserver: update target description creation for x86/linux >> gdb: move xcr0 == 0 check into i386_linux_core_read_description >> gdb/gdbserver: share x86/linux tdesc caching >> >> gdb/Makefile.in | 8 + >> gdb/amd64-linux-tdep.c | 32 +--- >> gdb/amd64-linux-tdep.h | 6 - >> gdb/arch/amd64-linux-tdesc.c | 61 +++++++ >> gdb/arch/amd64-linux-tdesc.h | 30 ++++ >> gdb/arch/amd64.c | 8 +- >> gdb/arch/i386-linux-tdesc.c | 51 ++++++ >> gdb/arch/i386-linux-tdesc.h | 29 ++++ >> gdb/arch/x86-linux-tdesc-features.c | 247 ++++++++++++++++++++++++++++ >> gdb/arch/x86-linux-tdesc-features.h | 62 +++++++ >> gdb/arch/x86-linux-tdesc.h | 37 +++++ >> gdb/configure.nat | 8 +- >> gdb/configure.tgt | 11 +- >> gdb/i386-linux-tdep.c | 42 ++--- >> gdb/i386-linux-tdep.h | 23 --- >> gdb/nat/x86-linux-tdesc.c | 120 ++++++++++++++ >> gdb/nat/x86-linux-tdesc.h | 60 +++++++ >> gdb/nat/x86-linux.c | 47 ++++++ >> gdb/nat/x86-linux.h | 28 ++++ >> gdb/x86-linux-nat.c | 123 ++------------ >> gdbserver/configure.srv | 11 ++ >> gdbserver/linux-amd64-ipa.cc | 50 ++---- >> gdbserver/linux-arm-low.cc | 6 +- >> gdbserver/linux-i386-ipa.cc | 26 +-- >> gdbserver/linux-low.cc | 2 +- >> gdbserver/linux-low.h | 2 +- >> gdbserver/linux-x86-low.cc | 192 +++++++-------------- >> gdbserver/linux-x86-tdesc.cc | 142 +--------------- >> gdbserver/linux-x86-tdesc.h | 56 ------- >> gdbsupport/x86-xstate.h | 20 +++ >> 30 files changed, 949 insertions(+), 591 deletions(-) >> create mode 100644 gdb/arch/amd64-linux-tdesc.c >> create mode 100644 gdb/arch/amd64-linux-tdesc.h >> create mode 100644 gdb/arch/i386-linux-tdesc.c >> create mode 100644 gdb/arch/i386-linux-tdesc.h >> create mode 100644 gdb/arch/x86-linux-tdesc-features.c >> create mode 100644 gdb/arch/x86-linux-tdesc-features.h >> create mode 100644 gdb/arch/x86-linux-tdesc.h >> create mode 100644 gdb/nat/x86-linux-tdesc.c >> create mode 100644 gdb/nat/x86-linux-tdesc.h >> delete mode 100644 gdbserver/linux-x86-tdesc.h >> >> >> base-commit: 16810e455feb26ef826a3ed876d6d7e6d24818b0 >> -- >> 2.25.4 > > > Hi Andrew, > > Thanks for doing this, I like what this does. > I am still in the middle of reviewing this. Though in first testing I saw > that I can't build a 32-bit gdbserver on Ubuntu 22.04. > I am building like this: > > export CFLAGS="-g -O0 -m32" > export CXXFLAGS="${CFLAGS} -m32" > export LDFLAGS="-m32" > > $src/gdb/configure \ > --build="x86_64-linux-gnu" \ > --host="i686-linux-gnu" \ > --disable-gdb > > make -s -j$(nproc) all-gdbserver > > Not sure if there is an easier way of doing this, but this works for master. > With your series I see this: > > CXX nat/x86-linux-tdesc.o > /gdb/gdbserver/../gdb/nat/x86-linux-tdesc.c: In function ?const target_desc* > x86_linux_tdesc_for_tid(int, tribool*, gdb::function_view<void(long long unsigned int)>, const char*, uint64_t*)?: > /gdb/gdbserver/../gdb/nat/x86-linux-tdesc.c:54:7: error: ?have_ptrace_getfpxregs? was not declared in this scope; > did you mean ?have_ptrace_getregset?? > 54 | if (have_ptrace_getfpxregs == -1) > | ^~~~~~~~~~~~~~~~~~~~~~ > | have_ptrace_getregset > /gdb/gdbserver/../gdb/nat/x86-linux-tdesc.c:56:7: error: ?elf_fpxregset_t? was not declared in this scope > 56 | elf_fpxregset_t fpxregs; > | ^~~~~~~~~~~~~~~ > In file included from /gdb/gdbserver/../gdb/nat/x86-linux-tdesc.c:29: > /gdb/gdbserver/../gdb/nat/x86-linux-tdesc.c:58:53: error: ?fpxregs? was not declared in this scope > 58 | if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) > | ^~~~~~~ > /gdb/gdbserver/../gdb/nat/gdb_ptrace.h:142:57: note: in definition of macro ?ptrace? > 142 | ptrace ((PTRACE_TYPE_ARG1) request, pid, addr, data) > | ^~~~ > make[1]: *** [Makefile:577: nat/x86-linux-tdesc.o] Error 1 > make: *** [Makefile:13074: install-gdbserver] Error 2 > > I didn't investigate further yet, sorry. Thanks for bringing this to my attention. I'll investigate this tomorrow. Thanks, Andrew > > Regards, > Felix > Intel Deutschland GmbH > Registered Address: Am Campeon 10, 85579 Neubiberg, Germany > Tel: +49 89 99 8853-0, www.intel.de <http://www.intel.de> > Managing Directors: Christin Eisenschmid, Sharon Heck, Tiffany Doon Silva > Chairperson of the Supervisory Board: Nicole Lau > Registered Office: Munich > Commercial Register: Amtsgericht Muenchen HRB 186928 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv5 00/11] x86/Linux Target Description Changes 2024-04-05 12:33 ` [PATCHv4 00/10] " Andrew Burgess ` (11 preceding siblings ...) 2024-04-25 13:35 ` Willgerodt, Felix @ 2024-04-26 15:01 ` Andrew Burgess 2024-04-26 15:01 ` [PATCHv5 01/11] gdbserver/ipa/x86: remove unneeded declarations Andrew Burgess ` (11 more replies) 12 siblings, 12 replies; 116+ messages in thread From: Andrew Burgess @ 2024-04-26 15:01 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, John Baldwin In v5: - Felix pointed out that building gdbserver with the '-m32' flag on an x86-64 host would fail. This is fixed in V5 with the addition of patch #4. This patch moves the have_ptrace_getfpxregs global into the gdb/nat/ directory and fixes the includes so that the the declaration is seen where needed, - I've rebased onto a slightly later commit. In v4: - I tried merging V3, but it turned out I broke pretty much everything that wasn't x86 based when configured with --enable-targets=all, - The problem was a failure to correctly split the shared code between the gdb/arch/ and gdb/nat/ directories, as a consequence, code which is needed on a non x86 based host to support x86 based targets wasn't available to the compilation, and the build failed, - In V4 I've gone through every patch and resplit the code in a way which I now believe is correct, I've done the following tests: + On a non x86 host I've built GDB to support only the current host as a target, to support all targets, and to support x86-64 and i386 linux targets, + On an i386 virtual machine I built GDB only for the host as a target, and for all targets. I regression tested the all targets build for unix, native-gdbserver, and native-extended-gdbserver, + On an x86-64 machine I've built GDB for only the current host as a target, and for all targets. I regression tested the all targets build for unix, native-gdbserver, and native-extended-gdbserver. - Only patches 6, 8, and 10 require significant review. All of the other patches are pretty trivial (though reviews always welcome). - I think there's more improvements that can be made to the x86 target description creation/lookup/caching. This series only changes the Linux lookup, and we still cache i386/amd64/x32 separately. In the future I think we can merge all x86 target description caching into a single data structure, this would be for all OS variants and all ABI variants. Though making that "grand unification" will certainly require some of the code in this series to change, I think the bulk of it will remain, and trying to do everything in one series is just going to result in an even larger series. I'd prefer to get these first patches merged, then come back to build on this work once this is merged and we know there's no problems with it. In v3: - Rebased. Nasty merge conflict with 4bb20a6244b7091 which I think I've resolved, but am unable to test. Reposting so the author of that other commit can validate. - Initial testing looks good. Full tests are still running. In v2: - Rebase to current upstream/master, no merge conflicts, - Retested. --- Andrew Burgess (11): gdbserver/ipa/x86: remove unneeded declarations gdbserver: convert have_ptrace_getregset to a tribool gdb/x86: move reading of cs and ds state into gdb/nat directory gdb/x86: move have_ptrace_getfpxregs global into gdb/nat directory gdbserver/x86: move no-xml code earlier in x86_linux_read_description gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition gdb/gdbserver: share some code relating to target description creation gdb/arch: assert that X86_XSTATE_MPX is not set for x32 gdbserver: update target description creation for x86/linux gdb: move xcr0 == 0 check into i386_linux_core_read_description gdb/gdbserver: share x86/linux tdesc caching gdb/Makefile.in | 9 +- gdb/amd64-linux-tdep.c | 32 +-- gdb/amd64-linux-tdep.h | 6 - gdb/arch/amd64-linux-tdesc.c | 61 +++++ gdb/arch/amd64-linux-tdesc.h | 30 +++ gdb/arch/amd64.c | 8 +- gdb/arch/i386-linux-tdesc.c | 51 ++++ .../i386-linux-tdesc.h} | 17 +- gdb/arch/x86-linux-tdesc-features.c | 247 ++++++++++++++++++ gdb/arch/x86-linux-tdesc-features.h | 62 +++++ gdb/arch/x86-linux-tdesc.h | 37 +++ gdb/configure.nat | 8 +- gdb/configure.tgt | 11 +- gdb/i386-linux-nat.c | 18 +- gdb/i386-linux-tdep.c | 42 +-- gdb/i386-linux-tdep.h | 23 -- gdb/nat/x86-linux-tdesc.c | 124 +++++++++ gdb/nat/x86-linux-tdesc.h | 60 +++++ gdb/nat/x86-linux.c | 56 ++++ gdb/nat/x86-linux.h | 44 ++++ gdb/x86-linux-nat.c | 126 +-------- gdbserver/configure.srv | 11 + gdbserver/linux-amd64-ipa.cc | 50 +--- gdbserver/linux-arm-low.cc | 6 +- gdbserver/linux-i386-ipa.cc | 26 +- gdbserver/linux-low.cc | 2 +- gdbserver/linux-low.h | 2 +- gdbserver/linux-x86-low.cc | 201 +++++--------- gdbserver/linux-x86-tdesc.cc | 142 +--------- gdbserver/linux-x86-tdesc.h | 56 ---- gdbsupport/x86-xstate.h | 20 ++ 31 files changed, 959 insertions(+), 629 deletions(-) create mode 100644 gdb/arch/amd64-linux-tdesc.c create mode 100644 gdb/arch/amd64-linux-tdesc.h create mode 100644 gdb/arch/i386-linux-tdesc.c rename gdb/{i386-linux-nat.h => arch/i386-linux-tdesc.h} (64%) create mode 100644 gdb/arch/x86-linux-tdesc-features.c create mode 100644 gdb/arch/x86-linux-tdesc-features.h create mode 100644 gdb/arch/x86-linux-tdesc.h create mode 100644 gdb/nat/x86-linux-tdesc.c create mode 100644 gdb/nat/x86-linux-tdesc.h delete mode 100644 gdbserver/linux-x86-tdesc.h base-commit: 39f0ac383169d7ef8b3169e483ae95dd185e41bd -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv5 01/11] gdbserver/ipa/x86: remove unneeded declarations 2024-04-26 15:01 ` [PATCHv5 00/11] " Andrew Burgess @ 2024-04-26 15:01 ` Andrew Burgess 2024-04-29 14:34 ` Willgerodt, Felix 2024-04-26 15:01 ` [PATCHv5 02/11] gdbserver: convert have_ptrace_getregset to a tribool Andrew Burgess ` (10 subsequent siblings) 11 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-04-26 15:01 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, John Baldwin Spotted some declarations in gdbserver/linux-amd64-ipa.cc that are no longer needed. These are: 1. 'init_registers_amd64_linux' - the comment claims this function is auto generated, but I don't believe that this is still the case. Also the function is not used in this file, 2. 'tdesc_amd64_linux' - this variable doesn't seem to exist any more, I suspect this was renamed to 'tdesc_amd64_linux_no_xml', but neither are used in this file, so lets remove the declaration. The amd64 in-process-agent still builds fine after this commit. There should be no user visible changes after this commit. --- gdbserver/linux-amd64-ipa.cc | 4 ---- 1 file changed, 4 deletions(-) diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index 13c8a5bb360..a6346750f49 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -23,10 +23,6 @@ #include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" -/* Defined in auto-generated file amd64-linux.c. */ -void init_registers_amd64_linux (void); -extern const struct target_desc *tdesc_amd64_linux; - /* fast tracepoints collect registers. */ #define FT_CR_RIP 0 -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 01/11] gdbserver/ipa/x86: remove unneeded declarations 2024-04-26 15:01 ` [PATCHv5 01/11] gdbserver/ipa/x86: remove unneeded declarations Andrew Burgess @ 2024-04-29 14:34 ` Willgerodt, Felix 2024-05-07 15:05 ` Andrew Burgess 0 siblings, 1 reply; 116+ messages in thread From: Willgerodt, Felix @ 2024-04-29 14:34 UTC (permalink / raw) To: Andrew Burgess, gdb-patches; +Cc: John Baldwin > -----Original Message----- > From: Andrew Burgess <aburgess@redhat.com> > Sent: Freitag, 26. April 2024 17:02 > To: gdb-patches@sourceware.org > Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix > <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> > Subject: [PATCHv5 01/11] gdbserver/ipa/x86: remove unneeded declarations > > Spotted some declarations in gdbserver/linux-amd64-ipa.cc that are no > longer needed. These are: > > 1. 'init_registers_amd64_linux' - the comment claims this function > is auto generated, but I don't believe that this is still the case. > Also the function is not used in this file, > > 2. 'tdesc_amd64_linux' - this variable doesn't seem to exist any > more, I suspect this was renamed to 'tdesc_amd64_linux_no_xml', but > neither are used in this file, so lets remove the declaration. > > The amd64 in-process-agent still builds fine after this commit. > > There should be no user visible changes after this commit. > --- > gdbserver/linux-amd64-ipa.cc | 4 ---- > 1 file changed, 4 deletions(-) > > diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc > index 13c8a5bb360..a6346750f49 100644 > --- a/gdbserver/linux-amd64-ipa.cc > +++ b/gdbserver/linux-amd64-ipa.cc > @@ -23,10 +23,6 @@ > #include "linux-x86-tdesc.h" > #include "gdbsupport/x86-xstate.h" > > -/* Defined in auto-generated file amd64-linux.c. */ > -void init_registers_amd64_linux (void); > -extern const struct target_desc *tdesc_amd64_linux; > - > /* fast tracepoints collect registers. */ > > #define FT_CR_RIP 0 > -- > 2.25.4 Approved-By: Felix Willgerodt <felix.willgerodt@intel.com> Thanks, Felix Intel Deutschland GmbH Registered Address: Am Campeon 10, 85579 Neubiberg, Germany Tel: +49 89 99 8853-0, www.intel.de <http://www.intel.de> Managing Directors: Christin Eisenschmid, Sharon Heck, Tiffany Doon Silva Chairperson of the Supervisory Board: Nicole Lau Registered Office: Munich Commercial Register: Amtsgericht Muenchen HRB 186928 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 01/11] gdbserver/ipa/x86: remove unneeded declarations 2024-04-29 14:34 ` Willgerodt, Felix @ 2024-05-07 15:05 ` Andrew Burgess 2024-05-08 7:49 ` Willgerodt, Felix 0 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-05-07 15:05 UTC (permalink / raw) To: Willgerodt, Felix, gdb-patches; +Cc: John Baldwin "Willgerodt, Felix" <felix.willgerodt@intel.com> writes: >> -----Original Message----- >> From: Andrew Burgess <aburgess@redhat.com> >> Sent: Freitag, 26. April 2024 17:02 >> To: gdb-patches@sourceware.org >> Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix >> <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> >> Subject: [PATCHv5 01/11] gdbserver/ipa/x86: remove unneeded declarations >> >> Spotted some declarations in gdbserver/linux-amd64-ipa.cc that are no >> longer needed. These are: >> >> 1. 'init_registers_amd64_linux' - the comment claims this function >> is auto generated, but I don't believe that this is still the case. >> Also the function is not used in this file, >> >> 2. 'tdesc_amd64_linux' - this variable doesn't seem to exist any >> more, I suspect this was renamed to 'tdesc_amd64_linux_no_xml', but >> neither are used in this file, so lets remove the declaration. >> >> The amd64 in-process-agent still builds fine after this commit. >> >> There should be no user visible changes after this commit. >> --- >> gdbserver/linux-amd64-ipa.cc | 4 ---- >> 1 file changed, 4 deletions(-) >> >> diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc >> index 13c8a5bb360..a6346750f49 100644 >> --- a/gdbserver/linux-amd64-ipa.cc >> +++ b/gdbserver/linux-amd64-ipa.cc >> @@ -23,10 +23,6 @@ >> #include "linux-x86-tdesc.h" >> #include "gdbsupport/x86-xstate.h" >> >> -/* Defined in auto-generated file amd64-linux.c. */ >> -void init_registers_amd64_linux (void); >> -extern const struct target_desc *tdesc_amd64_linux; >> - >> /* fast tracepoints collect registers. */ >> >> #define FT_CR_RIP 0 >> -- >> 2.25.4 > > Approved-By: Felix Willgerodt <felix.willgerodt@intel.com> Given this patch is pretty disjoint from the rest of the series I've gone ahead and pushed just this patch. Thanks, Andrew ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 01/11] gdbserver/ipa/x86: remove unneeded declarations 2024-05-07 15:05 ` Andrew Burgess @ 2024-05-08 7:49 ` Willgerodt, Felix 0 siblings, 0 replies; 116+ messages in thread From: Willgerodt, Felix @ 2024-05-08 7:49 UTC (permalink / raw) To: Andrew Burgess, gdb-patches; +Cc: John Baldwin > -----Original Message----- > From: Andrew Burgess <aburgess@redhat.com> > Sent: Dienstag, 7. Mai 2024 17:05 > To: Willgerodt, Felix <felix.willgerodt@intel.com>; gdb-patches@sourceware.org > Cc: John Baldwin <jhb@FreeBSD.org> > Subject: RE: [PATCHv5 01/11] gdbserver/ipa/x86: remove unneeded declarations > > "Willgerodt, Felix" <felix.willgerodt@intel.com> writes: > > >> -----Original Message----- > >> From: Andrew Burgess <aburgess@redhat.com> > >> Sent: Freitag, 26. April 2024 17:02 > >> To: gdb-patches@sourceware.org > >> Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix > >> <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> > >> Subject: [PATCHv5 01/11] gdbserver/ipa/x86: remove unneeded declarations > >> > >> Spotted some declarations in gdbserver/linux-amd64-ipa.cc that are no > >> longer needed. These are: > >> > >> 1. 'init_registers_amd64_linux' - the comment claims this function > >> is auto generated, but I don't believe that this is still the case. > >> Also the function is not used in this file, > >> > >> 2. 'tdesc_amd64_linux' - this variable doesn't seem to exist any > >> more, I suspect this was renamed to 'tdesc_amd64_linux_no_xml', but > >> neither are used in this file, so lets remove the declaration. > >> > >> The amd64 in-process-agent still builds fine after this commit. > >> > >> There should be no user visible changes after this commit. > >> --- > >> gdbserver/linux-amd64-ipa.cc | 4 ---- > >> 1 file changed, 4 deletions(-) > >> > >> diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc > >> index 13c8a5bb360..a6346750f49 100644 > >> --- a/gdbserver/linux-amd64-ipa.cc > >> +++ b/gdbserver/linux-amd64-ipa.cc > >> @@ -23,10 +23,6 @@ > >> #include "linux-x86-tdesc.h" > >> #include "gdbsupport/x86-xstate.h" > >> > >> -/* Defined in auto-generated file amd64-linux.c. */ > >> -void init_registers_amd64_linux (void); > >> -extern const struct target_desc *tdesc_amd64_linux; > >> - > >> /* fast tracepoints collect registers. */ > >> > >> #define FT_CR_RIP 0 > >> -- > >> 2.25.4 > > > > Approved-By: Felix Willgerodt <felix.willgerodt@intel.com> > > Given this patch is pretty disjoint from the rest of the series I've > gone ahead and pushed just this patch. > > Thanks, > Andrew Thanks! That is a good idea. Same for the others that you pushed. Felix Intel Deutschland GmbH Registered Address: Am Campeon 10, 85579 Neubiberg, Germany Tel: +49 89 99 8853-0, www.intel.de Managing Directors: Sean Fennelly, Jeffrey Schneiderman, Tiffany Doon Silva Chairperson of the Supervisory Board: Nicole Lau Registered Office: Munich Commercial Register: Amtsgericht Muenchen HRB 186928 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv5 02/11] gdbserver: convert have_ptrace_getregset to a tribool 2024-04-26 15:01 ` [PATCHv5 00/11] " Andrew Burgess 2024-04-26 15:01 ` [PATCHv5 01/11] gdbserver/ipa/x86: remove unneeded declarations Andrew Burgess @ 2024-04-26 15:01 ` Andrew Burgess 2024-04-29 14:34 ` Willgerodt, Felix 2024-04-26 15:01 ` [PATCHv5 03/11] gdb/x86: move reading of cs and ds state into gdb/nat directory Andrew Burgess ` (9 subsequent siblings) 11 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-04-26 15:01 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, John Baldwin Convert the have_ptrace_getregset global within gdbserver to a tribool. This brings the flag into alignment with the corresponding flag in GDB. The gdbserver have_ptrace_getregset variable is already used as a tribool, it just doesn't have the tribool type. In a future commit I plan to share more code between GDB and gdbserver, and having this variable be the same type in both code bases will make the sharing much easier. There should be no user visible changes after this commit. Approved-By: John Baldwin <jhb@FreeBSD.org> --- gdbserver/linux-arm-low.cc | 6 +++--- gdbserver/linux-low.cc | 2 +- gdbserver/linux-low.h | 2 +- gdbserver/linux-x86-low.cc | 10 +++++----- 4 files changed, 10 insertions(+), 10 deletions(-) diff --git a/gdbserver/linux-arm-low.cc b/gdbserver/linux-arm-low.cc index 17b64c09aff..eec4649b235 100644 --- a/gdbserver/linux-arm-low.cc +++ b/gdbserver/linux-arm-low.cc @@ -1006,9 +1006,9 @@ arm_target::low_arch_setup () /* Check if PTRACE_GETREGSET works. */ if (ptrace (PTRACE_GETREGSET, tid, NT_PRSTATUS, &iov) == 0) - have_ptrace_getregset = 1; + have_ptrace_getregset = TRIBOOL_TRUE; else - have_ptrace_getregset = 0; + have_ptrace_getregset = TRIBOOL_FALSE; } bool @@ -1121,7 +1121,7 @@ arm_target::get_regs_info () { const struct target_desc *tdesc = current_process ()->tdesc; - if (have_ptrace_getregset == 1 + if (have_ptrace_getregset == TRIBOOL_TRUE && (is_aarch32_linux_description (tdesc) || arm_linux_get_tdesc_fp_type (tdesc) == ARM_FP_TYPE_VFPV3)) return ®s_info_aarch32; diff --git a/gdbserver/linux-low.cc b/gdbserver/linux-low.cc index 9614cd7150c..ac7f9807ecc 100644 --- a/gdbserver/linux-low.cc +++ b/gdbserver/linux-low.cc @@ -134,7 +134,7 @@ typedef struct #endif /* Does the current host support PTRACE_GETREGSET? */ -int have_ptrace_getregset = -1; +enum tribool have_ptrace_getregset = TRIBOOL_UNKNOWN; /* Return TRUE if THREAD is the leader thread of the process. */ diff --git a/gdbserver/linux-low.h b/gdbserver/linux-low.h index d34d2738238..eaf87527338 100644 --- a/gdbserver/linux-low.h +++ b/gdbserver/linux-low.h @@ -951,7 +951,7 @@ void thread_db_notice_clone (struct thread_info *parent_thr, ptid_t child_ptid); bool thread_db_thread_handle (ptid_t ptid, gdb_byte **handle, int *handle_len); -extern int have_ptrace_getregset; +extern enum tribool have_ptrace_getregset; /* Search for the value with type MATCH in the auxv vector, with entries of length WORDSIZE bytes, of process with pid PID. If found, store the diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 2532603451a..2603fb2ac5d 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -898,7 +898,7 @@ x86_linux_read_description (void) if (ptrace (PTRACE_GETFPXREGS, tid, 0, (long) &fpxregs) < 0) { have_ptrace_getfpxregs = 0; - have_ptrace_getregset = 0; + have_ptrace_getregset = TRIBOOL_FALSE; return i386_linux_read_description (X86_XSTATE_X87); } else @@ -917,7 +917,7 @@ x86_linux_read_description (void) return tdesc_i386_linux_no_xml.get (); } - if (have_ptrace_getregset == -1) + if (have_ptrace_getregset == TRIBOOL_UNKNOWN) { uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; struct iovec iov; @@ -928,10 +928,10 @@ x86_linux_read_description (void) /* Check if PTRACE_GETREGSET works. */ if (ptrace (PTRACE_GETREGSET, tid, (unsigned int) NT_X86_XSTATE, (long) &iov) < 0) - have_ptrace_getregset = 0; + have_ptrace_getregset = TRIBOOL_FALSE; else { - have_ptrace_getregset = 1; + have_ptrace_getregset = TRIBOOL_TRUE; /* Get XCR0 from XSAVE extended state. */ xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET @@ -954,7 +954,7 @@ x86_linux_read_description (void) } /* Check the native XCR0 only if PTRACE_GETREGSET is available. */ - xcr0_features = (have_ptrace_getregset + xcr0_features = (have_ptrace_getregset == TRIBOOL_TRUE && (xcr0 & X86_XSTATE_ALL_MASK)); if (xcr0_features) -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 02/11] gdbserver: convert have_ptrace_getregset to a tribool 2024-04-26 15:01 ` [PATCHv5 02/11] gdbserver: convert have_ptrace_getregset to a tribool Andrew Burgess @ 2024-04-29 14:34 ` Willgerodt, Felix 2024-05-07 15:28 ` Andrew Burgess 0 siblings, 1 reply; 116+ messages in thread From: Willgerodt, Felix @ 2024-04-29 14:34 UTC (permalink / raw) To: Andrew Burgess, gdb-patches; +Cc: John Baldwin > -----Original Message----- > From: Andrew Burgess <aburgess@redhat.com> > Sent: Freitag, 26. April 2024 17:02 > To: gdb-patches@sourceware.org > Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix > <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> > Subject: [PATCHv5 02/11] gdbserver: convert have_ptrace_getregset to a tribool > > Convert the have_ptrace_getregset global within gdbserver to a > tribool. This brings the flag into alignment with the corresponding > flag in GDB. > > The gdbserver have_ptrace_getregset variable is already used as a > tribool, it just doesn't have the tribool type. > > In a future commit I plan to share more code between GDB and > gdbserver, and having this variable be the same type in both code > bases will make the sharing much easier. > > There should be no user visible changes after this commit. > > Approved-By: John Baldwin <jhb@FreeBSD.org> > --- > gdbserver/linux-arm-low.cc | 6 +++--- > gdbserver/linux-low.cc | 2 +- > gdbserver/linux-low.h | 2 +- > gdbserver/linux-x86-low.cc | 10 +++++----- > 4 files changed, 10 insertions(+), 10 deletions(-) Reviewed-By: Felix Willgerodt <felix.willgerodt@intel.com> Looks fine to me, x86 parts are approved. Thanks, Felix Intel Deutschland GmbH Registered Address: Am Campeon 10, 85579 Neubiberg, Germany Tel: +49 89 99 8853-0, www.intel.de <http://www.intel.de> Managing Directors: Christin Eisenschmid, Sharon Heck, Tiffany Doon Silva Chairperson of the Supervisory Board: Nicole Lau Registered Office: Munich Commercial Register: Amtsgericht Muenchen HRB 186928 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 02/11] gdbserver: convert have_ptrace_getregset to a tribool 2024-04-29 14:34 ` Willgerodt, Felix @ 2024-05-07 15:28 ` Andrew Burgess 0 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-07 15:28 UTC (permalink / raw) To: Willgerodt, Felix, gdb-patches; +Cc: John Baldwin "Willgerodt, Felix" <felix.willgerodt@intel.com> writes: >> -----Original Message----- >> From: Andrew Burgess <aburgess@redhat.com> >> Sent: Freitag, 26. April 2024 17:02 >> To: gdb-patches@sourceware.org >> Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix >> <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> >> Subject: [PATCHv5 02/11] gdbserver: convert have_ptrace_getregset to a tribool >> >> Convert the have_ptrace_getregset global within gdbserver to a >> tribool. This brings the flag into alignment with the corresponding >> flag in GDB. >> >> The gdbserver have_ptrace_getregset variable is already used as a >> tribool, it just doesn't have the tribool type. >> >> In a future commit I plan to share more code between GDB and >> gdbserver, and having this variable be the same type in both code >> bases will make the sharing much easier. >> >> There should be no user visible changes after this commit. >> >> Approved-By: John Baldwin <jhb@FreeBSD.org> >> --- >> gdbserver/linux-arm-low.cc | 6 +++--- >> gdbserver/linux-low.cc | 2 +- >> gdbserver/linux-low.h | 2 +- >> gdbserver/linux-x86-low.cc | 10 +++++----- >> 4 files changed, 10 insertions(+), 10 deletions(-) > > > Reviewed-By: Felix Willgerodt <felix.willgerodt@intel.com> > > Looks fine to me, x86 parts are approved. I've also pushed this patch from this series as it's only loosely related to the rest of the series. Thanks, Andrew ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv5 03/11] gdb/x86: move reading of cs and ds state into gdb/nat directory 2024-04-26 15:01 ` [PATCHv5 00/11] " Andrew Burgess 2024-04-26 15:01 ` [PATCHv5 01/11] gdbserver/ipa/x86: remove unneeded declarations Andrew Burgess 2024-04-26 15:01 ` [PATCHv5 02/11] gdbserver: convert have_ptrace_getregset to a tribool Andrew Burgess @ 2024-04-26 15:01 ` Andrew Burgess 2024-04-29 14:34 ` Willgerodt, Felix 2024-04-26 15:01 ` [PATCHv5 04/11] gdb/x86: move have_ptrace_getfpxregs global " Andrew Burgess ` (8 subsequent siblings) 11 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-04-26 15:01 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, John Baldwin This patch is part of a series that has the aim sharing the x86 Linux target description creation code between GDB and gdbserver. Within GDB part of this process involves reading the cs and ds state from the 'struct user_regs_struct' using a ptrace call. This isn't done by gdbserver, which is part of the motivation for this whole series; the approach gdbserver takes is inferior to the approach GDB takes (gdbserver relies on reading the file being debugged, and extracting similar information from the file headers). This commit moves the reading of cs and ds, which is used to figure out if a thread is 32-bit or 64-bit (or in x32 mode), into the gdb/nat directory so that the code can be shared with gdbserver, but at this point I'm not actually using the code in gdbserver, that will come later. As such there should be no user visible changes after this commit, GDB continues to do things as it did before (reading cs/ds), while gdbserver continues to use its own approach (which doesn't require reading cs/ds). Approved-By: John Baldwin <jhb@FreeBSD.org> --- gdb/nat/x86-linux.c | 47 +++++++++++++++++++++++++++++++++++++++++++++ gdb/nat/x86-linux.h | 28 +++++++++++++++++++++++++++ gdb/x86-linux-nat.c | 42 +++++----------------------------------- 3 files changed, 80 insertions(+), 37 deletions(-) diff --git a/gdb/nat/x86-linux.c b/gdb/nat/x86-linux.c index 7a21c8f2c26..ad3ed3c2289 100644 --- a/gdb/nat/x86-linux.c +++ b/gdb/nat/x86-linux.c @@ -19,6 +19,8 @@ #include "x86-linux.h" #include "x86-linux-dregs.h" +#include "nat/gdb_ptrace.h" +#include <sys/user.h> /* Per-thread arch-specific data we want to keep. */ @@ -79,3 +81,48 @@ x86_linux_prepare_to_resume (struct lwp_info *lwp) { x86_linux_update_debug_registers (lwp); } + +#ifdef __x86_64__ +/* Value of CS segment register: + 64bit process: 0x33 + 32bit process: 0x23 */ +#define AMD64_LINUX_USER64_CS 0x33 + +/* Value of DS segment register: + LP64 process: 0x0 + X32 process: 0x2b */ +#define AMD64_LINUX_X32_DS 0x2b +#endif + +/* See nat/x86-linux.h. */ + +x86_linux_arch_size +x86_linux_ptrace_get_arch_size (int tid) +{ +#ifdef __x86_64__ + unsigned long cs; + unsigned long ds; + + /* Get CS register. */ + errno = 0; + cs = ptrace (PTRACE_PEEKUSER, tid, + offsetof (struct user_regs_struct, cs), 0); + if (errno != 0) + perror_with_name (_("Couldn't get CS register")); + + bool is_64bit = cs == AMD64_LINUX_USER64_CS; + + /* Get DS register. */ + errno = 0; + ds = ptrace (PTRACE_PEEKUSER, tid, + offsetof (struct user_regs_struct, ds), 0); + if (errno != 0) + perror_with_name (_("Couldn't get DS register")); + + bool is_x32 = ds == AMD64_LINUX_X32_DS; + + return x86_linux_arch_size (is_64bit, is_x32); +#else + return x86_linux_arch_size (false, false); +#endif +} diff --git a/gdb/nat/x86-linux.h b/gdb/nat/x86-linux.h index 822882173f9..15153ea277e 100644 --- a/gdb/nat/x86-linux.h +++ b/gdb/nat/x86-linux.h @@ -47,4 +47,32 @@ extern void x86_linux_delete_thread (struct arch_lwp_info *arch_lwp); extern void x86_linux_prepare_to_resume (struct lwp_info *lwp); +/* Return value from x86_linux_ptrace_get_arch_size function. Indicates if + a thread is 32-bit, 64-bit, or x32. */ + +struct x86_linux_arch_size +{ + explicit x86_linux_arch_size (bool is_64bit, bool is_x32) + : m_is_64bit (is_64bit), + m_is_x32 (is_x32) + { + /* Nothing. */ + } + + bool is_64bit () const + { return m_is_64bit; } + + bool is_x32 () const + { return m_is_x32; } + +private: + bool m_is_64bit = false; + bool m_is_x32 = false; +}; + +/* Use ptrace calls to figure out if thread TID is 32-bit, 64-bit, or + 64-bit running in x32 mode. */ + +extern x86_linux_arch_size x86_linux_ptrace_get_arch_size (int tid); + #endif /* NAT_X86_LINUX_H */ diff --git a/gdb/x86-linux-nat.c b/gdb/x86-linux-nat.c index a3d8ffb60f1..f91db492d05 100644 --- a/gdb/x86-linux-nat.c +++ b/gdb/x86-linux-nat.c @@ -90,18 +90,6 @@ x86_linux_nat_target::post_startup_inferior (ptid_t ptid) linux_nat_target::post_startup_inferior (ptid); } -#ifdef __x86_64__ -/* Value of CS segment register: - 64bit process: 0x33 - 32bit process: 0x23 */ -#define AMD64_LINUX_USER64_CS 0x33 - -/* Value of DS segment register: - LP64 process: 0x0 - X32 process: 0x2b */ -#define AMD64_LINUX_X32_DS 0x2b -#endif - /* Get Linux/x86 target description from running target. */ const struct target_desc * @@ -121,31 +109,11 @@ x86_linux_nat_target::read_description () tid = inferior_ptid.pid (); #ifdef __x86_64__ - { - unsigned long cs; - unsigned long ds; - - /* Get CS register. */ - errno = 0; - cs = ptrace (PTRACE_PEEKUSER, tid, - offsetof (struct user_regs_struct, cs), 0); - if (errno != 0) - perror_with_name (_("Couldn't get CS register")); - - is_64bit = cs == AMD64_LINUX_USER64_CS; - - /* Get DS register. */ - errno = 0; - ds = ptrace (PTRACE_PEEKUSER, tid, - offsetof (struct user_regs_struct, ds), 0); - if (errno != 0) - perror_with_name (_("Couldn't get DS register")); - - is_x32 = ds == AMD64_LINUX_X32_DS; - - if (sizeof (void *) == 4 && is_64bit && !is_x32) - error (_("Can't debug 64-bit process with 32-bit GDB")); - } + + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); + is_64bit = arch_size.is_64bit (); + is_x32 = arch_size.is_x32 (); + #elif HAVE_PTRACE_GETFPXREGS if (have_ptrace_getfpxregs == -1) { -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 03/11] gdb/x86: move reading of cs and ds state into gdb/nat directory 2024-04-26 15:01 ` [PATCHv5 03/11] gdb/x86: move reading of cs and ds state into gdb/nat directory Andrew Burgess @ 2024-04-29 14:34 ` Willgerodt, Felix 0 siblings, 0 replies; 116+ messages in thread From: Willgerodt, Felix @ 2024-04-29 14:34 UTC (permalink / raw) To: Andrew Burgess, gdb-patches; +Cc: John Baldwin > -----Original Message----- > From: Andrew Burgess <aburgess@redhat.com> > Sent: Freitag, 26. April 2024 17:02 > To: gdb-patches@sourceware.org > Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix > <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> > Subject: [PATCHv5 03/11] gdb/x86: move reading of cs and ds state into gdb/nat > directory > > This patch is part of a series that has the aim sharing the x86 Linux > target description creation code between GDB and gdbserver. > > Within GDB part of this process involves reading the cs and ds state > from the 'struct user_regs_struct' using a ptrace call. > > This isn't done by gdbserver, which is part of the motivation for this > whole series; the approach gdbserver takes is inferior to the approach > GDB takes (gdbserver relies on reading the file being debugged, and > extracting similar information from the file headers). > > This commit moves the reading of cs and ds, which is used to figure > out if a thread is 32-bit or 64-bit (or in x32 mode), into the gdb/nat > directory so that the code can be shared with gdbserver, but at this > point I'm not actually using the code in gdbserver, that will come > later. > > As such there should be no user visible changes after this commit, GDB > continues to do things as it did before (reading cs/ds), while > gdbserver continues to use its own approach (which doesn't require > reading cs/ds). > > Approved-By: John Baldwin <jhb@FreeBSD.org> > --- > gdb/nat/x86-linux.c | 47 > +++++++++++++++++++++++++++++++++++++++++++++ > gdb/nat/x86-linux.h | 28 +++++++++++++++++++++++++++ > gdb/x86-linux-nat.c | 42 +++++----------------------------------- > 3 files changed, 80 insertions(+), 37 deletions(-) > > diff --git a/gdb/nat/x86-linux.c b/gdb/nat/x86-linux.c > index 7a21c8f2c26..ad3ed3c2289 100644 > --- a/gdb/nat/x86-linux.c > +++ b/gdb/nat/x86-linux.c > @@ -19,6 +19,8 @@ > > #include "x86-linux.h" > #include "x86-linux-dregs.h" > +#include "nat/gdb_ptrace.h" > +#include <sys/user.h> > > /* Per-thread arch-specific data we want to keep. */ > > @@ -79,3 +81,48 @@ x86_linux_prepare_to_resume (struct lwp_info *lwp) > { > x86_linux_update_debug_registers (lwp); > } > + > +#ifdef __x86_64__ > +/* Value of CS segment register: > + 64bit process: 0x33 > + 32bit process: 0x23 */ > +#define AMD64_LINUX_USER64_CS 0x33 > + > +/* Value of DS segment register: > + LP64 process: 0x0 > + X32 process: 0x2b */ > +#define AMD64_LINUX_X32_DS 0x2b > +#endif > + > +/* See nat/x86-linux.h. */ > + > +x86_linux_arch_size > +x86_linux_ptrace_get_arch_size (int tid) > +{ > +#ifdef __x86_64__ > + unsigned long cs; > + unsigned long ds; > + > + /* Get CS register. */ > + errno = 0; > + cs = ptrace (PTRACE_PEEKUSER, tid, > + offsetof (struct user_regs_struct, cs), 0); > + if (errno != 0) > + perror_with_name (_("Couldn't get CS register")); > + > + bool is_64bit = cs == AMD64_LINUX_USER64_CS; > + > + /* Get DS register. */ > + errno = 0; > + ds = ptrace (PTRACE_PEEKUSER, tid, > + offsetof (struct user_regs_struct, ds), 0); > + if (errno != 0) > + perror_with_name (_("Couldn't get DS register")); > + > + bool is_x32 = ds == AMD64_LINUX_X32_DS; > + > + return x86_linux_arch_size (is_64bit, is_x32); > +#else > + return x86_linux_arch_size (false, false); > +#endif The old function had: if (sizeof (void *) == 4 && is_64bit && !is_x32) error (_("Can't debug 64-bit process with 32-bit GDB")); That makes me wonder: Was this just dead code or do we actually need this still? I saw that you add a similar message back in later commits. But here there might be user visible change, contrary to your commit message. The rest looked fine to me. Thanks, Felix Intel Deutschland GmbH Registered Address: Am Campeon 10, 85579 Neubiberg, Germany Tel: +49 89 99 8853-0, www.intel.de <http://www.intel.de> Managing Directors: Christin Eisenschmid, Sharon Heck, Tiffany Doon Silva Chairperson of the Supervisory Board: Nicole Lau Registered Office: Munich Commercial Register: Amtsgericht Muenchen HRB 186928 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv5 04/11] gdb/x86: move have_ptrace_getfpxregs global into gdb/nat directory 2024-04-26 15:01 ` [PATCHv5 00/11] " Andrew Burgess ` (2 preceding siblings ...) 2024-04-26 15:01 ` [PATCHv5 03/11] gdb/x86: move reading of cs and ds state into gdb/nat directory Andrew Burgess @ 2024-04-26 15:01 ` Andrew Burgess 2024-04-29 14:34 ` Willgerodt, Felix 2024-04-26 15:01 ` [PATCHv5 05/11] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess ` (7 subsequent siblings) 11 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-04-26 15:01 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, John Baldwin The have_ptrace_getfpxregs global tracks whether GDB or gdbserver is running on a kernel that supports the GETFPXREGS ptrace request. Currently this global is declared twice (once in GDB and once in gdbserver), but it makes sense to move this global into the nat/ directory, and have a single declaration and definition. This is a refactor, there should be no user visible changes after this commit. --- gdb/Makefile.in | 1 - gdb/i386-linux-nat.c | 18 +----------------- gdb/i386-linux-nat.h | 26 -------------------------- gdb/nat/x86-linux.c | 9 +++++++++ gdb/nat/x86-linux.h | 16 ++++++++++++++++ gdb/x86-linux-nat.c | 3 --- gdbserver/linux-x86-low.cc | 11 ----------- 7 files changed, 26 insertions(+), 58 deletions(-) delete mode 100644 gdb/i386-linux-nat.h diff --git a/gdb/Makefile.in b/gdb/Makefile.in index 23894ea4a4d..353e45b3cec 100644 --- a/gdb/Makefile.in +++ b/gdb/Makefile.in @@ -1396,7 +1396,6 @@ HFILES_NO_SRCDIR = \ hppa-tdep.h \ i386-bsd-nat.h \ i386-darwin-tdep.h \ - i386-linux-nat.h \ i386-linux-tdep.h \ i386-tdep.h \ i387-tdep.h \ diff --git a/gdb/i386-linux-nat.c b/gdb/i386-linux-nat.c index 7278dd91ff0..32d9d998beb 100644 --- a/gdb/i386-linux-nat.c +++ b/gdb/i386-linux-nat.c @@ -26,7 +26,7 @@ #include "gregset.h" #include "gdb_proc_service.h" -#include "i386-linux-nat.h" +#include "nat/x86-linux.h" #include "i387-tdep.h" #include "i386-tdep.h" #include "i386-linux-tdep.h" @@ -80,22 +80,6 @@ int have_ptrace_getregs = 0 #endif ; - -/* Does the current host support the GETFPXREGS request? The header - file may or may not define it, and even if it is defined, the - kernel will return EIO if it's running on a pre-SSE processor. - - My instinct is to attach this to some architecture- or - target-specific data structure, but really, a particular GDB - process can only run on top of one kernel at a time. So it's okay - for this to be a simple variable. */ -int have_ptrace_getfpxregs = -#ifdef HAVE_PTRACE_GETFPXREGS - -1 -#else - 0 -#endif -; \f /* Accessing registers through the U area, one at a time. */ diff --git a/gdb/i386-linux-nat.h b/gdb/i386-linux-nat.h deleted file mode 100644 index c3d32aff1be..00000000000 --- a/gdb/i386-linux-nat.h +++ /dev/null @@ -1,26 +0,0 @@ -/* Native-dependent code for GNU/Linux i386. - - Copyright (C) 1999-2024 Free Software Foundation, Inc. - - This file is part of GDB. - - This program is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published by - the Free Software Foundation; either version 3 of the License, or - (at your option) any later version. - - This program is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - GNU General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see <http://www.gnu.org/licenses/>. */ - -#ifndef I386_LINUX_NAT_H -#define I386_LINUX_NAT_H 1 - -/* Does the current host support the GETFPXREGS request? */ -extern int have_ptrace_getfpxregs; - -#endif diff --git a/gdb/nat/x86-linux.c b/gdb/nat/x86-linux.c index ad3ed3c2289..35a3e960946 100644 --- a/gdb/nat/x86-linux.c +++ b/gdb/nat/x86-linux.c @@ -22,6 +22,15 @@ #include "nat/gdb_ptrace.h" #include <sys/user.h> +/* See nat/x86-linux.h. */ +int have_ptrace_getfpxregs = +#ifdef HAVE_PTRACE_GETFPXREGS + -1 +#else + 0 +#endif +; + /* Per-thread arch-specific data we want to keep. */ struct arch_lwp_info diff --git a/gdb/nat/x86-linux.h b/gdb/nat/x86-linux.h index 15153ea277e..061b3c2e394 100644 --- a/gdb/nat/x86-linux.h +++ b/gdb/nat/x86-linux.h @@ -22,6 +22,22 @@ #include "nat/linux-nat.h" +/* Does the current host support the GETFPXREGS request? The system header + file may or may not define it, but even if it is defined, the kernel + will return EIO if it's running on a pre-SSE processor. + + A value of -1 indicates that we don't know yet, we've not yet tried the + ptrace call; a value of 0 indicates we've previously tried the ptrace + call and it failed, indicating this request is not supported; and a + value of 1 indicates that we've previously tried the ptrace call and it + has been successful. + + My instinct is to attach this to some architecture- or target-specific + data structure, but really, a particular GDB process can only run on top + of one kernel at a time. So it's okay - for this to be a global + variable. */ +extern int have_ptrace_getfpxregs; + /* Set whether our local mirror of LWP's debug registers has been changed since the values were last written to the thread. Nonzero indicates that a change has been made, zero indicates no change. */ diff --git a/gdb/x86-linux-nat.c b/gdb/x86-linux-nat.c index f91db492d05..12965c7a21b 100644 --- a/gdb/x86-linux-nat.c +++ b/gdb/x86-linux-nat.c @@ -26,9 +26,6 @@ #include <sys/uio.h> #include "x86-nat.h" -#ifndef __x86_64__ -#include "i386-linux-nat.h" -#endif #include "x86-linux-nat.h" #include "i386-linux-tdep.h" #ifdef __x86_64__ diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 2603fb2ac5d..8199b0ccc09 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -852,17 +852,6 @@ static int use_xml; the process/thread is in. */ #define I386_LINUX_XSAVE_XCR0_OFFSET 464 -/* Does the current host support the GETFPXREGS request? The header - file may or may not define it, and even if it is defined, the - kernel will return EIO if it's running on a pre-SSE processor. */ -int have_ptrace_getfpxregs = -#ifdef HAVE_PTRACE_GETFPXREGS - -1 -#else - 0 -#endif -; - /* Get Linux/x86 target description from running target. */ static const struct target_desc * -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 04/11] gdb/x86: move have_ptrace_getfpxregs global into gdb/nat directory 2024-04-26 15:01 ` [PATCHv5 04/11] gdb/x86: move have_ptrace_getfpxregs global " Andrew Burgess @ 2024-04-29 14:34 ` Willgerodt, Felix 0 siblings, 0 replies; 116+ messages in thread From: Willgerodt, Felix @ 2024-04-29 14:34 UTC (permalink / raw) To: Andrew Burgess, gdb-patches; +Cc: John Baldwin > -----Original Message----- > From: Andrew Burgess <aburgess@redhat.com> > Sent: Freitag, 26. April 2024 17:02 > To: gdb-patches@sourceware.org > Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix > <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> > Subject: [PATCHv5 04/11] gdb/x86: move have_ptrace_getfpxregs global into > gdb/nat directory > > The have_ptrace_getfpxregs global tracks whether GDB or gdbserver is > running on a kernel that supports the GETFPXREGS ptrace request. > > Currently this global is declared twice (once in GDB and once in > gdbserver), but it makes sense to move this global into the nat/ > directory, and have a single declaration and definition. > > This is a refactor, there should be no user visible changes after this > commit. > --- > gdb/Makefile.in | 1 - > gdb/i386-linux-nat.c | 18 +----------------- > gdb/i386-linux-nat.h | 26 -------------------------- > gdb/nat/x86-linux.c | 9 +++++++++ > gdb/nat/x86-linux.h | 16 ++++++++++++++++ > gdb/x86-linux-nat.c | 3 --- > gdbserver/linux-x86-low.cc | 11 ----------- > 7 files changed, 26 insertions(+), 58 deletions(-) > delete mode 100644 gdb/i386-linux-nat.h This fixes my m32 gdbserver build and lgtm in general. But since we are already touching it, can't we make this a tribool as well? Thanks, Felix Intel Deutschland GmbH Registered Address: Am Campeon 10, 85579 Neubiberg, Germany Tel: +49 89 99 8853-0, www.intel.de <http://www.intel.de> Managing Directors: Christin Eisenschmid, Sharon Heck, Tiffany Doon Silva Chairperson of the Supervisory Board: Nicole Lau Registered Office: Munich Commercial Register: Amtsgericht Muenchen HRB 186928 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv5 05/11] gdbserver/x86: move no-xml code earlier in x86_linux_read_description 2024-04-26 15:01 ` [PATCHv5 00/11] " Andrew Burgess ` (3 preceding siblings ...) 2024-04-26 15:01 ` [PATCHv5 04/11] gdb/x86: move have_ptrace_getfpxregs global " Andrew Burgess @ 2024-04-26 15:01 ` Andrew Burgess 2024-04-29 14:34 ` Willgerodt, Felix 2024-04-26 15:01 ` [PATCHv5 06/11] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess ` (6 subsequent siblings) 11 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-04-26 15:01 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, John Baldwin This commit is part of a series that aims to share more of the x86 target description reading/generation code between GDB and gdbserver. There are a huge number of similarities between the code in gdbserver's x86_linux_read_description function and GDB's x86_linux_nat_target::read_description function, and it is this similarity that I plan, in a later commit, to share between GDB and gdbserver. However, one thing that is different in x86_linux_read_description is the code inside the '!use_xml' block. This is the code that handles the case where gdbserver is not allowed to send an XML target description back to GDB. In this case gdbserver uses some predefined, fixed, target descriptions. First, it's worth noting that I suspect this code is not tested any more. I couldn't find anything in the testsuite that tries to disable XML target description support. And the idea of having a single "fixed" target description really doesn't work well when we think about all the various x86 extensions that exist. Part of me would like to rip out the no-xml support in gdbserver (at least for x86), and if a GDB connects that doesn't support XML target descriptions, gdbserver can just give an error and drop the connection. GDB has supported XML target descriptions for 16 years now, I think it would be reasonable for our shipped gdbserver to drop support for the old way of doing things. Anyway.... this commit doesn't do that. What I did notice was that, over time, the '!use_xml' block appears to have "drifted" within the x86_linux_read_description function; it's now not the first check we do. Instead we make some ptrace calls and return a target description generated based on the result of these ptrace calls. Surely it only makes sense to generate variable target descriptions if we can send these back to GDB? So in this commit I propose to move the '!use_xml' block earlier in the x86_linux_read_description function. The benefit of this is that this leaves the later half of x86_linux_read_description much more similar to the GDB function x86_linux_nat_target::read_description and sets us up for potentially sharing code between GDB and gdbserver in a later commit. Approved-By: John Baldwin <jhb@FreeBSD.org> --- gdbserver/linux-x86-low.cc | 27 ++++++++++++++++----------- 1 file changed, 16 insertions(+), 11 deletions(-) diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 8199b0ccc09..ffb2a90731d 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -879,6 +879,22 @@ x86_linux_read_description (void) #endif } + /* If we are not allowed to send an XML target description then we need + to use the hard-wired target descriptions. This corresponds to GDB's + default machine for x86. + + This check needs to occur before any returns statements that might + generate some alternative target descriptions. */ + if (!use_xml) + { +#ifdef __x86_64__ + if (machine == EM_X86_64) + return tdesc_amd64_linux_no_xml.get (); + else +#endif + return tdesc_i386_linux_no_xml.get (); + } + #if !defined __x86_64__ && defined HAVE_PTRACE_GETFPXREGS if (machine == EM_386 && have_ptrace_getfpxregs == -1) { @@ -895,17 +911,6 @@ x86_linux_read_description (void) } #endif - if (!use_xml) - { - /* Don't use XML. */ -#ifdef __x86_64__ - if (machine == EM_X86_64) - return tdesc_amd64_linux_no_xml.get (); - else -#endif - return tdesc_i386_linux_no_xml.get (); - } - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) { uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 05/11] gdbserver/x86: move no-xml code earlier in x86_linux_read_description 2024-04-26 15:01 ` [PATCHv5 05/11] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess @ 2024-04-29 14:34 ` Willgerodt, Felix 2024-05-07 11:55 ` Luis Machado 0 siblings, 1 reply; 116+ messages in thread From: Willgerodt, Felix @ 2024-04-29 14:34 UTC (permalink / raw) To: Andrew Burgess, gdb-patches; +Cc: John Baldwin > -----Original Message----- > From: Andrew Burgess <aburgess@redhat.com> > Sent: Freitag, 26. April 2024 17:02 > To: gdb-patches@sourceware.org > Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix > <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> > Subject: [PATCHv5 05/11] gdbserver/x86: move no-xml code earlier in > x86_linux_read_description > > This commit is part of a series that aims to share more of the x86 > target description reading/generation code between GDB and gdbserver. > > There are a huge number of similarities between the code in > gdbserver's x86_linux_read_description function and GDB's > x86_linux_nat_target::read_description function, and it is this > similarity that I plan, in a later commit, to share between GDB and > gdbserver. > > However, one thing that is different in x86_linux_read_description is > the code inside the '!use_xml' block. This is the code that handles > the case where gdbserver is not allowed to send an XML target > description back to GDB. In this case gdbserver uses some predefined, > fixed, target descriptions. > > First, it's worth noting that I suspect this code is not tested any > more. I couldn't find anything in the testsuite that tries to disable > XML target description support. And the idea of having a single > "fixed" target description really doesn't work well when we think > about all the various x86 extensions that exist. Part of me would > like to rip out the no-xml support in gdbserver (at least for x86), > and if a GDB connects that doesn't support XML target descriptions, > gdbserver can just give an error and drop the connection. GDB has > supported XML target descriptions for 16 years now, I think it would > be reasonable for our shipped gdbserver to drop support for the old > way of doing things. Interesting. I for one would +1 this. Slightly related: I wonder if anyone really builds GDB without libexpat anymore and if we couldn't even think of making it mandatory. (Which doesn't mean dropping the ball on supporting stubs without XML.) > Anyway.... this commit doesn't do that. > > What I did notice was that, over time, the '!use_xml' block appears to > have "drifted" within the x86_linux_read_description function; it's > now not the first check we do. Instead we make some ptrace calls and > return a target description generated based on the result of these > ptrace calls. Surely it only makes sense to generate variable target > descriptions if we can send these back to GDB? > > So in this commit I propose to move the '!use_xml' block earlier in > the x86_linux_read_description function. > > The benefit of this is that this leaves the later half of > x86_linux_read_description much more similar to the GDB function > x86_linux_nat_target::read_description and sets us up for potentially > sharing code between GDB and gdbserver in a later commit. > > Approved-By: John Baldwin <jhb@FreeBSD.org> > --- > gdbserver/linux-x86-low.cc | 27 ++++++++++++++++----------- > 1 file changed, 16 insertions(+), 11 deletions(-) > Looks fine to me. Approved-By: Felix Willgerodt <felix.willgerodt@intel.com> Thanks, Felix Intel Deutschland GmbH Registered Address: Am Campeon 10, 85579 Neubiberg, Germany Tel: +49 89 99 8853-0, www.intel.de <http://www.intel.de> Managing Directors: Christin Eisenschmid, Sharon Heck, Tiffany Doon Silva Chairperson of the Supervisory Board: Nicole Lau Registered Office: Munich Commercial Register: Amtsgericht Muenchen HRB 186928 ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv5 05/11] gdbserver/x86: move no-xml code earlier in x86_linux_read_description 2024-04-29 14:34 ` Willgerodt, Felix @ 2024-05-07 11:55 ` Luis Machado 2024-05-07 15:43 ` Andrew Burgess 0 siblings, 1 reply; 116+ messages in thread From: Luis Machado @ 2024-05-07 11:55 UTC (permalink / raw) To: Willgerodt, Felix, Andrew Burgess, gdb-patches; +Cc: John Baldwin On 4/29/24 15:34, Willgerodt, Felix wrote: >> -----Original Message----- >> From: Andrew Burgess <aburgess@redhat.com> >> Sent: Freitag, 26. April 2024 17:02 >> To: gdb-patches@sourceware.org >> Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix >> <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> >> Subject: [PATCHv5 05/11] gdbserver/x86: move no-xml code earlier in >> x86_linux_read_description >> >> This commit is part of a series that aims to share more of the x86 >> target description reading/generation code between GDB and gdbserver. >> >> There are a huge number of similarities between the code in >> gdbserver's x86_linux_read_description function and GDB's >> x86_linux_nat_target::read_description function, and it is this >> similarity that I plan, in a later commit, to share between GDB and >> gdbserver. >> >> However, one thing that is different in x86_linux_read_description is >> the code inside the '!use_xml' block. This is the code that handles >> the case where gdbserver is not allowed to send an XML target >> description back to GDB. In this case gdbserver uses some predefined, >> fixed, target descriptions. >> >> First, it's worth noting that I suspect this code is not tested any >> more. I couldn't find anything in the testsuite that tries to disable >> XML target description support. And the idea of having a single >> "fixed" target description really doesn't work well when we think >> about all the various x86 extensions that exist. Part of me would >> like to rip out the no-xml support in gdbserver (at least for x86), >> and if a GDB connects that doesn't support XML target descriptions, >> gdbserver can just give an error and drop the connection. GDB has >> supported XML target descriptions for 16 years now, I think it would >> be reasonable for our shipped gdbserver to drop support for the old >> way of doing things. > > Interesting. I for one would +1 this. Slightly related: > I wonder if anyone really builds GDB without libexpat anymore and > if we couldn't even think of making it mandatory. (Which doesn't > mean dropping the ball on supporting stubs without XML.) I'd support making xml descriptions mandatory for some targets. AArch64 and 32-bit Arm both rely heavily on xml description for feature discovery. Off the top of my head, I think very old probes and the Linux Kernel's kgdb don't support xml. I'm not sure how often kgdb gets used though, but falling back to a default description should work for AArch64 and 32-bit Arm. ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv5 05/11] gdbserver/x86: move no-xml code earlier in x86_linux_read_description 2024-05-07 11:55 ` Luis Machado @ 2024-05-07 15:43 ` Andrew Burgess 2024-05-07 15:56 ` Luis Machado 2024-05-08 7:49 ` Willgerodt, Felix 0 siblings, 2 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-07 15:43 UTC (permalink / raw) To: Luis Machado, Willgerodt, Felix, gdb-patches; +Cc: John Baldwin Luis Machado <luis.machado@arm.com> writes: > On 4/29/24 15:34, Willgerodt, Felix wrote: >>> -----Original Message----- >>> From: Andrew Burgess <aburgess@redhat.com> >>> Sent: Freitag, 26. April 2024 17:02 >>> To: gdb-patches@sourceware.org >>> Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix >>> <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> >>> Subject: [PATCHv5 05/11] gdbserver/x86: move no-xml code earlier in >>> x86_linux_read_description >>> >>> This commit is part of a series that aims to share more of the x86 >>> target description reading/generation code between GDB and gdbserver. >>> >>> There are a huge number of similarities between the code in >>> gdbserver's x86_linux_read_description function and GDB's >>> x86_linux_nat_target::read_description function, and it is this >>> similarity that I plan, in a later commit, to share between GDB and >>> gdbserver. >>> >>> However, one thing that is different in x86_linux_read_description is >>> the code inside the '!use_xml' block. This is the code that handles >>> the case where gdbserver is not allowed to send an XML target >>> description back to GDB. In this case gdbserver uses some predefined, >>> fixed, target descriptions. >>> >>> First, it's worth noting that I suspect this code is not tested any >>> more. I couldn't find anything in the testsuite that tries to disable >>> XML target description support. And the idea of having a single >>> "fixed" target description really doesn't work well when we think >>> about all the various x86 extensions that exist. Part of me would >>> like to rip out the no-xml support in gdbserver (at least for x86), >>> and if a GDB connects that doesn't support XML target descriptions, >>> gdbserver can just give an error and drop the connection. GDB has >>> supported XML target descriptions for 16 years now, I think it would >>> be reasonable for our shipped gdbserver to drop support for the old >>> way of doing things. >> >> Interesting. I for one would +1 this. Slightly related: >> I wonder if anyone really builds GDB without libexpat anymore and >> if we couldn't even think of making it mandatory. (Which doesn't >> mean dropping the ball on supporting stubs without XML.) > > I'd support making xml descriptions mandatory for some targets. AArch64 and > 32-bit Arm both rely heavily on xml description for feature discovery. I might be wrong, but I don't think Felix was even suggesting we need to go that far. I think the suggestion was just that GDB itself should require libexpat, which would mean we have the potential to read XML target descriptions. This wouldn't mean that targets would be required to send XML target descriptions. My comment was specifically that x86 gdbserver should drop its no XML support, and if a user tries to connect with a no XML GDB then gdbserver would reject the connection. Moving to make libexpat a requirement would help in this case as a user would no longer be able to build a GDB which couldn't connect to an x86 gdbserver. > Off the top of my head, I think very old probes and the Linux Kernel's kgdb don't > support xml. I'm not sure how often kgdb gets used though, but falling back to a > default description should work for AArch64 and 32-bit Arm. I don't believe building GDB with libexpat support breaks the ability to connect to a target which doesn't support XML descriptions (if it does then that's a bug), so hopefully nobody else would be inconvenienced... except for those folk who are building GDB without libexpat support. I've added making these changes to my todo list, but I don't want to add them to this series. Thanks, Andrew ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv5 05/11] gdbserver/x86: move no-xml code earlier in x86_linux_read_description 2024-05-07 15:43 ` Andrew Burgess @ 2024-05-07 15:56 ` Luis Machado 2024-05-08 7:49 ` Willgerodt, Felix 1 sibling, 0 replies; 116+ messages in thread From: Luis Machado @ 2024-05-07 15:56 UTC (permalink / raw) To: Andrew Burgess, Willgerodt, Felix, gdb-patches; +Cc: John Baldwin On 5/7/24 16:43, Andrew Burgess wrote: > Luis Machado <luis.machado@arm.com> writes: > >> On 4/29/24 15:34, Willgerodt, Felix wrote: >>>> -----Original Message----- >>>> From: Andrew Burgess <aburgess@redhat.com> >>>> Sent: Freitag, 26. April 2024 17:02 >>>> To: gdb-patches@sourceware.org >>>> Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix >>>> <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> >>>> Subject: [PATCHv5 05/11] gdbserver/x86: move no-xml code earlier in >>>> x86_linux_read_description >>>> >>>> This commit is part of a series that aims to share more of the x86 >>>> target description reading/generation code between GDB and gdbserver. >>>> >>>> There are a huge number of similarities between the code in >>>> gdbserver's x86_linux_read_description function and GDB's >>>> x86_linux_nat_target::read_description function, and it is this >>>> similarity that I plan, in a later commit, to share between GDB and >>>> gdbserver. >>>> >>>> However, one thing that is different in x86_linux_read_description is >>>> the code inside the '!use_xml' block. This is the code that handles >>>> the case where gdbserver is not allowed to send an XML target >>>> description back to GDB. In this case gdbserver uses some predefined, >>>> fixed, target descriptions. >>>> >>>> First, it's worth noting that I suspect this code is not tested any >>>> more. I couldn't find anything in the testsuite that tries to disable >>>> XML target description support. And the idea of having a single >>>> "fixed" target description really doesn't work well when we think >>>> about all the various x86 extensions that exist. Part of me would >>>> like to rip out the no-xml support in gdbserver (at least for x86), >>>> and if a GDB connects that doesn't support XML target descriptions, >>>> gdbserver can just give an error and drop the connection. GDB has >>>> supported XML target descriptions for 16 years now, I think it would >>>> be reasonable for our shipped gdbserver to drop support for the old >>>> way of doing things. >>> >>> Interesting. I for one would +1 this. Slightly related: >>> I wonder if anyone really builds GDB without libexpat anymore and >>> if we couldn't even think of making it mandatory. (Which doesn't >>> mean dropping the ball on supporting stubs without XML.) >> >> I'd support making xml descriptions mandatory for some targets. AArch64 and >> 32-bit Arm both rely heavily on xml description for feature discovery. > > I might be wrong, but I don't think Felix was even suggesting we need to > go that far. I think the suggestion was just that GDB itself should > require libexpat, which would mean we have the potential to read XML > target descriptions. Sorry, I might've read too much into it. But making libexpat a requirement for gdb would be a good move in my opinion, nonetheless. Cross-building gdb sometimes yields a gdb with no libexpat and that is not very useful for AArch64/Arm debugging for instance. > > This wouldn't mean that targets would be required to send XML target > descriptions.> > My comment was specifically that x86 gdbserver should drop its no XML > support, and if a user tries to connect with a no XML GDB then gdbserver > would reject the connection. Understood. > > Moving to make libexpat a requirement would help in this case as a user > would no longer be able to build a GDB which couldn't connect to an x86 > gdbserver. > >> Off the top of my head, I think very old probes and the Linux Kernel's kgdb don't >> support xml. I'm not sure how often kgdb gets used though, but falling back to a >> default description should work for AArch64 and 32-bit Arm. > > I don't believe building GDB with libexpat support breaks the ability to > connect to a target which doesn't support XML descriptions (if it does > then that's a bug), so hopefully nobody else would be > inconvenienced... except for those folk who are building GDB without > libexpat support. Makes sense. Thanks for clarifying. > > I've added making these changes to my todo list, but I don't want to add > them to this series. > > Thanks, > Andrew > ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 05/11] gdbserver/x86: move no-xml code earlier in x86_linux_read_description 2024-05-07 15:43 ` Andrew Burgess 2024-05-07 15:56 ` Luis Machado @ 2024-05-08 7:49 ` Willgerodt, Felix 2024-05-08 13:18 ` Andrew Burgess 1 sibling, 1 reply; 116+ messages in thread From: Willgerodt, Felix @ 2024-05-08 7:49 UTC (permalink / raw) To: Andrew Burgess, Luis Machado, gdb-patches; +Cc: John Baldwin > -----Original Message----- > From: Andrew Burgess <aburgess@redhat.com> > Sent: Dienstag, 7. Mai 2024 17:44 > To: Luis Machado <luis.machado@arm.com>; Willgerodt, Felix > <felix.willgerodt@intel.com>; gdb-patches@sourceware.org > Cc: John Baldwin <jhb@FreeBSD.org> > Subject: Re: [PATCHv5 05/11] gdbserver/x86: move no-xml code earlier in > x86_linux_read_description > > Luis Machado <luis.machado@arm.com> writes: > > > On 4/29/24 15:34, Willgerodt, Felix wrote: > >>> -----Original Message----- > >>> From: Andrew Burgess <aburgess@redhat.com> > >>> Sent: Freitag, 26. April 2024 17:02 > >>> To: gdb-patches@sourceware.org > >>> Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix > >>> <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> > >>> Subject: [PATCHv5 05/11] gdbserver/x86: move no-xml code earlier in > >>> x86_linux_read_description > >>> > >>> This commit is part of a series that aims to share more of the x86 > >>> target description reading/generation code between GDB and gdbserver. > >>> > >>> There are a huge number of similarities between the code in > >>> gdbserver's x86_linux_read_description function and GDB's > >>> x86_linux_nat_target::read_description function, and it is this > >>> similarity that I plan, in a later commit, to share between GDB and > >>> gdbserver. > >>> > >>> However, one thing that is different in x86_linux_read_description is > >>> the code inside the '!use_xml' block. This is the code that handles > >>> the case where gdbserver is not allowed to send an XML target > >>> description back to GDB. In this case gdbserver uses some predefined, > >>> fixed, target descriptions. > >>> > >>> First, it's worth noting that I suspect this code is not tested any > >>> more. I couldn't find anything in the testsuite that tries to disable > >>> XML target description support. And the idea of having a single > >>> "fixed" target description really doesn't work well when we think > >>> about all the various x86 extensions that exist. Part of me would > >>> like to rip out the no-xml support in gdbserver (at least for x86), > >>> and if a GDB connects that doesn't support XML target descriptions, > >>> gdbserver can just give an error and drop the connection. GDB has > >>> supported XML target descriptions for 16 years now, I think it would > >>> be reasonable for our shipped gdbserver to drop support for the old > >>> way of doing things. > >> > >> Interesting. I for one would +1 this. Slightly related: > >> I wonder if anyone really builds GDB without libexpat anymore and > >> if we couldn't even think of making it mandatory. (Which doesn't > >> mean dropping the ball on supporting stubs without XML.) > > > > I'd support making xml descriptions mandatory for some targets. AArch64 and > > 32-bit Arm both rely heavily on xml description for feature discovery. > > I might be wrong, but I don't think Felix was even suggesting we need to > go that far. You are correct. > I think the suggestion was just that GDB itself should > require libexpat, which would mean we have the potential to read XML > target descriptions. > > This wouldn't mean that targets would be required to send XML target > descriptions. > > My comment was specifically that x86 gdbserver should drop its no XML > support, and if a user tries to connect with a no XML GDB then gdbserver > would reject the connection. > > Moving to make libexpat a requirement would help in this case as a user > would no longer be able to build a GDB which couldn't connect to an x86 > gdbserver. > > > Off the top of my head, I think very old probes and the Linux Kernel's kgdb don't > > support xml. I'm not sure how often kgdb gets used though, but falling back to a > > default description should work for AArch64 and 32-bit Arm. > > I don't believe building GDB with libexpat support breaks the ability to > connect to a target which doesn't support XML descriptions (if it does > then that's a bug), so hopefully nobody else would be > inconvenienced... except for those folk who are building GDB without > libexpat support. One immediate concern in my mind would be testability for no-xml stubs (without having looked at the current testing situation). Anyway, this is not really a topic for this series, just an interesting side discussion. Felix Intel Deutschland GmbH Registered Address: Am Campeon 10, 85579 Neubiberg, Germany Tel: +49 89 99 8853-0, www.intel.de Managing Directors: Sean Fennelly, Jeffrey Schneiderman, Tiffany Doon Silva Chairperson of the Supervisory Board: Nicole Lau Registered Office: Munich Commercial Register: Amtsgericht Muenchen HRB 186928 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 05/11] gdbserver/x86: move no-xml code earlier in x86_linux_read_description 2024-05-08 7:49 ` Willgerodt, Felix @ 2024-05-08 13:18 ` Andrew Burgess 0 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-08 13:18 UTC (permalink / raw) To: Willgerodt, Felix, Luis Machado, gdb-patches; +Cc: John Baldwin "Willgerodt, Felix" <felix.willgerodt@intel.com> writes: >> -----Original Message----- >> From: Andrew Burgess <aburgess@redhat.com> >> Sent: Dienstag, 7. Mai 2024 17:44 >> To: Luis Machado <luis.machado@arm.com>; Willgerodt, Felix >> <felix.willgerodt@intel.com>; gdb-patches@sourceware.org >> Cc: John Baldwin <jhb@FreeBSD.org> >> Subject: Re: [PATCHv5 05/11] gdbserver/x86: move no-xml code earlier in >> x86_linux_read_description >> >> Luis Machado <luis.machado@arm.com> writes: >> >> > On 4/29/24 15:34, Willgerodt, Felix wrote: >> >>> -----Original Message----- >> >>> From: Andrew Burgess <aburgess@redhat.com> >> >>> Sent: Freitag, 26. April 2024 17:02 >> >>> To: gdb-patches@sourceware.org >> >>> Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix >> >>> <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> >> >>> Subject: [PATCHv5 05/11] gdbserver/x86: move no-xml code earlier in >> >>> x86_linux_read_description >> >>> >> >>> This commit is part of a series that aims to share more of the x86 >> >>> target description reading/generation code between GDB and gdbserver. >> >>> >> >>> There are a huge number of similarities between the code in >> >>> gdbserver's x86_linux_read_description function and GDB's >> >>> x86_linux_nat_target::read_description function, and it is this >> >>> similarity that I plan, in a later commit, to share between GDB and >> >>> gdbserver. >> >>> >> >>> However, one thing that is different in x86_linux_read_description is >> >>> the code inside the '!use_xml' block. This is the code that handles >> >>> the case where gdbserver is not allowed to send an XML target >> >>> description back to GDB. In this case gdbserver uses some predefined, >> >>> fixed, target descriptions. >> >>> >> >>> First, it's worth noting that I suspect this code is not tested any >> >>> more. I couldn't find anything in the testsuite that tries to disable >> >>> XML target description support. And the idea of having a single >> >>> "fixed" target description really doesn't work well when we think >> >>> about all the various x86 extensions that exist. Part of me would >> >>> like to rip out the no-xml support in gdbserver (at least for x86), >> >>> and if a GDB connects that doesn't support XML target descriptions, >> >>> gdbserver can just give an error and drop the connection. GDB has >> >>> supported XML target descriptions for 16 years now, I think it would >> >>> be reasonable for our shipped gdbserver to drop support for the old >> >>> way of doing things. >> >> >> >> Interesting. I for one would +1 this. Slightly related: >> >> I wonder if anyone really builds GDB without libexpat anymore and >> >> if we couldn't even think of making it mandatory. (Which doesn't >> >> mean dropping the ball on supporting stubs without XML.) >> > >> > I'd support making xml descriptions mandatory for some targets. AArch64 and >> > 32-bit Arm both rely heavily on xml description for feature discovery. >> >> I might be wrong, but I don't think Felix was even suggesting we need to >> go that far. > > You are correct. > >> I think the suggestion was just that GDB itself should >> require libexpat, which would mean we have the potential to read XML >> target descriptions. >> >> This wouldn't mean that targets would be required to send XML target >> descriptions. >> >> My comment was specifically that x86 gdbserver should drop its no XML >> support, and if a user tries to connect with a no XML GDB then gdbserver >> would reject the connection. >> >> Moving to make libexpat a requirement would help in this case as a user >> would no longer be able to build a GDB which couldn't connect to an x86 >> gdbserver. >> >> > Off the top of my head, I think very old probes and the Linux Kernel's kgdb don't >> > support xml. I'm not sure how often kgdb gets used though, but falling back to a >> > default description should work for AArch64 and 32-bit Arm. >> >> I don't believe building GDB with libexpat support breaks the ability to >> connect to a target which doesn't support XML descriptions (if it does >> then that's a bug), so hopefully nobody else would be >> inconvenienced... except for those folk who are building GDB without >> libexpat support. > > One immediate concern in my mind would be testability for no-xml stubs > (without having looked at the current testing situation). Anyway, this is not > really a topic for this series, just an interesting side discussion. This don't think there would be a problem. Or, if there is, then that would be a legitimate bug that needs fixing. Right now I would not expect a user to have to build a no-xml GDB in order to talk to a no-xml target. I'd expect the user to fire up there standard XML supporting GDB, connect to the target, and have GDB "just work". If there are architectures supported by gdbserver that don't support sending back an XML description (I haven't checked) then they should work just as they do today, even if GDB supports reading XML. Thanks, Andrew ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv5 06/11] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition 2024-04-26 15:01 ` [PATCHv5 00/11] " Andrew Burgess ` (4 preceding siblings ...) 2024-04-26 15:01 ` [PATCHv5 05/11] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess @ 2024-04-26 15:01 ` Andrew Burgess 2024-04-29 14:34 ` Willgerodt, Felix 2024-04-26 15:01 ` [PATCHv5 07/11] gdb/gdbserver: share some code relating to target description creation Andrew Burgess ` (5 subsequent siblings) 11 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-04-26 15:01 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, John Baldwin Share the definition of I386_LINUX_XSAVE_XCR0_OFFSET between GDB and gdbserver. This commit is part of a series that aims to share more of the x86 target description creation code between GDB and gdbserver. The I386_LINUX_XSAVE_XCR0_OFFSET #define is used as part of the target description creation, and I noticed that this constant is defined separately for GDB and gdbserver. This commit moves the definition into gdbsupport/x86-xstate.h, which allows the #define to be shared. There should be no user visible changes after this commit. --- gdb/i386-linux-tdep.h | 20 -------------------- gdbserver/linux-x86-low.cc | 21 --------------------- gdbsupport/x86-xstate.h | 20 ++++++++++++++++++++ 3 files changed, 20 insertions(+), 41 deletions(-) diff --git a/gdb/i386-linux-tdep.h b/gdb/i386-linux-tdep.h index 5891747572b..07593c6a8ec 100644 --- a/gdb/i386-linux-tdep.h +++ b/gdb/i386-linux-tdep.h @@ -58,26 +58,6 @@ extern void i386_linux_report_signal_info (struct gdbarch *gdbarch, /* Return the target description according to XCR0. */ extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); -/* Format of XSAVE extended state is: - struct - { - fxsave_bytes[0..463] - sw_usable_bytes[464..511] - xstate_hdr_bytes[512..575] - extended state regions (AVX, MPX, AVX512, PKRU, etc.) - }; - - Same memory layout will be used for the coredump NT_X86_XSTATE - representing the XSAVE extended state registers. - - The first 8 bytes of the sw_usable_bytes[464..467] is the OS enabled - extended state mask, which is the same as the extended control register - 0 (the XFEATURE_ENABLED_MASK register), XCR0. We can use this mask - together with the mask saved in the xstate_hdr_bytes to determine what - states the processor/OS supports and what state, used or initialized, - the process/thread is in. */ -#define I386_LINUX_XSAVE_XCR0_OFFSET 464 - extern int i386_linux_gregset_reg_offset[]; /* Return x86 siginfo type. */ diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index ffb2a90731d..62612773a87 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -831,27 +831,6 @@ x86_target::low_siginfo_fixup (siginfo_t *ptrace, gdb_byte *inf, int direction) \f static int use_xml; -/* Format of XSAVE extended state is: - struct - { - fxsave_bytes[0..463] - sw_usable_bytes[464..511] - xstate_hdr_bytes[512..575] - avx_bytes[576..831] - future_state etc - }; - - Same memory layout will be used for the coredump NT_X86_XSTATE - representing the XSAVE extended state registers. - - The first 8 bytes of the sw_usable_bytes[464..467] is the OS enabled - extended state mask, which is the same as the extended control register - 0 (the XFEATURE_ENABLED_MASK register), XCR0. We can use this mask - together with the mask saved in the xstate_hdr_bytes to determine what - states the processor/OS supports and what state, used or initialized, - the process/thread is in. */ -#define I386_LINUX_XSAVE_XCR0_OFFSET 464 - /* Get Linux/x86 target description from running target. */ static const struct target_desc * diff --git a/gdbsupport/x86-xstate.h b/gdbsupport/x86-xstate.h index 89c1143fbe1..11b37544aa3 100644 --- a/gdbsupport/x86-xstate.h +++ b/gdbsupport/x86-xstate.h @@ -120,4 +120,24 @@ constexpr bool operator!= (const x86_xsave_layout &lhs, #define I387_MXCSR_INIT_VAL 0x1f80 +/* Format of XSAVE extended state is: + struct + { + fxsave_bytes[0..463] + sw_usable_bytes[464..511] + xstate_hdr_bytes[512..575] + extended state regions (AVX, MPX, AVX512, PKRU, etc.) + }; + + Same memory layout will be used for the coredump NT_X86_XSTATE + representing the XSAVE extended state registers. + + The first 8 bytes of the sw_usable_bytes[464..467] is the OS enabled + extended state mask, which is the same as the extended control register + 0 (the XFEATURE_ENABLED_MASK register), XCR0. We can use this mask + together with the mask saved in the xstate_hdr_bytes to determine what + states the processor/OS supports and what state, used or initialized, + the process/thread is in. */ +#define I386_LINUX_XSAVE_XCR0_OFFSET 464 + #endif /* COMMON_X86_XSTATE_H */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 06/11] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition 2024-04-26 15:01 ` [PATCHv5 06/11] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess @ 2024-04-29 14:34 ` Willgerodt, Felix 0 siblings, 0 replies; 116+ messages in thread From: Willgerodt, Felix @ 2024-04-29 14:34 UTC (permalink / raw) To: Andrew Burgess, gdb-patches; +Cc: John Baldwin > -----Original Message----- > From: Andrew Burgess <aburgess@redhat.com> > Sent: Freitag, 26. April 2024 17:02 > To: gdb-patches@sourceware.org > Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix > <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> > Subject: [PATCHv5 06/11] gdb/gdbserver: share > I386_LINUX_XSAVE_XCR0_OFFSET definition > > Share the definition of I386_LINUX_XSAVE_XCR0_OFFSET between GDB and > gdbserver. > > This commit is part of a series that aims to share more of the x86 > target description creation code between GDB and gdbserver. The > I386_LINUX_XSAVE_XCR0_OFFSET #define is used as part of the target > description creation, and I noticed that this constant is defined > separately for GDB and gdbserver. > > This commit moves the definition into gdbsupport/x86-xstate.h, which > allows the #define to be shared. > > There should be no user visible changes after this commit. > --- > gdb/i386-linux-tdep.h | 20 -------------------- > gdbserver/linux-x86-low.cc | 21 --------------------- > gdbsupport/x86-xstate.h | 20 ++++++++++++++++++++ > 3 files changed, 20 insertions(+), 41 deletions(-) > Approved-By: Felix Willgerodt <felix.willgerodt@intel.com> Thanks, Felix Intel Deutschland GmbH Registered Address: Am Campeon 10, 85579 Neubiberg, Germany Tel: +49 89 99 8853-0, www.intel.de <http://www.intel.de> Managing Directors: Christin Eisenschmid, Sharon Heck, Tiffany Doon Silva Chairperson of the Supervisory Board: Nicole Lau Registered Office: Munich Commercial Register: Amtsgericht Muenchen HRB 186928 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv5 07/11] gdb/gdbserver: share some code relating to target description creation 2024-04-26 15:01 ` [PATCHv5 00/11] " Andrew Burgess ` (5 preceding siblings ...) 2024-04-26 15:01 ` [PATCHv5 06/11] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess @ 2024-04-26 15:01 ` Andrew Burgess 2024-04-29 14:34 ` Willgerodt, Felix 2024-04-26 15:01 ` [PATCHv5 08/11] gdb/arch: assert that X86_XSTATE_MPX is not set for x32 Andrew Burgess ` (4 subsequent siblings) 11 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-04-26 15:01 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, John Baldwin This commit is part of a series to share more of the x86 target description creation code between GDB and gdbserver. Unlike previous commits which were mostly refactoring, this commit is the first that makes a real change, though that change should mostly be for gdbserver; I've largely adopted the "GDB" way of doing things for gdbserver, and this fixes a real gdbserver bug. On a x86-64 Linux target, running the test: gdb.server/connect-with-no-symbol-file.exp results in two core files being created. Both of these core files are from the inferior process, created after gdbserver has detached. In this test a gdbserver process is started and then, after gdbserver has started, but before GDB attaches, we either delete the inferior executable, or change its permissions so it can't be read. Only after doing this do we attempt to connect with GDB. As GDB connects to gdbserver, gdbserver attempts to figure out the target description so that it can send the description to GDB, this involves a call to x86_linux_read_description. In x86_linux_read_description one of the first things we do is try to figure out if the process is 32-bit or 64-bit. To do this we look up the executable via the thread-id, and then attempt to read the architecture size from the executable. This isn't going to work if the executable has been deleted, or is no longer readable. And so, as we can't read the executable, we default to an i386 target and use an i386 target description. A consequence of using an i386 target description is that addresses are assumed to be 32-bits. Here's an example session that shows the problems this causes. This is run on an x86-64 machine, and the test binary (xx.x) is a standard 64-bit x86-64 binary: shell_1$ gdbserver --once localhost :54321 /tmp/xx.x shell_2$ gdb -q (gdb) set sysroot (gdb) shell chmod 000 /tmp/xx.x (gdb) target remote :54321 Remote debugging using :54321 warning: /tmp/xx.x: Permission denied. 0xf7fd3110 in ?? () (gdb) show architecture The target architecture is set to "auto" (currently "i386"). (gdb) p/x $pc $1 = 0xf7fd3110 (gdb) info proc mappings process 2412639 Mapped address spaces: Start Addr End Addr Size Offset Perms objfile 0x400000 0x401000 0x1000 0x0 r--p /tmp/xx.x 0x401000 0x402000 0x1000 0x1000 r-xp /tmp/xx.x 0x402000 0x403000 0x1000 0x2000 r--p /tmp/xx.x 0x403000 0x405000 0x2000 0x2000 rw-p /tmp/xx.x 0xf7fcb000 0xf7fcf000 0x4000 0x0 r--p [vvar] 0xf7fcf000 0xf7fd1000 0x2000 0x0 r-xp [vdso] 0xf7fd1000 0xf7fd3000 0x2000 0x0 r--p /usr/lib64/ld-2.30.so 0xf7fd3000 0xf7ff3000 0x20000 0x2000 r-xp /usr/lib64/ld-2.30.so 0xf7ff3000 0xf7ffb000 0x8000 0x22000 r--p /usr/lib64/ld-2.30.so 0xf7ffc000 0xf7ffe000 0x2000 0x2a000 rw-p /usr/lib64/ld-2.30.so 0xf7ffe000 0xf7fff000 0x1000 0x0 rw-p 0xfffda000 0xfffff000 0x25000 0x0 rw-p [stack] 0xff600000 0xff601000 0x1000 0x0 r-xp [vsyscall] (gdb) info inferiors Num Description Connection Executable * 1 process 2412639 1 (remote :54321) (gdb) shell cat /proc/2412639/maps 00400000-00401000 r--p 00000000 fd:03 45907133 /tmp/xx.x 00401000-00402000 r-xp 00001000 fd:03 45907133 /tmp/xx.x 00402000-00403000 r--p 00002000 fd:03 45907133 /tmp/xx.x 00403000-00405000 rw-p 00002000 fd:03 45907133 /tmp/xx.x 7ffff7fcb000-7ffff7fcf000 r--p 00000000 00:00 0 [vvar] 7ffff7fcf000-7ffff7fd1000 r-xp 00000000 00:00 0 [vdso] 7ffff7fd1000-7ffff7fd3000 r--p 00000000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7fd3000-7ffff7ff3000 r-xp 00002000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7ff3000-7ffff7ffb000 r--p 00022000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7ffc000-7ffff7ffe000 rw-p 0002a000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7ffe000-7ffff7fff000 rw-p 00000000 00:00 0 7ffffffda000-7ffffffff000 rw-p 00000000 00:00 0 [stack] ffffffffff600000-ffffffffff601000 r-xp 00000000 00:00 0 [vsyscall] (gdb) Notice the difference between the mappings reported via GDB and those reported directly from the kernel via /proc/PID/maps, the addresses of every mapping is clamped to 32-bits for GDB, while the kernel reports real 64-bit addresses. Notice also that the $pc value is a 32-bit value. It appears to be within one of the mappings reported by GDB, but is outside any of the mappings reported from the kernel. And this is where the problem arises. When gdbserver detaches from the inferior we pass the inferior the address from which it should resume. Due to the 32/64 bit confusion we tell the inferior to resume from the 32-bit $pc value, which is not within any valid mapping, and so, as soon as the inferior resumes, it segfaults. If we look at how GDB (not gdbserver) figures out its target description then we see an interesting difference. GDB doesn't try to read the executable. Instead GDB uses ptrace to query the thread's state, and uses this to figure out the if the thread is 32 or 64 bit. If we update gdbserver to do it the "GDB" way then the above problem is resolved, gdbserver now sees the process as 64-bit, and when we detach from the inferior we give it the correct 64-bit address, and the inferior no longer segfaults. Now, I could just update the gdbserver code, but better, I think, to share one copy of the code between GDB and gdbserver in gdb/nat/. That is what this commit does. The cores of x86_linux_read_description from gdbserver and x86_linux_nat_target::read_description from GDB are moved into a new file gdb/nat/x86-linux-tdesc.c and combined into a single function x86_linux_tdesc_for_tid which is called from each location. This new function does things the GDB way, the only changes are to allow for the sharing; we now have a callback function to call the first time that the xcr0 state is read, this allows for GDB and gdbserver to perform their own initialisation as needed, and additionally, the new function takes a pointer for where to cache the xcr0 value, this isn't needed for this commit, but will be useful in a later commit where gdbserver will want to read this cached xcr0 value. Another thing to note about this commit is how the functions i386_linux_read_description and amd64_linux_read_description are handled. For now I've left these function as implemented separately in GDB and gdbserver. I've moved the declarations of these functions into gdb/arch/{i386,amd64}-linux-tdesc.h, but the implementations are left where they are. A later commit in this series will make these functions shared too, but doing this is not trivial, so I've left that for a separate commit. Merging the declarations as I've done here ensures that everyone implements the function to the same API, and once these functions are shared (in a later commit) we'll want a shared declaration anyway. --- gdb/Makefile.in | 3 + gdb/amd64-linux-tdep.c | 1 + gdb/amd64-linux-tdep.h | 6 -- gdb/arch/amd64-linux-tdesc.h | 30 +++++++ gdb/arch/i386-linux-tdesc.h | 29 +++++++ gdb/configure.nat | 4 +- gdb/i386-linux-tdep.c | 1 + gdb/i386-linux-tdep.h | 3 - gdb/nat/x86-linux-tdesc.c | 124 ++++++++++++++++++++++++++++ gdb/nat/x86-linux-tdesc.h | 60 ++++++++++++++ gdb/x86-linux-nat.c | 91 ++++----------------- gdbserver/configure.srv | 2 + gdbserver/linux-amd64-ipa.cc | 1 + gdbserver/linux-i386-ipa.cc | 1 + gdbserver/linux-x86-low.cc | 151 +++++++++++------------------------ gdbserver/linux-x86-tdesc.cc | 2 + gdbserver/linux-x86-tdesc.h | 7 -- 17 files changed, 317 insertions(+), 199 deletions(-) create mode 100644 gdb/arch/amd64-linux-tdesc.h create mode 100644 gdb/arch/i386-linux-tdesc.h create mode 100644 gdb/nat/x86-linux-tdesc.c create mode 100644 gdb/nat/x86-linux-tdesc.h diff --git a/gdb/Makefile.in b/gdb/Makefile.in index 353e45b3cec..a24b2232daa 100644 --- a/gdb/Makefile.in +++ b/gdb/Makefile.in @@ -1549,8 +1549,10 @@ HFILES_NO_SRCDIR = \ arch/aarch64-insn.h \ arch/aarch64-mte-linux.h \ arch/aarch64-scalable-linux.h \ + arch/amd64-linux-tdesc.h \ arch/arc.h \ arch/arm.h \ + arch/i386-linux-tdesc.h \ arch/i386.h \ arch/loongarch.h \ arch/ppc-linux-common.h \ @@ -1606,6 +1608,7 @@ HFILES_NO_SRCDIR = \ nat/x86-gcc-cpuid.h \ nat/x86-linux.h \ nat/x86-linux-dregs.h \ + nat/x86-linux-tdesc.h \ python/py-event.h \ python/py-events.h \ python/py-stopevent.h \ diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c index 9d560ac4fbf..bcb9868e79e 100644 --- a/gdb/amd64-linux-tdep.c +++ b/gdb/amd64-linux-tdep.c @@ -41,6 +41,7 @@ #include "arch/amd64.h" #include "target-descriptions.h" #include "expop.h" +#include "arch/amd64-linux-tdesc.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_AMD64 "syscalls/amd64-linux.xml" diff --git a/gdb/amd64-linux-tdep.h b/gdb/amd64-linux-tdep.h index 2003dcda78f..0ec49e7fe03 100644 --- a/gdb/amd64-linux-tdep.h +++ b/gdb/amd64-linux-tdep.h @@ -43,12 +43,6 @@ extern struct target_desc *tdesc_x32_linux; extern struct target_desc *tdesc_x32_avx_linux; extern struct target_desc *tdesc_x32_avx_avx512_linux; -/* Return the right amd64-linux target descriptions according to - XCR0_FEATURES_BIT and IS_X32. */ - -const target_desc *amd64_linux_read_description (uint64_t xcr0_features_bit, - bool is_x32); - /* Enum that defines the syscall identifiers for amd64 linux. Used for process record/replay, these will be translated into a gdb-canonical set of syscall ids in linux-record.c. */ diff --git a/gdb/arch/amd64-linux-tdesc.h b/gdb/arch/amd64-linux-tdesc.h new file mode 100644 index 00000000000..db425b60df6 --- /dev/null +++ b/gdb/arch/amd64-linux-tdesc.h @@ -0,0 +1,30 @@ +/* Target description related code for GNU/Linux x86-64. + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#ifndef ARCH_AMD64_LINUX_TDESC_H +#define ARCH_AMD64_LINUX_TDESC_H + +struct target_desc; + +/* Return the AMD64 target descriptions corresponding to XCR0 and IS_X32. */ + +extern const target_desc *amd64_linux_read_description (uint64_t xcr0, + bool is_x32); + +#endif /* ARCH_AMD64_LINUX_TDESC_H */ diff --git a/gdb/arch/i386-linux-tdesc.h b/gdb/arch/i386-linux-tdesc.h new file mode 100644 index 00000000000..0b736337a75 --- /dev/null +++ b/gdb/arch/i386-linux-tdesc.h @@ -0,0 +1,29 @@ +/* Target description related code for GNU/Linux i386. + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#ifndef ARCH_I386_LINUX_TDESC_H +#define ARCH_I386_LINUX_TDESC_H + +struct target_desc; + +/* Return the i386 target description corresponding to XCR0. */ + +extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); + +#endif /* ARCH_I386_LINUX_TDESC_H */ diff --git a/gdb/configure.nat b/gdb/configure.nat index 8b98511cef7..4bcc0696027 100644 --- a/gdb/configure.nat +++ b/gdb/configure.nat @@ -256,7 +256,7 @@ case ${gdb_host} in NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ nat/x86-xstate.o \ i386-linux-nat.o x86-linux-nat.o nat/linux-btrace.o \ - nat/x86-linux.o nat/x86-linux-dregs.o" + nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o" ;; ia64) # Host: Intel IA-64 running GNU/Linux @@ -322,7 +322,7 @@ case ${gdb_host} in NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ nat/x86-xstate.o amd64-nat.o amd64-linux-nat.o x86-linux-nat.o \ nat/linux-btrace.o \ - nat/x86-linux.o nat/x86-linux-dregs.o \ + nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o \ nat/amd64-linux-siginfo.o" ;; sparc) diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index 44730f204db..78ebc99d3df 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -39,6 +39,7 @@ #include "i387-tdep.h" #include "gdbsupport/x86-xstate.h" +#include "arch/i386-linux-tdesc.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_I386 "syscalls/i386-linux.xml" diff --git a/gdb/i386-linux-tdep.h b/gdb/i386-linux-tdep.h index 07593c6a8ec..e8691cd778e 100644 --- a/gdb/i386-linux-tdep.h +++ b/gdb/i386-linux-tdep.h @@ -55,9 +55,6 @@ extern void i386_linux_report_signal_info (struct gdbarch *gdbarch, struct ui_out *uiout, enum gdb_signal siggnal); -/* Return the target description according to XCR0. */ -extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); - extern int i386_linux_gregset_reg_offset[]; /* Return x86 siginfo type. */ diff --git a/gdb/nat/x86-linux-tdesc.c b/gdb/nat/x86-linux-tdesc.c new file mode 100644 index 00000000000..c2301e6a873 --- /dev/null +++ b/gdb/nat/x86-linux-tdesc.c @@ -0,0 +1,124 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#include "nat/x86-linux-tdesc.h" +#ifdef __x86_64__ +#include "arch/amd64.h" +#include "arch/amd64-linux-tdesc.h" +#endif +#include "arch/i386.h" +#include "arch/i386-linux-tdesc.h" + +#include "nat/x86-linux.h" +#include "nat/gdb_ptrace.h" +#include "nat/x86-xstate.h" + +#ifndef __x86_64__ +#include <sys/procfs.h> +#endif + +#include <sys/uio.h> +#include <elf.h> + +#ifndef IN_PROCESS_AGENT + +/* See nat/x86-linux-tdesc.h. */ + +const target_desc * +x86_linux_tdesc_for_tid (int tid, enum tribool *have_ptrace_getregset, + gdb::function_view<void (uint64_t)> xcr0_init_cb, + const char *error_msg, uint64_t *xcr0_storage) +{ +#ifdef __x86_64__ + + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); + bool is_64bit = arch_size.is_64bit (); + bool is_x32 = arch_size.is_x32 (); + + if (sizeof (void *) == 4 && is_64bit && !is_x32) + error ("%s", error_msg); + +#elif HAVE_PTRACE_GETFPXREGS + if (have_ptrace_getfpxregs == -1) + { + elf_fpxregset_t fpxregs; + + if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) + { + have_ptrace_getfpxregs = 0; + *have_ptrace_getregset = TRIBOOL_FALSE; + return i386_linux_read_description (X86_XSTATE_X87_MASK); + } + } +#endif + + if (*have_ptrace_getregset == TRIBOOL_UNKNOWN) + { + uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; + struct iovec iov; + + iov.iov_base = xstateregs; + iov.iov_len = sizeof (xstateregs); + + /* Check if PTRACE_GETREGSET works. */ + if (ptrace (PTRACE_GETREGSET, tid, + (unsigned int) NT_X86_XSTATE, &iov) < 0) + { + *have_ptrace_getregset = TRIBOOL_FALSE; + *xcr0_storage = 0; + } + else + { + *have_ptrace_getregset = TRIBOOL_TRUE; + + /* Get XCR0 from XSAVE extended state. */ + *xcr0_storage = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET + / sizeof (uint64_t))]; + +#ifdef __x86_64__ + /* No MPX on x32. */ + if (is_64bit && is_x32) + *xcr0_storage &= ~X86_XSTATE_MPX; +#endif /* __x86_64__ */ + + xcr0_init_cb (*xcr0_storage); + } + } + + /* Check the native XCR0 only if PTRACE_GETREGSET is available. If + PTRACE_GETREGSET is not available then set xcr0_features_bits to + zero so that the "no-features" descriptions are returned by the + switches below. */ + uint64_t xcr0_features_bits; + if (*have_ptrace_getregset == TRIBOOL_TRUE) + xcr0_features_bits = *xcr0_storage & X86_XSTATE_ALL_MASK; + else + xcr0_features_bits = 0; + +#ifdef __x86_64__ + if (is_64bit) + { + return amd64_linux_read_description (xcr0_features_bits, is_x32); + } + else +#endif + return i386_linux_read_description (xcr0_features_bits); +} + +#endif /* !IN_PROCESS_AGENT */ diff --git a/gdb/nat/x86-linux-tdesc.h b/gdb/nat/x86-linux-tdesc.h new file mode 100644 index 00000000000..8c481609876 --- /dev/null +++ b/gdb/nat/x86-linux-tdesc.h @@ -0,0 +1,60 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#ifndef NAT_X86_LINUX_TDESC_H +#define NAT_X86_LINUX_TDESC_H + +#include "gdbsupport/function-view.h" + +struct target_desc; + +/* Return the target description for Linux thread TID. + + When *HAVE_PTRACE_GETREGSET is TRIBOOL_UNKNOWN then the current value of + xcr0 is read using ptrace calls and stored into *XCR0_STORAGE. Then + XCR0_INIT_CB is called with the value of *XCR0_STORAGE and + *HAVE_PTRACE_GETREGSET is set to TRIBOOL_TRUE. + + If the attempt to read xcr0 using ptrace fails then *XCR0_STORAGE is set + to zero and *HAVE_PTRACE_GETREGSET is set to TRIBOOL_FALSE. + + The storage pointed to by XCR0_STORAGE must exist until the program + terminates, this storage is used to cache the xcr0 value. As such + XCR0_INIT_CB will only be called once if xcr0 is successfully read using + ptrace, or not at all if the ptrace call fails. + + This function returns a target description based on the extracted xcr0 + value along with other characteristics of the thread identified by TID. + + This function can return nullptr if we encounter a machine configuration + for which a target_desc cannot be created. Ideally this would not be + the case, we should be able to create a target description for every + possible machine configuration. See amd64_linux_read_description and + i386_linux_read_description for cases when nullptr might be + returned. + + ERROR_MSG is using in an error() call if we try to create a target + description for a 64-bit process but this is a 32-bit build of GDB. */ + +extern const target_desc * +x86_linux_tdesc_for_tid (int tid, enum tribool *have_ptrace_getregset, + gdb::function_view<void (uint64_t)> xcr0_init_cb, + const char *error_msg, uint64_t *xcr0_storage); + +#endif /* NAT_X86_LINUX_TDESC_H */ diff --git a/gdb/x86-linux-nat.c b/gdb/x86-linux-nat.c index 12965c7a21b..9fbccb28d21 100644 --- a/gdb/x86-linux-nat.c +++ b/gdb/x86-linux-nat.c @@ -38,6 +38,7 @@ #include "nat/x86-linux.h" #include "nat/x86-linux-dregs.h" #include "nat/linux-ptrace.h" +#include "nat/x86-linux-tdesc.h" /* linux_nat_target::low_new_fork implementation. */ @@ -92,90 +93,26 @@ x86_linux_nat_target::post_startup_inferior (ptid_t ptid) const struct target_desc * x86_linux_nat_target::read_description () { - int tid; - int is_64bit = 0; -#ifdef __x86_64__ - int is_x32; -#endif - static uint64_t xcr0; - uint64_t xcr0_features_bits; + static uint64_t xcr0_storage; if (inferior_ptid == null_ptid) return this->beneath ()->read_description (); - tid = inferior_ptid.pid (); - -#ifdef __x86_64__ - - x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); - is_64bit = arch_size.is_64bit (); - is_x32 = arch_size.is_x32 (); - -#elif HAVE_PTRACE_GETFPXREGS - if (have_ptrace_getfpxregs == -1) - { - elf_fpxregset_t fpxregs; - - if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) - { - have_ptrace_getfpxregs = 0; - have_ptrace_getregset = TRIBOOL_FALSE; - return i386_linux_read_description (X86_XSTATE_X87_MASK); - } - } -#endif - - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) - { - uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; - struct iovec iov; - - iov.iov_base = xstateregs; - iov.iov_len = sizeof (xstateregs); - - /* Check if PTRACE_GETREGSET works. */ - if (ptrace (PTRACE_GETREGSET, tid, - (unsigned int) NT_X86_XSTATE, &iov) < 0) - have_ptrace_getregset = TRIBOOL_FALSE; - else - { - have_ptrace_getregset = TRIBOOL_TRUE; - - /* Get XCR0 from XSAVE extended state. */ - xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET - / sizeof (uint64_t))]; - - m_xsave_layout = x86_fetch_xsave_layout (xcr0, x86_xsave_length ()); - } - } - - /* Check the native XCR0 only if PTRACE_GETREGSET is available. If - PTRACE_GETREGSET is not available then set xcr0_features_bits to - zero so that the "no-features" descriptions are returned by the - switches below. */ - if (have_ptrace_getregset == TRIBOOL_TRUE) - xcr0_features_bits = xcr0 & X86_XSTATE_ALL_MASK; - else - xcr0_features_bits = 0; - - if (is_64bit) - { -#ifdef __x86_64__ - return amd64_linux_read_description (xcr0_features_bits, is_x32); -#endif - } - else - { - const struct target_desc * tdesc - = i386_linux_read_description (xcr0_features_bits); + int tid = inferior_ptid.pid (); - if (tdesc == NULL) - tdesc = i386_linux_read_description (X86_XSTATE_SSE_MASK); + const char *error_msg + = _("Can't debug 64-bit process with 32-bit GDB"); - return tdesc; - } + /* Callback that is triggered the first time x86_linux_tdesc_for_tid + reads the xcr0 register. Setup other bits of state. */ + auto cb = [&] (uint64_t xcr0) + { + this->m_xsave_layout + = x86_fetch_xsave_layout (xcr0, x86_xsave_length ()); + }; - gdb_assert_not_reached ("failed to return tdesc"); + return x86_linux_tdesc_for_tid (tid, &have_ptrace_getregset, cb, + error_msg, &xcr0_storage); } \f diff --git a/gdbserver/configure.srv b/gdbserver/configure.srv index 9e861a75088..7a2702d78bf 100644 --- a/gdbserver/configure.srv +++ b/gdbserver/configure.srv @@ -109,6 +109,7 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/linux-btrace.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" + srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" srv_linux_usrregs=yes srv_linux_regsets=yes srv_linux_thread_db=yes @@ -371,6 +372,7 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/linux-btrace.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" + srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" srv_tgtobj="${srv_tgtobj} nat/amd64-linux-siginfo.o" srv_linux_usrregs=yes # This is for i386 progs. srv_linux_regsets=yes diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index a6346750f49..df5e6aca081 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -22,6 +22,7 @@ #include "tracepoint.h" #include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" +#include "arch/amd64-linux-tdesc.h" /* fast tracepoints collect registers. */ diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc index 8f14e0937d4..aa346fc9bc3 100644 --- a/gdbserver/linux-i386-ipa.cc +++ b/gdbserver/linux-i386-ipa.cc @@ -22,6 +22,7 @@ #include "tracepoint.h" #include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" +#include "arch/i386-linux-tdesc.h" /* GDB register numbers. */ diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 62612773a87..68d2f13537c 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -29,8 +29,11 @@ #ifdef __x86_64__ #include "nat/amd64-linux-siginfo.h" +#include "arch/amd64-linux-tdesc.h" #endif +#include "arch/i386-linux-tdesc.h" + #include "gdb_proc_service.h" /* Don't include elf/common.h if linux/elf.h got included by gdb_proc_service.h. */ @@ -46,6 +49,7 @@ #include "nat/x86-linux.h" #include "nat/x86-linux-dregs.h" #include "linux-x86-tdesc.h" +#include "nat/x86-linux-tdesc.h" #ifdef __x86_64__ static target_desc_up tdesc_amd64_linux_no_xml; @@ -831,32 +835,20 @@ x86_target::low_siginfo_fixup (siginfo_t *ptrace, gdb_byte *inf, int direction) \f static int use_xml; +/* Cached xcr0 value. This is initialised the first time + x86_linux_read_description is called. */ + +static uint64_t xcr0_storage; + /* Get Linux/x86 target description from running target. */ static const struct target_desc * x86_linux_read_description (void) { - unsigned int machine; - int is_elf64; - int xcr0_features; - int tid; - static uint64_t xcr0; - static int xsave_len; - struct regset_info *regset; - - tid = lwpid_of (current_thread); - - is_elf64 = linux_pid_exe_is_elf_64_file (tid, &machine); + int tid = lwpid_of (current_thread); - if (sizeof (void *) == 4) - { - if (is_elf64 > 0) - error (_("Can't debug 64-bit process with 32-bit GDBserver")); -#ifndef __x86_64__ - else if (machine == EM_X86_64) - error (_("Can't debug x86-64 process with 32-bit GDBserver")); -#endif - } + const char *error_msg + = _("Can't debug 64-bit process with 32-bit GDBserver"); /* If we are not allowed to send an XML target description then we need to use the hard-wired target descriptions. This corresponds to GDB's @@ -866,103 +858,54 @@ x86_linux_read_description (void) generate some alternative target descriptions. */ if (!use_xml) { + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); + bool is_64bit = arch_size.is_64bit (); + bool is_x32 = arch_size.is_x32 (); + + if (sizeof (void *) == 4 && is_64bit && !is_x32) + error ("%s", error_msg); + #ifdef __x86_64__ - if (machine == EM_X86_64) + if (is_64bit && !is_x32) return tdesc_amd64_linux_no_xml.get (); else #endif return tdesc_i386_linux_no_xml.get (); } -#if !defined __x86_64__ && defined HAVE_PTRACE_GETFPXREGS - if (machine == EM_386 && have_ptrace_getfpxregs == -1) - { - elf_fpxregset_t fpxregs; - - if (ptrace (PTRACE_GETFPXREGS, tid, 0, (long) &fpxregs) < 0) - { - have_ptrace_getfpxregs = 0; - have_ptrace_getregset = TRIBOOL_FALSE; - return i386_linux_read_description (X86_XSTATE_X87); - } - else - have_ptrace_getfpxregs = 1; - } -#endif - - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) - { - uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; - struct iovec iov; - - iov.iov_base = xstateregs; - iov.iov_len = sizeof (xstateregs); - - /* Check if PTRACE_GETREGSET works. */ - if (ptrace (PTRACE_GETREGSET, tid, - (unsigned int) NT_X86_XSTATE, (long) &iov) < 0) - have_ptrace_getregset = TRIBOOL_FALSE; - else - { - have_ptrace_getregset = TRIBOOL_TRUE; - - /* Get XCR0 from XSAVE extended state. */ - xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET - / sizeof (uint64_t))]; - - /* No MPX on x32. */ - if (machine == EM_X86_64 && !is_elf64) - xcr0 &= ~X86_XSTATE_MPX; - - xsave_len = x86_xsave_length (); - - /* Use PTRACE_GETREGSET if it is available. */ - for (regset = x86_regsets; - regset->fill_function != NULL; regset++) - if (regset->get_request == PTRACE_GETREGSET) - regset->size = xsave_len; - else if (regset->type != GENERAL_REGS) - regset->size = 0; - } - } - - /* Check the native XCR0 only if PTRACE_GETREGSET is available. */ - xcr0_features = (have_ptrace_getregset == TRIBOOL_TRUE - && (xcr0 & X86_XSTATE_ALL_MASK)); - - if (xcr0_features) - i387_set_xsave_mask (xcr0, xsave_len); + /* Callback that is triggered the first time x86_linux_tdesc_for_tid + reads the xcr0 register. Setup other bits of state */ + auto cb = [] (uint64_t xcr0) + { + i387_set_xsave_mask (xcr0, x86_xsave_length ()); + }; - if (machine == EM_X86_64) - { -#ifdef __x86_64__ - const target_desc *tdesc = NULL; + /* If have_ptrace_getregset is changed to true by calling + x86_linux_tdesc_for_tid then we will perform some additional + initialisation. */ + bool have_ptrace_getregset_is_unknown + = have_ptrace_getregset == TRIBOOL_UNKNOWN; - if (xcr0_features) - { - tdesc = amd64_linux_read_description (xcr0 & X86_XSTATE_ALL_MASK, - !is_elf64); - } + const target_desc *tdesc + = x86_linux_tdesc_for_tid (tid, &have_ptrace_getregset, cb, error_msg, + &xcr0_storage); - if (tdesc == NULL) - tdesc = amd64_linux_read_description (X86_XSTATE_SSE_MASK, !is_elf64); - return tdesc; -#endif - } - else + if (have_ptrace_getregset_is_unknown + && have_ptrace_getregset == TRIBOOL_TRUE) { - const target_desc *tdesc = NULL; - - if (xcr0_features) - tdesc = i386_linux_read_description (xcr0 & X86_XSTATE_ALL_MASK); - - if (tdesc == NULL) - tdesc = i386_linux_read_description (X86_XSTATE_SSE); - - return tdesc; + int xsave_len = x86_xsave_length (); + + /* Use PTRACE_GETREGSET if it is available. */ + for (regset_info *regset = x86_regsets; + regset->fill_function != nullptr; + regset++) + if (regset->get_request == PTRACE_GETREGSET) + regset->size = xsave_len; + else if (regset->type != GENERAL_REGS) + regset->size = 0; } - gdb_assert_not_reached ("failed to return tdesc"); + return tdesc; } /* Update all the target description of all processes; a new GDB diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc index cd3b5d80e37..af3735aa895 100644 --- a/gdbserver/linux-x86-tdesc.cc +++ b/gdbserver/linux-x86-tdesc.cc @@ -23,8 +23,10 @@ #include "gdbsupport/x86-xstate.h" #ifdef __x86_64__ #include "arch/amd64.h" +#include "arch/amd64-linux-tdesc.h" #endif #include "x86-tdesc.h" +#include "arch/i386-linux-tdesc.h" /* Return the right x86_linux_tdesc index for a given XCR0. Return X86_TDESC_LAST if can't find a match. */ diff --git a/gdbserver/linux-x86-tdesc.h b/gdbserver/linux-x86-tdesc.h index f9561b129ae..576aaf5e165 100644 --- a/gdbserver/linux-x86-tdesc.h +++ b/gdbserver/linux-x86-tdesc.h @@ -46,11 +46,4 @@ int amd64_get_ipa_tdesc_idx (const struct target_desc *tdesc); const struct target_desc *i386_get_ipa_tdesc (int idx); -#ifdef __x86_64__ -const struct target_desc *amd64_linux_read_description (uint64_t xcr0, - bool is_x32); -#endif - -const struct target_desc *i386_linux_read_description (uint64_t xcr0); - #endif /* GDBSERVER_LINUX_X86_TDESC_H */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 07/11] gdb/gdbserver: share some code relating to target description creation 2024-04-26 15:01 ` [PATCHv5 07/11] gdb/gdbserver: share some code relating to target description creation Andrew Burgess @ 2024-04-29 14:34 ` Willgerodt, Felix 2024-05-07 11:40 ` Andrew Burgess 0 siblings, 1 reply; 116+ messages in thread From: Willgerodt, Felix @ 2024-04-29 14:34 UTC (permalink / raw) To: Andrew Burgess, gdb-patches; +Cc: John Baldwin > -----Original Message----- > From: Andrew Burgess <aburgess@redhat.com> > Sent: Freitag, 26. April 2024 17:02 > To: gdb-patches@sourceware.org > Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix > <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> > Subject: [PATCHv5 07/11] gdb/gdbserver: share some code relating to target > description creation > > This commit is part of a series to share more of the x86 target > description creation code between GDB and gdbserver. > > Unlike previous commits which were mostly refactoring, this commit is > the first that makes a real change, though that change should mostly > be for gdbserver; I've largely adopted the "GDB" way of doing things > for gdbserver, and this fixes a real gdbserver bug. > > On a x86-64 Linux target, running the test: > > gdb.server/connect-with-no-symbol-file.exp > > results in two core files being created. Both of these core files are > from the inferior process, created after gdbserver has detached. > > In this test a gdbserver process is started and then, after gdbserver > has started, but before GDB attaches, we either delete the inferior > executable, or change its permissions so it can't be read. Only after > doing this do we attempt to connect with GDB. > > As GDB connects to gdbserver, gdbserver attempts to figure out the > target description so that it can send the description to GDB, this > involves a call to x86_linux_read_description. > > In x86_linux_read_description one of the first things we do is try to > figure out if the process is 32-bit or 64-bit. To do this we look up > the executable via the thread-id, and then attempt to read the > architecture size from the executable. This isn't going to work if > the executable has been deleted, or is no longer readable. > > And so, as we can't read the executable, we default to an i386 target > and use an i386 target description. > > A consequence of using an i386 target description is that addresses > are assumed to be 32-bits. Here's an example session that shows the > problems this causes. This is run on an x86-64 machine, and the test > binary (xx.x) is a standard 64-bit x86-64 binary: > > shell_1$ gdbserver --once localhost :54321 /tmp/xx.x > > shell_2$ gdb -q > (gdb) set sysroot > (gdb) shell chmod 000 /tmp/xx.x > (gdb) target remote :54321 > Remote debugging using :54321 > warning: /tmp/xx.x: Permission denied. > 0xf7fd3110 in ?? () > (gdb) show architecture > The target architecture is set to "auto" (currently "i386"). > (gdb) p/x $pc > $1 = 0xf7fd3110 > (gdb) info proc mappings > process 2412639 > Mapped address spaces: > > Start Addr End Addr Size Offset Perms objfile > 0x400000 0x401000 0x1000 0x0 r--p /tmp/xx.x > 0x401000 0x402000 0x1000 0x1000 r-xp /tmp/xx.x > 0x402000 0x403000 0x1000 0x2000 r--p /tmp/xx.x > 0x403000 0x405000 0x2000 0x2000 rw-p /tmp/xx.x > 0xf7fcb000 0xf7fcf000 0x4000 0x0 r--p [vvar] > 0xf7fcf000 0xf7fd1000 0x2000 0x0 r-xp [vdso] > 0xf7fd1000 0xf7fd3000 0x2000 0x0 r--p /usr/lib64/ld-2.30.so > 0xf7fd3000 0xf7ff3000 0x20000 0x2000 r-xp /usr/lib64/ld-2.30.so > 0xf7ff3000 0xf7ffb000 0x8000 0x22000 r--p /usr/lib64/ld-2.30.so > 0xf7ffc000 0xf7ffe000 0x2000 0x2a000 rw-p /usr/lib64/ld-2.30.so > 0xf7ffe000 0xf7fff000 0x1000 0x0 rw-p > 0xfffda000 0xfffff000 0x25000 0x0 rw-p [stack] > 0xff600000 0xff601000 0x1000 0x0 r-xp [vsyscall] > (gdb) info inferiors > Num Description Connection Executable > * 1 process 2412639 1 (remote :54321) > (gdb) shell cat /proc/2412639/maps > 00400000-00401000 r--p 00000000 fd:03 45907133 /tmp/xx.x > 00401000-00402000 r-xp 00001000 fd:03 45907133 /tmp/xx.x > 00402000-00403000 r--p 00002000 fd:03 45907133 /tmp/xx.x > 00403000-00405000 rw-p 00002000 fd:03 45907133 /tmp/xx.x > 7ffff7fcb000-7ffff7fcf000 r--p 00000000 00:00 0 [vvar] > 7ffff7fcf000-7ffff7fd1000 r-xp 00000000 00:00 0 [vdso] > 7ffff7fd1000-7ffff7fd3000 r--p 00000000 fd:00 143904 /usr/lib64/ld-2.30.so > 7ffff7fd3000-7ffff7ff3000 r-xp 00002000 fd:00 143904 /usr/lib64/ld-2.30.so > 7ffff7ff3000-7ffff7ffb000 r--p 00022000 fd:00 143904 /usr/lib64/ld-2.30.so > 7ffff7ffc000-7ffff7ffe000 rw-p 0002a000 fd:00 143904 /usr/lib64/ld-2.30.so > 7ffff7ffe000-7ffff7fff000 rw-p 00000000 00:00 0 > 7ffffffda000-7ffffffff000 rw-p 00000000 00:00 0 [stack] > ffffffffff600000-ffffffffff601000 r-xp 00000000 00:00 0 [vsyscall] > (gdb) > > Notice the difference between the mappings reported via GDB and those > reported directly from the kernel via /proc/PID/maps, the addresses of > every mapping is clamped to 32-bits for GDB, while the kernel reports > real 64-bit addresses. > > Notice also that the $pc value is a 32-bit value. It appears to be > within one of the mappings reported by GDB, but is outside any of the > mappings reported from the kernel. > > And this is where the problem arises. When gdbserver detaches from > the inferior we pass the inferior the address from which it should > resume. Due to the 32/64 bit confusion we tell the inferior to resume > from the 32-bit $pc value, which is not within any valid mapping, and > so, as soon as the inferior resumes, it segfaults. > > If we look at how GDB (not gdbserver) figures out its target > description then we see an interesting difference. GDB doesn't try to > read the executable. Instead GDB uses ptrace to query the thread's > state, and uses this to figure out the if the thread is 32 or 64 bit. > > If we update gdbserver to do it the "GDB" way then the above problem > is resolved, gdbserver now sees the process as 64-bit, and when we > detach from the inferior we give it the correct 64-bit address, and > the inferior no longer segfaults. > > Now, I could just update the gdbserver code, but better, I think, to > share one copy of the code between GDB and gdbserver in gdb/nat/. > That is what this commit does. > > The cores of x86_linux_read_description from gdbserver and > x86_linux_nat_target::read_description from GDB are moved into a new > file gdb/nat/x86-linux-tdesc.c and combined into a single function > x86_linux_tdesc_for_tid which is called from each location. > > This new function does things the GDB way, the only changes are to > allow for the sharing; we now have a callback function to call the > first time that the xcr0 state is read, this allows for GDB and > gdbserver to perform their own initialisation as needed, and > additionally, the new function takes a pointer for where to cache the > xcr0 value, this isn't needed for this commit, but will be useful in a > later commit where gdbserver will want to read this cached xcr0 > value. > > Another thing to note about this commit is how the functions > i386_linux_read_description and amd64_linux_read_description are > handled. For now I've left these function as implemented separately > in GDB and gdbserver. I've moved the declarations of these functions > into gdb/arch/{i386,amd64}-linux-tdesc.h, but the implementations are > left where they are. > > A later commit in this series will make these functions shared too, > but doing this is not trivial, so I've left that for a separate > commit. Merging the declarations as I've done here ensures that > everyone implements the function to the same API, and once these > functions are shared (in a later commit) we'll want a shared > declaration anyway. > --- > gdb/Makefile.in | 3 + > gdb/amd64-linux-tdep.c | 1 + > gdb/amd64-linux-tdep.h | 6 -- > gdb/arch/amd64-linux-tdesc.h | 30 +++++++ > gdb/arch/i386-linux-tdesc.h | 29 +++++++ > gdb/configure.nat | 4 +- > gdb/i386-linux-tdep.c | 1 + > gdb/i386-linux-tdep.h | 3 - > gdb/nat/x86-linux-tdesc.c | 124 ++++++++++++++++++++++++++++ > gdb/nat/x86-linux-tdesc.h | 60 ++++++++++++++ > gdb/x86-linux-nat.c | 91 ++++----------------- > gdbserver/configure.srv | 2 + > gdbserver/linux-amd64-ipa.cc | 1 + > gdbserver/linux-i386-ipa.cc | 1 + > gdbserver/linux-x86-low.cc | 151 +++++++++++------------------------ > gdbserver/linux-x86-tdesc.cc | 2 + > gdbserver/linux-x86-tdesc.h | 7 -- > 17 files changed, 317 insertions(+), 199 deletions(-) > create mode 100644 gdb/arch/amd64-linux-tdesc.h > create mode 100644 gdb/arch/i386-linux-tdesc.h > create mode 100644 gdb/nat/x86-linux-tdesc.c > create mode 100644 gdb/nat/x86-linux-tdesc.h > > diff --git a/gdb/Makefile.in b/gdb/Makefile.in > index 353e45b3cec..a24b2232daa 100644 > --- a/gdb/Makefile.in > +++ b/gdb/Makefile.in > @@ -1549,8 +1549,10 @@ HFILES_NO_SRCDIR = \ > arch/aarch64-insn.h \ > arch/aarch64-mte-linux.h \ > arch/aarch64-scalable-linux.h \ > + arch/amd64-linux-tdesc.h \ > arch/arc.h \ > arch/arm.h \ > + arch/i386-linux-tdesc.h \ > arch/i386.h \ > arch/loongarch.h \ > arch/ppc-linux-common.h \ > @@ -1606,6 +1608,7 @@ HFILES_NO_SRCDIR = \ > nat/x86-gcc-cpuid.h \ > nat/x86-linux.h \ > nat/x86-linux-dregs.h \ > + nat/x86-linux-tdesc.h \ > python/py-event.h \ > python/py-events.h \ > python/py-stopevent.h \ > diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c > index 9d560ac4fbf..bcb9868e79e 100644 > --- a/gdb/amd64-linux-tdep.c > +++ b/gdb/amd64-linux-tdep.c > @@ -41,6 +41,7 @@ > #include "arch/amd64.h" > #include "target-descriptions.h" > #include "expop.h" > +#include "arch/amd64-linux-tdesc.h" > > /* The syscall's XML filename for i386. */ > #define XML_SYSCALL_FILENAME_AMD64 "syscalls/amd64-linux.xml" > diff --git a/gdb/amd64-linux-tdep.h b/gdb/amd64-linux-tdep.h > index 2003dcda78f..0ec49e7fe03 100644 > --- a/gdb/amd64-linux-tdep.h > +++ b/gdb/amd64-linux-tdep.h > @@ -43,12 +43,6 @@ extern struct target_desc *tdesc_x32_linux; > extern struct target_desc *tdesc_x32_avx_linux; > extern struct target_desc *tdesc_x32_avx_avx512_linux; > > -/* Return the right amd64-linux target descriptions according to > - XCR0_FEATURES_BIT and IS_X32. */ > - > -const target_desc *amd64_linux_read_description (uint64_t xcr0_features_bit, > - bool is_x32); > - > /* Enum that defines the syscall identifiers for amd64 linux. > Used for process record/replay, these will be translated into > a gdb-canonical set of syscall ids in linux-record.c. */ > diff --git a/gdb/arch/amd64-linux-tdesc.h b/gdb/arch/amd64-linux-tdesc.h > new file mode 100644 > index 00000000000..db425b60df6 > --- /dev/null > +++ b/gdb/arch/amd64-linux-tdesc.h > @@ -0,0 +1,30 @@ > +/* Target description related code for GNU/Linux x86-64. > + > + Copyright (C) 2024 Free Software Foundation, Inc. > + > + This file is part of GDB. > + > + This program is free software; you can redistribute it and/or modify > + it under the terms of the GNU General Public License as published by > + the Free Software Foundation; either version 3 of the License, or > + (at your option) any later version. > + > + This program is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the > + GNU General Public License for more details. > + > + You should have received a copy of the GNU General Public License > + along with this program. If not, see <http://www.gnu.org/licenses/>. */ > + > +#ifndef ARCH_AMD64_LINUX_TDESC_H > +#define ARCH_AMD64_LINUX_TDESC_H > + > +struct target_desc; > + > +/* Return the AMD64 target descriptions corresponding to XCR0 and IS_X32. */ > + > +extern const target_desc *amd64_linux_read_description (uint64_t xcr0, > + bool is_x32); > + > +#endif /* ARCH_AMD64_LINUX_TDESC_H */ > diff --git a/gdb/arch/i386-linux-tdesc.h b/gdb/arch/i386-linux-tdesc.h > new file mode 100644 > index 00000000000..0b736337a75 > --- /dev/null > +++ b/gdb/arch/i386-linux-tdesc.h > @@ -0,0 +1,29 @@ > +/* Target description related code for GNU/Linux i386. > + > + Copyright (C) 2024 Free Software Foundation, Inc. > + > + This file is part of GDB. > + > + This program is free software; you can redistribute it and/or modify > + it under the terms of the GNU General Public License as published by > + the Free Software Foundation; either version 3 of the License, or > + (at your option) any later version. > + > + This program is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the > + GNU General Public License for more details. > + > + You should have received a copy of the GNU General Public License > + along with this program. If not, see <http://www.gnu.org/licenses/>. */ > + > +#ifndef ARCH_I386_LINUX_TDESC_H > +#define ARCH_I386_LINUX_TDESC_H > + > +struct target_desc; > + > +/* Return the i386 target description corresponding to XCR0. */ > + > +extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); > + > +#endif /* ARCH_I386_LINUX_TDESC_H */ > diff --git a/gdb/configure.nat b/gdb/configure.nat > index 8b98511cef7..4bcc0696027 100644 > --- a/gdb/configure.nat > +++ b/gdb/configure.nat > @@ -256,7 +256,7 @@ case ${gdb_host} in > NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ > nat/x86-xstate.o \ > i386-linux-nat.o x86-linux-nat.o nat/linux-btrace.o \ > - nat/x86-linux.o nat/x86-linux-dregs.o" > + nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o" > ;; > ia64) > # Host: Intel IA-64 running GNU/Linux > @@ -322,7 +322,7 @@ case ${gdb_host} in > NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ > nat/x86-xstate.o amd64-nat.o amd64-linux-nat.o x86-linux-nat.o > \ > nat/linux-btrace.o \ > - nat/x86-linux.o nat/x86-linux-dregs.o \ > + nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o \ > nat/amd64-linux-siginfo.o" > ;; > sparc) > diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c > index 44730f204db..78ebc99d3df 100644 > --- a/gdb/i386-linux-tdep.c > +++ b/gdb/i386-linux-tdep.c > @@ -39,6 +39,7 @@ > > #include "i387-tdep.h" > #include "gdbsupport/x86-xstate.h" > +#include "arch/i386-linux-tdesc.h" > > /* The syscall's XML filename for i386. */ > #define XML_SYSCALL_FILENAME_I386 "syscalls/i386-linux.xml" > diff --git a/gdb/i386-linux-tdep.h b/gdb/i386-linux-tdep.h > index 07593c6a8ec..e8691cd778e 100644 > --- a/gdb/i386-linux-tdep.h > +++ b/gdb/i386-linux-tdep.h > @@ -55,9 +55,6 @@ extern void i386_linux_report_signal_info (struct gdbarch > *gdbarch, > struct ui_out *uiout, > enum gdb_signal siggnal); > > -/* Return the target description according to XCR0. */ > -extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); > - > extern int i386_linux_gregset_reg_offset[]; > > /* Return x86 siginfo type. */ > diff --git a/gdb/nat/x86-linux-tdesc.c b/gdb/nat/x86-linux-tdesc.c > new file mode 100644 > index 00000000000..c2301e6a873 > --- /dev/null > +++ b/gdb/nat/x86-linux-tdesc.c > @@ -0,0 +1,124 @@ > +/* Target description related code for GNU/Linux x86 (i386 and x86-64). > + > + Copyright (C) 2024 Free Software Foundation, Inc. > + > + This file is part of GDB. > + > + This program is free software; you can redistribute it and/or modify > + it under the terms of the GNU General Public License as published by > + the Free Software Foundation; either version 3 of the License, or > + (at your option) any later version. > + > + This program is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the > + GNU General Public License for more details. > + > + You should have received a copy of the GNU General Public License > + along with this program. If not, see <http://www.gnu.org/licenses/>. */ > + > +#include "nat/x86-linux-tdesc.h" > +#ifdef __x86_64__ > +#include "arch/amd64.h" > +#include "arch/amd64-linux-tdesc.h" > +#endif > +#include "arch/i386.h" > +#include "arch/i386-linux-tdesc.h" > + > +#include "nat/x86-linux.h" > +#include "nat/gdb_ptrace.h" > +#include "nat/x86-xstate.h" > + > +#ifndef __x86_64__ > +#include <sys/procfs.h> > +#endif > + > +#include <sys/uio.h> > +#include <elf.h> > + > +#ifndef IN_PROCESS_AGENT > + > +/* See nat/x86-linux-tdesc.h. */ > + > +const target_desc * > +x86_linux_tdesc_for_tid (int tid, enum tribool *have_ptrace_getregset, > + gdb::function_view<void (uint64_t)> xcr0_init_cb, > + const char *error_msg, uint64_t *xcr0_storage) > +{ > +#ifdef __x86_64__ > + > + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); > + bool is_64bit = arch_size.is_64bit (); > + bool is_x32 = arch_size.is_x32 (); > + > + if (sizeof (void *) == 4 && is_64bit && !is_x32) > + error ("%s", error_msg); > + > +#elif HAVE_PTRACE_GETFPXREGS > + if (have_ptrace_getfpxregs == -1) > + { > + elf_fpxregset_t fpxregs; > + > + if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) > + { > + have_ptrace_getfpxregs = 0; > + *have_ptrace_getregset = TRIBOOL_FALSE; We tried PTRACE_GETFPXREGS, does that really indicate PTRACE_GETREGSET is not supported? I realize you just moved this code, but it feels weird to me. Then again this is Linux specific and it might actually be the case. > + return i386_linux_read_description (X86_XSTATE_X87_MASK); > + } > + } > +#endif > + > + if (*have_ptrace_getregset == TRIBOOL_UNKNOWN) > + { > + uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; > + struct iovec iov; > + > + iov.iov_base = xstateregs; > + iov.iov_len = sizeof (xstateregs); > + > + /* Check if PTRACE_GETREGSET works. */ > + if (ptrace (PTRACE_GETREGSET, tid, > + (unsigned int) NT_X86_XSTATE, &iov) < 0) > + { > + *have_ptrace_getregset = TRIBOOL_FALSE; > + *xcr0_storage = 0; > + } > + else > + { > + *have_ptrace_getregset = TRIBOOL_TRUE; > + > + /* Get XCR0 from XSAVE extended state. */ > + *xcr0_storage = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET > + / sizeof (uint64_t))]; > + > +#ifdef __x86_64__ > + /* No MPX on x32. */ > + if (is_64bit && is_x32) > + *xcr0_storage &= ~X86_XSTATE_MPX; > +#endif /* __x86_64__ */ > + > + xcr0_init_cb (*xcr0_storage); Needing a callback made me scratch my head. What could be the reason that GDB and gdbserver really want to do different things here? Digging into the code made me see that GDB saves just the xsave_layout. Gdbserver in addition saves xcr0, though I guess that might not even be needed if you have a tdesc already (or the xsave_layout?). And they are saving things in different locations (global in gdbserver, member in GDB), but that could be done without a callback I think. I realize that changing these things is a bigger tasks, that you might not want to do at all or mix with this series. But can we avoid a callback? E.g. by adding another return value, that only gdbserver uses for now? With a TODO comment explaining this? > + } > + } > + > + /* Check the native XCR0 only if PTRACE_GETREGSET is available. If > + PTRACE_GETREGSET is not available then set xcr0_features_bits to > + zero so that the "no-features" descriptions are returned by the > + switches below. */ > + uint64_t xcr0_features_bits; > + if (*have_ptrace_getregset == TRIBOOL_TRUE) > + xcr0_features_bits = *xcr0_storage & X86_XSTATE_ALL_MASK; > + else > + xcr0_features_bits = 0; > + > +#ifdef __x86_64__ > + if (is_64bit) > + { > + return amd64_linux_read_description (xcr0_features_bits, is_x32); > + } Nit: I think we could just drop the brackets here. > + else > +#endif > + return i386_linux_read_description (xcr0_features_bits); > +} > + > +#endif /* !IN_PROCESS_AGENT */ > diff --git a/gdb/nat/x86-linux-tdesc.h b/gdb/nat/x86-linux-tdesc.h > new file mode 100644 > index 00000000000..8c481609876 > --- /dev/null > +++ b/gdb/nat/x86-linux-tdesc.h > @@ -0,0 +1,60 @@ > +/* Target description related code for GNU/Linux x86 (i386 and x86-64). > + > + Copyright (C) 2024 Free Software Foundation, Inc. > + > + This file is part of GDB. > + > + This program is free software; you can redistribute it and/or modify > + it under the terms of the GNU General Public License as published by > + the Free Software Foundation; either version 3 of the License, or > + (at your option) any later version. > + > + This program is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the > + GNU General Public License for more details. > + > + You should have received a copy of the GNU General Public License > + along with this program. If not, see <http://www.gnu.org/licenses/>. */ > + > +#ifndef NAT_X86_LINUX_TDESC_H > +#define NAT_X86_LINUX_TDESC_H > + > +#include "gdbsupport/function-view.h" > + > +struct target_desc; > + > +/* Return the target description for Linux thread TID. > + > + When *HAVE_PTRACE_GETREGSET is TRIBOOL_UNKNOWN then the current > value of > + xcr0 is read using ptrace calls and stored into *XCR0_STORAGE. Then > + XCR0_INIT_CB is called with the value of *XCR0_STORAGE and > + *HAVE_PTRACE_GETREGSET is set to TRIBOOL_TRUE. > + > + If the attempt to read xcr0 using ptrace fails then *XCR0_STORAGE is set > + to zero and *HAVE_PTRACE_GETREGSET is set to TRIBOOL_FALSE. > + > + The storage pointed to by XCR0_STORAGE must exist until the program > + terminates, this storage is used to cache the xcr0 value. As such > + XCR0_INIT_CB will only be called once if xcr0 is successfully read using > + ptrace, or not at all if the ptrace call fails. > + > + This function returns a target description based on the extracted xcr0 > + value along with other characteristics of the thread identified by TID. > + > + This function can return nullptr if we encounter a machine configuration > + for which a target_desc cannot be created. Ideally this would not be > + the case, we should be able to create a target description for every > + possible machine configuration. See amd64_linux_read_description and > + i386_linux_read_description for cases when nullptr might be > + returned. > + > + ERROR_MSG is using in an error() call if we try to create a target > + description for a 64-bit process but this is a 32-bit build of GDB. */ > + > +extern const target_desc * > +x86_linux_tdesc_for_tid (int tid, enum tribool *have_ptrace_getregset, > + gdb::function_view<void (uint64_t)> xcr0_init_cb, > + const char *error_msg, uint64_t *xcr0_storage); > + > +#endif /* NAT_X86_LINUX_TDESC_H */ > diff --git a/gdb/x86-linux-nat.c b/gdb/x86-linux-nat.c > index 12965c7a21b..9fbccb28d21 100644 > --- a/gdb/x86-linux-nat.c > +++ b/gdb/x86-linux-nat.c > @@ -38,6 +38,7 @@ > #include "nat/x86-linux.h" > #include "nat/x86-linux-dregs.h" > #include "nat/linux-ptrace.h" > +#include "nat/x86-linux-tdesc.h" > > /* linux_nat_target::low_new_fork implementation. */ > > @@ -92,90 +93,26 @@ x86_linux_nat_target::post_startup_inferior (ptid_t ptid) > const struct target_desc * > x86_linux_nat_target::read_description () > { > - int tid; > - int is_64bit = 0; > -#ifdef __x86_64__ > - int is_x32; > -#endif > - static uint64_t xcr0; > - uint64_t xcr0_features_bits; > + static uint64_t xcr0_storage; > > if (inferior_ptid == null_ptid) > return this->beneath ()->read_description (); > > - tid = inferior_ptid.pid (); > - > -#ifdef __x86_64__ > - > - x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); > - is_64bit = arch_size.is_64bit (); > - is_x32 = arch_size.is_x32 (); > - > -#elif HAVE_PTRACE_GETFPXREGS > - if (have_ptrace_getfpxregs == -1) > - { > - elf_fpxregset_t fpxregs; > - > - if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) > - { > - have_ptrace_getfpxregs = 0; > - have_ptrace_getregset = TRIBOOL_FALSE; > - return i386_linux_read_description (X86_XSTATE_X87_MASK); > - } > - } > -#endif > - > - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) > - { > - uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; > - struct iovec iov; > - > - iov.iov_base = xstateregs; > - iov.iov_len = sizeof (xstateregs); > - > - /* Check if PTRACE_GETREGSET works. */ > - if (ptrace (PTRACE_GETREGSET, tid, > - (unsigned int) NT_X86_XSTATE, &iov) < 0) > - have_ptrace_getregset = TRIBOOL_FALSE; > - else > - { > - have_ptrace_getregset = TRIBOOL_TRUE; > - > - /* Get XCR0 from XSAVE extended state. */ > - xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET > - / sizeof (uint64_t))]; > - > - m_xsave_layout = x86_fetch_xsave_layout (xcr0, x86_xsave_length ()); > - } > - } > - > - /* Check the native XCR0 only if PTRACE_GETREGSET is available. If > - PTRACE_GETREGSET is not available then set xcr0_features_bits to > - zero so that the "no-features" descriptions are returned by the > - switches below. */ Should we s/switches/code? > - if (have_ptrace_getregset == TRIBOOL_TRUE) > - xcr0_features_bits = xcr0 & X86_XSTATE_ALL_MASK; > - else > - xcr0_features_bits = 0; > - > - if (is_64bit) > - { > -#ifdef __x86_64__ > - return amd64_linux_read_description (xcr0_features_bits, is_x32); > -#endif > - } > - else > - { > - const struct target_desc * tdesc > - = i386_linux_read_description (xcr0_features_bits); > + int tid = inferior_ptid.pid (); > > - if (tdesc == NULL) > - tdesc = i386_linux_read_description (X86_XSTATE_SSE_MASK); > + const char *error_msg > + = _("Can't debug 64-bit process with 32-bit GDB"); > > - return tdesc; > - } > + /* Callback that is triggered the first time x86_linux_tdesc_for_tid > + reads the xcr0 register. Setup other bits of state. */ > + auto cb = [&] (uint64_t xcr0) > + { > + this->m_xsave_layout > + = x86_fetch_xsave_layout (xcr0, x86_xsave_length ()); > + }; > > - gdb_assert_not_reached ("failed to return tdesc"); > + return x86_linux_tdesc_for_tid (tid, &have_ptrace_getregset, cb, > + error_msg, &xcr0_storage); > } > > > > > diff --git a/gdbserver/configure.srv b/gdbserver/configure.srv > index 9e861a75088..7a2702d78bf 100644 > --- a/gdbserver/configure.srv > +++ b/gdbserver/configure.srv > @@ -109,6 +109,7 @@ case "${gdbserver_host}" in > srv_tgtobj="${srv_tgtobj} nat/linux-btrace.o" > srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" > srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" > + srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" > srv_linux_usrregs=yes > srv_linux_regsets=yes > srv_linux_thread_db=yes > @@ -371,6 +372,7 @@ case "${gdbserver_host}" in > srv_tgtobj="${srv_tgtobj} nat/linux-btrace.o" > srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" > srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" > + srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" > srv_tgtobj="${srv_tgtobj} nat/amd64-linux-siginfo.o" > srv_linux_usrregs=yes # This is for i386 progs. > srv_linux_regsets=yes > diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc > index a6346750f49..df5e6aca081 100644 > --- a/gdbserver/linux-amd64-ipa.cc > +++ b/gdbserver/linux-amd64-ipa.cc > @@ -22,6 +22,7 @@ > #include "tracepoint.h" > #include "linux-x86-tdesc.h" > #include "gdbsupport/x86-xstate.h" > +#include "arch/amd64-linux-tdesc.h" > > /* fast tracepoints collect registers. */ > > diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc > index 8f14e0937d4..aa346fc9bc3 100644 > --- a/gdbserver/linux-i386-ipa.cc > +++ b/gdbserver/linux-i386-ipa.cc > @@ -22,6 +22,7 @@ > #include "tracepoint.h" > #include "linux-x86-tdesc.h" > #include "gdbsupport/x86-xstate.h" > +#include "arch/i386-linux-tdesc.h" > > /* GDB register numbers. */ > > diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc > index 62612773a87..68d2f13537c 100644 > --- a/gdbserver/linux-x86-low.cc > +++ b/gdbserver/linux-x86-low.cc > @@ -29,8 +29,11 @@ > > #ifdef __x86_64__ > #include "nat/amd64-linux-siginfo.h" > +#include "arch/amd64-linux-tdesc.h" > #endif > > +#include "arch/i386-linux-tdesc.h" > + > #include "gdb_proc_service.h" > /* Don't include elf/common.h if linux/elf.h got included by > gdb_proc_service.h. */ > @@ -46,6 +49,7 @@ > #include "nat/x86-linux.h" > #include "nat/x86-linux-dregs.h" > #include "linux-x86-tdesc.h" > +#include "nat/x86-linux-tdesc.h" > > #ifdef __x86_64__ > static target_desc_up tdesc_amd64_linux_no_xml; > @@ -831,32 +835,20 @@ x86_target::low_siginfo_fixup (siginfo_t *ptrace, > gdb_byte *inf, int direction) > > > > static int use_xml; > > +/* Cached xcr0 value. This is initialised the first time > + x86_linux_read_description is called. */ > + > +static uint64_t xcr0_storage; We already cache it in fp-i387.cc. It get's set by the callback you added. I think we should avoid having it twice. > /* Get Linux/x86 target description from running target. */ > > static const struct target_desc * > x86_linux_read_description (void) Not strictly necessary, but we could remove the void while we are touching this. This is a .cc file. Thanks, Felix Intel Deutschland GmbH Registered Address: Am Campeon 10, 85579 Neubiberg, Germany Tel: +49 89 99 8853-0, www.intel.de <http://www.intel.de> Managing Directors: Christin Eisenschmid, Sharon Heck, Tiffany Doon Silva Chairperson of the Supervisory Board: Nicole Lau Registered Office: Munich Commercial Register: Amtsgericht Muenchen HRB 186928 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 07/11] gdb/gdbserver: share some code relating to target description creation 2024-04-29 14:34 ` Willgerodt, Felix @ 2024-05-07 11:40 ` Andrew Burgess 0 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-07 11:40 UTC (permalink / raw) To: Willgerodt, Felix, gdb-patches; +Cc: John Baldwin "Willgerodt, Felix" <felix.willgerodt@intel.com> writes: >> -----Original Message----- >> From: Andrew Burgess <aburgess@redhat.com> >> Sent: Freitag, 26. April 2024 17:02 >> To: gdb-patches@sourceware.org >> Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix >> <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> >> Subject: [PATCHv5 07/11] gdb/gdbserver: share some code relating to target >> description creation >> >> This commit is part of a series to share more of the x86 target >> description creation code between GDB and gdbserver. >> >> Unlike previous commits which were mostly refactoring, this commit is >> the first that makes a real change, though that change should mostly >> be for gdbserver; I've largely adopted the "GDB" way of doing things >> for gdbserver, and this fixes a real gdbserver bug. >> >> On a x86-64 Linux target, running the test: >> >> gdb.server/connect-with-no-symbol-file.exp >> >> results in two core files being created. Both of these core files are >> from the inferior process, created after gdbserver has detached. >> >> In this test a gdbserver process is started and then, after gdbserver >> has started, but before GDB attaches, we either delete the inferior >> executable, or change its permissions so it can't be read. Only after >> doing this do we attempt to connect with GDB. >> >> As GDB connects to gdbserver, gdbserver attempts to figure out the >> target description so that it can send the description to GDB, this >> involves a call to x86_linux_read_description. >> >> In x86_linux_read_description one of the first things we do is try to >> figure out if the process is 32-bit or 64-bit. To do this we look up >> the executable via the thread-id, and then attempt to read the >> architecture size from the executable. This isn't going to work if >> the executable has been deleted, or is no longer readable. >> >> And so, as we can't read the executable, we default to an i386 target >> and use an i386 target description. >> >> A consequence of using an i386 target description is that addresses >> are assumed to be 32-bits. Here's an example session that shows the >> problems this causes. This is run on an x86-64 machine, and the test >> binary (xx.x) is a standard 64-bit x86-64 binary: >> >> shell_1$ gdbserver --once localhost :54321 /tmp/xx.x >> >> shell_2$ gdb -q >> (gdb) set sysroot >> (gdb) shell chmod 000 /tmp/xx.x >> (gdb) target remote :54321 >> Remote debugging using :54321 >> warning: /tmp/xx.x: Permission denied. >> 0xf7fd3110 in ?? () >> (gdb) show architecture >> The target architecture is set to "auto" (currently "i386"). >> (gdb) p/x $pc >> $1 = 0xf7fd3110 >> (gdb) info proc mappings >> process 2412639 >> Mapped address spaces: >> >> Start Addr End Addr Size Offset Perms objfile >> 0x400000 0x401000 0x1000 0x0 r--p /tmp/xx.x >> 0x401000 0x402000 0x1000 0x1000 r-xp /tmp/xx.x >> 0x402000 0x403000 0x1000 0x2000 r--p /tmp/xx.x >> 0x403000 0x405000 0x2000 0x2000 rw-p /tmp/xx.x >> 0xf7fcb000 0xf7fcf000 0x4000 0x0 r--p [vvar] >> 0xf7fcf000 0xf7fd1000 0x2000 0x0 r-xp [vdso] >> 0xf7fd1000 0xf7fd3000 0x2000 0x0 r--p /usr/lib64/ld-2.30.so >> 0xf7fd3000 0xf7ff3000 0x20000 0x2000 r-xp /usr/lib64/ld-2.30.so >> 0xf7ff3000 0xf7ffb000 0x8000 0x22000 r--p /usr/lib64/ld-2.30.so >> 0xf7ffc000 0xf7ffe000 0x2000 0x2a000 rw-p /usr/lib64/ld-2.30.so >> 0xf7ffe000 0xf7fff000 0x1000 0x0 rw-p >> 0xfffda000 0xfffff000 0x25000 0x0 rw-p [stack] >> 0xff600000 0xff601000 0x1000 0x0 r-xp [vsyscall] >> (gdb) info inferiors >> Num Description Connection Executable >> * 1 process 2412639 1 (remote :54321) >> (gdb) shell cat /proc/2412639/maps >> 00400000-00401000 r--p 00000000 fd:03 45907133 /tmp/xx.x >> 00401000-00402000 r-xp 00001000 fd:03 45907133 /tmp/xx.x >> 00402000-00403000 r--p 00002000 fd:03 45907133 /tmp/xx.x >> 00403000-00405000 rw-p 00002000 fd:03 45907133 /tmp/xx.x >> 7ffff7fcb000-7ffff7fcf000 r--p 00000000 00:00 0 [vvar] >> 7ffff7fcf000-7ffff7fd1000 r-xp 00000000 00:00 0 [vdso] >> 7ffff7fd1000-7ffff7fd3000 r--p 00000000 fd:00 143904 /usr/lib64/ld-2.30.so >> 7ffff7fd3000-7ffff7ff3000 r-xp 00002000 fd:00 143904 /usr/lib64/ld-2.30.so >> 7ffff7ff3000-7ffff7ffb000 r--p 00022000 fd:00 143904 /usr/lib64/ld-2.30.so >> 7ffff7ffc000-7ffff7ffe000 rw-p 0002a000 fd:00 143904 /usr/lib64/ld-2.30.so >> 7ffff7ffe000-7ffff7fff000 rw-p 00000000 00:00 0 >> 7ffffffda000-7ffffffff000 rw-p 00000000 00:00 0 [stack] >> ffffffffff600000-ffffffffff601000 r-xp 00000000 00:00 0 [vsyscall] >> (gdb) >> >> Notice the difference between the mappings reported via GDB and those >> reported directly from the kernel via /proc/PID/maps, the addresses of >> every mapping is clamped to 32-bits for GDB, while the kernel reports >> real 64-bit addresses. >> >> Notice also that the $pc value is a 32-bit value. It appears to be >> within one of the mappings reported by GDB, but is outside any of the >> mappings reported from the kernel. >> >> And this is where the problem arises. When gdbserver detaches from >> the inferior we pass the inferior the address from which it should >> resume. Due to the 32/64 bit confusion we tell the inferior to resume >> from the 32-bit $pc value, which is not within any valid mapping, and >> so, as soon as the inferior resumes, it segfaults. >> >> If we look at how GDB (not gdbserver) figures out its target >> description then we see an interesting difference. GDB doesn't try to >> read the executable. Instead GDB uses ptrace to query the thread's >> state, and uses this to figure out the if the thread is 32 or 64 bit. >> >> If we update gdbserver to do it the "GDB" way then the above problem >> is resolved, gdbserver now sees the process as 64-bit, and when we >> detach from the inferior we give it the correct 64-bit address, and >> the inferior no longer segfaults. >> >> Now, I could just update the gdbserver code, but better, I think, to >> share one copy of the code between GDB and gdbserver in gdb/nat/. >> That is what this commit does. >> >> The cores of x86_linux_read_description from gdbserver and >> x86_linux_nat_target::read_description from GDB are moved into a new >> file gdb/nat/x86-linux-tdesc.c and combined into a single function >> x86_linux_tdesc_for_tid which is called from each location. >> >> This new function does things the GDB way, the only changes are to >> allow for the sharing; we now have a callback function to call the >> first time that the xcr0 state is read, this allows for GDB and >> gdbserver to perform their own initialisation as needed, and >> additionally, the new function takes a pointer for where to cache the >> xcr0 value, this isn't needed for this commit, but will be useful in a >> later commit where gdbserver will want to read this cached xcr0 >> value. >> >> Another thing to note about this commit is how the functions >> i386_linux_read_description and amd64_linux_read_description are >> handled. For now I've left these function as implemented separately >> in GDB and gdbserver. I've moved the declarations of these functions >> into gdb/arch/{i386,amd64}-linux-tdesc.h, but the implementations are >> left where they are. >> >> A later commit in this series will make these functions shared too, >> but doing this is not trivial, so I've left that for a separate >> commit. Merging the declarations as I've done here ensures that >> everyone implements the function to the same API, and once these >> functions are shared (in a later commit) we'll want a shared >> declaration anyway. >> --- >> gdb/Makefile.in | 3 + >> gdb/amd64-linux-tdep.c | 1 + >> gdb/amd64-linux-tdep.h | 6 -- >> gdb/arch/amd64-linux-tdesc.h | 30 +++++++ >> gdb/arch/i386-linux-tdesc.h | 29 +++++++ >> gdb/configure.nat | 4 +- >> gdb/i386-linux-tdep.c | 1 + >> gdb/i386-linux-tdep.h | 3 - >> gdb/nat/x86-linux-tdesc.c | 124 ++++++++++++++++++++++++++++ >> gdb/nat/x86-linux-tdesc.h | 60 ++++++++++++++ >> gdb/x86-linux-nat.c | 91 ++++----------------- >> gdbserver/configure.srv | 2 + >> gdbserver/linux-amd64-ipa.cc | 1 + >> gdbserver/linux-i386-ipa.cc | 1 + >> gdbserver/linux-x86-low.cc | 151 +++++++++++------------------------ >> gdbserver/linux-x86-tdesc.cc | 2 + >> gdbserver/linux-x86-tdesc.h | 7 -- >> 17 files changed, 317 insertions(+), 199 deletions(-) >> create mode 100644 gdb/arch/amd64-linux-tdesc.h >> create mode 100644 gdb/arch/i386-linux-tdesc.h >> create mode 100644 gdb/nat/x86-linux-tdesc.c >> create mode 100644 gdb/nat/x86-linux-tdesc.h >> >> diff --git a/gdb/Makefile.in b/gdb/Makefile.in >> index 353e45b3cec..a24b2232daa 100644 >> --- a/gdb/Makefile.in >> +++ b/gdb/Makefile.in >> @@ -1549,8 +1549,10 @@ HFILES_NO_SRCDIR = \ >> arch/aarch64-insn.h \ >> arch/aarch64-mte-linux.h \ >> arch/aarch64-scalable-linux.h \ >> + arch/amd64-linux-tdesc.h \ >> arch/arc.h \ >> arch/arm.h \ >> + arch/i386-linux-tdesc.h \ >> arch/i386.h \ >> arch/loongarch.h \ >> arch/ppc-linux-common.h \ >> @@ -1606,6 +1608,7 @@ HFILES_NO_SRCDIR = \ >> nat/x86-gcc-cpuid.h \ >> nat/x86-linux.h \ >> nat/x86-linux-dregs.h \ >> + nat/x86-linux-tdesc.h \ >> python/py-event.h \ >> python/py-events.h \ >> python/py-stopevent.h \ >> diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c >> index 9d560ac4fbf..bcb9868e79e 100644 >> --- a/gdb/amd64-linux-tdep.c >> +++ b/gdb/amd64-linux-tdep.c >> @@ -41,6 +41,7 @@ >> #include "arch/amd64.h" >> #include "target-descriptions.h" >> #include "expop.h" >> +#include "arch/amd64-linux-tdesc.h" >> >> /* The syscall's XML filename for i386. */ >> #define XML_SYSCALL_FILENAME_AMD64 "syscalls/amd64-linux.xml" >> diff --git a/gdb/amd64-linux-tdep.h b/gdb/amd64-linux-tdep.h >> index 2003dcda78f..0ec49e7fe03 100644 >> --- a/gdb/amd64-linux-tdep.h >> +++ b/gdb/amd64-linux-tdep.h >> @@ -43,12 +43,6 @@ extern struct target_desc *tdesc_x32_linux; >> extern struct target_desc *tdesc_x32_avx_linux; >> extern struct target_desc *tdesc_x32_avx_avx512_linux; >> >> -/* Return the right amd64-linux target descriptions according to >> - XCR0_FEATURES_BIT and IS_X32. */ >> - >> -const target_desc *amd64_linux_read_description (uint64_t xcr0_features_bit, >> - bool is_x32); >> - >> /* Enum that defines the syscall identifiers for amd64 linux. >> Used for process record/replay, these will be translated into >> a gdb-canonical set of syscall ids in linux-record.c. */ >> diff --git a/gdb/arch/amd64-linux-tdesc.h b/gdb/arch/amd64-linux-tdesc.h >> new file mode 100644 >> index 00000000000..db425b60df6 >> --- /dev/null >> +++ b/gdb/arch/amd64-linux-tdesc.h >> @@ -0,0 +1,30 @@ >> +/* Target description related code for GNU/Linux x86-64. >> + >> + Copyright (C) 2024 Free Software Foundation, Inc. >> + >> + This file is part of GDB. >> + >> + This program is free software; you can redistribute it and/or modify >> + it under the terms of the GNU General Public License as published by >> + the Free Software Foundation; either version 3 of the License, or >> + (at your option) any later version. >> + >> + This program is distributed in the hope that it will be useful, >> + but WITHOUT ANY WARRANTY; without even the implied warranty of >> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the >> + GNU General Public License for more details. >> + >> + You should have received a copy of the GNU General Public License >> + along with this program. If not, see <http://www.gnu.org/licenses/>. */ >> + >> +#ifndef ARCH_AMD64_LINUX_TDESC_H >> +#define ARCH_AMD64_LINUX_TDESC_H >> + >> +struct target_desc; >> + >> +/* Return the AMD64 target descriptions corresponding to XCR0 and IS_X32. */ >> + >> +extern const target_desc *amd64_linux_read_description (uint64_t xcr0, >> + bool is_x32); >> + >> +#endif /* ARCH_AMD64_LINUX_TDESC_H */ >> diff --git a/gdb/arch/i386-linux-tdesc.h b/gdb/arch/i386-linux-tdesc.h >> new file mode 100644 >> index 00000000000..0b736337a75 >> --- /dev/null >> +++ b/gdb/arch/i386-linux-tdesc.h >> @@ -0,0 +1,29 @@ >> +/* Target description related code for GNU/Linux i386. >> + >> + Copyright (C) 2024 Free Software Foundation, Inc. >> + >> + This file is part of GDB. >> + >> + This program is free software; you can redistribute it and/or modify >> + it under the terms of the GNU General Public License as published by >> + the Free Software Foundation; either version 3 of the License, or >> + (at your option) any later version. >> + >> + This program is distributed in the hope that it will be useful, >> + but WITHOUT ANY WARRANTY; without even the implied warranty of >> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the >> + GNU General Public License for more details. >> + >> + You should have received a copy of the GNU General Public License >> + along with this program. If not, see <http://www.gnu.org/licenses/>. */ >> + >> +#ifndef ARCH_I386_LINUX_TDESC_H >> +#define ARCH_I386_LINUX_TDESC_H >> + >> +struct target_desc; >> + >> +/* Return the i386 target description corresponding to XCR0. */ >> + >> +extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); >> + >> +#endif /* ARCH_I386_LINUX_TDESC_H */ >> diff --git a/gdb/configure.nat b/gdb/configure.nat >> index 8b98511cef7..4bcc0696027 100644 >> --- a/gdb/configure.nat >> +++ b/gdb/configure.nat >> @@ -256,7 +256,7 @@ case ${gdb_host} in >> NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ >> nat/x86-xstate.o \ >> i386-linux-nat.o x86-linux-nat.o nat/linux-btrace.o \ >> - nat/x86-linux.o nat/x86-linux-dregs.o" >> + nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o" >> ;; >> ia64) >> # Host: Intel IA-64 running GNU/Linux >> @@ -322,7 +322,7 @@ case ${gdb_host} in >> NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ >> nat/x86-xstate.o amd64-nat.o amd64-linux-nat.o x86-linux-nat.o >> \ >> nat/linux-btrace.o \ >> - nat/x86-linux.o nat/x86-linux-dregs.o \ >> + nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o \ >> nat/amd64-linux-siginfo.o" >> ;; >> sparc) >> diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c >> index 44730f204db..78ebc99d3df 100644 >> --- a/gdb/i386-linux-tdep.c >> +++ b/gdb/i386-linux-tdep.c >> @@ -39,6 +39,7 @@ >> >> #include "i387-tdep.h" >> #include "gdbsupport/x86-xstate.h" >> +#include "arch/i386-linux-tdesc.h" >> >> /* The syscall's XML filename for i386. */ >> #define XML_SYSCALL_FILENAME_I386 "syscalls/i386-linux.xml" >> diff --git a/gdb/i386-linux-tdep.h b/gdb/i386-linux-tdep.h >> index 07593c6a8ec..e8691cd778e 100644 >> --- a/gdb/i386-linux-tdep.h >> +++ b/gdb/i386-linux-tdep.h >> @@ -55,9 +55,6 @@ extern void i386_linux_report_signal_info (struct gdbarch >> *gdbarch, >> struct ui_out *uiout, >> enum gdb_signal siggnal); >> >> -/* Return the target description according to XCR0. */ >> -extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); >> - >> extern int i386_linux_gregset_reg_offset[]; >> >> /* Return x86 siginfo type. */ >> diff --git a/gdb/nat/x86-linux-tdesc.c b/gdb/nat/x86-linux-tdesc.c >> new file mode 100644 >> index 00000000000..c2301e6a873 >> --- /dev/null >> +++ b/gdb/nat/x86-linux-tdesc.c >> @@ -0,0 +1,124 @@ >> +/* Target description related code for GNU/Linux x86 (i386 and x86-64). >> + >> + Copyright (C) 2024 Free Software Foundation, Inc. >> + >> + This file is part of GDB. >> + >> + This program is free software; you can redistribute it and/or modify >> + it under the terms of the GNU General Public License as published by >> + the Free Software Foundation; either version 3 of the License, or >> + (at your option) any later version. >> + >> + This program is distributed in the hope that it will be useful, >> + but WITHOUT ANY WARRANTY; without even the implied warranty of >> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the >> + GNU General Public License for more details. >> + >> + You should have received a copy of the GNU General Public License >> + along with this program. If not, see <http://www.gnu.org/licenses/>. */ >> + >> +#include "nat/x86-linux-tdesc.h" >> +#ifdef __x86_64__ >> +#include "arch/amd64.h" >> +#include "arch/amd64-linux-tdesc.h" >> +#endif >> +#include "arch/i386.h" >> +#include "arch/i386-linux-tdesc.h" >> + >> +#include "nat/x86-linux.h" >> +#include "nat/gdb_ptrace.h" >> +#include "nat/x86-xstate.h" >> + >> +#ifndef __x86_64__ >> +#include <sys/procfs.h> >> +#endif >> + >> +#include <sys/uio.h> >> +#include <elf.h> >> + >> +#ifndef IN_PROCESS_AGENT >> + >> +/* See nat/x86-linux-tdesc.h. */ >> + >> +const target_desc * >> +x86_linux_tdesc_for_tid (int tid, enum tribool *have_ptrace_getregset, >> + gdb::function_view<void (uint64_t)> xcr0_init_cb, >> + const char *error_msg, uint64_t *xcr0_storage) >> +{ >> +#ifdef __x86_64__ >> + >> + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); >> + bool is_64bit = arch_size.is_64bit (); >> + bool is_x32 = arch_size.is_x32 (); >> + >> + if (sizeof (void *) == 4 && is_64bit && !is_x32) >> + error ("%s", error_msg); >> + >> +#elif HAVE_PTRACE_GETFPXREGS >> + if (have_ptrace_getfpxregs == -1) >> + { >> + elf_fpxregset_t fpxregs; >> + >> + if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) >> + { >> + have_ptrace_getfpxregs = 0; >> + *have_ptrace_getregset = TRIBOOL_FALSE; > > We tried PTRACE_GETFPXREGS, does that really indicate > PTRACE_GETREGSET is not supported? I realize you just moved this code, > but it feels weird to me. Then again this is Linux specific and it might actually > be the case. Thanks for all your feedback on this series. I'm working through the comments, but wanted to reply to this point separately: I also wondered about this. I initially left if because, as you say, I'm just moving the code, and didn't want to change too much in one go. Given your feedback though I did trace this code back to this commit: 3a13a53b432d732be258d677b6afca969ce0d65f from April 2010. What's interesting is this part of the ChangeLog: (i386_linux_read_description): Set have_ptrace_getfpxregs and have_ptrace_getregset to 0 if ptrace PTRACE_GETFPXREGS failed. The function name has changed since then, but notice that the author specifically mentions their intention to set BOTH flags when PTRACE_GETFPXREGS fails. Based on that I'm inclined to leave this alone unless it can be shown that it's wrong. Thanks, Andrew > > >> + return i386_linux_read_description (X86_XSTATE_X87_MASK); >> + } >> + } >> +#endif >> + >> + if (*have_ptrace_getregset == TRIBOOL_UNKNOWN) >> + { >> + uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; >> + struct iovec iov; >> + >> + iov.iov_base = xstateregs; >> + iov.iov_len = sizeof (xstateregs); >> + >> + /* Check if PTRACE_GETREGSET works. */ >> + if (ptrace (PTRACE_GETREGSET, tid, >> + (unsigned int) NT_X86_XSTATE, &iov) < 0) >> + { >> + *have_ptrace_getregset = TRIBOOL_FALSE; >> + *xcr0_storage = 0; >> + } >> + else >> + { >> + *have_ptrace_getregset = TRIBOOL_TRUE; >> + >> + /* Get XCR0 from XSAVE extended state. */ >> + *xcr0_storage = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET >> + / sizeof (uint64_t))]; >> + >> +#ifdef __x86_64__ >> + /* No MPX on x32. */ >> + if (is_64bit && is_x32) >> + *xcr0_storage &= ~X86_XSTATE_MPX; >> +#endif /* __x86_64__ */ >> + >> + xcr0_init_cb (*xcr0_storage); > > > Needing a callback made me scratch my head. What could be the reason > that GDB and gdbserver really want to do different things here? > > Digging into the code made me see that GDB saves just the xsave_layout. > Gdbserver in addition saves xcr0, though I guess that might not even > be needed if you have a tdesc already (or the xsave_layout?). > And they are saving things in different locations (global in gdbserver, > member in GDB), but that could be done without a callback I think. > > I realize that changing these things is a bigger tasks, that you might > not want to do at all or mix with this series. But can we avoid > a callback? E.g. by adding another return value, that only gdbserver > uses for now? With a TODO comment explaining this? > > > >> + } >> + } >> + >> + /* Check the native XCR0 only if PTRACE_GETREGSET is available. If >> + PTRACE_GETREGSET is not available then set xcr0_features_bits to >> + zero so that the "no-features" descriptions are returned by the >> + switches below. */ >> + uint64_t xcr0_features_bits; >> + if (*have_ptrace_getregset == TRIBOOL_TRUE) >> + xcr0_features_bits = *xcr0_storage & X86_XSTATE_ALL_MASK; >> + else >> + xcr0_features_bits = 0; >> + >> +#ifdef __x86_64__ >> + if (is_64bit) >> + { >> + return amd64_linux_read_description (xcr0_features_bits, is_x32); >> + } > > Nit: I think we could just drop the brackets here. > >> + else >> +#endif >> + return i386_linux_read_description (xcr0_features_bits); >> +} >> + >> +#endif /* !IN_PROCESS_AGENT */ >> diff --git a/gdb/nat/x86-linux-tdesc.h b/gdb/nat/x86-linux-tdesc.h >> new file mode 100644 >> index 00000000000..8c481609876 >> --- /dev/null >> +++ b/gdb/nat/x86-linux-tdesc.h >> @@ -0,0 +1,60 @@ >> +/* Target description related code for GNU/Linux x86 (i386 and x86-64). >> + >> + Copyright (C) 2024 Free Software Foundation, Inc. >> + >> + This file is part of GDB. >> + >> + This program is free software; you can redistribute it and/or modify >> + it under the terms of the GNU General Public License as published by >> + the Free Software Foundation; either version 3 of the License, or >> + (at your option) any later version. >> + >> + This program is distributed in the hope that it will be useful, >> + but WITHOUT ANY WARRANTY; without even the implied warranty of >> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the >> + GNU General Public License for more details. >> + >> + You should have received a copy of the GNU General Public License >> + along with this program. If not, see <http://www.gnu.org/licenses/>. */ >> + >> +#ifndef NAT_X86_LINUX_TDESC_H >> +#define NAT_X86_LINUX_TDESC_H >> + >> +#include "gdbsupport/function-view.h" >> + >> +struct target_desc; >> + >> +/* Return the target description for Linux thread TID. >> + >> + When *HAVE_PTRACE_GETREGSET is TRIBOOL_UNKNOWN then the current >> value of >> + xcr0 is read using ptrace calls and stored into *XCR0_STORAGE. Then >> + XCR0_INIT_CB is called with the value of *XCR0_STORAGE and >> + *HAVE_PTRACE_GETREGSET is set to TRIBOOL_TRUE. >> + >> + If the attempt to read xcr0 using ptrace fails then *XCR0_STORAGE is set >> + to zero and *HAVE_PTRACE_GETREGSET is set to TRIBOOL_FALSE. >> + >> + The storage pointed to by XCR0_STORAGE must exist until the program >> + terminates, this storage is used to cache the xcr0 value. As such >> + XCR0_INIT_CB will only be called once if xcr0 is successfully read using >> + ptrace, or not at all if the ptrace call fails. >> + >> + This function returns a target description based on the extracted xcr0 >> + value along with other characteristics of the thread identified by TID. >> + >> + This function can return nullptr if we encounter a machine configuration >> + for which a target_desc cannot be created. Ideally this would not be >> + the case, we should be able to create a target description for every >> + possible machine configuration. See amd64_linux_read_description and >> + i386_linux_read_description for cases when nullptr might be >> + returned. >> + >> + ERROR_MSG is using in an error() call if we try to create a target >> + description for a 64-bit process but this is a 32-bit build of GDB. */ >> + >> +extern const target_desc * >> +x86_linux_tdesc_for_tid (int tid, enum tribool *have_ptrace_getregset, >> + gdb::function_view<void (uint64_t)> xcr0_init_cb, >> + const char *error_msg, uint64_t *xcr0_storage); >> + >> +#endif /* NAT_X86_LINUX_TDESC_H */ >> diff --git a/gdb/x86-linux-nat.c b/gdb/x86-linux-nat.c >> index 12965c7a21b..9fbccb28d21 100644 >> --- a/gdb/x86-linux-nat.c >> +++ b/gdb/x86-linux-nat.c >> @@ -38,6 +38,7 @@ >> #include "nat/x86-linux.h" >> #include "nat/x86-linux-dregs.h" >> #include "nat/linux-ptrace.h" >> +#include "nat/x86-linux-tdesc.h" >> >> /* linux_nat_target::low_new_fork implementation. */ >> >> @@ -92,90 +93,26 @@ x86_linux_nat_target::post_startup_inferior (ptid_t ptid) >> const struct target_desc * >> x86_linux_nat_target::read_description () >> { >> - int tid; >> - int is_64bit = 0; >> -#ifdef __x86_64__ >> - int is_x32; >> -#endif >> - static uint64_t xcr0; >> - uint64_t xcr0_features_bits; >> + static uint64_t xcr0_storage; >> >> if (inferior_ptid == null_ptid) >> return this->beneath ()->read_description (); >> >> - tid = inferior_ptid.pid (); >> - >> -#ifdef __x86_64__ >> - >> - x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); >> - is_64bit = arch_size.is_64bit (); >> - is_x32 = arch_size.is_x32 (); >> - >> -#elif HAVE_PTRACE_GETFPXREGS >> - if (have_ptrace_getfpxregs == -1) >> - { >> - elf_fpxregset_t fpxregs; >> - >> - if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) >> - { >> - have_ptrace_getfpxregs = 0; >> - have_ptrace_getregset = TRIBOOL_FALSE; >> - return i386_linux_read_description (X86_XSTATE_X87_MASK); >> - } >> - } >> -#endif >> - >> - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) >> - { >> - uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; >> - struct iovec iov; >> - >> - iov.iov_base = xstateregs; >> - iov.iov_len = sizeof (xstateregs); >> - >> - /* Check if PTRACE_GETREGSET works. */ >> - if (ptrace (PTRACE_GETREGSET, tid, >> - (unsigned int) NT_X86_XSTATE, &iov) < 0) >> - have_ptrace_getregset = TRIBOOL_FALSE; >> - else >> - { >> - have_ptrace_getregset = TRIBOOL_TRUE; >> - >> - /* Get XCR0 from XSAVE extended state. */ >> - xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET >> - / sizeof (uint64_t))]; >> - >> - m_xsave_layout = x86_fetch_xsave_layout (xcr0, x86_xsave_length ()); >> - } >> - } >> - >> - /* Check the native XCR0 only if PTRACE_GETREGSET is available. If >> - PTRACE_GETREGSET is not available then set xcr0_features_bits to >> - zero so that the "no-features" descriptions are returned by the >> - switches below. */ > > Should we s/switches/code? > >> - if (have_ptrace_getregset == TRIBOOL_TRUE) >> - xcr0_features_bits = xcr0 & X86_XSTATE_ALL_MASK; >> - else >> - xcr0_features_bits = 0; >> - >> - if (is_64bit) >> - { >> -#ifdef __x86_64__ >> - return amd64_linux_read_description (xcr0_features_bits, is_x32); >> -#endif >> - } >> - else >> - { >> - const struct target_desc * tdesc >> - = i386_linux_read_description (xcr0_features_bits); >> + int tid = inferior_ptid.pid (); >> >> - if (tdesc == NULL) >> - tdesc = i386_linux_read_description (X86_XSTATE_SSE_MASK); >> + const char *error_msg >> + = _("Can't debug 64-bit process with 32-bit GDB"); >> >> - return tdesc; >> - } >> + /* Callback that is triggered the first time x86_linux_tdesc_for_tid >> + reads the xcr0 register. Setup other bits of state. */ >> + auto cb = [&] (uint64_t xcr0) >> + { >> + this->m_xsave_layout >> + = x86_fetch_xsave_layout (xcr0, x86_xsave_length ()); >> + }; >> >> - gdb_assert_not_reached ("failed to return tdesc"); >> + return x86_linux_tdesc_for_tid (tid, &have_ptrace_getregset, cb, >> + error_msg, &xcr0_storage); >> } >> >> >> >> >> diff --git a/gdbserver/configure.srv b/gdbserver/configure.srv >> index 9e861a75088..7a2702d78bf 100644 >> --- a/gdbserver/configure.srv >> +++ b/gdbserver/configure.srv >> @@ -109,6 +109,7 @@ case "${gdbserver_host}" in >> srv_tgtobj="${srv_tgtobj} nat/linux-btrace.o" >> srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" >> srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" >> + srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" >> srv_linux_usrregs=yes >> srv_linux_regsets=yes >> srv_linux_thread_db=yes >> @@ -371,6 +372,7 @@ case "${gdbserver_host}" in >> srv_tgtobj="${srv_tgtobj} nat/linux-btrace.o" >> srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" >> srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" >> + srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" >> srv_tgtobj="${srv_tgtobj} nat/amd64-linux-siginfo.o" >> srv_linux_usrregs=yes # This is for i386 progs. >> srv_linux_regsets=yes >> diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc >> index a6346750f49..df5e6aca081 100644 >> --- a/gdbserver/linux-amd64-ipa.cc >> +++ b/gdbserver/linux-amd64-ipa.cc >> @@ -22,6 +22,7 @@ >> #include "tracepoint.h" >> #include "linux-x86-tdesc.h" >> #include "gdbsupport/x86-xstate.h" >> +#include "arch/amd64-linux-tdesc.h" >> >> /* fast tracepoints collect registers. */ >> >> diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc >> index 8f14e0937d4..aa346fc9bc3 100644 >> --- a/gdbserver/linux-i386-ipa.cc >> +++ b/gdbserver/linux-i386-ipa.cc >> @@ -22,6 +22,7 @@ >> #include "tracepoint.h" >> #include "linux-x86-tdesc.h" >> #include "gdbsupport/x86-xstate.h" >> +#include "arch/i386-linux-tdesc.h" >> >> /* GDB register numbers. */ >> >> diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc >> index 62612773a87..68d2f13537c 100644 >> --- a/gdbserver/linux-x86-low.cc >> +++ b/gdbserver/linux-x86-low.cc >> @@ -29,8 +29,11 @@ >> >> #ifdef __x86_64__ >> #include "nat/amd64-linux-siginfo.h" >> +#include "arch/amd64-linux-tdesc.h" >> #endif >> >> +#include "arch/i386-linux-tdesc.h" >> + >> #include "gdb_proc_service.h" >> /* Don't include elf/common.h if linux/elf.h got included by >> gdb_proc_service.h. */ >> @@ -46,6 +49,7 @@ >> #include "nat/x86-linux.h" >> #include "nat/x86-linux-dregs.h" >> #include "linux-x86-tdesc.h" >> +#include "nat/x86-linux-tdesc.h" >> >> #ifdef __x86_64__ >> static target_desc_up tdesc_amd64_linux_no_xml; >> @@ -831,32 +835,20 @@ x86_target::low_siginfo_fixup (siginfo_t *ptrace, >> gdb_byte *inf, int direction) >> >> >> >> static int use_xml; >> >> +/* Cached xcr0 value. This is initialised the first time >> + x86_linux_read_description is called. */ >> + >> +static uint64_t xcr0_storage; > > We already cache it in fp-i387.cc. It get's set by the callback > you added. I think we should avoid having it twice. > > >> /* Get Linux/x86 target description from running target. */ >> >> static const struct target_desc * >> x86_linux_read_description (void) > > Not strictly necessary, but we could remove the void while we are touching this. > This is a .cc file. > > Thanks, > Felix > Intel Deutschland GmbH > Registered Address: Am Campeon 10, 85579 Neubiberg, Germany > Tel: +49 89 99 8853-0, www.intel.de <http://www.intel.de> > Managing Directors: Christin Eisenschmid, Sharon Heck, Tiffany Doon Silva > Chairperson of the Supervisory Board: Nicole Lau > Registered Office: Munich > Commercial Register: Amtsgericht Muenchen HRB 186928 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv5 08/11] gdb/arch: assert that X86_XSTATE_MPX is not set for x32 2024-04-26 15:01 ` [PATCHv5 00/11] " Andrew Burgess ` (6 preceding siblings ...) 2024-04-26 15:01 ` [PATCHv5 07/11] gdb/gdbserver: share some code relating to target description creation Andrew Burgess @ 2024-04-26 15:01 ` Andrew Burgess 2024-04-29 14:34 ` Willgerodt, Felix 2024-04-26 15:01 ` [PATCHv5 09/11] gdbserver: update target description creation for x86/linux Andrew Burgess ` (3 subsequent siblings) 11 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-04-26 15:01 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, John Baldwin While rebasing this series past this commit: commit 4bb20a6244b7091a9a7a2ae35dfbd7e8db27550a Date: Wed Mar 20 04:13:18 2024 -0700 gdbserver: Clear X86_XSTATE_MPX bits in xcr0 on x32 I worried that there could be other paths that might result in an xcr0 value which has X86_XSTATE_MPX set in x32 mode. As everyone eventually calls amd64_create_target_description to build their target description, I figured we could assert in here that if X86_XSTATE_MPX is set then we should not be an x32 target, this will uncover any other bugs in this area. I'm not currently able to build/run any x32 binaries, so I have no way to test this, but the author of commit 4bb20a6244b7091 did test this series with that assert in place and didn't see any problems. Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=31511 --- gdb/arch/amd64.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-) diff --git a/gdb/arch/amd64.c b/gdb/arch/amd64.c index cb9683c6931..94d55d72a4e 100644 --- a/gdb/arch/amd64.c +++ b/gdb/arch/amd64.c @@ -65,8 +65,12 @@ amd64_create_target_description (uint64_t xcr0, bool is_x32, bool is_linux, if (xcr0 & X86_XSTATE_AVX) regnum = create_feature_i386_64bit_avx (tdesc.get (), regnum); - if ((xcr0 & X86_XSTATE_MPX) && !is_x32) - regnum = create_feature_i386_64bit_mpx (tdesc.get (), regnum); + if (xcr0 & X86_XSTATE_MPX) + { + /* MPX is not available on x32. */ + gdb_assert (!is_x32); + regnum = create_feature_i386_64bit_mpx (tdesc.get (), regnum); + } if (xcr0 & X86_XSTATE_AVX512) regnum = create_feature_i386_64bit_avx512 (tdesc.get (), regnum); -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 08/11] gdb/arch: assert that X86_XSTATE_MPX is not set for x32 2024-04-26 15:01 ` [PATCHv5 08/11] gdb/arch: assert that X86_XSTATE_MPX is not set for x32 Andrew Burgess @ 2024-04-29 14:34 ` Willgerodt, Felix 2024-05-07 16:08 ` Andrew Burgess 0 siblings, 1 reply; 116+ messages in thread From: Willgerodt, Felix @ 2024-04-29 14:34 UTC (permalink / raw) To: Andrew Burgess, gdb-patches; +Cc: John Baldwin > -----Original Message----- > From: Andrew Burgess <aburgess@redhat.com> > Sent: Freitag, 26. April 2024 17:02 > To: gdb-patches@sourceware.org > Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix > <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> > Subject: [PATCHv5 08/11] gdb/arch: assert that X86_XSTATE_MPX is not set for > x32 > > While rebasing this series past this commit: > > commit 4bb20a6244b7091a9a7a2ae35dfbd7e8db27550a > Date: Wed Mar 20 04:13:18 2024 -0700 > > gdbserver: Clear X86_XSTATE_MPX bits in xcr0 on x32 > > I worried that there could be other paths that might result in an xcr0 > value which has X86_XSTATE_MPX set in x32 mode. As everyone > eventually calls amd64_create_target_description to build their target > description, I figured we could assert in here that if X86_XSTATE_MPX > is set then we should not be an x32 target, this will uncover any > other bugs in this area. > > I'm not currently able to build/run any x32 binaries, so I have no way > to test this, but the author of commit 4bb20a6244b7091 did test this > series with that assert in place and didn't see any problems. > > Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=31511 > --- > gdb/arch/amd64.c | 8 ++++++-- > 1 file changed, 6 insertions(+), 2 deletions(-) > GDB was also checking this somewhere when I reviewed H.J.'s patch. Though it doesn't hurt to have it here in my eyes. Approved-By: Felix Willgerodt <felix.willgerodt@intel.com> Thanks, Felix Intel Deutschland GmbH Registered Address: Am Campeon 10, 85579 Neubiberg, Germany Tel: +49 89 99 8853-0, www.intel.de <http://www.intel.de> Managing Directors: Christin Eisenschmid, Sharon Heck, Tiffany Doon Silva Chairperson of the Supervisory Board: Nicole Lau Registered Office: Munich Commercial Register: Amtsgericht Muenchen HRB 186928 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 08/11] gdb/arch: assert that X86_XSTATE_MPX is not set for x32 2024-04-29 14:34 ` Willgerodt, Felix @ 2024-05-07 16:08 ` Andrew Burgess 0 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-07 16:08 UTC (permalink / raw) To: Willgerodt, Felix, gdb-patches; +Cc: John Baldwin "Willgerodt, Felix" <felix.willgerodt@intel.com> writes: >> -----Original Message----- >> From: Andrew Burgess <aburgess@redhat.com> >> Sent: Freitag, 26. April 2024 17:02 >> To: gdb-patches@sourceware.org >> Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix >> <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> >> Subject: [PATCHv5 08/11] gdb/arch: assert that X86_XSTATE_MPX is not set for >> x32 >> >> While rebasing this series past this commit: >> >> commit 4bb20a6244b7091a9a7a2ae35dfbd7e8db27550a >> Date: Wed Mar 20 04:13:18 2024 -0700 >> >> gdbserver: Clear X86_XSTATE_MPX bits in xcr0 on x32 >> >> I worried that there could be other paths that might result in an xcr0 >> value which has X86_XSTATE_MPX set in x32 mode. As everyone >> eventually calls amd64_create_target_description to build their target >> description, I figured we could assert in here that if X86_XSTATE_MPX >> is set then we should not be an x32 target, this will uncover any >> other bugs in this area. >> >> I'm not currently able to build/run any x32 binaries, so I have no way >> to test this, but the author of commit 4bb20a6244b7091 did test this >> series with that assert in place and didn't see any problems. >> >> Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=31511 >> --- >> gdb/arch/amd64.c | 8 ++++++-- >> 1 file changed, 6 insertions(+), 2 deletions(-) >> > > GDB was also checking this somewhere when I reviewed H.J.'s patch. > Though it doesn't hurt to have it here in my eyes. > > Approved-By: Felix Willgerodt <felix.willgerodt@intel.com> As this commit is only loosely tied to the rest of the series I've gone ahead and pushed this to upstream. Thanks, Andrew ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv5 09/11] gdbserver: update target description creation for x86/linux 2024-04-26 15:01 ` [PATCHv5 00/11] " Andrew Burgess ` (7 preceding siblings ...) 2024-04-26 15:01 ` [PATCHv5 08/11] gdb/arch: assert that X86_XSTATE_MPX is not set for x32 Andrew Burgess @ 2024-04-26 15:01 ` Andrew Burgess 2024-04-29 14:35 ` Willgerodt, Felix 2024-04-26 15:01 ` [PATCHv5 10/11] gdb: move xcr0 == 0 check into i386_linux_core_read_description Andrew Burgess ` (2 subsequent siblings) 11 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-04-26 15:01 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, John Baldwin This commit is part of a series which aims to share more of the target description creation between GDB and gdbserver for x86/Linux. After some refactoring earlier in this series the shared x86_linux_tdesc_for_tid function was added into nat/x86-linux-tdesc.c. However, this function still relies on amd64_linux_read_description and i386_linux_read_description which are implemented separately for both gdbserver and GDB. Given that at their core, all these functions do is: 1. take an xcr0 value as input, 2. mask out some feature bits, 3. look for a cached pre-generated target description and return it if found, 4. if no cached target description is found then call either amd64_create_target_description or i386_create_target_description to create a new target description, which is then added to the cache. Return the newly created target description. The inner functions amd64_create_target_description and i386_create_target_description are already shared between GDB and gdbserver (in the gdb/arch/ directory), so the only thing that the *_read_description functions really do is add the caching layer, and it feels like this really could be shared. However, we have a small problem. On the GDB side we create target descriptions using a different set of cpu features than on the gdbserver side! This means that for the exact same target, we might get a different target description when using native GDB vs using gdbserver. This surely feels like a mistake, I would expect to get the same target description on each. The table below shows the number of possible different target descriptions that we can create on the GDB side vs on the gdbserver side for each target type: | GDB | gdbserver ------|-----|---------- i386 | 64 | 7 amd64 | 32 | 7 x32 | 16 | 7 So in theory, all I want to do is move the GDB version of *_read_description into the arch/ directory and have gdbserver use that, then both GDB and gdbserver would be able to create any of the possible target descriptions. Unfortunately it's a little more complex than that due to the in process agent (IPA). When the IPA is in use, gdbserver sends a target description index to the IPA, and the IPA uses this to find the correct target description to use. ** START OF AN ASIDE ** Back in the day I suspect this approach made perfect sense. However since this commit: commit a8806230241d201f808d856eaae4d44088117b0c Date: Thu Dec 7 17:07:01 2017 +0000 Initialize target description early in IPA I think passing the index is now more trouble than its worth. We used to pass the index, and then use that index to lookup which target description to instantiate and use. However, the above commit fixed an issue where we can't call malloc() within (certain parts of) the IPA (apparently), so instead we now pre-compute _every_ possible target description within the IPA. The index is now only used to lookup which of the (many) pre-computed target descriptions to use. It would (I think) have been easier all around if the IPA just self-inspected, figured out its own xcr0 value, and used that to create the one target description that is required. So long as the xcr0 to target description code is shared (at compile time) with gdbserver, then we can be sure that the IPA will derive the same target description as gdbserver, and we would avoid all this index passing business, which has made this commit so very, very painful. ** END OF AN ASIDE ** Currently then for x86/linux, gdbserver sends a number between 0 and 7 to the IPA, and the IPA uses this to create a target description. However, I am proposing that gdbserver should now create one of (up to) 64 different target descriptions for i386, so this 0 to 7 index isn't going to be good enough any more (amd64 and x32 have slightly fewer possible target descriptions, but still more than 8, so the problem is the same). For a while I wondered if I was going to have to try and find some backward compatible solution for this mess. But after seeing how lightly the IPA is actually documented, I wonder if it is not the case that there is a tight coupling between a version of gdbserver and a version of the IPA? At least I'm hoping so. In this commit I have thrown out the old IPA target description index numbering scheme, and switched to a completely new numbering scheme. Instead of the index that is passed being arbitrary, the index is instead calculated from the set of cpu features that are present on the target. Within the IPA we can then reverse this logic to recreate the xcr0 value based on the index, and from the xcr0 value we can choose the correct target description. With the gdbserver to IPA numbering scheme issue resolved I have then update the gdbserver versions of amd64_linux_read_description and i386_linux_read_description so that they create target descriptions using the same set of cpu features as GDB itself. After this gdbserver should now always come up with the same target description as GDB does on any x86/Linux target. This commit does not introduce any new code sharing between GDB and gdbserver as previous commits in this series have done. Instead this commit is all about bringing GDB and gdbserver into alignment functionally so that the next commit(s) can merge the GDB and gdbserver versions of these functions. Approved-By: John Baldwin <jhb@FreeBSD.org> --- gdbserver/linux-amd64-ipa.cc | 43 +---- gdbserver/linux-i386-ipa.cc | 23 +-- gdbserver/linux-x86-low.cc | 15 +- gdbserver/linux-x86-tdesc.cc | 316 +++++++++++++++++++++++++---------- gdbserver/linux-x86-tdesc.h | 49 +++--- 5 files changed, 278 insertions(+), 168 deletions(-) diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index df5e6aca081..0c80812cc6f 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -164,47 +164,21 @@ supply_static_tracepoint_registers (struct regcache *regcache, #endif /* HAVE_UST */ -#if !defined __ILP32__ -/* Map the tdesc index to xcr0 mask. */ -static uint64_t idx2mask[X86_TDESC_LAST] = { - X86_XSTATE_X87_MASK, - X86_XSTATE_SSE_MASK, - X86_XSTATE_AVX_MASK, - X86_XSTATE_MPX_MASK, - X86_XSTATE_AVX_MPX_MASK, - X86_XSTATE_AVX_AVX512_MASK, - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, -}; -#endif - /* Return target_desc to use for IPA, given the tdesc index passed by gdbserver. */ const struct target_desc * get_ipa_tdesc (int idx) { - if (idx >= X86_TDESC_LAST) - { - internal_error ("unknown ipa tdesc index: %d", idx); - } + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); #if defined __ILP32__ - switch (idx) - { - case X86_TDESC_SSE: - return amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); - case X86_TDESC_AVX: - return amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); - case X86_TDESC_AVX_AVX512: - return amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, true); - default: - break; - } + bool is_x32 = true; #else - return amd64_linux_read_description (idx2mask[idx], false); + bool is_x32 = false; #endif - internal_error ("unknown ipa tdesc index: %d", idx); + return amd64_linux_read_description (xcr0, is_x32); } /* Allocate buffer for the jump pads. The branch instruction has a @@ -272,11 +246,10 @@ void initialize_low_tracepoint (void) { #if defined __ILP32__ - amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); - amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); - amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, true); + for (auto i = 0; i < x86_linux_x32_tdesc_count (); i++) + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), true); #else - for (auto i = 0; i < X86_TDESC_LAST; i++) - amd64_linux_read_description (idx2mask[i], false); + for (auto i = 0; i < x86_linux_amd64_tdesc_count (); i++) + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), false); #endif } diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc index aa346fc9bc3..c1c3152fb04 100644 --- a/gdbserver/linux-i386-ipa.cc +++ b/gdbserver/linux-i386-ipa.cc @@ -245,28 +245,15 @@ initialize_fast_tracepoint_trampoline_buffer (void) } } -/* Map the tdesc index to xcr0 mask. */ -static uint64_t idx2mask[X86_TDESC_LAST] = { - X86_XSTATE_X87_MASK, - X86_XSTATE_SSE_MASK, - X86_XSTATE_AVX_MASK, - X86_XSTATE_MPX_MASK, - X86_XSTATE_AVX_MPX_MASK, - X86_XSTATE_AVX_AVX512_MASK, - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, -}; - /* Return target_desc to use for IPA, given the tdesc index passed by gdbserver. */ const struct target_desc * get_ipa_tdesc (int idx) { - if (idx >= X86_TDESC_LAST) - { - internal_error ("unknown ipa tdesc index: %d", idx); - } - return i386_linux_read_description (idx2mask[idx]); + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); + + return i386_linux_read_description (xcr0); } /* Allocate buffer for the jump pads. On i386, we can reach an arbitrary @@ -288,6 +275,6 @@ void initialize_low_tracepoint (void) { initialize_fast_tracepoint_trampoline_buffer (); - for (auto i = 0; i < X86_TDESC_LAST; i++) - i386_linux_read_description (idx2mask[i]); + for (auto i = 0; i < x86_linux_i386_tdesc_count (); i++) + i386_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i)); } diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 68d2f13537c..6e23a53118b 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -2882,14 +2882,17 @@ x86_target::get_ipa_tdesc_idx () struct regcache *regcache = get_thread_regcache (current_thread, 0); const struct target_desc *tdesc = regcache->tdesc; + if (!use_xml) + { + if (tdesc == tdesc_i386_linux_no_xml.get () #ifdef __x86_64__ - return amd64_get_ipa_tdesc_idx (tdesc); -#endif - - if (tdesc == tdesc_i386_linux_no_xml.get ()) - return X86_TDESC_SSE; + || tdesc == tdesc_amd64_linux_no_xml.get () +#endif /* __x86_64__ */ + ) + return x86_linux_xcr0_to_tdesc_idx (X86_XSTATE_SSE_MASK); + } - return i386_get_ipa_tdesc_idx (tdesc); + return x86_linux_xcr0_to_tdesc_idx (xcr0_storage); } /* The linux target ops object. */ diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc index af3735aa895..5e12526bf17 100644 --- a/gdbserver/linux-x86-tdesc.cc +++ b/gdbserver/linux-x86-tdesc.cc @@ -28,96 +28,278 @@ #include "x86-tdesc.h" #include "arch/i386-linux-tdesc.h" -/* Return the right x86_linux_tdesc index for a given XCR0. Return - X86_TDESC_LAST if can't find a match. */ +/* A structure used to describe a single cpu feature that might, or might + not, be checked for when creating a target description for one of i386, + amd64, or x32. */ -static enum x86_linux_tdesc -xcr0_to_tdesc_idx (uint64_t xcr0, bool is_x32) +struct x86_tdesc_feature { + /* The cpu feature mask. This is a mask against an xcr0 value. */ + uint64_t feature; + + /* Is this feature checked when creating an i386 target description. */ + bool is_i386; + + /* Is this feature checked when creating an amd64 target description. */ + bool is_amd64; + + /* Is this feature checked when creating an x32 target description. */ + bool is_x32; +}; + +/* A constant table that describes all of the cpu features that are + checked when building a target description for i386, amd64, or x32. */ + +static constexpr x86_tdesc_feature x86_linux_all_tdesc_features[] = { + /* Feature, i386, amd64, x32. */ + { X86_XSTATE_PKRU, true, true, true }, + { X86_XSTATE_AVX512, true, true, true }, + { X86_XSTATE_AVX, true, true, true }, + { X86_XSTATE_MPX, true, true, false }, + { X86_XSTATE_SSE, true, false, false }, + { X86_XSTATE_X87, true, false, false } +}; + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an i386 target description. */ + +static constexpr uint64_t +x86_linux_i386_tdesc_feature_mask () { - if (xcr0 & X86_XSTATE_PKRU) - { - if (is_x32) - { - /* No x32 MPX and PKU, fall back to avx_avx512. */ - return X86_TDESC_AVX_AVX512; - } - else - return X86_TDESC_AVX_MPX_AVX512_PKU; - } - else if (xcr0 & X86_XSTATE_AVX512) - return X86_TDESC_AVX_AVX512; - else if ((xcr0 & X86_XSTATE_AVX_MPX_MASK) == X86_XSTATE_AVX_MPX_MASK) + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_i386) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an amd64 target description. */ + +static constexpr uint64_t +x86_linux_amd64_tdesc_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_amd64) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an x32 target description. */ + +static constexpr uint64_t +x86_linux_x32_tdesc_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_x32) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an i386 target description. */ + +static constexpr int +x86_linux_i386_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_i386) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an amd64 target description. */ + +static constexpr int +x86_linux_amd64_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_amd64) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an x32 target description. */ + +static constexpr int +x86_linux_x32_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_x32) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +#ifdef IN_PROCESS_AGENT + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_amd64_tdesc_count () +{ + return x86_linux_amd64_tdesc_count_1 (); +} + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_x32_tdesc_count () +{ + return x86_linux_x32_tdesc_count_1 (); +} + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_i386_tdesc_count () +{ + return x86_linux_i386_tdesc_count_1 (); +} + +#endif /* IN_PROCESS_AGENT */ + +/* Convert an xcr0 value into an integer. The integer will be passed to + the in-process-agent where it will then be passed to + x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ + +int +x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0) +{ + /* The following table shows which features are checked for when creating + the target descriptions (see nat/x86-linux-tdesc.c), the feature order + represents the bit order within the generated index number. + + i386 | x87 sse mpx avx avx512 pkru + amd64 | mpx avx avx512 pkru + i32 | avx avx512 pkru + + The features are ordered so that for each mode (i386, amd64, i32) the + generated index will form a continuous range. */ + + int idx = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) { - if (is_x32) /* No MPX on x32. */ - return X86_TDESC_AVX; - else - return X86_TDESC_AVX_MPX; + if ((xcr0 & x86_linux_all_tdesc_features[i].feature) + == x86_linux_all_tdesc_features[i].feature) + idx |= (1 << i); } - else if (xcr0 & X86_XSTATE_MPX) + + return idx; +} + + +#ifdef IN_PROCESS_AGENT + +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +uint64_t +x86_linux_tdesc_idx_to_xcr0 (int idx) +{ + uint64_t xcr0 = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) { - if (is_x32) /* No MPX on x32. */ - return X86_TDESC_AVX; - else - return X86_TDESC_MPX; + if ((idx & (1 << i)) != 0) + xcr0 |= x86_linux_all_tdesc_features[i].feature; } - else if (xcr0 & X86_XSTATE_AVX) - return X86_TDESC_AVX; - else if (xcr0 & X86_XSTATE_SSE) - return X86_TDESC_SSE; - else if (xcr0 & X86_XSTATE_X87) - return X86_TDESC_MMX; - else - return X86_TDESC_LAST; + + return xcr0; } +#endif /* IN_PROCESS_AGENT */ + #if defined __i386__ || !defined IN_PROCESS_AGENT -static struct target_desc *i386_tdescs[X86_TDESC_LAST] = { }; +/* A cache of all possible i386 target descriptions. */ -/* Return the target description according to XCR0. */ +static struct target_desc *i386_tdescs[x86_linux_i386_tdesc_count_1 ()] = { }; + +/* See nat/x86-linux-tdesc.h. */ const struct target_desc * i386_linux_read_description (uint64_t xcr0) { - enum x86_linux_tdesc idx = xcr0_to_tdesc_idx (xcr0, false); + xcr0 &= x86_linux_i386_tdesc_feature_mask (); + int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - if (idx == X86_TDESC_LAST) - return NULL; + gdb_assert (idx >= 0 && idx < x86_linux_i386_tdesc_count_1 ()); - struct target_desc **tdesc = &i386_tdescs[idx]; + target_desc **tdesc = &i386_tdescs[idx]; - if (*tdesc == NULL) + if (*tdesc == nullptr) { *tdesc = i386_create_target_description (xcr0, true, false); init_target_desc (*tdesc, i386_expedite_regs); } - return *tdesc;; + return *tdesc; } #endif #ifdef __x86_64__ -static target_desc *amd64_tdescs[X86_TDESC_LAST] = { }; -static target_desc *x32_tdescs[X86_TDESC_LAST] = { }; +/* A cache of all possible amd64 target descriptions. */ + +static target_desc *amd64_tdescs[x86_linux_amd64_tdesc_count_1 ()] = { }; + +/* A cache of all possible x32 target descriptions. */ + +static target_desc *x32_tdescs[x86_linux_x32_tdesc_count_1 ()] = { }; + +/* See nat/x86-linux-tdesc.h. */ const struct target_desc * amd64_linux_read_description (uint64_t xcr0, bool is_x32) { - enum x86_linux_tdesc idx = xcr0_to_tdesc_idx (xcr0, is_x32); + if (is_x32) + xcr0 &= x86_linux_x32_tdesc_feature_mask (); + else + xcr0 &= x86_linux_amd64_tdesc_feature_mask (); + + int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - if (idx == X86_TDESC_LAST) - return NULL; + if (is_x32) + gdb_assert (idx >= 0 && idx < x86_linux_x32_tdesc_count_1 ()); + else + gdb_assert (idx >= 0 && idx < x86_linux_amd64_tdesc_count_1 ()); - struct target_desc **tdesc = NULL; + target_desc **tdesc = nullptr; if (is_x32) tdesc = &x32_tdescs[idx]; else tdesc = &amd64_tdescs[idx]; - if (*tdesc == NULL) + if (*tdesc == nullptr) { *tdesc = amd64_create_target_description (xcr0, is_x32, true, true); @@ -127,39 +309,3 @@ amd64_linux_read_description (uint64_t xcr0, bool is_x32) } #endif - -#ifndef IN_PROCESS_AGENT - -int -i386_get_ipa_tdesc_idx (const struct target_desc *tdesc) -{ - for (int i = 0; i < X86_TDESC_LAST; i++) - { - if (tdesc == i386_tdescs[i]) - return i; - } - - /* If none tdesc is found, return the one with minimum features. */ - return X86_TDESC_MMX; -} - -#if defined __x86_64__ -int -amd64_get_ipa_tdesc_idx (const struct target_desc *tdesc) -{ - for (int i = 0; i < X86_TDESC_LAST; i++) - { - if (tdesc == amd64_tdescs[i]) - return i; - } - for (int i = 0; i < X86_TDESC_LAST; i++) - { - if (tdesc == x32_tdescs[i]) - return i; - } - - return X86_TDESC_SSE; -} - -#endif -#endif diff --git a/gdbserver/linux-x86-tdesc.h b/gdbserver/linux-x86-tdesc.h index 576aaf5e165..3d6e0e51833 100644 --- a/gdbserver/linux-x86-tdesc.h +++ b/gdbserver/linux-x86-tdesc.h @@ -21,29 +21,30 @@ #ifndef GDBSERVER_LINUX_X86_TDESC_H #define GDBSERVER_LINUX_X86_TDESC_H -/* Note: since IPA obviously knows what ABI it's running on (i386 vs x86_64 - vs x32), it's sufficient to pass only the register set here. This, - together with the ABI known at IPA compile time, maps to a tdesc. */ - -enum x86_linux_tdesc { - X86_TDESC_MMX = 0, - X86_TDESC_SSE = 1, - X86_TDESC_AVX = 2, - X86_TDESC_MPX = 3, - X86_TDESC_AVX_MPX = 4, - X86_TDESC_AVX_AVX512 = 5, - X86_TDESC_AVX_MPX_AVX512_PKU = 6, - X86_TDESC_LAST = 7, -}; - -#if defined __i386__ || !defined IN_PROCESS_AGENT -int i386_get_ipa_tdesc_idx (const struct target_desc *tdesc); -#endif - -#if defined __x86_64__ && !defined IN_PROCESS_AGENT -int amd64_get_ipa_tdesc_idx (const struct target_desc *tdesc); -#endif - -const struct target_desc *i386_get_ipa_tdesc (int idx); +/* Convert an xcr0 value into an integer. The integer will be passed to + the in-process-agent where it will then be passed to + x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ + +extern int x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0); + +#ifdef IN_PROCESS_AGENT + +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +extern uint64_t x86_linux_tdesc_idx_to_xcr0 (int idx); + +/* Within the in-process-agent we need to pre-initialise all of the target + descriptions, to do this we need to know how many target descriptions + there are for each different target type. These functions return the + target description count for the relevant target. */ + +extern int x86_linux_amd64_tdesc_count (); +extern int x86_linux_x32_tdesc_count (); +extern int x86_linux_i386_tdesc_count (); + + +#endif /* IN_PROCESS_AGENT */ #endif /* GDBSERVER_LINUX_X86_TDESC_H */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 09/11] gdbserver: update target description creation for x86/linux 2024-04-26 15:01 ` [PATCHv5 09/11] gdbserver: update target description creation for x86/linux Andrew Burgess @ 2024-04-29 14:35 ` Willgerodt, Felix 2024-05-07 14:24 ` Andrew Burgess 0 siblings, 1 reply; 116+ messages in thread From: Willgerodt, Felix @ 2024-04-29 14:35 UTC (permalink / raw) To: Andrew Burgess, gdb-patches; +Cc: John Baldwin > -----Original Message----- > From: Andrew Burgess <aburgess@redhat.com> > Sent: Freitag, 26. April 2024 17:02 > To: gdb-patches@sourceware.org > Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix > <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> > Subject: [PATCHv5 09/11] gdbserver: update target description creation for > x86/linux > > This commit is part of a series which aims to share more of the target > description creation between GDB and gdbserver for x86/Linux. > > After some refactoring earlier in this series the shared > x86_linux_tdesc_for_tid function was added into nat/x86-linux-tdesc.c. > However, this function still relies on amd64_linux_read_description > and i386_linux_read_description which are implemented separately for > both gdbserver and GDB. Given that at their core, all these functions > do is: > > 1. take an xcr0 value as input, > 2. mask out some feature bits, > 3. look for a cached pre-generated target description and return it > if found, > 4. if no cached target description is found then call either > amd64_create_target_description or > i386_create_target_description to create a new target > description, which is then added to the cache. Return the newly > created target description. > > The inner functions amd64_create_target_description and > i386_create_target_description are already shared between GDB and > gdbserver (in the gdb/arch/ directory), so the only thing that > the *_read_description functions really do is add the caching layer, > and it feels like this really could be shared. > > However, we have a small problem. > > On the GDB side we create target descriptions using a different set of > cpu features than on the gdbserver side! This means that for the > exact same target, we might get a different target description when > using native GDB vs using gdbserver. This surely feels like a > mistake, I would expect to get the same target description on each. > > The table below shows the number of possible different target > descriptions that we can create on the GDB side vs on the gdbserver > side for each target type: > > | GDB | gdbserver > ------|-----|---------- > i386 | 64 | 7 > amd64 | 32 | 7 > x32 | 16 | 7 I would have somehow expected to have the highest number for 64-bit CPUs. As it should include 32-bit and x32. Do you know why that isn't the case and 32-bit has double? Or is my problem that this is reflecting GDB code, which shares stuff between amd64 and i386? Maybe this is also solved with the comments I made in code below. > So in theory, all I want to do is move the GDB version > of *_read_description into the arch/ directory and have gdbserver use > that, then both GDB and gdbserver would be able to create any of the > possible target descriptions. > > Unfortunately it's a little more complex than that due to the in > process agent (IPA). > > When the IPA is in use, gdbserver sends a target description index to > the IPA, and the IPA uses this to find the correct target description > to use. > > ** START OF AN ASIDE ** > > Back in the day I suspect this approach made perfect sense. However > since this commit: > > commit a8806230241d201f808d856eaae4d44088117b0c > Date: Thu Dec 7 17:07:01 2017 +0000 > > Initialize target description early in IPA > > I think passing the index is now more trouble than its worth. > > We used to pass the index, and then use that index to lookup which > target description to instantiate and use. However, the above commit > fixed an issue where we can't call malloc() within (certain parts of) > the IPA (apparently), so instead we now pre-compute _every_ possible > target description within the IPA. The index is now only used to > lookup which of the (many) pre-computed target descriptions to use. > > It would (I think) have been easier all around if the IPA just > self-inspected, figured out its own xcr0 value, and used that to > create the one target description that is required. So long as the > xcr0 to target description code is shared (at compile time) with > gdbserver, then we can be sure that the IPA will derive the same > target description as gdbserver, and we would avoid all this index > passing business, which has made this commit so very, very painful. > > ** END OF AN ASIDE ** > > Currently then for x86/linux, gdbserver sends a number between 0 and 7 > to the IPA, and the IPA uses this to create a target description. > > However, I am proposing that gdbserver should now create one of (up > to) 64 different target descriptions for i386, so this 0 to 7 index > isn't going to be good enough any more (amd64 and x32 have slightly > fewer possible target descriptions, but still more than 8, so the > problem is the same). > > For a while I wondered if I was going to have to try and find some > backward compatible solution for this mess. But after seeing how > lightly the IPA is actually documented, I wonder if it is not the case > that there is a tight coupling between a version of gdbserver and a > version of the IPA? At least I'm hoping so. > > In this commit I have thrown out the old IPA target description index > numbering scheme, and switched to a completely new numbering scheme. > Instead of the index that is passed being arbitrary, the index is > instead calculated from the set of cpu features that are present on > the target. Within the IPA we can then reverse this logic to recreate > the xcr0 value based on the index, and from the xcr0 value we can > choose the correct target description. > > With the gdbserver to IPA numbering scheme issue resolved I have then > update the gdbserver versions of amd64_linux_read_description and > i386_linux_read_description so that they create target descriptions > using the same set of cpu features as GDB itself. > > After this gdbserver should now always come up with the same target > description as GDB does on any x86/Linux target. > > This commit does not introduce any new code sharing between GDB and > gdbserver as previous commits in this series have done. Instead this > commit is all about bringing GDB and gdbserver into alignment > functionally so that the next commit(s) can merge the GDB and > gdbserver versions of these functions. > > Approved-By: John Baldwin <jhb@FreeBSD.org> > --- > gdbserver/linux-amd64-ipa.cc | 43 +---- > gdbserver/linux-i386-ipa.cc | 23 +-- > gdbserver/linux-x86-low.cc | 15 +- > gdbserver/linux-x86-tdesc.cc | 316 +++++++++++++++++++++++++---------- > gdbserver/linux-x86-tdesc.h | 49 +++--- > 5 files changed, 278 insertions(+), 168 deletions(-) > > diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc > index df5e6aca081..0c80812cc6f 100644 > --- a/gdbserver/linux-amd64-ipa.cc > +++ b/gdbserver/linux-amd64-ipa.cc > @@ -164,47 +164,21 @@ supply_static_tracepoint_registers (struct regcache > *regcache, > > #endif /* HAVE_UST */ > > -#if !defined __ILP32__ > -/* Map the tdesc index to xcr0 mask. */ > -static uint64_t idx2mask[X86_TDESC_LAST] = { > - X86_XSTATE_X87_MASK, > - X86_XSTATE_SSE_MASK, > - X86_XSTATE_AVX_MASK, > - X86_XSTATE_MPX_MASK, > - X86_XSTATE_AVX_MPX_MASK, > - X86_XSTATE_AVX_AVX512_MASK, > - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, > -}; > -#endif > - > /* Return target_desc to use for IPA, given the tdesc index passed by > gdbserver. */ > > const struct target_desc * > get_ipa_tdesc (int idx) > { > - if (idx >= X86_TDESC_LAST) > - { > - internal_error ("unknown ipa tdesc index: %d", idx); > - } > + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); > > #if defined __ILP32__ > - switch (idx) > - { > - case X86_TDESC_SSE: > - return amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); > - case X86_TDESC_AVX: > - return amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); > - case X86_TDESC_AVX_AVX512: > - return amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, > true); > - default: > - break; > - } > + bool is_x32 = true; > #else > - return amd64_linux_read_description (idx2mask[idx], false); > + bool is_x32 = false; > #endif > > - internal_error ("unknown ipa tdesc index: %d", idx); > + return amd64_linux_read_description (xcr0, is_x32); > } > > /* Allocate buffer for the jump pads. The branch instruction has a > @@ -272,11 +246,10 @@ void > initialize_low_tracepoint (void) > { > #if defined __ILP32__ > - amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); > - amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); > - amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, true); > + for (auto i = 0; i < x86_linux_x32_tdesc_count (); i++) > + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), true); > #else > - for (auto i = 0; i < X86_TDESC_LAST; i++) > - amd64_linux_read_description (idx2mask[i], false); > + for (auto i = 0; i < x86_linux_amd64_tdesc_count (); i++) > + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), false); I know this was already here and I know there are different opinions on this. But to me using auto here (and in similar locations in this patch) is just wrong. So I would vote for making this a proper type. But this is a bit of my personal "pet peeve" ;) > #endif > } > diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc > index aa346fc9bc3..c1c3152fb04 100644 > --- a/gdbserver/linux-i386-ipa.cc > +++ b/gdbserver/linux-i386-ipa.cc > @@ -245,28 +245,15 @@ initialize_fast_tracepoint_trampoline_buffer (void) > } > } > > -/* Map the tdesc index to xcr0 mask. */ > -static uint64_t idx2mask[X86_TDESC_LAST] = { > - X86_XSTATE_X87_MASK, > - X86_XSTATE_SSE_MASK, > - X86_XSTATE_AVX_MASK, > - X86_XSTATE_MPX_MASK, > - X86_XSTATE_AVX_MPX_MASK, > - X86_XSTATE_AVX_AVX512_MASK, > - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, > -}; > - > /* Return target_desc to use for IPA, given the tdesc index passed by > gdbserver. */ > > const struct target_desc * > get_ipa_tdesc (int idx) > { > - if (idx >= X86_TDESC_LAST) > - { > - internal_error ("unknown ipa tdesc index: %d", idx); > - } > - return i386_linux_read_description (idx2mask[idx]); > + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); > + > + return i386_linux_read_description (xcr0); > } > > /* Allocate buffer for the jump pads. On i386, we can reach an arbitrary > @@ -288,6 +275,6 @@ void > initialize_low_tracepoint (void) > { > initialize_fast_tracepoint_trampoline_buffer (); > - for (auto i = 0; i < X86_TDESC_LAST; i++) > - i386_linux_read_description (idx2mask[i]); > + for (auto i = 0; i < x86_linux_i386_tdesc_count (); i++) > + i386_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i)); > } > diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc > index 68d2f13537c..6e23a53118b 100644 > --- a/gdbserver/linux-x86-low.cc > +++ b/gdbserver/linux-x86-low.cc > @@ -2882,14 +2882,17 @@ x86_target::get_ipa_tdesc_idx () > struct regcache *regcache = get_thread_regcache (current_thread, 0); > const struct target_desc *tdesc = regcache->tdesc; > > + if (!use_xml) > + { > + if (tdesc == tdesc_i386_linux_no_xml.get () > #ifdef __x86_64__ > - return amd64_get_ipa_tdesc_idx (tdesc); > -#endif > - > - if (tdesc == tdesc_i386_linux_no_xml.get ()) > - return X86_TDESC_SSE; > + || tdesc == tdesc_amd64_linux_no_xml.get () > +#endif /* __x86_64__ */ > + ) > + return x86_linux_xcr0_to_tdesc_idx (X86_XSTATE_SSE_MASK); What happens if neither of the two is true? Do we need to assert this? Or do we need to just return the X86_XSTATE_SSE_MASK index without checking, as this can't ever happen? > + } > > - return i386_get_ipa_tdesc_idx (tdesc); > + return x86_linux_xcr0_to_tdesc_idx (xcr0_storage); > } > > /* The linux target ops object. */ > diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc > index af3735aa895..5e12526bf17 100644 > --- a/gdbserver/linux-x86-tdesc.cc > +++ b/gdbserver/linux-x86-tdesc.cc > @@ -28,96 +28,278 @@ > #include "x86-tdesc.h" > #include "arch/i386-linux-tdesc.h" > > -/* Return the right x86_linux_tdesc index for a given XCR0. Return > - X86_TDESC_LAST if can't find a match. */ > +/* A structure used to describe a single cpu feature that might, or might > + not, be checked for when creating a target description for one of i386, > + amd64, or x32. */ > > -static enum x86_linux_tdesc > -xcr0_to_tdesc_idx (uint64_t xcr0, bool is_x32) > +struct x86_tdesc_feature { > + /* The cpu feature mask. This is a mask against an xcr0 value. */ > + uint64_t feature; Can we elaborate the comment? Why do we need to mask anything? Or maybe we can refer to the table below. > + /* Is this feature checked when creating an i386 target description. */ > + bool is_i386; > + > + /* Is this feature checked when creating an amd64 target description. */ > + bool is_amd64; > + > + /* Is this feature checked when creating an x32 target description. */ > + bool is_x32; > +}; > + > +/* A constant table that describes all of the cpu features that are > + checked when building a target description for i386, amd64, or x32. */ > + I am missing a bit of elaboration on why we can't only rely on XCR0. And the table doesn't describe all CPU features that are checked. It just describes all XSTATE features. There is also segments and orig_rax and in the near future a new register for CET, which are independent of XSTATE. > +static constexpr x86_tdesc_feature x86_linux_all_tdesc_features[] = { > + /* Feature, i386, amd64, x32. */ > + { X86_XSTATE_PKRU, true, true, true }, > + { X86_XSTATE_AVX512, true, true, true }, > + { X86_XSTATE_AVX, true, true, true }, > + { X86_XSTATE_MPX, true, true, false }, > + { X86_XSTATE_SSE, true, false, false }, > + { X86_XSTATE_X87, true, false, false } > +}; I have never looked at IPA code much. But this table looked a bit weird to me. I get that MPX is not supported for x32. Though MPX is already deprecated and will be removed once GDB 15 is branched (at least that is the plan). But why does amd64 not have x87 and SSE? I do see e.g. st0 and xmm0 on amd64. After digging a bit: In arch/amd64.c, we call create_feature_i386_64bit_sse() without any check for SSE in XCR0. And I see that we have st0 in the "core" registers with EAX etc. But in arch/i386.c it is different, and we add SSE based on X86_XSTATE_SSE. And while st0 is also in the "core" registers with EAX etc., we only add them based on X86_XSTATE_X87. To me this looks wrong. Why would CPUs without x87 not add EAX? And even if it isn't for some reason, do we really want to support such old CPUs? In my view we could just align the two. If we would do that, and deprecate MPX, the table looks unnecessary. Or did I miss something else? This is a complicated series. Like others I am wondering how many users IPA really has and if it is worth maintaining. But I have no data. Is it in any distros? Thanks, Felix Intel Deutschland GmbH Registered Address: Am Campeon 10, 85579 Neubiberg, Germany Tel: +49 89 99 8853-0, www.intel.de <http://www.intel.de> Managing Directors: Christin Eisenschmid, Sharon Heck, Tiffany Doon Silva Chairperson of the Supervisory Board: Nicole Lau Registered Office: Munich Commercial Register: Amtsgericht Muenchen HRB 186928 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 09/11] gdbserver: update target description creation for x86/linux 2024-04-29 14:35 ` Willgerodt, Felix @ 2024-05-07 14:24 ` Andrew Burgess 2024-05-08 7:47 ` Willgerodt, Felix 0 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-05-07 14:24 UTC (permalink / raw) To: Willgerodt, Felix, gdb-patches; +Cc: John Baldwin "Willgerodt, Felix" <felix.willgerodt@intel.com> writes: >> -----Original Message----- >> From: Andrew Burgess <aburgess@redhat.com> >> Sent: Freitag, 26. April 2024 17:02 >> To: gdb-patches@sourceware.org >> Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix >> <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> >> Subject: [PATCHv5 09/11] gdbserver: update target description creation for >> x86/linux >> >> This commit is part of a series which aims to share more of the target >> description creation between GDB and gdbserver for x86/Linux. >> >> After some refactoring earlier in this series the shared >> x86_linux_tdesc_for_tid function was added into nat/x86-linux-tdesc.c. >> However, this function still relies on amd64_linux_read_description >> and i386_linux_read_description which are implemented separately for >> both gdbserver and GDB. Given that at their core, all these functions >> do is: >> >> 1. take an xcr0 value as input, >> 2. mask out some feature bits, >> 3. look for a cached pre-generated target description and return it >> if found, >> 4. if no cached target description is found then call either >> amd64_create_target_description or >> i386_create_target_description to create a new target >> description, which is then added to the cache. Return the newly >> created target description. >> >> The inner functions amd64_create_target_description and >> i386_create_target_description are already shared between GDB and >> gdbserver (in the gdb/arch/ directory), so the only thing that >> the *_read_description functions really do is add the caching layer, >> and it feels like this really could be shared. >> >> However, we have a small problem. >> >> On the GDB side we create target descriptions using a different set of >> cpu features than on the gdbserver side! This means that for the >> exact same target, we might get a different target description when >> using native GDB vs using gdbserver. This surely feels like a >> mistake, I would expect to get the same target description on each. >> >> The table below shows the number of possible different target >> descriptions that we can create on the GDB side vs on the gdbserver >> side for each target type: >> >> | GDB | gdbserver >> ------|-----|---------- >> i386 | 64 | 7 >> amd64 | 32 | 7 >> x32 | 16 | 7 > > I would have somehow expected to have the highest number for 64-bit > CPUs. As it should include 32-bit and x32. > Do you know why that isn't the case and 32-bit has double? > Or is my problem that this is reflecting GDB code, which shares stuff > between amd64 and i386? The table reflects the different tdesc split by category, not the total number that a user might get on a particular system. So on an _actual_ i386 CPU, then sure, the user will (on the GDB side) only have a choice of 64 different tdesc. On an _actual_ amd64 CPU the user could run an i386 compiled binary, in which case GDB will return one of those same 64 tdesc. Or the user might run an amd64 compiled binary, in which case GDB will return one of the 32 amd64 tdesc. So yes, from a user's point of view on amd64 there are 96 different tdesc, but GDB splits these into 3 different categories (i386, amd64, and x32), which is what this table represents. > Maybe this is also solved with the comments I made in code below. > > >> So in theory, all I want to do is move the GDB version >> of *_read_description into the arch/ directory and have gdbserver use >> that, then both GDB and gdbserver would be able to create any of the >> possible target descriptions. >> >> Unfortunately it's a little more complex than that due to the in >> process agent (IPA). >> >> When the IPA is in use, gdbserver sends a target description index to >> the IPA, and the IPA uses this to find the correct target description >> to use. >> >> ** START OF AN ASIDE ** >> >> Back in the day I suspect this approach made perfect sense. However >> since this commit: >> >> commit a8806230241d201f808d856eaae4d44088117b0c >> Date: Thu Dec 7 17:07:01 2017 +0000 >> >> Initialize target description early in IPA >> >> I think passing the index is now more trouble than its worth. >> >> We used to pass the index, and then use that index to lookup which >> target description to instantiate and use. However, the above commit >> fixed an issue where we can't call malloc() within (certain parts of) >> the IPA (apparently), so instead we now pre-compute _every_ possible >> target description within the IPA. The index is now only used to >> lookup which of the (many) pre-computed target descriptions to use. >> >> It would (I think) have been easier all around if the IPA just >> self-inspected, figured out its own xcr0 value, and used that to >> create the one target description that is required. So long as the >> xcr0 to target description code is shared (at compile time) with >> gdbserver, then we can be sure that the IPA will derive the same >> target description as gdbserver, and we would avoid all this index >> passing business, which has made this commit so very, very painful. >> >> ** END OF AN ASIDE ** >> >> Currently then for x86/linux, gdbserver sends a number between 0 and 7 >> to the IPA, and the IPA uses this to create a target description. >> >> However, I am proposing that gdbserver should now create one of (up >> to) 64 different target descriptions for i386, so this 0 to 7 index >> isn't going to be good enough any more (amd64 and x32 have slightly >> fewer possible target descriptions, but still more than 8, so the >> problem is the same). >> >> For a while I wondered if I was going to have to try and find some >> backward compatible solution for this mess. But after seeing how >> lightly the IPA is actually documented, I wonder if it is not the case >> that there is a tight coupling between a version of gdbserver and a >> version of the IPA? At least I'm hoping so. >> >> In this commit I have thrown out the old IPA target description index >> numbering scheme, and switched to a completely new numbering scheme. >> Instead of the index that is passed being arbitrary, the index is >> instead calculated from the set of cpu features that are present on >> the target. Within the IPA we can then reverse this logic to recreate >> the xcr0 value based on the index, and from the xcr0 value we can >> choose the correct target description. >> >> With the gdbserver to IPA numbering scheme issue resolved I have then >> update the gdbserver versions of amd64_linux_read_description and >> i386_linux_read_description so that they create target descriptions >> using the same set of cpu features as GDB itself. >> >> After this gdbserver should now always come up with the same target >> description as GDB does on any x86/Linux target. >> >> This commit does not introduce any new code sharing between GDB and >> gdbserver as previous commits in this series have done. Instead this >> commit is all about bringing GDB and gdbserver into alignment >> functionally so that the next commit(s) can merge the GDB and >> gdbserver versions of these functions. >> >> Approved-By: John Baldwin <jhb@FreeBSD.org> >> --- >> gdbserver/linux-amd64-ipa.cc | 43 +---- >> gdbserver/linux-i386-ipa.cc | 23 +-- >> gdbserver/linux-x86-low.cc | 15 +- >> gdbserver/linux-x86-tdesc.cc | 316 +++++++++++++++++++++++++---------- >> gdbserver/linux-x86-tdesc.h | 49 +++--- >> 5 files changed, 278 insertions(+), 168 deletions(-) >> >> diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc >> index df5e6aca081..0c80812cc6f 100644 >> --- a/gdbserver/linux-amd64-ipa.cc >> +++ b/gdbserver/linux-amd64-ipa.cc >> @@ -164,47 +164,21 @@ supply_static_tracepoint_registers (struct regcache >> *regcache, >> >> #endif /* HAVE_UST */ >> >> -#if !defined __ILP32__ >> -/* Map the tdesc index to xcr0 mask. */ >> -static uint64_t idx2mask[X86_TDESC_LAST] = { >> - X86_XSTATE_X87_MASK, >> - X86_XSTATE_SSE_MASK, >> - X86_XSTATE_AVX_MASK, >> - X86_XSTATE_MPX_MASK, >> - X86_XSTATE_AVX_MPX_MASK, >> - X86_XSTATE_AVX_AVX512_MASK, >> - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, >> -}; >> -#endif >> - >> /* Return target_desc to use for IPA, given the tdesc index passed by >> gdbserver. */ >> >> const struct target_desc * >> get_ipa_tdesc (int idx) >> { >> - if (idx >= X86_TDESC_LAST) >> - { >> - internal_error ("unknown ipa tdesc index: %d", idx); >> - } >> + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); >> >> #if defined __ILP32__ >> - switch (idx) >> - { >> - case X86_TDESC_SSE: >> - return amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); >> - case X86_TDESC_AVX: >> - return amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); >> - case X86_TDESC_AVX_AVX512: >> - return amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, >> true); >> - default: >> - break; >> - } >> + bool is_x32 = true; >> #else >> - return amd64_linux_read_description (idx2mask[idx], false); >> + bool is_x32 = false; >> #endif >> >> - internal_error ("unknown ipa tdesc index: %d", idx); >> + return amd64_linux_read_description (xcr0, is_x32); >> } >> >> /* Allocate buffer for the jump pads. The branch instruction has a >> @@ -272,11 +246,10 @@ void >> initialize_low_tracepoint (void) >> { >> #if defined __ILP32__ >> - amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); >> - amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); >> - amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, true); >> + for (auto i = 0; i < x86_linux_x32_tdesc_count (); i++) >> + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), true); >> #else >> - for (auto i = 0; i < X86_TDESC_LAST; i++) >> - amd64_linux_read_description (idx2mask[i], false); >> + for (auto i = 0; i < x86_linux_amd64_tdesc_count (); i++) >> + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), false); > > I know this was already here and I know there are different opinions on this. > But to me using auto here (and in similar locations in this patch) is just wrong. > So I would vote for making this a proper type. > But this is a bit of my personal "pet peeve" ;) > > >> #endif >> } >> diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc >> index aa346fc9bc3..c1c3152fb04 100644 >> --- a/gdbserver/linux-i386-ipa.cc >> +++ b/gdbserver/linux-i386-ipa.cc >> @@ -245,28 +245,15 @@ initialize_fast_tracepoint_trampoline_buffer (void) >> } >> } >> >> -/* Map the tdesc index to xcr0 mask. */ >> -static uint64_t idx2mask[X86_TDESC_LAST] = { >> - X86_XSTATE_X87_MASK, >> - X86_XSTATE_SSE_MASK, >> - X86_XSTATE_AVX_MASK, >> - X86_XSTATE_MPX_MASK, >> - X86_XSTATE_AVX_MPX_MASK, >> - X86_XSTATE_AVX_AVX512_MASK, >> - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, >> -}; >> - >> /* Return target_desc to use for IPA, given the tdesc index passed by >> gdbserver. */ >> >> const struct target_desc * >> get_ipa_tdesc (int idx) >> { >> - if (idx >= X86_TDESC_LAST) >> - { >> - internal_error ("unknown ipa tdesc index: %d", idx); >> - } >> - return i386_linux_read_description (idx2mask[idx]); >> + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); >> + >> + return i386_linux_read_description (xcr0); >> } >> >> /* Allocate buffer for the jump pads. On i386, we can reach an arbitrary >> @@ -288,6 +275,6 @@ void >> initialize_low_tracepoint (void) >> { >> initialize_fast_tracepoint_trampoline_buffer (); >> - for (auto i = 0; i < X86_TDESC_LAST; i++) >> - i386_linux_read_description (idx2mask[i]); >> + for (auto i = 0; i < x86_linux_i386_tdesc_count (); i++) >> + i386_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i)); >> } >> diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc >> index 68d2f13537c..6e23a53118b 100644 >> --- a/gdbserver/linux-x86-low.cc >> +++ b/gdbserver/linux-x86-low.cc >> @@ -2882,14 +2882,17 @@ x86_target::get_ipa_tdesc_idx () >> struct regcache *regcache = get_thread_regcache (current_thread, 0); >> const struct target_desc *tdesc = regcache->tdesc; >> >> + if (!use_xml) >> + { >> + if (tdesc == tdesc_i386_linux_no_xml.get () >> #ifdef __x86_64__ >> - return amd64_get_ipa_tdesc_idx (tdesc); >> -#endif >> - >> - if (tdesc == tdesc_i386_linux_no_xml.get ()) >> - return X86_TDESC_SSE; >> + || tdesc == tdesc_amd64_linux_no_xml.get () >> +#endif /* __x86_64__ */ >> + ) >> + return x86_linux_xcr0_to_tdesc_idx (X86_XSTATE_SSE_MASK); > > What happens if neither of the two is true? Do we need to assert this? > Or do we need to just return the X86_XSTATE_SSE_MASK index > without checking, as this can't ever happen? You're right. I've changed the 'if' into 'gdb_assert', and made the return of an index based on X86_XSTATE_SSE_MASK unconditional (within the '!use_xml' block. > > >> + } >> >> - return i386_get_ipa_tdesc_idx (tdesc); >> + return x86_linux_xcr0_to_tdesc_idx (xcr0_storage); >> } >> >> /* The linux target ops object. */ >> diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc >> index af3735aa895..5e12526bf17 100644 >> --- a/gdbserver/linux-x86-tdesc.cc >> +++ b/gdbserver/linux-x86-tdesc.cc >> @@ -28,96 +28,278 @@ >> #include "x86-tdesc.h" >> #include "arch/i386-linux-tdesc.h" >> >> -/* Return the right x86_linux_tdesc index for a given XCR0. Return >> - X86_TDESC_LAST if can't find a match. */ >> +/* A structure used to describe a single cpu feature that might, or might >> + not, be checked for when creating a target description for one of i386, >> + amd64, or x32. */ >> >> -static enum x86_linux_tdesc >> -xcr0_to_tdesc_idx (uint64_t xcr0, bool is_x32) >> +struct x86_tdesc_feature { >> + /* The cpu feature mask. This is a mask against an xcr0 value. */ >> + uint64_t feature; > > Can we elaborate the comment? Why do we need to mask anything? > Or maybe we can refer to the table below. > > >> + /* Is this feature checked when creating an i386 target description. */ >> + bool is_i386; >> + >> + /* Is this feature checked when creating an amd64 target description. */ >> + bool is_amd64; >> + >> + /* Is this feature checked when creating an x32 target description. */ >> + bool is_x32; >> +}; >> + >> +/* A constant table that describes all of the cpu features that are >> + checked when building a target description for i386, amd64, or x32. */ >> + > > I am missing a bit of elaboration on why we can't only rely on XCR0. And > the table doesn't describe all CPU features that are checked. It just describes > all XSTATE features. There is also segments and orig_rax and in the near future > a new register for CET, which are independent of XSTATE. I can rename the data structures to make it clearer that this all relates to xstate/xcr0. I was aware about the segments flag that is passed into the tdesc creation, but even with prompting I can't see anything related to orig_rax -- how is that fed into the tdesc creation process? > > >> +static constexpr x86_tdesc_feature x86_linux_all_tdesc_features[] = { >> + /* Feature, i386, amd64, x32. */ >> + { X86_XSTATE_PKRU, true, true, true }, >> + { X86_XSTATE_AVX512, true, true, true }, >> + { X86_XSTATE_AVX, true, true, true }, >> + { X86_XSTATE_MPX, true, true, false }, >> + { X86_XSTATE_SSE, true, false, false }, >> + { X86_XSTATE_X87, true, false, false } >> +}; > > I have never looked at IPA code much. But this table looked a bit weird to me. > I get that MPX is not supported for x32. Though MPX is already deprecated and will > be removed once GDB 15 is branched (at least that is the plan). You must excuse my lack of up to date knowledge of different x86 features, but is there not hardware in the wild with the MPX feature? Wont removing support for this mean folk will be unable to debug on that hardware? > But why does amd64 not have x87 and SSE? I do see e.g. st0 and xmm0 on amd64. > > After digging a bit: > In arch/amd64.c, we call create_feature_i386_64bit_sse() without any check for > SSE in XCR0. And I see that we have st0 in the "core" registers with EAX etc. > But in arch/i386.c it is different, and we add SSE based on X86_XSTATE_SSE. > And while st0 is also in the "core" registers with EAX etc., we only add them based > on X86_XSTATE_X87. To me this looks wrong. Why would CPUs without x87 > not add EAX? And even if it isn't for some reason, do we really want to support > such old CPUs? In my view we could just align the two. > > If we would do that, and deprecate MPX, the table looks unnecessary. Or did I miss > something else? That looks like a reasonable conclusion. Honestly, pretty much everything in this patch is based on the idea of merging gdbserver and GDB code, while keeping functionality as similar as possible. > > This is a complicated series. Like others I am wondering how many users IPA > really has and if it is worth maintaining. But I have no data. Is it in any distros? I have no idea. But I really don't want to tie this series (which fixes actual bugs caused by tdesc mismatch between GDB and gdbserver) to the removal of IPA. As far as I can tell the IPA is tested as part of the gdb.trace/ sub-directory, so it's not like we're removing some untested broken corner of GDB. We'd be removing a somewhat tested, at least minimally working, feature. I'll take another pass over this patch, maybe I can convince myself that we can just check for all features for all machine types, in which case the table driven approach becomes unnecessary. Also, maybe I can split the series in two. Like so many of my series it just seems to keep growing. While addressing your earlier comments I've already spotted another bug, which is going to push the patch count to at least 12 *sigh*. Lets see what I can come up with for v6! Thanks, Andrew > > > Thanks, > Felix > Intel Deutschland GmbH > Registered Address: Am Campeon 10, 85579 Neubiberg, Germany > Tel: +49 89 99 8853-0, www.intel.de <http://www.intel.de> > Managing Directors: Christin Eisenschmid, Sharon Heck, Tiffany Doon Silva > Chairperson of the Supervisory Board: Nicole Lau > Registered Office: Munich > Commercial Register: Amtsgericht Muenchen HRB 186928 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 09/11] gdbserver: update target description creation for x86/linux 2024-05-07 14:24 ` Andrew Burgess @ 2024-05-08 7:47 ` Willgerodt, Felix 2024-05-08 13:28 ` Andrew Burgess 0 siblings, 1 reply; 116+ messages in thread From: Willgerodt, Felix @ 2024-05-08 7:47 UTC (permalink / raw) To: Andrew Burgess, gdb-patches; +Cc: John Baldwin > -----Original Message----- > From: Andrew Burgess <aburgess@redhat.com> > Sent: Dienstag, 7. Mai 2024 16:25 > To: Willgerodt, Felix <felix.willgerodt@intel.com>; gdb-patches@sourceware.org > Cc: John Baldwin <jhb@FreeBSD.org> > Subject: RE: [PATCHv5 09/11] gdbserver: update target description creation for > x86/linux > > "Willgerodt, Felix" <felix.willgerodt@intel.com> writes: > > >> -----Original Message----- > >> From: Andrew Burgess <aburgess@redhat.com> > >> Sent: Freitag, 26. April 2024 17:02 > >> To: gdb-patches@sourceware.org > >> Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix > >> <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> > >> Subject: [PATCHv5 09/11] gdbserver: update target description creation for > >> x86/linux > >> > >> This commit is part of a series which aims to share more of the target > >> description creation between GDB and gdbserver for x86/Linux. > >> > >> After some refactoring earlier in this series the shared > >> x86_linux_tdesc_for_tid function was added into nat/x86-linux-tdesc.c. > >> However, this function still relies on amd64_linux_read_description > >> and i386_linux_read_description which are implemented separately for > >> both gdbserver and GDB. Given that at their core, all these functions > >> do is: > >> > >> 1. take an xcr0 value as input, > >> 2. mask out some feature bits, > >> 3. look for a cached pre-generated target description and return it > >> if found, > >> 4. if no cached target description is found then call either > >> amd64_create_target_description or > >> i386_create_target_description to create a new target > >> description, which is then added to the cache. Return the newly > >> created target description. > >> > >> The inner functions amd64_create_target_description and > >> i386_create_target_description are already shared between GDB and > >> gdbserver (in the gdb/arch/ directory), so the only thing that > >> the *_read_description functions really do is add the caching layer, > >> and it feels like this really could be shared. > >> > >> However, we have a small problem. > >> > >> On the GDB side we create target descriptions using a different set of > >> cpu features than on the gdbserver side! This means that for the > >> exact same target, we might get a different target description when > >> using native GDB vs using gdbserver. This surely feels like a > >> mistake, I would expect to get the same target description on each. > >> > >> The table below shows the number of possible different target > >> descriptions that we can create on the GDB side vs on the gdbserver > >> side for each target type: > >> > >> | GDB | gdbserver > >> ------|-----|---------- > >> i386 | 64 | 7 > >> amd64 | 32 | 7 > >> x32 | 16 | 7 > > > > I would have somehow expected to have the highest number for 64-bit > > CPUs. As it should include 32-bit and x32. > > Do you know why that isn't the case and 32-bit has double? > > Or is my problem that this is reflecting GDB code, which shares stuff > > between amd64 and i386? > > The table reflects the different tdesc split by category, not the total > number that a user might get on a particular system. > > So on an _actual_ i386 CPU, then sure, the user will (on the GDB side) > only have a choice of 64 different tdesc. > > On an _actual_ amd64 CPU the user could run an i386 compiled binary, in > which case GDB will return one of those same 64 tdesc. Or the user > might run an amd64 compiled binary, in which case GDB will return one of > the 32 amd64 tdesc. > > So yes, from a user's point of view on amd64 there are 96 different > tdesc, but GDB splits these into 3 different categories (i386, amd64, > and x32), which is what this table represents. Thanks for explaining. Understood. > > Maybe this is also solved with the comments I made in code below. > > > > > >> So in theory, all I want to do is move the GDB version > >> of *_read_description into the arch/ directory and have gdbserver use > >> that, then both GDB and gdbserver would be able to create any of the > >> possible target descriptions. > >> > >> Unfortunately it's a little more complex than that due to the in > >> process agent (IPA). > >> > >> When the IPA is in use, gdbserver sends a target description index to > >> the IPA, and the IPA uses this to find the correct target description > >> to use. > >> > >> ** START OF AN ASIDE ** > >> > >> Back in the day I suspect this approach made perfect sense. However > >> since this commit: > >> > >> commit a8806230241d201f808d856eaae4d44088117b0c > >> Date: Thu Dec 7 17:07:01 2017 +0000 > >> > >> Initialize target description early in IPA > >> > >> I think passing the index is now more trouble than its worth. > >> > >> We used to pass the index, and then use that index to lookup which > >> target description to instantiate and use. However, the above commit > >> fixed an issue where we can't call malloc() within (certain parts of) > >> the IPA (apparently), so instead we now pre-compute _every_ possible > >> target description within the IPA. The index is now only used to > >> lookup which of the (many) pre-computed target descriptions to use. > >> > >> It would (I think) have been easier all around if the IPA just > >> self-inspected, figured out its own xcr0 value, and used that to > >> create the one target description that is required. So long as the > >> xcr0 to target description code is shared (at compile time) with > >> gdbserver, then we can be sure that the IPA will derive the same > >> target description as gdbserver, and we would avoid all this index > >> passing business, which has made this commit so very, very painful. > >> > >> ** END OF AN ASIDE ** > >> > >> Currently then for x86/linux, gdbserver sends a number between 0 and 7 > >> to the IPA, and the IPA uses this to create a target description. > >> > >> However, I am proposing that gdbserver should now create one of (up > >> to) 64 different target descriptions for i386, so this 0 to 7 index > >> isn't going to be good enough any more (amd64 and x32 have slightly > >> fewer possible target descriptions, but still more than 8, so the > >> problem is the same). > >> > >> For a while I wondered if I was going to have to try and find some > >> backward compatible solution for this mess. But after seeing how > >> lightly the IPA is actually documented, I wonder if it is not the case > >> that there is a tight coupling between a version of gdbserver and a > >> version of the IPA? At least I'm hoping so. > >> > >> In this commit I have thrown out the old IPA target description index > >> numbering scheme, and switched to a completely new numbering scheme. > >> Instead of the index that is passed being arbitrary, the index is > >> instead calculated from the set of cpu features that are present on > >> the target. Within the IPA we can then reverse this logic to recreate > >> the xcr0 value based on the index, and from the xcr0 value we can > >> choose the correct target description. > >> > >> With the gdbserver to IPA numbering scheme issue resolved I have then > >> update the gdbserver versions of amd64_linux_read_description and > >> i386_linux_read_description so that they create target descriptions > >> using the same set of cpu features as GDB itself. > >> > >> After this gdbserver should now always come up with the same target > >> description as GDB does on any x86/Linux target. > >> > >> This commit does not introduce any new code sharing between GDB and > >> gdbserver as previous commits in this series have done. Instead this > >> commit is all about bringing GDB and gdbserver into alignment > >> functionally so that the next commit(s) can merge the GDB and > >> gdbserver versions of these functions. > >> > >> Approved-By: John Baldwin <jhb@FreeBSD.org> > >> --- > >> gdbserver/linux-amd64-ipa.cc | 43 +---- > >> gdbserver/linux-i386-ipa.cc | 23 +-- > >> gdbserver/linux-x86-low.cc | 15 +- > >> gdbserver/linux-x86-tdesc.cc | 316 +++++++++++++++++++++++++---------- > >> gdbserver/linux-x86-tdesc.h | 49 +++--- > >> 5 files changed, 278 insertions(+), 168 deletions(-) > >> > >> diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc > >> index df5e6aca081..0c80812cc6f 100644 > >> --- a/gdbserver/linux-amd64-ipa.cc > >> +++ b/gdbserver/linux-amd64-ipa.cc > >> @@ -164,47 +164,21 @@ supply_static_tracepoint_registers (struct regcache > >> *regcache, > >> > >> #endif /* HAVE_UST */ > >> > >> -#if !defined __ILP32__ > >> -/* Map the tdesc index to xcr0 mask. */ > >> -static uint64_t idx2mask[X86_TDESC_LAST] = { > >> - X86_XSTATE_X87_MASK, > >> - X86_XSTATE_SSE_MASK, > >> - X86_XSTATE_AVX_MASK, > >> - X86_XSTATE_MPX_MASK, > >> - X86_XSTATE_AVX_MPX_MASK, > >> - X86_XSTATE_AVX_AVX512_MASK, > >> - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, > >> -}; > >> -#endif > >> - > >> /* Return target_desc to use for IPA, given the tdesc index passed by > >> gdbserver. */ > >> > >> const struct target_desc * > >> get_ipa_tdesc (int idx) > >> { > >> - if (idx >= X86_TDESC_LAST) > >> - { > >> - internal_error ("unknown ipa tdesc index: %d", idx); > >> - } > >> + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); > >> > >> #if defined __ILP32__ > >> - switch (idx) > >> - { > >> - case X86_TDESC_SSE: > >> - return amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); > >> - case X86_TDESC_AVX: > >> - return amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); > >> - case X86_TDESC_AVX_AVX512: > >> - return amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, > >> true); > >> - default: > >> - break; > >> - } > >> + bool is_x32 = true; > >> #else > >> - return amd64_linux_read_description (idx2mask[idx], false); > >> + bool is_x32 = false; > >> #endif > >> > >> - internal_error ("unknown ipa tdesc index: %d", idx); > >> + return amd64_linux_read_description (xcr0, is_x32); > >> } > >> > >> /* Allocate buffer for the jump pads. The branch instruction has a > >> @@ -272,11 +246,10 @@ void > >> initialize_low_tracepoint (void) > >> { > >> #if defined __ILP32__ > >> - amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); > >> - amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); > >> - amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, true); > >> + for (auto i = 0; i < x86_linux_x32_tdesc_count (); i++) > >> + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), true); > >> #else > >> - for (auto i = 0; i < X86_TDESC_LAST; i++) > >> - amd64_linux_read_description (idx2mask[i], false); > >> + for (auto i = 0; i < x86_linux_amd64_tdesc_count (); i++) > >> + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), false); > > > > I know this was already here and I know there are different opinions on this. > > But to me using auto here (and in similar locations in this patch) is just wrong. > > So I would vote for making this a proper type. > > But this is a bit of my personal "pet peeve" ;) > > > > > >> #endif > >> } > >> diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc > >> index aa346fc9bc3..c1c3152fb04 100644 > >> --- a/gdbserver/linux-i386-ipa.cc > >> +++ b/gdbserver/linux-i386-ipa.cc > >> @@ -245,28 +245,15 @@ initialize_fast_tracepoint_trampoline_buffer (void) > >> } > >> } > >> > >> -/* Map the tdesc index to xcr0 mask. */ > >> -static uint64_t idx2mask[X86_TDESC_LAST] = { > >> - X86_XSTATE_X87_MASK, > >> - X86_XSTATE_SSE_MASK, > >> - X86_XSTATE_AVX_MASK, > >> - X86_XSTATE_MPX_MASK, > >> - X86_XSTATE_AVX_MPX_MASK, > >> - X86_XSTATE_AVX_AVX512_MASK, > >> - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, > >> -}; > >> - > >> /* Return target_desc to use for IPA, given the tdesc index passed by > >> gdbserver. */ > >> > >> const struct target_desc * > >> get_ipa_tdesc (int idx) > >> { > >> - if (idx >= X86_TDESC_LAST) > >> - { > >> - internal_error ("unknown ipa tdesc index: %d", idx); > >> - } > >> - return i386_linux_read_description (idx2mask[idx]); > >> + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); > >> + > >> + return i386_linux_read_description (xcr0); > >> } > >> > >> /* Allocate buffer for the jump pads. On i386, we can reach an arbitrary > >> @@ -288,6 +275,6 @@ void > >> initialize_low_tracepoint (void) > >> { > >> initialize_fast_tracepoint_trampoline_buffer (); > >> - for (auto i = 0; i < X86_TDESC_LAST; i++) > >> - i386_linux_read_description (idx2mask[i]); > >> + for (auto i = 0; i < x86_linux_i386_tdesc_count (); i++) > >> + i386_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i)); > >> } > >> diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc > >> index 68d2f13537c..6e23a53118b 100644 > >> --- a/gdbserver/linux-x86-low.cc > >> +++ b/gdbserver/linux-x86-low.cc > >> @@ -2882,14 +2882,17 @@ x86_target::get_ipa_tdesc_idx () > >> struct regcache *regcache = get_thread_regcache (current_thread, 0); > >> const struct target_desc *tdesc = regcache->tdesc; > >> > >> + if (!use_xml) > >> + { > >> + if (tdesc == tdesc_i386_linux_no_xml.get () > >> #ifdef __x86_64__ > >> - return amd64_get_ipa_tdesc_idx (tdesc); > >> -#endif > >> - > >> - if (tdesc == tdesc_i386_linux_no_xml.get ()) > >> - return X86_TDESC_SSE; > >> + || tdesc == tdesc_amd64_linux_no_xml.get () > >> +#endif /* __x86_64__ */ > >> + ) > >> + return x86_linux_xcr0_to_tdesc_idx (X86_XSTATE_SSE_MASK); > > > > What happens if neither of the two is true? Do we need to assert this? > > Or do we need to just return the X86_XSTATE_SSE_MASK index > > without checking, as this can't ever happen? > > You're right. I've changed the 'if' into 'gdb_assert', and made the > return of an index based on X86_XSTATE_SSE_MASK unconditional (within > the '!use_xml' block. > > > > > > >> + } > >> > >> - return i386_get_ipa_tdesc_idx (tdesc); > >> + return x86_linux_xcr0_to_tdesc_idx (xcr0_storage); > >> } > >> > >> /* The linux target ops object. */ > >> diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc > >> index af3735aa895..5e12526bf17 100644 > >> --- a/gdbserver/linux-x86-tdesc.cc > >> +++ b/gdbserver/linux-x86-tdesc.cc > >> @@ -28,96 +28,278 @@ > >> #include "x86-tdesc.h" > >> #include "arch/i386-linux-tdesc.h" > >> > >> -/* Return the right x86_linux_tdesc index for a given XCR0. Return > >> - X86_TDESC_LAST if can't find a match. */ > >> +/* A structure used to describe a single cpu feature that might, or might > >> + not, be checked for when creating a target description for one of i386, > >> + amd64, or x32. */ > >> > >> -static enum x86_linux_tdesc > >> -xcr0_to_tdesc_idx (uint64_t xcr0, bool is_x32) > >> +struct x86_tdesc_feature { > >> + /* The cpu feature mask. This is a mask against an xcr0 value. */ > >> + uint64_t feature; > > > > Can we elaborate the comment? Why do we need to mask anything? > > Or maybe we can refer to the table below. > > > > > >> + /* Is this feature checked when creating an i386 target description. */ > >> + bool is_i386; > >> + > >> + /* Is this feature checked when creating an amd64 target description. */ > >> + bool is_amd64; > >> + > >> + /* Is this feature checked when creating an x32 target description. */ > >> + bool is_x32; > >> +}; > >> + > >> +/* A constant table that describes all of the cpu features that are > >> + checked when building a target description for i386, amd64, or x32. */ > >> + > > > > I am missing a bit of elaboration on why we can't only rely on XCR0. And > > the table doesn't describe all CPU features that are checked. It just describes > > all XSTATE features. There is also segments and orig_rax and in the near future > > a new register for CET, which are independent of XSTATE. > > I can rename the data structures to make it clearer that this all > relates to xstate/xcr0. That would be good. A side node: Over the last couple of days I remember that we will need something aside from XCR0 also for AVX10 (basically avx512 but 256-bit registers only) and probably AMX as well. And APX. Mostly some CPUID bits or a register in the case of AMX. But I see that you already responded in patch 11 to my fear about the map. > I was aware about the segments flag that is > passed into the tdesc creation, but even with prompting I can't see > anything related to orig_rax -- how is that fed into the tdesc creation > process? Both segments and orig_rax are added unconditionally in amd64-linux-tdep.c (or the i386 equivalent). They are controlled via the "bool segments" and "bool is_linux" args of "*_create_target_description()". Unfortunately my knowledge is pretty spotty on these two as well, and I didn't try to dig out the patches that added them. > > > > > >> +static constexpr x86_tdesc_feature x86_linux_all_tdesc_features[] = { > >> + /* Feature, i386, amd64, x32. */ > >> + { X86_XSTATE_PKRU, true, true, true }, > >> + { X86_XSTATE_AVX512, true, true, true }, > >> + { X86_XSTATE_AVX, true, true, true }, > >> + { X86_XSTATE_MPX, true, true, false }, > >> + { X86_XSTATE_SSE, true, false, false }, > >> + { X86_XSTATE_X87, true, false, false } > >> +}; > > > > I have never looked at IPA code much. But this table looked a bit weird to me. > > I get that MPX is not supported for x32. Though MPX is already deprecated and > will > > be removed once GDB 15 is branched (at least that is the plan). > > You must excuse my lack of up to date knowledge of different x86 > features, but is there not hardware in the wild with the MPX feature? > Wont removing support for this mean folk will be unable to debug on that > hardware? Intel deprecated MPX in 2019 for various reasons: https://en.wikipedia.org/wiki/Intel_MPX I think that means no-one really should use MPX anymore, even if their CPU supports it. Since GCC, glibc and Linux don't support it anymore for a while, we also deprecated it: https://sourceware.org/git/?p=binutils-gdb.git;a=commit;h=7650ea38908e98a5823a334286813eca2ff5719e Another argument for that deprecation is Intel APX, which has been announced for future CPUs. It will re-use the MPX XSTATE area. While there is a new/different bit for APX in the XSTATE header, which would allow us to keep support for both types of registers, it is still a much needed deprecation in my view. The code is already complex enough. > > But why does amd64 not have x87 and SSE? I do see e.g. st0 and xmm0 on > amd64. > > > > After digging a bit: > > In arch/amd64.c, we call create_feature_i386_64bit_sse() without any check > for > > SSE in XCR0. And I see that we have st0 in the "core" registers with EAX etc. > > But in arch/i386.c it is different, and we add SSE based on X86_XSTATE_SSE. > > And while st0 is also in the "core" registers with EAX etc., we only add them > based > > on X86_XSTATE_X87. To me this looks wrong. Why would CPUs without x87 > > not add EAX? And even if it isn't for some reason, do we really want to support > > such old CPUs? In my view we could just align the two. > > > > If we would do that, and deprecate MPX, the table looks unnecessary. Or did I > miss > > something else? > > That looks like a reasonable conclusion. > > Honestly, pretty much everything in this patch is based on the idea of > merging gdbserver and GDB code, while keeping functionality as similar > as possible. Which is a really good thing. Thanks for tackling this! > > > > This is a complicated series. Like others I am wondering how many users IPA > > really has and if it is worth maintaining. But I have no data. Is it in any distros? > > I have no idea. But I really don't want to tie this series (which fixes > actual bugs caused by tdesc mismatch between GDB and gdbserver) to the > removal of IPA. > Totally fine with me. > As far as I can tell the IPA is tested as part of the gdb.trace/ > sub-directory, so it's not like we're removing some untested broken > corner of GDB. We'd be removing a somewhat tested, at least minimally > working, feature. > Ok by me, I was only wondering. It would for sure need a separate discussion and series. Thanks, Felix Intel Deutschland GmbH Registered Address: Am Campeon 10, 85579 Neubiberg, Germany Tel: +49 89 99 8853-0, www.intel.de Managing Directors: Sean Fennelly, Jeffrey Schneiderman, Tiffany Doon Silva Chairperson of the Supervisory Board: Nicole Lau Registered Office: Munich Commercial Register: Amtsgericht Muenchen HRB 186928 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 09/11] gdbserver: update target description creation for x86/linux 2024-05-08 7:47 ` Willgerodt, Felix @ 2024-05-08 13:28 ` Andrew Burgess 0 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-08 13:28 UTC (permalink / raw) To: Willgerodt, Felix, gdb-patches; +Cc: John Baldwin "Willgerodt, Felix" <felix.willgerodt@intel.com> writes: >> -----Original Message----- >> From: Andrew Burgess <aburgess@redhat.com> >> Sent: Dienstag, 7. Mai 2024 16:25 >> To: Willgerodt, Felix <felix.willgerodt@intel.com>; gdb-patches@sourceware.org >> Cc: John Baldwin <jhb@FreeBSD.org> >> Subject: RE: [PATCHv5 09/11] gdbserver: update target description creation for >> x86/linux >> >> "Willgerodt, Felix" <felix.willgerodt@intel.com> writes: >> >> >> -----Original Message----- >> >> From: Andrew Burgess <aburgess@redhat.com> >> >> Sent: Freitag, 26. April 2024 17:02 >> >> To: gdb-patches@sourceware.org >> >> Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix >> >> <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> >> >> Subject: [PATCHv5 09/11] gdbserver: update target description creation for >> >> x86/linux >> >> >> >> This commit is part of a series which aims to share more of the target >> >> description creation between GDB and gdbserver for x86/Linux. >> >> >> >> After some refactoring earlier in this series the shared >> >> x86_linux_tdesc_for_tid function was added into nat/x86-linux-tdesc.c. >> >> However, this function still relies on amd64_linux_read_description >> >> and i386_linux_read_description which are implemented separately for >> >> both gdbserver and GDB. Given that at their core, all these functions >> >> do is: >> >> >> >> 1. take an xcr0 value as input, >> >> 2. mask out some feature bits, >> >> 3. look for a cached pre-generated target description and return it >> >> if found, >> >> 4. if no cached target description is found then call either >> >> amd64_create_target_description or >> >> i386_create_target_description to create a new target >> >> description, which is then added to the cache. Return the newly >> >> created target description. >> >> >> >> The inner functions amd64_create_target_description and >> >> i386_create_target_description are already shared between GDB and >> >> gdbserver (in the gdb/arch/ directory), so the only thing that >> >> the *_read_description functions really do is add the caching layer, >> >> and it feels like this really could be shared. >> >> >> >> However, we have a small problem. >> >> >> >> On the GDB side we create target descriptions using a different set of >> >> cpu features than on the gdbserver side! This means that for the >> >> exact same target, we might get a different target description when >> >> using native GDB vs using gdbserver. This surely feels like a >> >> mistake, I would expect to get the same target description on each. >> >> >> >> The table below shows the number of possible different target >> >> descriptions that we can create on the GDB side vs on the gdbserver >> >> side for each target type: >> >> >> >> | GDB | gdbserver >> >> ------|-----|---------- >> >> i386 | 64 | 7 >> >> amd64 | 32 | 7 >> >> x32 | 16 | 7 >> > >> > I would have somehow expected to have the highest number for 64-bit >> > CPUs. As it should include 32-bit and x32. >> > Do you know why that isn't the case and 32-bit has double? >> > Or is my problem that this is reflecting GDB code, which shares stuff >> > between amd64 and i386? >> >> The table reflects the different tdesc split by category, not the total >> number that a user might get on a particular system. >> >> So on an _actual_ i386 CPU, then sure, the user will (on the GDB side) >> only have a choice of 64 different tdesc. >> >> On an _actual_ amd64 CPU the user could run an i386 compiled binary, in >> which case GDB will return one of those same 64 tdesc. Or the user >> might run an amd64 compiled binary, in which case GDB will return one of >> the 32 amd64 tdesc. >> >> So yes, from a user's point of view on amd64 there are 96 different >> tdesc, but GDB splits these into 3 different categories (i386, amd64, >> and x32), which is what this table represents. > > Thanks for explaining. Understood. > > >> > Maybe this is also solved with the comments I made in code below. >> > >> > >> >> So in theory, all I want to do is move the GDB version >> >> of *_read_description into the arch/ directory and have gdbserver use >> >> that, then both GDB and gdbserver would be able to create any of the >> >> possible target descriptions. >> >> >> >> Unfortunately it's a little more complex than that due to the in >> >> process agent (IPA). >> >> >> >> When the IPA is in use, gdbserver sends a target description index to >> >> the IPA, and the IPA uses this to find the correct target description >> >> to use. >> >> >> >> ** START OF AN ASIDE ** >> >> >> >> Back in the day I suspect this approach made perfect sense. However >> >> since this commit: >> >> >> >> commit a8806230241d201f808d856eaae4d44088117b0c >> >> Date: Thu Dec 7 17:07:01 2017 +0000 >> >> >> >> Initialize target description early in IPA >> >> >> >> I think passing the index is now more trouble than its worth. >> >> >> >> We used to pass the index, and then use that index to lookup which >> >> target description to instantiate and use. However, the above commit >> >> fixed an issue where we can't call malloc() within (certain parts of) >> >> the IPA (apparently), so instead we now pre-compute _every_ possible >> >> target description within the IPA. The index is now only used to >> >> lookup which of the (many) pre-computed target descriptions to use. >> >> >> >> It would (I think) have been easier all around if the IPA just >> >> self-inspected, figured out its own xcr0 value, and used that to >> >> create the one target description that is required. So long as the >> >> xcr0 to target description code is shared (at compile time) with >> >> gdbserver, then we can be sure that the IPA will derive the same >> >> target description as gdbserver, and we would avoid all this index >> >> passing business, which has made this commit so very, very painful. >> >> >> >> ** END OF AN ASIDE ** >> >> >> >> Currently then for x86/linux, gdbserver sends a number between 0 and 7 >> >> to the IPA, and the IPA uses this to create a target description. >> >> >> >> However, I am proposing that gdbserver should now create one of (up >> >> to) 64 different target descriptions for i386, so this 0 to 7 index >> >> isn't going to be good enough any more (amd64 and x32 have slightly >> >> fewer possible target descriptions, but still more than 8, so the >> >> problem is the same). >> >> >> >> For a while I wondered if I was going to have to try and find some >> >> backward compatible solution for this mess. But after seeing how >> >> lightly the IPA is actually documented, I wonder if it is not the case >> >> that there is a tight coupling between a version of gdbserver and a >> >> version of the IPA? At least I'm hoping so. >> >> >> >> In this commit I have thrown out the old IPA target description index >> >> numbering scheme, and switched to a completely new numbering scheme. >> >> Instead of the index that is passed being arbitrary, the index is >> >> instead calculated from the set of cpu features that are present on >> >> the target. Within the IPA we can then reverse this logic to recreate >> >> the xcr0 value based on the index, and from the xcr0 value we can >> >> choose the correct target description. >> >> >> >> With the gdbserver to IPA numbering scheme issue resolved I have then >> >> update the gdbserver versions of amd64_linux_read_description and >> >> i386_linux_read_description so that they create target descriptions >> >> using the same set of cpu features as GDB itself. >> >> >> >> After this gdbserver should now always come up with the same target >> >> description as GDB does on any x86/Linux target. >> >> >> >> This commit does not introduce any new code sharing between GDB and >> >> gdbserver as previous commits in this series have done. Instead this >> >> commit is all about bringing GDB and gdbserver into alignment >> >> functionally so that the next commit(s) can merge the GDB and >> >> gdbserver versions of these functions. >> >> >> >> Approved-By: John Baldwin <jhb@FreeBSD.org> >> >> --- >> >> gdbserver/linux-amd64-ipa.cc | 43 +---- >> >> gdbserver/linux-i386-ipa.cc | 23 +-- >> >> gdbserver/linux-x86-low.cc | 15 +- >> >> gdbserver/linux-x86-tdesc.cc | 316 +++++++++++++++++++++++++---------- >> >> gdbserver/linux-x86-tdesc.h | 49 +++--- >> >> 5 files changed, 278 insertions(+), 168 deletions(-) >> >> >> >> diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc >> >> index df5e6aca081..0c80812cc6f 100644 >> >> --- a/gdbserver/linux-amd64-ipa.cc >> >> +++ b/gdbserver/linux-amd64-ipa.cc >> >> @@ -164,47 +164,21 @@ supply_static_tracepoint_registers (struct regcache >> >> *regcache, >> >> >> >> #endif /* HAVE_UST */ >> >> >> >> -#if !defined __ILP32__ >> >> -/* Map the tdesc index to xcr0 mask. */ >> >> -static uint64_t idx2mask[X86_TDESC_LAST] = { >> >> - X86_XSTATE_X87_MASK, >> >> - X86_XSTATE_SSE_MASK, >> >> - X86_XSTATE_AVX_MASK, >> >> - X86_XSTATE_MPX_MASK, >> >> - X86_XSTATE_AVX_MPX_MASK, >> >> - X86_XSTATE_AVX_AVX512_MASK, >> >> - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, >> >> -}; >> >> -#endif >> >> - >> >> /* Return target_desc to use for IPA, given the tdesc index passed by >> >> gdbserver. */ >> >> >> >> const struct target_desc * >> >> get_ipa_tdesc (int idx) >> >> { >> >> - if (idx >= X86_TDESC_LAST) >> >> - { >> >> - internal_error ("unknown ipa tdesc index: %d", idx); >> >> - } >> >> + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); >> >> >> >> #if defined __ILP32__ >> >> - switch (idx) >> >> - { >> >> - case X86_TDESC_SSE: >> >> - return amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); >> >> - case X86_TDESC_AVX: >> >> - return amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); >> >> - case X86_TDESC_AVX_AVX512: >> >> - return amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, >> >> true); >> >> - default: >> >> - break; >> >> - } >> >> + bool is_x32 = true; >> >> #else >> >> - return amd64_linux_read_description (idx2mask[idx], false); >> >> + bool is_x32 = false; >> >> #endif >> >> >> >> - internal_error ("unknown ipa tdesc index: %d", idx); >> >> + return amd64_linux_read_description (xcr0, is_x32); >> >> } >> >> >> >> /* Allocate buffer for the jump pads. The branch instruction has a >> >> @@ -272,11 +246,10 @@ void >> >> initialize_low_tracepoint (void) >> >> { >> >> #if defined __ILP32__ >> >> - amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); >> >> - amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); >> >> - amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, true); >> >> + for (auto i = 0; i < x86_linux_x32_tdesc_count (); i++) >> >> + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), true); >> >> #else >> >> - for (auto i = 0; i < X86_TDESC_LAST; i++) >> >> - amd64_linux_read_description (idx2mask[i], false); >> >> + for (auto i = 0; i < x86_linux_amd64_tdesc_count (); i++) >> >> + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), false); >> > >> > I know this was already here and I know there are different opinions on this. >> > But to me using auto here (and in similar locations in this patch) is just wrong. >> > So I would vote for making this a proper type. >> > But this is a bit of my personal "pet peeve" ;) >> > >> > >> >> #endif >> >> } >> >> diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc >> >> index aa346fc9bc3..c1c3152fb04 100644 >> >> --- a/gdbserver/linux-i386-ipa.cc >> >> +++ b/gdbserver/linux-i386-ipa.cc >> >> @@ -245,28 +245,15 @@ initialize_fast_tracepoint_trampoline_buffer (void) >> >> } >> >> } >> >> >> >> -/* Map the tdesc index to xcr0 mask. */ >> >> -static uint64_t idx2mask[X86_TDESC_LAST] = { >> >> - X86_XSTATE_X87_MASK, >> >> - X86_XSTATE_SSE_MASK, >> >> - X86_XSTATE_AVX_MASK, >> >> - X86_XSTATE_MPX_MASK, >> >> - X86_XSTATE_AVX_MPX_MASK, >> >> - X86_XSTATE_AVX_AVX512_MASK, >> >> - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, >> >> -}; >> >> - >> >> /* Return target_desc to use for IPA, given the tdesc index passed by >> >> gdbserver. */ >> >> >> >> const struct target_desc * >> >> get_ipa_tdesc (int idx) >> >> { >> >> - if (idx >= X86_TDESC_LAST) >> >> - { >> >> - internal_error ("unknown ipa tdesc index: %d", idx); >> >> - } >> >> - return i386_linux_read_description (idx2mask[idx]); >> >> + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); >> >> + >> >> + return i386_linux_read_description (xcr0); >> >> } >> >> >> >> /* Allocate buffer for the jump pads. On i386, we can reach an arbitrary >> >> @@ -288,6 +275,6 @@ void >> >> initialize_low_tracepoint (void) >> >> { >> >> initialize_fast_tracepoint_trampoline_buffer (); >> >> - for (auto i = 0; i < X86_TDESC_LAST; i++) >> >> - i386_linux_read_description (idx2mask[i]); >> >> + for (auto i = 0; i < x86_linux_i386_tdesc_count (); i++) >> >> + i386_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i)); >> >> } >> >> diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc >> >> index 68d2f13537c..6e23a53118b 100644 >> >> --- a/gdbserver/linux-x86-low.cc >> >> +++ b/gdbserver/linux-x86-low.cc >> >> @@ -2882,14 +2882,17 @@ x86_target::get_ipa_tdesc_idx () >> >> struct regcache *regcache = get_thread_regcache (current_thread, 0); >> >> const struct target_desc *tdesc = regcache->tdesc; >> >> >> >> + if (!use_xml) >> >> + { >> >> + if (tdesc == tdesc_i386_linux_no_xml.get () >> >> #ifdef __x86_64__ >> >> - return amd64_get_ipa_tdesc_idx (tdesc); >> >> -#endif >> >> - >> >> - if (tdesc == tdesc_i386_linux_no_xml.get ()) >> >> - return X86_TDESC_SSE; >> >> + || tdesc == tdesc_amd64_linux_no_xml.get () >> >> +#endif /* __x86_64__ */ >> >> + ) >> >> + return x86_linux_xcr0_to_tdesc_idx (X86_XSTATE_SSE_MASK); >> > >> > What happens if neither of the two is true? Do we need to assert this? >> > Or do we need to just return the X86_XSTATE_SSE_MASK index >> > without checking, as this can't ever happen? >> >> You're right. I've changed the 'if' into 'gdb_assert', and made the >> return of an index based on X86_XSTATE_SSE_MASK unconditional (within >> the '!use_xml' block. >> >> > >> > >> >> + } >> >> >> >> - return i386_get_ipa_tdesc_idx (tdesc); >> >> + return x86_linux_xcr0_to_tdesc_idx (xcr0_storage); >> >> } >> >> >> >> /* The linux target ops object. */ >> >> diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc >> >> index af3735aa895..5e12526bf17 100644 >> >> --- a/gdbserver/linux-x86-tdesc.cc >> >> +++ b/gdbserver/linux-x86-tdesc.cc >> >> @@ -28,96 +28,278 @@ >> >> #include "x86-tdesc.h" >> >> #include "arch/i386-linux-tdesc.h" >> >> >> >> -/* Return the right x86_linux_tdesc index for a given XCR0. Return >> >> - X86_TDESC_LAST if can't find a match. */ >> >> +/* A structure used to describe a single cpu feature that might, or might >> >> + not, be checked for when creating a target description for one of i386, >> >> + amd64, or x32. */ >> >> >> >> -static enum x86_linux_tdesc >> >> -xcr0_to_tdesc_idx (uint64_t xcr0, bool is_x32) >> >> +struct x86_tdesc_feature { >> >> + /* The cpu feature mask. This is a mask against an xcr0 value. */ >> >> + uint64_t feature; >> > >> > Can we elaborate the comment? Why do we need to mask anything? >> > Or maybe we can refer to the table below. >> > >> > >> >> + /* Is this feature checked when creating an i386 target description. */ >> >> + bool is_i386; >> >> + >> >> + /* Is this feature checked when creating an amd64 target description. */ >> >> + bool is_amd64; >> >> + >> >> + /* Is this feature checked when creating an x32 target description. */ >> >> + bool is_x32; >> >> +}; >> >> + >> >> +/* A constant table that describes all of the cpu features that are >> >> + checked when building a target description for i386, amd64, or x32. */ >> >> + >> > >> > I am missing a bit of elaboration on why we can't only rely on XCR0. And >> > the table doesn't describe all CPU features that are checked. It just describes >> > all XSTATE features. There is also segments and orig_rax and in the near future >> > a new register for CET, which are independent of XSTATE. >> >> I can rename the data structures to make it clearer that this all >> relates to xstate/xcr0. > > That would be good. I've done this in the new version that I'll post shortly. > > A side node: Over the last couple of days I remember that we will need something > aside from XCR0 also for AVX10 (basically avx512 but 256-bit registers only) and > probably AMX as well. And APX. Mostly some CPUID bits or a register in the > case of AMX. But I see that you already responded in patch 11 to my fear about > the map. > > >> I was aware about the segments flag that is >> passed into the tdesc creation, but even with prompting I can't see >> anything related to orig_rax -- how is that fed into the tdesc creation >> process? > > Both segments and orig_rax are added unconditionally in amd64-linux-tdep.c > (or the i386 equivalent). They are controlled via the "bool segments" and > "bool is_linux" args of "*_create_target_description()". Unfortunately my > knowledge is pretty spotty on these two as well, and I didn't try to dig > out the patches that added them. I did eventually figure out that this is what we were talking about. And the reason that this doesn't matter, as far as the caching in this series is concerned, is that these fields always take fixed values for Linux targets, which is the only target this series touches. As I said in the patch #11 thread, beyond this series I definitely want to look into merging _all_ x86 target description caching, for all sub-architecture variants (i386, amd64, x32), on all OS (Linux, FreeBSD, etc), and this will mean the caching needs to start considering things outside of xcr0. The plan I have in mind would 100% be extendable to cover the AVX10 case you mention above. But for me, step #1 is merging the gdb/gdbserver code, then we can start expanding to cover more cases. > > >> > >> > >> >> +static constexpr x86_tdesc_feature x86_linux_all_tdesc_features[] = { >> >> + /* Feature, i386, amd64, x32. */ >> >> + { X86_XSTATE_PKRU, true, true, true }, >> >> + { X86_XSTATE_AVX512, true, true, true }, >> >> + { X86_XSTATE_AVX, true, true, true }, >> >> + { X86_XSTATE_MPX, true, true, false }, >> >> + { X86_XSTATE_SSE, true, false, false }, >> >> + { X86_XSTATE_X87, true, false, false } >> >> +}; >> > >> > I have never looked at IPA code much. But this table looked a bit weird to me. >> > I get that MPX is not supported for x32. Though MPX is already deprecated and >> will >> > be removed once GDB 15 is branched (at least that is the plan). >> >> You must excuse my lack of up to date knowledge of different x86 >> features, but is there not hardware in the wild with the MPX feature? >> Wont removing support for this mean folk will be unable to debug on that >> hardware? > > Intel deprecated MPX in 2019 for various reasons: > https://en.wikipedia.org/wiki/Intel_MPX > I think that means no-one really should use MPX anymore, even if their CPU > supports it. > > Since GCC, glibc and Linux don't support it anymore for a while, we also > deprecated it: > https://sourceware.org/git/?p=binutils-gdb.git;a=commit;h=7650ea38908e98a5823a334286813eca2ff5719e > > Another argument for that deprecation is Intel APX, which has been announced > for future CPUs. It will re-use the MPX XSTATE area. While there is a new/different > bit for APX in the XSTATE header, which would allow us to keep support > for both types of registers, it is still a much needed deprecation in my view. > The code is already complex enough. I'm happy to defer to your @intel email address :) So long as we're not going to leave users of older hardware in a bind then I'm good. Thanks, Andrew > > >> > But why does amd64 not have x87 and SSE? I do see e.g. st0 and xmm0 on >> amd64. >> > >> > After digging a bit: >> > In arch/amd64.c, we call create_feature_i386_64bit_sse() without any check >> for >> > SSE in XCR0. And I see that we have st0 in the "core" registers with EAX etc. >> > But in arch/i386.c it is different, and we add SSE based on X86_XSTATE_SSE. >> > And while st0 is also in the "core" registers with EAX etc., we only add them >> based >> > on X86_XSTATE_X87. To me this looks wrong. Why would CPUs without x87 >> > not add EAX? And even if it isn't for some reason, do we really want to support >> > such old CPUs? In my view we could just align the two. >> > >> > If we would do that, and deprecate MPX, the table looks unnecessary. Or did I >> miss >> > something else? >> >> That looks like a reasonable conclusion. >> >> Honestly, pretty much everything in this patch is based on the idea of >> merging gdbserver and GDB code, while keeping functionality as similar >> as possible. > > Which is a really good thing. Thanks for tackling this! > >> > >> > This is a complicated series. Like others I am wondering how many users IPA >> > really has and if it is worth maintaining. But I have no data. Is it in any distros? >> >> I have no idea. But I really don't want to tie this series (which fixes >> actual bugs caused by tdesc mismatch between GDB and gdbserver) to the >> removal of IPA. >> > > Totally fine with me. > >> As far as I can tell the IPA is tested as part of the gdb.trace/ >> sub-directory, so it's not like we're removing some untested broken >> corner of GDB. We'd be removing a somewhat tested, at least minimally >> working, feature. >> > > Ok by me, I was only wondering. It would for sure need a separate discussion > and series. > > Thanks, > Felix > Intel Deutschland GmbH > Registered Address: Am Campeon 10, 85579 Neubiberg, Germany > Tel: +49 89 99 8853-0, www.intel.de > Managing Directors: Sean Fennelly, Jeffrey Schneiderman, Tiffany Doon Silva > Chairperson of the Supervisory Board: Nicole Lau > Registered Office: Munich > Commercial Register: Amtsgericht Muenchen HRB 186928 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv5 10/11] gdb: move xcr0 == 0 check into i386_linux_core_read_description 2024-04-26 15:01 ` [PATCHv5 00/11] " Andrew Burgess ` (8 preceding siblings ...) 2024-04-26 15:01 ` [PATCHv5 09/11] gdbserver: update target description creation for x86/linux Andrew Burgess @ 2024-04-26 15:01 ` Andrew Burgess 2024-04-29 14:35 ` Willgerodt, Felix 2024-04-26 15:01 ` [PATCHv5 11/11] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess 2024-05-08 16:46 ` [PATCHv6 0/9] x86/Linux Target Description Changes Andrew Burgess 11 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-04-26 15:01 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, John Baldwin Currently, in i386_linux_core_read_description, if GDB fails to extract an xcr0 value from the core file, then we will have a default zero value for the xcr0 variable, we still call the i386_linux_read_description function, which checks for this zero value and returns nullptr. Back in i386_linux_core_read_description we spot the nullptr return value from i386_linux_read_description and call i386_linux_read_description again, but this time passing a default value for xcr0. In the next commit I plan to rework i386_linux_read_description, and in so doing I will remove the check for xcr0 == 0, this is inline with how the amd64 code is written. However, this means that the 'xcr0 == 0' check needs to move up the stack to i386_linux_core_read_description, again, this brings the i386 code into line with the amd64 code. This is just a refactor in preparation for the next commit, there should be no user visible changes after this commit. --- gdb/i386-linux-tdep.c | 15 ++++++++------- 1 file changed, 8 insertions(+), 7 deletions(-) diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index 78ebc99d3df..c796f87780b 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -714,15 +714,16 @@ i386_linux_core_read_description (struct gdbarch *gdbarch, /* Linux/i386. */ x86_xsave_layout layout; uint64_t xcr0 = i386_linux_core_read_xsave_info (abfd, layout); - const struct target_desc *tdesc = i386_linux_read_description (xcr0); - if (tdesc != NULL) - return tdesc; + if (xcr0 == 0) + { + if (bfd_get_section_by_name (abfd, ".reg-xfp") != nullptr) + xcr0 = X86_XSTATE_SSE_MASK; + else + xcr0 = X86_XSTATE_X87_MASK; + } - if (bfd_get_section_by_name (abfd, ".reg-xfp") != NULL) - return i386_linux_read_description (X86_XSTATE_SSE_MASK); - else - return i386_linux_read_description (X86_XSTATE_X87_MASK); + return i386_linux_read_description (xcr0); } /* Similar to i386_supply_fpregset, but use XSAVE extended state. */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 10/11] gdb: move xcr0 == 0 check into i386_linux_core_read_description 2024-04-26 15:01 ` [PATCHv5 10/11] gdb: move xcr0 == 0 check into i386_linux_core_read_description Andrew Burgess @ 2024-04-29 14:35 ` Willgerodt, Felix 0 siblings, 0 replies; 116+ messages in thread From: Willgerodt, Felix @ 2024-04-29 14:35 UTC (permalink / raw) To: Andrew Burgess, gdb-patches; +Cc: John Baldwin > -----Original Message----- > From: Andrew Burgess <aburgess@redhat.com> > Sent: Freitag, 26. April 2024 17:02 > To: gdb-patches@sourceware.org > Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix > <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> > Subject: [PATCHv5 10/11] gdb: move xcr0 == 0 check into > i386_linux_core_read_description > > Currently, in i386_linux_core_read_description, if GDB fails to > extract an xcr0 value from the core file, then we will have a default > zero value for the xcr0 variable, we still call the > i386_linux_read_description function, which checks for this zero value > and returns nullptr. > > Back in i386_linux_core_read_description we spot the nullptr return > value from i386_linux_read_description and call > i386_linux_read_description again, but this time passing a default > value for xcr0. > > In the next commit I plan to rework i386_linux_read_description, and > in so doing I will remove the check for xcr0 == 0, this is inline with > how the amd64 code is written. > > However, this means that the 'xcr0 == 0' check needs to move up the > stack to i386_linux_core_read_description, again, this brings the i386 > code into line with the amd64 code. > > This is just a refactor in preparation for the next commit, there > should be no user visible changes after this commit. > --- > gdb/i386-linux-tdep.c | 15 ++++++++------- > 1 file changed, 8 insertions(+), 7 deletions(-) > Approved-By: Felix Willgerodt <felix.willgerodt@intel.com> Thanks, Felix Intel Deutschland GmbH Registered Address: Am Campeon 10, 85579 Neubiberg, Germany Tel: +49 89 99 8853-0, www.intel.de <http://www.intel.de> Managing Directors: Christin Eisenschmid, Sharon Heck, Tiffany Doon Silva Chairperson of the Supervisory Board: Nicole Lau Registered Office: Munich Commercial Register: Amtsgericht Muenchen HRB 186928 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv5 11/11] gdb/gdbserver: share x86/linux tdesc caching 2024-04-26 15:01 ` [PATCHv5 00/11] " Andrew Burgess ` (9 preceding siblings ...) 2024-04-26 15:01 ` [PATCHv5 10/11] gdb: move xcr0 == 0 check into i386_linux_core_read_description Andrew Burgess @ 2024-04-26 15:01 ` Andrew Burgess 2024-04-29 14:35 ` Willgerodt, Felix 2024-05-08 16:46 ` [PATCHv6 0/9] x86/Linux Target Description Changes Andrew Burgess 11 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-04-26 15:01 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, John Baldwin This commit builds on the previous series of commits to share the target description caching code between GDB and gdbserver for x86/Linux targets. The objective of this commit is to move the four functions (2 each of) i386_linux_read_description and amd64_linux_read_description into the gdb/arch/ directory and combine them so we have just a single copy of each. Then GDB, gdbserver, and the in-process-agent (IPA) will link against these shared functions. It is worth reading the description of the previous commit(s) to see why this merging is not as simple as it seems, but in summary, gdbserver used to generate a different set of possible target descriptions than GDB. The previous commit(s) fixed this, so now it should be simpler to share the functions. One curiosity with this patch is the function x86_linux_post_init_tdesc. On the gdbserver side the two functions amd64_linux_read_description and i386_linux_read_description have some functionality that is not present on the GDB side, that is some additional configuration that is performed as each target description is created to setup the expedited registers. To support this I've added the function x86_linux_post_init_tdesc. This function is called from the two *_linux_read_description functions, but is implemented separately for GDB and gdbserver. An alternative approach that avoids adding x86_linux_post_init_tdesc would be to have x86_linux_tdesc_for_tid return a non-const target description, in x86_target::low_arch_setup we could then inspect the target description to figure out if it is 64-bit or not, and modify the target description as needed. In the end I figured that adding the x86_linux_post_init_tdesc function was good enough, so went with that solution. The contents of gdbserver/linux-x86-low.cc have moved to gdb/arch/x86-linux-tdesc-features.c, and gdbserver/linux-x86-tdesc.h has moved to gdb/arch/x86-linux-tdesc-features.h, this change leads to some updates in the #includes in the gdbserver/ directory. For testing I've done the following: - Built on x86-64 GNU/Linux for all targets, and just for the native target, - Build on i386 GNU/Linux for all targets, and just for the native target, - Build on a 64-bit, non-x86 GNU/Linux for all targets, just for the native target, and for targets x86_64-*-linux and i386-*-linux. This did flush out a bunch of build issues where I'd failed to add a required file in a configure.* file, but I think everything should now be good. --- gdb/Makefile.in | 5 + gdb/amd64-linux-tdep.c | 31 -- gdb/arch/amd64-linux-tdesc.c | 61 ++++ gdb/arch/i386-linux-tdesc.c | 51 +++ gdb/arch/x86-linux-tdesc-features.c | 247 +++++++++++++++ .../arch/x86-linux-tdesc-features.h | 50 +-- gdb/arch/x86-linux-tdesc.h | 37 +++ gdb/configure.nat | 6 +- gdb/configure.tgt | 11 +- gdb/i386-linux-tdep.c | 26 +- gdbserver/configure.srv | 9 + gdbserver/linux-amd64-ipa.cc | 2 +- gdbserver/linux-i386-ipa.cc | 2 +- gdbserver/linux-x86-low.cc | 2 +- gdbserver/linux-x86-tdesc.cc | 292 +----------------- 15 files changed, 469 insertions(+), 363 deletions(-) create mode 100644 gdb/arch/amd64-linux-tdesc.c create mode 100644 gdb/arch/i386-linux-tdesc.c create mode 100644 gdb/arch/x86-linux-tdesc-features.c rename gdbserver/linux-x86-tdesc.h => gdb/arch/x86-linux-tdesc-features.h (52%) create mode 100644 gdb/arch/x86-linux-tdesc.h diff --git a/gdb/Makefile.in b/gdb/Makefile.in index a24b2232daa..1e49ae396f4 100644 --- a/gdb/Makefile.in +++ b/gdb/Makefile.in @@ -748,6 +748,7 @@ ALL_64_TARGET_OBS = \ arch/aarch64-insn.o \ arch/aarch64-mte-linux.o \ arch/aarch64-scalable-linux.o \ + arch/amd64-linux-tdesc.o \ arch/amd64.o \ arch/riscv.o \ bpf-tdep.o \ @@ -788,9 +789,11 @@ ALL_TARGET_OBS = \ arch/arm.o \ arch/arm-get-next-pcs.o \ arch/arm-linux.o \ + arch/i386-linux-tdesc.o \ arch/i386.o \ arch/loongarch.o \ arch/ppc-linux-common.o \ + arch/x86-linux-tdesc-features.o \ arm-bsd-tdep.o \ arm-fbsd-tdep.o \ arm-linux-tdep.o \ @@ -1558,6 +1561,8 @@ HFILES_NO_SRCDIR = \ arch/ppc-linux-common.h \ arch/ppc-linux-tdesc.h \ arch/riscv.h \ + arch/x86-linux-tdesc-features.h \ + arch/x86-linux-tdesc.h \ cli/cli-cmds.h \ cli/cli-decode.h \ cli/cli-script.h \ diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c index bcb9868e79e..c707745cd9a 100644 --- a/gdb/amd64-linux-tdep.c +++ b/gdb/amd64-linux-tdep.c @@ -1577,37 +1577,6 @@ amd64_linux_record_signal (struct gdbarch *gdbarch, return 0; } -const target_desc * -amd64_linux_read_description (uint64_t xcr0_features_bit, bool is_x32) -{ - static target_desc *amd64_linux_tdescs \ - [2/*AVX*/][2/*MPX*/][2/*AVX512*/][2/*PKRU*/] = {}; - static target_desc *x32_linux_tdescs \ - [2/*AVX*/][2/*AVX512*/][2/*PKRU*/] = {}; - - target_desc **tdesc; - - if (is_x32) - { - tdesc = &x32_linux_tdescs[(xcr0_features_bit & X86_XSTATE_AVX) ? 1 : 0 ] - [(xcr0_features_bit & X86_XSTATE_AVX512) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_PKRU) ? 1 : 0]; - } - else - { - tdesc = &amd64_linux_tdescs[(xcr0_features_bit & X86_XSTATE_AVX) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_MPX) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_AVX512) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_PKRU) ? 1 : 0]; - } - - if (*tdesc == NULL) - *tdesc = amd64_create_target_description (xcr0_features_bit, is_x32, - true, true); - - return *tdesc; -} - /* Get Linux/x86 target description from core dump. */ static const struct target_desc * diff --git a/gdb/arch/amd64-linux-tdesc.c b/gdb/arch/amd64-linux-tdesc.c new file mode 100644 index 00000000000..63b5ddfcece --- /dev/null +++ b/gdb/arch/amd64-linux-tdesc.c @@ -0,0 +1,61 @@ +/* Target description related code for GNU/Linux x86-64. + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#include "arch/x86-linux-tdesc.h" +#include "arch/amd64-linux-tdesc.h" +#include "arch/amd64.h" +#include "arch/x86-linux-tdesc-features.h" + + +/* See arch/amd64-linux-tdesc.h. */ + +const struct target_desc * +amd64_linux_read_description (uint64_t xcr0, bool is_x32) +{ + /* The type used for the amd64 and x32 target description caches. */ + using tdesc_cache_type = std::unordered_map<uint64_t, const target_desc_up>; + + /* Caches for the previously seen amd64 and x32 target descriptions, + indexed by the xcr0 value that created the target description. These + need to be static within this function to ensure they are initialised + before first use. */ + static tdesc_cache_type amd64_tdesc_cache, x32_tdesc_cache; + + tdesc_cache_type &tdesc_cache = is_x32 ? x32_tdesc_cache : amd64_tdesc_cache; + + xcr0 &= is_x32 + ? x86_linux_x32_tdesc_feature_mask () + : x86_linux_amd64_tdesc_feature_mask (); + + const auto it = tdesc_cache.find (xcr0); + if (it != tdesc_cache.end ()) + return it->second.get (); + + /* Create the previously unseen target description. */ + target_desc_up tdesc (amd64_create_target_description (xcr0, is_x32, + true, true)); + x86_linux_post_init_tdesc (tdesc.get (), true); + + /* Add to the cache, and return a pointer borrowed from the + target_desc_up. This is safe as the cache (and the pointers contained + within it) are not deleted until GDB exits. */ + target_desc *ptr = tdesc.get (); + tdesc_cache.emplace (xcr0, std::move (tdesc)); + return ptr; +} diff --git a/gdb/arch/i386-linux-tdesc.c b/gdb/arch/i386-linux-tdesc.c new file mode 100644 index 00000000000..6d1f1793457 --- /dev/null +++ b/gdb/arch/i386-linux-tdesc.c @@ -0,0 +1,51 @@ +/* Target description related code for GNU/Linux i386. + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#include "arch/x86-linux-tdesc.h" +#include "arch/i386-linux-tdesc.h" +#include "arch/i386.h" +#include "arch/x86-linux-tdesc-features.h" + +/* See arch/i386-linux-tdesc.h. */ + +const target_desc * +i386_linux_read_description (uint64_t xcr0) +{ + /* Cache of previously seen i386 target descriptions, indexed by the xcr0 + value that created the target description. This needs to be static + within this function to ensure it is initialised before first use. */ + static std::unordered_map<uint64_t, const target_desc_up> i386_tdesc_cache; + + xcr0 &= x86_linux_i386_tdesc_feature_mask (); + + const auto it = i386_tdesc_cache.find (xcr0); + if (it != i386_tdesc_cache.end ()) + return it->second.get (); + + /* Create the previously unseen target description. */ + target_desc_up tdesc (i386_create_target_description (xcr0, true, false)); + x86_linux_post_init_tdesc (tdesc.get (), false); + + /* Add to the cache, and return a pointer borrowed from the + target_desc_up. This is safe as the cache (and the pointers contained + within it) are not deleted until GDB exits. */ + target_desc *ptr = tdesc.get (); + i386_tdesc_cache.emplace (xcr0, std::move (tdesc)); + return ptr; +} diff --git a/gdb/arch/x86-linux-tdesc-features.c b/gdb/arch/x86-linux-tdesc-features.c new file mode 100644 index 00000000000..91abd9b1928 --- /dev/null +++ b/gdb/arch/x86-linux-tdesc-features.c @@ -0,0 +1,247 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#include "arch/x86-linux-tdesc-features.h" + +/* A structure used to describe a single cpu feature that might, or might + not, be checked for when creating a target description for one of i386, + amd64, or x32. */ + +struct x86_tdesc_feature { + /* The cpu feature mask. This is a mask against an xcr0 value. */ + uint64_t feature; + + /* Is this feature checked when creating an i386 target description. */ + bool is_i386; + + /* Is this feature checked when creating an amd64 target description. */ + bool is_amd64; + + /* Is this feature checked when creating an x32 target description. */ + bool is_x32; +}; + +/* A constant table that describes all of the cpu features that are + checked when building a target description for i386, amd64, or x32. */ + +static constexpr x86_tdesc_feature x86_linux_all_tdesc_features[] = { + /* Feature, i386, amd64, x32. */ + { X86_XSTATE_PKRU, true, true, true }, + { X86_XSTATE_AVX512, true, true, true }, + { X86_XSTATE_AVX, true, true, true }, + { X86_XSTATE_MPX, true, true, false }, + { X86_XSTATE_SSE, true, false, false }, + { X86_XSTATE_X87, true, false, false } +}; + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an i386 target description. */ + +static constexpr uint64_t +x86_linux_i386_tdesc_feature_mask_1 () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_i386) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an amd64 target description. */ + +static constexpr uint64_t +x86_linux_amd64_tdesc_feature_mask_1 () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_amd64) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the cpu features + that are checked for when building an x32 target description. */ + +static constexpr uint64_t +x86_linux_x32_tdesc_feature_mask_1 () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_x32) + mask |= entry.feature; + + return mask; +} + +/* See arch/x86-linux-tdesc-features.h. */ + +uint64_t +x86_linux_amd64_tdesc_feature_mask () +{ + return x86_linux_amd64_tdesc_feature_mask_1 (); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +uint64_t +x86_linux_x32_tdesc_feature_mask () +{ + return x86_linux_x32_tdesc_feature_mask_1 (); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +uint64_t +x86_linux_i386_tdesc_feature_mask () +{ + return x86_linux_i386_tdesc_feature_mask_1 (); +} + +#ifdef GDBSERVER + +/* See arch/x86-linux-tdesc-features.h. */ + +int +x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0) +{ + /* The following table shows which features are checked for when creating + the target descriptions (see nat/x86-linux-tdesc.c), the feature order + represents the bit order within the generated index number. + + i386 | x87 sse mpx avx avx512 pkru + amd64 | mpx avx avx512 pkru + i32 | avx avx512 pkru + + The features are ordered so that for each mode (i386, amd64, i32) the + generated index will form a continuous range. */ + + int idx = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) + { + if ((xcr0 & x86_linux_all_tdesc_features[i].feature) + == x86_linux_all_tdesc_features[i].feature) + idx |= (1 << i); + } + + return idx; +} + +#endif /* GDBSERVER */ + +#ifdef IN_PROCESS_AGENT + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an i386 target description. */ + +static constexpr int +x86_linux_i386_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_i386) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an amd64 target description. */ + +static constexpr int +x86_linux_amd64_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_amd64) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of cpu + features that are checked for when building an x32 target description. */ + +static constexpr int +x86_linux_x32_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_tdesc_features) + if (entry.is_x32) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +int +x86_linux_amd64_tdesc_count () +{ + return x86_linux_amd64_tdesc_count_1 (); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +int +x86_linux_x32_tdesc_count () +{ + return x86_linux_x32_tdesc_count_1 (); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +int +x86_linux_i386_tdesc_count () +{ + return x86_linux_i386_tdesc_count_1 (); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +uint64_t +x86_linux_tdesc_idx_to_xcr0 (int idx) +{ + uint64_t xcr0 = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) + { + if ((idx & (1 << i)) != 0) + xcr0 |= x86_linux_all_tdesc_features[i].feature; + } + + return xcr0; +} + +#endif /* IN_PROCESS_AGENT */ diff --git a/gdbserver/linux-x86-tdesc.h b/gdb/arch/x86-linux-tdesc-features.h similarity index 52% rename from gdbserver/linux-x86-tdesc.h rename to gdb/arch/x86-linux-tdesc-features.h index 3d6e0e51833..0d3db587174 100644 --- a/gdbserver/linux-x86-tdesc.h +++ b/gdb/arch/x86-linux-tdesc-features.h @@ -1,7 +1,6 @@ -/* Low level support for x86 (i386 and x86-64), shared between gdbserver - and IPA. +/* Target description related code for GNU/Linux x86 (i386 and x86-64). - Copyright (C) 2016-2024 Free Software Foundation, Inc. + Copyright (C) 2024 Free Software Foundation, Inc. This file is part of GDB. @@ -18,33 +17,46 @@ You should have received a copy of the GNU General Public License along with this program. If not, see <http://www.gnu.org/licenses/>. */ -#ifndef GDBSERVER_LINUX_X86_TDESC_H -#define GDBSERVER_LINUX_X86_TDESC_H +#ifndef ARCH_X86_LINUX_TDESC_FEATURES_H +#define ARCH_X86_LINUX_TDESC_FEATURES_H -/* Convert an xcr0 value into an integer. The integer will be passed to - the in-process-agent where it will then be passed to - x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ +#include "gdbsupport/x86-xstate.h" +#include "gdbsupport/gdb_assert.h" -extern int x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0); +/* Return a mask of X86_STATE_* feature flags. The returned mask indicates + the set of features which are checked for when creating the target + description for each of amd64, x32, and i386. */ -#ifdef IN_PROCESS_AGENT +extern uint64_t x86_linux_amd64_tdesc_feature_mask (); +extern uint64_t x86_linux_x32_tdesc_feature_mask (); +extern uint64_t x86_linux_i386_tdesc_feature_mask (); -/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) - into an xcr0 value which can then be used to create a target - description. */ +#ifdef GDBSERVER -extern uint64_t x86_linux_tdesc_idx_to_xcr0 (int idx); +/* Convert an xcr0 value into an integer. The integer will be passed from + gdbserver to the in-process-agent where it will then be passed through + x86_linux_tdesc_idx_to_xcr0 to get back the original xcr0 value. */ + +extern int x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0); -/* Within the in-process-agent we need to pre-initialise all of the target - descriptions, to do this we need to know how many target descriptions - there are for each different target type. These functions return the - target description count for the relevant target. */ +#endif /* GDBSERVER */ + +#ifdef IN_PROCESS_AGENT + +/* Return the maximum possible number of target descriptions for each of + amd64, x32, and i386. These are used by the in-process-agent to + generate every possible target description. */ extern int x86_linux_amd64_tdesc_count (); extern int x86_linux_x32_tdesc_count (); extern int x86_linux_i386_tdesc_count (); +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +extern uint64_t x86_linux_tdesc_idx_to_xcr0 (int idx); #endif /* IN_PROCESS_AGENT */ -#endif /* GDBSERVER_LINUX_X86_TDESC_H */ +#endif /* ARCH_X86_LINUX_TDESC_FEATURES_H */ diff --git a/gdb/arch/x86-linux-tdesc.h b/gdb/arch/x86-linux-tdesc.h new file mode 100644 index 00000000000..152592fcf76 --- /dev/null +++ b/gdb/arch/x86-linux-tdesc.h @@ -0,0 +1,37 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#ifndef ARCH_X86_LINUX_TDESC_H +#define ARCH_X86_LINUX_TDESC_H + +struct target_desc; + +/* This function is called from amd64_linux_read_description and + i386_linux_read_description after a new target description has been + created, TDESC is the new target description, IS_64BIT will be true + when called from amd64_linux_read_description, otherwise IS_64BIT will + be false. If the *_linux_read_description functions found a cached + target description then this function will not be called. + + Both GDB and gdbserver have their own implementations of this + function. */ + +extern void x86_linux_post_init_tdesc (target_desc *tdesc, bool is_64bit); + +#endif /* ARCH_X86_LINUX_TDESC_H */ diff --git a/gdb/configure.nat b/gdb/configure.nat index 4bcc0696027..24e1824b01c 100644 --- a/gdb/configure.nat +++ b/gdb/configure.nat @@ -256,7 +256,8 @@ case ${gdb_host} in NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ nat/x86-xstate.o \ i386-linux-nat.o x86-linux-nat.o nat/linux-btrace.o \ - nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o" + nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o \ + arch/i386-linux-tdesc.o arch/x86-linux-tdesc-features.o" ;; ia64) # Host: Intel IA-64 running GNU/Linux @@ -323,7 +324,8 @@ case ${gdb_host} in nat/x86-xstate.o amd64-nat.o amd64-linux-nat.o x86-linux-nat.o \ nat/linux-btrace.o \ nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o \ - nat/amd64-linux-siginfo.o" + nat/amd64-linux-siginfo.o arch/x86-linux-tdesc-features.o \ + arch/i386-linux-tdesc.o arch/amd64-linux-tdesc.o" ;; sparc) # Host: GNU/Linux UltraSPARC diff --git a/gdb/configure.tgt b/gdb/configure.tgt index 47a674201f9..8326c458eb1 100644 --- a/gdb/configure.tgt +++ b/gdb/configure.tgt @@ -320,10 +320,13 @@ i[34567]86-*-linux*) gdb_target_obs="i386-linux-tdep.o \ glibc-tdep.o \ solib-svr4.o symfile-mem.o \ - linux-tdep.o linux-record.o" + linux-tdep.o linux-record.o \ + arch/i386-linux-tdesc.o \ + arch/x86-linux-tdesc-features.o" if test "x$enable_64_bit_bfd" = "xyes"; then # Target: GNU/Linux x86-64 - gdb_target_obs="amd64-linux-tdep.o ${gdb_target_obs}" + gdb_target_obs="amd64-linux-tdep.o \ + arch/amd64-linux-tdesc.o ${gdb_target_obs}" fi ;; i[34567]86-*-gnu*) @@ -718,7 +721,9 @@ x86_64-*-linux*) # Target: GNU/Linux x86-64 gdb_target_obs="amd64-linux-tdep.o ${i386_tobjs} \ i386-linux-tdep.o glibc-tdep.o \ - solib-svr4.o symfile-mem.o linux-tdep.o linux-record.o" + solib-svr4.o symfile-mem.o linux-tdep.o linux-record.o \ + arch/i386-linux-tdesc.o arch/amd64-linux-tdesc.o \ + arch/x86-linux-tdesc-features.o" ;; x86_64-*-freebsd* | x86_64-*-kfreebsd*-gnu) # Target: FreeBSD/amd64 diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index c796f87780b..d4d820b9f98 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -40,6 +40,7 @@ #include "i387-tdep.h" #include "gdbsupport/x86-xstate.h" #include "arch/i386-linux-tdesc.h" +#include "arch/x86-linux-tdesc.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_I386 "syscalls/i386-linux.xml" @@ -679,29 +680,12 @@ i386_linux_core_read_x86_xsave_layout (struct gdbarch *gdbarch, layout) != 0; } -/* See i386-linux-tdep.h. */ +/* See arch/x86-linux-tdesc.h. */ -const struct target_desc * -i386_linux_read_description (uint64_t xcr0) +void +x86_linux_post_init_tdesc (target_desc *tdesc, bool is_64bit) { - if (xcr0 == 0) - return NULL; - - static struct target_desc *i386_linux_tdescs \ - [2/*X87*/][2/*SSE*/][2/*AVX*/][2/*MPX*/][2/*AVX512*/][2/*PKRU*/] = {}; - struct target_desc **tdesc; - - tdesc = &i386_linux_tdescs[(xcr0 & X86_XSTATE_X87) ? 1 : 0] - [(xcr0 & X86_XSTATE_SSE) ? 1 : 0] - [(xcr0 & X86_XSTATE_AVX) ? 1 : 0] - [(xcr0 & X86_XSTATE_MPX) ? 1 : 0] - [(xcr0 & X86_XSTATE_AVX512) ? 1 : 0] - [(xcr0 & X86_XSTATE_PKRU) ? 1 : 0]; - - if (*tdesc == NULL) - *tdesc = i386_create_target_description (xcr0, true, false); - - return *tdesc; + /* Nothing. */ } /* Get Linux/x86 target description from core dump. */ diff --git a/gdbserver/configure.srv b/gdbserver/configure.srv index 7a2702d78bf..e17b5cf280c 100644 --- a/gdbserver/configure.srv +++ b/gdbserver/configure.srv @@ -110,12 +110,16 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" + srv_tgtobj="${srv_tgtobj} arch/x86-linux-tdesc-features.o" + srv_tgtobj="${srv_tgtobj} arch/i386-linux-tdesc.o" srv_linux_usrregs=yes srv_linux_regsets=yes srv_linux_thread_db=yes srv_linux_btrace=yes ipa_obj="linux-i386-ipa.o linux-x86-tdesc-ipa.o" ipa_obj="${ipa_obj} arch/i386-ipa.o" + ipa_obj="${ipa_obj} arch/x86-linux-tdesc-features-ipa.o" + ipa_obj="${ipa_obj} arch/i386-linux-tdesc-ipa.o" ;; i[34567]86-*-mingw*) srv_regobj="" srv_tgtobj="x86-low.o nat/x86-dregs.o win32-low.o" @@ -374,12 +378,17 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" srv_tgtobj="${srv_tgtobj} nat/amd64-linux-siginfo.o" + srv_tgtobj="${srv_tgtobj} arch/x86-linux-tdesc-features.o" + srv_tgtobj="${srv_tgtobj} arch/amd64-linux-tdesc.o" + srv_tgtobj="${srv_tgtobj} arch/i386-linux-tdesc.o" srv_linux_usrregs=yes # This is for i386 progs. srv_linux_regsets=yes srv_linux_thread_db=yes srv_linux_btrace=yes ipa_obj="linux-amd64-ipa.o linux-x86-tdesc-ipa.o" ipa_obj="${ipa_obj} arch/amd64-ipa.o" + ipa_obj="${ipa_obj} arch/x86-linux-tdesc-features-ipa.o" + ipa_obj="${ipa_obj} arch/amd64-linux-tdesc-ipa.o" ;; x86_64-*-mingw*) srv_regobj="" srv_tgtobj="x86-low.o nat/x86-dregs.o" diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index 0c80812cc6f..89857f208b8 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -20,9 +20,9 @@ #include <sys/mman.h> #include "tracepoint.h" -#include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" #include "arch/amd64-linux-tdesc.h" +#include "arch/x86-linux-tdesc-features.h" /* fast tracepoints collect registers. */ diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc index c1c3152fb04..8100c9f9840 100644 --- a/gdbserver/linux-i386-ipa.cc +++ b/gdbserver/linux-i386-ipa.cc @@ -20,9 +20,9 @@ #include <sys/mman.h> #include "tracepoint.h" -#include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" #include "arch/i386-linux-tdesc.h" +#include "arch/x86-linux-tdesc-features.h" /* GDB register numbers. */ diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 6e23a53118b..b089e818211 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -33,6 +33,7 @@ #endif #include "arch/i386-linux-tdesc.h" +#include "arch/x86-linux-tdesc-features.h" #include "gdb_proc_service.h" /* Don't include elf/common.h if linux/elf.h got included by @@ -48,7 +49,6 @@ #include "nat/linux-nat.h" #include "nat/x86-linux.h" #include "nat/x86-linux-dregs.h" -#include "linux-x86-tdesc.h" #include "nat/x86-linux-tdesc.h" #ifdef __x86_64__ diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc index 5e12526bf17..13c80762605 100644 --- a/gdbserver/linux-x86-tdesc.cc +++ b/gdbserver/linux-x86-tdesc.cc @@ -17,295 +17,19 @@ You should have received a copy of the GNU General Public License along with this program. If not, see <http://www.gnu.org/licenses/>. */ +#include "arch/x86-linux-tdesc.h" #include "tdesc.h" -#include "linux-x86-tdesc.h" -#include "arch/i386.h" -#include "gdbsupport/x86-xstate.h" -#ifdef __x86_64__ -#include "arch/amd64.h" -#include "arch/amd64-linux-tdesc.h" -#endif #include "x86-tdesc.h" -#include "arch/i386-linux-tdesc.h" - -/* A structure used to describe a single cpu feature that might, or might - not, be checked for when creating a target description for one of i386, - amd64, or x32. */ - -struct x86_tdesc_feature { - /* The cpu feature mask. This is a mask against an xcr0 value. */ - uint64_t feature; - - /* Is this feature checked when creating an i386 target description. */ - bool is_i386; - - /* Is this feature checked when creating an amd64 target description. */ - bool is_amd64; - - /* Is this feature checked when creating an x32 target description. */ - bool is_x32; -}; - -/* A constant table that describes all of the cpu features that are - checked when building a target description for i386, amd64, or x32. */ - -static constexpr x86_tdesc_feature x86_linux_all_tdesc_features[] = { - /* Feature, i386, amd64, x32. */ - { X86_XSTATE_PKRU, true, true, true }, - { X86_XSTATE_AVX512, true, true, true }, - { X86_XSTATE_AVX, true, true, true }, - { X86_XSTATE_MPX, true, true, false }, - { X86_XSTATE_SSE, true, false, false }, - { X86_XSTATE_X87, true, false, false } -}; - -/* Return a compile time constant which is a mask of all the cpu features - that are checked for when building an i386 target description. */ - -static constexpr uint64_t -x86_linux_i386_tdesc_feature_mask () -{ - uint64_t mask = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_i386) - mask |= entry.feature; - - return mask; -} - -/* Return a compile time constant which is a mask of all the cpu features - that are checked for when building an amd64 target description. */ - -static constexpr uint64_t -x86_linux_amd64_tdesc_feature_mask () -{ - uint64_t mask = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_amd64) - mask |= entry.feature; - - return mask; -} -/* Return a compile time constant which is a mask of all the cpu features - that are checked for when building an x32 target description. */ +/* See arch/x86-linux-tdesc.h. */ -static constexpr uint64_t -x86_linux_x32_tdesc_feature_mask () +void +x86_linux_post_init_tdesc (target_desc *tdesc, bool is_64bit) { - uint64_t mask = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_x32) - mask |= entry.feature; - - return mask; -} - -/* Return a compile time constant which is a count of the number of cpu - features that are checked for when building an i386 target description. */ - -static constexpr int -x86_linux_i386_tdesc_count_1 () -{ - uint64_t count = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_i386) - ++count; - - gdb_assert (count > 0); - - return (1 << count); -} - -/* Return a compile time constant which is a count of the number of cpu - features that are checked for when building an amd64 target description. */ - -static constexpr int -x86_linux_amd64_tdesc_count_1 () -{ - uint64_t count = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_amd64) - ++count; - - gdb_assert (count > 0); - - return (1 << count); -} - -/* Return a compile time constant which is a count of the number of cpu - features that are checked for when building an x32 target description. */ - -static constexpr int -x86_linux_x32_tdesc_count_1 () -{ - uint64_t count = 0; - - for (const auto &entry : x86_linux_all_tdesc_features) - if (entry.is_x32) - ++count; - - gdb_assert (count > 0); - - return (1 << count); -} - -#ifdef IN_PROCESS_AGENT - -/* See linux-x86-tdesc.h. */ - -int -x86_linux_amd64_tdesc_count () -{ - return x86_linux_amd64_tdesc_count_1 (); -} - -/* See linux-x86-tdesc.h. */ - -int -x86_linux_x32_tdesc_count () -{ - return x86_linux_x32_tdesc_count_1 (); -} - -/* See linux-x86-tdesc.h. */ - -int -x86_linux_i386_tdesc_count () -{ - return x86_linux_i386_tdesc_count_1 (); -} - -#endif /* IN_PROCESS_AGENT */ - -/* Convert an xcr0 value into an integer. The integer will be passed to - the in-process-agent where it will then be passed to - x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ - -int -x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0) -{ - /* The following table shows which features are checked for when creating - the target descriptions (see nat/x86-linux-tdesc.c), the feature order - represents the bit order within the generated index number. - - i386 | x87 sse mpx avx avx512 pkru - amd64 | mpx avx avx512 pkru - i32 | avx avx512 pkru - - The features are ordered so that for each mode (i386, amd64, i32) the - generated index will form a continuous range. */ - - int idx = 0; - - for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) - { - if ((xcr0 & x86_linux_all_tdesc_features[i].feature) - == x86_linux_all_tdesc_features[i].feature) - idx |= (1 << i); - } - - return idx; -} - - -#ifdef IN_PROCESS_AGENT - -/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) - into an xcr0 value which can then be used to create a target - description. */ - -uint64_t -x86_linux_tdesc_idx_to_xcr0 (int idx) -{ - uint64_t xcr0 = 0; - - for (int i = 0; i < ARRAY_SIZE (x86_linux_all_tdesc_features); ++i) - { - if ((idx & (1 << i)) != 0) - xcr0 |= x86_linux_all_tdesc_features[i].feature; - } - - return xcr0; -} - -#endif /* IN_PROCESS_AGENT */ - -#if defined __i386__ || !defined IN_PROCESS_AGENT - -/* A cache of all possible i386 target descriptions. */ - -static struct target_desc *i386_tdescs[x86_linux_i386_tdesc_count_1 ()] = { }; - -/* See nat/x86-linux-tdesc.h. */ - -const struct target_desc * -i386_linux_read_description (uint64_t xcr0) -{ - xcr0 &= x86_linux_i386_tdesc_feature_mask (); - int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - - gdb_assert (idx >= 0 && idx < x86_linux_i386_tdesc_count_1 ()); - - target_desc **tdesc = &i386_tdescs[idx]; - - if (*tdesc == nullptr) - { - *tdesc = i386_create_target_description (xcr0, true, false); - - init_target_desc (*tdesc, i386_expedite_regs); - } - - return *tdesc; -} -#endif - #ifdef __x86_64__ - -/* A cache of all possible amd64 target descriptions. */ - -static target_desc *amd64_tdescs[x86_linux_amd64_tdesc_count_1 ()] = { }; - -/* A cache of all possible x32 target descriptions. */ - -static target_desc *x32_tdescs[x86_linux_x32_tdesc_count_1 ()] = { }; - -/* See nat/x86-linux-tdesc.h. */ - -const struct target_desc * -amd64_linux_read_description (uint64_t xcr0, bool is_x32) -{ - if (is_x32) - xcr0 &= x86_linux_x32_tdesc_feature_mask (); + if (is_64bit) + init_target_desc (tdesc, amd64_expedite_regs); else - xcr0 &= x86_linux_amd64_tdesc_feature_mask (); - - int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - - if (is_x32) - gdb_assert (idx >= 0 && idx < x86_linux_x32_tdesc_count_1 ()); - else - gdb_assert (idx >= 0 && idx < x86_linux_amd64_tdesc_count_1 ()); - - target_desc **tdesc = nullptr; - - if (is_x32) - tdesc = &x32_tdescs[idx]; - else - tdesc = &amd64_tdescs[idx]; - - if (*tdesc == nullptr) - { - *tdesc = amd64_create_target_description (xcr0, is_x32, true, true); - - init_target_desc (*tdesc, amd64_expedite_regs); - } - return *tdesc; -} - #endif + init_target_desc (tdesc, i386_expedite_regs); +} -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 11/11] gdb/gdbserver: share x86/linux tdesc caching 2024-04-26 15:01 ` [PATCHv5 11/11] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess @ 2024-04-29 14:35 ` Willgerodt, Felix 2024-05-07 14:50 ` Andrew Burgess 0 siblings, 1 reply; 116+ messages in thread From: Willgerodt, Felix @ 2024-04-29 14:35 UTC (permalink / raw) To: Andrew Burgess, gdb-patches; +Cc: John Baldwin > -----Original Message----- > From: Andrew Burgess <aburgess@redhat.com> > Sent: Freitag, 26. April 2024 17:02 > To: gdb-patches@sourceware.org > Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix > <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> > Subject: [PATCHv5 11/11] gdb/gdbserver: share x86/linux tdesc caching > > This commit builds on the previous series of commits to share the > target description caching code between GDB and gdbserver for > x86/Linux targets. > > The objective of this commit is to move the four functions (2 each of) > i386_linux_read_description and amd64_linux_read_description into the > gdb/arch/ directory and combine them so we have just a single copy of > each. Then GDB, gdbserver, and the in-process-agent (IPA) will link > against these shared functions. > > It is worth reading the description of the previous commit(s) to see > why this merging is not as simple as it seems, but in summary, > gdbserver used to generate a different set of possible target > descriptions than GDB. The previous commit(s) fixed this, so now it > should be simpler to share the functions. > > One curiosity with this patch is the function > x86_linux_post_init_tdesc. On the gdbserver side the two functions > amd64_linux_read_description and i386_linux_read_description have some > functionality that is not present on the GDB side, that is some > additional configuration that is performed as each target description > is created to setup the expedited registers. > > To support this I've added the function x86_linux_post_init_tdesc. > This function is called from the two *_linux_read_description > functions, but is implemented separately for GDB and gdbserver. > > An alternative approach that avoids adding x86_linux_post_init_tdesc > would be to have x86_linux_tdesc_for_tid return a non-const target > description, in x86_target::low_arch_setup we could then inspect the > target description to figure out if it is 64-bit or not, and modify > the target description as needed. In the end I figured that adding > the x86_linux_post_init_tdesc function was good enough, so went with > that solution. > > The contents of gdbserver/linux-x86-low.cc have moved to > gdb/arch/x86-linux-tdesc-features.c, and gdbserver/linux-x86-tdesc.h > has moved to gdb/arch/x86-linux-tdesc-features.h, this change leads to > some updates in the #includes in the gdbserver/ directory. > > For testing I've done the following: > > - Built on x86-64 GNU/Linux for all targets, and just for the native > target, > > - Build on i386 GNU/Linux for all targets, and just for the native > target, > > - Build on a 64-bit, non-x86 GNU/Linux for all targets, just for the > native target, and for targets x86_64-*-linux and i386-*-linux. > > This did flush out a bunch of build issues where I'd failed to add a > required file in a configure.* file, but I think everything should now > be good. > --- > gdb/Makefile.in | 5 + > gdb/amd64-linux-tdep.c | 31 -- > gdb/arch/amd64-linux-tdesc.c | 61 ++++ > gdb/arch/i386-linux-tdesc.c | 51 +++ > gdb/arch/x86-linux-tdesc-features.c | 247 +++++++++++++++ > .../arch/x86-linux-tdesc-features.h | 50 +-- > gdb/arch/x86-linux-tdesc.h | 37 +++ > gdb/configure.nat | 6 +- > gdb/configure.tgt | 11 +- > gdb/i386-linux-tdep.c | 26 +- > gdbserver/configure.srv | 9 + > gdbserver/linux-amd64-ipa.cc | 2 +- > gdbserver/linux-i386-ipa.cc | 2 +- > gdbserver/linux-x86-low.cc | 2 +- > gdbserver/linux-x86-tdesc.cc | 292 +----------------- > 15 files changed, 469 insertions(+), 363 deletions(-) > create mode 100644 gdb/arch/amd64-linux-tdesc.c > create mode 100644 gdb/arch/i386-linux-tdesc.c > create mode 100644 gdb/arch/x86-linux-tdesc-features.c > rename gdbserver/linux-x86-tdesc.h => gdb/arch/x86-linux-tdesc-features.h > (52%) > create mode 100644 gdb/arch/x86-linux-tdesc.h > > diff --git a/gdb/Makefile.in b/gdb/Makefile.in > index a24b2232daa..1e49ae396f4 100644 > --- a/gdb/Makefile.in > +++ b/gdb/Makefile.in > @@ -748,6 +748,7 @@ ALL_64_TARGET_OBS = \ > arch/aarch64-insn.o \ > arch/aarch64-mte-linux.o \ > arch/aarch64-scalable-linux.o \ > + arch/amd64-linux-tdesc.o \ > arch/amd64.o \ > arch/riscv.o \ > bpf-tdep.o \ > @@ -788,9 +789,11 @@ ALL_TARGET_OBS = \ > arch/arm.o \ > arch/arm-get-next-pcs.o \ > arch/arm-linux.o \ > + arch/i386-linux-tdesc.o \ > arch/i386.o \ > arch/loongarch.o \ > arch/ppc-linux-common.o \ > + arch/x86-linux-tdesc-features.o \ > arm-bsd-tdep.o \ > arm-fbsd-tdep.o \ > arm-linux-tdep.o \ > @@ -1558,6 +1561,8 @@ HFILES_NO_SRCDIR = \ > arch/ppc-linux-common.h \ > arch/ppc-linux-tdesc.h \ > arch/riscv.h \ > + arch/x86-linux-tdesc-features.h \ > + arch/x86-linux-tdesc.h \ > cli/cli-cmds.h \ > cli/cli-decode.h \ > cli/cli-script.h \ > diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c > index bcb9868e79e..c707745cd9a 100644 > --- a/gdb/amd64-linux-tdep.c > +++ b/gdb/amd64-linux-tdep.c > @@ -1577,37 +1577,6 @@ amd64_linux_record_signal (struct gdbarch *gdbarch, > return 0; > } > > -const target_desc * > -amd64_linux_read_description (uint64_t xcr0_features_bit, bool is_x32) > -{ > - static target_desc *amd64_linux_tdescs \ > - [2/*AVX*/][2/*MPX*/][2/*AVX512*/][2/*PKRU*/] = {}; > - static target_desc *x32_linux_tdescs \ > - [2/*AVX*/][2/*AVX512*/][2/*PKRU*/] = {}; > - > - target_desc **tdesc; > - > - if (is_x32) > - { > - tdesc = &x32_linux_tdescs[(xcr0_features_bit & X86_XSTATE_AVX) ? 1 : 0 ] > - [(xcr0_features_bit & X86_XSTATE_AVX512) ? 1 : 0] > - [(xcr0_features_bit & X86_XSTATE_PKRU) ? 1 : 0]; > - } > - else > - { > - tdesc = &amd64_linux_tdescs[(xcr0_features_bit & X86_XSTATE_AVX) ? 1 : 0] > - [(xcr0_features_bit & X86_XSTATE_MPX) ? 1 : 0] > - [(xcr0_features_bit & X86_XSTATE_AVX512) ? 1 : 0] > - [(xcr0_features_bit & X86_XSTATE_PKRU) ? 1 : 0]; > - } > - > - if (*tdesc == NULL) > - *tdesc = amd64_create_target_description (xcr0_features_bit, is_x32, > - true, true); > - > - return *tdesc; > -} > - > /* Get Linux/x86 target description from core dump. */ > > static const struct target_desc * > diff --git a/gdb/arch/amd64-linux-tdesc.c b/gdb/arch/amd64-linux-tdesc.c > new file mode 100644 > index 00000000000..63b5ddfcece > --- /dev/null > +++ b/gdb/arch/amd64-linux-tdesc.c > @@ -0,0 +1,61 @@ > +/* Target description related code for GNU/Linux x86-64. > + > + Copyright (C) 2024 Free Software Foundation, Inc. > + > + This file is part of GDB. > + > + This program is free software; you can redistribute it and/or modify > + it under the terms of the GNU General Public License as published by > + the Free Software Foundation; either version 3 of the License, or > + (at your option) any later version. > + > + This program is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the > + GNU General Public License for more details. > + > + You should have received a copy of the GNU General Public License > + along with this program. If not, see <http://www.gnu.org/licenses/>. */ > + > +#include "arch/x86-linux-tdesc.h" > +#include "arch/amd64-linux-tdesc.h" > +#include "arch/amd64.h" > +#include "arch/x86-linux-tdesc-features.h" > + > + > +/* See arch/amd64-linux-tdesc.h. */ > + > +const struct target_desc * > +amd64_linux_read_description (uint64_t xcr0, bool is_x32) > +{ > + /* The type used for the amd64 and x32 target description caches. */ > + using tdesc_cache_type = std::unordered_map<uint64_t, const > target_desc_up>; > + > + /* Caches for the previously seen amd64 and x32 target descriptions, > + indexed by the xcr0 value that created the target description. These > + need to be static within this function to ensure they are initialised > + before first use. */ > + static tdesc_cache_type amd64_tdesc_cache, x32_tdesc_cache; > + Sorry for this long comment, it isn't strictly related to your series, as you don't change the status quo. But since you are changing the cache, I find it relevant. I wonder if caching of target descriptions based on xcr0 alone is really good enough. It isn't always the only thing used for determining registers. I realize you don't change that fact, though with the map we would need to hash xcr0 with the other factors for the key. Arguably that could still be viewed as better than the current way. Right now there is already the segments and "orig_rax" register for amd64. Though I don't know if IPA really supports them and I am not super familiar with their details. It seems like we just add them unconditionally or based on whether or not we are on Linux. I am not really sure that means we don't care about them for our cache (especially the Linux part), but it seems that is the current state. But in the near future there will be the shadow stack pointer register for CET, which we can only read with a separate ptrace call. See this patch, which we want to post soon: https://github.com/intel/gdb/commit/afff428f38a1f03ac0127797deda524d81ec156e I must admit I never really fully understood why this caching is done. Is it for a single GDB session connecting to different gdbserver sessions one after the other? Or for inferior re-starts? Or for runtime target description updates similar to SVE on aarch64? (Which we will also add at some point for AMX on x86.) Or a mix of all? Thanks, Felix Intel Deutschland GmbH Registered Address: Am Campeon 10, 85579 Neubiberg, Germany Tel: +49 89 99 8853-0, www.intel.de <http://www.intel.de> Managing Directors: Christin Eisenschmid, Sharon Heck, Tiffany Doon Silva Chairperson of the Supervisory Board: Nicole Lau Registered Office: Munich Commercial Register: Amtsgericht Muenchen HRB 186928 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 11/11] gdb/gdbserver: share x86/linux tdesc caching 2024-04-29 14:35 ` Willgerodt, Felix @ 2024-05-07 14:50 ` Andrew Burgess 2024-05-08 7:49 ` Willgerodt, Felix 0 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-05-07 14:50 UTC (permalink / raw) To: Willgerodt, Felix, gdb-patches; +Cc: John Baldwin "Willgerodt, Felix" <felix.willgerodt@intel.com> writes: >> -----Original Message----- >> From: Andrew Burgess <aburgess@redhat.com> >> Sent: Freitag, 26. April 2024 17:02 >> To: gdb-patches@sourceware.org >> Cc: Andrew Burgess <aburgess@redhat.com>; Willgerodt, Felix >> <felix.willgerodt@intel.com>; John Baldwin <jhb@FreeBSD.org> >> Subject: [PATCHv5 11/11] gdb/gdbserver: share x86/linux tdesc caching >> >> This commit builds on the previous series of commits to share the >> target description caching code between GDB and gdbserver for >> x86/Linux targets. >> >> The objective of this commit is to move the four functions (2 each of) >> i386_linux_read_description and amd64_linux_read_description into the >> gdb/arch/ directory and combine them so we have just a single copy of >> each. Then GDB, gdbserver, and the in-process-agent (IPA) will link >> against these shared functions. >> >> It is worth reading the description of the previous commit(s) to see >> why this merging is not as simple as it seems, but in summary, >> gdbserver used to generate a different set of possible target >> descriptions than GDB. The previous commit(s) fixed this, so now it >> should be simpler to share the functions. >> >> One curiosity with this patch is the function >> x86_linux_post_init_tdesc. On the gdbserver side the two functions >> amd64_linux_read_description and i386_linux_read_description have some >> functionality that is not present on the GDB side, that is some >> additional configuration that is performed as each target description >> is created to setup the expedited registers. >> >> To support this I've added the function x86_linux_post_init_tdesc. >> This function is called from the two *_linux_read_description >> functions, but is implemented separately for GDB and gdbserver. >> >> An alternative approach that avoids adding x86_linux_post_init_tdesc >> would be to have x86_linux_tdesc_for_tid return a non-const target >> description, in x86_target::low_arch_setup we could then inspect the >> target description to figure out if it is 64-bit or not, and modify >> the target description as needed. In the end I figured that adding >> the x86_linux_post_init_tdesc function was good enough, so went with >> that solution. >> >> The contents of gdbserver/linux-x86-low.cc have moved to >> gdb/arch/x86-linux-tdesc-features.c, and gdbserver/linux-x86-tdesc.h >> has moved to gdb/arch/x86-linux-tdesc-features.h, this change leads to >> some updates in the #includes in the gdbserver/ directory. >> >> For testing I've done the following: >> >> - Built on x86-64 GNU/Linux for all targets, and just for the native >> target, >> >> - Build on i386 GNU/Linux for all targets, and just for the native >> target, >> >> - Build on a 64-bit, non-x86 GNU/Linux for all targets, just for the >> native target, and for targets x86_64-*-linux and i386-*-linux. >> >> This did flush out a bunch of build issues where I'd failed to add a >> required file in a configure.* file, but I think everything should now >> be good. >> --- >> gdb/Makefile.in | 5 + >> gdb/amd64-linux-tdep.c | 31 -- >> gdb/arch/amd64-linux-tdesc.c | 61 ++++ >> gdb/arch/i386-linux-tdesc.c | 51 +++ >> gdb/arch/x86-linux-tdesc-features.c | 247 +++++++++++++++ >> .../arch/x86-linux-tdesc-features.h | 50 +-- >> gdb/arch/x86-linux-tdesc.h | 37 +++ >> gdb/configure.nat | 6 +- >> gdb/configure.tgt | 11 +- >> gdb/i386-linux-tdep.c | 26 +- >> gdbserver/configure.srv | 9 + >> gdbserver/linux-amd64-ipa.cc | 2 +- >> gdbserver/linux-i386-ipa.cc | 2 +- >> gdbserver/linux-x86-low.cc | 2 +- >> gdbserver/linux-x86-tdesc.cc | 292 +----------------- >> 15 files changed, 469 insertions(+), 363 deletions(-) >> create mode 100644 gdb/arch/amd64-linux-tdesc.c >> create mode 100644 gdb/arch/i386-linux-tdesc.c >> create mode 100644 gdb/arch/x86-linux-tdesc-features.c >> rename gdbserver/linux-x86-tdesc.h => gdb/arch/x86-linux-tdesc-features.h >> (52%) >> create mode 100644 gdb/arch/x86-linux-tdesc.h >> >> diff --git a/gdb/Makefile.in b/gdb/Makefile.in >> index a24b2232daa..1e49ae396f4 100644 >> --- a/gdb/Makefile.in >> +++ b/gdb/Makefile.in >> @@ -748,6 +748,7 @@ ALL_64_TARGET_OBS = \ >> arch/aarch64-insn.o \ >> arch/aarch64-mte-linux.o \ >> arch/aarch64-scalable-linux.o \ >> + arch/amd64-linux-tdesc.o \ >> arch/amd64.o \ >> arch/riscv.o \ >> bpf-tdep.o \ >> @@ -788,9 +789,11 @@ ALL_TARGET_OBS = \ >> arch/arm.o \ >> arch/arm-get-next-pcs.o \ >> arch/arm-linux.o \ >> + arch/i386-linux-tdesc.o \ >> arch/i386.o \ >> arch/loongarch.o \ >> arch/ppc-linux-common.o \ >> + arch/x86-linux-tdesc-features.o \ >> arm-bsd-tdep.o \ >> arm-fbsd-tdep.o \ >> arm-linux-tdep.o \ >> @@ -1558,6 +1561,8 @@ HFILES_NO_SRCDIR = \ >> arch/ppc-linux-common.h \ >> arch/ppc-linux-tdesc.h \ >> arch/riscv.h \ >> + arch/x86-linux-tdesc-features.h \ >> + arch/x86-linux-tdesc.h \ >> cli/cli-cmds.h \ >> cli/cli-decode.h \ >> cli/cli-script.h \ >> diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c >> index bcb9868e79e..c707745cd9a 100644 >> --- a/gdb/amd64-linux-tdep.c >> +++ b/gdb/amd64-linux-tdep.c >> @@ -1577,37 +1577,6 @@ amd64_linux_record_signal (struct gdbarch *gdbarch, >> return 0; >> } >> >> -const target_desc * >> -amd64_linux_read_description (uint64_t xcr0_features_bit, bool is_x32) >> -{ >> - static target_desc *amd64_linux_tdescs \ >> - [2/*AVX*/][2/*MPX*/][2/*AVX512*/][2/*PKRU*/] = {}; >> - static target_desc *x32_linux_tdescs \ >> - [2/*AVX*/][2/*AVX512*/][2/*PKRU*/] = {}; >> - >> - target_desc **tdesc; >> - >> - if (is_x32) >> - { >> - tdesc = &x32_linux_tdescs[(xcr0_features_bit & X86_XSTATE_AVX) ? 1 : 0 ] >> - [(xcr0_features_bit & X86_XSTATE_AVX512) ? 1 : 0] >> - [(xcr0_features_bit & X86_XSTATE_PKRU) ? 1 : 0]; >> - } >> - else >> - { >> - tdesc = &amd64_linux_tdescs[(xcr0_features_bit & X86_XSTATE_AVX) ? 1 : 0] >> - [(xcr0_features_bit & X86_XSTATE_MPX) ? 1 : 0] >> - [(xcr0_features_bit & X86_XSTATE_AVX512) ? 1 : 0] >> - [(xcr0_features_bit & X86_XSTATE_PKRU) ? 1 : 0]; >> - } >> - >> - if (*tdesc == NULL) >> - *tdesc = amd64_create_target_description (xcr0_features_bit, is_x32, >> - true, true); >> - >> - return *tdesc; >> -} >> - >> /* Get Linux/x86 target description from core dump. */ >> >> static const struct target_desc * >> diff --git a/gdb/arch/amd64-linux-tdesc.c b/gdb/arch/amd64-linux-tdesc.c >> new file mode 100644 >> index 00000000000..63b5ddfcece >> --- /dev/null >> +++ b/gdb/arch/amd64-linux-tdesc.c >> @@ -0,0 +1,61 @@ >> +/* Target description related code for GNU/Linux x86-64. >> + >> + Copyright (C) 2024 Free Software Foundation, Inc. >> + >> + This file is part of GDB. >> + >> + This program is free software; you can redistribute it and/or modify >> + it under the terms of the GNU General Public License as published by >> + the Free Software Foundation; either version 3 of the License, or >> + (at your option) any later version. >> + >> + This program is distributed in the hope that it will be useful, >> + but WITHOUT ANY WARRANTY; without even the implied warranty of >> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the >> + GNU General Public License for more details. >> + >> + You should have received a copy of the GNU General Public License >> + along with this program. If not, see <http://www.gnu.org/licenses/>. */ >> + >> +#include "arch/x86-linux-tdesc.h" >> +#include "arch/amd64-linux-tdesc.h" >> +#include "arch/amd64.h" >> +#include "arch/x86-linux-tdesc-features.h" >> + >> + >> +/* See arch/amd64-linux-tdesc.h. */ >> + >> +const struct target_desc * >> +amd64_linux_read_description (uint64_t xcr0, bool is_x32) >> +{ >> + /* The type used for the amd64 and x32 target description caches. */ >> + using tdesc_cache_type = std::unordered_map<uint64_t, const >> target_desc_up>; >> + >> + /* Caches for the previously seen amd64 and x32 target descriptions, >> + indexed by the xcr0 value that created the target description. These >> + need to be static within this function to ensure they are initialised >> + before first use. */ >> + static tdesc_cache_type amd64_tdesc_cache, x32_tdesc_cache; >> + > > Sorry for this long comment, it isn't strictly related to your series, as you don't > change the status quo. But since you are changing the cache, I find it relevant. > > I wonder if caching of target descriptions based on xcr0 alone is really > good enough. It isn't always the only thing used for determining registers. > I realize you don't change that fact, though with the map we would need to > hash xcr0 with the other factors for the key. Arguably that could still be > viewed as better than the current way. > > Right now there is already the segments and "orig_rax" register for amd64. > Though I don't know if IPA really supports them and I am not super familiar with > their details. It seems like we just add them unconditionally or based on whether > or not we are on Linux. I am not really sure that means we don't care about them > for our cache (especially the Linux part), but it seems that is the current state. > > But in the near future there will be the shadow stack pointer register for CET, > which we can only read with a separate ptrace call. > See this patch, which we want to post soon: > https://github.com/intel/gdb/commit/afff428f38a1f03ac0127797deda524d81ec156e > > I must admit I never really fully understood why this caching is done. > Is it for a single GDB session connecting to different gdbserver sessions one > after the other? Or for inferior re-starts? > Or for runtime target description updates similar to SVE on aarch64? > (Which we will also add at some point for AMX on x86.) > Or a mix of all? You are 100% correct. And I agree with everything you've said. I also spotted this issue and had a choice to make. I could: (a) fix the caching mechanism first, which would require duplicating the effort in GDB and gdbserver, then (b) merge the GDB and gdbserver code. Or I could: (a) merge the GDB and gdbserver code, then (b) fix the caching mechanism. As you can see, I opted to merge the code first, but this was already 11 patches. Which is going to be at least 12 in v6. Realistically, if I tried to "fix" caching at the same time then either each patch is going to get much bigger (and break the one change per patch rule) or the series is going to start getting really long. I've stopped where I have to try and keep the number of patches down; I don't want to overload reviewers. Plus, if I write too much on top of earlier patches and get asked to rewrite things the rewrites become more work. My plan after this series is merged was to unify all of the x86 tdesc caching, not just for Linux, but for all x86 targets (e.g. FreeBSD). I was going to adopt an approach similar to the approach I used for risc-v, which I think works pretty well, that is have a single struct that can holds all the features that impact tdesc creation, give that struct a hash() function, and then use this as the key for caching created tdesc. I'll take a look and see if I can do something part way towards this solution as part of this patch which would address your immediate concerns. Thanks, Andrew ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 11/11] gdb/gdbserver: share x86/linux tdesc caching 2024-05-07 14:50 ` Andrew Burgess @ 2024-05-08 7:49 ` Willgerodt, Felix 2024-05-08 16:09 ` Andrew Burgess 0 siblings, 1 reply; 116+ messages in thread From: Willgerodt, Felix @ 2024-05-08 7:49 UTC (permalink / raw) To: Andrew Burgess, gdb-patches; +Cc: John Baldwin > >> diff --git a/gdb/arch/amd64-linux-tdesc.c b/gdb/arch/amd64-linux-tdesc.c > >> new file mode 100644 > >> index 00000000000..63b5ddfcece > >> --- /dev/null > >> +++ b/gdb/arch/amd64-linux-tdesc.c > >> @@ -0,0 +1,61 @@ > >> +/* Target description related code for GNU/Linux x86-64. > >> + > >> + Copyright (C) 2024 Free Software Foundation, Inc. > >> + > >> + This file is part of GDB. > >> + > >> + This program is free software; you can redistribute it and/or modify > >> + it under the terms of the GNU General Public License as published by > >> + the Free Software Foundation; either version 3 of the License, or > >> + (at your option) any later version. > >> + > >> + This program is distributed in the hope that it will be useful, > >> + but WITHOUT ANY WARRANTY; without even the implied warranty of > >> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the > >> + GNU General Public License for more details. > >> + > >> + You should have received a copy of the GNU General Public License > >> + along with this program. If not, see <http://www.gnu.org/licenses/>. */ > >> + > >> +#include "arch/x86-linux-tdesc.h" > >> +#include "arch/amd64-linux-tdesc.h" > >> +#include "arch/amd64.h" > >> +#include "arch/x86-linux-tdesc-features.h" > >> + > >> + > >> +/* See arch/amd64-linux-tdesc.h. */ > >> + > >> +const struct target_desc * > >> +amd64_linux_read_description (uint64_t xcr0, bool is_x32) > >> +{ > >> + /* The type used for the amd64 and x32 target description caches. */ > >> + using tdesc_cache_type = std::unordered_map<uint64_t, const > >> target_desc_up>; > >> + > >> + /* Caches for the previously seen amd64 and x32 target descriptions, > >> + indexed by the xcr0 value that created the target description. These > >> + need to be static within this function to ensure they are initialised > >> + before first use. */ > >> + static tdesc_cache_type amd64_tdesc_cache, x32_tdesc_cache; > >> + > > > > Sorry for this long comment, it isn't strictly related to your series, as you don't > > change the status quo. But since you are changing the cache, I find it relevant. > > > > I wonder if caching of target descriptions based on xcr0 alone is really > > good enough. It isn't always the only thing used for determining registers. > > I realize you don't change that fact, though with the map we would need to > > hash xcr0 with the other factors for the key. Arguably that could still be > > viewed as better than the current way. > > > > Right now there is already the segments and "orig_rax" register for amd64. > > Though I don't know if IPA really supports them and I am not super familiar with > > their details. It seems like we just add them unconditionally or based on > whether > > or not we are on Linux. I am not really sure that means we don't care about > them > > for our cache (especially the Linux part), but it seems that is the current state. > > > > But in the near future there will be the shadow stack pointer register for CET, > > which we can only read with a separate ptrace call. > > See this patch, which we want to post soon: > > > https://github.com/intel/gdb/commit/afff428f38a1f03ac0127797deda524d81ec1 > 56e > > > > I must admit I never really fully understood why this caching is done. > > Is it for a single GDB session connecting to different gdbserver sessions one > > after the other? Or for inferior re-starts? > > Or for runtime target description updates similar to SVE on aarch64? > > (Which we will also add at some point for AMX on x86.) > > Or a mix of all? > > You are 100% correct. And I agree with everything you've said. > > I also spotted this issue and had a choice to make. > > I could: > > (a) fix the caching mechanism first, which would require duplicating > the effort in GDB and gdbserver, then > > (b) merge the GDB and gdbserver code. > > Or I could: > > (a) merge the GDB and gdbserver code, then > > (b) fix the caching mechanism. > > As you can see, I opted to merge the code first, but this was already 11 > patches. Which is going to be at least 12 in v6. > > Realistically, if I tried to "fix" caching at the same time then either > each patch is going to get much bigger (and break the one change per > patch rule) or the series is going to start getting really long. > > I've stopped where I have to try and keep the number of patches down; I > don't want to overload reviewers. Plus, if I write too much on top of > earlier patches and get asked to rewrite things the rewrites become more > work. I agree that we need to take this step by step and try to keep this series focused on the actual goal. For your sake and the sake of reviewers. > My plan after this series is merged was to unify all of the x86 tdesc > caching, not just for Linux, but for all x86 targets (e.g. FreeBSD). I > was going to adopt an approach similar to the approach I used for > risc-v, which I think works pretty well, that is have a single struct > that can holds all the features that impact tdesc creation, give that > struct a hash() function, and then use this as the key for caching > created tdesc. > > I'll take a look and see if I can do something part way towards this > solution as part of this patch which would address your immediate > concerns. Thanks a lot, I wasn't aware that risc-v already has a solution for this. Just for the record, I also would find it ok to not change to a map for this series yet. Though I for sure won't object to doing it here either. Felix Intel Deutschland GmbH Registered Address: Am Campeon 10, 85579 Neubiberg, Germany Tel: +49 89 99 8853-0, www.intel.de Managing Directors: Sean Fennelly, Jeffrey Schneiderman, Tiffany Doon Silva Chairperson of the Supervisory Board: Nicole Lau Registered Office: Munich Commercial Register: Amtsgericht Muenchen HRB 186928 ^ permalink raw reply [flat|nested] 116+ messages in thread
* RE: [PATCHv5 11/11] gdb/gdbserver: share x86/linux tdesc caching 2024-05-08 7:49 ` Willgerodt, Felix @ 2024-05-08 16:09 ` Andrew Burgess 0 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-08 16:09 UTC (permalink / raw) To: Willgerodt, Felix, gdb-patches; +Cc: John Baldwin "Willgerodt, Felix" <felix.willgerodt@intel.com> writes: >> >> diff --git a/gdb/arch/amd64-linux-tdesc.c b/gdb/arch/amd64-linux-tdesc.c >> >> new file mode 100644 >> >> index 00000000000..63b5ddfcece >> >> --- /dev/null >> >> +++ b/gdb/arch/amd64-linux-tdesc.c >> >> @@ -0,0 +1,61 @@ >> >> +/* Target description related code for GNU/Linux x86-64. >> >> + >> >> + Copyright (C) 2024 Free Software Foundation, Inc. >> >> + >> >> + This file is part of GDB. >> >> + >> >> + This program is free software; you can redistribute it and/or modify >> >> + it under the terms of the GNU General Public License as published by >> >> + the Free Software Foundation; either version 3 of the License, or >> >> + (at your option) any later version. >> >> + >> >> + This program is distributed in the hope that it will be useful, >> >> + but WITHOUT ANY WARRANTY; without even the implied warranty of >> >> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the >> >> + GNU General Public License for more details. >> >> + >> >> + You should have received a copy of the GNU General Public License >> >> + along with this program. If not, see <http://www.gnu.org/licenses/>. */ >> >> + >> >> +#include "arch/x86-linux-tdesc.h" >> >> +#include "arch/amd64-linux-tdesc.h" >> >> +#include "arch/amd64.h" >> >> +#include "arch/x86-linux-tdesc-features.h" >> >> + >> >> + >> >> +/* See arch/amd64-linux-tdesc.h. */ >> >> + >> >> +const struct target_desc * >> >> +amd64_linux_read_description (uint64_t xcr0, bool is_x32) >> >> +{ >> >> + /* The type used for the amd64 and x32 target description caches. */ >> >> + using tdesc_cache_type = std::unordered_map<uint64_t, const >> >> target_desc_up>; >> >> + >> >> + /* Caches for the previously seen amd64 and x32 target descriptions, >> >> + indexed by the xcr0 value that created the target description. These >> >> + need to be static within this function to ensure they are initialised >> >> + before first use. */ >> >> + static tdesc_cache_type amd64_tdesc_cache, x32_tdesc_cache; >> >> + >> > >> > Sorry for this long comment, it isn't strictly related to your series, as you don't >> > change the status quo. But since you are changing the cache, I find it relevant. >> > >> > I wonder if caching of target descriptions based on xcr0 alone is really >> > good enough. It isn't always the only thing used for determining registers. >> > I realize you don't change that fact, though with the map we would need to >> > hash xcr0 with the other factors for the key. Arguably that could still be >> > viewed as better than the current way. >> > >> > Right now there is already the segments and "orig_rax" register for amd64. >> > Though I don't know if IPA really supports them and I am not super familiar with >> > their details. It seems like we just add them unconditionally or based on >> whether >> > or not we are on Linux. I am not really sure that means we don't care about >> them >> > for our cache (especially the Linux part), but it seems that is the current state. >> > >> > But in the near future there will be the shadow stack pointer register for CET, >> > which we can only read with a separate ptrace call. >> > See this patch, which we want to post soon: >> > >> https://github.com/intel/gdb/commit/afff428f38a1f03ac0127797deda524d81ec1 >> 56e >> > >> > I must admit I never really fully understood why this caching is done. >> > Is it for a single GDB session connecting to different gdbserver sessions one >> > after the other? Or for inferior re-starts? >> > Or for runtime target description updates similar to SVE on aarch64? >> > (Which we will also add at some point for AMX on x86.) >> > Or a mix of all? >> >> You are 100% correct. And I agree with everything you've said. >> >> I also spotted this issue and had a choice to make. >> >> I could: >> >> (a) fix the caching mechanism first, which would require duplicating >> the effort in GDB and gdbserver, then >> >> (b) merge the GDB and gdbserver code. >> >> Or I could: >> >> (a) merge the GDB and gdbserver code, then >> >> (b) fix the caching mechanism. >> >> As you can see, I opted to merge the code first, but this was already 11 >> patches. Which is going to be at least 12 in v6. >> >> Realistically, if I tried to "fix" caching at the same time then either >> each patch is going to get much bigger (and break the one change per >> patch rule) or the series is going to start getting really long. >> >> I've stopped where I have to try and keep the number of patches down; I >> don't want to overload reviewers. Plus, if I write too much on top of >> earlier patches and get asked to rewrite things the rewrites become more >> work. > > I agree that we need to take this step by step and try to keep this series > focused on the actual goal. For your sake and the sake of reviewers. > >> My plan after this series is merged was to unify all of the x86 tdesc >> caching, not just for Linux, but for all x86 targets (e.g. FreeBSD). I >> was going to adopt an approach similar to the approach I used for >> risc-v, which I think works pretty well, that is have a single struct >> that can holds all the features that impact tdesc creation, give that >> struct a hash() function, and then use this as the key for caching >> created tdesc. >> >> I'll take a look and see if I can do something part way towards this >> solution as part of this patch which would address your immediate >> concerns. > > Thanks a lot, I wasn't aware that risc-v already has a solution for this. > > Just for the record, I also would find it ok to not change to a > map for this series yet. Though I for sure won't object to doing it > here either. I'd like to update my response to your feedback a little. You originally asked: > I wonder if caching of target descriptions based on xcr0 alone is really > good enough. It isn't always the only thing used for determining registers. > I realize you don't change that fact, though with the map we would need to > hash xcr0 with the other factors for the key. Arguably that could still be > viewed as better than the current way. And don't think I was clear enough in my reply. As far as this patch is concerned, caching based only on xcr0 is absolutely fine. Remember that this series is only for Linux, and for Linux the other (non xcr0) flags that are used when creating a target description (the is_linux and segments arguments to {amd64,i386}_create_target_description) are always given fixed values. As such we don't need to consider these flags at all when caching target descriptions (for x86 Linux). Once this series is merged, as I've said, I'd like to look at extending the caching to all x86 targets, in which case we will need to consider all the flags. The reason that I switched from an array to a map in this commit is because I wanted to avoid hard coding the array sizes for each x86 target type (i386, amd64, x32): - Only in gdb/arch/x86-linux-tdesc-features.c could we make use of the constexpr x86_linux_*_tdesc_count_1() functions to create fixed C-style arrays of the required sizes, this would require all of the caching code to live in this file, which didn't seem right, - I could move to a dynamic array like structure, e.g. std::vector, I could then grow the cache as needed, but I figure if I'm going to change the data structure anyway, then I might as well go with a map, - So I switched to use a map structure, keyed off the xcr0, which is fine, because (for x86 Linux) that's the only value that matters. I'll update the commit message to try and explain all this a bit better. Thanks, Andrew ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv6 0/9] x86/Linux Target Description Changes 2024-04-26 15:01 ` [PATCHv5 00/11] " Andrew Burgess ` (10 preceding siblings ...) 2024-04-26 15:01 ` [PATCHv5 11/11] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess @ 2024-05-08 16:46 ` Andrew Burgess 2024-05-08 16:46 ` [PATCHv6 1/9] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess ` (9 more replies) 11 siblings, 10 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-08 16:46 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, jhb In v6: - I merged some of the smaller patches which had been approved, and which were only loosely connected to this series, i.e. they were refactoring, or trivial cleanup patches. - I've reordered some of the remaining patches, moving the smaller patches towards the start of the series, these are largely unchanged and have mostly already been reviewed and/or approved. If there's no negative feedback for these smaller patches in this new order then I'll likely merge some more of these smaller ones to try and get the size of this series down. - Patches #1, #2, #3 are pretty much unchaged, and are either already approved or have at least been reviewed. - Patch #4 is new, though this is pretty similar to patch #3, I don't expect any problems from this patch. - Patch #5 has an error restored that Felix pointed out I'd deleted by mistake. The error was, as Felix commented, restored (in a different way) in a later patch anyway, but dropping it in this patch was a mistake. - Patch #6 has already been approved. - Patch #7 has had some changes after feedback from Felix, the x86_linux_tdesc_for_tid function has dropped its callback argument, and we now take two pointers in which the function caches two different pieces of state. I think this is actually a huge improvement, thanks Felix! I've also removed passing have_ptrace_getfpxregs by pointer to this function. Instead, after patch #4, we can now access have_ptrace_getfpxregs as a global, just like we do for have_ptrace_getregset, this is another great improvement I think. I've also fixed a bug in x86_linux_tdesc_for_tid which was present in the original code, if we're NOT on x86-64, and we do have HAVE_PTRACE_GETFPXREGS defined then we will now always call i386_linux_read_description. Previously, we'd only call i386_linux_read_description when have_ptrace_getfpxregs was TRIBOOL_UNKNOWN, which was the first time x86_linux_tdesc_for_tid was called. - Patch #8, I've renamed some of the data structures, and reworded some of the comments, to try and make it clearer that the code added in this commit is all about checking xstate (xcr0) feature bits, and is not claiming to represent all possible state which might ever need to be checked when creating a tdesc. Felix correctly pointed out that the previous naming / commenting was rather misleading, and though I believe the code was (and is) doing the right thing, the comments gave the impression that things had been overlooked. - Patch #9, the core ideas of this patch are unchanged. I've reworked the commit message to try and explain myself better. In v5: - Felix pointed out that building gdbserver with the '-m32' flag on an x86-64 host would fail. This is fixed in V5 with the addition of patch #4. This patch moves the have_ptrace_getfpxregs global into the gdb/nat/ directory and fixes the includes so that the the declaration is seen where needed, - I've rebased onto a slightly later commit. In v4: - I tried merging V3, but it turned out I broke pretty much everything that wasn't x86 based when configured with --enable-targets=all, - The problem was a failure to correctly split the shared code between the gdb/arch/ and gdb/nat/ directories, as a consequence, code which is needed on a non x86 based host to support x86 based targets wasn't available to the compilation, and the build failed, - In V4 I've gone through every patch and resplit the code in a way which I now believe is correct, I've done the following tests: + On a non x86 host I've built GDB to support only the current host as a target, to support all targets, and to support x86-64 and i386 linux targets, + On an i386 virtual machine I built GDB only for the host as a target, and for all targets. I regression tested the all targets build for unix, native-gdbserver, and native-extended-gdbserver, + On an x86-64 machine I've built GDB for only the current host as a target, and for all targets. I regression tested the all targets build for unix, native-gdbserver, and native-extended-gdbserver. - Only patches 6, 8, and 10 require significant review. All of the other patches are pretty trivial (though reviews always welcome). - I think there's more improvements that can be made to the x86 target description creation/lookup/caching. This series only changes the Linux lookup, and we still cache i386/amd64/x32 separately. In the future I think we can merge all x86 target description caching into a single data structure, this would be for all OS variants and all ABI variants. Though making that "grand unification" will certainly require some of the code in this series to change, I think the bulk of it will remain, and trying to do everything in one series is just going to result in an even larger series. I'd prefer to get these first patches merged, then come back to build on this work once this is merged and we know there's no problems with it. In v3: - Rebased. Nasty merge conflict with 4bb20a6244b7091 which I think I've resolved, but am unable to test. Reposting so the author of that other commit can validate. - Initial testing looks good. Full tests are still running. In v2: - Rebase to current upstream/master, no merge conflicts, - Retested. --- Andrew Burgess (9): gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition gdbserver/x86: move no-xml code earlier in x86_linux_read_description gdb/x86: move have_ptrace_getfpxregs global into gdb/nat directory gdb/x86: move have_ptrace_getregset global into gdb/nat directory gdb/x86: move reading of cs and ds state into gdb/nat directory gdb: move xcr0 == 0 check into i386_linux_core_read_description gdb/gdbserver: share some code relating to target description creation gdbserver: update target description creation for x86/linux gdb/gdbserver: share x86/linux tdesc caching gdb/Makefile.in | 9 +- gdb/amd64-linux-tdep.c | 32 +-- gdb/amd64-linux-tdep.h | 6 - gdb/arch/amd64-linux-tdesc.c | 61 ++++ gdb/arch/amd64-linux-tdesc.h | 30 ++ gdb/arch/i386-linux-tdesc.c | 51 ++++ .../i386-linux-tdesc.h} | 17 +- gdb/arch/x86-linux-tdesc-features.c | 267 ++++++++++++++++++ gdb/arch/x86-linux-tdesc-features.h | 62 ++++ gdb/arch/x86-linux-tdesc.h | 37 +++ gdb/configure.nat | 8 +- gdb/configure.tgt | 11 +- gdb/i386-linux-nat.c | 26 +- gdb/i386-linux-tdep.c | 42 +-- gdb/i386-linux-tdep.h | 23 -- gdb/linux-nat.c | 3 - gdb/linux-nat.h | 3 - gdb/nat/x86-linux-tdesc.c | 128 +++++++++ gdb/nat/x86-linux-tdesc.h | 54 ++++ gdb/nat/x86-linux.c | 59 ++++ gdb/nat/x86-linux.h | 45 +++ gdb/x86-linux-nat.c | 120 +------- gdbserver/configure.srv | 11 + gdbserver/i387-fp.cc | 9 +- gdbserver/i387-fp.h | 4 +- gdbserver/linux-amd64-ipa.cc | 46 +-- gdbserver/linux-i386-ipa.cc | 26 +- gdbserver/linux-low.cc | 3 - gdbserver/linux-low.h | 2 - gdbserver/linux-x86-low.cc | 208 +++++--------- gdbserver/linux-x86-tdesc.cc | 142 +--------- gdbserver/linux-x86-tdesc.h | 56 ---- gdbsupport/x86-xstate.h | 20 ++ 33 files changed, 980 insertions(+), 641 deletions(-) create mode 100644 gdb/arch/amd64-linux-tdesc.c create mode 100644 gdb/arch/amd64-linux-tdesc.h create mode 100644 gdb/arch/i386-linux-tdesc.c rename gdb/{i386-linux-nat.h => arch/i386-linux-tdesc.h} (64%) create mode 100644 gdb/arch/x86-linux-tdesc-features.c create mode 100644 gdb/arch/x86-linux-tdesc-features.h create mode 100644 gdb/arch/x86-linux-tdesc.h create mode 100644 gdb/nat/x86-linux-tdesc.c create mode 100644 gdb/nat/x86-linux-tdesc.h delete mode 100644 gdbserver/linux-x86-tdesc.h base-commit: 868883583e7520ff1bd99fcb224d2b33a990edff -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv6 1/9] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition 2024-05-08 16:46 ` [PATCHv6 0/9] x86/Linux Target Description Changes Andrew Burgess @ 2024-05-08 16:46 ` Andrew Burgess 2024-05-08 16:46 ` [PATCHv6 2/9] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess ` (8 subsequent siblings) 9 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-08 16:46 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, jhb Share the definition of I386_LINUX_XSAVE_XCR0_OFFSET between GDB and gdbserver. This commit moves the definition into gdbsupport/x86-xstate.h, which allows the #define to be shared. There should be no user visible changes after this commit. Approved-By: Felix Willgerodt <felix.willgerodt@intel.com> --- gdb/i386-linux-tdep.h | 20 -------------------- gdbserver/linux-x86-low.cc | 21 --------------------- gdbsupport/x86-xstate.h | 20 ++++++++++++++++++++ 3 files changed, 20 insertions(+), 41 deletions(-) diff --git a/gdb/i386-linux-tdep.h b/gdb/i386-linux-tdep.h index 5891747572b..07593c6a8ec 100644 --- a/gdb/i386-linux-tdep.h +++ b/gdb/i386-linux-tdep.h @@ -58,26 +58,6 @@ extern void i386_linux_report_signal_info (struct gdbarch *gdbarch, /* Return the target description according to XCR0. */ extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); -/* Format of XSAVE extended state is: - struct - { - fxsave_bytes[0..463] - sw_usable_bytes[464..511] - xstate_hdr_bytes[512..575] - extended state regions (AVX, MPX, AVX512, PKRU, etc.) - }; - - Same memory layout will be used for the coredump NT_X86_XSTATE - representing the XSAVE extended state registers. - - The first 8 bytes of the sw_usable_bytes[464..467] is the OS enabled - extended state mask, which is the same as the extended control register - 0 (the XFEATURE_ENABLED_MASK register), XCR0. We can use this mask - together with the mask saved in the xstate_hdr_bytes to determine what - states the processor/OS supports and what state, used or initialized, - the process/thread is in. */ -#define I386_LINUX_XSAVE_XCR0_OFFSET 464 - extern int i386_linux_gregset_reg_offset[]; /* Return x86 siginfo type. */ diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 2603fb2ac5d..6de44aeb2d9 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -831,27 +831,6 @@ x86_target::low_siginfo_fixup (siginfo_t *ptrace, gdb_byte *inf, int direction) \f static int use_xml; -/* Format of XSAVE extended state is: - struct - { - fxsave_bytes[0..463] - sw_usable_bytes[464..511] - xstate_hdr_bytes[512..575] - avx_bytes[576..831] - future_state etc - }; - - Same memory layout will be used for the coredump NT_X86_XSTATE - representing the XSAVE extended state registers. - - The first 8 bytes of the sw_usable_bytes[464..467] is the OS enabled - extended state mask, which is the same as the extended control register - 0 (the XFEATURE_ENABLED_MASK register), XCR0. We can use this mask - together with the mask saved in the xstate_hdr_bytes to determine what - states the processor/OS supports and what state, used or initialized, - the process/thread is in. */ -#define I386_LINUX_XSAVE_XCR0_OFFSET 464 - /* Does the current host support the GETFPXREGS request? The header file may or may not define it, and even if it is defined, the kernel will return EIO if it's running on a pre-SSE processor. */ diff --git a/gdbsupport/x86-xstate.h b/gdbsupport/x86-xstate.h index 89c1143fbe1..11b37544aa3 100644 --- a/gdbsupport/x86-xstate.h +++ b/gdbsupport/x86-xstate.h @@ -120,4 +120,24 @@ constexpr bool operator!= (const x86_xsave_layout &lhs, #define I387_MXCSR_INIT_VAL 0x1f80 +/* Format of XSAVE extended state is: + struct + { + fxsave_bytes[0..463] + sw_usable_bytes[464..511] + xstate_hdr_bytes[512..575] + extended state regions (AVX, MPX, AVX512, PKRU, etc.) + }; + + Same memory layout will be used for the coredump NT_X86_XSTATE + representing the XSAVE extended state registers. + + The first 8 bytes of the sw_usable_bytes[464..467] is the OS enabled + extended state mask, which is the same as the extended control register + 0 (the XFEATURE_ENABLED_MASK register), XCR0. We can use this mask + together with the mask saved in the xstate_hdr_bytes to determine what + states the processor/OS supports and what state, used or initialized, + the process/thread is in. */ +#define I386_LINUX_XSAVE_XCR0_OFFSET 464 + #endif /* COMMON_X86_XSTATE_H */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv6 2/9] gdbserver/x86: move no-xml code earlier in x86_linux_read_description 2024-05-08 16:46 ` [PATCHv6 0/9] x86/Linux Target Description Changes Andrew Burgess 2024-05-08 16:46 ` [PATCHv6 1/9] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess @ 2024-05-08 16:46 ` Andrew Burgess 2024-05-08 16:46 ` [PATCHv6 3/9] gdb/x86: move have_ptrace_getfpxregs global into gdb/nat directory Andrew Burgess ` (7 subsequent siblings) 9 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-08 16:46 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, jhb This commit is part of a series that aims to share more of the x86 target description reading/generation code between GDB and gdbserver. There are a huge number of similarities between the code in gdbserver's x86_linux_read_description function and GDB's x86_linux_nat_target::read_description function, and it is this similarity that I plan, in a later commit, to share between GDB and gdbserver. However, one thing that is different in x86_linux_read_description is the code inside the '!use_xml' block. This is the code that handles the case where gdbserver is not allowed to send an XML target description back to GDB. In this case gdbserver uses some predefined, fixed, target descriptions. First, it's worth noting that I suspect this code is not tested any more. I couldn't find anything in the testsuite that tries to disable XML target description support. And the idea of having a single "fixed" target description really doesn't work well when we think about all the various x86 extensions that exist. Part of me would like to rip out the no-xml support in gdbserver (at least for x86), and if a GDB connects that doesn't support XML target descriptions, gdbserver can just give an error and drop the connection. GDB has supported XML target descriptions for 16 years now, I think it would be reasonable for our shipped gdbserver to drop support for the old way of doing things. Anyway.... this commit doesn't do that. What I did notice was that, over time, the '!use_xml' block appears to have "drifted" within the x86_linux_read_description function; it's now not the first check we do. Instead we make some ptrace calls and return a target description generated based on the result of these ptrace calls. Surely it only makes sense to generate variable target descriptions if we can send these back to GDB? So in this commit I propose to move the '!use_xml' block earlier in the x86_linux_read_description function. The benefit of this is that this leaves the later half of x86_linux_read_description much more similar to the GDB function x86_linux_nat_target::read_description and sets us up for potentially sharing code between GDB and gdbserver in a later commit. Approved-By: John Baldwin <jhb@FreeBSD.org> Approved-By: Felix Willgerodt <felix.willgerodt@intel.com> --- gdbserver/linux-x86-low.cc | 27 ++++++++++++++++----------- 1 file changed, 16 insertions(+), 11 deletions(-) diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 6de44aeb2d9..e8ef3667eb4 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -869,6 +869,22 @@ x86_linux_read_description (void) #endif } + /* If we are not allowed to send an XML target description then we need + to use the hard-wired target descriptions. This corresponds to GDB's + default machine for x86. + + This check needs to occur before any returns statements that might + generate some alternative target descriptions. */ + if (!use_xml) + { +#ifdef __x86_64__ + if (machine == EM_X86_64) + return tdesc_amd64_linux_no_xml.get (); + else +#endif + return tdesc_i386_linux_no_xml.get (); + } + #if !defined __x86_64__ && defined HAVE_PTRACE_GETFPXREGS if (machine == EM_386 && have_ptrace_getfpxregs == -1) { @@ -885,17 +901,6 @@ x86_linux_read_description (void) } #endif - if (!use_xml) - { - /* Don't use XML. */ -#ifdef __x86_64__ - if (machine == EM_X86_64) - return tdesc_amd64_linux_no_xml.get (); - else -#endif - return tdesc_i386_linux_no_xml.get (); - } - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) { uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv6 3/9] gdb/x86: move have_ptrace_getfpxregs global into gdb/nat directory 2024-05-08 16:46 ` [PATCHv6 0/9] x86/Linux Target Description Changes Andrew Burgess 2024-05-08 16:46 ` [PATCHv6 1/9] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess 2024-05-08 16:46 ` [PATCHv6 2/9] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess @ 2024-05-08 16:46 ` Andrew Burgess 2024-05-08 22:52 ` John Baldwin 2024-05-08 16:46 ` [PATCHv6 4/9] gdb/x86: move have_ptrace_getregset " Andrew Burgess ` (6 subsequent siblings) 9 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-05-08 16:46 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, jhb The have_ptrace_getfpxregs global tracks whether GDB or gdbserver is running on a kernel that supports the GETFPXREGS ptrace request. Currently this global is declared twice (once in GDB and once in gdbserver), but it makes sense to move this global into the nat/ directory, and have a single declaration and definition. While moving this variable I have converted it to a tribool, as that was what it really was, if even used the same numbering as the tribool enum (-1, 0, 1). Where have_ptrace_getfpxregs was used I have updated in the obvious way. However, while making this change I noticed what I think is a bug in x86_linux_nat_target::read_description and x86_linux_read_description, both of these functions can be called multiple times, but in both cases we only end up calling i386_linux_read_description the first time through in the event that PTRACE_GETFPXREGS is not supported. This is because initially have_ptrace_getfpxregs will be TRIBOOL_UNKNOWN, but after the ptrace call fails we set have_ptrace_getfpxregs to TRIBOOL_FALSE. The next time we attempt to read the target description we'll skip the ptrace call, and so skip the call to i386_linux_read_description. I've not tried to address this preexisting bug in this commit, this is purely a refactor, there should be no user visible changes after this commit. In later commits I'll merge the gdbserver and GDB code together into the nat/ directory, and after that I'll try to address this bug. Reviewed-By: Felix Willgerodt <felix.willgerodt@intel.com> --- gdb/Makefile.in | 1 - gdb/i386-linux-nat.c | 26 +++++--------------------- gdb/i386-linux-nat.h | 26 -------------------------- gdb/nat/x86-linux.c | 9 +++++++++ gdb/nat/x86-linux.h | 14 ++++++++++++++ gdb/x86-linux-nat.c | 7 ++----- gdbserver/linux-x86-low.cc | 17 +++-------------- 7 files changed, 33 insertions(+), 67 deletions(-) delete mode 100644 gdb/i386-linux-nat.h diff --git a/gdb/Makefile.in b/gdb/Makefile.in index 234d0051f49..d1c9c2f2859 100644 --- a/gdb/Makefile.in +++ b/gdb/Makefile.in @@ -1397,7 +1397,6 @@ HFILES_NO_SRCDIR = \ hppa-tdep.h \ i386-bsd-nat.h \ i386-darwin-tdep.h \ - i386-linux-nat.h \ i386-linux-tdep.h \ i386-tdep.h \ i387-tdep.h \ diff --git a/gdb/i386-linux-nat.c b/gdb/i386-linux-nat.c index 7278dd91ff0..c54c9b8f9c1 100644 --- a/gdb/i386-linux-nat.c +++ b/gdb/i386-linux-nat.c @@ -26,7 +26,7 @@ #include "gregset.h" #include "gdb_proc_service.h" -#include "i386-linux-nat.h" +#include "nat/x86-linux.h" #include "i387-tdep.h" #include "i386-tdep.h" #include "i386-linux-tdep.h" @@ -80,22 +80,6 @@ int have_ptrace_getregs = 0 #endif ; - -/* Does the current host support the GETFPXREGS request? The header - file may or may not define it, and even if it is defined, the - kernel will return EIO if it's running on a pre-SSE processor. - - My instinct is to attach this to some architecture- or - target-specific data structure, but really, a particular GDB - process can only run on top of one kernel at a time. So it's okay - for this to be a simple variable. */ -int have_ptrace_getfpxregs = -#ifdef HAVE_PTRACE_GETFPXREGS - -1 -#else - 0 -#endif -; \f /* Accessing registers through the U area, one at a time. */ @@ -388,14 +372,14 @@ fetch_fpxregs (struct regcache *regcache, int tid) { elf_fpxregset_t fpxregs; - if (! have_ptrace_getfpxregs) + if (have_ptrace_getfpxregs == TRIBOOL_FALSE) return 0; if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) { if (errno == EIO) { - have_ptrace_getfpxregs = 0; + have_ptrace_getfpxregs = TRIBOOL_FALSE; return 0; } @@ -415,14 +399,14 @@ store_fpxregs (const struct regcache *regcache, int tid, int regno) { elf_fpxregset_t fpxregs; - if (! have_ptrace_getfpxregs) + if (have_ptrace_getfpxregs == TRIBOOL_FALSE) return 0; if (ptrace (PTRACE_GETFPXREGS, tid, 0, &fpxregs) == -1) { if (errno == EIO) { - have_ptrace_getfpxregs = 0; + have_ptrace_getfpxregs = TRIBOOL_FALSE; return 0; } diff --git a/gdb/i386-linux-nat.h b/gdb/i386-linux-nat.h deleted file mode 100644 index c3d32aff1be..00000000000 --- a/gdb/i386-linux-nat.h +++ /dev/null @@ -1,26 +0,0 @@ -/* Native-dependent code for GNU/Linux i386. - - Copyright (C) 1999-2024 Free Software Foundation, Inc. - - This file is part of GDB. - - This program is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published by - the Free Software Foundation; either version 3 of the License, or - (at your option) any later version. - - This program is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - GNU General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see <http://www.gnu.org/licenses/>. */ - -#ifndef I386_LINUX_NAT_H -#define I386_LINUX_NAT_H 1 - -/* Does the current host support the GETFPXREGS request? */ -extern int have_ptrace_getfpxregs; - -#endif diff --git a/gdb/nat/x86-linux.c b/gdb/nat/x86-linux.c index 7a21c8f2c26..f394dc5d5e1 100644 --- a/gdb/nat/x86-linux.c +++ b/gdb/nat/x86-linux.c @@ -20,6 +20,15 @@ #include "x86-linux.h" #include "x86-linux-dregs.h" +/* See nat/x86-linux.h. */ +tribool have_ptrace_getfpxregs = +#ifdef HAVE_PTRACE_GETFPXREGS + TRIBOOL_UNKNOWN +#else + TRIBOOL_FALSE +#endif +; + /* Per-thread arch-specific data we want to keep. */ struct arch_lwp_info diff --git a/gdb/nat/x86-linux.h b/gdb/nat/x86-linux.h index 822882173f9..936e20154a0 100644 --- a/gdb/nat/x86-linux.h +++ b/gdb/nat/x86-linux.h @@ -22,6 +22,20 @@ #include "nat/linux-nat.h" +/* Does the current host support the GETFPXREGS request? The system header + file may or may not define it, but even if it is defined, the kernel + will return EIO if it's running on a pre-SSE processor. + + Initially this will be TRIBOOL_UNKNOWN and should be changed to + TRIBOOL_FALSE if the ptrace call is attempted and fails or changed to + TRIBOOL_TRUE if the ptrace call is attempted and succeeds. + + My instinct is to attach this to some architecture- or target-specific + data structure, but really, a particular GDB process can only run on top + of one kernel at a time. So it's okay - for this to be a global + variable. */ +extern tribool have_ptrace_getfpxregs; + /* Set whether our local mirror of LWP's debug registers has been changed since the values were last written to the thread. Nonzero indicates that a change has been made, zero indicates no change. */ diff --git a/gdb/x86-linux-nat.c b/gdb/x86-linux-nat.c index a3d8ffb60f1..b42884ad57f 100644 --- a/gdb/x86-linux-nat.c +++ b/gdb/x86-linux-nat.c @@ -26,9 +26,6 @@ #include <sys/uio.h> #include "x86-nat.h" -#ifndef __x86_64__ -#include "i386-linux-nat.h" -#endif #include "x86-linux-nat.h" #include "i386-linux-tdep.h" #ifdef __x86_64__ @@ -147,13 +144,13 @@ x86_linux_nat_target::read_description () error (_("Can't debug 64-bit process with 32-bit GDB")); } #elif HAVE_PTRACE_GETFPXREGS - if (have_ptrace_getfpxregs == -1) + if (have_ptrace_getfpxregs == TRIBOOL_UNKNOWN) { elf_fpxregset_t fpxregs; if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) { - have_ptrace_getfpxregs = 0; + have_ptrace_getfpxregs = TRIBOOL_FALSE; have_ptrace_getregset = TRIBOOL_FALSE; return i386_linux_read_description (X86_XSTATE_X87_MASK); } diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index e8ef3667eb4..7937055c060 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -831,17 +831,6 @@ x86_target::low_siginfo_fixup (siginfo_t *ptrace, gdb_byte *inf, int direction) \f static int use_xml; -/* Does the current host support the GETFPXREGS request? The header - file may or may not define it, and even if it is defined, the - kernel will return EIO if it's running on a pre-SSE processor. */ -int have_ptrace_getfpxregs = -#ifdef HAVE_PTRACE_GETFPXREGS - -1 -#else - 0 -#endif -; - /* Get Linux/x86 target description from running target. */ static const struct target_desc * @@ -886,18 +875,18 @@ x86_linux_read_description (void) } #if !defined __x86_64__ && defined HAVE_PTRACE_GETFPXREGS - if (machine == EM_386 && have_ptrace_getfpxregs == -1) + if (machine == EM_386 && have_ptrace_getfpxregs == TRIBOOL_UNKNOWN) { elf_fpxregset_t fpxregs; if (ptrace (PTRACE_GETFPXREGS, tid, 0, (long) &fpxregs) < 0) { - have_ptrace_getfpxregs = 0; + have_ptrace_getfpxregs = TRIBOOL_FALSE; have_ptrace_getregset = TRIBOOL_FALSE; return i386_linux_read_description (X86_XSTATE_X87); } else - have_ptrace_getfpxregs = 1; + have_ptrace_getfpxregs = TRIBOOL_TRUE; } #endif -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv6 3/9] gdb/x86: move have_ptrace_getfpxregs global into gdb/nat directory 2024-05-08 16:46 ` [PATCHv6 3/9] gdb/x86: move have_ptrace_getfpxregs global into gdb/nat directory Andrew Burgess @ 2024-05-08 22:52 ` John Baldwin 0 siblings, 0 replies; 116+ messages in thread From: John Baldwin @ 2024-05-08 22:52 UTC (permalink / raw) To: Andrew Burgess, gdb-patches; +Cc: felix.willgerodt On 5/8/24 9:46 AM, Andrew Burgess wrote: > The have_ptrace_getfpxregs global tracks whether GDB or gdbserver is > running on a kernel that supports the GETFPXREGS ptrace request. > > Currently this global is declared twice (once in GDB and once in > gdbserver), but it makes sense to move this global into the nat/ > directory, and have a single declaration and definition. > > While moving this variable I have converted it to a tribool, as that > was what it really was, if even used the same numbering as the tribool > enum (-1, 0, 1). Where have_ptrace_getfpxregs was used I have updated > in the obvious way. > > However, while making this change I noticed what I think is a bug in > x86_linux_nat_target::read_description and x86_linux_read_description, > both of these functions can be called multiple times, but in both > cases we only end up calling i386_linux_read_description the first > time through in the event that PTRACE_GETFPXREGS is not supported. > This is because initially have_ptrace_getfpxregs will be > TRIBOOL_UNKNOWN, but after the ptrace call fails we set > have_ptrace_getfpxregs to TRIBOOL_FALSE. The next time we attempt to > read the target description we'll skip the ptrace call, and so skip > the call to i386_linux_read_description. > > I've not tried to address this preexisting bug in this commit, this is > purely a refactor, there should be no user visible changes after this > commit. In later commits I'll merge the gdbserver and GDB code > together into the nat/ directory, and after that I'll try to address > this bug. > > Reviewed-By: Felix Willgerodt <felix.willgerodt@intel.com> > --- > gdb/Makefile.in | 1 - > gdb/i386-linux-nat.c | 26 +++++--------------------- > gdb/i386-linux-nat.h | 26 -------------------------- > gdb/nat/x86-linux.c | 9 +++++++++ > gdb/nat/x86-linux.h | 14 ++++++++++++++ > gdb/x86-linux-nat.c | 7 ++----- > gdbserver/linux-x86-low.cc | 17 +++-------------- > 7 files changed, 33 insertions(+), 67 deletions(-) > delete mode 100644 gdb/i386-linux-nat.h > > diff --git a/gdb/nat/x86-linux.c b/gdb/nat/x86-linux.c > index 7a21c8f2c26..f394dc5d5e1 100644 > --- a/gdb/nat/x86-linux.c > +++ b/gdb/nat/x86-linux.c > @@ -20,6 +20,15 @@ > #include "x86-linux.h" > #include "x86-linux-dregs.h" > > +/* See nat/x86-linux.h. */ > +tribool have_ptrace_getfpxregs = > +#ifdef HAVE_PTRACE_GETFPXREGS > + TRIBOOL_UNKNOWN > +#else > + TRIBOOL_FALSE > +#endif > +; > + Should this be under #ifdef __i386__ or the like? My understanding is that this request isn't supported for Linux x86-64, just for i386? -- John Baldwin ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv6 4/9] gdb/x86: move have_ptrace_getregset global into gdb/nat directory 2024-05-08 16:46 ` [PATCHv6 0/9] x86/Linux Target Description Changes Andrew Burgess ` (2 preceding siblings ...) 2024-05-08 16:46 ` [PATCHv6 3/9] gdb/x86: move have_ptrace_getfpxregs global into gdb/nat directory Andrew Burgess @ 2024-05-08 16:46 ` Andrew Burgess 2024-05-08 22:53 ` John Baldwin 2024-05-08 16:46 ` [PATCHv6 5/9] gdb/x86: move reading of cs and ds state " Andrew Burgess ` (5 subsequent siblings) 9 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-05-08 16:46 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, jhb Just like the previous commit, this commit moves have_ptrace_getregset into the nat/ directory. This global is currently declared and defined in both GDB and gdbserver, but this can be merged and shared. There should be no user visible changes after this commit. --- gdb/linux-nat.c | 3 --- gdb/linux-nat.h | 3 --- gdb/nat/x86-linux.c | 3 +++ gdb/nat/x86-linux.h | 3 +++ gdbserver/linux-low.cc | 3 --- gdbserver/linux-low.h | 2 -- 6 files changed, 6 insertions(+), 11 deletions(-) diff --git a/gdb/linux-nat.c b/gdb/linux-nat.c index 48ecd3627ca..38d3af56f39 100644 --- a/gdb/linux-nat.c +++ b/gdb/linux-nat.c @@ -243,9 +243,6 @@ PTRACE_PEEKTEXT/PTRACE_POKETEXT or process_vm_readv/process_vm_writev: struct linux_nat_target *linux_target; -/* Does the current host support PTRACE_GETREGSET? */ -enum tribool have_ptrace_getregset = TRIBOOL_UNKNOWN; - /* When true, print debug messages relating to the linux native target. */ static bool debug_linux_nat; diff --git a/gdb/linux-nat.h b/gdb/linux-nat.h index 4dcbe9e170a..fd393df5020 100644 --- a/gdb/linux-nat.h +++ b/gdb/linux-nat.h @@ -299,9 +299,6 @@ lwp_info_range all_lwps (); lwp_info_safe_range all_lwps_safe (); -/* Does the current host support PTRACE_GETREGSET? */ -extern enum tribool have_ptrace_getregset; - /* Called from the LWP layer to inform the thread_db layer that PARENT spawned CHILD. Both LWPs are currently stopped. This function does whatever is required to have the child LWP under the diff --git a/gdb/nat/x86-linux.c b/gdb/nat/x86-linux.c index f394dc5d5e1..ed1d06bd623 100644 --- a/gdb/nat/x86-linux.c +++ b/gdb/nat/x86-linux.c @@ -29,6 +29,9 @@ tribool have_ptrace_getfpxregs = #endif ; +/* See nat/x86-linux.h. */ +tribool have_ptrace_getregset = TRIBOOL_UNKNOWN; + /* Per-thread arch-specific data we want to keep. */ struct arch_lwp_info diff --git a/gdb/nat/x86-linux.h b/gdb/nat/x86-linux.h index 936e20154a0..cb5f44f4b49 100644 --- a/gdb/nat/x86-linux.h +++ b/gdb/nat/x86-linux.h @@ -36,6 +36,9 @@ variable. */ extern tribool have_ptrace_getfpxregs; +/* Does the current host support PTRACE_GETREGSET? */ +extern tribool have_ptrace_getregset; + /* Set whether our local mirror of LWP's debug registers has been changed since the values were last written to the thread. Nonzero indicates that a change has been made, zero indicates no change. */ diff --git a/gdbserver/linux-low.cc b/gdbserver/linux-low.cc index ac7f9807ecc..4ba56cd34ff 100644 --- a/gdbserver/linux-low.cc +++ b/gdbserver/linux-low.cc @@ -133,9 +133,6 @@ typedef struct } Elf64_auxv_t; #endif -/* Does the current host support PTRACE_GETREGSET? */ -enum tribool have_ptrace_getregset = TRIBOOL_UNKNOWN; - /* Return TRUE if THREAD is the leader thread of the process. */ static bool diff --git a/gdbserver/linux-low.h b/gdbserver/linux-low.h index eaf87527338..273c04626b8 100644 --- a/gdbserver/linux-low.h +++ b/gdbserver/linux-low.h @@ -951,8 +951,6 @@ void thread_db_notice_clone (struct thread_info *parent_thr, ptid_t child_ptid); bool thread_db_thread_handle (ptid_t ptid, gdb_byte **handle, int *handle_len); -extern enum tribool have_ptrace_getregset; - /* Search for the value with type MATCH in the auxv vector, with entries of length WORDSIZE bytes, of process with pid PID. If found, store the value in *VALP and return 1. If not found or if there is an error, -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv6 4/9] gdb/x86: move have_ptrace_getregset global into gdb/nat directory 2024-05-08 16:46 ` [PATCHv6 4/9] gdb/x86: move have_ptrace_getregset " Andrew Burgess @ 2024-05-08 22:53 ` John Baldwin 0 siblings, 0 replies; 116+ messages in thread From: John Baldwin @ 2024-05-08 22:53 UTC (permalink / raw) To: Andrew Burgess, gdb-patches; +Cc: felix.willgerodt On 5/8/24 9:46 AM, Andrew Burgess wrote: > Just like the previous commit, this commit moves have_ptrace_getregset > into the nat/ directory. This global is currently declared and > defined in both GDB and gdbserver, but this can be merged and shared. > > There should be no user visible changes after this commit. Approved-By: John Baldwin <jhb@FreeBSD.org> -- John Baldwin ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv6 5/9] gdb/x86: move reading of cs and ds state into gdb/nat directory 2024-05-08 16:46 ` [PATCHv6 0/9] x86/Linux Target Description Changes Andrew Burgess ` (3 preceding siblings ...) 2024-05-08 16:46 ` [PATCHv6 4/9] gdb/x86: move have_ptrace_getregset " Andrew Burgess @ 2024-05-08 16:46 ` Andrew Burgess 2024-05-08 16:46 ` [PATCHv6 6/9] gdb: move xcr0 == 0 check into i386_linux_core_read_description Andrew Burgess ` (4 subsequent siblings) 9 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-08 16:46 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, jhb This patch is part of a series that has the aim sharing the x86 Linux target description creation code between GDB and gdbserver. Within GDB part of this process involves reading the cs and ds state from the 'struct user_regs_struct' using a ptrace call. This isn't done by gdbserver, which is part of the motivation for this whole series; the approach gdbserver takes is inferior to the approach GDB takes (gdbserver relies on reading the file being debugged, and extracting similar information from the file headers). This commit moves the reading of cs and ds, which is used to figure out if a thread is 32-bit or 64-bit (or in x32 mode), into the gdb/nat directory so that the code can be shared with gdbserver, but at this point I'm not actually using the code in gdbserver, that will come later. As such there should be no user visible changes after this commit, GDB continues to do things as it did before (reading cs/ds), while gdbserver continues to use its own approach (which doesn't require reading cs/ds). Approved-By: John Baldwin <jhb@FreeBSD.org> Reviewed-By: Felix Willgerodt <felix.willgerodt@intel.com> --- gdb/nat/x86-linux.c | 47 +++++++++++++++++++++++++++++++++++++++++++++ gdb/nat/x86-linux.h | 28 +++++++++++++++++++++++++++ gdb/x86-linux-nat.c | 45 ++++++++----------------------------------- 3 files changed, 83 insertions(+), 37 deletions(-) diff --git a/gdb/nat/x86-linux.c b/gdb/nat/x86-linux.c index ed1d06bd623..145684367d7 100644 --- a/gdb/nat/x86-linux.c +++ b/gdb/nat/x86-linux.c @@ -19,6 +19,8 @@ #include "x86-linux.h" #include "x86-linux-dregs.h" +#include "nat/gdb_ptrace.h" +#include <sys/user.h> /* See nat/x86-linux.h. */ tribool have_ptrace_getfpxregs = @@ -91,3 +93,48 @@ x86_linux_prepare_to_resume (struct lwp_info *lwp) { x86_linux_update_debug_registers (lwp); } + +#ifdef __x86_64__ +/* Value of CS segment register: + 64bit process: 0x33 + 32bit process: 0x23 */ +#define AMD64_LINUX_USER64_CS 0x33 + +/* Value of DS segment register: + LP64 process: 0x0 + X32 process: 0x2b */ +#define AMD64_LINUX_X32_DS 0x2b +#endif + +/* See nat/x86-linux.h. */ + +x86_linux_arch_size +x86_linux_ptrace_get_arch_size (int tid) +{ +#ifdef __x86_64__ + unsigned long cs; + unsigned long ds; + + /* Get CS register. */ + errno = 0; + cs = ptrace (PTRACE_PEEKUSER, tid, + offsetof (struct user_regs_struct, cs), 0); + if (errno != 0) + perror_with_name (_("Couldn't get CS register")); + + bool is_64bit = cs == AMD64_LINUX_USER64_CS; + + /* Get DS register. */ + errno = 0; + ds = ptrace (PTRACE_PEEKUSER, tid, + offsetof (struct user_regs_struct, ds), 0); + if (errno != 0) + perror_with_name (_("Couldn't get DS register")); + + bool is_x32 = ds == AMD64_LINUX_X32_DS; + + return x86_linux_arch_size (is_64bit, is_x32); +#else + return x86_linux_arch_size (false, false); +#endif +} diff --git a/gdb/nat/x86-linux.h b/gdb/nat/x86-linux.h index cb5f44f4b49..a76c5092f52 100644 --- a/gdb/nat/x86-linux.h +++ b/gdb/nat/x86-linux.h @@ -64,4 +64,32 @@ extern void x86_linux_delete_thread (struct arch_lwp_info *arch_lwp); extern void x86_linux_prepare_to_resume (struct lwp_info *lwp); +/* Return value from x86_linux_ptrace_get_arch_size function. Indicates if + a thread is 32-bit, 64-bit, or x32. */ + +struct x86_linux_arch_size +{ + explicit x86_linux_arch_size (bool is_64bit, bool is_x32) + : m_is_64bit (is_64bit), + m_is_x32 (is_x32) + { + /* Nothing. */ + } + + bool is_64bit () const + { return m_is_64bit; } + + bool is_x32 () const + { return m_is_x32; } + +private: + bool m_is_64bit = false; + bool m_is_x32 = false; +}; + +/* Use ptrace calls to figure out if thread TID is 32-bit, 64-bit, or + 64-bit running in x32 mode. */ + +extern x86_linux_arch_size x86_linux_ptrace_get_arch_size (int tid); + #endif /* NAT_X86_LINUX_H */ diff --git a/gdb/x86-linux-nat.c b/gdb/x86-linux-nat.c index b42884ad57f..285a4880103 100644 --- a/gdb/x86-linux-nat.c +++ b/gdb/x86-linux-nat.c @@ -87,18 +87,6 @@ x86_linux_nat_target::post_startup_inferior (ptid_t ptid) linux_nat_target::post_startup_inferior (ptid); } -#ifdef __x86_64__ -/* Value of CS segment register: - 64bit process: 0x33 - 32bit process: 0x23 */ -#define AMD64_LINUX_USER64_CS 0x33 - -/* Value of DS segment register: - LP64 process: 0x0 - X32 process: 0x2b */ -#define AMD64_LINUX_X32_DS 0x2b -#endif - /* Get Linux/x86 target description from running target. */ const struct target_desc * @@ -118,31 +106,14 @@ x86_linux_nat_target::read_description () tid = inferior_ptid.pid (); #ifdef __x86_64__ - { - unsigned long cs; - unsigned long ds; - - /* Get CS register. */ - errno = 0; - cs = ptrace (PTRACE_PEEKUSER, tid, - offsetof (struct user_regs_struct, cs), 0); - if (errno != 0) - perror_with_name (_("Couldn't get CS register")); - - is_64bit = cs == AMD64_LINUX_USER64_CS; - - /* Get DS register. */ - errno = 0; - ds = ptrace (PTRACE_PEEKUSER, tid, - offsetof (struct user_regs_struct, ds), 0); - if (errno != 0) - perror_with_name (_("Couldn't get DS register")); - - is_x32 = ds == AMD64_LINUX_X32_DS; - - if (sizeof (void *) == 4 && is_64bit && !is_x32) - error (_("Can't debug 64-bit process with 32-bit GDB")); - } + + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); + is_64bit = arch_size.is_64bit (); + is_x32 = arch_size.is_x32 (); + + if (sizeof (void *) == 4 && is_64bit && !is_x32) + error (_("Can't debug 64-bit process with 32-bit GDB")); + #elif HAVE_PTRACE_GETFPXREGS if (have_ptrace_getfpxregs == TRIBOOL_UNKNOWN) { -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv6 6/9] gdb: move xcr0 == 0 check into i386_linux_core_read_description 2024-05-08 16:46 ` [PATCHv6 0/9] x86/Linux Target Description Changes Andrew Burgess ` (4 preceding siblings ...) 2024-05-08 16:46 ` [PATCHv6 5/9] gdb/x86: move reading of cs and ds state " Andrew Burgess @ 2024-05-08 16:46 ` Andrew Burgess 2024-05-08 22:54 ` John Baldwin 2024-05-08 16:46 ` [PATCHv6 7/9] gdb/gdbserver: share some code relating to target description creation Andrew Burgess ` (3 subsequent siblings) 9 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-05-08 16:46 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, jhb Currently, in i386_linux_core_read_description, if GDB fails to extract an xcr0 value from the core file, then we will have a default zero value for the xcr0 variable, we still call the i386_linux_read_description function, which checks for this zero value and returns nullptr. Back in i386_linux_core_read_description we spot the nullptr return value from i386_linux_read_description and call i386_linux_read_description again, but this time passing a default value for xcr0. In the next commit I plan to rework i386_linux_read_description, and in so doing I will remove the check for xcr0 == 0, this is inline with how the amd64 code is written. However, this means that the 'xcr0 == 0' check needs to move up the stack to i386_linux_core_read_description, again, this brings the i386 code into line with the amd64 code. This is just a refactor in preparation for the next commit, there should be no user visible changes after this commit. Approved-By: Felix Willgerodt <felix.willgerodt@intel.com> --- gdb/i386-linux-tdep.c | 15 ++++++++------- 1 file changed, 8 insertions(+), 7 deletions(-) diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index a78f03fac8d..511e43f3b6f 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -714,15 +714,16 @@ i386_linux_core_read_description (struct gdbarch *gdbarch, /* Linux/i386. */ x86_xsave_layout layout; uint64_t xcr0 = i386_linux_core_read_xsave_info (abfd, layout); - const struct target_desc *tdesc = i386_linux_read_description (xcr0); - if (tdesc != NULL) - return tdesc; + if (xcr0 == 0) + { + if (bfd_get_section_by_name (abfd, ".reg-xfp") != nullptr) + xcr0 = X86_XSTATE_SSE_MASK; + else + xcr0 = X86_XSTATE_X87_MASK; + } - if (bfd_get_section_by_name (abfd, ".reg-xfp") != NULL) - return i386_linux_read_description (X86_XSTATE_SSE_MASK); - else - return i386_linux_read_description (X86_XSTATE_X87_MASK); + return i386_linux_read_description (xcr0); } /* Similar to i386_supply_fpregset, but use XSAVE extended state. */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv6 6/9] gdb: move xcr0 == 0 check into i386_linux_core_read_description 2024-05-08 16:46 ` [PATCHv6 6/9] gdb: move xcr0 == 0 check into i386_linux_core_read_description Andrew Burgess @ 2024-05-08 22:54 ` John Baldwin 0 siblings, 0 replies; 116+ messages in thread From: John Baldwin @ 2024-05-08 22:54 UTC (permalink / raw) To: Andrew Burgess, gdb-patches; +Cc: felix.willgerodt On 5/8/24 9:46 AM, Andrew Burgess wrote: > Currently, in i386_linux_core_read_description, if GDB fails to > extract an xcr0 value from the core file, then we will have a default > zero value for the xcr0 variable, we still call the > i386_linux_read_description function, which checks for this zero value > and returns nullptr. > > Back in i386_linux_core_read_description we spot the nullptr return > value from i386_linux_read_description and call > i386_linux_read_description again, but this time passing a default > value for xcr0. > > In the next commit I plan to rework i386_linux_read_description, and > in so doing I will remove the check for xcr0 == 0, this is inline with > how the amd64 code is written. > > However, this means that the 'xcr0 == 0' check needs to move up the > stack to i386_linux_core_read_description, again, this brings the i386 > code into line with the amd64 code. > > This is just a refactor in preparation for the next commit, there > should be no user visible changes after this commit. > > Approved-By: Felix Willgerodt <felix.willgerodt@intel.com> Approved-By: John Baldwin <jhb@FreeBSD.org> -- John Baldwin ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv6 7/9] gdb/gdbserver: share some code relating to target description creation 2024-05-08 16:46 ` [PATCHv6 0/9] x86/Linux Target Description Changes Andrew Burgess ` (5 preceding siblings ...) 2024-05-08 16:46 ` [PATCHv6 6/9] gdb: move xcr0 == 0 check into i386_linux_core_read_description Andrew Burgess @ 2024-05-08 16:46 ` Andrew Burgess 2024-05-08 22:58 ` John Baldwin 2024-05-08 16:46 ` [PATCHv6 8/9] gdbserver: update target description creation for x86/linux Andrew Burgess ` (2 subsequent siblings) 9 siblings, 1 reply; 116+ messages in thread From: Andrew Burgess @ 2024-05-08 16:46 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, jhb This commit is part of a series to share more of the x86 target description creation code between GDB and gdbserver. Unlike previous commits which were mostly refactoring, this commit is the first that makes a real change, though that change should mostly be for gdbserver; I've largely adopted the "GDB" way of doing things for gdbserver, and this fixes a real gdbserver bug. On a x86-64 Linux target, running the test: gdb.server/connect-with-no-symbol-file.exp results in two core files being created. Both of these core files are from the inferior process, created after gdbserver has detached. In this test a gdbserver process is started and then, after gdbserver has started, but before GDB attaches, we either delete the inferior executable, or change its permissions so it can't be read. Only after doing this do we attempt to connect with GDB. As GDB connects to gdbserver, gdbserver attempts to figure out the target description so that it can send the description to GDB, this involves a call to x86_linux_read_description. In x86_linux_read_description one of the first things we do is try to figure out if the process is 32-bit or 64-bit. To do this we look up the executable via the thread-id, and then attempt to read the architecture size from the executable. This isn't going to work if the executable has been deleted, or is no longer readable. And so, as we can't read the executable, we default to an i386 target and use an i386 target description. A consequence of using an i386 target description is that addresses are assumed to be 32-bits. Here's an example session that shows the problems this causes. This is run on an x86-64 machine, and the test binary (xx.x) is a standard 64-bit x86-64 binary: shell_1$ gdbserver --once localhost :54321 /tmp/xx.x shell_2$ gdb -q (gdb) set sysroot (gdb) shell chmod 000 /tmp/xx.x (gdb) target remote :54321 Remote debugging using :54321 warning: /tmp/xx.x: Permission denied. 0xf7fd3110 in ?? () (gdb) show architecture The target architecture is set to "auto" (currently "i386"). (gdb) p/x $pc $1 = 0xf7fd3110 (gdb) info proc mappings process 2412639 Mapped address spaces: Start Addr End Addr Size Offset Perms objfile 0x400000 0x401000 0x1000 0x0 r--p /tmp/xx.x 0x401000 0x402000 0x1000 0x1000 r-xp /tmp/xx.x 0x402000 0x403000 0x1000 0x2000 r--p /tmp/xx.x 0x403000 0x405000 0x2000 0x2000 rw-p /tmp/xx.x 0xf7fcb000 0xf7fcf000 0x4000 0x0 r--p [vvar] 0xf7fcf000 0xf7fd1000 0x2000 0x0 r-xp [vdso] 0xf7fd1000 0xf7fd3000 0x2000 0x0 r--p /usr/lib64/ld-2.30.so 0xf7fd3000 0xf7ff3000 0x20000 0x2000 r-xp /usr/lib64/ld-2.30.so 0xf7ff3000 0xf7ffb000 0x8000 0x22000 r--p /usr/lib64/ld-2.30.so 0xf7ffc000 0xf7ffe000 0x2000 0x2a000 rw-p /usr/lib64/ld-2.30.so 0xf7ffe000 0xf7fff000 0x1000 0x0 rw-p 0xfffda000 0xfffff000 0x25000 0x0 rw-p [stack] 0xff600000 0xff601000 0x1000 0x0 r-xp [vsyscall] (gdb) info inferiors Num Description Connection Executable * 1 process 2412639 1 (remote :54321) (gdb) shell cat /proc/2412639/maps 00400000-00401000 r--p 00000000 fd:03 45907133 /tmp/xx.x 00401000-00402000 r-xp 00001000 fd:03 45907133 /tmp/xx.x 00402000-00403000 r--p 00002000 fd:03 45907133 /tmp/xx.x 00403000-00405000 rw-p 00002000 fd:03 45907133 /tmp/xx.x 7ffff7fcb000-7ffff7fcf000 r--p 00000000 00:00 0 [vvar] 7ffff7fcf000-7ffff7fd1000 r-xp 00000000 00:00 0 [vdso] 7ffff7fd1000-7ffff7fd3000 r--p 00000000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7fd3000-7ffff7ff3000 r-xp 00002000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7ff3000-7ffff7ffb000 r--p 00022000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7ffc000-7ffff7ffe000 rw-p 0002a000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7ffe000-7ffff7fff000 rw-p 00000000 00:00 0 7ffffffda000-7ffffffff000 rw-p 00000000 00:00 0 [stack] ffffffffff600000-ffffffffff601000 r-xp 00000000 00:00 0 [vsyscall] (gdb) Notice the difference between the mappings reported via GDB and those reported directly from the kernel via /proc/PID/maps, the addresses of every mapping is clamped to 32-bits for GDB, while the kernel reports real 64-bit addresses. Notice also that the $pc value is a 32-bit value. It appears to be within one of the mappings reported by GDB, but is outside any of the mappings reported from the kernel. And this is where the problem arises. When gdbserver detaches from the inferior we pass the inferior the address from which it should resume. Due to the 32/64 bit confusion we tell the inferior to resume from the 32-bit $pc value, which is not within any valid mapping, and so, as soon as the inferior resumes, it segfaults. If we look at how GDB (not gdbserver) figures out its target description then we see an interesting difference. GDB doesn't try to read the executable. Instead GDB uses ptrace to query the thread's state, and uses this to figure out the if the thread is 32 or 64 bit. If we update gdbserver to do it the "GDB" way then the above problem is resolved, gdbserver now sees the process as 64-bit, and when we detach from the inferior we give it the correct 64-bit address, and the inferior no longer segfaults. Now, I could just update the gdbserver code, but better, I think, to share one copy of the code between GDB and gdbserver in gdb/nat/. That is what this commit does. The cores of x86_linux_read_description from gdbserver and x86_linux_nat_target::read_description from GDB are moved into a new file gdb/nat/x86-linux-tdesc.c and combined into a single function x86_linux_tdesc_for_tid which is called from each location. This new function does things mostly the GDB way, some changes are needed to allow for the sharing; we now take some pointers for where the shared code can cache the xcr0 and xsave layout values. Another thing to note about this commit is how the functions i386_linux_read_description and amd64_linux_read_description are handled. For now I've left these function as implemented separately in GDB and gdbserver. I've moved the declarations of these functions into gdb/arch/{i386,amd64}-linux-tdesc.h, but the implementations are left where they are. A later commit in this series will make these functions shared too, but doing this is not trivial, so I've left that for a separate commit. Merging the declarations as I've done here ensures that everyone implements the function to the same API, and once these functions are shared (in a later commit) we'll want a shared declaration anyway. Reviewed-By: Felix Willgerodt <felix.willgerodt@intel.com> --- gdb/Makefile.in | 3 + gdb/amd64-linux-tdep.c | 1 + gdb/amd64-linux-tdep.h | 6 -- gdb/arch/amd64-linux-tdesc.h | 30 +++++++ gdb/arch/i386-linux-tdesc.h | 29 +++++++ gdb/configure.nat | 4 +- gdb/i386-linux-tdep.c | 1 + gdb/i386-linux-tdep.h | 3 - gdb/nat/x86-linux-tdesc.c | 128 ++++++++++++++++++++++++++++++ gdb/nat/x86-linux-tdesc.h | 54 +++++++++++++ gdb/x86-linux-nat.c | 88 ++------------------- gdbserver/configure.srv | 2 + gdbserver/i387-fp.cc | 9 +-- gdbserver/i387-fp.h | 4 +- gdbserver/linux-amd64-ipa.cc | 1 + gdbserver/linux-i386-ipa.cc | 1 + gdbserver/linux-x86-low.cc | 147 ++++++++++------------------------- gdbserver/linux-x86-tdesc.cc | 2 + gdbserver/linux-x86-tdesc.h | 7 -- 19 files changed, 310 insertions(+), 210 deletions(-) create mode 100644 gdb/arch/amd64-linux-tdesc.h create mode 100644 gdb/arch/i386-linux-tdesc.h create mode 100644 gdb/nat/x86-linux-tdesc.c create mode 100644 gdb/nat/x86-linux-tdesc.h diff --git a/gdb/Makefile.in b/gdb/Makefile.in index d1c9c2f2859..9c2c69b02ab 100644 --- a/gdb/Makefile.in +++ b/gdb/Makefile.in @@ -1550,8 +1550,10 @@ HFILES_NO_SRCDIR = \ arch/aarch64-insn.h \ arch/aarch64-mte-linux.h \ arch/aarch64-scalable-linux.h \ + arch/amd64-linux-tdesc.h \ arch/arc.h \ arch/arm.h \ + arch/i386-linux-tdesc.h \ arch/i386.h \ arch/loongarch.h \ arch/ppc-linux-common.h \ @@ -1607,6 +1609,7 @@ HFILES_NO_SRCDIR = \ nat/x86-gcc-cpuid.h \ nat/x86-linux.h \ nat/x86-linux-dregs.h \ + nat/x86-linux-tdesc.h \ python/py-event.h \ python/py-events.h \ python/py-stopevent.h \ diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c index c52b0436872..f9647dce1e6 100644 --- a/gdb/amd64-linux-tdep.c +++ b/gdb/amd64-linux-tdep.c @@ -42,6 +42,7 @@ #include "arch/amd64.h" #include "target-descriptions.h" #include "expop.h" +#include "arch/amd64-linux-tdesc.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_AMD64 "syscalls/amd64-linux.xml" diff --git a/gdb/amd64-linux-tdep.h b/gdb/amd64-linux-tdep.h index 2003dcda78f..0ec49e7fe03 100644 --- a/gdb/amd64-linux-tdep.h +++ b/gdb/amd64-linux-tdep.h @@ -43,12 +43,6 @@ extern struct target_desc *tdesc_x32_linux; extern struct target_desc *tdesc_x32_avx_linux; extern struct target_desc *tdesc_x32_avx_avx512_linux; -/* Return the right amd64-linux target descriptions according to - XCR0_FEATURES_BIT and IS_X32. */ - -const target_desc *amd64_linux_read_description (uint64_t xcr0_features_bit, - bool is_x32); - /* Enum that defines the syscall identifiers for amd64 linux. Used for process record/replay, these will be translated into a gdb-canonical set of syscall ids in linux-record.c. */ diff --git a/gdb/arch/amd64-linux-tdesc.h b/gdb/arch/amd64-linux-tdesc.h new file mode 100644 index 00000000000..db425b60df6 --- /dev/null +++ b/gdb/arch/amd64-linux-tdesc.h @@ -0,0 +1,30 @@ +/* Target description related code for GNU/Linux x86-64. + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#ifndef ARCH_AMD64_LINUX_TDESC_H +#define ARCH_AMD64_LINUX_TDESC_H + +struct target_desc; + +/* Return the AMD64 target descriptions corresponding to XCR0 and IS_X32. */ + +extern const target_desc *amd64_linux_read_description (uint64_t xcr0, + bool is_x32); + +#endif /* ARCH_AMD64_LINUX_TDESC_H */ diff --git a/gdb/arch/i386-linux-tdesc.h b/gdb/arch/i386-linux-tdesc.h new file mode 100644 index 00000000000..0b736337a75 --- /dev/null +++ b/gdb/arch/i386-linux-tdesc.h @@ -0,0 +1,29 @@ +/* Target description related code for GNU/Linux i386. + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#ifndef ARCH_I386_LINUX_TDESC_H +#define ARCH_I386_LINUX_TDESC_H + +struct target_desc; + +/* Return the i386 target description corresponding to XCR0. */ + +extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); + +#endif /* ARCH_I386_LINUX_TDESC_H */ diff --git a/gdb/configure.nat b/gdb/configure.nat index 8b98511cef7..4bcc0696027 100644 --- a/gdb/configure.nat +++ b/gdb/configure.nat @@ -256,7 +256,7 @@ case ${gdb_host} in NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ nat/x86-xstate.o \ i386-linux-nat.o x86-linux-nat.o nat/linux-btrace.o \ - nat/x86-linux.o nat/x86-linux-dregs.o" + nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o" ;; ia64) # Host: Intel IA-64 running GNU/Linux @@ -322,7 +322,7 @@ case ${gdb_host} in NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ nat/x86-xstate.o amd64-nat.o amd64-linux-nat.o x86-linux-nat.o \ nat/linux-btrace.o \ - nat/x86-linux.o nat/x86-linux-dregs.o \ + nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o \ nat/amd64-linux-siginfo.o" ;; sparc) diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index 511e43f3b6f..6a7490680c1 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -40,6 +40,7 @@ #include "i387-tdep.h" #include "gdbsupport/x86-xstate.h" +#include "arch/i386-linux-tdesc.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_I386 "syscalls/i386-linux.xml" diff --git a/gdb/i386-linux-tdep.h b/gdb/i386-linux-tdep.h index 07593c6a8ec..e8691cd778e 100644 --- a/gdb/i386-linux-tdep.h +++ b/gdb/i386-linux-tdep.h @@ -55,9 +55,6 @@ extern void i386_linux_report_signal_info (struct gdbarch *gdbarch, struct ui_out *uiout, enum gdb_signal siggnal); -/* Return the target description according to XCR0. */ -extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); - extern int i386_linux_gregset_reg_offset[]; /* Return x86 siginfo type. */ diff --git a/gdb/nat/x86-linux-tdesc.c b/gdb/nat/x86-linux-tdesc.c new file mode 100644 index 00000000000..ebe94d475ce --- /dev/null +++ b/gdb/nat/x86-linux-tdesc.c @@ -0,0 +1,128 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#include "nat/x86-linux-tdesc.h" +#ifdef __x86_64__ +#include "arch/amd64.h" +#include "arch/amd64-linux-tdesc.h" +#endif +#include "arch/i386.h" +#include "arch/i386-linux-tdesc.h" + +#include "nat/x86-linux.h" +#include "nat/gdb_ptrace.h" +#include "nat/x86-xstate.h" +#include "gdbsupport/x86-xstate.h" + +#ifndef __x86_64__ +#include <sys/procfs.h> +#endif + +#include <sys/uio.h> +#include <elf.h> + +#ifndef IN_PROCESS_AGENT + +/* See nat/x86-linux-tdesc.h. */ + +const target_desc * +x86_linux_tdesc_for_tid (int tid, const char *error_msg, + uint64_t *xcr0_storage, + x86_xsave_layout *xsave_layout_storage) +{ +#ifdef __x86_64__ + + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); + bool is_64bit = arch_size.is_64bit (); + bool is_x32 = arch_size.is_x32 (); + + if (sizeof (void *) == 4 && is_64bit && !is_x32) + error ("%s", error_msg); + +#elif HAVE_PTRACE_GETFPXREGS + if (have_ptrace_getfpxregs == TRIBOOL_UNKNOWN) + { + elf_fpxregset_t fpxregs; + + if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) + { + have_ptrace_getfpxregs = TRIBOOL_FALSE; + have_ptrace_getregset = TRIBOOL_FALSE; + } + else + have_ptrace_getfpxregs = TRIBOOL_TRUE; + } + + if (have_ptrace_getfpxregs == TRIBOOL_FALSE) + return i386_linux_read_description (X86_XSTATE_X87_MASK); +#endif + + if (have_ptrace_getregset == TRIBOOL_UNKNOWN) + { + uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; + struct iovec iov; + + iov.iov_base = xstateregs; + iov.iov_len = sizeof (xstateregs); + + /* Check if PTRACE_GETREGSET works. */ + if (ptrace (PTRACE_GETREGSET, tid, + (unsigned int) NT_X86_XSTATE, &iov) < 0) + { + have_ptrace_getregset = TRIBOOL_FALSE; + *xcr0_storage = 0; + } + else + { + have_ptrace_getregset = TRIBOOL_TRUE; + + /* Get XCR0 from XSAVE extended state. */ + *xcr0_storage = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET + / sizeof (uint64_t))]; + +#ifdef __x86_64__ + /* No MPX on x32. */ + if (is_64bit && is_x32) + *xcr0_storage &= ~X86_XSTATE_MPX; +#endif /* __x86_64__ */ + + *xsave_layout_storage + = x86_fetch_xsave_layout (*xcr0_storage, x86_xsave_length ()); + } + } + + /* Check the native XCR0 only if PTRACE_GETREGSET is available. If + PTRACE_GETREGSET is not available then set xcr0_features_bits to + zero so that the "no-features" descriptions are returned by the + code below. */ + uint64_t xcr0_features_bits; + if (have_ptrace_getregset == TRIBOOL_TRUE) + xcr0_features_bits = *xcr0_storage & X86_XSTATE_ALL_MASK; + else + xcr0_features_bits = 0; + +#ifdef __x86_64__ + if (is_64bit) + return amd64_linux_read_description (xcr0_features_bits, is_x32); + else +#endif + return i386_linux_read_description (xcr0_features_bits); +} + +#endif /* !IN_PROCESS_AGENT */ diff --git a/gdb/nat/x86-linux-tdesc.h b/gdb/nat/x86-linux-tdesc.h new file mode 100644 index 00000000000..4035c49fc06 --- /dev/null +++ b/gdb/nat/x86-linux-tdesc.h @@ -0,0 +1,54 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#ifndef NAT_X86_LINUX_TDESC_H +#define NAT_X86_LINUX_TDESC_H + +#include "gdbsupport/function-view.h" + +struct target_desc; +struct x86_xsave_layout; + +/* Return the target description for Linux thread TID. + + The storage pointed to by XCR0_STORAGE AND XSAVE_LAYOUT_STORAGE must + exist until the program (GDB or gdbserver) terminates, this storage is + used to cache the xcr0 and xsave layout values. The values pointed to + by these arguments are only updated at most once, the first time this + function is called. + + This function returns a target description based on the extracted xcr0 + value along with other characteristics of the thread identified by TID. + + This function can return nullptr if we encounter a machine configuration + for which a target_desc cannot be created. Ideally this would not be + the case, we should be able to create a target description for every + possible machine configuration. See amd64_linux_read_description and + i386_linux_read_description for cases when nullptr might be + returned. + + ERROR_MSG is using in an error() call if we try to create a target + description for a 64-bit process but this is a 32-bit build of GDB. */ + +extern const target_desc * +x86_linux_tdesc_for_tid (int tid, const char *error_msg, + uint64_t *xcr0_storage, + x86_xsave_layout *xsave_layout_storage); + +#endif /* NAT_X86_LINUX_TDESC_H */ diff --git a/gdb/x86-linux-nat.c b/gdb/x86-linux-nat.c index 285a4880103..43a72893562 100644 --- a/gdb/x86-linux-nat.c +++ b/gdb/x86-linux-nat.c @@ -38,6 +38,7 @@ #include "nat/x86-linux.h" #include "nat/x86-linux-dregs.h" #include "nat/linux-ptrace.h" +#include "nat/x86-linux-tdesc.h" /* linux_nat_target::low_new_fork implementation. */ @@ -92,93 +93,18 @@ x86_linux_nat_target::post_startup_inferior (ptid_t ptid) const struct target_desc * x86_linux_nat_target::read_description () { - int tid; - int is_64bit = 0; -#ifdef __x86_64__ - int is_x32; -#endif - static uint64_t xcr0; - uint64_t xcr0_features_bits; + static uint64_t xcr0_storage; if (inferior_ptid == null_ptid) return this->beneath ()->read_description (); - tid = inferior_ptid.pid (); - -#ifdef __x86_64__ - - x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); - is_64bit = arch_size.is_64bit (); - is_x32 = arch_size.is_x32 (); - - if (sizeof (void *) == 4 && is_64bit && !is_x32) - error (_("Can't debug 64-bit process with 32-bit GDB")); - -#elif HAVE_PTRACE_GETFPXREGS - if (have_ptrace_getfpxregs == TRIBOOL_UNKNOWN) - { - elf_fpxregset_t fpxregs; - - if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) - { - have_ptrace_getfpxregs = TRIBOOL_FALSE; - have_ptrace_getregset = TRIBOOL_FALSE; - return i386_linux_read_description (X86_XSTATE_X87_MASK); - } - } -#endif - - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) - { - uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; - struct iovec iov; - - iov.iov_base = xstateregs; - iov.iov_len = sizeof (xstateregs); - - /* Check if PTRACE_GETREGSET works. */ - if (ptrace (PTRACE_GETREGSET, tid, - (unsigned int) NT_X86_XSTATE, &iov) < 0) - have_ptrace_getregset = TRIBOOL_FALSE; - else - { - have_ptrace_getregset = TRIBOOL_TRUE; - - /* Get XCR0 from XSAVE extended state. */ - xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET - / sizeof (uint64_t))]; - - m_xsave_layout = x86_fetch_xsave_layout (xcr0, x86_xsave_length ()); - } - } - - /* Check the native XCR0 only if PTRACE_GETREGSET is available. If - PTRACE_GETREGSET is not available then set xcr0_features_bits to - zero so that the "no-features" descriptions are returned by the - switches below. */ - if (have_ptrace_getregset == TRIBOOL_TRUE) - xcr0_features_bits = xcr0 & X86_XSTATE_ALL_MASK; - else - xcr0_features_bits = 0; + int tid = inferior_ptid.pid (); - if (is_64bit) - { -#ifdef __x86_64__ - return amd64_linux_read_description (xcr0_features_bits, is_x32); -#endif - } - else - { - const struct target_desc * tdesc - = i386_linux_read_description (xcr0_features_bits); - - if (tdesc == NULL) - tdesc = i386_linux_read_description (X86_XSTATE_SSE_MASK); - - return tdesc; - } + const char *error_msg + = _("Can't debug 64-bit process with 32-bit GDB"); - gdb_assert_not_reached ("failed to return tdesc"); + return x86_linux_tdesc_for_tid (tid, error_msg, &xcr0_storage, + &this->m_xsave_layout); } \f diff --git a/gdbserver/configure.srv b/gdbserver/configure.srv index 9e861a75088..7a2702d78bf 100644 --- a/gdbserver/configure.srv +++ b/gdbserver/configure.srv @@ -109,6 +109,7 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/linux-btrace.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" + srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" srv_linux_usrregs=yes srv_linux_regsets=yes srv_linux_thread_db=yes @@ -371,6 +372,7 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/linux-btrace.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" + srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" srv_tgtobj="${srv_tgtobj} nat/amd64-linux-siginfo.o" srv_linux_usrregs=yes # This is for i386 progs. srv_linux_regsets=yes diff --git a/gdbserver/i387-fp.cc b/gdbserver/i387-fp.cc index 62cafd87204..4d8bcb5edfa 100644 --- a/gdbserver/i387-fp.cc +++ b/gdbserver/i387-fp.cc @@ -21,7 +21,7 @@ #include "nat/x86-xstate.h" /* Default to SSE. */ -static unsigned long long x86_xcr0 = X86_XSTATE_SSE_MASK; +static uint64_t x86_xcr0 = X86_XSTATE_SSE_MASK; static const int num_mpx_bnd_registers = 4; static const int num_mpx_cfg_registers = 2; @@ -944,9 +944,8 @@ i387_xsave_to_cache (struct regcache *regcache, const void *buf) /* See i387-fp.h. */ -void -i387_set_xsave_mask (uint64_t xcr0, int len) +std::pair<uint64_t *, x86_xsave_layout *> +i387_get_xsave_storage () { - x86_xcr0 = xcr0; - xsave_layout = x86_fetch_xsave_layout (xcr0, len); + return { &x86_xcr0, &xsave_layout }; } diff --git a/gdbserver/i387-fp.h b/gdbserver/i387-fp.h index 450466efb75..4ee21da8461 100644 --- a/gdbserver/i387-fp.h +++ b/gdbserver/i387-fp.h @@ -19,6 +19,8 @@ #ifndef GDBSERVER_I387_FP_H #define GDBSERVER_I387_FP_H +struct x86_xsave_layout; + void i387_cache_to_fsave (struct regcache *regcache, void *buf); void i387_fsave_to_cache (struct regcache *regcache, const void *buf); @@ -30,6 +32,6 @@ void i387_xsave_to_cache (struct regcache *regcache, const void *buf); /* Set the XSAVE mask and fetch the XSAVE layout via CPUID. */ -void i387_set_xsave_mask (uint64_t xcr0, int len); +std::pair<uint64_t *, x86_xsave_layout *> i387_get_xsave_storage (); #endif /* GDBSERVER_I387_FP_H */ diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index a6346750f49..df5e6aca081 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -22,6 +22,7 @@ #include "tracepoint.h" #include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" +#include "arch/amd64-linux-tdesc.h" /* fast tracepoints collect registers. */ diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc index 8f14e0937d4..aa346fc9bc3 100644 --- a/gdbserver/linux-i386-ipa.cc +++ b/gdbserver/linux-i386-ipa.cc @@ -22,6 +22,7 @@ #include "tracepoint.h" #include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" +#include "arch/i386-linux-tdesc.h" /* GDB register numbers. */ diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 7937055c060..5d60f6b6fe1 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -29,8 +29,11 @@ #ifdef __x86_64__ #include "nat/amd64-linux-siginfo.h" +#include "arch/amd64-linux-tdesc.h" #endif +#include "arch/i386-linux-tdesc.h" + #include "gdb_proc_service.h" /* Don't include elf/common.h if linux/elf.h got included by gdb_proc_service.h. */ @@ -46,6 +49,7 @@ #include "nat/x86-linux.h" #include "nat/x86-linux-dregs.h" #include "linux-x86-tdesc.h" +#include "nat/x86-linux-tdesc.h" #ifdef __x86_64__ static target_desc_up tdesc_amd64_linux_no_xml; @@ -834,29 +838,12 @@ static int use_xml; /* Get Linux/x86 target description from running target. */ static const struct target_desc * -x86_linux_read_description (void) +x86_linux_read_description () { - unsigned int machine; - int is_elf64; - int xcr0_features; - int tid; - static uint64_t xcr0; - static int xsave_len; - struct regset_info *regset; - - tid = lwpid_of (current_thread); + int tid = lwpid_of (current_thread); - is_elf64 = linux_pid_exe_is_elf_64_file (tid, &machine); - - if (sizeof (void *) == 4) - { - if (is_elf64 > 0) - error (_("Can't debug 64-bit process with 32-bit GDBserver")); -#ifndef __x86_64__ - else if (machine == EM_X86_64) - error (_("Can't debug x86-64 process with 32-bit GDBserver")); -#endif - } + const char *error_msg + = _("Can't debug 64-bit process with 32-bit GDBserver"); /* If we are not allowed to send an XML target description then we need to use the hard-wired target descriptions. This corresponds to GDB's @@ -866,103 +853,53 @@ x86_linux_read_description (void) generate some alternative target descriptions. */ if (!use_xml) { + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); + bool is_64bit = arch_size.is_64bit (); + bool is_x32 = arch_size.is_x32 (); + + if (sizeof (void *) == 4 && is_64bit && !is_x32) + error ("%s", error_msg); + #ifdef __x86_64__ - if (machine == EM_X86_64) + if (is_64bit && !is_x32) return tdesc_amd64_linux_no_xml.get (); else #endif return tdesc_i386_linux_no_xml.get (); } -#if !defined __x86_64__ && defined HAVE_PTRACE_GETFPXREGS - if (machine == EM_386 && have_ptrace_getfpxregs == TRIBOOL_UNKNOWN) - { - elf_fpxregset_t fpxregs; - - if (ptrace (PTRACE_GETFPXREGS, tid, 0, (long) &fpxregs) < 0) - { - have_ptrace_getfpxregs = TRIBOOL_FALSE; - have_ptrace_getregset = TRIBOOL_FALSE; - return i386_linux_read_description (X86_XSTATE_X87); - } - else - have_ptrace_getfpxregs = TRIBOOL_TRUE; - } -#endif - - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) - { - uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; - struct iovec iov; - - iov.iov_base = xstateregs; - iov.iov_len = sizeof (xstateregs); - - /* Check if PTRACE_GETREGSET works. */ - if (ptrace (PTRACE_GETREGSET, tid, - (unsigned int) NT_X86_XSTATE, (long) &iov) < 0) - have_ptrace_getregset = TRIBOOL_FALSE; - else - { - have_ptrace_getregset = TRIBOOL_TRUE; - - /* Get XCR0 from XSAVE extended state. */ - xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET - / sizeof (uint64_t))]; - - /* No MPX on x32. */ - if (machine == EM_X86_64 && !is_elf64) - xcr0 &= ~X86_XSTATE_MPX; - - xsave_len = x86_xsave_length (); - - /* Use PTRACE_GETREGSET if it is available. */ - for (regset = x86_regsets; - regset->fill_function != NULL; regset++) - if (regset->get_request == PTRACE_GETREGSET) - regset->size = xsave_len; - else if (regset->type != GENERAL_REGS) - regset->size = 0; - } - } - - /* Check the native XCR0 only if PTRACE_GETREGSET is available. */ - xcr0_features = (have_ptrace_getregset == TRIBOOL_TRUE - && (xcr0 & X86_XSTATE_ALL_MASK)); + /* If have_ptrace_getregset is changed to true by calling + x86_linux_tdesc_for_tid then we will perform some additional + initialisation. */ + bool have_ptrace_getregset_is_unknown + = have_ptrace_getregset == TRIBOOL_UNKNOWN; - if (xcr0_features) - i387_set_xsave_mask (xcr0, xsave_len); + /* Get pointers to where we should store the xcr0 and xsave_layout + values. These will be filled in by x86_linux_tdesc_for_tid the first + time that the function is called. Subsequent calls will not modify + the stored values. */ + std::pair<uint64_t *, x86_xsave_layout *> storage + = i387_get_xsave_storage (); - if (machine == EM_X86_64) - { -#ifdef __x86_64__ - const target_desc *tdesc = NULL; + const target_desc *tdesc + = x86_linux_tdesc_for_tid (tid, error_msg, storage.first, storage.second); - if (xcr0_features) - { - tdesc = amd64_linux_read_description (xcr0 & X86_XSTATE_ALL_MASK, - !is_elf64); - } - - if (tdesc == NULL) - tdesc = amd64_linux_read_description (X86_XSTATE_SSE_MASK, !is_elf64); - return tdesc; -#endif - } - else + if (have_ptrace_getregset_is_unknown + && have_ptrace_getregset == TRIBOOL_TRUE) { - const target_desc *tdesc = NULL; - - if (xcr0_features) - tdesc = i386_linux_read_description (xcr0 & X86_XSTATE_ALL_MASK); - - if (tdesc == NULL) - tdesc = i386_linux_read_description (X86_XSTATE_SSE); - - return tdesc; + int xsave_len = x86_xsave_length (); + + /* Use PTRACE_GETREGSET if it is available. */ + for (regset_info *regset = x86_regsets; + regset->fill_function != nullptr; + regset++) + if (regset->get_request == PTRACE_GETREGSET) + regset->size = xsave_len; + else if (regset->type != GENERAL_REGS) + regset->size = 0; } - gdb_assert_not_reached ("failed to return tdesc"); + return tdesc; } /* Update all the target description of all processes; a new GDB diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc index cd3b5d80e37..af3735aa895 100644 --- a/gdbserver/linux-x86-tdesc.cc +++ b/gdbserver/linux-x86-tdesc.cc @@ -23,8 +23,10 @@ #include "gdbsupport/x86-xstate.h" #ifdef __x86_64__ #include "arch/amd64.h" +#include "arch/amd64-linux-tdesc.h" #endif #include "x86-tdesc.h" +#include "arch/i386-linux-tdesc.h" /* Return the right x86_linux_tdesc index for a given XCR0. Return X86_TDESC_LAST if can't find a match. */ diff --git a/gdbserver/linux-x86-tdesc.h b/gdbserver/linux-x86-tdesc.h index f9561b129ae..576aaf5e165 100644 --- a/gdbserver/linux-x86-tdesc.h +++ b/gdbserver/linux-x86-tdesc.h @@ -46,11 +46,4 @@ int amd64_get_ipa_tdesc_idx (const struct target_desc *tdesc); const struct target_desc *i386_get_ipa_tdesc (int idx); -#ifdef __x86_64__ -const struct target_desc *amd64_linux_read_description (uint64_t xcr0, - bool is_x32); -#endif - -const struct target_desc *i386_linux_read_description (uint64_t xcr0); - #endif /* GDBSERVER_LINUX_X86_TDESC_H */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* Re: [PATCHv6 7/9] gdb/gdbserver: share some code relating to target description creation 2024-05-08 16:46 ` [PATCHv6 7/9] gdb/gdbserver: share some code relating to target description creation Andrew Burgess @ 2024-05-08 22:58 ` John Baldwin 0 siblings, 0 replies; 116+ messages in thread From: John Baldwin @ 2024-05-08 22:58 UTC (permalink / raw) To: Andrew Burgess, gdb-patches; +Cc: felix.willgerodt On 5/8/24 9:46 AM, Andrew Burgess wrote: > This commit is part of a series to share more of the x86 target > description creation code between GDB and gdbserver. > > Unlike previous commits which were mostly refactoring, this commit is > the first that makes a real change, though that change should mostly > be for gdbserver; I've largely adopted the "GDB" way of doing things > for gdbserver, and this fixes a real gdbserver bug. > > On a x86-64 Linux target, running the test: > > gdb.server/connect-with-no-symbol-file.exp > > results in two core files being created. Both of these core files are > from the inferior process, created after gdbserver has detached. > > In this test a gdbserver process is started and then, after gdbserver > has started, but before GDB attaches, we either delete the inferior > executable, or change its permissions so it can't be read. Only after > doing this do we attempt to connect with GDB. > > As GDB connects to gdbserver, gdbserver attempts to figure out the > target description so that it can send the description to GDB, this > involves a call to x86_linux_read_description. > > In x86_linux_read_description one of the first things we do is try to > figure out if the process is 32-bit or 64-bit. To do this we look up > the executable via the thread-id, and then attempt to read the > architecture size from the executable. This isn't going to work if > the executable has been deleted, or is no longer readable. > > And so, as we can't read the executable, we default to an i386 target > and use an i386 target description. > > A consequence of using an i386 target description is that addresses > are assumed to be 32-bits. Here's an example session that shows the > problems this causes. This is run on an x86-64 machine, and the test > binary (xx.x) is a standard 64-bit x86-64 binary: > > shell_1$ gdbserver --once localhost :54321 /tmp/xx.x > > shell_2$ gdb -q > (gdb) set sysroot > (gdb) shell chmod 000 /tmp/xx.x > (gdb) target remote :54321 > Remote debugging using :54321 > warning: /tmp/xx.x: Permission denied. > 0xf7fd3110 in ?? () > (gdb) show architecture > The target architecture is set to "auto" (currently "i386"). > (gdb) p/x $pc > $1 = 0xf7fd3110 > (gdb) info proc mappings > process 2412639 > Mapped address spaces: > > Start Addr End Addr Size Offset Perms objfile > 0x400000 0x401000 0x1000 0x0 r--p /tmp/xx.x > 0x401000 0x402000 0x1000 0x1000 r-xp /tmp/xx.x > 0x402000 0x403000 0x1000 0x2000 r--p /tmp/xx.x > 0x403000 0x405000 0x2000 0x2000 rw-p /tmp/xx.x > 0xf7fcb000 0xf7fcf000 0x4000 0x0 r--p [vvar] > 0xf7fcf000 0xf7fd1000 0x2000 0x0 r-xp [vdso] > 0xf7fd1000 0xf7fd3000 0x2000 0x0 r--p /usr/lib64/ld-2.30.so > 0xf7fd3000 0xf7ff3000 0x20000 0x2000 r-xp /usr/lib64/ld-2.30.so > 0xf7ff3000 0xf7ffb000 0x8000 0x22000 r--p /usr/lib64/ld-2.30.so > 0xf7ffc000 0xf7ffe000 0x2000 0x2a000 rw-p /usr/lib64/ld-2.30.so > 0xf7ffe000 0xf7fff000 0x1000 0x0 rw-p > 0xfffda000 0xfffff000 0x25000 0x0 rw-p [stack] > 0xff600000 0xff601000 0x1000 0x0 r-xp [vsyscall] > (gdb) info inferiors > Num Description Connection Executable > * 1 process 2412639 1 (remote :54321) > (gdb) shell cat /proc/2412639/maps > 00400000-00401000 r--p 00000000 fd:03 45907133 /tmp/xx.x > 00401000-00402000 r-xp 00001000 fd:03 45907133 /tmp/xx.x > 00402000-00403000 r--p 00002000 fd:03 45907133 /tmp/xx.x > 00403000-00405000 rw-p 00002000 fd:03 45907133 /tmp/xx.x > 7ffff7fcb000-7ffff7fcf000 r--p 00000000 00:00 0 [vvar] > 7ffff7fcf000-7ffff7fd1000 r-xp 00000000 00:00 0 [vdso] > 7ffff7fd1000-7ffff7fd3000 r--p 00000000 fd:00 143904 /usr/lib64/ld-2.30.so > 7ffff7fd3000-7ffff7ff3000 r-xp 00002000 fd:00 143904 /usr/lib64/ld-2.30.so > 7ffff7ff3000-7ffff7ffb000 r--p 00022000 fd:00 143904 /usr/lib64/ld-2.30.so > 7ffff7ffc000-7ffff7ffe000 rw-p 0002a000 fd:00 143904 /usr/lib64/ld-2.30.so > 7ffff7ffe000-7ffff7fff000 rw-p 00000000 00:00 0 > 7ffffffda000-7ffffffff000 rw-p 00000000 00:00 0 [stack] > ffffffffff600000-ffffffffff601000 r-xp 00000000 00:00 0 [vsyscall] > (gdb) > > Notice the difference between the mappings reported via GDB and those > reported directly from the kernel via /proc/PID/maps, the addresses of > every mapping is clamped to 32-bits for GDB, while the kernel reports > real 64-bit addresses. > > Notice also that the $pc value is a 32-bit value. It appears to be > within one of the mappings reported by GDB, but is outside any of the > mappings reported from the kernel. > > And this is where the problem arises. When gdbserver detaches from > the inferior we pass the inferior the address from which it should > resume. Due to the 32/64 bit confusion we tell the inferior to resume > from the 32-bit $pc value, which is not within any valid mapping, and > so, as soon as the inferior resumes, it segfaults. > > If we look at how GDB (not gdbserver) figures out its target > description then we see an interesting difference. GDB doesn't try to > read the executable. Instead GDB uses ptrace to query the thread's > state, and uses this to figure out the if the thread is 32 or 64 bit. > > If we update gdbserver to do it the "GDB" way then the above problem > is resolved, gdbserver now sees the process as 64-bit, and when we > detach from the inferior we give it the correct 64-bit address, and > the inferior no longer segfaults. > > Now, I could just update the gdbserver code, but better, I think, to > share one copy of the code between GDB and gdbserver in gdb/nat/. > That is what this commit does. > > The cores of x86_linux_read_description from gdbserver and > x86_linux_nat_target::read_description from GDB are moved into a new > file gdb/nat/x86-linux-tdesc.c and combined into a single function > x86_linux_tdesc_for_tid which is called from each location. > > This new function does things mostly the GDB way, some changes are > needed to allow for the sharing; we now take some pointers for where > the shared code can cache the xcr0 and xsave layout values. > > Another thing to note about this commit is how the functions > i386_linux_read_description and amd64_linux_read_description are > handled. For now I've left these function as implemented separately > in GDB and gdbserver. I've moved the declarations of these functions > into gdb/arch/{i386,amd64}-linux-tdesc.h, but the implementations are > left where they are. > > A later commit in this series will make these functions shared too, > but doing this is not trivial, so I've left that for a separate > commit. Merging the declarations as I've done here ensures that > everyone implements the function to the same API, and once these > functions are shared (in a later commit) we'll want a shared > declaration anyway. > > Reviewed-By: Felix Willgerodt <felix.willgerodt@intel.com> Acked-By: John Baldwin <jhb@FreeBSD.org> -- John Baldwin ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv6 8/9] gdbserver: update target description creation for x86/linux 2024-05-08 16:46 ` [PATCHv6 0/9] x86/Linux Target Description Changes Andrew Burgess ` (6 preceding siblings ...) 2024-05-08 16:46 ` [PATCHv6 7/9] gdb/gdbserver: share some code relating to target description creation Andrew Burgess @ 2024-05-08 16:46 ` Andrew Burgess 2024-05-08 16:46 ` [PATCHv6 9/9] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 0/9] x86/Linux Target Description Changes Andrew Burgess 9 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-08 16:46 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, jhb This commit is part of a series which aims to share more of the target description creation between GDB and gdbserver for x86/Linux. After some refactoring earlier in this series the shared x86_linux_tdesc_for_tid function was added into nat/x86-linux-tdesc.c. However, this function still relies on amd64_linux_read_description and i386_linux_read_description which are implemented separately for both gdbserver and GDB. Given that at their core, all these functions do is: 1. take an xcr0 value as input, 2. mask out some feature bits, 3. look for a cached pre-generated target description and return it if found, 4. if no cached target description is found then call either amd64_create_target_description or i386_create_target_description to create a new target description, which is then added to the cache. Return the newly created target description. The inner functions amd64_create_target_description and i386_create_target_description are already shared between GDB and gdbserver (in the gdb/arch/ directory), so the only thing that the *_read_description functions really do is add the caching layer, and it feels like this really could be shared. However, we have a small problem. Despite using the same {amd64,i386}_create_target_description functions in both GDB and gdbserver to create the target descriptions, on the gdbserver side we cache target descriptions based on a reduced set of xcr0 feature bits. What this means is that, in theory, different xcr0 values can map to the same cache entry, which could result in the wrong target description being used. However, I'm not sure if this can actually happen in reality. Within gdbserver we already split the target description cache based on i386, amd64, and x32. I suspect within a given gdbserver session we'll only see at most one target description for each of these. The cache conflicting problem is caused by xcr0_to_tdesc_idx, which maps an xcr0 value to a enum x86_linux_tdesc value, and there are only 7 usable values in enum x86_linux_tdesc. In contrast, on the GDB side there are 64, 32, and 16 cache slots for i386, amd64, and x32 respectively. On the GDB side it is much more important to cache things correctly as a single GDB session might connect to multiple different remote targets, each of which might have slightly different x86 architectures. And so, if we want to merge the target description caching between GDB and gdbserver, then we need to first update gdbserver so that it caches in the same way as GDB, that is, it needs to adopt a mechanism that allows for the same number of cache slots of each of i386, amd64, and x32. In this way, when the caching is shared, GDB's behaviour will not change. Unfortunately it's a little more complex than that due to the in process agent (IPA). When the IPA is in use, gdbserver sends a target description index to the IPA, and the IPA uses this to find the correct target description to use, the IPA having first generated every possible target description. Interestingly, there is certainly a bug here which results from only having 7 values in enum x86_linux_tdesc. As multiple possible target descriptions in gdbserver map to the same enum x86_linux_tdesc value, then, when the enum x86_linux_tdesc value is sent to the IPA there is no way for gdbserver to know that the IPA will select the correct target description. This bug will get fixed by this commit. ** START OF AN ASIDE ** Back in the day I suspect this approach of sending a target description index made perfect sense. However since this commit: commit a8806230241d201f808d856eaae4d44088117b0c Date: Thu Dec 7 17:07:01 2017 +0000 Initialize target description early in IPA I think that passing an index was probably a bad idea. We used to pass the index, and then use that index to lookup which target description to instantiate and use, the target description was not generated until the index arrived. However, the above commit fixed an issue where we can't call malloc() within (certain parts of) the IPA (apparently), so instead we now pre-compute _every_ possible target description within the IPA. The index is only used to lookup which of the (many) pre-computed target descriptions to use. It would (I think) have been easier all around if the IPA just self-inspected, figured out its own xcr0 value, and used that to create the one target description that is required. So long as the xcr0 to target description code is shared (at compile time) with gdbserver, then we can be sure that the IPA will derive the same target description as gdbserver, and we would avoid all this index passing business, which has made this commit so very, very painful. I did look at how a process might derive its own xcr0 value, but I don't believe this is actually that simple, so for now I've just doubled down on the index passing approach. While reviewing earlier iterations of this patch there has been discussion about the possibility of removing the IPA from GDB. That would certainly make all of the code touched in this patch much simpler, but I don't really want to do that as part of this series. ** END OF AN ASIDE ** Currently then for x86/linux, gdbserver sends a number between 0 and 7 to the IPA, and the IPA uses this to create a target description. However, I am proposing that gdbserver should now create one of (up to) 64 different target descriptions for i386, so this 0 to 7 index isn't going to be good enough any more (amd64 and x32 have slightly fewer possible target descriptions, but still more than 8, so the problem is the same). For a while I wondered if I was going to have to try and find some backward compatible solution for this mess. But after seeing how lightly the IPA is actually documented, I wonder if it is not the case that there is a tight coupling between a version of gdbserver and a version of the IPA? At least I'm hoping so, and that's what I've assumed in this commit. In this commit I have thrown out the old IPA target description index numbering scheme, and switched to a completely new numbering scheme. Instead of the index that is passed being arbitrary, the index is instead calculated from the set of xcr0 features that are present on the target. Within the IPA we can then reverse this logic to recreate the xcr0 value based on the index, and from the xcr0 value we can choose the correct target description. With the gdbserver to IPA numbering scheme issue resolved I have then update the gdbserver versions of amd64_linux_read_description and i386_linux_read_description so that they cache target descriptions using the same set of xcr0 features as GDB itself. After this gdbserver should now always come up with the same target description as GDB does on any x86/Linux target. This commit does not introduce any new code sharing between GDB and gdbserver as previous commits in this series have done. Instead this commit is all about bringing GDB and gdbserver into alignment functionally so that the next commit(s) can merge the GDB and gdbserver versions of these functions. Notes On The Implementation --------------------------- Previously, within gdbserver, target descriptions were cached within arrays. These arrays were sized based on enum x86_linux_tdesc and xcr0_to_tdesc_idx returned the array (cache) index. Now we need different array lengths for each of i386, amd64, and x32. And the index to use within each array is calculated based on which xcr0 bits are set and valid for a particular target type. I really wanted to avoid having fixed array sizes, or having the set of relevant xcr0 bits encoded in multiple places. The solution I came up with was to create a single data structure which would contain a list of xcr0 bits along with flags to indicate which of the i386, amd64, and x32 targets the bit is relevant for. By making the table constexpr, and adding some constexpr helper functions, it is possible to calculate the sizes for the cache arrays at compile time, as well as the bit masks needed to each target type. During review it was pointed out[1] that possibly the failure to check the SSE and X87 bits for amd64/x32 targets might be an error, however, if this is the case then this is an issue that existed long before this patch. I'd really like to keep this patch focused on reworking the existing code and try to avoid changing how target descriptions are actually created, mostly out of fear that I'll break something. [1] https://inbox.sourceware.org/gdb-patches/MN2PR11MB4566070607318EE7E669A5E28E1B2@MN2PR11MB4566.namprd11.prod.outlook.com Approved-By: John Baldwin <jhb@FreeBSD.org> --- gdbserver/linux-amd64-ipa.cc | 43 +---- gdbserver/linux-i386-ipa.cc | 23 +-- gdbserver/linux-x86-low.cc | 24 ++- gdbserver/linux-x86-tdesc.cc | 335 ++++++++++++++++++++++++++--------- gdbserver/linux-x86-tdesc.h | 48 ++--- 5 files changed, 306 insertions(+), 167 deletions(-) diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index df5e6aca081..0c80812cc6f 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -164,47 +164,21 @@ supply_static_tracepoint_registers (struct regcache *regcache, #endif /* HAVE_UST */ -#if !defined __ILP32__ -/* Map the tdesc index to xcr0 mask. */ -static uint64_t idx2mask[X86_TDESC_LAST] = { - X86_XSTATE_X87_MASK, - X86_XSTATE_SSE_MASK, - X86_XSTATE_AVX_MASK, - X86_XSTATE_MPX_MASK, - X86_XSTATE_AVX_MPX_MASK, - X86_XSTATE_AVX_AVX512_MASK, - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, -}; -#endif - /* Return target_desc to use for IPA, given the tdesc index passed by gdbserver. */ const struct target_desc * get_ipa_tdesc (int idx) { - if (idx >= X86_TDESC_LAST) - { - internal_error ("unknown ipa tdesc index: %d", idx); - } + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); #if defined __ILP32__ - switch (idx) - { - case X86_TDESC_SSE: - return amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); - case X86_TDESC_AVX: - return amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); - case X86_TDESC_AVX_AVX512: - return amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, true); - default: - break; - } + bool is_x32 = true; #else - return amd64_linux_read_description (idx2mask[idx], false); + bool is_x32 = false; #endif - internal_error ("unknown ipa tdesc index: %d", idx); + return amd64_linux_read_description (xcr0, is_x32); } /* Allocate buffer for the jump pads. The branch instruction has a @@ -272,11 +246,10 @@ void initialize_low_tracepoint (void) { #if defined __ILP32__ - amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); - amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); - amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, true); + for (auto i = 0; i < x86_linux_x32_tdesc_count (); i++) + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), true); #else - for (auto i = 0; i < X86_TDESC_LAST; i++) - amd64_linux_read_description (idx2mask[i], false); + for (auto i = 0; i < x86_linux_amd64_tdesc_count (); i++) + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), false); #endif } diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc index aa346fc9bc3..c1c3152fb04 100644 --- a/gdbserver/linux-i386-ipa.cc +++ b/gdbserver/linux-i386-ipa.cc @@ -245,28 +245,15 @@ initialize_fast_tracepoint_trampoline_buffer (void) } } -/* Map the tdesc index to xcr0 mask. */ -static uint64_t idx2mask[X86_TDESC_LAST] = { - X86_XSTATE_X87_MASK, - X86_XSTATE_SSE_MASK, - X86_XSTATE_AVX_MASK, - X86_XSTATE_MPX_MASK, - X86_XSTATE_AVX_MPX_MASK, - X86_XSTATE_AVX_AVX512_MASK, - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, -}; - /* Return target_desc to use for IPA, given the tdesc index passed by gdbserver. */ const struct target_desc * get_ipa_tdesc (int idx) { - if (idx >= X86_TDESC_LAST) - { - internal_error ("unknown ipa tdesc index: %d", idx); - } - return i386_linux_read_description (idx2mask[idx]); + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); + + return i386_linux_read_description (xcr0); } /* Allocate buffer for the jump pads. On i386, we can reach an arbitrary @@ -288,6 +275,6 @@ void initialize_low_tracepoint (void) { initialize_fast_tracepoint_trampoline_buffer (); - for (auto i = 0; i < X86_TDESC_LAST; i++) - i386_linux_read_description (idx2mask[i]); + for (auto i = 0; i < x86_linux_i386_tdesc_count (); i++) + i386_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i)); } diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 5d60f6b6fe1..86dbe165eca 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -2876,14 +2876,28 @@ x86_target::get_ipa_tdesc_idx () struct regcache *regcache = get_thread_regcache (current_thread, 0); const struct target_desc *tdesc = regcache->tdesc; + if (!use_xml) + { + /* If USE_XML is false then we should be using one of these target + descriptions, see x86_linux_read_description for where we choose + one of these. Both of these descriptions are created from this + fixed xcr0 value X86_XSTATE_SSE_MASK. */ + gdb_assert (tdesc == tdesc_i386_linux_no_xml.get () #ifdef __x86_64__ - return amd64_get_ipa_tdesc_idx (tdesc); -#endif + || tdesc == tdesc_amd64_linux_no_xml.get () +#endif /* __x86_64__ */ + ); + return x86_linux_xcr0_to_tdesc_idx (X86_XSTATE_SSE_MASK); + } - if (tdesc == tdesc_i386_linux_no_xml.get ()) - return X86_TDESC_SSE; + /* The xcr0 value and xsave layout value are cached when the target + description is read. Grab their cache location, and use the cached + value to calculate a tdesc index. */ + std::pair<uint64_t *, x86_xsave_layout *> storage + = i387_get_xsave_storage (); + uint64_t xcr0 = *storage.first; - return i386_get_ipa_tdesc_idx (tdesc); + return x86_linux_xcr0_to_tdesc_idx (xcr0); } /* The linux target ops object. */ diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc index af3735aa895..f5636898d71 100644 --- a/gdbserver/linux-x86-tdesc.cc +++ b/gdbserver/linux-x86-tdesc.cc @@ -28,96 +28,297 @@ #include "x86-tdesc.h" #include "arch/i386-linux-tdesc.h" -/* Return the right x86_linux_tdesc index for a given XCR0. Return - X86_TDESC_LAST if can't find a match. */ +/* A structure used to describe a single xstate feature bit that might, or + might not, be checked for when creating a target description for one of + i386, amd64, or x32. -static enum x86_linux_tdesc -xcr0_to_tdesc_idx (uint64_t xcr0, bool is_x32) + The different CPU/ABI types check for different xstate features when + creating a target description. + + We want to cache target descriptions, and this is currently done in + three separate caches, one each for i386, amd64, and x32. Additionally, + the caching we're discussing here is Linux only, and for Linux, the only + thing that has an impact on target description creation is the xcr0 + value. + + In order to ensure the cache functions correctly we need to filter out + only those xcr0 feature bits that are relevant, we can then cache target + descriptions based on the relevant feature bits. Two xcr0 values might + be different, but have the same relevant feature bits. In this case we + would expect the two xcr0 values to map to the same cache entry. */ + +struct x86_xstate_feature { + /* The xstate feature mask. This is a mask against an xcr0 value. */ + uint64_t feature; + + /* Is this feature checked when creating an i386 target description. */ + bool is_i386; + + /* Is this feature checked when creating an amd64 target description. */ + bool is_amd64; + + /* Is this feature checked when creating an x32 target description. */ + bool is_x32; +}; + +/* A constant table that describes all of the xstate features that are + checked when building a target description for i386, amd64, or x32. + + If in the future, due to simplifications or refactoring, this table ever + ends up with 'true' for every xcr0 feature on every target type, then this + is an indication that this table should probably be removed, and that the + rest of the code in this file can be simplified. */ + +static constexpr x86_xstate_feature x86_linux_all_xstate_features[] = { + /* Feature, i386, amd64, x32. */ + { X86_XSTATE_PKRU, true, true, true }, + { X86_XSTATE_AVX512, true, true, true }, + { X86_XSTATE_AVX, true, true, true }, + { X86_XSTATE_MPX, true, true, false }, + { X86_XSTATE_SSE, true, false, false }, + { X86_XSTATE_X87, true, false, false } +}; + +/* Return a compile time constant which is a mask of all the xstate features + that are checked for when building an i386 target description. */ + +static constexpr uint64_t +x86_linux_i386_xcr0_feature_mask () { - if (xcr0 & X86_XSTATE_PKRU) - { - if (is_x32) - { - /* No x32 MPX and PKU, fall back to avx_avx512. */ - return X86_TDESC_AVX_AVX512; - } - else - return X86_TDESC_AVX_MPX_AVX512_PKU; - } - else if (xcr0 & X86_XSTATE_AVX512) - return X86_TDESC_AVX_AVX512; - else if ((xcr0 & X86_XSTATE_AVX_MPX_MASK) == X86_XSTATE_AVX_MPX_MASK) + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_i386) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the xstate features + that are checked for when building an amd64 target description. */ + +static constexpr uint64_t +x86_linux_amd64_xcr0_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_amd64) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the xstate features + that are checked for when building an x32 target description. */ + +static constexpr uint64_t +x86_linux_x32_xcr0_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_x32) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a count of the number of xstate + features that are checked for when building an i386 target description. */ + +static constexpr int +x86_linux_i386_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_i386) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of xstate + features that are checked for when building an amd64 target description. */ + +static constexpr int +x86_linux_amd64_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_amd64) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of xstate + features that are checked for when building an x32 target description. */ + +static constexpr int +x86_linux_x32_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_x32) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +#ifdef IN_PROCESS_AGENT + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_amd64_tdesc_count () +{ + return x86_linux_amd64_tdesc_count_1 (); +} + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_x32_tdesc_count () +{ + return x86_linux_x32_tdesc_count_1 (); +} + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_i386_tdesc_count () +{ + return x86_linux_i386_tdesc_count_1 (); +} + +#endif /* IN_PROCESS_AGENT */ + +/* Convert an xcr0 value into an integer. The integer will be passed to + the in-process-agent where it will then be passed to + x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ + +int +x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0) +{ + /* The following table shows which features are checked for when creating + the target descriptions (see nat/x86-linux-tdesc.c), the feature order + represents the bit order within the generated index number. + + i386 | x87 sse mpx avx avx512 pkru + amd64 | mpx avx avx512 pkru + i32 | avx avx512 pkru + + The features are ordered so that for each mode (i386, amd64, i32) the + generated index will form a continuous range. */ + + int idx = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_xstate_features); ++i) { - if (is_x32) /* No MPX on x32. */ - return X86_TDESC_AVX; - else - return X86_TDESC_AVX_MPX; + if ((xcr0 & x86_linux_all_xstate_features[i].feature) + == x86_linux_all_xstate_features[i].feature) + idx |= (1 << i); } - else if (xcr0 & X86_XSTATE_MPX) + + return idx; +} + +#ifdef IN_PROCESS_AGENT + +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +uint64_t +x86_linux_tdesc_idx_to_xcr0 (int idx) +{ + uint64_t xcr0 = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_xstate_features); ++i) { - if (is_x32) /* No MPX on x32. */ - return X86_TDESC_AVX; - else - return X86_TDESC_MPX; + if ((idx & (1 << i)) != 0) + xcr0 |= x86_linux_all_xstate_features[i].feature; } - else if (xcr0 & X86_XSTATE_AVX) - return X86_TDESC_AVX; - else if (xcr0 & X86_XSTATE_SSE) - return X86_TDESC_SSE; - else if (xcr0 & X86_XSTATE_X87) - return X86_TDESC_MMX; - else - return X86_TDESC_LAST; + + return xcr0; } +#endif /* IN_PROCESS_AGENT */ + #if defined __i386__ || !defined IN_PROCESS_AGENT -static struct target_desc *i386_tdescs[X86_TDESC_LAST] = { }; +/* A cache of all possible i386 target descriptions. */ -/* Return the target description according to XCR0. */ +static struct target_desc *i386_tdescs[x86_linux_i386_tdesc_count_1 ()] = { }; + +/* See nat/x86-linux-tdesc.h. */ const struct target_desc * i386_linux_read_description (uint64_t xcr0) { - enum x86_linux_tdesc idx = xcr0_to_tdesc_idx (xcr0, false); + xcr0 &= x86_linux_i386_xcr0_feature_mask (); + int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - if (idx == X86_TDESC_LAST) - return NULL; + gdb_assert (idx >= 0 && idx < x86_linux_i386_tdesc_count_1 ()); - struct target_desc **tdesc = &i386_tdescs[idx]; + target_desc **tdesc = &i386_tdescs[idx]; - if (*tdesc == NULL) + if (*tdesc == nullptr) { *tdesc = i386_create_target_description (xcr0, true, false); init_target_desc (*tdesc, i386_expedite_regs); } - return *tdesc;; + return *tdesc; } #endif #ifdef __x86_64__ -static target_desc *amd64_tdescs[X86_TDESC_LAST] = { }; -static target_desc *x32_tdescs[X86_TDESC_LAST] = { }; +/* A cache of all possible amd64 target descriptions. */ + +static target_desc *amd64_tdescs[x86_linux_amd64_tdesc_count_1 ()] = { }; + +/* A cache of all possible x32 target descriptions. */ + +static target_desc *x32_tdescs[x86_linux_x32_tdesc_count_1 ()] = { }; + +/* See nat/x86-linux-tdesc.h. */ const struct target_desc * amd64_linux_read_description (uint64_t xcr0, bool is_x32) { - enum x86_linux_tdesc idx = xcr0_to_tdesc_idx (xcr0, is_x32); + if (is_x32) + xcr0 &= x86_linux_x32_xcr0_feature_mask (); + else + xcr0 &= x86_linux_amd64_xcr0_feature_mask (); + + int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - if (idx == X86_TDESC_LAST) - return NULL; + if (is_x32) + gdb_assert (idx >= 0 && idx < x86_linux_x32_tdesc_count_1 ()); + else + gdb_assert (idx >= 0 && idx < x86_linux_amd64_tdesc_count_1 ()); - struct target_desc **tdesc = NULL; + target_desc **tdesc = nullptr; if (is_x32) tdesc = &x32_tdescs[idx]; else tdesc = &amd64_tdescs[idx]; - if (*tdesc == NULL) + if (*tdesc == nullptr) { *tdesc = amd64_create_target_description (xcr0, is_x32, true, true); @@ -127,39 +328,3 @@ amd64_linux_read_description (uint64_t xcr0, bool is_x32) } #endif - -#ifndef IN_PROCESS_AGENT - -int -i386_get_ipa_tdesc_idx (const struct target_desc *tdesc) -{ - for (int i = 0; i < X86_TDESC_LAST; i++) - { - if (tdesc == i386_tdescs[i]) - return i; - } - - /* If none tdesc is found, return the one with minimum features. */ - return X86_TDESC_MMX; -} - -#if defined __x86_64__ -int -amd64_get_ipa_tdesc_idx (const struct target_desc *tdesc) -{ - for (int i = 0; i < X86_TDESC_LAST; i++) - { - if (tdesc == amd64_tdescs[i]) - return i; - } - for (int i = 0; i < X86_TDESC_LAST; i++) - { - if (tdesc == x32_tdescs[i]) - return i; - } - - return X86_TDESC_SSE; -} - -#endif -#endif diff --git a/gdbserver/linux-x86-tdesc.h b/gdbserver/linux-x86-tdesc.h index 576aaf5e165..c06af21da48 100644 --- a/gdbserver/linux-x86-tdesc.h +++ b/gdbserver/linux-x86-tdesc.h @@ -21,29 +21,29 @@ #ifndef GDBSERVER_LINUX_X86_TDESC_H #define GDBSERVER_LINUX_X86_TDESC_H -/* Note: since IPA obviously knows what ABI it's running on (i386 vs x86_64 - vs x32), it's sufficient to pass only the register set here. This, - together with the ABI known at IPA compile time, maps to a tdesc. */ - -enum x86_linux_tdesc { - X86_TDESC_MMX = 0, - X86_TDESC_SSE = 1, - X86_TDESC_AVX = 2, - X86_TDESC_MPX = 3, - X86_TDESC_AVX_MPX = 4, - X86_TDESC_AVX_AVX512 = 5, - X86_TDESC_AVX_MPX_AVX512_PKU = 6, - X86_TDESC_LAST = 7, -}; - -#if defined __i386__ || !defined IN_PROCESS_AGENT -int i386_get_ipa_tdesc_idx (const struct target_desc *tdesc); -#endif - -#if defined __x86_64__ && !defined IN_PROCESS_AGENT -int amd64_get_ipa_tdesc_idx (const struct target_desc *tdesc); -#endif - -const struct target_desc *i386_get_ipa_tdesc (int idx); +/* Convert an xcr0 value into an integer. The integer will be passed to + the in-process-agent where it will then be passed to + x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ + +extern int x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0); + +#ifdef IN_PROCESS_AGENT + +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +extern uint64_t x86_linux_tdesc_idx_to_xcr0 (int idx); + +/* Within the in-process-agent we need to pre-initialise all of the target + descriptions, to do this we need to know how many target descriptions + there are for each different target type. These functions return the + target description count for the relevant target. */ + +extern int x86_linux_amd64_tdesc_count (); +extern int x86_linux_x32_tdesc_count (); +extern int x86_linux_i386_tdesc_count (); + +#endif /* IN_PROCESS_AGENT */ #endif /* GDBSERVER_LINUX_X86_TDESC_H */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv6 9/9] gdb/gdbserver: share x86/linux tdesc caching 2024-05-08 16:46 ` [PATCHv6 0/9] x86/Linux Target Description Changes Andrew Burgess ` (7 preceding siblings ...) 2024-05-08 16:46 ` [PATCHv6 8/9] gdbserver: update target description creation for x86/linux Andrew Burgess @ 2024-05-08 16:46 ` Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 0/9] x86/Linux Target Description Changes Andrew Burgess 9 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-08 16:46 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, jhb This commit builds on the previous series of commits to share the target description caching code between GDB and gdbserver for x86/Linux targets. The objective of this commit is to move the four functions (2 each of) i386_linux_read_description and amd64_linux_read_description into the gdb/arch/ directory and combine them so we have just a single copy of each. Then GDB, gdbserver, and the in-process-agent (IPA) will link against these shared functions. One curiosity with this patch is the function x86_linux_post_init_tdesc. On the gdbserver side the two functions amd64_linux_read_description and i386_linux_read_description have some functionality that is not present on the GDB side, there is some additional configuration that is performed as each target description is created, to setup the expedited registers. To support this I've added the function x86_linux_post_init_tdesc. This function is called from the two *_linux_read_description functions, but is implemented separately for GDB and gdbserver. An alternative approach that avoids adding x86_linux_post_init_tdesc would be to have x86_linux_tdesc_for_tid return a non-const target description, then in x86_target::low_arch_setup we could inspect the target description to figure out if it is 64-bit or not, and modify the target description as needed. In the end I think that adding the x86_linux_post_init_tdesc function is the simpler solution. The contents of gdbserver/linux-x86-low.cc have moved to gdb/arch/x86-linux-tdesc-features.c, and gdbserver/linux-x86-tdesc.h has moved to gdb/arch/x86-linux-tdesc-features.h, this change leads to some updates in the #includes in the gdbserver/ directory. This commit also changes how target descriptions are cached. Previously both GDB and gdbserver used static C-style arrays to act as the tdesc cache. This was fine, except for two problems. Either the C-style arrays would need to be placed in x86-linux-tdesc-features.c, which would allow us to use the x86_linux_*_tdesc_count_1() functions to size the arrays for us, or we'd need to hard code the array sizes using separate #defines, which we'd then have to keep in sync with the rest of the code in x86-linux-tdesc-features.c. Given both of these problems I decided a better solution would be to just switch to using a std::unordered_map to act as the cache. This will resize automatically, and we can use the xcr0 value as the key. At first inspection, using xcr0 might seem to be a problem; after all the {i386,amd64}_create_target_description functions take more than just the xcr0 value. However, this patch is only for x86/Linux targets, and for x86/Linux all of the other flags passed to the tdesc creation functions have constant values and so are irrelevant when we consider tdesc caching. For testing I've done the following: - Built on x86-64 GNU/Linux for all targets, and just for the native target, - Build on i386 GNU/Linux for all targets, and just for the native target, - Build on a 64-bit, non-x86 GNU/Linux for all targets, just for the native target, and for targets x86_64-*-linux and i386-*-linux. --- gdb/Makefile.in | 5 + gdb/amd64-linux-tdep.c | 31 -- gdb/arch/amd64-linux-tdesc.c | 61 ++++ gdb/arch/i386-linux-tdesc.c | 51 +++ gdb/arch/x86-linux-tdesc-features.c | 267 +++++++++++++++ .../arch/x86-linux-tdesc-features.h | 49 ++- gdb/arch/x86-linux-tdesc.h | 37 +++ gdb/configure.nat | 6 +- gdb/configure.tgt | 11 +- gdb/i386-linux-tdep.c | 26 +- gdbserver/configure.srv | 9 + gdbserver/linux-amd64-ipa.cc | 2 +- gdbserver/linux-i386-ipa.cc | 2 +- gdbserver/linux-x86-low.cc | 2 +- gdbserver/linux-x86-tdesc.cc | 311 +----------------- 15 files changed, 489 insertions(+), 381 deletions(-) create mode 100644 gdb/arch/amd64-linux-tdesc.c create mode 100644 gdb/arch/i386-linux-tdesc.c create mode 100644 gdb/arch/x86-linux-tdesc-features.c rename gdbserver/linux-x86-tdesc.h => gdb/arch/x86-linux-tdesc-features.h (52%) create mode 100644 gdb/arch/x86-linux-tdesc.h diff --git a/gdb/Makefile.in b/gdb/Makefile.in index 9c2c69b02ab..800021551b9 100644 --- a/gdb/Makefile.in +++ b/gdb/Makefile.in @@ -748,6 +748,7 @@ ALL_64_TARGET_OBS = \ arch/aarch64-insn.o \ arch/aarch64-mte-linux.o \ arch/aarch64-scalable-linux.o \ + arch/amd64-linux-tdesc.o \ arch/amd64.o \ arch/riscv.o \ bpf-tdep.o \ @@ -788,9 +789,11 @@ ALL_TARGET_OBS = \ arch/arm.o \ arch/arm-get-next-pcs.o \ arch/arm-linux.o \ + arch/i386-linux-tdesc.o \ arch/i386.o \ arch/loongarch.o \ arch/ppc-linux-common.o \ + arch/x86-linux-tdesc-features.o \ arm-bsd-tdep.o \ arm-fbsd-tdep.o \ arm-linux-tdep.o \ @@ -1559,6 +1562,8 @@ HFILES_NO_SRCDIR = \ arch/ppc-linux-common.h \ arch/ppc-linux-tdesc.h \ arch/riscv.h \ + arch/x86-linux-tdesc-features.h \ + arch/x86-linux-tdesc.h \ cli/cli-cmds.h \ cli/cli-decode.h \ cli/cli-script.h \ diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c index f9647dce1e6..d7662cac572 100644 --- a/gdb/amd64-linux-tdep.c +++ b/gdb/amd64-linux-tdep.c @@ -1578,37 +1578,6 @@ amd64_linux_record_signal (struct gdbarch *gdbarch, return 0; } -const target_desc * -amd64_linux_read_description (uint64_t xcr0_features_bit, bool is_x32) -{ - static target_desc *amd64_linux_tdescs \ - [2/*AVX*/][2/*MPX*/][2/*AVX512*/][2/*PKRU*/] = {}; - static target_desc *x32_linux_tdescs \ - [2/*AVX*/][2/*AVX512*/][2/*PKRU*/] = {}; - - target_desc **tdesc; - - if (is_x32) - { - tdesc = &x32_linux_tdescs[(xcr0_features_bit & X86_XSTATE_AVX) ? 1 : 0 ] - [(xcr0_features_bit & X86_XSTATE_AVX512) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_PKRU) ? 1 : 0]; - } - else - { - tdesc = &amd64_linux_tdescs[(xcr0_features_bit & X86_XSTATE_AVX) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_MPX) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_AVX512) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_PKRU) ? 1 : 0]; - } - - if (*tdesc == NULL) - *tdesc = amd64_create_target_description (xcr0_features_bit, is_x32, - true, true); - - return *tdesc; -} - /* Get Linux/x86 target description from core dump. */ static const struct target_desc * diff --git a/gdb/arch/amd64-linux-tdesc.c b/gdb/arch/amd64-linux-tdesc.c new file mode 100644 index 00000000000..794cd5c815d --- /dev/null +++ b/gdb/arch/amd64-linux-tdesc.c @@ -0,0 +1,61 @@ +/* Target description related code for GNU/Linux x86-64. + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#include "arch/x86-linux-tdesc.h" +#include "arch/amd64-linux-tdesc.h" +#include "arch/amd64.h" +#include "arch/x86-linux-tdesc-features.h" + + +/* See arch/amd64-linux-tdesc.h. */ + +const struct target_desc * +amd64_linux_read_description (uint64_t xcr0, bool is_x32) +{ + /* The type used for the amd64 and x32 target description caches. */ + using tdesc_cache_type = std::unordered_map<uint64_t, const target_desc_up>; + + /* Caches for the previously seen amd64 and x32 target descriptions, + indexed by the xcr0 value that created the target description. These + need to be static within this function to ensure they are initialised + before first use. */ + static tdesc_cache_type amd64_tdesc_cache, x32_tdesc_cache; + + tdesc_cache_type &tdesc_cache = is_x32 ? x32_tdesc_cache : amd64_tdesc_cache; + + xcr0 &= is_x32 + ? x86_linux_x32_xcr0_feature_mask () + : x86_linux_amd64_xcr0_feature_mask (); + + const auto it = tdesc_cache.find (xcr0); + if (it != tdesc_cache.end ()) + return it->second.get (); + + /* Create the previously unseen target description. */ + target_desc_up tdesc (amd64_create_target_description (xcr0, is_x32, + true, true)); + x86_linux_post_init_tdesc (tdesc.get (), true); + + /* Add to the cache, and return a pointer borrowed from the + target_desc_up. This is safe as the cache (and the pointers contained + within it) are not deleted until GDB exits. */ + target_desc *ptr = tdesc.get (); + tdesc_cache.emplace (xcr0, std::move (tdesc)); + return ptr; +} diff --git a/gdb/arch/i386-linux-tdesc.c b/gdb/arch/i386-linux-tdesc.c new file mode 100644 index 00000000000..78b0cef841e --- /dev/null +++ b/gdb/arch/i386-linux-tdesc.c @@ -0,0 +1,51 @@ +/* Target description related code for GNU/Linux i386. + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#include "arch/x86-linux-tdesc.h" +#include "arch/i386-linux-tdesc.h" +#include "arch/i386.h" +#include "arch/x86-linux-tdesc-features.h" + +/* See arch/i386-linux-tdesc.h. */ + +const target_desc * +i386_linux_read_description (uint64_t xcr0) +{ + /* Cache of previously seen i386 target descriptions, indexed by the xcr0 + value that created the target description. This needs to be static + within this function to ensure it is initialised before first use. */ + static std::unordered_map<uint64_t, const target_desc_up> i386_tdesc_cache; + + xcr0 &= x86_linux_i386_xcr0_feature_mask (); + + const auto it = i386_tdesc_cache.find (xcr0); + if (it != i386_tdesc_cache.end ()) + return it->second.get (); + + /* Create the previously unseen target description. */ + target_desc_up tdesc (i386_create_target_description (xcr0, true, false)); + x86_linux_post_init_tdesc (tdesc.get (), false); + + /* Add to the cache, and return a pointer borrowed from the + target_desc_up. This is safe as the cache (and the pointers contained + within it) are not deleted until GDB exits. */ + target_desc *ptr = tdesc.get (); + i386_tdesc_cache.emplace (xcr0, std::move (tdesc)); + return ptr; +} diff --git a/gdb/arch/x86-linux-tdesc-features.c b/gdb/arch/x86-linux-tdesc-features.c new file mode 100644 index 00000000000..55444d91bd5 --- /dev/null +++ b/gdb/arch/x86-linux-tdesc-features.c @@ -0,0 +1,267 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#include "arch/x86-linux-tdesc-features.h" + +/* A structure used to describe a single xstate feature bit that might, or + might not, be checked for when creating a target description for one of + i386, amd64, or x32. + + The different CPU/ABI types check for different xstate features when + creating a target description. + + We want to cache target descriptions, and this is currently done in + three separate caches, one each for i386, amd64, and x32. Additionally, + the caching we're discussing here is Linux only, and for Linux, the only + thing that has an impact on target description creation is the xcr0 + value. + + In order to ensure the cache functions correctly we need to filter out + only those xcr0 feature bits that are relevant, we can then cache target + descriptions based on the relevant feature bits. Two xcr0 values might + be different, but have the same relevant feature bits. In this case we + would expect the two xcr0 values to map to the same cache entry. */ + +struct x86_xstate_feature { + /* The xstate feature mask. This is a mask against an xcr0 value. */ + uint64_t feature; + + /* Is this feature checked when creating an i386 target description. */ + bool is_i386; + + /* Is this feature checked when creating an amd64 target description. */ + bool is_amd64; + + /* Is this feature checked when creating an x32 target description. */ + bool is_x32; +}; + +/* A constant table that describes all of the xstate features that are + checked when building a target description for i386, amd64, or x32. + + If in the future, due to simplifications or refactoring, this table ever + ends up with 'true' for every xcr0 feature on every target type, then this + is an indication that this table should probably be removed, and that the + rest of the code in this file can be simplified. */ + +static constexpr x86_xstate_feature x86_linux_all_xstate_features[] = { + /* Feature, i386, amd64, x32. */ + { X86_XSTATE_PKRU, true, true, true }, + { X86_XSTATE_AVX512, true, true, true }, + { X86_XSTATE_AVX, true, true, true }, + { X86_XSTATE_MPX, true, true, false }, + { X86_XSTATE_SSE, true, false, false }, + { X86_XSTATE_X87, true, false, false } +}; + +/* Return a compile time constant which is a mask of all the xstate features + that are checked for when building an i386 target description. */ + +static constexpr uint64_t +x86_linux_i386_xcr0_feature_mask_1 () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_i386) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the xstate features + that are checked for when building an amd64 target description. */ + +static constexpr uint64_t +x86_linux_amd64_xcr0_feature_mask_1 () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_amd64) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the xstate features + that are checked for when building an x32 target description. */ + +static constexpr uint64_t +x86_linux_x32_xcr0_feature_mask_1 () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_x32) + mask |= entry.feature; + + return mask; +} + +/* See arch/x86-linux-tdesc-features.h. */ + +uint64_t +x86_linux_amd64_xcr0_feature_mask () +{ + return x86_linux_amd64_xcr0_feature_mask_1 (); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +uint64_t +x86_linux_x32_xcr0_feature_mask () +{ + return x86_linux_x32_xcr0_feature_mask_1 (); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +uint64_t +x86_linux_i386_xcr0_feature_mask () +{ + return x86_linux_i386_xcr0_feature_mask_1 (); +} + +#ifdef GDBSERVER + +/* See arch/x86-linux-tdesc-features.h. */ + +int +x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0) +{ + /* The following table shows which features are checked for when creating + the target descriptions (see nat/x86-linux-tdesc.c), the feature order + represents the bit order within the generated index number. + + i386 | x87 sse mpx avx avx512 pkru + amd64 | mpx avx avx512 pkru + i32 | avx avx512 pkru + + The features are ordered so that for each mode (i386, amd64, i32) the + generated index will form a continuous range. */ + + int idx = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_xstate_features); ++i) + { + if ((xcr0 & x86_linux_all_xstate_features[i].feature) + == x86_linux_all_xstate_features[i].feature) + idx |= (1 << i); + } + + return idx; +} + +#endif /* GDBSERVER */ + +#ifdef IN_PROCESS_AGENT + +/* Return a compile time constant which is a count of the number of xstate + features that are checked for when building an i386 target description. */ + +static constexpr int +x86_linux_i386_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_i386) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of xstate + features that are checked for when building an amd64 target description. */ + +static constexpr int +x86_linux_amd64_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_amd64) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of xstate + features that are checked for when building an x32 target description. */ + +static constexpr int +x86_linux_x32_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_x32) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +int +x86_linux_amd64_tdesc_count () +{ + return x86_linux_amd64_tdesc_count_1 (); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +int +x86_linux_x32_tdesc_count () +{ + return x86_linux_x32_tdesc_count_1 (); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +int +x86_linux_i386_tdesc_count () +{ + return x86_linux_i386_tdesc_count_1 (); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +uint64_t +x86_linux_tdesc_idx_to_xcr0 (int idx) +{ + uint64_t xcr0 = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_xstate_features); ++i) + { + if ((idx & (1 << i)) != 0) + xcr0 |= x86_linux_all_xstate_features[i].feature; + } + + return xcr0; +} + +#endif /* IN_PROCESS_AGENT */ diff --git a/gdbserver/linux-x86-tdesc.h b/gdb/arch/x86-linux-tdesc-features.h similarity index 52% rename from gdbserver/linux-x86-tdesc.h rename to gdb/arch/x86-linux-tdesc-features.h index c06af21da48..fb0017e46db 100644 --- a/gdbserver/linux-x86-tdesc.h +++ b/gdb/arch/x86-linux-tdesc-features.h @@ -1,7 +1,6 @@ -/* Low level support for x86 (i386 and x86-64), shared between gdbserver - and IPA. +/* Target description related code for GNU/Linux x86 (i386 and x86-64). - Copyright (C) 2016-2024 Free Software Foundation, Inc. + Copyright (C) 2024 Free Software Foundation, Inc. This file is part of GDB. @@ -18,32 +17,46 @@ You should have received a copy of the GNU General Public License along with this program. If not, see <http://www.gnu.org/licenses/>. */ -#ifndef GDBSERVER_LINUX_X86_TDESC_H -#define GDBSERVER_LINUX_X86_TDESC_H +#ifndef ARCH_X86_LINUX_TDESC_FEATURES_H +#define ARCH_X86_LINUX_TDESC_FEATURES_H -/* Convert an xcr0 value into an integer. The integer will be passed to - the in-process-agent where it will then be passed to - x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ +#include "gdbsupport/x86-xstate.h" +#include "gdbsupport/gdb_assert.h" + +/* Return a mask of X86_STATE_* feature flags. The returned mask indicates + the set of features which are checked for when creating the target + description for each of amd64, x32, and i386. */ + +extern uint64_t x86_linux_amd64_xcr0_feature_mask (); +extern uint64_t x86_linux_x32_xcr0_feature_mask (); +extern uint64_t x86_linux_i386_xcr0_feature_mask (); + +#ifdef GDBSERVER + +/* Convert an xcr0 value into an integer. The integer will be passed from + gdbserver to the in-process-agent where it will then be passed through + x86_linux_tdesc_idx_to_xcr0 to get back the original xcr0 value. */ extern int x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0); +#endif /* GDBSERVER */ + #ifdef IN_PROCESS_AGENT +/* Return the maximum possible number of target descriptions for each of + amd64, x32, and i386. These are used by the in-process-agent to + generate every possible target description. */ + +extern int x86_linux_amd64_tdesc_count (); +extern int x86_linux_x32_tdesc_count (); +extern int x86_linux_i386_tdesc_count (); + /* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) into an xcr0 value which can then be used to create a target description. */ extern uint64_t x86_linux_tdesc_idx_to_xcr0 (int idx); -/* Within the in-process-agent we need to pre-initialise all of the target - descriptions, to do this we need to know how many target descriptions - there are for each different target type. These functions return the - target description count for the relevant target. */ - -extern int x86_linux_amd64_tdesc_count (); -extern int x86_linux_x32_tdesc_count (); -extern int x86_linux_i386_tdesc_count (); - #endif /* IN_PROCESS_AGENT */ -#endif /* GDBSERVER_LINUX_X86_TDESC_H */ +#endif /* ARCH_X86_LINUX_TDESC_FEATURES_H */ diff --git a/gdb/arch/x86-linux-tdesc.h b/gdb/arch/x86-linux-tdesc.h new file mode 100644 index 00000000000..152592fcf76 --- /dev/null +++ b/gdb/arch/x86-linux-tdesc.h @@ -0,0 +1,37 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#ifndef ARCH_X86_LINUX_TDESC_H +#define ARCH_X86_LINUX_TDESC_H + +struct target_desc; + +/* This function is called from amd64_linux_read_description and + i386_linux_read_description after a new target description has been + created, TDESC is the new target description, IS_64BIT will be true + when called from amd64_linux_read_description, otherwise IS_64BIT will + be false. If the *_linux_read_description functions found a cached + target description then this function will not be called. + + Both GDB and gdbserver have their own implementations of this + function. */ + +extern void x86_linux_post_init_tdesc (target_desc *tdesc, bool is_64bit); + +#endif /* ARCH_X86_LINUX_TDESC_H */ diff --git a/gdb/configure.nat b/gdb/configure.nat index 4bcc0696027..24e1824b01c 100644 --- a/gdb/configure.nat +++ b/gdb/configure.nat @@ -256,7 +256,8 @@ case ${gdb_host} in NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ nat/x86-xstate.o \ i386-linux-nat.o x86-linux-nat.o nat/linux-btrace.o \ - nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o" + nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o \ + arch/i386-linux-tdesc.o arch/x86-linux-tdesc-features.o" ;; ia64) # Host: Intel IA-64 running GNU/Linux @@ -323,7 +324,8 @@ case ${gdb_host} in nat/x86-xstate.o amd64-nat.o amd64-linux-nat.o x86-linux-nat.o \ nat/linux-btrace.o \ nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o \ - nat/amd64-linux-siginfo.o" + nat/amd64-linux-siginfo.o arch/x86-linux-tdesc-features.o \ + arch/i386-linux-tdesc.o arch/amd64-linux-tdesc.o" ;; sparc) # Host: GNU/Linux UltraSPARC diff --git a/gdb/configure.tgt b/gdb/configure.tgt index 47a674201f9..8326c458eb1 100644 --- a/gdb/configure.tgt +++ b/gdb/configure.tgt @@ -320,10 +320,13 @@ i[34567]86-*-linux*) gdb_target_obs="i386-linux-tdep.o \ glibc-tdep.o \ solib-svr4.o symfile-mem.o \ - linux-tdep.o linux-record.o" + linux-tdep.o linux-record.o \ + arch/i386-linux-tdesc.o \ + arch/x86-linux-tdesc-features.o" if test "x$enable_64_bit_bfd" = "xyes"; then # Target: GNU/Linux x86-64 - gdb_target_obs="amd64-linux-tdep.o ${gdb_target_obs}" + gdb_target_obs="amd64-linux-tdep.o \ + arch/amd64-linux-tdesc.o ${gdb_target_obs}" fi ;; i[34567]86-*-gnu*) @@ -718,7 +721,9 @@ x86_64-*-linux*) # Target: GNU/Linux x86-64 gdb_target_obs="amd64-linux-tdep.o ${i386_tobjs} \ i386-linux-tdep.o glibc-tdep.o \ - solib-svr4.o symfile-mem.o linux-tdep.o linux-record.o" + solib-svr4.o symfile-mem.o linux-tdep.o linux-record.o \ + arch/i386-linux-tdesc.o arch/amd64-linux-tdesc.o \ + arch/x86-linux-tdesc-features.o" ;; x86_64-*-freebsd* | x86_64-*-kfreebsd*-gnu) # Target: FreeBSD/amd64 diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index 6a7490680c1..6debed21b89 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -41,6 +41,7 @@ #include "i387-tdep.h" #include "gdbsupport/x86-xstate.h" #include "arch/i386-linux-tdesc.h" +#include "arch/x86-linux-tdesc.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_I386 "syscalls/i386-linux.xml" @@ -680,29 +681,12 @@ i386_linux_core_read_x86_xsave_layout (struct gdbarch *gdbarch, layout) != 0; } -/* See i386-linux-tdep.h. */ +/* See arch/x86-linux-tdesc.h. */ -const struct target_desc * -i386_linux_read_description (uint64_t xcr0) +void +x86_linux_post_init_tdesc (target_desc *tdesc, bool is_64bit) { - if (xcr0 == 0) - return NULL; - - static struct target_desc *i386_linux_tdescs \ - [2/*X87*/][2/*SSE*/][2/*AVX*/][2/*MPX*/][2/*AVX512*/][2/*PKRU*/] = {}; - struct target_desc **tdesc; - - tdesc = &i386_linux_tdescs[(xcr0 & X86_XSTATE_X87) ? 1 : 0] - [(xcr0 & X86_XSTATE_SSE) ? 1 : 0] - [(xcr0 & X86_XSTATE_AVX) ? 1 : 0] - [(xcr0 & X86_XSTATE_MPX) ? 1 : 0] - [(xcr0 & X86_XSTATE_AVX512) ? 1 : 0] - [(xcr0 & X86_XSTATE_PKRU) ? 1 : 0]; - - if (*tdesc == NULL) - *tdesc = i386_create_target_description (xcr0, true, false); - - return *tdesc; + /* Nothing. */ } /* Get Linux/x86 target description from core dump. */ diff --git a/gdbserver/configure.srv b/gdbserver/configure.srv index 7a2702d78bf..e17b5cf280c 100644 --- a/gdbserver/configure.srv +++ b/gdbserver/configure.srv @@ -110,12 +110,16 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" + srv_tgtobj="${srv_tgtobj} arch/x86-linux-tdesc-features.o" + srv_tgtobj="${srv_tgtobj} arch/i386-linux-tdesc.o" srv_linux_usrregs=yes srv_linux_regsets=yes srv_linux_thread_db=yes srv_linux_btrace=yes ipa_obj="linux-i386-ipa.o linux-x86-tdesc-ipa.o" ipa_obj="${ipa_obj} arch/i386-ipa.o" + ipa_obj="${ipa_obj} arch/x86-linux-tdesc-features-ipa.o" + ipa_obj="${ipa_obj} arch/i386-linux-tdesc-ipa.o" ;; i[34567]86-*-mingw*) srv_regobj="" srv_tgtobj="x86-low.o nat/x86-dregs.o win32-low.o" @@ -374,12 +378,17 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" srv_tgtobj="${srv_tgtobj} nat/amd64-linux-siginfo.o" + srv_tgtobj="${srv_tgtobj} arch/x86-linux-tdesc-features.o" + srv_tgtobj="${srv_tgtobj} arch/amd64-linux-tdesc.o" + srv_tgtobj="${srv_tgtobj} arch/i386-linux-tdesc.o" srv_linux_usrregs=yes # This is for i386 progs. srv_linux_regsets=yes srv_linux_thread_db=yes srv_linux_btrace=yes ipa_obj="linux-amd64-ipa.o linux-x86-tdesc-ipa.o" ipa_obj="${ipa_obj} arch/amd64-ipa.o" + ipa_obj="${ipa_obj} arch/x86-linux-tdesc-features-ipa.o" + ipa_obj="${ipa_obj} arch/amd64-linux-tdesc-ipa.o" ;; x86_64-*-mingw*) srv_regobj="" srv_tgtobj="x86-low.o nat/x86-dregs.o" diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index 0c80812cc6f..89857f208b8 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -20,9 +20,9 @@ #include <sys/mman.h> #include "tracepoint.h" -#include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" #include "arch/amd64-linux-tdesc.h" +#include "arch/x86-linux-tdesc-features.h" /* fast tracepoints collect registers. */ diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc index c1c3152fb04..8100c9f9840 100644 --- a/gdbserver/linux-i386-ipa.cc +++ b/gdbserver/linux-i386-ipa.cc @@ -20,9 +20,9 @@ #include <sys/mman.h> #include "tracepoint.h" -#include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" #include "arch/i386-linux-tdesc.h" +#include "arch/x86-linux-tdesc-features.h" /* GDB register numbers. */ diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 86dbe165eca..5c28fec6df6 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -33,6 +33,7 @@ #endif #include "arch/i386-linux-tdesc.h" +#include "arch/x86-linux-tdesc-features.h" #include "gdb_proc_service.h" /* Don't include elf/common.h if linux/elf.h got included by @@ -48,7 +49,6 @@ #include "nat/linux-nat.h" #include "nat/x86-linux.h" #include "nat/x86-linux-dregs.h" -#include "linux-x86-tdesc.h" #include "nat/x86-linux-tdesc.h" #ifdef __x86_64__ diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc index f5636898d71..13c80762605 100644 --- a/gdbserver/linux-x86-tdesc.cc +++ b/gdbserver/linux-x86-tdesc.cc @@ -17,314 +17,19 @@ You should have received a copy of the GNU General Public License along with this program. If not, see <http://www.gnu.org/licenses/>. */ +#include "arch/x86-linux-tdesc.h" #include "tdesc.h" -#include "linux-x86-tdesc.h" -#include "arch/i386.h" -#include "gdbsupport/x86-xstate.h" -#ifdef __x86_64__ -#include "arch/amd64.h" -#include "arch/amd64-linux-tdesc.h" -#endif #include "x86-tdesc.h" -#include "arch/i386-linux-tdesc.h" - -/* A structure used to describe a single xstate feature bit that might, or - might not, be checked for when creating a target description for one of - i386, amd64, or x32. - - The different CPU/ABI types check for different xstate features when - creating a target description. - - We want to cache target descriptions, and this is currently done in - three separate caches, one each for i386, amd64, and x32. Additionally, - the caching we're discussing here is Linux only, and for Linux, the only - thing that has an impact on target description creation is the xcr0 - value. - - In order to ensure the cache functions correctly we need to filter out - only those xcr0 feature bits that are relevant, we can then cache target - descriptions based on the relevant feature bits. Two xcr0 values might - be different, but have the same relevant feature bits. In this case we - would expect the two xcr0 values to map to the same cache entry. */ - -struct x86_xstate_feature { - /* The xstate feature mask. This is a mask against an xcr0 value. */ - uint64_t feature; - - /* Is this feature checked when creating an i386 target description. */ - bool is_i386; - - /* Is this feature checked when creating an amd64 target description. */ - bool is_amd64; - - /* Is this feature checked when creating an x32 target description. */ - bool is_x32; -}; - -/* A constant table that describes all of the xstate features that are - checked when building a target description for i386, amd64, or x32. - - If in the future, due to simplifications or refactoring, this table ever - ends up with 'true' for every xcr0 feature on every target type, then this - is an indication that this table should probably be removed, and that the - rest of the code in this file can be simplified. */ - -static constexpr x86_xstate_feature x86_linux_all_xstate_features[] = { - /* Feature, i386, amd64, x32. */ - { X86_XSTATE_PKRU, true, true, true }, - { X86_XSTATE_AVX512, true, true, true }, - { X86_XSTATE_AVX, true, true, true }, - { X86_XSTATE_MPX, true, true, false }, - { X86_XSTATE_SSE, true, false, false }, - { X86_XSTATE_X87, true, false, false } -}; - -/* Return a compile time constant which is a mask of all the xstate features - that are checked for when building an i386 target description. */ - -static constexpr uint64_t -x86_linux_i386_xcr0_feature_mask () -{ - uint64_t mask = 0; - - for (const auto &entry : x86_linux_all_xstate_features) - if (entry.is_i386) - mask |= entry.feature; - - return mask; -} - -/* Return a compile time constant which is a mask of all the xstate features - that are checked for when building an amd64 target description. */ - -static constexpr uint64_t -x86_linux_amd64_xcr0_feature_mask () -{ - uint64_t mask = 0; - - for (const auto &entry : x86_linux_all_xstate_features) - if (entry.is_amd64) - mask |= entry.feature; - - return mask; -} - -/* Return a compile time constant which is a mask of all the xstate features - that are checked for when building an x32 target description. */ - -static constexpr uint64_t -x86_linux_x32_xcr0_feature_mask () -{ - uint64_t mask = 0; - - for (const auto &entry : x86_linux_all_xstate_features) - if (entry.is_x32) - mask |= entry.feature; - - return mask; -} - -/* Return a compile time constant which is a count of the number of xstate - features that are checked for when building an i386 target description. */ - -static constexpr int -x86_linux_i386_tdesc_count_1 () -{ - uint64_t count = 0; - - for (const auto &entry : x86_linux_all_xstate_features) - if (entry.is_i386) - ++count; - - gdb_assert (count > 0); - - return (1 << count); -} - -/* Return a compile time constant which is a count of the number of xstate - features that are checked for when building an amd64 target description. */ - -static constexpr int -x86_linux_amd64_tdesc_count_1 () -{ - uint64_t count = 0; - - for (const auto &entry : x86_linux_all_xstate_features) - if (entry.is_amd64) - ++count; - - gdb_assert (count > 0); - - return (1 << count); -} - -/* Return a compile time constant which is a count of the number of xstate - features that are checked for when building an x32 target description. */ - -static constexpr int -x86_linux_x32_tdesc_count_1 () -{ - uint64_t count = 0; - - for (const auto &entry : x86_linux_all_xstate_features) - if (entry.is_x32) - ++count; - - gdb_assert (count > 0); - - return (1 << count); -} - -#ifdef IN_PROCESS_AGENT - -/* See linux-x86-tdesc.h. */ - -int -x86_linux_amd64_tdesc_count () -{ - return x86_linux_amd64_tdesc_count_1 (); -} - -/* See linux-x86-tdesc.h. */ - -int -x86_linux_x32_tdesc_count () -{ - return x86_linux_x32_tdesc_count_1 (); -} - -/* See linux-x86-tdesc.h. */ - -int -x86_linux_i386_tdesc_count () -{ - return x86_linux_i386_tdesc_count_1 (); -} - -#endif /* IN_PROCESS_AGENT */ - -/* Convert an xcr0 value into an integer. The integer will be passed to - the in-process-agent where it will then be passed to - x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ - -int -x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0) -{ - /* The following table shows which features are checked for when creating - the target descriptions (see nat/x86-linux-tdesc.c), the feature order - represents the bit order within the generated index number. - - i386 | x87 sse mpx avx avx512 pkru - amd64 | mpx avx avx512 pkru - i32 | avx avx512 pkru - - The features are ordered so that for each mode (i386, amd64, i32) the - generated index will form a continuous range. */ - - int idx = 0; - - for (int i = 0; i < ARRAY_SIZE (x86_linux_all_xstate_features); ++i) - { - if ((xcr0 & x86_linux_all_xstate_features[i].feature) - == x86_linux_all_xstate_features[i].feature) - idx |= (1 << i); - } - - return idx; -} - -#ifdef IN_PROCESS_AGENT - -/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) - into an xcr0 value which can then be used to create a target - description. */ - -uint64_t -x86_linux_tdesc_idx_to_xcr0 (int idx) -{ - uint64_t xcr0 = 0; - - for (int i = 0; i < ARRAY_SIZE (x86_linux_all_xstate_features); ++i) - { - if ((idx & (1 << i)) != 0) - xcr0 |= x86_linux_all_xstate_features[i].feature; - } - - return xcr0; -} - -#endif /* IN_PROCESS_AGENT */ - -#if defined __i386__ || !defined IN_PROCESS_AGENT -/* A cache of all possible i386 target descriptions. */ +/* See arch/x86-linux-tdesc.h. */ -static struct target_desc *i386_tdescs[x86_linux_i386_tdesc_count_1 ()] = { }; - -/* See nat/x86-linux-tdesc.h. */ - -const struct target_desc * -i386_linux_read_description (uint64_t xcr0) +void +x86_linux_post_init_tdesc (target_desc *tdesc, bool is_64bit) { - xcr0 &= x86_linux_i386_xcr0_feature_mask (); - int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - - gdb_assert (idx >= 0 && idx < x86_linux_i386_tdesc_count_1 ()); - - target_desc **tdesc = &i386_tdescs[idx]; - - if (*tdesc == nullptr) - { - *tdesc = i386_create_target_description (xcr0, true, false); - - init_target_desc (*tdesc, i386_expedite_regs); - } - - return *tdesc; -} -#endif - #ifdef __x86_64__ - -/* A cache of all possible amd64 target descriptions. */ - -static target_desc *amd64_tdescs[x86_linux_amd64_tdesc_count_1 ()] = { }; - -/* A cache of all possible x32 target descriptions. */ - -static target_desc *x32_tdescs[x86_linux_x32_tdesc_count_1 ()] = { }; - -/* See nat/x86-linux-tdesc.h. */ - -const struct target_desc * -amd64_linux_read_description (uint64_t xcr0, bool is_x32) -{ - if (is_x32) - xcr0 &= x86_linux_x32_xcr0_feature_mask (); - else - xcr0 &= x86_linux_amd64_xcr0_feature_mask (); - - int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - - if (is_x32) - gdb_assert (idx >= 0 && idx < x86_linux_x32_tdesc_count_1 ()); - else - gdb_assert (idx >= 0 && idx < x86_linux_amd64_tdesc_count_1 ()); - - target_desc **tdesc = nullptr; - - if (is_x32) - tdesc = &x32_tdescs[idx]; + if (is_64bit) + init_target_desc (tdesc, amd64_expedite_regs); else - tdesc = &amd64_tdescs[idx]; - - if (*tdesc == nullptr) - { - *tdesc = amd64_create_target_description (xcr0, is_x32, true, true); - - init_target_desc (*tdesc, amd64_expedite_regs); - } - return *tdesc; -} - #endif + init_target_desc (tdesc, i386_expedite_regs); +} -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv7 0/9] x86/Linux Target Description Changes 2024-05-08 16:46 ` [PATCHv6 0/9] x86/Linux Target Description Changes Andrew Burgess ` (8 preceding siblings ...) 2024-05-08 16:46 ` [PATCHv6 9/9] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess @ 2024-05-11 10:08 ` Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 1/9] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess ` (8 more replies) 9 siblings, 9 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-11 10:08 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, jhb In v7: - Patch #3, based on John's feedback I've moved the have_ptrace_getfpxregs global into an i386 specific file within the gdb/nat/ directory. There's some tweaking of includes as a result, but nothing much really changes. I agree with John that this was probably the right thing to do. - Patch #4, the v6 patch broke the ARM build (I have access to AArch64, but not ARM), ARM is the other target to use the have_ptrace_getregset global, but I'd assumed this was x86 only! In v7 all I've done is move the declaration of the have_ptrace_getregset global into the gdb/nat/ directory, I've left the definitions alone. This allows me to reference the global from code I will (later in the series) add to the gdb/nat/ directory, but I don't have to worry about testing the ARM build as the definitions haven't moved. This could be cleaned up _more_ later, moving the definition of have_ptrace_getregset into the gdb/nat/ directory, but I don't plan to do this as part of this series. - No other patches have changed. - Rebased the series onto something closer to HEAD of current master. In v6: - I merged some of the smaller patches which had been approved, and which were only loosely connected to this series, i.e. they were refactoring, or trivial cleanup patches. - I've reordered some of the remaining patches, moving the smaller patches towards the start of the series, these are largely unchanged and have mostly already been reviewed and/or approved. If there's no negative feedback for these smaller patches in this new order then I'll likely merge some more of these smaller ones to try and get the size of this series down. - Patches #1, #2, #3 are pretty much unchaged, and are either already approved or have at least been reviewed. - Patch #4 is new, though this is pretty similar to patch #3, I don't expect any problems from this patch. - Patch #5 has an error restored that Felix pointed out I'd deleted by mistake. The error was, as Felix commented, restored (in a different way) in a later patch anyway, but dropping it in this patch was a mistake. - Patch #6 has already been approved. - Patch #7 has had some changes after feedback from Felix, the x86_linux_tdesc_for_tid function has dropped its callback argument, and we now take two pointers in which the function caches two different pieces of state. I think this is actually a huge improvement, thanks Felix! I've also removed passing have_ptrace_getfpxregs by pointer to this function. Instead, after patch #4, we can now access have_ptrace_getfpxregs as a global, just like we do for have_ptrace_getregset, this is another great improvement I think. I've also fixed a bug in x86_linux_tdesc_for_tid which was present in the original code, if we're NOT on x86-64, and we do have HAVE_PTRACE_GETFPXREGS defined then we will now always call i386_linux_read_description. Previously, we'd only call i386_linux_read_description when have_ptrace_getfpxregs was TRIBOOL_UNKNOWN, which was the first time x86_linux_tdesc_for_tid was called. - Patch #8, I've renamed some of the data structures, and reworded some of the comments, to try and make it clearer that the code added in this commit is all about checking xstate (xcr0) feature bits, and is not claiming to represent all possible state which might ever need to be checked when creating a tdesc. Felix correctly pointed out that the previous naming / commenting was rather misleading, and though I believe the code was (and is) doing the right thing, the comments gave the impression that things had been overlooked. - Patch #9, the core ideas of this patch are unchanged. I've reworked the commit message to try and explain myself better. In v5: - Felix pointed out that building gdbserver with the '-m32' flag on an x86-64 host would fail. This is fixed in V5 with the addition of patch #4. This patch moves the have_ptrace_getfpxregs global into the gdb/nat/ directory and fixes the includes so that the the declaration is seen where needed, - I've rebased onto a slightly later commit. In v4: - I tried merging V3, but it turned out I broke pretty much everything that wasn't x86 based when configured with --enable-targets=all, - The problem was a failure to correctly split the shared code between the gdb/arch/ and gdb/nat/ directories, as a consequence, code which is needed on a non x86 based host to support x86 based targets wasn't available to the compilation, and the build failed, - In V4 I've gone through every patch and resplit the code in a way which I now believe is correct, I've done the following tests: + On a non x86 host I've built GDB to support only the current host as a target, to support all targets, and to support x86-64 and i386 linux targets, + On an i386 virtual machine I built GDB only for the host as a target, and for all targets. I regression tested the all targets build for unix, native-gdbserver, and native-extended-gdbserver, + On an x86-64 machine I've built GDB for only the current host as a target, and for all targets. I regression tested the all targets build for unix, native-gdbserver, and native-extended-gdbserver. - Only patches 6, 8, and 10 require significant review. All of the other patches are pretty trivial (though reviews always welcome). - I think there's more improvements that can be made to the x86 target description creation/lookup/caching. This series only changes the Linux lookup, and we still cache i386/amd64/x32 separately. In the future I think we can merge all x86 target description caching into a single data structure, this would be for all OS variants and all ABI variants. Though making that "grand unification" will certainly require some of the code in this series to change, I think the bulk of it will remain, and trying to do everything in one series is just going to result in an even larger series. I'd prefer to get these first patches merged, then come back to build on this work once this is merged and we know there's no problems with it. In v3: - Rebased. Nasty merge conflict with 4bb20a6244b7091 which I think I've resolved, but am unable to test. Reposting so the author of that other commit can validate. - Initial testing looks good. Full tests are still running. In v2: - Rebase to current upstream/master, no merge conflicts, - Retested. --- Andrew Burgess (9): gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition gdbserver/x86: move no-xml code earlier in x86_linux_read_description gdb/x86: move have_ptrace_getfpxregs global into gdb/nat directory gdb: move have_ptrace_getregset declaration into gdb/nat directory gdb/x86: move reading of cs and ds state into gdb/nat directory gdb: move xcr0 == 0 check into i386_linux_core_read_description gdb/gdbserver: share some code relating to target description creation gdbserver: update target description creation for x86/linux gdb/gdbserver: share x86/linux tdesc caching gdb/Makefile.in | 10 +- gdb/amd64-linux-tdep.c | 32 +-- gdb/amd64-linux-tdep.h | 6 - gdb/arch/amd64-linux-tdesc.c | 61 +++++ gdb/arch/amd64-linux-tdesc.h | 30 +++ gdb/arch/i386-linux-tdesc.c | 51 ++++ gdb/arch/i386-linux-tdesc.h | 29 +++ gdb/arch/x86-linux-tdesc-features.c | 267 +++++++++++++++++++++ gdb/arch/x86-linux-tdesc-features.h | 62 +++++ gdb/arch/x86-linux-tdesc.h | 37 +++ gdb/configure.nat | 9 +- gdb/configure.tgt | 11 +- gdb/i386-linux-nat.c | 26 +- gdb/i386-linux-tdep.c | 42 ++-- gdb/i386-linux-tdep.h | 23 -- gdb/linux-nat.c | 2 +- gdb/linux-nat.h | 3 - gdb/{i386-linux-nat.h => nat/i386-linux.c} | 15 +- gdb/nat/i386-linux.h | 37 +++ gdb/nat/linux-nat.h | 3 + gdb/nat/x86-linux-tdesc.c | 129 ++++++++++ gdb/nat/x86-linux-tdesc.h | 54 +++++ gdb/nat/x86-linux.c | 47 ++++ gdb/nat/x86-linux.h | 28 +++ gdb/x86-linux-nat.c | 119 +-------- gdbserver/configure.srv | 12 + gdbserver/i387-fp.cc | 9 +- gdbserver/i387-fp.h | 4 +- gdbserver/linux-amd64-ipa.cc | 46 +--- gdbserver/linux-i386-ipa.cc | 26 +- gdbserver/linux-low.cc | 2 +- gdbserver/linux-low.h | 2 - gdbserver/linux-x86-low.cc | 210 ++++++---------- gdbserver/linux-x86-tdesc.cc | 142 +---------- gdbserver/linux-x86-tdesc.h | 56 ----- gdbsupport/x86-xstate.h | 20 ++ 36 files changed, 1028 insertions(+), 634 deletions(-) create mode 100644 gdb/arch/amd64-linux-tdesc.c create mode 100644 gdb/arch/amd64-linux-tdesc.h create mode 100644 gdb/arch/i386-linux-tdesc.c create mode 100644 gdb/arch/i386-linux-tdesc.h create mode 100644 gdb/arch/x86-linux-tdesc-features.c create mode 100644 gdb/arch/x86-linux-tdesc-features.h create mode 100644 gdb/arch/x86-linux-tdesc.h rename gdb/{i386-linux-nat.h => nat/i386-linux.c} (77%) create mode 100644 gdb/nat/i386-linux.h create mode 100644 gdb/nat/x86-linux-tdesc.c create mode 100644 gdb/nat/x86-linux-tdesc.h delete mode 100644 gdbserver/linux-x86-tdesc.h base-commit: a4f76c0765a0b9c643dc91d5a398a1cd9519572b -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv7 1/9] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition 2024-05-11 10:08 ` [PATCHv7 0/9] x86/Linux Target Description Changes Andrew Burgess @ 2024-05-11 10:08 ` Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 2/9] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess ` (7 subsequent siblings) 8 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-11 10:08 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, jhb Share the definition of I386_LINUX_XSAVE_XCR0_OFFSET between GDB and gdbserver. This commit moves the definition into gdbsupport/x86-xstate.h, which allows the #define to be shared. There should be no user visible changes after this commit. Approved-By: Felix Willgerodt <felix.willgerodt@intel.com> --- gdb/i386-linux-tdep.h | 20 -------------------- gdbserver/linux-x86-low.cc | 21 --------------------- gdbsupport/x86-xstate.h | 20 ++++++++++++++++++++ 3 files changed, 20 insertions(+), 41 deletions(-) diff --git a/gdb/i386-linux-tdep.h b/gdb/i386-linux-tdep.h index 5891747572b..07593c6a8ec 100644 --- a/gdb/i386-linux-tdep.h +++ b/gdb/i386-linux-tdep.h @@ -58,26 +58,6 @@ extern void i386_linux_report_signal_info (struct gdbarch *gdbarch, /* Return the target description according to XCR0. */ extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); -/* Format of XSAVE extended state is: - struct - { - fxsave_bytes[0..463] - sw_usable_bytes[464..511] - xstate_hdr_bytes[512..575] - extended state regions (AVX, MPX, AVX512, PKRU, etc.) - }; - - Same memory layout will be used for the coredump NT_X86_XSTATE - representing the XSAVE extended state registers. - - The first 8 bytes of the sw_usable_bytes[464..467] is the OS enabled - extended state mask, which is the same as the extended control register - 0 (the XFEATURE_ENABLED_MASK register), XCR0. We can use this mask - together with the mask saved in the xstate_hdr_bytes to determine what - states the processor/OS supports and what state, used or initialized, - the process/thread is in. */ -#define I386_LINUX_XSAVE_XCR0_OFFSET 464 - extern int i386_linux_gregset_reg_offset[]; /* Return x86 siginfo type. */ diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 2603fb2ac5d..6de44aeb2d9 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -831,27 +831,6 @@ x86_target::low_siginfo_fixup (siginfo_t *ptrace, gdb_byte *inf, int direction) \f static int use_xml; -/* Format of XSAVE extended state is: - struct - { - fxsave_bytes[0..463] - sw_usable_bytes[464..511] - xstate_hdr_bytes[512..575] - avx_bytes[576..831] - future_state etc - }; - - Same memory layout will be used for the coredump NT_X86_XSTATE - representing the XSAVE extended state registers. - - The first 8 bytes of the sw_usable_bytes[464..467] is the OS enabled - extended state mask, which is the same as the extended control register - 0 (the XFEATURE_ENABLED_MASK register), XCR0. We can use this mask - together with the mask saved in the xstate_hdr_bytes to determine what - states the processor/OS supports and what state, used or initialized, - the process/thread is in. */ -#define I386_LINUX_XSAVE_XCR0_OFFSET 464 - /* Does the current host support the GETFPXREGS request? The header file may or may not define it, and even if it is defined, the kernel will return EIO if it's running on a pre-SSE processor. */ diff --git a/gdbsupport/x86-xstate.h b/gdbsupport/x86-xstate.h index 89c1143fbe1..11b37544aa3 100644 --- a/gdbsupport/x86-xstate.h +++ b/gdbsupport/x86-xstate.h @@ -120,4 +120,24 @@ constexpr bool operator!= (const x86_xsave_layout &lhs, #define I387_MXCSR_INIT_VAL 0x1f80 +/* Format of XSAVE extended state is: + struct + { + fxsave_bytes[0..463] + sw_usable_bytes[464..511] + xstate_hdr_bytes[512..575] + extended state regions (AVX, MPX, AVX512, PKRU, etc.) + }; + + Same memory layout will be used for the coredump NT_X86_XSTATE + representing the XSAVE extended state registers. + + The first 8 bytes of the sw_usable_bytes[464..467] is the OS enabled + extended state mask, which is the same as the extended control register + 0 (the XFEATURE_ENABLED_MASK register), XCR0. We can use this mask + together with the mask saved in the xstate_hdr_bytes to determine what + states the processor/OS supports and what state, used or initialized, + the process/thread is in. */ +#define I386_LINUX_XSAVE_XCR0_OFFSET 464 + #endif /* COMMON_X86_XSTATE_H */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv7 2/9] gdbserver/x86: move no-xml code earlier in x86_linux_read_description 2024-05-11 10:08 ` [PATCHv7 0/9] x86/Linux Target Description Changes Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 1/9] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess @ 2024-05-11 10:08 ` Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 3/9] gdb/x86: move have_ptrace_getfpxregs global into gdb/nat directory Andrew Burgess ` (6 subsequent siblings) 8 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-11 10:08 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, jhb This commit is part of a series that aims to share more of the x86 target description reading/generation code between GDB and gdbserver. There are a huge number of similarities between the code in gdbserver's x86_linux_read_description function and GDB's x86_linux_nat_target::read_description function, and it is this similarity that I plan, in a later commit, to share between GDB and gdbserver. However, one thing that is different in x86_linux_read_description is the code inside the '!use_xml' block. This is the code that handles the case where gdbserver is not allowed to send an XML target description back to GDB. In this case gdbserver uses some predefined, fixed, target descriptions. First, it's worth noting that I suspect this code is not tested any more. I couldn't find anything in the testsuite that tries to disable XML target description support. And the idea of having a single "fixed" target description really doesn't work well when we think about all the various x86 extensions that exist. Part of me would like to rip out the no-xml support in gdbserver (at least for x86), and if a GDB connects that doesn't support XML target descriptions, gdbserver can just give an error and drop the connection. GDB has supported XML target descriptions for 16 years now, I think it would be reasonable for our shipped gdbserver to drop support for the old way of doing things. Anyway.... this commit doesn't do that. What I did notice was that, over time, the '!use_xml' block appears to have "drifted" within the x86_linux_read_description function; it's now not the first check we do. Instead we make some ptrace calls and return a target description generated based on the result of these ptrace calls. Surely it only makes sense to generate variable target descriptions if we can send these back to GDB? So in this commit I propose to move the '!use_xml' block earlier in the x86_linux_read_description function. The benefit of this is that this leaves the later half of x86_linux_read_description much more similar to the GDB function x86_linux_nat_target::read_description and sets us up for potentially sharing code between GDB and gdbserver in a later commit. Approved-By: John Baldwin <jhb@FreeBSD.org> Approved-By: Felix Willgerodt <felix.willgerodt@intel.com> --- gdbserver/linux-x86-low.cc | 27 ++++++++++++++++----------- 1 file changed, 16 insertions(+), 11 deletions(-) diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 6de44aeb2d9..e8ef3667eb4 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -869,6 +869,22 @@ x86_linux_read_description (void) #endif } + /* If we are not allowed to send an XML target description then we need + to use the hard-wired target descriptions. This corresponds to GDB's + default machine for x86. + + This check needs to occur before any returns statements that might + generate some alternative target descriptions. */ + if (!use_xml) + { +#ifdef __x86_64__ + if (machine == EM_X86_64) + return tdesc_amd64_linux_no_xml.get (); + else +#endif + return tdesc_i386_linux_no_xml.get (); + } + #if !defined __x86_64__ && defined HAVE_PTRACE_GETFPXREGS if (machine == EM_386 && have_ptrace_getfpxregs == -1) { @@ -885,17 +901,6 @@ x86_linux_read_description (void) } #endif - if (!use_xml) - { - /* Don't use XML. */ -#ifdef __x86_64__ - if (machine == EM_X86_64) - return tdesc_amd64_linux_no_xml.get (); - else -#endif - return tdesc_i386_linux_no_xml.get (); - } - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) { uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv7 3/9] gdb/x86: move have_ptrace_getfpxregs global into gdb/nat directory 2024-05-11 10:08 ` [PATCHv7 0/9] x86/Linux Target Description Changes Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 1/9] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 2/9] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess @ 2024-05-11 10:08 ` Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 4/9] gdb: move have_ptrace_getregset declaration " Andrew Burgess ` (5 subsequent siblings) 8 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-11 10:08 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, jhb The have_ptrace_getfpxregs global tracks whether GDB or gdbserver is running on a kernel that supports the GETFPXREGS ptrace request. Currently this global is declared twice (once in GDB and once in gdbserver), I think it makes sense to move this global into the nat/ directory, and have a single declaration and definition. While moving this variable I have converted it to a tribool, as that was what it really was, if even used the same numbering as the tribool enum (-1, 0, 1). Where have_ptrace_getfpxregs was used I have updated in the obvious way. However, while making this change I noticed what I think is a bug in x86_linux_nat_target::read_description and x86_linux_read_description, both of these functions can be called multiple times, but in both cases we only end up calling i386_linux_read_description the first time through in the event that PTRACE_GETFPXREGS is not supported. This is because initially have_ptrace_getfpxregs will be TRIBOOL_UNKNOWN, but after the ptrace call fails we set have_ptrace_getfpxregs to TRIBOOL_FALSE. The next time we attempt to read the target description we'll skip the ptrace call, and so skip the call to i386_linux_read_description. I've not tried to address this preexisting bug in this commit, this is purely a refactor, there should be no user visible changes after this commit. In later commits I'll merge the gdbserver and GDB code together into the nat/ directory, and after that I'll try to address this bug. Reviewed-By: Felix Willgerodt <felix.willgerodt@intel.com> --- gdb/Makefile.in | 2 +- gdb/configure.nat | 2 +- gdb/i386-linux-nat.c | 26 +++------------ gdb/{i386-linux-nat.h => nat/i386-linux.c} | 15 +++++---- gdb/nat/i386-linux.h | 37 ++++++++++++++++++++++ gdb/x86-linux-nat.c | 6 ++-- gdbserver/configure.srv | 1 + gdbserver/linux-x86-low.cc | 19 +++-------- 8 files changed, 62 insertions(+), 46 deletions(-) rename gdb/{i386-linux-nat.h => nat/i386-linux.c} (77%) create mode 100644 gdb/nat/i386-linux.h diff --git a/gdb/Makefile.in b/gdb/Makefile.in index 234d0051f49..bdb8a3683b5 100644 --- a/gdb/Makefile.in +++ b/gdb/Makefile.in @@ -1397,7 +1397,6 @@ HFILES_NO_SRCDIR = \ hppa-tdep.h \ i386-bsd-nat.h \ i386-darwin-tdep.h \ - i386-linux-nat.h \ i386-linux-tdep.h \ i386-tdep.h \ i387-tdep.h \ @@ -1594,6 +1593,7 @@ HFILES_NO_SRCDIR = \ nat/gdb_ptrace.h \ nat/gdb_thread_db.h \ nat/fork-inferior.h \ + nat/i386-linux.h \ nat/linux-btrace.h \ nat/linux-namespaces.h \ nat/linux-nat.h \ diff --git a/gdb/configure.nat b/gdb/configure.nat index 8b98511cef7..25268bb268b 100644 --- a/gdb/configure.nat +++ b/gdb/configure.nat @@ -256,7 +256,7 @@ case ${gdb_host} in NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ nat/x86-xstate.o \ i386-linux-nat.o x86-linux-nat.o nat/linux-btrace.o \ - nat/x86-linux.o nat/x86-linux-dregs.o" + nat/x86-linux.o nat/x86-linux-dregs.o nat/i386-linux.o" ;; ia64) # Host: Intel IA-64 running GNU/Linux diff --git a/gdb/i386-linux-nat.c b/gdb/i386-linux-nat.c index 7278dd91ff0..bfe3fb84d0a 100644 --- a/gdb/i386-linux-nat.c +++ b/gdb/i386-linux-nat.c @@ -26,7 +26,7 @@ #include "gregset.h" #include "gdb_proc_service.h" -#include "i386-linux-nat.h" +#include "nat/i386-linux.h" #include "i387-tdep.h" #include "i386-tdep.h" #include "i386-linux-tdep.h" @@ -80,22 +80,6 @@ int have_ptrace_getregs = 0 #endif ; - -/* Does the current host support the GETFPXREGS request? The header - file may or may not define it, and even if it is defined, the - kernel will return EIO if it's running on a pre-SSE processor. - - My instinct is to attach this to some architecture- or - target-specific data structure, but really, a particular GDB - process can only run on top of one kernel at a time. So it's okay - for this to be a simple variable. */ -int have_ptrace_getfpxregs = -#ifdef HAVE_PTRACE_GETFPXREGS - -1 -#else - 0 -#endif -; \f /* Accessing registers through the U area, one at a time. */ @@ -388,14 +372,14 @@ fetch_fpxregs (struct regcache *regcache, int tid) { elf_fpxregset_t fpxregs; - if (! have_ptrace_getfpxregs) + if (have_ptrace_getfpxregs == TRIBOOL_FALSE) return 0; if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) { if (errno == EIO) { - have_ptrace_getfpxregs = 0; + have_ptrace_getfpxregs = TRIBOOL_FALSE; return 0; } @@ -415,14 +399,14 @@ store_fpxregs (const struct regcache *regcache, int tid, int regno) { elf_fpxregset_t fpxregs; - if (! have_ptrace_getfpxregs) + if (have_ptrace_getfpxregs == TRIBOOL_FALSE) return 0; if (ptrace (PTRACE_GETFPXREGS, tid, 0, &fpxregs) == -1) { if (errno == EIO) { - have_ptrace_getfpxregs = 0; + have_ptrace_getfpxregs = TRIBOOL_FALSE; return 0; } diff --git a/gdb/i386-linux-nat.h b/gdb/nat/i386-linux.c similarity index 77% rename from gdb/i386-linux-nat.h rename to gdb/nat/i386-linux.c index c3d32aff1be..64605be4fc5 100644 --- a/gdb/i386-linux-nat.h +++ b/gdb/nat/i386-linux.c @@ -1,6 +1,6 @@ /* Native-dependent code for GNU/Linux i386. - Copyright (C) 1999-2024 Free Software Foundation, Inc. + Copyright (C) 2024 Free Software Foundation, Inc. This file is part of GDB. @@ -17,10 +17,13 @@ You should have received a copy of the GNU General Public License along with this program. If not, see <http://www.gnu.org/licenses/>. */ -#ifndef I386_LINUX_NAT_H -#define I386_LINUX_NAT_H 1 - -/* Does the current host support the GETFPXREGS request? */ -extern int have_ptrace_getfpxregs; +#include "nat/i386-linux.h" +/* See nat/i386-linux.h. */ +tribool have_ptrace_getfpxregs = +#ifdef HAVE_PTRACE_GETFPXREGS + TRIBOOL_UNKNOWN +#else + TRIBOOL_FALSE #endif +; diff --git a/gdb/nat/i386-linux.h b/gdb/nat/i386-linux.h new file mode 100644 index 00000000000..c93626b6439 --- /dev/null +++ b/gdb/nat/i386-linux.h @@ -0,0 +1,37 @@ +/* Native-dependent code for GNU/Linux i386. + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#ifndef NAT_I386_LINUX_H +#define NAT_I386_LINUX_H + +/* Does the current host support the GETFPXREGS request? The system header + file may or may not define it, but even if it is defined, the kernel + will return EIO if it's running on a pre-SSE processor. + + Initially this will be TRIBOOL_UNKNOWN and should be changed to + TRIBOOL_FALSE if the ptrace call is attempted and fails or changed to + TRIBOOL_TRUE if the ptrace call is attempted and succeeds. + + My instinct is to attach this to some architecture- or target-specific + data structure, but really, a particular GDB process can only run on top + of one kernel at a time. So it's okay - for this to be a global + variable. */ +extern tribool have_ptrace_getfpxregs; + +#endif /* NAT_I386_LINUX_H */ diff --git a/gdb/x86-linux-nat.c b/gdb/x86-linux-nat.c index a3d8ffb60f1..762151ff378 100644 --- a/gdb/x86-linux-nat.c +++ b/gdb/x86-linux-nat.c @@ -27,7 +27,7 @@ #include "x86-nat.h" #ifndef __x86_64__ -#include "i386-linux-nat.h" +#include "nat/i386-linux.h" #endif #include "x86-linux-nat.h" #include "i386-linux-tdep.h" @@ -147,13 +147,13 @@ x86_linux_nat_target::read_description () error (_("Can't debug 64-bit process with 32-bit GDB")); } #elif HAVE_PTRACE_GETFPXREGS - if (have_ptrace_getfpxregs == -1) + if (have_ptrace_getfpxregs == TRIBOOL_UNKNOWN) { elf_fpxregset_t fpxregs; if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) { - have_ptrace_getfpxregs = 0; + have_ptrace_getfpxregs = TRIBOOL_FALSE; have_ptrace_getregset = TRIBOOL_FALSE; return i386_linux_read_description (X86_XSTATE_X87_MASK); } diff --git a/gdbserver/configure.srv b/gdbserver/configure.srv index 9e861a75088..8e882d2159b 100644 --- a/gdbserver/configure.srv +++ b/gdbserver/configure.srv @@ -109,6 +109,7 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/linux-btrace.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" + srv_tgtobj="${srv_tgtobj} nat/i386-linux.o" srv_linux_usrregs=yes srv_linux_regsets=yes srv_linux_thread_db=yes diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index e8ef3667eb4..2d99a82f566 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -29,6 +29,8 @@ #ifdef __x86_64__ #include "nat/amd64-linux-siginfo.h" +#else +#include "nat/i386-linux.h" #endif #include "gdb_proc_service.h" @@ -831,17 +833,6 @@ x86_target::low_siginfo_fixup (siginfo_t *ptrace, gdb_byte *inf, int direction) \f static int use_xml; -/* Does the current host support the GETFPXREGS request? The header - file may or may not define it, and even if it is defined, the - kernel will return EIO if it's running on a pre-SSE processor. */ -int have_ptrace_getfpxregs = -#ifdef HAVE_PTRACE_GETFPXREGS - -1 -#else - 0 -#endif -; - /* Get Linux/x86 target description from running target. */ static const struct target_desc * @@ -886,18 +877,18 @@ x86_linux_read_description (void) } #if !defined __x86_64__ && defined HAVE_PTRACE_GETFPXREGS - if (machine == EM_386 && have_ptrace_getfpxregs == -1) + if (machine == EM_386 && have_ptrace_getfpxregs == TRIBOOL_UNKNOWN) { elf_fpxregset_t fpxregs; if (ptrace (PTRACE_GETFPXREGS, tid, 0, (long) &fpxregs) < 0) { - have_ptrace_getfpxregs = 0; + have_ptrace_getfpxregs = TRIBOOL_FALSE; have_ptrace_getregset = TRIBOOL_FALSE; return i386_linux_read_description (X86_XSTATE_X87); } else - have_ptrace_getfpxregs = 1; + have_ptrace_getfpxregs = TRIBOOL_TRUE; } #endif -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv7 4/9] gdb: move have_ptrace_getregset declaration into gdb/nat directory 2024-05-11 10:08 ` [PATCHv7 0/9] x86/Linux Target Description Changes Andrew Burgess ` (2 preceding siblings ...) 2024-05-11 10:08 ` [PATCHv7 3/9] gdb/x86: move have_ptrace_getfpxregs global into gdb/nat directory Andrew Burgess @ 2024-05-11 10:08 ` Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 5/9] gdb/x86: move reading of cs and ds state " Andrew Burgess ` (4 subsequent siblings) 8 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-11 10:08 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, jhb In a later commit I want to access have_ptrace_getregset from a .c file in the nat/ directory. To achieve this I need access to the declaration of have_ptrace_getregset. Currently have_ptrace_getregset is declared (and defined) twice, once in GDB and once in gdbserver. This commit moves the declaration into nat/linux-nat.h, but leaves the two definitions where they are. Now, in my later commit, I can pull in the declaration from nat/linux-nat.h. There should be no user visible changes after this commit. --- gdb/linux-nat.c | 2 +- gdb/linux-nat.h | 3 --- gdb/nat/linux-nat.h | 3 +++ gdbserver/linux-low.cc | 2 +- gdbserver/linux-low.h | 2 -- 5 files changed, 5 insertions(+), 7 deletions(-) diff --git a/gdb/linux-nat.c b/gdb/linux-nat.c index 48ecd3627ca..a342e06fd53 100644 --- a/gdb/linux-nat.c +++ b/gdb/linux-nat.c @@ -243,7 +243,7 @@ PTRACE_PEEKTEXT/PTRACE_POKETEXT or process_vm_readv/process_vm_writev: struct linux_nat_target *linux_target; -/* Does the current host support PTRACE_GETREGSET? */ +/* See nat/linux-nat.h. */ enum tribool have_ptrace_getregset = TRIBOOL_UNKNOWN; /* When true, print debug messages relating to the linux native target. */ diff --git a/gdb/linux-nat.h b/gdb/linux-nat.h index 4dcbe9e170a..fd393df5020 100644 --- a/gdb/linux-nat.h +++ b/gdb/linux-nat.h @@ -299,9 +299,6 @@ lwp_info_range all_lwps (); lwp_info_safe_range all_lwps_safe (); -/* Does the current host support PTRACE_GETREGSET? */ -extern enum tribool have_ptrace_getregset; - /* Called from the LWP layer to inform the thread_db layer that PARENT spawned CHILD. Both LWPs are currently stopped. This function does whatever is required to have the child LWP under the diff --git a/gdb/nat/linux-nat.h b/gdb/nat/linux-nat.h index d4a36cbb601..b9fba7924b9 100644 --- a/gdb/nat/linux-nat.h +++ b/gdb/nat/linux-nat.h @@ -36,6 +36,9 @@ struct arch_lwp_info; instead SIGTRAP with bit 7 set. */ #define SYSCALL_SIGTRAP (SIGTRAP | 0x80) +/* Does the current host support PTRACE_GETREGSET? */ +extern tribool have_ptrace_getregset; + /* Return the ptid of the current lightweight process. With NPTL threads and LWPs map 1:1, so this is equivalent to returning the ptid of the current thread. This function must be provided by diff --git a/gdbserver/linux-low.cc b/gdbserver/linux-low.cc index ac7f9807ecc..266c7de8fb8 100644 --- a/gdbserver/linux-low.cc +++ b/gdbserver/linux-low.cc @@ -133,7 +133,7 @@ typedef struct } Elf64_auxv_t; #endif -/* Does the current host support PTRACE_GETREGSET? */ +/* See nat/linux-nat.h. */ enum tribool have_ptrace_getregset = TRIBOOL_UNKNOWN; /* Return TRUE if THREAD is the leader thread of the process. */ diff --git a/gdbserver/linux-low.h b/gdbserver/linux-low.h index eaf87527338..273c04626b8 100644 --- a/gdbserver/linux-low.h +++ b/gdbserver/linux-low.h @@ -951,8 +951,6 @@ void thread_db_notice_clone (struct thread_info *parent_thr, ptid_t child_ptid); bool thread_db_thread_handle (ptid_t ptid, gdb_byte **handle, int *handle_len); -extern enum tribool have_ptrace_getregset; - /* Search for the value with type MATCH in the auxv vector, with entries of length WORDSIZE bytes, of process with pid PID. If found, store the value in *VALP and return 1. If not found or if there is an error, -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv7 5/9] gdb/x86: move reading of cs and ds state into gdb/nat directory 2024-05-11 10:08 ` [PATCHv7 0/9] x86/Linux Target Description Changes Andrew Burgess ` (3 preceding siblings ...) 2024-05-11 10:08 ` [PATCHv7 4/9] gdb: move have_ptrace_getregset declaration " Andrew Burgess @ 2024-05-11 10:08 ` Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 6/9] gdb: move xcr0 == 0 check into i386_linux_core_read_description Andrew Burgess ` (3 subsequent siblings) 8 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-11 10:08 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, jhb This patch is part of a series that has the aim sharing the x86 Linux target description creation code between GDB and gdbserver. Within GDB part of this process involves reading the cs and ds state from the 'struct user_regs_struct' using a ptrace call. This isn't done by gdbserver, which is part of the motivation for this whole series; the approach gdbserver takes is inferior to the approach GDB takes (gdbserver relies on reading the file being debugged, and extracting similar information from the file headers). This commit moves the reading of cs and ds, which is used to figure out if a thread is 32-bit or 64-bit (or in x32 mode), into the gdb/nat directory so that the code can be shared with gdbserver, but at this point I'm not actually using the code in gdbserver, that will come later. As such there should be no user visible changes after this commit, GDB continues to do things as it did before (reading cs/ds), while gdbserver continues to use its own approach (which doesn't require reading cs/ds). Approved-By: John Baldwin <jhb@FreeBSD.org> Reviewed-By: Felix Willgerodt <felix.willgerodt@intel.com> --- gdb/nat/x86-linux.c | 47 +++++++++++++++++++++++++++++++++++++++++++++ gdb/nat/x86-linux.h | 28 +++++++++++++++++++++++++++ gdb/x86-linux-nat.c | 45 ++++++++----------------------------------- 3 files changed, 83 insertions(+), 37 deletions(-) diff --git a/gdb/nat/x86-linux.c b/gdb/nat/x86-linux.c index 7a21c8f2c26..ad3ed3c2289 100644 --- a/gdb/nat/x86-linux.c +++ b/gdb/nat/x86-linux.c @@ -19,6 +19,8 @@ #include "x86-linux.h" #include "x86-linux-dregs.h" +#include "nat/gdb_ptrace.h" +#include <sys/user.h> /* Per-thread arch-specific data we want to keep. */ @@ -79,3 +81,48 @@ x86_linux_prepare_to_resume (struct lwp_info *lwp) { x86_linux_update_debug_registers (lwp); } + +#ifdef __x86_64__ +/* Value of CS segment register: + 64bit process: 0x33 + 32bit process: 0x23 */ +#define AMD64_LINUX_USER64_CS 0x33 + +/* Value of DS segment register: + LP64 process: 0x0 + X32 process: 0x2b */ +#define AMD64_LINUX_X32_DS 0x2b +#endif + +/* See nat/x86-linux.h. */ + +x86_linux_arch_size +x86_linux_ptrace_get_arch_size (int tid) +{ +#ifdef __x86_64__ + unsigned long cs; + unsigned long ds; + + /* Get CS register. */ + errno = 0; + cs = ptrace (PTRACE_PEEKUSER, tid, + offsetof (struct user_regs_struct, cs), 0); + if (errno != 0) + perror_with_name (_("Couldn't get CS register")); + + bool is_64bit = cs == AMD64_LINUX_USER64_CS; + + /* Get DS register. */ + errno = 0; + ds = ptrace (PTRACE_PEEKUSER, tid, + offsetof (struct user_regs_struct, ds), 0); + if (errno != 0) + perror_with_name (_("Couldn't get DS register")); + + bool is_x32 = ds == AMD64_LINUX_X32_DS; + + return x86_linux_arch_size (is_64bit, is_x32); +#else + return x86_linux_arch_size (false, false); +#endif +} diff --git a/gdb/nat/x86-linux.h b/gdb/nat/x86-linux.h index 822882173f9..15153ea277e 100644 --- a/gdb/nat/x86-linux.h +++ b/gdb/nat/x86-linux.h @@ -47,4 +47,32 @@ extern void x86_linux_delete_thread (struct arch_lwp_info *arch_lwp); extern void x86_linux_prepare_to_resume (struct lwp_info *lwp); +/* Return value from x86_linux_ptrace_get_arch_size function. Indicates if + a thread is 32-bit, 64-bit, or x32. */ + +struct x86_linux_arch_size +{ + explicit x86_linux_arch_size (bool is_64bit, bool is_x32) + : m_is_64bit (is_64bit), + m_is_x32 (is_x32) + { + /* Nothing. */ + } + + bool is_64bit () const + { return m_is_64bit; } + + bool is_x32 () const + { return m_is_x32; } + +private: + bool m_is_64bit = false; + bool m_is_x32 = false; +}; + +/* Use ptrace calls to figure out if thread TID is 32-bit, 64-bit, or + 64-bit running in x32 mode. */ + +extern x86_linux_arch_size x86_linux_ptrace_get_arch_size (int tid); + #endif /* NAT_X86_LINUX_H */ diff --git a/gdb/x86-linux-nat.c b/gdb/x86-linux-nat.c index 762151ff378..ed76c1760bc 100644 --- a/gdb/x86-linux-nat.c +++ b/gdb/x86-linux-nat.c @@ -90,18 +90,6 @@ x86_linux_nat_target::post_startup_inferior (ptid_t ptid) linux_nat_target::post_startup_inferior (ptid); } -#ifdef __x86_64__ -/* Value of CS segment register: - 64bit process: 0x33 - 32bit process: 0x23 */ -#define AMD64_LINUX_USER64_CS 0x33 - -/* Value of DS segment register: - LP64 process: 0x0 - X32 process: 0x2b */ -#define AMD64_LINUX_X32_DS 0x2b -#endif - /* Get Linux/x86 target description from running target. */ const struct target_desc * @@ -121,31 +109,14 @@ x86_linux_nat_target::read_description () tid = inferior_ptid.pid (); #ifdef __x86_64__ - { - unsigned long cs; - unsigned long ds; - - /* Get CS register. */ - errno = 0; - cs = ptrace (PTRACE_PEEKUSER, tid, - offsetof (struct user_regs_struct, cs), 0); - if (errno != 0) - perror_with_name (_("Couldn't get CS register")); - - is_64bit = cs == AMD64_LINUX_USER64_CS; - - /* Get DS register. */ - errno = 0; - ds = ptrace (PTRACE_PEEKUSER, tid, - offsetof (struct user_regs_struct, ds), 0); - if (errno != 0) - perror_with_name (_("Couldn't get DS register")); - - is_x32 = ds == AMD64_LINUX_X32_DS; - - if (sizeof (void *) == 4 && is_64bit && !is_x32) - error (_("Can't debug 64-bit process with 32-bit GDB")); - } + + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); + is_64bit = arch_size.is_64bit (); + is_x32 = arch_size.is_x32 (); + + if (sizeof (void *) == 4 && is_64bit && !is_x32) + error (_("Can't debug 64-bit process with 32-bit GDB")); + #elif HAVE_PTRACE_GETFPXREGS if (have_ptrace_getfpxregs == TRIBOOL_UNKNOWN) { -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv7 6/9] gdb: move xcr0 == 0 check into i386_linux_core_read_description 2024-05-11 10:08 ` [PATCHv7 0/9] x86/Linux Target Description Changes Andrew Burgess ` (4 preceding siblings ...) 2024-05-11 10:08 ` [PATCHv7 5/9] gdb/x86: move reading of cs and ds state " Andrew Burgess @ 2024-05-11 10:08 ` Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 7/9] gdb/gdbserver: share some code relating to target description creation Andrew Burgess ` (2 subsequent siblings) 8 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-11 10:08 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, jhb Currently, in i386_linux_core_read_description, if GDB fails to extract an xcr0 value from the core file, then we will have a default zero value for the xcr0 variable, we still call the i386_linux_read_description function, which checks for this zero value and returns nullptr. Back in i386_linux_core_read_description we spot the nullptr return value from i386_linux_read_description and call i386_linux_read_description again, but this time passing a default value for xcr0. In the next commit I plan to rework i386_linux_read_description, and in so doing I will remove the check for xcr0 == 0, this is inline with how the amd64 code is written. However, this means that the 'xcr0 == 0' check needs to move up the stack to i386_linux_core_read_description, again, this brings the i386 code into line with the amd64 code. This is just a refactor in preparation for the next commit, there should be no user visible changes after this commit. Approved-By: Felix Willgerodt <felix.willgerodt@intel.com> Approved-By: John Baldwin <jhb@FreeBSD.org> --- gdb/i386-linux-tdep.c | 15 ++++++++------- 1 file changed, 8 insertions(+), 7 deletions(-) diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index a78f03fac8d..511e43f3b6f 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -714,15 +714,16 @@ i386_linux_core_read_description (struct gdbarch *gdbarch, /* Linux/i386. */ x86_xsave_layout layout; uint64_t xcr0 = i386_linux_core_read_xsave_info (abfd, layout); - const struct target_desc *tdesc = i386_linux_read_description (xcr0); - if (tdesc != NULL) - return tdesc; + if (xcr0 == 0) + { + if (bfd_get_section_by_name (abfd, ".reg-xfp") != nullptr) + xcr0 = X86_XSTATE_SSE_MASK; + else + xcr0 = X86_XSTATE_X87_MASK; + } - if (bfd_get_section_by_name (abfd, ".reg-xfp") != NULL) - return i386_linux_read_description (X86_XSTATE_SSE_MASK); - else - return i386_linux_read_description (X86_XSTATE_X87_MASK); + return i386_linux_read_description (xcr0); } /* Similar to i386_supply_fpregset, but use XSAVE extended state. */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv7 7/9] gdb/gdbserver: share some code relating to target description creation 2024-05-11 10:08 ` [PATCHv7 0/9] x86/Linux Target Description Changes Andrew Burgess ` (5 preceding siblings ...) 2024-05-11 10:08 ` [PATCHv7 6/9] gdb: move xcr0 == 0 check into i386_linux_core_read_description Andrew Burgess @ 2024-05-11 10:08 ` Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 8/9] gdbserver: update target description creation for x86/linux Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 9/9] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess 8 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-11 10:08 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, jhb This commit is part of a series to share more of the x86 target description creation code between GDB and gdbserver. Unlike previous commits which were mostly refactoring, this commit is the first that makes a real change, though that change should mostly be for gdbserver; I've largely adopted the "GDB" way of doing things for gdbserver, and this fixes a real gdbserver bug. On a x86-64 Linux target, running the test: gdb.server/connect-with-no-symbol-file.exp results in two core files being created. Both of these core files are from the inferior process, created after gdbserver has detached. In this test a gdbserver process is started and then, after gdbserver has started, but before GDB attaches, we either delete the inferior executable, or change its permissions so it can't be read. Only after doing this do we attempt to connect with GDB. As GDB connects to gdbserver, gdbserver attempts to figure out the target description so that it can send the description to GDB, this involves a call to x86_linux_read_description. In x86_linux_read_description one of the first things we do is try to figure out if the process is 32-bit or 64-bit. To do this we look up the executable via the thread-id, and then attempt to read the architecture size from the executable. This isn't going to work if the executable has been deleted, or is no longer readable. And so, as we can't read the executable, we default to an i386 target and use an i386 target description. A consequence of using an i386 target description is that addresses are assumed to be 32-bits. Here's an example session that shows the problems this causes. This is run on an x86-64 machine, and the test binary (xx.x) is a standard 64-bit x86-64 binary: shell_1$ gdbserver --once localhost :54321 /tmp/xx.x shell_2$ gdb -q (gdb) set sysroot (gdb) shell chmod 000 /tmp/xx.x (gdb) target remote :54321 Remote debugging using :54321 warning: /tmp/xx.x: Permission denied. 0xf7fd3110 in ?? () (gdb) show architecture The target architecture is set to "auto" (currently "i386"). (gdb) p/x $pc $1 = 0xf7fd3110 (gdb) info proc mappings process 2412639 Mapped address spaces: Start Addr End Addr Size Offset Perms objfile 0x400000 0x401000 0x1000 0x0 r--p /tmp/xx.x 0x401000 0x402000 0x1000 0x1000 r-xp /tmp/xx.x 0x402000 0x403000 0x1000 0x2000 r--p /tmp/xx.x 0x403000 0x405000 0x2000 0x2000 rw-p /tmp/xx.x 0xf7fcb000 0xf7fcf000 0x4000 0x0 r--p [vvar] 0xf7fcf000 0xf7fd1000 0x2000 0x0 r-xp [vdso] 0xf7fd1000 0xf7fd3000 0x2000 0x0 r--p /usr/lib64/ld-2.30.so 0xf7fd3000 0xf7ff3000 0x20000 0x2000 r-xp /usr/lib64/ld-2.30.so 0xf7ff3000 0xf7ffb000 0x8000 0x22000 r--p /usr/lib64/ld-2.30.so 0xf7ffc000 0xf7ffe000 0x2000 0x2a000 rw-p /usr/lib64/ld-2.30.so 0xf7ffe000 0xf7fff000 0x1000 0x0 rw-p 0xfffda000 0xfffff000 0x25000 0x0 rw-p [stack] 0xff600000 0xff601000 0x1000 0x0 r-xp [vsyscall] (gdb) info inferiors Num Description Connection Executable * 1 process 2412639 1 (remote :54321) (gdb) shell cat /proc/2412639/maps 00400000-00401000 r--p 00000000 fd:03 45907133 /tmp/xx.x 00401000-00402000 r-xp 00001000 fd:03 45907133 /tmp/xx.x 00402000-00403000 r--p 00002000 fd:03 45907133 /tmp/xx.x 00403000-00405000 rw-p 00002000 fd:03 45907133 /tmp/xx.x 7ffff7fcb000-7ffff7fcf000 r--p 00000000 00:00 0 [vvar] 7ffff7fcf000-7ffff7fd1000 r-xp 00000000 00:00 0 [vdso] 7ffff7fd1000-7ffff7fd3000 r--p 00000000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7fd3000-7ffff7ff3000 r-xp 00002000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7ff3000-7ffff7ffb000 r--p 00022000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7ffc000-7ffff7ffe000 rw-p 0002a000 fd:00 143904 /usr/lib64/ld-2.30.so 7ffff7ffe000-7ffff7fff000 rw-p 00000000 00:00 0 7ffffffda000-7ffffffff000 rw-p 00000000 00:00 0 [stack] ffffffffff600000-ffffffffff601000 r-xp 00000000 00:00 0 [vsyscall] (gdb) Notice the difference between the mappings reported via GDB and those reported directly from the kernel via /proc/PID/maps, the addresses of every mapping is clamped to 32-bits for GDB, while the kernel reports real 64-bit addresses. Notice also that the $pc value is a 32-bit value. It appears to be within one of the mappings reported by GDB, but is outside any of the mappings reported from the kernel. And this is where the problem arises. When gdbserver detaches from the inferior we pass the inferior the address from which it should resume. Due to the 32/64 bit confusion we tell the inferior to resume from the 32-bit $pc value, which is not within any valid mapping, and so, as soon as the inferior resumes, it segfaults. If we look at how GDB (not gdbserver) figures out its target description then we see an interesting difference. GDB doesn't try to read the executable. Instead GDB uses ptrace to query the thread's state, and uses this to figure out the if the thread is 32 or 64 bit. If we update gdbserver to do it the "GDB" way then the above problem is resolved, gdbserver now sees the process as 64-bit, and when we detach from the inferior we give it the correct 64-bit address, and the inferior no longer segfaults. Now, I could just update the gdbserver code, but better, I think, to share one copy of the code between GDB and gdbserver in gdb/nat/. That is what this commit does. The cores of x86_linux_read_description from gdbserver and x86_linux_nat_target::read_description from GDB are moved into a new file gdb/nat/x86-linux-tdesc.c and combined into a single function x86_linux_tdesc_for_tid which is called from each location. This new function does things mostly the GDB way, some changes are needed to allow for the sharing; we now take some pointers for where the shared code can cache the xcr0 and xsave layout values. Another thing to note about this commit is how the functions i386_linux_read_description and amd64_linux_read_description are handled. For now I've left these function as implemented separately in GDB and gdbserver. I've moved the declarations of these functions into gdb/arch/{i386,amd64}-linux-tdesc.h, but the implementations are left where they are. A later commit in this series will make these functions shared too, but doing this is not trivial, so I've left that for a separate commit. Merging the declarations as I've done here ensures that everyone implements the function to the same API, and once these functions are shared (in a later commit) we'll want a shared declaration anyway. Reviewed-By: Felix Willgerodt <felix.willgerodt@intel.com> Acked-By: John Baldwin <jhb@FreeBSD.org> --- gdb/Makefile.in | 3 + gdb/amd64-linux-tdep.c | 1 + gdb/amd64-linux-tdep.h | 6 -- gdb/arch/amd64-linux-tdesc.h | 30 +++++++ gdb/arch/i386-linux-tdesc.h | 29 +++++++ gdb/configure.nat | 5 +- gdb/i386-linux-tdep.c | 1 + gdb/i386-linux-tdep.h | 3 - gdb/nat/x86-linux-tdesc.c | 129 ++++++++++++++++++++++++++++++ gdb/nat/x86-linux-tdesc.h | 54 +++++++++++++ gdb/x86-linux-nat.c | 88 ++------------------- gdbserver/configure.srv | 2 + gdbserver/i387-fp.cc | 9 +-- gdbserver/i387-fp.h | 4 +- gdbserver/linux-amd64-ipa.cc | 1 + gdbserver/linux-i386-ipa.cc | 1 + gdbserver/linux-x86-low.cc | 147 ++++++++++------------------------- gdbserver/linux-x86-tdesc.cc | 2 + gdbserver/linux-x86-tdesc.h | 7 -- 19 files changed, 312 insertions(+), 210 deletions(-) create mode 100644 gdb/arch/amd64-linux-tdesc.h create mode 100644 gdb/arch/i386-linux-tdesc.h create mode 100644 gdb/nat/x86-linux-tdesc.c create mode 100644 gdb/nat/x86-linux-tdesc.h diff --git a/gdb/Makefile.in b/gdb/Makefile.in index bdb8a3683b5..13ca417706b 100644 --- a/gdb/Makefile.in +++ b/gdb/Makefile.in @@ -1550,8 +1550,10 @@ HFILES_NO_SRCDIR = \ arch/aarch64-insn.h \ arch/aarch64-mte-linux.h \ arch/aarch64-scalable-linux.h \ + arch/amd64-linux-tdesc.h \ arch/arc.h \ arch/arm.h \ + arch/i386-linux-tdesc.h \ arch/i386.h \ arch/loongarch.h \ arch/ppc-linux-common.h \ @@ -1608,6 +1610,7 @@ HFILES_NO_SRCDIR = \ nat/x86-gcc-cpuid.h \ nat/x86-linux.h \ nat/x86-linux-dregs.h \ + nat/x86-linux-tdesc.h \ python/py-event.h \ python/py-events.h \ python/py-stopevent.h \ diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c index c52b0436872..f9647dce1e6 100644 --- a/gdb/amd64-linux-tdep.c +++ b/gdb/amd64-linux-tdep.c @@ -42,6 +42,7 @@ #include "arch/amd64.h" #include "target-descriptions.h" #include "expop.h" +#include "arch/amd64-linux-tdesc.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_AMD64 "syscalls/amd64-linux.xml" diff --git a/gdb/amd64-linux-tdep.h b/gdb/amd64-linux-tdep.h index 2003dcda78f..0ec49e7fe03 100644 --- a/gdb/amd64-linux-tdep.h +++ b/gdb/amd64-linux-tdep.h @@ -43,12 +43,6 @@ extern struct target_desc *tdesc_x32_linux; extern struct target_desc *tdesc_x32_avx_linux; extern struct target_desc *tdesc_x32_avx_avx512_linux; -/* Return the right amd64-linux target descriptions according to - XCR0_FEATURES_BIT and IS_X32. */ - -const target_desc *amd64_linux_read_description (uint64_t xcr0_features_bit, - bool is_x32); - /* Enum that defines the syscall identifiers for amd64 linux. Used for process record/replay, these will be translated into a gdb-canonical set of syscall ids in linux-record.c. */ diff --git a/gdb/arch/amd64-linux-tdesc.h b/gdb/arch/amd64-linux-tdesc.h new file mode 100644 index 00000000000..db425b60df6 --- /dev/null +++ b/gdb/arch/amd64-linux-tdesc.h @@ -0,0 +1,30 @@ +/* Target description related code for GNU/Linux x86-64. + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#ifndef ARCH_AMD64_LINUX_TDESC_H +#define ARCH_AMD64_LINUX_TDESC_H + +struct target_desc; + +/* Return the AMD64 target descriptions corresponding to XCR0 and IS_X32. */ + +extern const target_desc *amd64_linux_read_description (uint64_t xcr0, + bool is_x32); + +#endif /* ARCH_AMD64_LINUX_TDESC_H */ diff --git a/gdb/arch/i386-linux-tdesc.h b/gdb/arch/i386-linux-tdesc.h new file mode 100644 index 00000000000..0b736337a75 --- /dev/null +++ b/gdb/arch/i386-linux-tdesc.h @@ -0,0 +1,29 @@ +/* Target description related code for GNU/Linux i386. + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#ifndef ARCH_I386_LINUX_TDESC_H +#define ARCH_I386_LINUX_TDESC_H + +struct target_desc; + +/* Return the i386 target description corresponding to XCR0. */ + +extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); + +#endif /* ARCH_I386_LINUX_TDESC_H */ diff --git a/gdb/configure.nat b/gdb/configure.nat index 25268bb268b..a30b07ef8c5 100644 --- a/gdb/configure.nat +++ b/gdb/configure.nat @@ -256,7 +256,8 @@ case ${gdb_host} in NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ nat/x86-xstate.o \ i386-linux-nat.o x86-linux-nat.o nat/linux-btrace.o \ - nat/x86-linux.o nat/x86-linux-dregs.o nat/i386-linux.o" + nat/x86-linux.o nat/x86-linux-dregs.o nat/i386-linux.o \ + nat/x86-linux-tdesc.o" ;; ia64) # Host: Intel IA-64 running GNU/Linux @@ -322,7 +323,7 @@ case ${gdb_host} in NATDEPFILES="${NATDEPFILES} x86-nat.o nat/x86-dregs.o \ nat/x86-xstate.o amd64-nat.o amd64-linux-nat.o x86-linux-nat.o \ nat/linux-btrace.o \ - nat/x86-linux.o nat/x86-linux-dregs.o \ + nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o \ nat/amd64-linux-siginfo.o" ;; sparc) diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index 511e43f3b6f..6a7490680c1 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -40,6 +40,7 @@ #include "i387-tdep.h" #include "gdbsupport/x86-xstate.h" +#include "arch/i386-linux-tdesc.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_I386 "syscalls/i386-linux.xml" diff --git a/gdb/i386-linux-tdep.h b/gdb/i386-linux-tdep.h index 07593c6a8ec..e8691cd778e 100644 --- a/gdb/i386-linux-tdep.h +++ b/gdb/i386-linux-tdep.h @@ -55,9 +55,6 @@ extern void i386_linux_report_signal_info (struct gdbarch *gdbarch, struct ui_out *uiout, enum gdb_signal siggnal); -/* Return the target description according to XCR0. */ -extern const struct target_desc *i386_linux_read_description (uint64_t xcr0); - extern int i386_linux_gregset_reg_offset[]; /* Return x86 siginfo type. */ diff --git a/gdb/nat/x86-linux-tdesc.c b/gdb/nat/x86-linux-tdesc.c new file mode 100644 index 00000000000..b065928f8ba --- /dev/null +++ b/gdb/nat/x86-linux-tdesc.c @@ -0,0 +1,129 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#include "nat/x86-linux-tdesc.h" +#ifdef __x86_64__ +#include "arch/amd64.h" +#include "arch/amd64-linux-tdesc.h" +#endif +#include "arch/i386.h" +#include "arch/i386-linux-tdesc.h" + +#include "nat/x86-linux.h" +#include "nat/gdb_ptrace.h" +#include "nat/x86-xstate.h" +#include "gdbsupport/x86-xstate.h" + +#ifndef __x86_64__ +#include <sys/procfs.h> +#include "nat/i386-linux.h" +#endif + +#include <sys/uio.h> +#include <elf.h> + +#ifndef IN_PROCESS_AGENT + +/* See nat/x86-linux-tdesc.h. */ + +const target_desc * +x86_linux_tdesc_for_tid (int tid, const char *error_msg, + uint64_t *xcr0_storage, + x86_xsave_layout *xsave_layout_storage) +{ +#ifdef __x86_64__ + + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); + bool is_64bit = arch_size.is_64bit (); + bool is_x32 = arch_size.is_x32 (); + + if (sizeof (void *) == 4 && is_64bit && !is_x32) + error ("%s", error_msg); + +#elif HAVE_PTRACE_GETFPXREGS + if (have_ptrace_getfpxregs == TRIBOOL_UNKNOWN) + { + elf_fpxregset_t fpxregs; + + if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) + { + have_ptrace_getfpxregs = TRIBOOL_FALSE; + have_ptrace_getregset = TRIBOOL_FALSE; + } + else + have_ptrace_getfpxregs = TRIBOOL_TRUE; + } + + if (have_ptrace_getfpxregs == TRIBOOL_FALSE) + return i386_linux_read_description (X86_XSTATE_X87_MASK); +#endif + + if (have_ptrace_getregset == TRIBOOL_UNKNOWN) + { + uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; + struct iovec iov; + + iov.iov_base = xstateregs; + iov.iov_len = sizeof (xstateregs); + + /* Check if PTRACE_GETREGSET works. */ + if (ptrace (PTRACE_GETREGSET, tid, + (unsigned int) NT_X86_XSTATE, &iov) < 0) + { + have_ptrace_getregset = TRIBOOL_FALSE; + *xcr0_storage = 0; + } + else + { + have_ptrace_getregset = TRIBOOL_TRUE; + + /* Get XCR0 from XSAVE extended state. */ + *xcr0_storage = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET + / sizeof (uint64_t))]; + +#ifdef __x86_64__ + /* No MPX on x32. */ + if (is_64bit && is_x32) + *xcr0_storage &= ~X86_XSTATE_MPX; +#endif /* __x86_64__ */ + + *xsave_layout_storage + = x86_fetch_xsave_layout (*xcr0_storage, x86_xsave_length ()); + } + } + + /* Check the native XCR0 only if PTRACE_GETREGSET is available. If + PTRACE_GETREGSET is not available then set xcr0_features_bits to + zero so that the "no-features" descriptions are returned by the + code below. */ + uint64_t xcr0_features_bits; + if (have_ptrace_getregset == TRIBOOL_TRUE) + xcr0_features_bits = *xcr0_storage & X86_XSTATE_ALL_MASK; + else + xcr0_features_bits = 0; + +#ifdef __x86_64__ + if (is_64bit) + return amd64_linux_read_description (xcr0_features_bits, is_x32); + else +#endif + return i386_linux_read_description (xcr0_features_bits); +} + +#endif /* !IN_PROCESS_AGENT */ diff --git a/gdb/nat/x86-linux-tdesc.h b/gdb/nat/x86-linux-tdesc.h new file mode 100644 index 00000000000..4035c49fc06 --- /dev/null +++ b/gdb/nat/x86-linux-tdesc.h @@ -0,0 +1,54 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#ifndef NAT_X86_LINUX_TDESC_H +#define NAT_X86_LINUX_TDESC_H + +#include "gdbsupport/function-view.h" + +struct target_desc; +struct x86_xsave_layout; + +/* Return the target description for Linux thread TID. + + The storage pointed to by XCR0_STORAGE AND XSAVE_LAYOUT_STORAGE must + exist until the program (GDB or gdbserver) terminates, this storage is + used to cache the xcr0 and xsave layout values. The values pointed to + by these arguments are only updated at most once, the first time this + function is called. + + This function returns a target description based on the extracted xcr0 + value along with other characteristics of the thread identified by TID. + + This function can return nullptr if we encounter a machine configuration + for which a target_desc cannot be created. Ideally this would not be + the case, we should be able to create a target description for every + possible machine configuration. See amd64_linux_read_description and + i386_linux_read_description for cases when nullptr might be + returned. + + ERROR_MSG is using in an error() call if we try to create a target + description for a 64-bit process but this is a 32-bit build of GDB. */ + +extern const target_desc * +x86_linux_tdesc_for_tid (int tid, const char *error_msg, + uint64_t *xcr0_storage, + x86_xsave_layout *xsave_layout_storage); + +#endif /* NAT_X86_LINUX_TDESC_H */ diff --git a/gdb/x86-linux-nat.c b/gdb/x86-linux-nat.c index ed76c1760bc..7c97f84ef56 100644 --- a/gdb/x86-linux-nat.c +++ b/gdb/x86-linux-nat.c @@ -41,6 +41,7 @@ #include "nat/x86-linux.h" #include "nat/x86-linux-dregs.h" #include "nat/linux-ptrace.h" +#include "nat/x86-linux-tdesc.h" /* linux_nat_target::low_new_fork implementation. */ @@ -95,93 +96,18 @@ x86_linux_nat_target::post_startup_inferior (ptid_t ptid) const struct target_desc * x86_linux_nat_target::read_description () { - int tid; - int is_64bit = 0; -#ifdef __x86_64__ - int is_x32; -#endif - static uint64_t xcr0; - uint64_t xcr0_features_bits; + static uint64_t xcr0_storage; if (inferior_ptid == null_ptid) return this->beneath ()->read_description (); - tid = inferior_ptid.pid (); - -#ifdef __x86_64__ - - x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); - is_64bit = arch_size.is_64bit (); - is_x32 = arch_size.is_x32 (); - - if (sizeof (void *) == 4 && is_64bit && !is_x32) - error (_("Can't debug 64-bit process with 32-bit GDB")); - -#elif HAVE_PTRACE_GETFPXREGS - if (have_ptrace_getfpxregs == TRIBOOL_UNKNOWN) - { - elf_fpxregset_t fpxregs; - - if (ptrace (PTRACE_GETFPXREGS, tid, 0, (int) &fpxregs) < 0) - { - have_ptrace_getfpxregs = TRIBOOL_FALSE; - have_ptrace_getregset = TRIBOOL_FALSE; - return i386_linux_read_description (X86_XSTATE_X87_MASK); - } - } -#endif - - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) - { - uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; - struct iovec iov; - - iov.iov_base = xstateregs; - iov.iov_len = sizeof (xstateregs); - - /* Check if PTRACE_GETREGSET works. */ - if (ptrace (PTRACE_GETREGSET, tid, - (unsigned int) NT_X86_XSTATE, &iov) < 0) - have_ptrace_getregset = TRIBOOL_FALSE; - else - { - have_ptrace_getregset = TRIBOOL_TRUE; - - /* Get XCR0 from XSAVE extended state. */ - xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET - / sizeof (uint64_t))]; - - m_xsave_layout = x86_fetch_xsave_layout (xcr0, x86_xsave_length ()); - } - } - - /* Check the native XCR0 only if PTRACE_GETREGSET is available. If - PTRACE_GETREGSET is not available then set xcr0_features_bits to - zero so that the "no-features" descriptions are returned by the - switches below. */ - if (have_ptrace_getregset == TRIBOOL_TRUE) - xcr0_features_bits = xcr0 & X86_XSTATE_ALL_MASK; - else - xcr0_features_bits = 0; + int tid = inferior_ptid.pid (); - if (is_64bit) - { -#ifdef __x86_64__ - return amd64_linux_read_description (xcr0_features_bits, is_x32); -#endif - } - else - { - const struct target_desc * tdesc - = i386_linux_read_description (xcr0_features_bits); - - if (tdesc == NULL) - tdesc = i386_linux_read_description (X86_XSTATE_SSE_MASK); - - return tdesc; - } + const char *error_msg + = _("Can't debug 64-bit process with 32-bit GDB"); - gdb_assert_not_reached ("failed to return tdesc"); + return x86_linux_tdesc_for_tid (tid, error_msg, &xcr0_storage, + &this->m_xsave_layout); } \f diff --git a/gdbserver/configure.srv b/gdbserver/configure.srv index 8e882d2159b..538845b96d5 100644 --- a/gdbserver/configure.srv +++ b/gdbserver/configure.srv @@ -110,6 +110,7 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" srv_tgtobj="${srv_tgtobj} nat/i386-linux.o" + srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" srv_linux_usrregs=yes srv_linux_regsets=yes srv_linux_thread_db=yes @@ -372,6 +373,7 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/linux-btrace.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" + srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" srv_tgtobj="${srv_tgtobj} nat/amd64-linux-siginfo.o" srv_linux_usrregs=yes # This is for i386 progs. srv_linux_regsets=yes diff --git a/gdbserver/i387-fp.cc b/gdbserver/i387-fp.cc index 62cafd87204..4d8bcb5edfa 100644 --- a/gdbserver/i387-fp.cc +++ b/gdbserver/i387-fp.cc @@ -21,7 +21,7 @@ #include "nat/x86-xstate.h" /* Default to SSE. */ -static unsigned long long x86_xcr0 = X86_XSTATE_SSE_MASK; +static uint64_t x86_xcr0 = X86_XSTATE_SSE_MASK; static const int num_mpx_bnd_registers = 4; static const int num_mpx_cfg_registers = 2; @@ -944,9 +944,8 @@ i387_xsave_to_cache (struct regcache *regcache, const void *buf) /* See i387-fp.h. */ -void -i387_set_xsave_mask (uint64_t xcr0, int len) +std::pair<uint64_t *, x86_xsave_layout *> +i387_get_xsave_storage () { - x86_xcr0 = xcr0; - xsave_layout = x86_fetch_xsave_layout (xcr0, len); + return { &x86_xcr0, &xsave_layout }; } diff --git a/gdbserver/i387-fp.h b/gdbserver/i387-fp.h index 450466efb75..4ee21da8461 100644 --- a/gdbserver/i387-fp.h +++ b/gdbserver/i387-fp.h @@ -19,6 +19,8 @@ #ifndef GDBSERVER_I387_FP_H #define GDBSERVER_I387_FP_H +struct x86_xsave_layout; + void i387_cache_to_fsave (struct regcache *regcache, void *buf); void i387_fsave_to_cache (struct regcache *regcache, const void *buf); @@ -30,6 +32,6 @@ void i387_xsave_to_cache (struct regcache *regcache, const void *buf); /* Set the XSAVE mask and fetch the XSAVE layout via CPUID. */ -void i387_set_xsave_mask (uint64_t xcr0, int len); +std::pair<uint64_t *, x86_xsave_layout *> i387_get_xsave_storage (); #endif /* GDBSERVER_I387_FP_H */ diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index a6346750f49..df5e6aca081 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -22,6 +22,7 @@ #include "tracepoint.h" #include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" +#include "arch/amd64-linux-tdesc.h" /* fast tracepoints collect registers. */ diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc index 8f14e0937d4..aa346fc9bc3 100644 --- a/gdbserver/linux-i386-ipa.cc +++ b/gdbserver/linux-i386-ipa.cc @@ -22,6 +22,7 @@ #include "tracepoint.h" #include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" +#include "arch/i386-linux-tdesc.h" /* GDB register numbers. */ diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 2d99a82f566..1c3d55f17fc 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -29,10 +29,13 @@ #ifdef __x86_64__ #include "nat/amd64-linux-siginfo.h" +#include "arch/amd64-linux-tdesc.h" #else #include "nat/i386-linux.h" #endif +#include "arch/i386-linux-tdesc.h" + #include "gdb_proc_service.h" /* Don't include elf/common.h if linux/elf.h got included by gdb_proc_service.h. */ @@ -48,6 +51,7 @@ #include "nat/x86-linux.h" #include "nat/x86-linux-dregs.h" #include "linux-x86-tdesc.h" +#include "nat/x86-linux-tdesc.h" #ifdef __x86_64__ static target_desc_up tdesc_amd64_linux_no_xml; @@ -836,29 +840,12 @@ static int use_xml; /* Get Linux/x86 target description from running target. */ static const struct target_desc * -x86_linux_read_description (void) +x86_linux_read_description () { - unsigned int machine; - int is_elf64; - int xcr0_features; - int tid; - static uint64_t xcr0; - static int xsave_len; - struct regset_info *regset; - - tid = lwpid_of (current_thread); + int tid = lwpid_of (current_thread); - is_elf64 = linux_pid_exe_is_elf_64_file (tid, &machine); - - if (sizeof (void *) == 4) - { - if (is_elf64 > 0) - error (_("Can't debug 64-bit process with 32-bit GDBserver")); -#ifndef __x86_64__ - else if (machine == EM_X86_64) - error (_("Can't debug x86-64 process with 32-bit GDBserver")); -#endif - } + const char *error_msg + = _("Can't debug 64-bit process with 32-bit GDBserver"); /* If we are not allowed to send an XML target description then we need to use the hard-wired target descriptions. This corresponds to GDB's @@ -868,103 +855,53 @@ x86_linux_read_description (void) generate some alternative target descriptions. */ if (!use_xml) { + x86_linux_arch_size arch_size = x86_linux_ptrace_get_arch_size (tid); + bool is_64bit = arch_size.is_64bit (); + bool is_x32 = arch_size.is_x32 (); + + if (sizeof (void *) == 4 && is_64bit && !is_x32) + error ("%s", error_msg); + #ifdef __x86_64__ - if (machine == EM_X86_64) + if (is_64bit && !is_x32) return tdesc_amd64_linux_no_xml.get (); else #endif return tdesc_i386_linux_no_xml.get (); } -#if !defined __x86_64__ && defined HAVE_PTRACE_GETFPXREGS - if (machine == EM_386 && have_ptrace_getfpxregs == TRIBOOL_UNKNOWN) - { - elf_fpxregset_t fpxregs; - - if (ptrace (PTRACE_GETFPXREGS, tid, 0, (long) &fpxregs) < 0) - { - have_ptrace_getfpxregs = TRIBOOL_FALSE; - have_ptrace_getregset = TRIBOOL_FALSE; - return i386_linux_read_description (X86_XSTATE_X87); - } - else - have_ptrace_getfpxregs = TRIBOOL_TRUE; - } -#endif - - if (have_ptrace_getregset == TRIBOOL_UNKNOWN) - { - uint64_t xstateregs[(X86_XSTATE_SSE_SIZE / sizeof (uint64_t))]; - struct iovec iov; - - iov.iov_base = xstateregs; - iov.iov_len = sizeof (xstateregs); - - /* Check if PTRACE_GETREGSET works. */ - if (ptrace (PTRACE_GETREGSET, tid, - (unsigned int) NT_X86_XSTATE, (long) &iov) < 0) - have_ptrace_getregset = TRIBOOL_FALSE; - else - { - have_ptrace_getregset = TRIBOOL_TRUE; - - /* Get XCR0 from XSAVE extended state. */ - xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET - / sizeof (uint64_t))]; - - /* No MPX on x32. */ - if (machine == EM_X86_64 && !is_elf64) - xcr0 &= ~X86_XSTATE_MPX; - - xsave_len = x86_xsave_length (); - - /* Use PTRACE_GETREGSET if it is available. */ - for (regset = x86_regsets; - regset->fill_function != NULL; regset++) - if (regset->get_request == PTRACE_GETREGSET) - regset->size = xsave_len; - else if (regset->type != GENERAL_REGS) - regset->size = 0; - } - } - - /* Check the native XCR0 only if PTRACE_GETREGSET is available. */ - xcr0_features = (have_ptrace_getregset == TRIBOOL_TRUE - && (xcr0 & X86_XSTATE_ALL_MASK)); + /* If have_ptrace_getregset is changed to true by calling + x86_linux_tdesc_for_tid then we will perform some additional + initialisation. */ + bool have_ptrace_getregset_is_unknown + = have_ptrace_getregset == TRIBOOL_UNKNOWN; - if (xcr0_features) - i387_set_xsave_mask (xcr0, xsave_len); + /* Get pointers to where we should store the xcr0 and xsave_layout + values. These will be filled in by x86_linux_tdesc_for_tid the first + time that the function is called. Subsequent calls will not modify + the stored values. */ + std::pair<uint64_t *, x86_xsave_layout *> storage + = i387_get_xsave_storage (); - if (machine == EM_X86_64) - { -#ifdef __x86_64__ - const target_desc *tdesc = NULL; + const target_desc *tdesc + = x86_linux_tdesc_for_tid (tid, error_msg, storage.first, storage.second); - if (xcr0_features) - { - tdesc = amd64_linux_read_description (xcr0 & X86_XSTATE_ALL_MASK, - !is_elf64); - } - - if (tdesc == NULL) - tdesc = amd64_linux_read_description (X86_XSTATE_SSE_MASK, !is_elf64); - return tdesc; -#endif - } - else + if (have_ptrace_getregset_is_unknown + && have_ptrace_getregset == TRIBOOL_TRUE) { - const target_desc *tdesc = NULL; - - if (xcr0_features) - tdesc = i386_linux_read_description (xcr0 & X86_XSTATE_ALL_MASK); - - if (tdesc == NULL) - tdesc = i386_linux_read_description (X86_XSTATE_SSE); - - return tdesc; + int xsave_len = x86_xsave_length (); + + /* Use PTRACE_GETREGSET if it is available. */ + for (regset_info *regset = x86_regsets; + regset->fill_function != nullptr; + regset++) + if (regset->get_request == PTRACE_GETREGSET) + regset->size = xsave_len; + else if (regset->type != GENERAL_REGS) + regset->size = 0; } - gdb_assert_not_reached ("failed to return tdesc"); + return tdesc; } /* Update all the target description of all processes; a new GDB diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc index cd3b5d80e37..af3735aa895 100644 --- a/gdbserver/linux-x86-tdesc.cc +++ b/gdbserver/linux-x86-tdesc.cc @@ -23,8 +23,10 @@ #include "gdbsupport/x86-xstate.h" #ifdef __x86_64__ #include "arch/amd64.h" +#include "arch/amd64-linux-tdesc.h" #endif #include "x86-tdesc.h" +#include "arch/i386-linux-tdesc.h" /* Return the right x86_linux_tdesc index for a given XCR0. Return X86_TDESC_LAST if can't find a match. */ diff --git a/gdbserver/linux-x86-tdesc.h b/gdbserver/linux-x86-tdesc.h index f9561b129ae..576aaf5e165 100644 --- a/gdbserver/linux-x86-tdesc.h +++ b/gdbserver/linux-x86-tdesc.h @@ -46,11 +46,4 @@ int amd64_get_ipa_tdesc_idx (const struct target_desc *tdesc); const struct target_desc *i386_get_ipa_tdesc (int idx); -#ifdef __x86_64__ -const struct target_desc *amd64_linux_read_description (uint64_t xcr0, - bool is_x32); -#endif - -const struct target_desc *i386_linux_read_description (uint64_t xcr0); - #endif /* GDBSERVER_LINUX_X86_TDESC_H */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv7 8/9] gdbserver: update target description creation for x86/linux 2024-05-11 10:08 ` [PATCHv7 0/9] x86/Linux Target Description Changes Andrew Burgess ` (6 preceding siblings ...) 2024-05-11 10:08 ` [PATCHv7 7/9] gdb/gdbserver: share some code relating to target description creation Andrew Burgess @ 2024-05-11 10:08 ` Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 9/9] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess 8 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-11 10:08 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, jhb This commit is part of a series which aims to share more of the target description creation between GDB and gdbserver for x86/Linux. After some refactoring earlier in this series the shared x86_linux_tdesc_for_tid function was added into nat/x86-linux-tdesc.c. However, this function still relies on amd64_linux_read_description and i386_linux_read_description which are implemented separately for both gdbserver and GDB. Given that at their core, all these functions do is: 1. take an xcr0 value as input, 2. mask out some feature bits, 3. look for a cached pre-generated target description and return it if found, 4. if no cached target description is found then call either amd64_create_target_description or i386_create_target_description to create a new target description, which is then added to the cache. Return the newly created target description. The inner functions amd64_create_target_description and i386_create_target_description are already shared between GDB and gdbserver (in the gdb/arch/ directory), so the only thing that the *_read_description functions really do is add the caching layer, and it feels like this really could be shared. However, we have a small problem. Despite using the same {amd64,i386}_create_target_description functions in both GDB and gdbserver to create the target descriptions, on the gdbserver side we cache target descriptions based on a reduced set of xcr0 feature bits. What this means is that, in theory, different xcr0 values can map to the same cache entry, which could result in the wrong target description being used. However, I'm not sure if this can actually happen in reality. Within gdbserver we already split the target description cache based on i386, amd64, and x32. I suspect within a given gdbserver session we'll only see at most one target description for each of these. The cache conflicting problem is caused by xcr0_to_tdesc_idx, which maps an xcr0 value to a enum x86_linux_tdesc value, and there are only 7 usable values in enum x86_linux_tdesc. In contrast, on the GDB side there are 64, 32, and 16 cache slots for i386, amd64, and x32 respectively. On the GDB side it is much more important to cache things correctly as a single GDB session might connect to multiple different remote targets, each of which might have slightly different x86 architectures. And so, if we want to merge the target description caching between GDB and gdbserver, then we need to first update gdbserver so that it caches in the same way as GDB, that is, it needs to adopt a mechanism that allows for the same number of cache slots of each of i386, amd64, and x32. In this way, when the caching is shared, GDB's behaviour will not change. Unfortunately it's a little more complex than that due to the in process agent (IPA). When the IPA is in use, gdbserver sends a target description index to the IPA, and the IPA uses this to find the correct target description to use, the IPA having first generated every possible target description. Interestingly, there is certainly a bug here which results from only having 7 values in enum x86_linux_tdesc. As multiple possible target descriptions in gdbserver map to the same enum x86_linux_tdesc value, then, when the enum x86_linux_tdesc value is sent to the IPA there is no way for gdbserver to know that the IPA will select the correct target description. This bug will get fixed by this commit. ** START OF AN ASIDE ** Back in the day I suspect this approach of sending a target description index made perfect sense. However since this commit: commit a8806230241d201f808d856eaae4d44088117b0c Date: Thu Dec 7 17:07:01 2017 +0000 Initialize target description early in IPA I think that passing an index was probably a bad idea. We used to pass the index, and then use that index to lookup which target description to instantiate and use, the target description was not generated until the index arrived. However, the above commit fixed an issue where we can't call malloc() within (certain parts of) the IPA (apparently), so instead we now pre-compute _every_ possible target description within the IPA. The index is only used to lookup which of the (many) pre-computed target descriptions to use. It would (I think) have been easier all around if the IPA just self-inspected, figured out its own xcr0 value, and used that to create the one target description that is required. So long as the xcr0 to target description code is shared (at compile time) with gdbserver, then we can be sure that the IPA will derive the same target description as gdbserver, and we would avoid all this index passing business, which has made this commit so very, very painful. I did look at how a process might derive its own xcr0 value, but I don't believe this is actually that simple, so for now I've just doubled down on the index passing approach. While reviewing earlier iterations of this patch there has been discussion about the possibility of removing the IPA from GDB. That would certainly make all of the code touched in this patch much simpler, but I don't really want to do that as part of this series. ** END OF AN ASIDE ** Currently then for x86/linux, gdbserver sends a number between 0 and 7 to the IPA, and the IPA uses this to create a target description. However, I am proposing that gdbserver should now create one of (up to) 64 different target descriptions for i386, so this 0 to 7 index isn't going to be good enough any more (amd64 and x32 have slightly fewer possible target descriptions, but still more than 8, so the problem is the same). For a while I wondered if I was going to have to try and find some backward compatible solution for this mess. But after seeing how lightly the IPA is actually documented, I wonder if it is not the case that there is a tight coupling between a version of gdbserver and a version of the IPA? At least I'm hoping so, and that's what I've assumed in this commit. In this commit I have thrown out the old IPA target description index numbering scheme, and switched to a completely new numbering scheme. Instead of the index that is passed being arbitrary, the index is instead calculated from the set of xcr0 features that are present on the target. Within the IPA we can then reverse this logic to recreate the xcr0 value based on the index, and from the xcr0 value we can choose the correct target description. With the gdbserver to IPA numbering scheme issue resolved I have then update the gdbserver versions of amd64_linux_read_description and i386_linux_read_description so that they cache target descriptions using the same set of xcr0 features as GDB itself. After this gdbserver should now always come up with the same target description as GDB does on any x86/Linux target. This commit does not introduce any new code sharing between GDB and gdbserver as previous commits in this series have done. Instead this commit is all about bringing GDB and gdbserver into alignment functionally so that the next commit(s) can merge the GDB and gdbserver versions of these functions. Notes On The Implementation --------------------------- Previously, within gdbserver, target descriptions were cached within arrays. These arrays were sized based on enum x86_linux_tdesc and xcr0_to_tdesc_idx returned the array (cache) index. Now we need different array lengths for each of i386, amd64, and x32. And the index to use within each array is calculated based on which xcr0 bits are set and valid for a particular target type. I really wanted to avoid having fixed array sizes, or having the set of relevant xcr0 bits encoded in multiple places. The solution I came up with was to create a single data structure which would contain a list of xcr0 bits along with flags to indicate which of the i386, amd64, and x32 targets the bit is relevant for. By making the table constexpr, and adding some constexpr helper functions, it is possible to calculate the sizes for the cache arrays at compile time, as well as the bit masks needed to each target type. During review it was pointed out[1] that possibly the failure to check the SSE and X87 bits for amd64/x32 targets might be an error, however, if this is the case then this is an issue that existed long before this patch. I'd really like to keep this patch focused on reworking the existing code and try to avoid changing how target descriptions are actually created, mostly out of fear that I'll break something. [1] https://inbox.sourceware.org/gdb-patches/MN2PR11MB4566070607318EE7E669A5E28E1B2@MN2PR11MB4566.namprd11.prod.outlook.com Approved-By: John Baldwin <jhb@FreeBSD.org> --- gdbserver/linux-amd64-ipa.cc | 43 +---- gdbserver/linux-i386-ipa.cc | 23 +-- gdbserver/linux-x86-low.cc | 24 ++- gdbserver/linux-x86-tdesc.cc | 335 ++++++++++++++++++++++++++--------- gdbserver/linux-x86-tdesc.h | 48 ++--- 5 files changed, 306 insertions(+), 167 deletions(-) diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index df5e6aca081..0c80812cc6f 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -164,47 +164,21 @@ supply_static_tracepoint_registers (struct regcache *regcache, #endif /* HAVE_UST */ -#if !defined __ILP32__ -/* Map the tdesc index to xcr0 mask. */ -static uint64_t idx2mask[X86_TDESC_LAST] = { - X86_XSTATE_X87_MASK, - X86_XSTATE_SSE_MASK, - X86_XSTATE_AVX_MASK, - X86_XSTATE_MPX_MASK, - X86_XSTATE_AVX_MPX_MASK, - X86_XSTATE_AVX_AVX512_MASK, - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, -}; -#endif - /* Return target_desc to use for IPA, given the tdesc index passed by gdbserver. */ const struct target_desc * get_ipa_tdesc (int idx) { - if (idx >= X86_TDESC_LAST) - { - internal_error ("unknown ipa tdesc index: %d", idx); - } + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); #if defined __ILP32__ - switch (idx) - { - case X86_TDESC_SSE: - return amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); - case X86_TDESC_AVX: - return amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); - case X86_TDESC_AVX_AVX512: - return amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, true); - default: - break; - } + bool is_x32 = true; #else - return amd64_linux_read_description (idx2mask[idx], false); + bool is_x32 = false; #endif - internal_error ("unknown ipa tdesc index: %d", idx); + return amd64_linux_read_description (xcr0, is_x32); } /* Allocate buffer for the jump pads. The branch instruction has a @@ -272,11 +246,10 @@ void initialize_low_tracepoint (void) { #if defined __ILP32__ - amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); - amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); - amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, true); + for (auto i = 0; i < x86_linux_x32_tdesc_count (); i++) + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), true); #else - for (auto i = 0; i < X86_TDESC_LAST; i++) - amd64_linux_read_description (idx2mask[i], false); + for (auto i = 0; i < x86_linux_amd64_tdesc_count (); i++) + amd64_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i), false); #endif } diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc index aa346fc9bc3..c1c3152fb04 100644 --- a/gdbserver/linux-i386-ipa.cc +++ b/gdbserver/linux-i386-ipa.cc @@ -245,28 +245,15 @@ initialize_fast_tracepoint_trampoline_buffer (void) } } -/* Map the tdesc index to xcr0 mask. */ -static uint64_t idx2mask[X86_TDESC_LAST] = { - X86_XSTATE_X87_MASK, - X86_XSTATE_SSE_MASK, - X86_XSTATE_AVX_MASK, - X86_XSTATE_MPX_MASK, - X86_XSTATE_AVX_MPX_MASK, - X86_XSTATE_AVX_AVX512_MASK, - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, -}; - /* Return target_desc to use for IPA, given the tdesc index passed by gdbserver. */ const struct target_desc * get_ipa_tdesc (int idx) { - if (idx >= X86_TDESC_LAST) - { - internal_error ("unknown ipa tdesc index: %d", idx); - } - return i386_linux_read_description (idx2mask[idx]); + uint64_t xcr0 = x86_linux_tdesc_idx_to_xcr0 (idx); + + return i386_linux_read_description (xcr0); } /* Allocate buffer for the jump pads. On i386, we can reach an arbitrary @@ -288,6 +275,6 @@ void initialize_low_tracepoint (void) { initialize_fast_tracepoint_trampoline_buffer (); - for (auto i = 0; i < X86_TDESC_LAST; i++) - i386_linux_read_description (idx2mask[i]); + for (auto i = 0; i < x86_linux_i386_tdesc_count (); i++) + i386_linux_read_description (x86_linux_tdesc_idx_to_xcr0 (i)); } diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index 1c3d55f17fc..ab225c57fa2 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -2878,14 +2878,28 @@ x86_target::get_ipa_tdesc_idx () struct regcache *regcache = get_thread_regcache (current_thread, 0); const struct target_desc *tdesc = regcache->tdesc; + if (!use_xml) + { + /* If USE_XML is false then we should be using one of these target + descriptions, see x86_linux_read_description for where we choose + one of these. Both of these descriptions are created from this + fixed xcr0 value X86_XSTATE_SSE_MASK. */ + gdb_assert (tdesc == tdesc_i386_linux_no_xml.get () #ifdef __x86_64__ - return amd64_get_ipa_tdesc_idx (tdesc); -#endif + || tdesc == tdesc_amd64_linux_no_xml.get () +#endif /* __x86_64__ */ + ); + return x86_linux_xcr0_to_tdesc_idx (X86_XSTATE_SSE_MASK); + } - if (tdesc == tdesc_i386_linux_no_xml.get ()) - return X86_TDESC_SSE; + /* The xcr0 value and xsave layout value are cached when the target + description is read. Grab their cache location, and use the cached + value to calculate a tdesc index. */ + std::pair<uint64_t *, x86_xsave_layout *> storage + = i387_get_xsave_storage (); + uint64_t xcr0 = *storage.first; - return i386_get_ipa_tdesc_idx (tdesc); + return x86_linux_xcr0_to_tdesc_idx (xcr0); } /* The linux target ops object. */ diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc index af3735aa895..f5636898d71 100644 --- a/gdbserver/linux-x86-tdesc.cc +++ b/gdbserver/linux-x86-tdesc.cc @@ -28,96 +28,297 @@ #include "x86-tdesc.h" #include "arch/i386-linux-tdesc.h" -/* Return the right x86_linux_tdesc index for a given XCR0. Return - X86_TDESC_LAST if can't find a match. */ +/* A structure used to describe a single xstate feature bit that might, or + might not, be checked for when creating a target description for one of + i386, amd64, or x32. -static enum x86_linux_tdesc -xcr0_to_tdesc_idx (uint64_t xcr0, bool is_x32) + The different CPU/ABI types check for different xstate features when + creating a target description. + + We want to cache target descriptions, and this is currently done in + three separate caches, one each for i386, amd64, and x32. Additionally, + the caching we're discussing here is Linux only, and for Linux, the only + thing that has an impact on target description creation is the xcr0 + value. + + In order to ensure the cache functions correctly we need to filter out + only those xcr0 feature bits that are relevant, we can then cache target + descriptions based on the relevant feature bits. Two xcr0 values might + be different, but have the same relevant feature bits. In this case we + would expect the two xcr0 values to map to the same cache entry. */ + +struct x86_xstate_feature { + /* The xstate feature mask. This is a mask against an xcr0 value. */ + uint64_t feature; + + /* Is this feature checked when creating an i386 target description. */ + bool is_i386; + + /* Is this feature checked when creating an amd64 target description. */ + bool is_amd64; + + /* Is this feature checked when creating an x32 target description. */ + bool is_x32; +}; + +/* A constant table that describes all of the xstate features that are + checked when building a target description for i386, amd64, or x32. + + If in the future, due to simplifications or refactoring, this table ever + ends up with 'true' for every xcr0 feature on every target type, then this + is an indication that this table should probably be removed, and that the + rest of the code in this file can be simplified. */ + +static constexpr x86_xstate_feature x86_linux_all_xstate_features[] = { + /* Feature, i386, amd64, x32. */ + { X86_XSTATE_PKRU, true, true, true }, + { X86_XSTATE_AVX512, true, true, true }, + { X86_XSTATE_AVX, true, true, true }, + { X86_XSTATE_MPX, true, true, false }, + { X86_XSTATE_SSE, true, false, false }, + { X86_XSTATE_X87, true, false, false } +}; + +/* Return a compile time constant which is a mask of all the xstate features + that are checked for when building an i386 target description. */ + +static constexpr uint64_t +x86_linux_i386_xcr0_feature_mask () { - if (xcr0 & X86_XSTATE_PKRU) - { - if (is_x32) - { - /* No x32 MPX and PKU, fall back to avx_avx512. */ - return X86_TDESC_AVX_AVX512; - } - else - return X86_TDESC_AVX_MPX_AVX512_PKU; - } - else if (xcr0 & X86_XSTATE_AVX512) - return X86_TDESC_AVX_AVX512; - else if ((xcr0 & X86_XSTATE_AVX_MPX_MASK) == X86_XSTATE_AVX_MPX_MASK) + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_i386) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the xstate features + that are checked for when building an amd64 target description. */ + +static constexpr uint64_t +x86_linux_amd64_xcr0_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_amd64) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the xstate features + that are checked for when building an x32 target description. */ + +static constexpr uint64_t +x86_linux_x32_xcr0_feature_mask () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_x32) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a count of the number of xstate + features that are checked for when building an i386 target description. */ + +static constexpr int +x86_linux_i386_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_i386) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of xstate + features that are checked for when building an amd64 target description. */ + +static constexpr int +x86_linux_amd64_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_amd64) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of xstate + features that are checked for when building an x32 target description. */ + +static constexpr int +x86_linux_x32_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_x32) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +#ifdef IN_PROCESS_AGENT + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_amd64_tdesc_count () +{ + return x86_linux_amd64_tdesc_count_1 (); +} + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_x32_tdesc_count () +{ + return x86_linux_x32_tdesc_count_1 (); +} + +/* See linux-x86-tdesc.h. */ + +int +x86_linux_i386_tdesc_count () +{ + return x86_linux_i386_tdesc_count_1 (); +} + +#endif /* IN_PROCESS_AGENT */ + +/* Convert an xcr0 value into an integer. The integer will be passed to + the in-process-agent where it will then be passed to + x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ + +int +x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0) +{ + /* The following table shows which features are checked for when creating + the target descriptions (see nat/x86-linux-tdesc.c), the feature order + represents the bit order within the generated index number. + + i386 | x87 sse mpx avx avx512 pkru + amd64 | mpx avx avx512 pkru + i32 | avx avx512 pkru + + The features are ordered so that for each mode (i386, amd64, i32) the + generated index will form a continuous range. */ + + int idx = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_xstate_features); ++i) { - if (is_x32) /* No MPX on x32. */ - return X86_TDESC_AVX; - else - return X86_TDESC_AVX_MPX; + if ((xcr0 & x86_linux_all_xstate_features[i].feature) + == x86_linux_all_xstate_features[i].feature) + idx |= (1 << i); } - else if (xcr0 & X86_XSTATE_MPX) + + return idx; +} + +#ifdef IN_PROCESS_AGENT + +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +uint64_t +x86_linux_tdesc_idx_to_xcr0 (int idx) +{ + uint64_t xcr0 = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_xstate_features); ++i) { - if (is_x32) /* No MPX on x32. */ - return X86_TDESC_AVX; - else - return X86_TDESC_MPX; + if ((idx & (1 << i)) != 0) + xcr0 |= x86_linux_all_xstate_features[i].feature; } - else if (xcr0 & X86_XSTATE_AVX) - return X86_TDESC_AVX; - else if (xcr0 & X86_XSTATE_SSE) - return X86_TDESC_SSE; - else if (xcr0 & X86_XSTATE_X87) - return X86_TDESC_MMX; - else - return X86_TDESC_LAST; + + return xcr0; } +#endif /* IN_PROCESS_AGENT */ + #if defined __i386__ || !defined IN_PROCESS_AGENT -static struct target_desc *i386_tdescs[X86_TDESC_LAST] = { }; +/* A cache of all possible i386 target descriptions. */ -/* Return the target description according to XCR0. */ +static struct target_desc *i386_tdescs[x86_linux_i386_tdesc_count_1 ()] = { }; + +/* See nat/x86-linux-tdesc.h. */ const struct target_desc * i386_linux_read_description (uint64_t xcr0) { - enum x86_linux_tdesc idx = xcr0_to_tdesc_idx (xcr0, false); + xcr0 &= x86_linux_i386_xcr0_feature_mask (); + int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - if (idx == X86_TDESC_LAST) - return NULL; + gdb_assert (idx >= 0 && idx < x86_linux_i386_tdesc_count_1 ()); - struct target_desc **tdesc = &i386_tdescs[idx]; + target_desc **tdesc = &i386_tdescs[idx]; - if (*tdesc == NULL) + if (*tdesc == nullptr) { *tdesc = i386_create_target_description (xcr0, true, false); init_target_desc (*tdesc, i386_expedite_regs); } - return *tdesc;; + return *tdesc; } #endif #ifdef __x86_64__ -static target_desc *amd64_tdescs[X86_TDESC_LAST] = { }; -static target_desc *x32_tdescs[X86_TDESC_LAST] = { }; +/* A cache of all possible amd64 target descriptions. */ + +static target_desc *amd64_tdescs[x86_linux_amd64_tdesc_count_1 ()] = { }; + +/* A cache of all possible x32 target descriptions. */ + +static target_desc *x32_tdescs[x86_linux_x32_tdesc_count_1 ()] = { }; + +/* See nat/x86-linux-tdesc.h. */ const struct target_desc * amd64_linux_read_description (uint64_t xcr0, bool is_x32) { - enum x86_linux_tdesc idx = xcr0_to_tdesc_idx (xcr0, is_x32); + if (is_x32) + xcr0 &= x86_linux_x32_xcr0_feature_mask (); + else + xcr0 &= x86_linux_amd64_xcr0_feature_mask (); + + int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - if (idx == X86_TDESC_LAST) - return NULL; + if (is_x32) + gdb_assert (idx >= 0 && idx < x86_linux_x32_tdesc_count_1 ()); + else + gdb_assert (idx >= 0 && idx < x86_linux_amd64_tdesc_count_1 ()); - struct target_desc **tdesc = NULL; + target_desc **tdesc = nullptr; if (is_x32) tdesc = &x32_tdescs[idx]; else tdesc = &amd64_tdescs[idx]; - if (*tdesc == NULL) + if (*tdesc == nullptr) { *tdesc = amd64_create_target_description (xcr0, is_x32, true, true); @@ -127,39 +328,3 @@ amd64_linux_read_description (uint64_t xcr0, bool is_x32) } #endif - -#ifndef IN_PROCESS_AGENT - -int -i386_get_ipa_tdesc_idx (const struct target_desc *tdesc) -{ - for (int i = 0; i < X86_TDESC_LAST; i++) - { - if (tdesc == i386_tdescs[i]) - return i; - } - - /* If none tdesc is found, return the one with minimum features. */ - return X86_TDESC_MMX; -} - -#if defined __x86_64__ -int -amd64_get_ipa_tdesc_idx (const struct target_desc *tdesc) -{ - for (int i = 0; i < X86_TDESC_LAST; i++) - { - if (tdesc == amd64_tdescs[i]) - return i; - } - for (int i = 0; i < X86_TDESC_LAST; i++) - { - if (tdesc == x32_tdescs[i]) - return i; - } - - return X86_TDESC_SSE; -} - -#endif -#endif diff --git a/gdbserver/linux-x86-tdesc.h b/gdbserver/linux-x86-tdesc.h index 576aaf5e165..c06af21da48 100644 --- a/gdbserver/linux-x86-tdesc.h +++ b/gdbserver/linux-x86-tdesc.h @@ -21,29 +21,29 @@ #ifndef GDBSERVER_LINUX_X86_TDESC_H #define GDBSERVER_LINUX_X86_TDESC_H -/* Note: since IPA obviously knows what ABI it's running on (i386 vs x86_64 - vs x32), it's sufficient to pass only the register set here. This, - together with the ABI known at IPA compile time, maps to a tdesc. */ - -enum x86_linux_tdesc { - X86_TDESC_MMX = 0, - X86_TDESC_SSE = 1, - X86_TDESC_AVX = 2, - X86_TDESC_MPX = 3, - X86_TDESC_AVX_MPX = 4, - X86_TDESC_AVX_AVX512 = 5, - X86_TDESC_AVX_MPX_AVX512_PKU = 6, - X86_TDESC_LAST = 7, -}; - -#if defined __i386__ || !defined IN_PROCESS_AGENT -int i386_get_ipa_tdesc_idx (const struct target_desc *tdesc); -#endif - -#if defined __x86_64__ && !defined IN_PROCESS_AGENT -int amd64_get_ipa_tdesc_idx (const struct target_desc *tdesc); -#endif - -const struct target_desc *i386_get_ipa_tdesc (int idx); +/* Convert an xcr0 value into an integer. The integer will be passed to + the in-process-agent where it will then be passed to + x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ + +extern int x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0); + +#ifdef IN_PROCESS_AGENT + +/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) + into an xcr0 value which can then be used to create a target + description. */ + +extern uint64_t x86_linux_tdesc_idx_to_xcr0 (int idx); + +/* Within the in-process-agent we need to pre-initialise all of the target + descriptions, to do this we need to know how many target descriptions + there are for each different target type. These functions return the + target description count for the relevant target. */ + +extern int x86_linux_amd64_tdesc_count (); +extern int x86_linux_x32_tdesc_count (); +extern int x86_linux_i386_tdesc_count (); + +#endif /* IN_PROCESS_AGENT */ #endif /* GDBSERVER_LINUX_X86_TDESC_H */ -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
* [PATCHv7 9/9] gdb/gdbserver: share x86/linux tdesc caching 2024-05-11 10:08 ` [PATCHv7 0/9] x86/Linux Target Description Changes Andrew Burgess ` (7 preceding siblings ...) 2024-05-11 10:08 ` [PATCHv7 8/9] gdbserver: update target description creation for x86/linux Andrew Burgess @ 2024-05-11 10:08 ` Andrew Burgess 8 siblings, 0 replies; 116+ messages in thread From: Andrew Burgess @ 2024-05-11 10:08 UTC (permalink / raw) To: gdb-patches; +Cc: Andrew Burgess, felix.willgerodt, jhb This commit builds on the previous series of commits to share the target description caching code between GDB and gdbserver for x86/Linux targets. The objective of this commit is to move the four functions (2 each of) i386_linux_read_description and amd64_linux_read_description into the gdb/arch/ directory and combine them so we have just a single copy of each. Then GDB, gdbserver, and the in-process-agent (IPA) will link against these shared functions. One curiosity with this patch is the function x86_linux_post_init_tdesc. On the gdbserver side the two functions amd64_linux_read_description and i386_linux_read_description have some functionality that is not present on the GDB side, there is some additional configuration that is performed as each target description is created, to setup the expedited registers. To support this I've added the function x86_linux_post_init_tdesc. This function is called from the two *_linux_read_description functions, but is implemented separately for GDB and gdbserver. An alternative approach that avoids adding x86_linux_post_init_tdesc would be to have x86_linux_tdesc_for_tid return a non-const target description, then in x86_target::low_arch_setup we could inspect the target description to figure out if it is 64-bit or not, and modify the target description as needed. In the end I think that adding the x86_linux_post_init_tdesc function is the simpler solution. The contents of gdbserver/linux-x86-low.cc have moved to gdb/arch/x86-linux-tdesc-features.c, and gdbserver/linux-x86-tdesc.h has moved to gdb/arch/x86-linux-tdesc-features.h, this change leads to some updates in the #includes in the gdbserver/ directory. This commit also changes how target descriptions are cached. Previously both GDB and gdbserver used static C-style arrays to act as the tdesc cache. This was fine, except for two problems. Either the C-style arrays would need to be placed in x86-linux-tdesc-features.c, which would allow us to use the x86_linux_*_tdesc_count_1() functions to size the arrays for us, or we'd need to hard code the array sizes using separate #defines, which we'd then have to keep in sync with the rest of the code in x86-linux-tdesc-features.c. Given both of these problems I decided a better solution would be to just switch to using a std::unordered_map to act as the cache. This will resize automatically, and we can use the xcr0 value as the key. At first inspection, using xcr0 might seem to be a problem; after all the {i386,amd64}_create_target_description functions take more than just the xcr0 value. However, this patch is only for x86/Linux targets, and for x86/Linux all of the other flags passed to the tdesc creation functions have constant values and so are irrelevant when we consider tdesc caching. For testing I've done the following: - Built on x86-64 GNU/Linux for all targets, and just for the native target, - Build on i386 GNU/Linux for all targets, and just for the native target, - Build on a 64-bit, non-x86 GNU/Linux for all targets, just for the native target, and for targets x86_64-*-linux and i386-*-linux. --- gdb/Makefile.in | 5 + gdb/amd64-linux-tdep.c | 31 -- gdb/arch/amd64-linux-tdesc.c | 61 ++++ gdb/arch/i386-linux-tdesc.c | 51 +++ gdb/arch/x86-linux-tdesc-features.c | 267 +++++++++++++++ .../arch/x86-linux-tdesc-features.h | 49 ++- gdb/arch/x86-linux-tdesc.h | 37 +++ gdb/configure.nat | 6 +- gdb/configure.tgt | 11 +- gdb/i386-linux-tdep.c | 26 +- gdbserver/configure.srv | 9 + gdbserver/linux-amd64-ipa.cc | 2 +- gdbserver/linux-i386-ipa.cc | 2 +- gdbserver/linux-x86-low.cc | 2 +- gdbserver/linux-x86-tdesc.cc | 311 +----------------- 15 files changed, 489 insertions(+), 381 deletions(-) create mode 100644 gdb/arch/amd64-linux-tdesc.c create mode 100644 gdb/arch/i386-linux-tdesc.c create mode 100644 gdb/arch/x86-linux-tdesc-features.c rename gdbserver/linux-x86-tdesc.h => gdb/arch/x86-linux-tdesc-features.h (52%) create mode 100644 gdb/arch/x86-linux-tdesc.h diff --git a/gdb/Makefile.in b/gdb/Makefile.in index 13ca417706b..7d912f00fb3 100644 --- a/gdb/Makefile.in +++ b/gdb/Makefile.in @@ -748,6 +748,7 @@ ALL_64_TARGET_OBS = \ arch/aarch64-insn.o \ arch/aarch64-mte-linux.o \ arch/aarch64-scalable-linux.o \ + arch/amd64-linux-tdesc.o \ arch/amd64.o \ arch/riscv.o \ bpf-tdep.o \ @@ -788,9 +789,11 @@ ALL_TARGET_OBS = \ arch/arm.o \ arch/arm-get-next-pcs.o \ arch/arm-linux.o \ + arch/i386-linux-tdesc.o \ arch/i386.o \ arch/loongarch.o \ arch/ppc-linux-common.o \ + arch/x86-linux-tdesc-features.o \ arm-bsd-tdep.o \ arm-fbsd-tdep.o \ arm-linux-tdep.o \ @@ -1559,6 +1562,8 @@ HFILES_NO_SRCDIR = \ arch/ppc-linux-common.h \ arch/ppc-linux-tdesc.h \ arch/riscv.h \ + arch/x86-linux-tdesc-features.h \ + arch/x86-linux-tdesc.h \ cli/cli-cmds.h \ cli/cli-decode.h \ cli/cli-script.h \ diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c index f9647dce1e6..d7662cac572 100644 --- a/gdb/amd64-linux-tdep.c +++ b/gdb/amd64-linux-tdep.c @@ -1578,37 +1578,6 @@ amd64_linux_record_signal (struct gdbarch *gdbarch, return 0; } -const target_desc * -amd64_linux_read_description (uint64_t xcr0_features_bit, bool is_x32) -{ - static target_desc *amd64_linux_tdescs \ - [2/*AVX*/][2/*MPX*/][2/*AVX512*/][2/*PKRU*/] = {}; - static target_desc *x32_linux_tdescs \ - [2/*AVX*/][2/*AVX512*/][2/*PKRU*/] = {}; - - target_desc **tdesc; - - if (is_x32) - { - tdesc = &x32_linux_tdescs[(xcr0_features_bit & X86_XSTATE_AVX) ? 1 : 0 ] - [(xcr0_features_bit & X86_XSTATE_AVX512) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_PKRU) ? 1 : 0]; - } - else - { - tdesc = &amd64_linux_tdescs[(xcr0_features_bit & X86_XSTATE_AVX) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_MPX) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_AVX512) ? 1 : 0] - [(xcr0_features_bit & X86_XSTATE_PKRU) ? 1 : 0]; - } - - if (*tdesc == NULL) - *tdesc = amd64_create_target_description (xcr0_features_bit, is_x32, - true, true); - - return *tdesc; -} - /* Get Linux/x86 target description from core dump. */ static const struct target_desc * diff --git a/gdb/arch/amd64-linux-tdesc.c b/gdb/arch/amd64-linux-tdesc.c new file mode 100644 index 00000000000..794cd5c815d --- /dev/null +++ b/gdb/arch/amd64-linux-tdesc.c @@ -0,0 +1,61 @@ +/* Target description related code for GNU/Linux x86-64. + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#include "arch/x86-linux-tdesc.h" +#include "arch/amd64-linux-tdesc.h" +#include "arch/amd64.h" +#include "arch/x86-linux-tdesc-features.h" + + +/* See arch/amd64-linux-tdesc.h. */ + +const struct target_desc * +amd64_linux_read_description (uint64_t xcr0, bool is_x32) +{ + /* The type used for the amd64 and x32 target description caches. */ + using tdesc_cache_type = std::unordered_map<uint64_t, const target_desc_up>; + + /* Caches for the previously seen amd64 and x32 target descriptions, + indexed by the xcr0 value that created the target description. These + need to be static within this function to ensure they are initialised + before first use. */ + static tdesc_cache_type amd64_tdesc_cache, x32_tdesc_cache; + + tdesc_cache_type &tdesc_cache = is_x32 ? x32_tdesc_cache : amd64_tdesc_cache; + + xcr0 &= is_x32 + ? x86_linux_x32_xcr0_feature_mask () + : x86_linux_amd64_xcr0_feature_mask (); + + const auto it = tdesc_cache.find (xcr0); + if (it != tdesc_cache.end ()) + return it->second.get (); + + /* Create the previously unseen target description. */ + target_desc_up tdesc (amd64_create_target_description (xcr0, is_x32, + true, true)); + x86_linux_post_init_tdesc (tdesc.get (), true); + + /* Add to the cache, and return a pointer borrowed from the + target_desc_up. This is safe as the cache (and the pointers contained + within it) are not deleted until GDB exits. */ + target_desc *ptr = tdesc.get (); + tdesc_cache.emplace (xcr0, std::move (tdesc)); + return ptr; +} diff --git a/gdb/arch/i386-linux-tdesc.c b/gdb/arch/i386-linux-tdesc.c new file mode 100644 index 00000000000..78b0cef841e --- /dev/null +++ b/gdb/arch/i386-linux-tdesc.c @@ -0,0 +1,51 @@ +/* Target description related code for GNU/Linux i386. + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#include "arch/x86-linux-tdesc.h" +#include "arch/i386-linux-tdesc.h" +#include "arch/i386.h" +#include "arch/x86-linux-tdesc-features.h" + +/* See arch/i386-linux-tdesc.h. */ + +const target_desc * +i386_linux_read_description (uint64_t xcr0) +{ + /* Cache of previously seen i386 target descriptions, indexed by the xcr0 + value that created the target description. This needs to be static + within this function to ensure it is initialised before first use. */ + static std::unordered_map<uint64_t, const target_desc_up> i386_tdesc_cache; + + xcr0 &= x86_linux_i386_xcr0_feature_mask (); + + const auto it = i386_tdesc_cache.find (xcr0); + if (it != i386_tdesc_cache.end ()) + return it->second.get (); + + /* Create the previously unseen target description. */ + target_desc_up tdesc (i386_create_target_description (xcr0, true, false)); + x86_linux_post_init_tdesc (tdesc.get (), false); + + /* Add to the cache, and return a pointer borrowed from the + target_desc_up. This is safe as the cache (and the pointers contained + within it) are not deleted until GDB exits. */ + target_desc *ptr = tdesc.get (); + i386_tdesc_cache.emplace (xcr0, std::move (tdesc)); + return ptr; +} diff --git a/gdb/arch/x86-linux-tdesc-features.c b/gdb/arch/x86-linux-tdesc-features.c new file mode 100644 index 00000000000..55444d91bd5 --- /dev/null +++ b/gdb/arch/x86-linux-tdesc-features.c @@ -0,0 +1,267 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#include "arch/x86-linux-tdesc-features.h" + +/* A structure used to describe a single xstate feature bit that might, or + might not, be checked for when creating a target description for one of + i386, amd64, or x32. + + The different CPU/ABI types check for different xstate features when + creating a target description. + + We want to cache target descriptions, and this is currently done in + three separate caches, one each for i386, amd64, and x32. Additionally, + the caching we're discussing here is Linux only, and for Linux, the only + thing that has an impact on target description creation is the xcr0 + value. + + In order to ensure the cache functions correctly we need to filter out + only those xcr0 feature bits that are relevant, we can then cache target + descriptions based on the relevant feature bits. Two xcr0 values might + be different, but have the same relevant feature bits. In this case we + would expect the two xcr0 values to map to the same cache entry. */ + +struct x86_xstate_feature { + /* The xstate feature mask. This is a mask against an xcr0 value. */ + uint64_t feature; + + /* Is this feature checked when creating an i386 target description. */ + bool is_i386; + + /* Is this feature checked when creating an amd64 target description. */ + bool is_amd64; + + /* Is this feature checked when creating an x32 target description. */ + bool is_x32; +}; + +/* A constant table that describes all of the xstate features that are + checked when building a target description for i386, amd64, or x32. + + If in the future, due to simplifications or refactoring, this table ever + ends up with 'true' for every xcr0 feature on every target type, then this + is an indication that this table should probably be removed, and that the + rest of the code in this file can be simplified. */ + +static constexpr x86_xstate_feature x86_linux_all_xstate_features[] = { + /* Feature, i386, amd64, x32. */ + { X86_XSTATE_PKRU, true, true, true }, + { X86_XSTATE_AVX512, true, true, true }, + { X86_XSTATE_AVX, true, true, true }, + { X86_XSTATE_MPX, true, true, false }, + { X86_XSTATE_SSE, true, false, false }, + { X86_XSTATE_X87, true, false, false } +}; + +/* Return a compile time constant which is a mask of all the xstate features + that are checked for when building an i386 target description. */ + +static constexpr uint64_t +x86_linux_i386_xcr0_feature_mask_1 () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_i386) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the xstate features + that are checked for when building an amd64 target description. */ + +static constexpr uint64_t +x86_linux_amd64_xcr0_feature_mask_1 () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_amd64) + mask |= entry.feature; + + return mask; +} + +/* Return a compile time constant which is a mask of all the xstate features + that are checked for when building an x32 target description. */ + +static constexpr uint64_t +x86_linux_x32_xcr0_feature_mask_1 () +{ + uint64_t mask = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_x32) + mask |= entry.feature; + + return mask; +} + +/* See arch/x86-linux-tdesc-features.h. */ + +uint64_t +x86_linux_amd64_xcr0_feature_mask () +{ + return x86_linux_amd64_xcr0_feature_mask_1 (); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +uint64_t +x86_linux_x32_xcr0_feature_mask () +{ + return x86_linux_x32_xcr0_feature_mask_1 (); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +uint64_t +x86_linux_i386_xcr0_feature_mask () +{ + return x86_linux_i386_xcr0_feature_mask_1 (); +} + +#ifdef GDBSERVER + +/* See arch/x86-linux-tdesc-features.h. */ + +int +x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0) +{ + /* The following table shows which features are checked for when creating + the target descriptions (see nat/x86-linux-tdesc.c), the feature order + represents the bit order within the generated index number. + + i386 | x87 sse mpx avx avx512 pkru + amd64 | mpx avx avx512 pkru + i32 | avx avx512 pkru + + The features are ordered so that for each mode (i386, amd64, i32) the + generated index will form a continuous range. */ + + int idx = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_xstate_features); ++i) + { + if ((xcr0 & x86_linux_all_xstate_features[i].feature) + == x86_linux_all_xstate_features[i].feature) + idx |= (1 << i); + } + + return idx; +} + +#endif /* GDBSERVER */ + +#ifdef IN_PROCESS_AGENT + +/* Return a compile time constant which is a count of the number of xstate + features that are checked for when building an i386 target description. */ + +static constexpr int +x86_linux_i386_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_i386) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of xstate + features that are checked for when building an amd64 target description. */ + +static constexpr int +x86_linux_amd64_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_amd64) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* Return a compile time constant which is a count of the number of xstate + features that are checked for when building an x32 target description. */ + +static constexpr int +x86_linux_x32_tdesc_count_1 () +{ + uint64_t count = 0; + + for (const auto &entry : x86_linux_all_xstate_features) + if (entry.is_x32) + ++count; + + gdb_assert (count > 0); + + return (1 << count); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +int +x86_linux_amd64_tdesc_count () +{ + return x86_linux_amd64_tdesc_count_1 (); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +int +x86_linux_x32_tdesc_count () +{ + return x86_linux_x32_tdesc_count_1 (); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +int +x86_linux_i386_tdesc_count () +{ + return x86_linux_i386_tdesc_count_1 (); +} + +/* See arch/x86-linux-tdesc-features.h. */ + +uint64_t +x86_linux_tdesc_idx_to_xcr0 (int idx) +{ + uint64_t xcr0 = 0; + + for (int i = 0; i < ARRAY_SIZE (x86_linux_all_xstate_features); ++i) + { + if ((idx & (1 << i)) != 0) + xcr0 |= x86_linux_all_xstate_features[i].feature; + } + + return xcr0; +} + +#endif /* IN_PROCESS_AGENT */ diff --git a/gdbserver/linux-x86-tdesc.h b/gdb/arch/x86-linux-tdesc-features.h similarity index 52% rename from gdbserver/linux-x86-tdesc.h rename to gdb/arch/x86-linux-tdesc-features.h index c06af21da48..fb0017e46db 100644 --- a/gdbserver/linux-x86-tdesc.h +++ b/gdb/arch/x86-linux-tdesc-features.h @@ -1,7 +1,6 @@ -/* Low level support for x86 (i386 and x86-64), shared between gdbserver - and IPA. +/* Target description related code for GNU/Linux x86 (i386 and x86-64). - Copyright (C) 2016-2024 Free Software Foundation, Inc. + Copyright (C) 2024 Free Software Foundation, Inc. This file is part of GDB. @@ -18,32 +17,46 @@ You should have received a copy of the GNU General Public License along with this program. If not, see <http://www.gnu.org/licenses/>. */ -#ifndef GDBSERVER_LINUX_X86_TDESC_H -#define GDBSERVER_LINUX_X86_TDESC_H +#ifndef ARCH_X86_LINUX_TDESC_FEATURES_H +#define ARCH_X86_LINUX_TDESC_FEATURES_H -/* Convert an xcr0 value into an integer. The integer will be passed to - the in-process-agent where it will then be passed to - x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ +#include "gdbsupport/x86-xstate.h" +#include "gdbsupport/gdb_assert.h" + +/* Return a mask of X86_STATE_* feature flags. The returned mask indicates + the set of features which are checked for when creating the target + description for each of amd64, x32, and i386. */ + +extern uint64_t x86_linux_amd64_xcr0_feature_mask (); +extern uint64_t x86_linux_x32_xcr0_feature_mask (); +extern uint64_t x86_linux_i386_xcr0_feature_mask (); + +#ifdef GDBSERVER + +/* Convert an xcr0 value into an integer. The integer will be passed from + gdbserver to the in-process-agent where it will then be passed through + x86_linux_tdesc_idx_to_xcr0 to get back the original xcr0 value. */ extern int x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0); +#endif /* GDBSERVER */ + #ifdef IN_PROCESS_AGENT +/* Return the maximum possible number of target descriptions for each of + amd64, x32, and i386. These are used by the in-process-agent to + generate every possible target description. */ + +extern int x86_linux_amd64_tdesc_count (); +extern int x86_linux_x32_tdesc_count (); +extern int x86_linux_i386_tdesc_count (); + /* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) into an xcr0 value which can then be used to create a target description. */ extern uint64_t x86_linux_tdesc_idx_to_xcr0 (int idx); -/* Within the in-process-agent we need to pre-initialise all of the target - descriptions, to do this we need to know how many target descriptions - there are for each different target type. These functions return the - target description count for the relevant target. */ - -extern int x86_linux_amd64_tdesc_count (); -extern int x86_linux_x32_tdesc_count (); -extern int x86_linux_i386_tdesc_count (); - #endif /* IN_PROCESS_AGENT */ -#endif /* GDBSERVER_LINUX_X86_TDESC_H */ +#endif /* ARCH_X86_LINUX_TDESC_FEATURES_H */ diff --git a/gdb/arch/x86-linux-tdesc.h b/gdb/arch/x86-linux-tdesc.h new file mode 100644 index 00000000000..152592fcf76 --- /dev/null +++ b/gdb/arch/x86-linux-tdesc.h @@ -0,0 +1,37 @@ +/* Target description related code for GNU/Linux x86 (i386 and x86-64). + + Copyright (C) 2024 Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. */ + +#ifndef ARCH_X86_LINUX_TDESC_H +#define ARCH_X86_LINUX_TDESC_H + +struct target_desc; + +/* This function is called from amd64_linux_read_description and + i386_linux_read_description after a new target description has been + created, TDESC is the new target description, IS_64BIT will be true + when called from amd64_linux_read_description, otherwise IS_64BIT will + be false. If the *_linux_read_description functions found a cached + target description then this function will not be called. + + Both GDB and gdbserver have their own implementations of this + function. */ + +extern void x86_linux_post_init_tdesc (target_desc *tdesc, bool is_64bit); + +#endif /* ARCH_X86_LINUX_TDESC_H */ diff --git a/gdb/configure.nat b/gdb/configure.nat index a30b07ef8c5..5d6669f34d7 100644 --- a/gdb/configure.nat +++ b/gdb/configure.nat @@ -257,7 +257,8 @@ case ${gdb_host} in nat/x86-xstate.o \ i386-linux-nat.o x86-linux-nat.o nat/linux-btrace.o \ nat/x86-linux.o nat/x86-linux-dregs.o nat/i386-linux.o \ - nat/x86-linux-tdesc.o" + nat/x86-linux-tdesc.o arch/i386-linux-tdesc.o \ + arch/x86-linux-tdesc-features.o" ;; ia64) # Host: Intel IA-64 running GNU/Linux @@ -324,7 +325,8 @@ case ${gdb_host} in nat/x86-xstate.o amd64-nat.o amd64-linux-nat.o x86-linux-nat.o \ nat/linux-btrace.o \ nat/x86-linux.o nat/x86-linux-dregs.o nat/x86-linux-tdesc.o \ - nat/amd64-linux-siginfo.o" + nat/amd64-linux-siginfo.o arch/x86-linux-tdesc-features.o \ + arch/i386-linux-tdesc.o arch/amd64-linux-tdesc.o" ;; sparc) # Host: GNU/Linux UltraSPARC diff --git a/gdb/configure.tgt b/gdb/configure.tgt index 47a674201f9..8326c458eb1 100644 --- a/gdb/configure.tgt +++ b/gdb/configure.tgt @@ -320,10 +320,13 @@ i[34567]86-*-linux*) gdb_target_obs="i386-linux-tdep.o \ glibc-tdep.o \ solib-svr4.o symfile-mem.o \ - linux-tdep.o linux-record.o" + linux-tdep.o linux-record.o \ + arch/i386-linux-tdesc.o \ + arch/x86-linux-tdesc-features.o" if test "x$enable_64_bit_bfd" = "xyes"; then # Target: GNU/Linux x86-64 - gdb_target_obs="amd64-linux-tdep.o ${gdb_target_obs}" + gdb_target_obs="amd64-linux-tdep.o \ + arch/amd64-linux-tdesc.o ${gdb_target_obs}" fi ;; i[34567]86-*-gnu*) @@ -718,7 +721,9 @@ x86_64-*-linux*) # Target: GNU/Linux x86-64 gdb_target_obs="amd64-linux-tdep.o ${i386_tobjs} \ i386-linux-tdep.o glibc-tdep.o \ - solib-svr4.o symfile-mem.o linux-tdep.o linux-record.o" + solib-svr4.o symfile-mem.o linux-tdep.o linux-record.o \ + arch/i386-linux-tdesc.o arch/amd64-linux-tdesc.o \ + arch/x86-linux-tdesc-features.o" ;; x86_64-*-freebsd* | x86_64-*-kfreebsd*-gnu) # Target: FreeBSD/amd64 diff --git a/gdb/i386-linux-tdep.c b/gdb/i386-linux-tdep.c index 6a7490680c1..6debed21b89 100644 --- a/gdb/i386-linux-tdep.c +++ b/gdb/i386-linux-tdep.c @@ -41,6 +41,7 @@ #include "i387-tdep.h" #include "gdbsupport/x86-xstate.h" #include "arch/i386-linux-tdesc.h" +#include "arch/x86-linux-tdesc.h" /* The syscall's XML filename for i386. */ #define XML_SYSCALL_FILENAME_I386 "syscalls/i386-linux.xml" @@ -680,29 +681,12 @@ i386_linux_core_read_x86_xsave_layout (struct gdbarch *gdbarch, layout) != 0; } -/* See i386-linux-tdep.h. */ +/* See arch/x86-linux-tdesc.h. */ -const struct target_desc * -i386_linux_read_description (uint64_t xcr0) +void +x86_linux_post_init_tdesc (target_desc *tdesc, bool is_64bit) { - if (xcr0 == 0) - return NULL; - - static struct target_desc *i386_linux_tdescs \ - [2/*X87*/][2/*SSE*/][2/*AVX*/][2/*MPX*/][2/*AVX512*/][2/*PKRU*/] = {}; - struct target_desc **tdesc; - - tdesc = &i386_linux_tdescs[(xcr0 & X86_XSTATE_X87) ? 1 : 0] - [(xcr0 & X86_XSTATE_SSE) ? 1 : 0] - [(xcr0 & X86_XSTATE_AVX) ? 1 : 0] - [(xcr0 & X86_XSTATE_MPX) ? 1 : 0] - [(xcr0 & X86_XSTATE_AVX512) ? 1 : 0] - [(xcr0 & X86_XSTATE_PKRU) ? 1 : 0]; - - if (*tdesc == NULL) - *tdesc = i386_create_target_description (xcr0, true, false); - - return *tdesc; + /* Nothing. */ } /* Get Linux/x86 target description from core dump. */ diff --git a/gdbserver/configure.srv b/gdbserver/configure.srv index 538845b96d5..e24e40e85c1 100644 --- a/gdbserver/configure.srv +++ b/gdbserver/configure.srv @@ -111,12 +111,16 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" srv_tgtobj="${srv_tgtobj} nat/i386-linux.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" + srv_tgtobj="${srv_tgtobj} arch/x86-linux-tdesc-features.o" + srv_tgtobj="${srv_tgtobj} arch/i386-linux-tdesc.o" srv_linux_usrregs=yes srv_linux_regsets=yes srv_linux_thread_db=yes srv_linux_btrace=yes ipa_obj="linux-i386-ipa.o linux-x86-tdesc-ipa.o" ipa_obj="${ipa_obj} arch/i386-ipa.o" + ipa_obj="${ipa_obj} arch/x86-linux-tdesc-features-ipa.o" + ipa_obj="${ipa_obj} arch/i386-linux-tdesc-ipa.o" ;; i[34567]86-*-mingw*) srv_regobj="" srv_tgtobj="x86-low.o nat/x86-dregs.o win32-low.o" @@ -375,12 +379,17 @@ case "${gdbserver_host}" in srv_tgtobj="${srv_tgtobj} nat/x86-linux-dregs.o" srv_tgtobj="${srv_tgtobj} nat/x86-linux-tdesc.o" srv_tgtobj="${srv_tgtobj} nat/amd64-linux-siginfo.o" + srv_tgtobj="${srv_tgtobj} arch/x86-linux-tdesc-features.o" + srv_tgtobj="${srv_tgtobj} arch/amd64-linux-tdesc.o" + srv_tgtobj="${srv_tgtobj} arch/i386-linux-tdesc.o" srv_linux_usrregs=yes # This is for i386 progs. srv_linux_regsets=yes srv_linux_thread_db=yes srv_linux_btrace=yes ipa_obj="linux-amd64-ipa.o linux-x86-tdesc-ipa.o" ipa_obj="${ipa_obj} arch/amd64-ipa.o" + ipa_obj="${ipa_obj} arch/x86-linux-tdesc-features-ipa.o" + ipa_obj="${ipa_obj} arch/amd64-linux-tdesc-ipa.o" ;; x86_64-*-mingw*) srv_regobj="" srv_tgtobj="x86-low.o nat/x86-dregs.o" diff --git a/gdbserver/linux-amd64-ipa.cc b/gdbserver/linux-amd64-ipa.cc index 0c80812cc6f..89857f208b8 100644 --- a/gdbserver/linux-amd64-ipa.cc +++ b/gdbserver/linux-amd64-ipa.cc @@ -20,9 +20,9 @@ #include <sys/mman.h> #include "tracepoint.h" -#include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" #include "arch/amd64-linux-tdesc.h" +#include "arch/x86-linux-tdesc-features.h" /* fast tracepoints collect registers. */ diff --git a/gdbserver/linux-i386-ipa.cc b/gdbserver/linux-i386-ipa.cc index c1c3152fb04..8100c9f9840 100644 --- a/gdbserver/linux-i386-ipa.cc +++ b/gdbserver/linux-i386-ipa.cc @@ -20,9 +20,9 @@ #include <sys/mman.h> #include "tracepoint.h" -#include "linux-x86-tdesc.h" #include "gdbsupport/x86-xstate.h" #include "arch/i386-linux-tdesc.h" +#include "arch/x86-linux-tdesc-features.h" /* GDB register numbers. */ diff --git a/gdbserver/linux-x86-low.cc b/gdbserver/linux-x86-low.cc index ab225c57fa2..5b2dc68a1e8 100644 --- a/gdbserver/linux-x86-low.cc +++ b/gdbserver/linux-x86-low.cc @@ -35,6 +35,7 @@ #endif #include "arch/i386-linux-tdesc.h" +#include "arch/x86-linux-tdesc-features.h" #include "gdb_proc_service.h" /* Don't include elf/common.h if linux/elf.h got included by @@ -50,7 +51,6 @@ #include "nat/linux-nat.h" #include "nat/x86-linux.h" #include "nat/x86-linux-dregs.h" -#include "linux-x86-tdesc.h" #include "nat/x86-linux-tdesc.h" #ifdef __x86_64__ diff --git a/gdbserver/linux-x86-tdesc.cc b/gdbserver/linux-x86-tdesc.cc index f5636898d71..13c80762605 100644 --- a/gdbserver/linux-x86-tdesc.cc +++ b/gdbserver/linux-x86-tdesc.cc @@ -17,314 +17,19 @@ You should have received a copy of the GNU General Public License along with this program. If not, see <http://www.gnu.org/licenses/>. */ +#include "arch/x86-linux-tdesc.h" #include "tdesc.h" -#include "linux-x86-tdesc.h" -#include "arch/i386.h" -#include "gdbsupport/x86-xstate.h" -#ifdef __x86_64__ -#include "arch/amd64.h" -#include "arch/amd64-linux-tdesc.h" -#endif #include "x86-tdesc.h" -#include "arch/i386-linux-tdesc.h" - -/* A structure used to describe a single xstate feature bit that might, or - might not, be checked for when creating a target description for one of - i386, amd64, or x32. - - The different CPU/ABI types check for different xstate features when - creating a target description. - - We want to cache target descriptions, and this is currently done in - three separate caches, one each for i386, amd64, and x32. Additionally, - the caching we're discussing here is Linux only, and for Linux, the only - thing that has an impact on target description creation is the xcr0 - value. - - In order to ensure the cache functions correctly we need to filter out - only those xcr0 feature bits that are relevant, we can then cache target - descriptions based on the relevant feature bits. Two xcr0 values might - be different, but have the same relevant feature bits. In this case we - would expect the two xcr0 values to map to the same cache entry. */ - -struct x86_xstate_feature { - /* The xstate feature mask. This is a mask against an xcr0 value. */ - uint64_t feature; - - /* Is this feature checked when creating an i386 target description. */ - bool is_i386; - - /* Is this feature checked when creating an amd64 target description. */ - bool is_amd64; - - /* Is this feature checked when creating an x32 target description. */ - bool is_x32; -}; - -/* A constant table that describes all of the xstate features that are - checked when building a target description for i386, amd64, or x32. - - If in the future, due to simplifications or refactoring, this table ever - ends up with 'true' for every xcr0 feature on every target type, then this - is an indication that this table should probably be removed, and that the - rest of the code in this file can be simplified. */ - -static constexpr x86_xstate_feature x86_linux_all_xstate_features[] = { - /* Feature, i386, amd64, x32. */ - { X86_XSTATE_PKRU, true, true, true }, - { X86_XSTATE_AVX512, true, true, true }, - { X86_XSTATE_AVX, true, true, true }, - { X86_XSTATE_MPX, true, true, false }, - { X86_XSTATE_SSE, true, false, false }, - { X86_XSTATE_X87, true, false, false } -}; - -/* Return a compile time constant which is a mask of all the xstate features - that are checked for when building an i386 target description. */ - -static constexpr uint64_t -x86_linux_i386_xcr0_feature_mask () -{ - uint64_t mask = 0; - - for (const auto &entry : x86_linux_all_xstate_features) - if (entry.is_i386) - mask |= entry.feature; - - return mask; -} - -/* Return a compile time constant which is a mask of all the xstate features - that are checked for when building an amd64 target description. */ - -static constexpr uint64_t -x86_linux_amd64_xcr0_feature_mask () -{ - uint64_t mask = 0; - - for (const auto &entry : x86_linux_all_xstate_features) - if (entry.is_amd64) - mask |= entry.feature; - - return mask; -} - -/* Return a compile time constant which is a mask of all the xstate features - that are checked for when building an x32 target description. */ - -static constexpr uint64_t -x86_linux_x32_xcr0_feature_mask () -{ - uint64_t mask = 0; - - for (const auto &entry : x86_linux_all_xstate_features) - if (entry.is_x32) - mask |= entry.feature; - - return mask; -} - -/* Return a compile time constant which is a count of the number of xstate - features that are checked for when building an i386 target description. */ - -static constexpr int -x86_linux_i386_tdesc_count_1 () -{ - uint64_t count = 0; - - for (const auto &entry : x86_linux_all_xstate_features) - if (entry.is_i386) - ++count; - - gdb_assert (count > 0); - - return (1 << count); -} - -/* Return a compile time constant which is a count of the number of xstate - features that are checked for when building an amd64 target description. */ - -static constexpr int -x86_linux_amd64_tdesc_count_1 () -{ - uint64_t count = 0; - - for (const auto &entry : x86_linux_all_xstate_features) - if (entry.is_amd64) - ++count; - - gdb_assert (count > 0); - - return (1 << count); -} - -/* Return a compile time constant which is a count of the number of xstate - features that are checked for when building an x32 target description. */ - -static constexpr int -x86_linux_x32_tdesc_count_1 () -{ - uint64_t count = 0; - - for (const auto &entry : x86_linux_all_xstate_features) - if (entry.is_x32) - ++count; - - gdb_assert (count > 0); - - return (1 << count); -} - -#ifdef IN_PROCESS_AGENT - -/* See linux-x86-tdesc.h. */ - -int -x86_linux_amd64_tdesc_count () -{ - return x86_linux_amd64_tdesc_count_1 (); -} - -/* See linux-x86-tdesc.h. */ - -int -x86_linux_x32_tdesc_count () -{ - return x86_linux_x32_tdesc_count_1 (); -} - -/* See linux-x86-tdesc.h. */ - -int -x86_linux_i386_tdesc_count () -{ - return x86_linux_i386_tdesc_count_1 (); -} - -#endif /* IN_PROCESS_AGENT */ - -/* Convert an xcr0 value into an integer. The integer will be passed to - the in-process-agent where it will then be passed to - x86_linux_tdesc_idx_to_xcr0 to get back the xcr0 value. */ - -int -x86_linux_xcr0_to_tdesc_idx (uint64_t xcr0) -{ - /* The following table shows which features are checked for when creating - the target descriptions (see nat/x86-linux-tdesc.c), the feature order - represents the bit order within the generated index number. - - i386 | x87 sse mpx avx avx512 pkru - amd64 | mpx avx avx512 pkru - i32 | avx avx512 pkru - - The features are ordered so that for each mode (i386, amd64, i32) the - generated index will form a continuous range. */ - - int idx = 0; - - for (int i = 0; i < ARRAY_SIZE (x86_linux_all_xstate_features); ++i) - { - if ((xcr0 & x86_linux_all_xstate_features[i].feature) - == x86_linux_all_xstate_features[i].feature) - idx |= (1 << i); - } - - return idx; -} - -#ifdef IN_PROCESS_AGENT - -/* Convert an index number (as returned from x86_linux_xcr0_to_tdesc_idx) - into an xcr0 value which can then be used to create a target - description. */ - -uint64_t -x86_linux_tdesc_idx_to_xcr0 (int idx) -{ - uint64_t xcr0 = 0; - - for (int i = 0; i < ARRAY_SIZE (x86_linux_all_xstate_features); ++i) - { - if ((idx & (1 << i)) != 0) - xcr0 |= x86_linux_all_xstate_features[i].feature; - } - - return xcr0; -} - -#endif /* IN_PROCESS_AGENT */ - -#if defined __i386__ || !defined IN_PROCESS_AGENT -/* A cache of all possible i386 target descriptions. */ +/* See arch/x86-linux-tdesc.h. */ -static struct target_desc *i386_tdescs[x86_linux_i386_tdesc_count_1 ()] = { }; - -/* See nat/x86-linux-tdesc.h. */ - -const struct target_desc * -i386_linux_read_description (uint64_t xcr0) +void +x86_linux_post_init_tdesc (target_desc *tdesc, bool is_64bit) { - xcr0 &= x86_linux_i386_xcr0_feature_mask (); - int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - - gdb_assert (idx >= 0 && idx < x86_linux_i386_tdesc_count_1 ()); - - target_desc **tdesc = &i386_tdescs[idx]; - - if (*tdesc == nullptr) - { - *tdesc = i386_create_target_description (xcr0, true, false); - - init_target_desc (*tdesc, i386_expedite_regs); - } - - return *tdesc; -} -#endif - #ifdef __x86_64__ - -/* A cache of all possible amd64 target descriptions. */ - -static target_desc *amd64_tdescs[x86_linux_amd64_tdesc_count_1 ()] = { }; - -/* A cache of all possible x32 target descriptions. */ - -static target_desc *x32_tdescs[x86_linux_x32_tdesc_count_1 ()] = { }; - -/* See nat/x86-linux-tdesc.h. */ - -const struct target_desc * -amd64_linux_read_description (uint64_t xcr0, bool is_x32) -{ - if (is_x32) - xcr0 &= x86_linux_x32_xcr0_feature_mask (); - else - xcr0 &= x86_linux_amd64_xcr0_feature_mask (); - - int idx = x86_linux_xcr0_to_tdesc_idx (xcr0); - - if (is_x32) - gdb_assert (idx >= 0 && idx < x86_linux_x32_tdesc_count_1 ()); - else - gdb_assert (idx >= 0 && idx < x86_linux_amd64_tdesc_count_1 ()); - - target_desc **tdesc = nullptr; - - if (is_x32) - tdesc = &x32_tdescs[idx]; + if (is_64bit) + init_target_desc (tdesc, amd64_expedite_regs); else - tdesc = &amd64_tdescs[idx]; - - if (*tdesc == nullptr) - { - *tdesc = amd64_create_target_description (xcr0, is_x32, true, true); - - init_target_desc (*tdesc, amd64_expedite_regs); - } - return *tdesc; -} - #endif + init_target_desc (tdesc, i386_expedite_regs); +} -- 2.25.4 ^ permalink raw reply [flat|nested] 116+ messages in thread
end of thread, other threads:[~2024-05-11 10:12 UTC | newest] Thread overview: 116+ messages (download: mbox.gz / follow: Atom feed) -- links below jump to the message on this page -- 2024-02-01 15:28 [PATCH 0/7] x86/Linux Target Description Changes Andrew Burgess 2024-02-01 15:28 ` [PATCH 1/7] gdbserver: convert have_ptrace_getregset to a tribool Andrew Burgess 2024-02-01 15:28 ` [PATCH 2/7] gdb/x86: move reading of cs and ds state into gdb/nat directory Andrew Burgess 2024-02-01 15:28 ` [PATCH 3/7] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess 2024-02-01 15:28 ` [PATCH 4/7] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess 2024-02-01 15:28 ` [PATCH 5/7] gdb/gdbserver: share some code relating to target description creation Andrew Burgess 2024-02-01 15:28 ` [PATCH 6/7] gdbserver: update target description creation for x86/linux Andrew Burgess 2024-02-01 15:28 ` [PATCH 7/7] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess 2024-03-05 17:00 ` [PATCHv2 0/7] x86/Linux Target Description Changes Andrew Burgess 2024-03-05 17:00 ` [PATCHv2 1/7] gdbserver: convert have_ptrace_getregset to a tribool Andrew Burgess 2024-03-05 17:00 ` [PATCHv2 2/7] gdb/x86: move reading of cs and ds state into gdb/nat directory Andrew Burgess 2024-03-05 17:00 ` [PATCHv2 3/7] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess 2024-03-05 17:00 ` [PATCHv2 4/7] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess 2024-03-05 17:00 ` [PATCHv2 5/7] gdb/gdbserver: share some code relating to target description creation Andrew Burgess 2024-03-05 17:00 ` [PATCHv2 6/7] gdbserver: update target description creation for x86/linux Andrew Burgess 2024-03-19 16:01 ` John Baldwin 2024-03-19 18:34 ` Andrew Burgess 2024-03-21 17:28 ` John Baldwin 2024-03-26 10:01 ` Luis Machado 2024-03-26 15:31 ` Tom Tromey 2024-03-05 17:00 ` [PATCHv2 7/7] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess 2024-03-19 16:05 ` [PATCHv2 0/7] x86/Linux Target Description Changes John Baldwin 2024-03-23 16:35 ` [PATCHv3 0/8] " Andrew Burgess 2024-03-23 16:35 ` [PATCHv3 1/8] gdbserver: convert have_ptrace_getregset to a tribool Andrew Burgess 2024-03-23 16:35 ` [PATCHv3 2/8] gdb/x86: move reading of cs and ds state into gdb/nat directory Andrew Burgess 2024-03-23 16:35 ` [PATCHv3 3/8] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess 2024-03-23 16:35 ` [PATCHv3 4/8] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess 2024-03-23 16:35 ` [PATCHv3 5/8] gdb/gdbserver: share some code relating to target description creation Andrew Burgess 2024-03-23 16:35 ` [PATCHv3 6/8] gdb/arch: assert that X86_XSTATE_MPX is not set for x32 Andrew Burgess 2024-03-23 16:35 ` [PATCHv3 7/8] gdbserver: update target description creation for x86/linux Andrew Burgess 2024-03-23 16:35 ` [PATCHv3 8/8] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess 2024-03-26 12:17 ` Andrew Burgess 2024-03-25 17:20 ` [PATCHv3 0/8] x86/Linux Target Description Changes Andrew Burgess 2024-03-25 18:26 ` Simon Marchi 2024-03-26 12:15 ` Andrew Burgess 2024-03-26 13:51 ` H.J. Lu 2024-03-26 14:16 ` H.J. Lu 2024-03-26 16:36 ` Andrew Burgess 2024-03-26 19:03 ` Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 00/10] " Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 01/10] gdbserver/ipa/x86: remove unneeded declarations Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 02/10] gdbserver: convert have_ptrace_getregset to a tribool Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 03/10] gdb/x86: move reading of cs and ds state into gdb/nat directory Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 04/10] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 05/10] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 06/10] gdb/gdbserver: share some code relating to target description creation Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 07/10] gdb/arch: assert that X86_XSTATE_MPX is not set for x32 Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 08/10] gdbserver: update target description creation for x86/linux Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 09/10] gdb: move xcr0 == 0 check into i386_linux_core_read_description Andrew Burgess 2024-04-05 12:33 ` [PATCHv4 10/10] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess 2024-04-09 18:37 ` [PATCHv4 00/10] x86/Linux Target Description Changes John Baldwin 2024-04-25 13:35 ` Willgerodt, Felix 2024-04-25 16:06 ` Andrew Burgess 2024-04-26 15:01 ` [PATCHv5 00/11] " Andrew Burgess 2024-04-26 15:01 ` [PATCHv5 01/11] gdbserver/ipa/x86: remove unneeded declarations Andrew Burgess 2024-04-29 14:34 ` Willgerodt, Felix 2024-05-07 15:05 ` Andrew Burgess 2024-05-08 7:49 ` Willgerodt, Felix 2024-04-26 15:01 ` [PATCHv5 02/11] gdbserver: convert have_ptrace_getregset to a tribool Andrew Burgess 2024-04-29 14:34 ` Willgerodt, Felix 2024-05-07 15:28 ` Andrew Burgess 2024-04-26 15:01 ` [PATCHv5 03/11] gdb/x86: move reading of cs and ds state into gdb/nat directory Andrew Burgess 2024-04-29 14:34 ` Willgerodt, Felix 2024-04-26 15:01 ` [PATCHv5 04/11] gdb/x86: move have_ptrace_getfpxregs global " Andrew Burgess 2024-04-29 14:34 ` Willgerodt, Felix 2024-04-26 15:01 ` [PATCHv5 05/11] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess 2024-04-29 14:34 ` Willgerodt, Felix 2024-05-07 11:55 ` Luis Machado 2024-05-07 15:43 ` Andrew Burgess 2024-05-07 15:56 ` Luis Machado 2024-05-08 7:49 ` Willgerodt, Felix 2024-05-08 13:18 ` Andrew Burgess 2024-04-26 15:01 ` [PATCHv5 06/11] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess 2024-04-29 14:34 ` Willgerodt, Felix 2024-04-26 15:01 ` [PATCHv5 07/11] gdb/gdbserver: share some code relating to target description creation Andrew Burgess 2024-04-29 14:34 ` Willgerodt, Felix 2024-05-07 11:40 ` Andrew Burgess 2024-04-26 15:01 ` [PATCHv5 08/11] gdb/arch: assert that X86_XSTATE_MPX is not set for x32 Andrew Burgess 2024-04-29 14:34 ` Willgerodt, Felix 2024-05-07 16:08 ` Andrew Burgess 2024-04-26 15:01 ` [PATCHv5 09/11] gdbserver: update target description creation for x86/linux Andrew Burgess 2024-04-29 14:35 ` Willgerodt, Felix 2024-05-07 14:24 ` Andrew Burgess 2024-05-08 7:47 ` Willgerodt, Felix 2024-05-08 13:28 ` Andrew Burgess 2024-04-26 15:01 ` [PATCHv5 10/11] gdb: move xcr0 == 0 check into i386_linux_core_read_description Andrew Burgess 2024-04-29 14:35 ` Willgerodt, Felix 2024-04-26 15:01 ` [PATCHv5 11/11] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess 2024-04-29 14:35 ` Willgerodt, Felix 2024-05-07 14:50 ` Andrew Burgess 2024-05-08 7:49 ` Willgerodt, Felix 2024-05-08 16:09 ` Andrew Burgess 2024-05-08 16:46 ` [PATCHv6 0/9] x86/Linux Target Description Changes Andrew Burgess 2024-05-08 16:46 ` [PATCHv6 1/9] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess 2024-05-08 16:46 ` [PATCHv6 2/9] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess 2024-05-08 16:46 ` [PATCHv6 3/9] gdb/x86: move have_ptrace_getfpxregs global into gdb/nat directory Andrew Burgess 2024-05-08 22:52 ` John Baldwin 2024-05-08 16:46 ` [PATCHv6 4/9] gdb/x86: move have_ptrace_getregset " Andrew Burgess 2024-05-08 22:53 ` John Baldwin 2024-05-08 16:46 ` [PATCHv6 5/9] gdb/x86: move reading of cs and ds state " Andrew Burgess 2024-05-08 16:46 ` [PATCHv6 6/9] gdb: move xcr0 == 0 check into i386_linux_core_read_description Andrew Burgess 2024-05-08 22:54 ` John Baldwin 2024-05-08 16:46 ` [PATCHv6 7/9] gdb/gdbserver: share some code relating to target description creation Andrew Burgess 2024-05-08 22:58 ` John Baldwin 2024-05-08 16:46 ` [PATCHv6 8/9] gdbserver: update target description creation for x86/linux Andrew Burgess 2024-05-08 16:46 ` [PATCHv6 9/9] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 0/9] x86/Linux Target Description Changes Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 1/9] gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 2/9] gdbserver/x86: move no-xml code earlier in x86_linux_read_description Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 3/9] gdb/x86: move have_ptrace_getfpxregs global into gdb/nat directory Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 4/9] gdb: move have_ptrace_getregset declaration " Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 5/9] gdb/x86: move reading of cs and ds state " Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 6/9] gdb: move xcr0 == 0 check into i386_linux_core_read_description Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 7/9] gdb/gdbserver: share some code relating to target description creation Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 8/9] gdbserver: update target description creation for x86/linux Andrew Burgess 2024-05-11 10:08 ` [PATCHv7 9/9] gdb/gdbserver: share x86/linux tdesc caching Andrew Burgess
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox