From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 22693 invoked by alias); 17 Dec 2006 18:23:23 -0000 Received: (qmail 22684 invoked by uid 22791); 17 Dec 2006 18:23:22 -0000 X-Spam-Check-By: sourceware.org Received: from 195.22.55.53.adsl.nextra.cz (HELO host0.dyn.jankratochvil.net) (195.22.55.53) by sourceware.org (qpsmtpd/0.31) with ESMTP; Sun, 17 Dec 2006 18:23:12 +0000 Received: from host0.dyn.jankratochvil.net (localhost [127.0.0.1]) by host0.dyn.jankratochvil.net (8.13.8/8.13.8) with ESMTP id kBHIN8VT002780; Sun, 17 Dec 2006 19:23:09 +0100 Received: (from jkratoch@localhost) by host0.dyn.jankratochvil.net (8.13.8/8.13.8/Submit) id kBHIN8vM002779; Sun, 17 Dec 2006 19:23:08 +0100 Date: Sun, 17 Dec 2006 18:23:00 -0000 From: Jan Kratochvil To: Mark Kettenis Cc: gdb-patches@sourceware.org Subject: Re: [PATCH] Re: Unwinding CFI gcc practice of assumed `same value' regs Message-ID: <20061217182307.GA739@host0.dyn.jankratochvil.net> References: <20061211224022.AD76E1800E7@magilla.sf.frob.com> <20061212155233.GH29911@devserv.devel.redhat.com> <20061213204603.GA11741@host0.dyn.jankratochvil.net> <20061213210357.GA27039@nevyn.them.org> <23342.82.92.89.47.1166046052.squirrel@webmail.xs4all.nl> <20061213214540.GA28965@nevyn.them.org> <5950.82.92.89.47.1166046772.squirrel@webmail.xs4all.nl> <20061213221111.GA26406@host0.dyn.jankratochvil.net> <20061216234551.GA6766@host0.dyn.jankratochvil.net> <5614.82.92.89.47.1166360768.squirrel@webmail.xs4all.nl> Mime-Version: 1.0 Content-Type: multipart/mixed; boundary="ew6BAiZeqk4r7MaW" Content-Disposition: inline In-Reply-To: <5614.82.92.89.47.1166360768.squirrel@webmail.xs4all.nl> User-Agent: Mutt/1.4.2.2i X-IsSubscribed: yes Mailing-List: contact gdb-patches-help@sourceware.org; run by ezmlm Precedence: bulk List-Id: List-Subscribe: List-Archive: List-Post: List-Help: , Sender: gdb-patches-owner@sourceware.org X-SW-Source: 2006-12/txt/msg00223.txt.bz2 --ew6BAiZeqk4r7MaW Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Content-length: 2953 Hi Mark, attached updated patch; no code changes were made, only wording got updated. On Sun, 17 Dec 2006 14:06:08 +0100, Mark Kettenis wrote: > > On Wed, 13 Dec 2006 23:11:11 +0100, Jan Kratochvil wrote: > > > > On the other hand as a legacy compatibility I wrote the attached patch to > > fix the GDB excessive non-CFI amd64 `clone' unwind as 0x0. Other > > platforms(and functions?) could be carbon-copied from this one. > > Do people indeed think the legacy support is worth the extra complication? The patch was provided AS IS. It is a mandatory and useful part of the customer support. It may not be worth the maintenance for the upstream development but this decision does not belong to me. > > I consider it the same approach as the way GDB currently on amd64 detects > > signal frames by checking the instructions - the CFI for signal frames > > has been also recently checked to the development glibc as was the > > `clone' CFI. > > Hmm, now if the CFI on 'clone' is incorrect and doesn't mark the return > address as undefined, things will still fail. You're sure there are no > official glibc releases with bogus CFI for 'clone'? This question is very intentionally out of the scope of this patch. This patch only fixes clearly broken unwind of ``clone' without any CFI'. Another patch may address a broken `clone' CFI but I do not have a bugreport/reproducibility for such case. > Anyway, most of it is irrelevant for ChangeLog. New functions, defines, etc. > should just be listed as New function, new define, etc. Thanks [6.8.1 Change Log Concepts]; it is really not easy to perceive all the GNU Coding Standards details. > > +static const unsigned char linux_clone_code[] = > > +{ > > +/* libc/sysdeps/unix/sysv/linux/x86_64/clone.S */ > > +/* #ifdef RESET_PID */ > > +/* ... */ > > +/* movl $SYS_ify(getpid), %eax */ > > + 0x48, 0xc7, 0xc0, 0x27, 0x00, 0x00, 0x00, > > +/* syscall */ > > + 0x0f, 0x05, > > +/* movl %eax, %fs:PID */ > > + 0x64, 0x89, 0x04, 0x25, 0x94, 0x00, 0x00, 0x00, > > +/* movl %eax, %fs:TID */ > > + 0x64, 0x89, 0x04, 0x25, 0x90, 0x00, 0x00, 0x00, > > +/* #endif */ > > +/* |* Set up arguments for the function call. *| */ > > +/* popq %rax |* Function to call. *| */ > > + 0x58, > > +/* popq %rdi |* Argument. *| */ > > + 0x5f, > > +/* call *%rax$ */ > > + 0xff, 0xd0 > > +}; > > The #ifdefs here make me a bit nervous Regarding GDB compilation or GLIBC compatibility? I tried to verbatim copy the original source code. It suggests the possible point of a failure for systems with glibc compiled `!RESET_PID'. > Sorry, again I don't understand your explanation here. Do you have a > collegue who is a native english speaker who can help you with writing > english? My code comments always have been a second grade text as nobody was reading it before. I understand it should change if submitting the code for review. Regards, Jan --ew6BAiZeqk4r7MaW Content-Type: text/plain; charset=us-ascii Content-Disposition: inline; filename="gdb-6.5-clone-is-outermost2.patch" Content-length: 8504 2006-12-17 Jan Kratochvil * gdb/amd64-linux-tdep.c (linux_clone_code): New variable. (LINUX_CLONE_LEN): New definition. (amd64_linux_clone_running, amd64_linux_outermost_frame): New function. (amd64_linux_init_abi): Initialize `outermost_frame_p'. * gdb/i386-tdep.c (i386_gdbarch_init): Likewise. * gdb/i386-tdep.h (gdbarch_tdep): Add `outermost_frame_p' member. * gdb/amd64-tdep.c (amd64_frame_this_id): Call `outermost_frame_p'. 2006-12-17 Jan Kratochvil * gdb.threads/bt-clone-stop.exp, gdb.threads/bt-clone-stop.c: New file. --- ./gdb/amd64-linux-tdep.c 19 Aug 2006 15:15:18 -0000 1.12 +++ ./gdb/amd64-linux-tdep.c 17 Dec 2006 17:59:05 -0000 @@ -235,6 +235,75 @@ amd64_linux_register_reggroup_p (struct /* Set the program counter for process PTID to PC. */ +/* Detect the outermost frame; during unwind of + #5 0x000000305cec68c3 in clone () from /lib64/tls/libc.so.6 + avoid the additional bogus frame + #6 0x0000000000000000 in ?? + We compare if the `linux_clone_code' block is _before_ unwound PC. */ + +static const unsigned char linux_clone_code[] = +{ +/* libc/sysdeps/unix/sysv/linux/x86_64/clone.S */ +/* #ifdef RESET_PID */ +/* ... */ +/* movl $SYS_ify(getpid), %eax */ + 0x48, 0xc7, 0xc0, 0x27, 0x00, 0x00, 0x00, +/* syscall */ + 0x0f, 0x05, +/* movl %eax, %fs:PID */ + 0x64, 0x89, 0x04, 0x25, 0x94, 0x00, 0x00, 0x00, +/* movl %eax, %fs:TID */ + 0x64, 0x89, 0x04, 0x25, 0x90, 0x00, 0x00, 0x00, +/* #endif */ +/* |* Set up arguments for the function call. *| */ +/* popq %rax |* Function to call. *| */ + 0x58, +/* popq %rdi |* Argument. *| */ + 0x5f, +/* call *%rax$ */ + 0xff, 0xd0 +}; + +#define LINUX_CLONE_LEN (sizeof linux_clone_code) + +static int +amd64_linux_clone_running (struct frame_info *next_frame) +{ + CORE_ADDR pc = frame_pc_unwind (next_frame); + unsigned char buf[LINUX_CLONE_LEN]; + + if (!safe_frame_unwind_memory (next_frame, pc - LINUX_CLONE_LEN, buf, + LINUX_CLONE_LEN)) + return 0; + + if (memcmp (buf, linux_clone_code, LINUX_CLONE_LEN) != 0) + return 0; + + return 1; +} + +static int +amd64_linux_outermost_frame (struct frame_info *next_frame) +{ + CORE_ADDR pc = frame_pc_unwind (next_frame); + char *name; + + find_pc_partial_function (pc, &name, NULL, NULL); + + /* If we have NAME, we can optimize the search. + `clone' NAME still needs to have the code checked as its name may be + present in the user code. + `__clone' NAME should not be present in the user code but in the initial + parts of the `__clone' implementation the unwind still makes sense. + More detailed unwinding decision would be too much sensitive to possible + subtle changes in specific glibc revisions. */ + if (name == NULL || strcmp (name, "clone") == 0 + || strcmp ("__clone", name) == 0) + return (amd64_linux_clone_running (next_frame) != 0); + + return 0; +} + static void amd64_linux_write_pc (CORE_ADDR pc, ptid_t ptid) { @@ -273,6 +342,8 @@ amd64_linux_init_abi (struct gdbarch_inf tdep->sc_reg_offset = amd64_linux_sc_reg_offset; tdep->sc_num_regs = ARRAY_SIZE (amd64_linux_sc_reg_offset); + tdep->outermost_frame_p = amd64_linux_outermost_frame; + /* GNU/Linux uses SVR4-style shared libraries. */ set_solib_svr4_fetch_link_map_offsets (gdbarch, svr4_lp64_fetch_link_map_offsets); --- ./gdb/amd64-tdep.c 19 Aug 2006 15:15:18 -0000 1.32 +++ ./gdb/amd64-tdep.c 17 Dec 2006 17:59:05 -0000 @@ -879,11 +879,16 @@ amd64_frame_this_id (struct frame_info * { struct amd64_frame_cache *cache = amd64_frame_cache (next_frame, this_cache); + struct gdbarch_tdep *tdep = gdbarch_tdep (current_gdbarch); /* This marks the outermost frame. */ if (cache->base == 0) return; + /* Detect OS dependent outermost frames; such as `clone'. */ + if (tdep->outermost_frame_p && tdep->outermost_frame_p (next_frame)) + return; + (*this_id) = frame_id_build (cache->base + 16, cache->pc); } --- ./gdb/i386-tdep.c 8 Aug 2006 21:36:46 -0000 1.225 +++ ./gdb/i386-tdep.c 17 Dec 2006 17:59:10 -0000 @@ -2313,6 +2313,9 @@ i386_gdbarch_init (struct gdbarch_info i tdep->sc_pc_offset = -1; tdep->sc_sp_offset = -1; + /* Unwinding stops on i386 automatically. */ + tdep->outermost_frame_p = NULL; + /* The format used for `long double' on almost all i386 targets is the i387 extended floating-point format. In fact, of all targets in the GCC 2.95 tree, only OSF/1 does it different, and insists --- ./gdb/i386-tdep.h 21 Jan 2006 20:59:50 -0000 1.46 +++ ./gdb/i386-tdep.h 17 Dec 2006 17:59:10 -0000 @@ -104,6 +104,9 @@ struct gdbarch_tdep is deprecated, please use `sc_reg_offset' instead. */ int sc_pc_offset; int sc_sp_offset; + + /* Detect OS dependent outermost frames; such as `clone'. */ + int (*outermost_frame_p) (struct frame_info *next_frame); }; /* Floating-point registers. */ --- /dev/null 1 Jan 1970 00:00:00 -0000 +++ ./gdb/testsuite/gdb.threads/bt-clone-stop.c 17 Dec 2006 17:59:19 -0000 @@ -0,0 +1,39 @@ +/* This testcase is part of GDB, the GNU debugger. + + Copyright 2006 Free Software Foundation, Inc. + + 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 2 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, write to the Free Software + Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, + MA 02110-1301, USA. */ + + +#include +#include +#include + + +void *threader (void *arg) +{ + assert (0); + return NULL; +} + +int main (void) +{ + pthread_t t1; + + pthread_create (&t1, NULL, threader, (void *) NULL); + for (;;) + pause(); +} --- /dev/null 1 Jan 1970 00:00:00 -0000 +++ ./gdb/testsuite/gdb.threads/bt-clone-stop.exp 17 Dec 2006 17:59:19 -0000 @@ -0,0 +1,61 @@ +# Copyright 2006 Free Software Foundation, Inc. + +# 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 2 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, write to the Free Software +# Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. + +# Backtraced `clone' must not have `PC == 0' as its previous frame. + +if $tracelevel then { + strace $tracelevel +} + +set testfile bt-clone-stop +set srcfile ${testfile}.c +set binfile ${objdir}/${subdir}/${testfile} +if { [gdb_compile_pthreads "${srcdir}/${subdir}/${srcfile}" "${binfile}" executable {debug}] != "" } { + untested "Couldn't compile test program" + return -1 +} + +# Get things started. + +gdb_exit +gdb_start +gdb_reinitialize_dir $srcdir/$subdir +gdb_load ${binfile} + +# threader: threader.c:8: threader: Assertion `0' failed. +# Program received signal SIGABRT, Aborted. + +gdb_test "run" \ + "Program received signal SIGABRT.*" \ + "run" + +# Former gdb unwind (the first function is `clone'): +# #5 0x0000003421ecd62d in ?? () from /lib64/libc.so.6 +# #6 0x0000000000000000 in ?? () +# (gdb) +# Tested `amd64_linux_outermost_frame' functionality should omit the line `#6'. +# +# Two `-re' cases below must be in this order (1st is a subset of the 2nd one). +# Unhandled case below should not happen and it is fortunately handled by +# `amd64_linux_outermost_frame' as FAIL (and result `0x0 entry output invalid'). +gdb_test_multiple "bt" "0x0 entry output invalid" { + -re "in threader \\(.*\n#\[0-9\]* *0x0* in .*$gdb_prompt $" { + fail "0x0 entry found" + } + -re "in threader \\(.*$gdb_prompt $" { + pass "0x0 entry not found" + } +} --ew6BAiZeqk4r7MaW--