From: Yao Qi <qiyaoltc@gmail.com>
To: Pedro Alves <palves@redhat.com>
Cc: Yao Qi <qiyaoltc@gmail.com>, gdb-patches@sourceware.org
Subject: Re: [PATCH] aarch64 multi-arch part 6: HW breakpoint on unaligned address
Date: Thu, 15 Oct 2015 08:14:00 -0000 [thread overview]
Message-ID: <86fv1c4kg7.fsf@gmail.com> (raw)
In-Reply-To: <561D4008.90009@redhat.com> (Pedro Alves's message of "Tue, 13 Oct 2015 18:31:52 +0100")
Pedro Alves <palves@redhat.com> writes:
>>>> aarch64_point_is_aligned (int is_watchpoint, CORE_ADDR addr, int len)
>>>> {
>>>> - unsigned int alignment = is_watchpoint ? AARCH64_HWP_ALIGNMENT
>>>> - : AARCH64_HBP_ALIGNMENT;
>>>> + unsigned int alignment = 0;
>>>> +
>>>> + if (is_watchpoint)
>>>> + alignment = AARCH64_HWP_ALIGNMENT;
>>>> + else
>>>> + {
>>>> + /* Set alignment to 2 only if the current process is 32-bit,
>>>> + since thumb instruction can be 2-byte aligned. Otherwise, set
>>>> + alignment to AARCH64_HBP_ALIGNMENT. */
>>>> + alignment = 2;
>>>
>>> Is some other code doing what the comment says? I'm not seeing
>>> any obvious 32-bit check.
>>
>> No, I don't do the 32-bit check here. Ideally, we should set alignment
>> to 2 only when the process is 32-bit, and still use 4 as alignment
>> otherwise. However, I don't find an easy way to do the 32-bit check
>> here, because this code is used by both GDB and GDBserver. We can do
>> the 32-bit check in GDB and GDBserver respectively, and pass the result
>> to nat/aarch64-linux-hw-point.c, but I don't like putting information down
>> multiple levels like this.
>
> At least the comment should be updated. It's quite misleading as is.
In order to do 32-bit check in nat/aarch64-linux-hw-point.c, I add a new
regcache interface regcache_register_size which is defined in both GDB
and GDBserver. It has two arguments, regcache and number, which looks
more reasonable than register_size, IMO. With regcache_register_size in
place, we can check 32-bit like this,
struct regcache *regcache
= get_thread_regcache_for_ptid (current_lwp_ptid ());
/* Set alignment to 2 only if the current process is 32-bit,
since thumb instruction can be 2-byte aligned. Otherwise, set
alignment to AARCH64_HBP_ALIGNMENT. */
if (regcache_register_size (regcache, 0) == 8)
alignment = AARCH64_HBP_ALIGNMENT;
else
alignment = 2;
on the other hand, a lot of register_size calls in GDB and GDBserver can
be replaced by regcache_register_size. This can be done separately.
Here is the patch V2, regression tested on aarch64-linux.
--
Yao (齐尧)
From 65e2b12411ee8431c24bdeb17bc3eee65f32c1b0 Mon Sep 17 00:00:00 2001
From: Yao Qi <yao.qi@linaro.org>
Date: Thu, 9 Jul 2015 09:28:27 +0100
Subject: [PATCH] aarch64 multi-arch part 6: HW breakpoint on unaligned address
Nowadays, both aarch64 GDB and linux kernel assumes that address for
setting breakpoint should be 4-byte aligned. However that is not true
after we support multi-arch, because thumb instruction can be at 2-byte
aligned address. Patch http://lists.infradead.org/pipermail/linux-arm-kernel/2015-October/375141.html
to linux kernel is to teach kernel to handle 2-byte aligned address for
HW breakpoint, while this patch is to teach aarch64 GDB handle 2-byte
aligned address.
First of all, we call gdbarch_breakpoint_from_pc to get the instruction
length rather than using hard-coded 4. Secondly, in GDBserver, we set
length back to 2 if it is 3, because GDB encode 3 in it to indicate it
is a 32-bit thumb breakpoint. Then we relax the address alignment
check from 4-byte aligned to 2-byte aligned.
This patch enables some tests (such as gdb.base/break-idempotent.exp,
gdb.base/cond-eval-mode.exp, gdb.base/watchpoint-reuse-slot.exp,) and
fixes many fails (such as gdb.base/hbreak2.exp) when the program is
compiled in thumb mode on aarch64.
Regression tested on aarch64-linux, both native and gdbserver. This
is the last patch of multi-arch work.
gdb:
2015-10-14 Yao Qi <yao.qi@linaro.org>
* aarch64-linux-nat.c (aarch64_linux_insert_hw_breakpoint):
Call gdbarch_breakpoint_from_pc to instruction length.
(aarch64_linux_remove_hw_breakpoint): Likewise.
* common/common-regcache.h (regcache_register_size): Declare.
* nat/aarch64-linux-hw-point.c: Include "common-regcache.h".
(aarch64_point_is_aligned): Set alignment to 2 for breakpoint if
the process is 32bit, otherwise set alignment to 4.
(aarch64_handle_breakpoint): Update comments.
* regcache.c (regcache_register_size): New function.
gdb/gdbserver:
2015-10-14 Yao Qi <yao.qi@linaro.org>
* linux-aarch64-low.c (aarch64_insert_point): Set len to 2
if it is 3.
(aarch64_remove_point): Likewise.
* regcache.c (regcache_register_size): New function.
diff --git a/gdb/aarch64-linux-nat.c b/gdb/aarch64-linux-nat.c
index c9f439f..4d3d55a 100644
--- a/gdb/aarch64-linux-nat.c
+++ b/gdb/aarch64-linux-nat.c
@@ -608,11 +608,13 @@ aarch64_linux_insert_hw_breakpoint (struct target_ops *self,
{
int ret;
CORE_ADDR addr = bp_tgt->placed_address = bp_tgt->reqstd_address;
- const int len = 4;
+ int len;
const enum target_hw_bp_type type = hw_execute;
struct aarch64_debug_reg_state *state
= aarch64_get_debug_reg_state (ptid_get_pid (inferior_ptid));
+ gdbarch_breakpoint_from_pc (gdbarch, &addr, &len);
+
if (show_debug_regs)
fprintf_unfiltered
(gdb_stdlog,
@@ -640,11 +642,13 @@ aarch64_linux_remove_hw_breakpoint (struct target_ops *self,
{
int ret;
CORE_ADDR addr = bp_tgt->placed_address;
- const int len = 4;
+ int len = 4;
const enum target_hw_bp_type type = hw_execute;
struct aarch64_debug_reg_state *state
= aarch64_get_debug_reg_state (ptid_get_pid (inferior_ptid));
+ gdbarch_breakpoint_from_pc (gdbarch, &addr, &len);
+
if (show_debug_regs)
fprintf_unfiltered
(gdb_stdlog, "remove_hw_breakpoint on entry (addr=0x%08lx, len=%d))\n",
diff --git a/gdb/common/common-regcache.h b/gdb/common/common-regcache.h
index a922316..c470603 100644
--- a/gdb/common/common-regcache.h
+++ b/gdb/common/common-regcache.h
@@ -28,6 +28,11 @@
extern struct regcache *get_thread_regcache_for_ptid (ptid_t ptid);
+/* Return the size of register numbered N in REGCACHE. This function
+ must be provided by the client. */
+
+extern int regcache_register_size (const struct regcache *regcache, int n);
+
/* Read the PC register. This function must be provided by the
client. */
diff --git a/gdb/gdbserver/linux-aarch64-low.c b/gdb/gdbserver/linux-aarch64-low.c
index 9cefdda..780c5e3 100644
--- a/gdb/gdbserver/linux-aarch64-low.c
+++ b/gdb/gdbserver/linux-aarch64-low.c
@@ -315,9 +315,17 @@ aarch64_insert_point (enum raw_bkpt_type type, CORE_ADDR addr,
ret = -1;
}
else
- ret =
- aarch64_handle_breakpoint (targ_type, addr, len, 1 /* is_insert */,
- state);
+ {
+ if (len == 3)
+ {
+ /* LEN is 3 means the breakpoint is set on a 32-bit thumb
+ instruction. Set it to 2 to correctly encode length bit
+ mask in hardware/watchpoint control register. */
+ len = 2;
+ }
+ ret = aarch64_handle_breakpoint (targ_type, addr, len,
+ 1 /* is_insert */, state);
+ }
if (show_debug_regs)
aarch64_show_debug_reg_state (state, "insert_point", addr, len,
@@ -353,9 +361,17 @@ aarch64_remove_point (enum raw_bkpt_type type, CORE_ADDR addr,
aarch64_handle_watchpoint (targ_type, addr, len, 0 /* is_insert */,
state);
else
- ret =
- aarch64_handle_breakpoint (targ_type, addr, len, 0 /* is_insert */,
- state);
+ {
+ if (len == 3)
+ {
+ /* LEN is 3 means the breakpoint is set on a 32-bit thumb
+ instruction. Set it to 2 to correctly encode length bit
+ mask in hardware/watchpoint control register. */
+ len = 2;
+ }
+ ret = aarch64_handle_breakpoint (targ_type, addr, len,
+ 0 /* is_insert */, state);
+ }
if (show_debug_regs)
aarch64_show_debug_reg_state (state, "remove_point", addr, len,
diff --git a/gdb/gdbserver/regcache.c b/gdb/gdbserver/regcache.c
index f79063c..e11b173 100644
--- a/gdb/gdbserver/regcache.c
+++ b/gdb/gdbserver/regcache.c
@@ -316,6 +316,14 @@ register_size (const struct target_desc *tdesc, int n)
return tdesc->reg_defs[n].size / 8;
}
+/* See common/common-regcache.h. */
+
+int
+regcache_register_size (const struct regcache *regcache, int n)
+{
+ return register_size (regcache->tdesc, n);
+}
+
static unsigned char *
register_data (struct regcache *regcache, int n, int fetch)
{
diff --git a/gdb/nat/aarch64-linux-hw-point.c b/gdb/nat/aarch64-linux-hw-point.c
index bca6ec1..1a5fa6a 100644
--- a/gdb/nat/aarch64-linux-hw-point.c
+++ b/gdb/nat/aarch64-linux-hw-point.c
@@ -18,6 +18,7 @@
#include "common-defs.h"
#include "break-common.h"
+#include "common-regcache.h"
#include "nat/linux-nat.h"
#include "aarch64-linux-hw-point.h"
@@ -112,8 +113,23 @@ aarch64_point_encode_ctrl_reg (enum target_hw_bp_type type, int len)
static int
aarch64_point_is_aligned (int is_watchpoint, CORE_ADDR addr, int len)
{
- unsigned int alignment = is_watchpoint ? AARCH64_HWP_ALIGNMENT
- : AARCH64_HBP_ALIGNMENT;
+ unsigned int alignment = 0;
+
+ if (is_watchpoint)
+ alignment = AARCH64_HWP_ALIGNMENT;
+ else
+ {
+ struct regcache *regcache
+ = get_thread_regcache_for_ptid (current_lwp_ptid ());
+
+ /* Set alignment to 2 only if the current process is 32-bit,
+ since thumb instruction can be 2-byte aligned. Otherwise, set
+ alignment to AARCH64_HBP_ALIGNMENT. */
+ if (regcache_register_size (regcache, 0) == 8)
+ alignment = AARCH64_HBP_ALIGNMENT;
+ else
+ alignment = 2;
+ }
if (addr & (alignment - 1))
return 0;
@@ -445,7 +461,7 @@ aarch64_handle_breakpoint (enum target_hw_bp_type type, CORE_ADDR addr,
struct aarch64_debug_reg_state *state)
{
/* The hardware breakpoint on AArch64 should always be 4-byte
- aligned. */
+ aligned, but on AArch32, it can be 2-byte aligned. */
if (!aarch64_point_is_aligned (0 /* is_watchpoint */ , addr, len))
return -1;
diff --git a/gdb/regcache.c b/gdb/regcache.c
index 74d883a..5ee31fb 100644
--- a/gdb/regcache.c
+++ b/gdb/regcache.c
@@ -179,6 +179,14 @@ register_size (struct gdbarch *gdbarch, int regnum)
return size;
}
+/* See common/common-regcache.h. */
+
+int
+regcache_register_size (const struct regcache *regcache, int n)
+{
+ return register_size (get_regcache_arch (regcache), n);
+}
+
/* The register cache for storing raw register values. */
struct regcache
next prev parent reply other threads:[~2015-10-15 8:14 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-10-13 10:11 Yao Qi
2015-10-13 11:07 ` Pedro Alves
2015-10-13 15:26 ` Yao Qi
2015-10-13 16:58 ` Andrew Pinski
2015-10-13 17:31 ` Pedro Alves
2015-10-15 8:14 ` Yao Qi [this message]
2015-10-15 13:02 ` Pedro Alves
2015-10-15 14:10 ` Yao Qi
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=86fv1c4kg7.fsf@gmail.com \
--to=qiyaoltc@gmail.com \
--cc=gdb-patches@sourceware.org \
--cc=palves@redhat.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox