This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [PATCH] Re: Unwinding CFI gcc practice of assumed `same value' regs
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
2006-12-17 Jan Kratochvil <jan.kratochvil@redhat.com>
* 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 <jan.kratochvil@redhat.com>
* 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 <pthread.h>
+#include <unistd.h>
+#include <assert.h>
+
+
+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"
+ }
+}