[patch] Fix segfault on a failed watchpoint update

[patch] Fix segfault on a failed watchpoint update

[Jan Kratochvil]

[-- Attachment #1: Type: text/plain, Size: 3222 bytes --]

Hi,

currently GDB HEAD segfaults during one case for the PIE inferiors (failing as
it is the unpatched CVS HEAD), segfault fix attached.


Regards,
Jan


crash reproducibility:

$ runtest --target_board=unix/-fPIE/-pie gdb.base/watchpoint.exp
WARNING: Couldn't find the global config file.
Test Run By jkratoch on Sun Mar  2 09:31:14 2008
Native configuration is x86_64-unknown-linux-gnu

                === gdb tests ===

Schedule of variations:
    unix/-fPIE/-pie

Running target unix/-fPIE/-pie
Using /usr/share/dejagnu/baseboards/unix.exp as board description file for target.
Using /usr/share/dejagnu/config/unix.exp as generic interface file for target.
Using ../.././gdb/testsuite/config/unix.exp as tool-and-target-specific interface file.
Running ../.././gdb/testsuite/gdb.base/watchpoint.exp ...
FAIL: gdb.base/watchpoint.exp: run to marker1 in test_simple_watchpoint
ERROR: Process no longer exists
ERROR: Couldn't send disable 3 to GDB.
ERROR: Delete all breakpoints in delete_breakpoints (timeout)
ERROR: Delete all breakpoints in delete_breakpoints (timeout)
FAIL: gdb.base/watchpoint.exp: watch tests suppressed
FAIL: gdb.base/watchpoint.exp: (timeout) disable fast watches
FAIL: gdb.base/watchpoint.exp: (timeout) show disable fast watches
FAIL: gdb.base/watchpoint.exp: (timeout) set slow conditional watch
FAIL: gdb.base/watchpoint.exp: (timeout) trigger slow conditional watch
FAIL: gdb.base/watchpoint.exp: (timeout) rwatch disallowed when can-use-hw-watchpoints cleared
ERROR: Delete all breakpoints in delete_breakpoints (timeout)

                === gdb Summary ===

# of expected passes            7
# of unexpected failures        7
# of unresolved testcases       5
.../sources/gdb/testsuite/../../gdb/gdb version  6.8.50.20080302-cvs -nx


...
(gdb) PASS: gdb.base/watchpoint.exp: disable watchpoint in test_simple_watchpoint
run
Starting program: /home/jkratoch/redhat/sources/gdb/testsuite/gdb.base/watchpoint
Error in re-setting breakpoint 1: Cannot access memory at address 0x86c
Error in re-setting breakpoint 2: Cannot access memory at address 0x872
Error in re-setting breakpoint 3: Cannot access memory at address 0x20127c
Error in re-setting breakpoint 1: Cannot access memory at address 0x86c
Error in re-setting breakpoint 2: Cannot access memory at address 0x872
Error in re-setting breakpoint 3: Cannot access memory at address 0x20127c
Error in re-setting breakpoint 1: Cannot access memory at address 0x86c
Error in re-setting breakpoint 2: Cannot access memory at address 0x872
Error in re-setting breakpoint 3: Cannot access memory at address 0x20127c

Program exited normally.
(gdb) FAIL: gdb.base/watchpoint.exp: run to marker1 in test_simple_watchpoint
info watch
*** glibc detected *** /home/jkratoch/redhat/sources/gdb/testsuite/../../gdb/gdb: double free or corruption (!prev): 0x00000000023f9d20 ***
======= Backtrace: =========
/lib64/libc.so.6[0x3b98672832]
/lib64/libc.so.6(cfree+0x8c)[0x3b98675f2c]
/home/jkratoch/redhat/sources/gdb/testsuite/../../gdb/gdb(xfree+0x1c)[0x44efb6]
/home/jkratoch/redhat/sources/gdb/testsuite/../../gdb/gdb(free_all_values+0x2a)[0x4cd65a]
/home/jkratoch/redhat/sources/gdb/testsuite/../../gdb/gdb(execute_command+0x26)[0x44b9df]
...

[-- Attachment #2: gdb-value-update_watchpoint-exception.patch --]
[-- Type: text/plain, Size: 1045 bytes --]

2008-03-02  Jan Kratochvil  <jan.kratochvil@redhat.com>

	* breakpoint.c (update_watchpoint): Move B->VAL setting behind the
	VALUE_CONTENTS call.

--- ./gdb/breakpoint.c	27 Feb 2008 20:27:49 -0000	1.304
+++ ./gdb/breakpoint.c	2 Mar 2008 08:27:20 -0000
@@ -913,13 +917,15 @@ update_watchpoint (struct breakpoint *b,
 	 during evaluation, and set watchoints at addresses as needed.
 	 Those values are explicitly deleted here.  */
       v = evaluate_expression (b->exp);
+      /* Fetch the value first as if ti fails and breaks out by the exception
+         there will be no stale V left in B->VAL.  */
+      value_contents (v);
       /* Avoid setting b->val if it's already set.  The meaning of
 	 b->val is 'the last value' user saw, and we should update
 	 it only if we reported that last value to user.  As it
 	 happens, the code that reports it updates b->val directly.  */
       if (b->val == NULL)
 	b->val = v;
-      value_contents (v);
       value_release_to_mark (mark);
 
       /* Look at each value on the value chain.  */

Re: [patch] Fix segfault on a failed watchpoint update

[Daniel Jacobowitz]

On Sun, Mar 02, 2008 at 09:40:50AM +0100, Jan Kratochvil wrote:
> Hi,
> 
> currently GDB HEAD segfaults during one case for the PIE inferiors (failing as
> it is the unpatched CVS HEAD), segfault fix attached.

Thanks, I can see what's wrong.  I'm about to commit another patch
which should fix this (just waiting for doc review, so probably
Monday).  Could you let me know if this fixes it?

http://sourceware.org/ml/gdb-patches/2008-02/msg00472.html

-- 
Daniel Jacobowitz
CodeSourcery

Re: [patch] Fix segfault on a failed watchpoint update

[Jan Kratochvil]

On Sun, 02 Mar 2008 16:08:49 +0100, Daniel Jacobowitz wrote:
> On Sun, Mar 02, 2008 at 09:40:50AM +0100, Jan Kratochvil wrote:
> > Hi,
> > 
> > currently GDB HEAD segfaults during one case for the PIE inferiors (failing as
> > it is the unpatched CVS HEAD), segfault fix attached.
> 
> Thanks, I can see what's wrong.  I'm about to commit another patch
> which should fix this (just waiting for doc review, so probably
> Monday).  Could you let me know if this fixes it?
> 
> http://sourceware.org/ml/gdb-patches/2008-02/msg00472.html

Yes, this patch also fixes the segfault.


Thanks,
Jan