[patch] PR 12028 "GDB crashes on a double free during overload resolution"

[patch] PR 12028 "GDB crashes on a double free during overload resolution"

[sami wagiaalla]

[-- Attachment #1: Type: text/plain, Size: 245 bytes --]

old_cleanups was being set twice making the later call to 
discard_cleanups ignore the first 'make_cleanup' request.

The patch is proposed for both head and the 7.2 branch.

This has been regression tested on x8664 with gcc-4.4.4-10.fc13

Sami

[-- Attachment #2: pr12028.patch --]
[-- Type: text/x-patch, Size: 2582 bytes --]

Fix PR 12028: "GDB crashes on a double free during overload resolution "

2010-09-16  Sami Wagiaalla  <swagiaal@redhat.com>

	PR C++/12028
	* valops.c (find_oload_champ_namespace_loop): removed incorrect
	'old_cleanups' reassignment.

2010-09-16  Sami Wagiaalla  <swagiaal@redhat.com>

	* gdb.cp/pr12028.cc: New.
	* gdb.cp/pr12028.exp: New.

diff --git a/gdb/testsuite/gdb.cp/pr12028.cc b/gdb/testsuite/gdb.cp/pr12028.cc
new file mode 100644
index 0000000..0fcab6b
--- /dev/null
+++ b/gdb/testsuite/gdb.cp/pr12028.cc
@@ -0,0 +1,21 @@
+class A{};
+class B{};
+class C: public B {};
+
+namespace D{
+  int foo (A) { return 11; }
+  int foo (C) { return 12; }
+}
+
+int main()
+{
+  A a;
+  B b;
+  C c;
+
+  D::foo (a);
+  //  D::foo (b);
+  D::foo (c);
+
+  return 0;
+}
diff --git a/gdb/testsuite/gdb.cp/pr12028.exp b/gdb/testsuite/gdb.cp/pr12028.exp
new file mode 100644
index 0000000..746c6b5
--- /dev/null
+++ b/gdb/testsuite/gdb.cp/pr12028.exp
@@ -0,0 +1,29 @@
+# Copyright 2008 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 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/>.
+
+set testfile pr12028
+set srcfile ${testfile}.cc
+if { [prepare_for_testing ${testfile}.exp ${testfile} ${srcfile} {debug c++}] } {
+    return -1
+}
+
+############################################
+
+if ![runto_main] then {
+    perror "couldn't run to breakpoint main"
+    continue
+}
+
+gdb_test "p D::foo(b)" "Cannot resolve function foo to any overloaded instance"
diff --git a/gdb/valops.c b/gdb/valops.c
index 7fbad10..4e83a04 100644
--- a/gdb/valops.c
+++ b/gdb/valops.c
@@ -2715,7 +2715,7 @@ find_oload_champ_namespace_loop (struct type **arg_types, int nargs,
      function symbol to start off with.)  */
 
   old_cleanups = make_cleanup (xfree, *oload_syms);
-  old_cleanups = make_cleanup (xfree, *oload_champ_bv);
+  make_cleanup (xfree, *oload_champ_bv);
   new_namespace = alloca (namespace_len + 1);
   strncpy (new_namespace, qualified_name, namespace_len);
   new_namespace[namespace_len] = '\0';

Re: [patch] PR 12028 "GDB crashes on a double free during overload resolution"

[Doug Evans]

On Fri, Sep 17, 2010 at 7:41 AM, sami wagiaalla <swagiaal@redhat.com> wrote:
> old_cleanups was being set twice making the later call to discard_cleanups
> ignore the first 'make_cleanup' request.
>
> The patch is proposed for both head and the 7.2 branch.
>
> This has been regression tested on x8664 with gcc-4.4.4-10.fc13
>
> Sami
>

Hi.  This patch is ok.

Looking at the function though, is this a memory leak?
i.e. you want to do_cleanups here, not discard them (right?).

  else
    {
      *oload_syms = new_oload_syms;
      *oload_champ = new_oload_champ;
      *oload_champ_bv = new_oload_champ_bv;
      discard_cleanups (old_cleanups);
      return 0;
    }

Re: [patch] PR 12028 "GDB crashes on a double free during overload resolution"

[sami wagiaalla]

[-- Attachment #1: Type: text/plain, Size: 888 bytes --]

On 09/18/2010 12:03 PM, Doug Evans wrote:
> On Fri, Sep 17, 2010 at 7:41 AM, sami wagiaalla<swagiaal@redhat.com>  wrote:
>> old_cleanups was being set twice making the later call to discard_cleanups
>> ignore the first 'make_cleanup' request.
>>
>> The patch is proposed for both head and the 7.2 branch.
>>
>> This has been regression tested on x8664 with gcc-4.4.4-10.fc13
>>
>> Sami
>>
>
> Hi.  This patch is ok.
>

Thanks for the review.

> Looking at the function though, is this a memory leak?
> i.e. you want to do_cleanups here, not discard them (right?).
>
>    else
>      {
>        *oload_syms = new_oload_syms;
>        *oload_champ = new_oload_champ;
>        *oload_champ_bv = new_oload_champ_bv;
>        discard_cleanups (old_cleanups);
>        return 0;
>      }

I agree. I have attached a patch which has been regression tested on 
x8664 with gcc-4.4.4-10.fc13

Sami

[-- Attachment #2: oload_memory_leak.patch --]
[-- Type: text/x-patch, Size: 621 bytes --]

Fix find_oload_champ_namespace_loop memory leak.

2010-09-22  Sami Wagiaalla  <swagiaal@redhat.com>

	* valops.c (find_oload_champ_namespace_loop): replace incorrect
	discard_cleanups do_cleanups.

diff --git a/gdb/valops.c b/gdb/valops.c
index bbbf66e..13c83ff 100644
--- a/gdb/valops.c
+++ b/gdb/valops.c
@@ -2760,7 +2760,7 @@ find_oload_champ_namespace_loop (struct type **arg_types, int nargs,
       *oload_syms = new_oload_syms;
       *oload_champ = new_oload_champ;
       *oload_champ_bv = new_oload_champ_bv;
-      discard_cleanups (old_cleanups);
+      do_cleanups (old_cleanups);
       return 0;
     }
 }

Re: [patch] PR 12028 "GDB crashes on a double free during overload resolution"

[Tom Tromey]

>>>>> "Sami" == sami wagiaalla <swagiaal@redhat.com> writes:

Sami> I agree. I have attached a patch which has been regression tested on
Sami> x8664 with gcc-4.4.4-10.fc13

This patch is also ok, assuming that it is still valgrind-clean for the
test in question.

Tom

Re: [patch] PR 12028 "GDB crashes on a double free during overload resolution"

[sami wagiaalla]

On 09/22/2010 03:19 PM, Tom Tromey wrote:
>>>>>> "Sami" == sami wagiaalla<swagiaal@redhat.com>  writes:
>
> Sami>  I agree. I have attached a patch which has been regression tested on
> Sami>  x8664 with gcc-4.4.4-10.fc13
>
> This patch is also ok, assuming that it is still valgrind-clean for the
> test in question.
>

I ran valgrind like this:

valgrind ./gdb/gdb ./gdb/testsuite/gdb.cp/pr12028 -ex 'sta' -ex 'p 
D::foo(b)' -ex 'q'

There are no glaring errors. The original double free error is still 
gone of course and there are no others.

Incidentally, there is this error:

==8467== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 27 from 9)
==8467== Use of uninitialised value of size 8
==8467==    at 0x57C966: bcache_full (bcache.c:189)
==8467==    by 0x52E94F: add_psymbol_to_list (psymtab.c:1358)
==8467==    by 0x596634: add_partial_symbol (dwarf2read.c:3794)
==8467==    by 0x596A7A: add_partial_subprogram (dwarf2read.c:3882)

Worth figuring out but it is unrelated to this patch series. So I am 
committing the above patch.

Re: [patch] PR 12028 "GDB crashes on a double free during overload resolution"

[Tom Tromey]

>>>>> "Sami" == sami wagiaalla <swagiaal@redhat.com> writes:

Sami> Incidentally, there is this error:
Sami> ==8467== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 27 from 9)
Sami> ==8467== Use of uninitialised value of size 8
Sami> ==8467==    at 0x57C966: bcache_full (bcache.c:189)
Sami> ==8467==    by 0x52E94F: add_psymbol_to_list (psymtab.c:1358)
Sami> ==8467==    by 0x596634: add_partial_symbol (dwarf2read.c:3794)
Sami> ==8467==    by 0x596A7A: add_partial_subprogram (dwarf2read.c:3882)

Sami> Worth figuring out but it is unrelated to this patch series.

This is fallout from your earlier bcache changes.

I'm regtesting the appended.  I already ran it through valgrind, it
definitely fixes the above report.  I will check it in once the regtest
reports no regressions.

Tom

2010-09-27  Tom Tromey  <tromey@redhat.com>

	* bcache.c (expand_hash_table): Use hash_function, not hash.

Index: bcache.c
===================================================================
RCS file: /cvs/src/src/gdb/bcache.c,v
retrieving revision 1.28
diff -u -r1.28 bcache.c
--- bcache.c	31 Aug 2010 17:26:08 -0000	1.28
+++ bcache.c	27 Sep 2010 17:16:41 -0000
@@ -184,7 +184,8 @@
 	  struct bstring **new_bucket;
 	  next = s->next;
 
-	  new_bucket = &new_buckets[(hash (&s->d.data, s->length)
+	  new_bucket = &new_buckets[(bcache->hash_function (&s->d.data,
+							    s->length)
 				     % new_num_buckets)];
 	  s->next = *new_bucket;
 	  *new_bucket = s;

Re: [patch] PR 12028 "GDB crashes on a double free during overload resolution"

[sami wagiaalla]

On 09/27/2010 01:18 PM, Tom Tromey wrote:
>>>>>> "Sami" == sami wagiaalla<swagiaal@redhat.com>  writes:
>
> Sami>  Incidentally, there is this error:
> Sami>  ==8467== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 27 from 9)
> Sami>  ==8467== Use of uninitialised value of size 8
> Sami>  ==8467==    at 0x57C966: bcache_full (bcache.c:189)
> Sami>  ==8467==    by 0x52E94F: add_psymbol_to_list (psymtab.c:1358)
> Sami>  ==8467==    by 0x596634: add_partial_symbol (dwarf2read.c:3794)
> Sami>  ==8467==    by 0x596A7A: add_partial_subprogram (dwarf2read.c:3882)
>
> Sami>  Worth figuring out but it is unrelated to this patch series.
>
> This is fallout from your earlier bcache changes.
>

Totally my bad, I should have guessed that and investigated further. 
Thanks for tracking this down and for the patch.