* [PATCH, CHKP] Fix cdtor merge for instrumented functions
@ 2015-04-02 15:15 Ilya Enkovich
2015-04-02 20:21 ` Jan Hubicka
0 siblings, 1 reply; 4+ messages in thread
From: Ilya Enkovich @ 2015-04-02 15:15 UTC (permalink / raw)
To: gcc-patches
Hi,
This patch doesn't allow instrumentation thunks calls while merging constructors and destructors. Not isntrumented code is not affeceted. Bootstrapped and tested on x86_64-unknown-linux-gnu. OK for trunk?
Thanks,
Ilya
--
gcc/
2015-04-02 Ilya Enkovich <ilya.enkovich@intel.com>
* ipa.c (ipa_cdtor_merge): Skip instrumentation thunks.
gcc/testsuite/
2015-04-02 Ilya Enkovich <ilya.enkovich@intel.com>
* gcc.dg/lto/chkp-ctor-merge_0.c: New.
diff --git a/gcc/ipa.c b/gcc/ipa.c
index b3752de..84ab542 100644
--- a/gcc/ipa.c
+++ b/gcc/ipa.c
@@ -1159,8 +1159,10 @@ ipa_cdtor_merge (void)
{
struct cgraph_node *node;
FOR_EACH_DEFINED_FUNCTION (node)
- if (DECL_STATIC_CONSTRUCTOR (node->decl)
- || DECL_STATIC_DESTRUCTOR (node->decl))
+ if ((DECL_STATIC_CONSTRUCTOR (node->decl)
+ || DECL_STATIC_DESTRUCTOR (node->decl))
+ && !(node->thunk.thunk_p
+ && node->thunk.add_pointer_bounds_args))
record_cdtor_fn (node);
build_cdtor_fns ();
static_ctors.release ();
diff --git a/gcc/testsuite/gcc.dg/lto/chkp-ctor-merge_0.c b/gcc/testsuite/gcc.dg/lto/chkp-ctor-merge_0.c
new file mode 100644
index 0000000..ac4095b
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/lto/chkp-ctor-merge_0.c
@@ -0,0 +1,23 @@
+/* { dg-lto-do run } */
+/* { dg-require-effective-target mpx } */
+/* { dg-lto-options { { -O2 -flto -fcheck-pointer-bounds -mmpx -nodefaultlibs -lc } } } */
+
+int glob = 1;
+
+void __attribute__((constructor))
+ctor1 ()
+{
+ glob += 1;
+}
+
+
+void __attribute__((constructor))
+ctor2 ()
+{
+ glob -= 2;
+}
+
+int main (int argc, const char **argv)
+{
+ return glob;
+}
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH, CHKP] Fix cdtor merge for instrumented functions
2015-04-02 15:15 [PATCH, CHKP] Fix cdtor merge for instrumented functions Ilya Enkovich
@ 2015-04-02 20:21 ` Jan Hubicka
2015-04-03 15:03 ` Ilya Enkovich
0 siblings, 1 reply; 4+ messages in thread
From: Jan Hubicka @ 2015-04-02 20:21 UTC (permalink / raw)
To: Ilya Enkovich; +Cc: gcc-patches
> Hi,
>
> This patch doesn't allow instrumentation thunks calls while merging constructors and destructors. Not isntrumented code is not affeceted. Bootstrapped and tested on x86_64-unknown-linux-gnu. OK for trunk?
>
> Thanks,
> Ilya
> --
> gcc/
>
> 2015-04-02 Ilya Enkovich <ilya.enkovich@intel.com>
>
> * ipa.c (ipa_cdtor_merge): Skip instrumentation thunks.
So the problem here is that you do have two names for the function, one that
is not instrumented and other that is instrumented? I am bit surprised we get
instrumentation on ctors that should not take or return pointer parameter,
but I see one can trigger that at least by manually adding constructor attribute.
I think what you need is to drop DECL_STATIC_CONSTRUCTOR/DESTRUCTURO flags when
producing the transparent alias.
Honza
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH, CHKP] Fix cdtor merge for instrumented functions
2015-04-02 20:21 ` Jan Hubicka
@ 2015-04-03 15:03 ` Ilya Enkovich
2015-04-03 17:01 ` Jan Hubicka
0 siblings, 1 reply; 4+ messages in thread
From: Ilya Enkovich @ 2015-04-03 15:03 UTC (permalink / raw)
To: Jan Hubicka; +Cc: gcc-patches
On 02 Apr 22:21, Jan Hubicka wrote:
> > Hi,
> >
> > This patch doesn't allow instrumentation thunks calls while merging constructors and destructors. Not isntrumented code is not affeceted. Bootstrapped and tested on x86_64-unknown-linux-gnu. OK for trunk?
> >
> > Thanks,
> > Ilya
> > --
> > gcc/
> >
> > 2015-04-02 Ilya Enkovich <ilya.enkovich@intel.com>
> >
> > * ipa.c (ipa_cdtor_merge): Skip instrumentation thunks.
>
> So the problem here is that you do have two names for the function, one that
> is not instrumented and other that is instrumented? I am bit surprised we get
> instrumentation on ctors that should not take or return pointer parameter,
> but I see one can trigger that at least by manually adding constructor attribute.
>
> I think what you need is to drop DECL_STATIC_CONSTRUCTOR/DESTRUCTURO flags when
> producing the transparent alias.
>
> Honza
Dropping flag is a good option. Here is a corresponding patch. Bootstrapped and tested on x86_64-unknown-linux-gnu. Does it look OK?
Thanks,
Ilya
--
gcc/
2015-04-03 Ilya Enkovich <ilya.enkovich@intel.com>
* ipa-chkp.c (chkp_maybe_create_clone): Reset cdtor
flags for instrumentation thunk.
(chkp_produce_thunks): Likewise.
gcc/testsuite/
2015-04-03 Ilya Enkovich <ilya.enkovich@intel.com>
* gcc.dg/lto/chkp-ctor-merge_0.c: New.
diff --git a/gcc/ipa-chkp.c b/gcc/ipa-chkp.c
index a9933e2..0c16f71 100644
--- a/gcc/ipa-chkp.c
+++ b/gcc/ipa-chkp.c
@@ -550,6 +550,9 @@ chkp_maybe_create_clone (tree fndecl)
clone->thunk.thunk_p = true;
clone->thunk.add_pointer_bounds_args = true;
clone->create_edge (node, NULL, 0, CGRAPH_FREQ_BASE);
+ /* Thunk shouldn't be a cdtor. */
+ DECL_STATIC_CONSTRUCTOR (clone->decl) = 0;
+ DECL_STATIC_DESTRUCTOR (clone->decl) = 0;
}
else
{
@@ -713,6 +716,9 @@ chkp_produce_thunks (bool early)
0, CGRAPH_FREQ_BASE);
node->create_reference (node->instrumented_version,
IPA_REF_CHKP, NULL);
+ /* Thunk shouldn't be a cdtor. */
+ DECL_STATIC_CONSTRUCTOR (node->decl) = 0;
+ DECL_STATIC_DESTRUCTOR (node->decl) = 0;
}
}
diff --git a/gcc/testsuite/gcc.dg/lto/chkp-ctor-merge_0.c b/gcc/testsuite/gcc.dg/lto/chkp-ctor-merge_0.c
new file mode 100644
index 0000000..ac4095b
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/lto/chkp-ctor-merge_0.c
@@ -0,0 +1,23 @@
+/* { dg-lto-do run } */
+/* { dg-require-effective-target mpx } */
+/* { dg-lto-options { { -O2 -flto -fcheck-pointer-bounds -mmpx -nodefaultlibs -lc } } } */
+
+int glob = 1;
+
+void __attribute__((constructor))
+ctor1 ()
+{
+ glob += 1;
+}
+
+
+void __attribute__((constructor))
+ctor2 ()
+{
+ glob -= 2;
+}
+
+int main (int argc, const char **argv)
+{
+ return glob;
+}
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH, CHKP] Fix cdtor merge for instrumented functions
2015-04-03 15:03 ` Ilya Enkovich
@ 2015-04-03 17:01 ` Jan Hubicka
0 siblings, 0 replies; 4+ messages in thread
From: Jan Hubicka @ 2015-04-03 17:01 UTC (permalink / raw)
To: Ilya Enkovich; +Cc: Jan Hubicka, gcc-patches
> On 02 Apr 22:21, Jan Hubicka wrote:
> > > Hi,
> > >
> > > This patch doesn't allow instrumentation thunks calls while merging constructors and destructors. Not isntrumented code is not affeceted. Bootstrapped and tested on x86_64-unknown-linux-gnu. OK for trunk?
> > >
> > > Thanks,
> > > Ilya
> > > --
> > > gcc/
> > >
> > > 2015-04-02 Ilya Enkovich <ilya.enkovich@intel.com>
> > >
> > > * ipa.c (ipa_cdtor_merge): Skip instrumentation thunks.
> >
> > So the problem here is that you do have two names for the function, one that
> > is not instrumented and other that is instrumented? I am bit surprised we get
> > instrumentation on ctors that should not take or return pointer parameter,
> > but I see one can trigger that at least by manually adding constructor attribute.
> >
> > I think what you need is to drop DECL_STATIC_CONSTRUCTOR/DESTRUCTURO flags when
> > producing the transparent alias.
> >
> > Honza
>
> Dropping flag is a good option. Here is a corresponding patch. Bootstrapped and tested on x86_64-unknown-linux-gnu. Does it look OK?
>
> Thanks,
> Ilya
> --
> gcc/
>
> 2015-04-03 Ilya Enkovich <ilya.enkovich@intel.com>
>
> * ipa-chkp.c (chkp_maybe_create_clone): Reset cdtor
> flags for instrumentation thunk.
> (chkp_produce_thunks): Likewise.
OK,
Thanks!
Honza
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2015-04-03 17:01 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-04-02 15:15 [PATCH, CHKP] Fix cdtor merge for instrumented functions Ilya Enkovich
2015-04-02 20:21 ` Jan Hubicka
2015-04-03 15:03 ` Ilya Enkovich
2015-04-03 17:01 ` Jan Hubicka
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).