https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #27 from Jakub Jelinek ---
Author: jakub
Date: Thu Mar 19 07:55:22 2015
New Revision: 221509
URL: https://gcc.gnu.org/viewcvs?rev=221509&root=gcc&view=rev
Log:
PR sanitizer/64265
* g++.dg/tsan/pr64265.C: New test.
Added:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #26 from Jakub Jelinek ---
Author: jakub
Date: Mon Jan 5 21:47:51 2015
New Revision: 219202
URL: https://gcc.gnu.org/viewcvs?rev=219202&root=gcc&view=rev
Log:
PR sanitizer/64265
* gimplify.c (gimplify_function_tree): Add TSA
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
Jakub Jelinek changed:
What|Removed |Added
Status|NEW |RESOLVED
Resolution|---
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #24 from Jakub Jelinek ---
Author: jakub
Date: Mon Dec 15 09:50:11 2014
New Revision: 218736
URL: https://gcc.gnu.org/viewcvs?rev=218736&root=gcc&view=rev
Log:
PR sanitizer/64265
* tsan.c (instrument_func_entry): Insert __tsa
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #23 from Jakub Jelinek ---
Author: jakub
Date: Mon Dec 15 09:46:21 2014
New Revision: 218735
URL: https://gcc.gnu.org/viewcvs?rev=218735&root=gcc&view=rev
Log:
PR sanitizer/64265
* tsan.c (instrument_func_entry): Insert __tsa
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #22 from Jakub Jelinek ---
Author: jakub
Date: Mon Dec 15 09:37:47 2014
New Revision: 218734
URL: https://gcc.gnu.org/viewcvs?rev=218734&root=gcc&view=rev
Log:
PR sanitizer/64265
* tsan.c (instrument_func_entry): Insert __tsa
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #21 from Jakub Jelinek ---
FYI, the #c12 patch needs more work, in particular the inliner probably has to
drop the TSAN_FUNC_EXIT () internal calls, otherwise after inlining there can
be multiple of them which is undesirable, as tsan
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #20 from Dmitry Vyukov ---
No, TSAN_GO is not defined for C/C++ tsan. It's only for race detector for Go
language.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #19 from Bernd Edlinger ---
oh, I see now, in tsan/tsan_rtl.cc
// Shadow stack maintenance can be replaced with
// stack unwinding during trace switch (which presumably must be faster).
DCHECK_GE(thr->shadow_stack_pos, thr->sha
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #18 from Bernd Edlinger ---
(In reply to Jakub Jelinek from comment #15)
> I've been running the tests for quite a while and RSS didn't increase in top
> at all.
>
> As for "and no calls to other functions", sure, I haven't changed a
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #17 from Dmitry Vyukov ---
Great.
Jakub, then you can go for gcc support whenever you have time. It's not super
priority as we managed to live without exceptions support so far.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #16 from Kostya Serebryany ---
> Kostya, can you say anything about llvm? On the tsan issue you said:
> "We'll need a kind of RAII for tsan entry/exit hooks. When we are adding
> tsan instrumentation, we need to create a fake class ob
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #15 from Jakub Jelinek ---
I've been running the tests for quite a while and RSS didn't increase in top at
all.
As for "and no calls to other functions", sure, I haven't changed anything on
that logic.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #14 from Bernd Edlinger ---
(In reply to Jakub Jelinek from comment #7)
> Note, I don't see any kind of memory leak on any of the testcases.
> Sure, calling __tsan_func_entry many times is of course wrong.
> As for #c5, clang doesn't
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #13 from Dmitry Vyukov ---
> ... we actually don't want any __tsan_func_{entry,exit} calls if there are no
> memory accesses in the function...
... and no calls to other functions, because these functions can contain memory
accesses
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #12 from Jakub Jelinek ---
Created attachment 34271
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=34271&action=edit
gcc5-pr64265-2.patch
Incremental patch to handle the exceptions, completely untested (don't have
spare cycles
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #11 from Dmitry Vyukov ---
>Doing it in gimplify_function_tree is pretty straightforward
That's good!
>So, the question is just if you want to do it that way...
Kostya, can you say anything about llvm? On the tsan issue you said:
"
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #10 from Jakub Jelinek ---
Created attachment 34270
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=34270&action=edit
gcc5-pr64265.patch
Untested patch to fix just the func entry issue.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #9 from Jakub Jelinek ---
Doing it in gimplify_function_tree is pretty straightforward, after all, we
already have there code to handle
if (flag_instrument_function_entry_exit
&& !DECL_NO_INSTRUMENT_FUNCTION_ENTRY_EXIT (fndecl
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
Dmitry Vyukov changed:
What|Removed |Added
CC||dvyukov at google dot com
--- Comment #8
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #7 from Jakub Jelinek ---
Note, I don't see any kind of memory leak on any of the testcases.
Sure, calling __tsan_func_entry many times is of course wrong.
As for #c5, clang doesn't call __tsan_func_exit in that case either. Dmitry?
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #6 from Jakub Jelinek ---
Seems there are more such spots that insert stmts at gsi_after_labels of
single_succ of entry block - e.g. ipa-split.c, omp-low.c, tree-inline.c,
tree-into-ssa.c, tree-profile.c, tree-ssa-reassoc.c at least.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
Bernd Edlinger changed:
What|Removed |Added
CC||edlinger at gcc dot gnu.org
--- Comment
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #4 from Bernd Edlinger ---
and now I see that his example is mis-compiled too:
cat test1.cpp
int test1(int x)
{
abc:
x=x+1;
__builtin_printf("Test %d\n", x);
if (x<9)
goto abc;
return 0;
}
is transformed to this in test
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
--- Comment #3 from Richard Biener ---
static void
instrument_func_entry (void)
{
basic_block succ_bb;
gimple_stmt_iterator gsi;
tree ret_addr, builtin_decl;
gimple g;
succ_bb = single_succ (ENTRY_BLOCK_PTR_FOR_FN (cfun));
gsi = gsi_
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64265
Richard Biener changed:
What|Removed |Added
Status|UNCONFIRMED |NEW
Last reconfirmed|
26 matches
Mail list logo
