I can't yet reproduce this failure on the command line. Still, it seems to happen once in a while, for -O2 here: https://gcc.gnu.org/ml/gcc-regression/2015-08/msg00147.html . AFAIU, given the dg-shouldfail, the failure actually means that once in a while the executable succeeds and fails to give the expected warning and runtime error. Expected warning (which is combined with return status 66): ... ================== WARNING: ThreadSanitizer: data race (pid=17100) Read of size 4 at 0x000000601544 by thread T1: #0 baz3 src/gcc/testsuite/c-c++-common/tsan/pr65400-1.c:46 (pr65400-1.exe+0x000000400cbb) #1 baz5 src/gcc/testsuite/c-c++-common/tsan/pr65400-1.c:61 (pr65400-1.exe+0x000000400d71) #2 tf src/gcc/testsuite/c-c++-common/tsan/pr65400-1.c:68 (pr65400-1.exe+0x000000400d97) Previous write of size 4 at 0x000000601544 by main thread: #0 main src/gcc/testsuite/c-c++-common/tsan/pr65400-1.c:79 (pr65400-1.exe+0x000000400e12) Location is global 'v' of size 4 at 0x000000601544 (pr65400-1.exe+0x000000601544) Thread T1 (tid=17102, running) created by main thread at: #0 pthread_create src/libsanitizer/tsan/tsan_interceptors.cc:876 (libtsan.so.0+0x00000002af04) #1 main src/gcc/testsuite/c-c++-common/tsan/pr65400-1.c:77 (pr65400-1.exe+0x000000400dea) SUMMARY: ThreadSanitizer: data race src/gcc/testsuite/c-c++-common/tsan/pr65400-1.c:46 in baz3 ================== ThreadSanitizer: reported 1 warnings ...
(In reply to vries from comment #0) > I can't yet reproduce this failure on the command line. That includes 10000 iterations of running the executable, for both -O2 and -O0.
If you change types of: int v; int q; int o; to 'long long', does it fix the failure?
(In reply to Dmitry Vyukov from comment #2) > If you change types of: > > int v; > int q; > int o; > > to 'long long', does it fix the failure? As long as I don't have a reliable way to reproduce the failure, I can't answer that question.
I've observed this failure now in two seperate build & test runs.
Author: law Date: Mon Feb 1 22:03:57 2016 New Revision: 233053 URL: https://gcc.gnu.org/viewcvs?rev=233053&root=gcc&view=rev Log: PR tree-optimization/68580 * params.def (FSM_MAXIMUM_PHI_ARGUMENTS): New param. * tree-ssa-threadbackward.c (fsm_find_control_statement_thread_paths): Do not try to walk through large PHI nodes. Modified: trunk/gcc/ChangeLog trunk/gcc/params.def trunk/gcc/tree-ssa-threadbackward.c
https://gcc.gnu.org/ml/gcc-cvs/2016-02/msg00030.html : Author: law Date: Mon Feb 1 22:05:58 2016 New Revision: 233054 URL: https://gcc.gnu.org/viewcvs?rev=233054&root=gcc&view=rev Log: Fix PR # in last change Modified: trunk/gcc/ChangeLog
Ran into it once more while testing couple of patches on top of r233103.
I managed to trigger the test returning 0 from the command line, by running it in parallel: ... $ ( export LD_LIBRARY_PATH=$(pwd -P)/install/lib64; for cnt in $(seq 1 400); do (./a.out ; echo $? ) > log.$cnt 2>&1 & done ) $ egrep '^0$' log.* | wc -l 54 ... I can't be sure if that is the reason that the test is failing (by returning 0) during running the testsuite, but it's certainly possible (I'm running it with -j12). So, I suspect that the test fails because pthread_create fails with EAGAIN due to insufficient resources: ... int main () { pthread_t th; barrier_init (&barrier, 2); if (pthread_create (&th, NULL, tf, NULL)) return 0; ... Indeed in the testcase we do not distinguish between: - return 0, case pthread_create failed, and - return 0 at end of main, -fsanitize=thread failed to catch the race condition. It would probably be good to print an error message when pthread_create fails, such that when looking at the fail in the testlog, we can easily see that that is the reason it failed, and not expect it to reproduce otherwise.
Created attachment 37675 [details] tentative patch
posted RFC: https://gcc.gnu.org/ml/gcc-patches/2016-02/msg00949.html
Alternative patch proposed by Bernd Edlinger : https://gcc.gnu.org/ml/gcc-patches/2016-02/msg00976.html . Patch approved by sanitizer reviewer : https://gcc.gnu.org/ml/gcc-patches/2016-02/msg00985.html .
Author: edlinger Date: Thu Feb 18 22:27:09 2016 New Revision: 233538 URL: https://gcc.gnu.org/viewcvs?rev=233538&root=gcc&view=rev Log: PR testsuite/68580 * c-c++-common/tsan/pr65400-1.c (v, q, o): Make 8-byte aligned. Modified: trunk/gcc/testsuite/ChangeLog trunk/gcc/testsuite/c-c++-common/tsan/pr65400-1.c
Author: edlinger Date: Sat Feb 20 05:58:00 2016 New Revision: 233581 URL: https://gcc.gnu.org/viewcvs?rev=233581&root=gcc&view=rev Log: 2016-02-20 Bernd Edlinger <bernd.edlinger@hotmail.de> Backport from mainline 2016-02-18 Bernd Edlinger <bernd.edlinger@hotmail.de> PR testsuite/68580 * c-c++-common/tsan/pr65400-1.c (v, q, o): Make 8-byte aligned. Modified: branches/gcc-5-branch/gcc/testsuite/ChangeLog branches/gcc-5-branch/gcc/testsuite/c-c++-common/tsan/pr65400-1.c
Patch with testcase fix committed to trunk and 5 branch. Testcase not present in 4.9 branch. Marking resolved-fixed.