[Bug testsuite/62060] g++.dg/tsan/cond_race.C triggers heap-use-after-free

[chefmax at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62060

--- Comment #2 from Maxim Ostapenko <chefmax at gcc dot gnu.org> ---
(In reply to vries from comment #1)
> Tentative patch:
> ...
> diff --git a/gcc/testsuite/g++.dg/tsan/cond_race.C
> b/gcc/testsuite/g++.dg/tsan/cond_race.C
> index a937614..90dfb19 100644
> --- a/gcc/testsuite/g++.dg/tsan/cond_race.C
> +++ b/gcc/testsuite/g++.dg/tsan/cond_race.C
> @@ -1,5 +1,5 @@
>  /* { dg-shouldfail "tsan" } */
> -/* { dg-output "ThreadSanitizer: data race.*" } */
> +/* { dg-output "ThreadSanitizer: (data race|heap-use-after-free).*" } */
>  /* { dg-output "pthread_cond_signal.*" } */
>  
>  #include <stdio.h>
> ...

This test was copied from LLVM compiler-rt testsuite. I see that compiler-rt
developers added sleep (1) right after pthread_mutex_unlock to avoid this
problem. Perhaps we should do the same?

diff --git a/gcc/testsuite/g++.dg/tsan/cond_race.C
b/gcc/testsuite/g++.dg/tsan/cond_race.C
index a937614..805465d 100644
--- a/gcc/testsuite/g++.dg/tsan/cond_race.C
+++ b/gcc/testsuite/g++.dg/tsan/cond_race.C
@@ -5,6 +5,7 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <pthread.h>
+#include <unistd.h>

 struct Ctx {
   pthread_mutex_t m;
@@ -31,6 +32,8 @@ int main() {
   while (!c->done)
     pthread_cond_wait(&c->c, &c->m);
   pthread_mutex_unlock(&c->m);
+  // w/o this sleep, it can be reported as use-after-free
+  sleep(1);
   delete c;
   pthread_join(th, 0);
 }