Mark success return value as volatile to work around rescheduling

Resolves #15921

The test case nptl/tst-cleanup2 fails on s390x and power6 due to
instruction sheduling in gcc.  This was reported in gcc:

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58034

but it was concluded that gcc is allowed to assume that the first
argument to sprintf is a character array - NULL not being a valid
character array.

2 files changed, 13 insertions, 1 deletions

diff --git a/nptl/ChangeLog b/nptl/ChangeLog
index d1efbb43e5..7511f701ec 100644
--- a/nptl/ChangeLog
+++ b/nptl/ChangeLog
@@ -1,3 +1,9 @@
+2013-09-03  Siddhesh Poyarekar  <siddhesh@redhat.com>
+
+	[BZ #15921]
+	* tst-cleanup2.c (do_test): New volatile variable RET to
+	return success.
+
 2013-08-30   Ondřej Bílka  <neleai@seznam.cz>
 
 	* sysdeps/pthread/pthread.h: Fix typos.
diff --git a/nptl/tst-cleanup2.c b/nptl/tst-cleanup2.c
index 5bd16095a6..65af0f2018 100644
--- a/nptl/tst-cleanup2.c
+++ b/nptl/tst-cleanup2.c
@@ -34,6 +34,12 @@ static int
 do_test (void)
 {
   char *p = NULL;
+  /* gcc can overwrite the success written value by scheduling instructions
+     around sprintf.  It is allowed to do this since according to C99 the first
+     argument of sprintf is a character array and NULL is not a valid character
+     array.  Mark the return value as volatile so that it gets reloaded on
+     return.  */
+  volatile int ret = 0;
   struct sigaction sa;
 
   sa.sa_handler = sig_handler;
@@ -50,7 +56,7 @@ do_test (void)
   if (setjmp (jmpbuf))
     {
       puts ("Exiting main...");
-      return 0;
+      return ret;
     }
 
   sprintf (p, "This should segv\n");