diff mbox

Fix string/tst-xbzero-opt if build with gcc head.

Message ID a1bd8b21-c935-125f-0b9c-381f0ed19b5a@linux.ibm.com
State Superseded
Headers show

Commit Message

Stefan Liebler July 12, 2018, 2:22 p.m. UTC
Fix string/tst-xbzero-opt is build with gcc head.

On s390x, the test string/tst-xbzero-opt is failing if build with GCC head:
FAIL: no clear/prepare: expected 32 got 0
FAIL: no clear/test: expected some got 0
FAIL: ordinary clear/prepare: expected 32 got 0
INFO: ordinary clear/test: found 0 patterns (memset not eliminated)
PASS: explicit clear/prepare: expected 32 got 32
PASS: explicit clear/test: expected 0 got 0

In setup_no_clear / setup_ordinary_clear, GCC is omitting the memcpy 
loop in prepare_test_buffer. Thus count_test_patterns does not find any 
of the test_pattern.

This patch introduces a compiler barrier just after filling the buffer.

If okay, shall this be committed before or after the release?

Bye
Stefan

---

ChangeLog:

	* string/tst-xbzero-opt.c (prepare_test_buffer):
	Add compiler barrier.

Comments

Zack Weinberg July 12, 2018, 4:42 p.m. UTC | #1
On Thu, Jul 12, 2018 at 10:22 AM, Stefan Liebler <stli@linux.ibm.com> wrote:
> Fix string/tst-xbzero-opt is build with gcc head.
...
> In setup_no_clear / setup_ordinary_clear, GCC is omitting the memcpy loop in
> prepare_test_buffer. Thus count_test_patterns does not find any of the
> test_pattern.
>
> This patch introduces a compiler barrier just after filling the buffer.

I think I understand why the call to swapcontext in
prepare_test_buffer is not a sufficient compiler barrier, but I am not
a fan of asm volatile ("" ::: "memory"), because I fully expect some
future compiler to decide that there are no actual assembly
instructions being inserted so the statement can be completely
ignored.  I would prefer us to find some kind of construct that
actually does make externally-visible side effects depend on the
contents of 'buf' in terms of the C abstract machine.

zw
diff mbox

Patch

commit 65f9b078c5053faa93e1f572282463685a869864
Author: Stefan Liebler <stli@linux.ibm.com>
Date:   Thu Jul 12 16:07:26 2018 +0200

    Fix string/tst-xbzero-opt if build with gcc head.
    
    On s390x, the test string/tst-xbzero-opt is failing if build with gcc head:
    FAIL: no clear/prepare: expected 32 got 0
    FAIL: no clear/test: expected some got 0
    FAIL: ordinary clear/prepare: expected 32 got 0
    INFO: ordinary clear/test: found 0 patterns (memset not eliminated)
    PASS: explicit clear/prepare: expected 32 got 32
    PASS: explicit clear/test: expected 0 got 0
    
    In setup_no_clear / setup_ordinary_clear, GCC is omitting the memcpy loop
    in prepare_test_buffer. Thus count_test_patterns does not find any of the
    test_pattern.
    
    This patch introduces a compiler barrier just after filling the buffer
    and the filling of the test_pattern is not omitted.
    
    ChangeLog:
    
            * string/tst-xbzero-opt.c (prepare_test_buffer): Add compiler barrier.

diff --git a/string/tst-xbzero-opt.c b/string/tst-xbzero-opt.c
index cf7041f37a..4c2be0c197 100644
--- a/string/tst-xbzero-opt.c
+++ b/string/tst-xbzero-opt.c
@@ -106,6 +106,9 @@  prepare_test_buffer (unsigned char *buf)
   for (unsigned int i = 0; i < PATTERN_REPS; i++)
     memcpy (buf + i*PATTERN_SIZE, test_pattern, PATTERN_SIZE);
 
+  /* Force the compiler to really copy the pattern to buf.  */
+  __asm__ __volatile__ ("" ::: "memory");
+
   if (swapcontext (&uc_co, &uc_main))
     abort ();
 }