Skip to content

Commit

Permalink
Browse files Browse the repository at this point in the history
Don't do lock elision on an error checking mutex (bug 17514)
Error checking mutexes are not supposed to be subject to lock elision.
That would defeat the error checking nature of the mutex because lock
elision doesn't record ownership.
  • Loading branch information
Andreas Schwab committed Jan 25, 2016
1 parent 9200e58 commit 5aded6f
Show file tree
Hide file tree
Showing 4 changed files with 76 additions and 4 deletions.
14 changes: 12 additions & 2 deletions ChangeLog
@@ -1,3 +1,11 @@
2016-01-25 Andreas Schwab <schwab@suse.de>

[BZ #17514]
* nptl/pthread_mutex_timedlock.c (pthread_mutex_timedlock)
<case PTHREAD_MUTEX_ERRORCHECK_NP>: Don't do lock elision.
* nptl/Makefile (tests): Add tst-mutex-errorcheck.
* nptl/tst-mutex-errorcheck.c: New file.

2016-01-25 Paul E. Murphy <murphyp@linux.vnet.ibm.com>

[BZ #18560]
Expand All @@ -24,8 +32,10 @@
2016-01-24 David S. Miller <davem@davemloft.net>

* sysdeps/sparc/sparc32/fpu/e_sqrtl.c: New file.
* sysdeps/sparc/sparc32/soft-fp/q_sqrt.c (__ieee754_sqrtl): Remove alias.
* sysdeps/unix/sysv/linux/sparc/sparc32/libm.abilist: Add __sqrtl_finite.
* sysdeps/sparc/sparc32/soft-fp/q_sqrt.c (__ieee754_sqrtl): Remove
alias.
* sysdeps/unix/sysv/linux/sparc/sparc32/libm.abilist: Add
__sqrtl_finite.

* sysdeps/sparc/fpu/libm-test-ulps: Update.

Expand Down
2 changes: 1 addition & 1 deletion nptl/Makefile
Expand Up @@ -288,7 +288,7 @@ tests = tst-typesizes \
tst-initializers1 $(addprefix tst-initializers1-,\
c89 gnu89 c99 gnu99 c11 gnu11) \
tst-bad-schedattr \
tst-thread_local1
tst-thread_local1 tst-mutex-errorcheck
xtests = tst-setuid1 tst-setuid1-static tst-setuid2 \
tst-mutexpp1 tst-mutexpp6 tst-mutexpp10
test-srcs = tst-oddstacklimit
Expand Down
3 changes: 2 additions & 1 deletion nptl/pthread_mutex_timedlock.c
Expand Up @@ -89,7 +89,8 @@ pthread_mutex_timedlock (pthread_mutex_t *mutex,
if (__glibc_unlikely (mutex->__data.__owner == id))
return EDEADLK;

/* FALLTHROUGH */
/* Don't do lock elision on an error checking mutex. */
goto simple;

case PTHREAD_MUTEX_TIMED_NP:
FORCE_ELISION (mutex, goto elision);
Expand Down
61 changes: 61 additions & 0 deletions nptl/tst-mutex-errorcheck.c
@@ -0,0 +1,61 @@
/* Check that error checking mutexes are not subject to lock elision.
Copyright (C) 2016 Free Software Foundation, Inc.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
<http://www.gnu.org/licenses/>. */

#include <stdio.h>
#include <errno.h>
#include <time.h>
#include <pthread.h>

static int
do_test (void)
{
struct timespec tms = { 0 };
pthread_mutex_t mutex;
pthread_mutexattr_t mutexattr;
int ret = 0;

if (pthread_mutexattr_init (&mutexattr) != 0)
return 1;
if (pthread_mutexattr_settype (&mutexattr, PTHREAD_MUTEX_ERRORCHECK) != 0)
return 1;

if (pthread_mutex_init (&mutex, &mutexattr) != 0)
return 1;
if (pthread_mutexattr_destroy (&mutexattr) != 0)
return 1;

/* The call to pthread_mutex_timedlock erroneously enabled lock elision
on the mutex, which then triggered an assertion failure in
pthread_mutex_unlock. It would also defeat the error checking nature
of the mutex. */
if (pthread_mutex_timedlock (&mutex, &tms) != 0)
return 1;
if (pthread_mutex_timedlock (&mutex, &tms) != EDEADLK)
{
printf ("Failed error checking on locked mutex\n");
ret = 1;
}

if (pthread_mutex_unlock (&mutex) != 0)
ret = 1;

return ret;
}

#define TEST_FUNCTION do_test ()
#include "../test-skeleton.c"

0 comments on commit 5aded6f

Please sign in to comment.