Re: svn commit: r232209 - in head: lib/libthr/thread sys/kern
On Sun, Mar 18, 2012 at 05:56:02PM -0700, Julian Elischer wrote: On 3/18/12 11:50 AM, Pawel Jakub Dawidek wrote: On Mon, Feb 27, 2012 at 01:38:52PM +, David Xu wrote: Author: davidxu Date: Mon Feb 27 13:38:52 2012 New Revision: 232209 URL: http://svn.freebsd.org/changeset/base/232209 Log: Follow changes made in revision 232144, pass absolute timeout to kernel, this eliminates a clock_gettime() syscall. This or some other change has broken CLOCK_MONOTONIC usage with condition variables. You should be able to reproduce this by something like this: pthread_cond_t cv; pthread_condattr_t attr; pthread_mutex_t lock; struct timespec ts; int error; (void)pthread_mutex_init(lock, NULL); (void)pthread_condattr_init(attr); (void)pthread_condattr_setclock(attr, CLOCK_MONOTONIC); (void)pthread_cond_init(cv,attr); (void)pthread_condattr_destroy(attr); (void)clock_gettime(CLOCK_MONOTONIC,ts); ts.tv_sec += 10; (void)pthread_mutex_lock(lock); (void)pthread_cond_timedwait(cv,lock,ts); (void)pthread_mutex_unlock(lock); This should timeout after 10 seconds, but pthread_cond_timedwait(3) returns immediately with ETIMEDOUT. CLOCK_REALTIME works properly. Bascially pthread_condattr_setclock(attr, CLOCK_MONOTONIC) is no-op. If you change CLOCK_MONOTONIC to CLOCK_REALTIME in clock_gettime(2) call, it will timeout after 10 seconds. this has been broken for a while for me in fact fixing this was, I thought, one of the reasons for this work.. glad I'm not the omnly person seeing it though. for me it broke sysutils/fio from ports For me it breaks sbin/hastd and openbsm/auditdistd, but my feeling is that it is was broken recently. This stuff use to work not so long ago. -- Pawel Jakub Dawidek http://www.wheelsystems.com FreeBSD committer http://www.FreeBSD.org Am I Evil? Yes, I Am! http://tupytaj.pl pgpOKwkORW0RY.pgp Description: PGP signature
Re: svn commit: r232209 - in head: lib/libthr/thread sys/kern
On 2012/3/19 15:33, Pawel Jakub Dawidek wrote: On Sun, Mar 18, 2012 at 05:56:02PM -0700, Julian Elischer wrote: On 3/18/12 11:50 AM, Pawel Jakub Dawidek wrote: On Mon, Feb 27, 2012 at 01:38:52PM +, David Xu wrote: Author: davidxu Date: Mon Feb 27 13:38:52 2012 New Revision: 232209 URL: http://svn.freebsd.org/changeset/base/232209 Log: Follow changes made in revision 232144, pass absolute timeout to kernel, this eliminates a clock_gettime() syscall. This or some other change has broken CLOCK_MONOTONIC usage with condition variables. You should be able to reproduce this by something like this: pthread_cond_t cv; pthread_condattr_t attr; pthread_mutex_t lock; struct timespec ts; int error; (void)pthread_mutex_init(lock, NULL); (void)pthread_condattr_init(attr); (void)pthread_condattr_setclock(attr, CLOCK_MONOTONIC); (void)pthread_cond_init(cv,attr); (void)pthread_condattr_destroy(attr); (void)clock_gettime(CLOCK_MONOTONIC,ts); ts.tv_sec += 10; (void)pthread_mutex_lock(lock); (void)pthread_cond_timedwait(cv,lock,ts); (void)pthread_mutex_unlock(lock); This should timeout after 10 seconds, but pthread_cond_timedwait(3) returns immediately with ETIMEDOUT. CLOCK_REALTIME works properly. Bascially pthread_condattr_setclock(attr, CLOCK_MONOTONIC) is no-op. If you change CLOCK_MONOTONIC to CLOCK_REALTIME in clock_gettime(2) call, it will timeout after 10 seconds. this has been broken for a while for me in fact fixing this was, I thought, one of the reasons for this work.. glad I'm not the omnly person seeing it though. for me it broke sysutils/fio from ports For me it breaks sbin/hastd and openbsm/auditdistd, but my feeling is that it is was broken recently. This stuff use to work not so long ago. Hi, Revision 233134 should have fixed the problem. http://svn.freebsd.org/changeset/base/233134 ___ svn-src-head@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/svn-src-head To unsubscribe, send any mail to svn-src-head-unsubscr...@freebsd.org
Re: svn commit: r232209 - in head: lib/libthr/thread sys/kern
On Mon, Mar 19, 2012 at 04:16:58PM +0800, David Xu wrote: Revision 233134 should have fixed the problem. http://svn.freebsd.org/changeset/base/233134 Works now, thanks for the quick fix. -- Pawel Jakub Dawidek http://www.wheelsystems.com FreeBSD committer http://www.FreeBSD.org Am I Evil? Yes, I Am! http://tupytaj.pl pgpYEhdhNTPFI.pgp Description: PGP signature
Re: svn commit: r232209 - in head: lib/libthr/thread sys/kern
On Mon, Feb 27, 2012 at 01:38:52PM +, David Xu wrote: Author: davidxu Date: Mon Feb 27 13:38:52 2012 New Revision: 232209 URL: http://svn.freebsd.org/changeset/base/232209 Log: Follow changes made in revision 232144, pass absolute timeout to kernel, this eliminates a clock_gettime() syscall. This or some other change has broken CLOCK_MONOTONIC usage with condition variables. You should be able to reproduce this by something like this: pthread_cond_t cv; pthread_condattr_t attr; pthread_mutex_t lock; struct timespec ts; int error; (void)pthread_mutex_init(lock, NULL); (void)pthread_condattr_init(attr); (void)pthread_condattr_setclock(attr, CLOCK_MONOTONIC); (void)pthread_cond_init(cv, attr); (void)pthread_condattr_destroy(attr); (void)clock_gettime(CLOCK_MONOTONIC, ts); ts.tv_sec += 10; (void)pthread_mutex_lock(lock); (void)pthread_cond_timedwait(cv, lock, ts); (void)pthread_mutex_unlock(lock); This should timeout after 10 seconds, but pthread_cond_timedwait(3) returns immediately with ETIMEDOUT. CLOCK_REALTIME works properly. Bascially pthread_condattr_setclock(attr, CLOCK_MONOTONIC) is no-op. If you change CLOCK_MONOTONIC to CLOCK_REALTIME in clock_gettime(2) call, it will timeout after 10 seconds. -- Pawel Jakub Dawidek http://www.wheelsystems.com FreeBSD committer http://www.FreeBSD.org Am I Evil? Yes, I Am! http://tupytaj.pl pgp18JNvAfgtP.pgp Description: PGP signature
Re: svn commit: r232209 - in head: lib/libthr/thread sys/kern
On 3/18/12 11:50 AM, Pawel Jakub Dawidek wrote: On Mon, Feb 27, 2012 at 01:38:52PM +, David Xu wrote: Author: davidxu Date: Mon Feb 27 13:38:52 2012 New Revision: 232209 URL: http://svn.freebsd.org/changeset/base/232209 Log: Follow changes made in revision 232144, pass absolute timeout to kernel, this eliminates a clock_gettime() syscall. This or some other change has broken CLOCK_MONOTONIC usage with condition variables. You should be able to reproduce this by something like this: pthread_cond_t cv; pthread_condattr_t attr; pthread_mutex_t lock; struct timespec ts; int error; (void)pthread_mutex_init(lock, NULL); (void)pthread_condattr_init(attr); (void)pthread_condattr_setclock(attr, CLOCK_MONOTONIC); (void)pthread_cond_init(cv,attr); (void)pthread_condattr_destroy(attr); (void)clock_gettime(CLOCK_MONOTONIC,ts); ts.tv_sec += 10; (void)pthread_mutex_lock(lock); (void)pthread_cond_timedwait(cv,lock,ts); (void)pthread_mutex_unlock(lock); This should timeout after 10 seconds, but pthread_cond_timedwait(3) returns immediately with ETIMEDOUT. CLOCK_REALTIME works properly. Bascially pthread_condattr_setclock(attr, CLOCK_MONOTONIC) is no-op. If you change CLOCK_MONOTONIC to CLOCK_REALTIME in clock_gettime(2) call, it will timeout after 10 seconds. this has been broken for a while for me in fact fixing this was, I thought, one of the reasons for this work.. glad I'm not the omnly person seeing it though. for me it broke sysutils/fio from ports ___ svn-src-head@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/svn-src-head To unsubscribe, send any mail to svn-src-head-unsubscr...@freebsd.org
svn commit: r232209 - in head: lib/libthr/thread sys/kern
Author: davidxu
Date: Mon Feb 27 13:38:52 2012
New Revision: 232209
URL: http://svn.freebsd.org/changeset/base/232209
Log:
Follow changes made in revision 232144, pass absolute timeout to kernel,
this eliminates a clock_gettime() syscall.
Modified:
head/lib/libthr/thread/thr_rwlock.c
head/lib/libthr/thread/thr_umtx.c
head/lib/libthr/thread/thr_umtx.h
head/sys/kern/kern_umtx.c
Modified: head/lib/libthr/thread/thr_rwlock.c
==
--- head/lib/libthr/thread/thr_rwlock.c Mon Feb 27 13:04:09 2012
(r232208)
+++ head/lib/libthr/thread/thr_rwlock.c Mon Feb 27 13:38:52 2012
(r232209)
@@ -123,7 +123,6 @@ rwlock_rdlock_common(pthread_rwlock_t *r
{
struct pthread *curthread = _get_curthread();
pthread_rwlock_t prwlock;
- struct timespec ts, ts2, *tsp;
int flags;
int ret;
@@ -162,18 +161,8 @@ rwlock_rdlock_common(pthread_rwlock_t *r
return (EINVAL);
for (;;) {
- if (abstime) {
- clock_gettime(CLOCK_REALTIME, ts);
- TIMESPEC_SUB(ts2, abstime, ts);
- if (ts2.tv_sec 0 ||
- (ts2.tv_sec == 0 ts2.tv_nsec = 0))
- return (ETIMEDOUT);
- tsp = ts2;
- } else
- tsp = NULL;
-
/* goto kernel and lock it */
- ret = __thr_rwlock_rdlock(prwlock-lock, flags, tsp);
+ ret = __thr_rwlock_rdlock(prwlock-lock, flags, abstime);
if (ret != EINTR)
break;
@@ -255,7 +244,6 @@ rwlock_wrlock_common (pthread_rwlock_t *
{
struct pthread *curthread = _get_curthread();
pthread_rwlock_t prwlock;
- struct timespec ts, ts2, *tsp;
int ret;
CHECK_AND_INIT_RWLOCK
@@ -275,18 +263,8 @@ rwlock_wrlock_common (pthread_rwlock_t *
return (EINVAL);
for (;;) {
- if (abstime != NULL) {
- clock_gettime(CLOCK_REALTIME, ts);
- TIMESPEC_SUB(ts2, abstime, ts);
- if (ts2.tv_sec 0 ||
- (ts2.tv_sec == 0 ts2.tv_nsec = 0))
- return (ETIMEDOUT);
- tsp = ts2;
- } else
- tsp = NULL;
-
/* goto kernel and lock it */
- ret = __thr_rwlock_wrlock(prwlock-lock, tsp);
+ ret = __thr_rwlock_wrlock(prwlock-lock, abstime);
if (ret == 0) {
prwlock-owner = curthread;
break;
Modified: head/lib/libthr/thread/thr_umtx.c
==
--- head/lib/libthr/thread/thr_umtx.c Mon Feb 27 13:04:09 2012
(r232208)
+++ head/lib/libthr/thread/thr_umtx.c Mon Feb 27 13:38:52 2012
(r232209)
@@ -265,15 +265,42 @@ _thr_ucond_broadcast(struct ucond *cv)
}
int
-__thr_rwlock_rdlock(struct urwlock *rwlock, int flags, struct timespec *tsp)
+__thr_rwlock_rdlock(struct urwlock *rwlock, int flags,
+ const struct timespec *tsp)
{
- return _umtx_op_err(rwlock, UMTX_OP_RW_RDLOCK, flags, NULL, tsp);
+ struct _umtx_time timeout, *tm_p;
+ size_t tm_size;
+
+ if (tsp == NULL) {
+ tm_p = NULL;
+ tm_size = 0;
+ } else {
+ timeout._timeout = *tsp;
+ timeout._flags = UMTX_ABSTIME;
+ timeout._clockid = CLOCK_REALTIME;
+ tm_p = timeout;
+ tm_size = sizeof(timeout);
+ }
+ return _umtx_op_err(rwlock, UMTX_OP_RW_RDLOCK, flags, (void *)tm_size,
tm_p);
}
int
-__thr_rwlock_wrlock(struct urwlock *rwlock, struct timespec *tsp)
+__thr_rwlock_wrlock(struct urwlock *rwlock, const struct timespec *tsp)
{
- return _umtx_op_err(rwlock, UMTX_OP_RW_WRLOCK, 0, NULL, tsp);
+ struct _umtx_time timeout, *tm_p;
+ size_t tm_size;
+
+ if (tsp == NULL) {
+ tm_p = NULL;
+ tm_size = 0;
+ } else {
+ timeout._timeout = *tsp;
+ timeout._flags = UMTX_ABSTIME;
+ timeout._clockid = CLOCK_REALTIME;
+ tm_p = timeout;
+ tm_size = sizeof(timeout);
+ }
+ return _umtx_op_err(rwlock, UMTX_OP_RW_WRLOCK, 0, (void *)tm_size,
tm_p);
}
int
Modified: head/lib/libthr/thread/thr_umtx.h
==
--- head/lib/libthr/thread/thr_umtx.h Mon Feb 27 13:04:09 2012
(r232208)
+++ head/lib/libthr/thread/thr_umtx.h Mon Feb 27 13:38:52 2012
(r232209)
@@ -60,8 +60,10 @@ void _thr_ucond_init(struct ucond *cv) _
int _thr_ucond_signal(struct ucond *cv) __hidden;
int _thr_ucond_broadcast(struct ucond *cv) __hidden;
Re: svn commit: r232209 - in head: lib/libthr/thread sys/kern
On 27 February 2012 14:38, David Xu davi...@freebsd.org wrote: Author: davidxu Date: Mon Feb 27 13:38:52 2012 New Revision: 232209 URL: http://svn.freebsd.org/changeset/base/232209 Log: Follow changes made in revision 232144, pass absolute timeout to kernel, this eliminates a clock_gettime() syscall. Any chance of a MFC? I use rwlocks a lot; removing a syscall from all operations looks like a nice improvement. ___ svn-src-head@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/svn-src-head To unsubscribe, send any mail to svn-src-head-unsubscr...@freebsd.org
Re: svn commit: r232209 - in head: lib/libthr/thread sys/kern
On 2012/2/27 22:29, Ivan Voras wrote: On 27 February 2012 14:38, David Xudavi...@freebsd.org wrote: Author: davidxu Date: Mon Feb 27 13:38:52 2012 New Revision: 232209 URL: http://svn.freebsd.org/changeset/base/232209 Log: Follow changes made in revision 232144, pass absolute timeout to kernel, this eliminates a clock_gettime() syscall. Any chance of a MFC? I use rwlocks a lot; removing a syscall from all operations looks like a nice improvement. Yes, I will do a MFC when it is ready. ;-) ___ svn-src-head@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/svn-src-head To unsubscribe, send any mail to svn-src-head-unsubscr...@freebsd.org
