Hi Vicentiu, I'm back from vacation, but quite busy with paperwork etc. I'll review your patch tomorrow. Regards, Sergey On Mon, Aug 01, 2016 at 02:46:03PM +0300, vicentiu wrote:

revision-id: 5486f3458859fa4bd161150b9ac0e9c0f633a0e3 (mariadb-10.2.1-9-g5486f34) parent(s): 08683a726773f8cdf16a4a3dfb3920e5f7842481 author: Vicențiu Ciorbaru committer: Vicențiu Ciorbaru timestamp: 2016-08-01 14:43:41 +0300 message:

MDEV-10341: InnoDB: Failing assertion: mutex_own(mutex) - mutex_exit_func

Fix memory barrier issues on releasing mutexes. We must have

--- storage/innobase/include/os0sync.h | 17 ++++++++--------- storage/innobase/include/sync0sync.ic | 2 ++ storage/xtradb/include/os0sync.h | 16 ++++++++-------- storage/xtradb/include/sync0sync.ic | 2 ++ 4 files changed, 20 insertions(+), 17 deletions(-)

diff --git a/storage/innobase/include/os0sync.h b/storage/innobase/include/os0sync.h index 1cf4e9c..08911e7 100644 --- a/storage/innobase/include/os0sync.h +++ b/storage/innobase/include/os0sync.h @@ -479,20 +479,13 @@ os_atomic_test_and_set(volatile lock_word_t* ptr) }

/** Do an atomic release. - -In theory __sync_lock_release should be used to release the lock. -Unfortunately, it does not work properly alone. The workaround is -that more conservative __sync_lock_test_and_set is used instead. - -Performance regression was observed at some conditions for Intel -architecture. Disable release barrier on Intel architecture for now. @param[in,out] ptr Memory location to write to @return the previous value */ inline -lock_word_t +void os_atomic_clear(volatile lock_word_t* ptr) { - return(__sync_lock_test_and_set(ptr, 0)); + __sync_lock_release(ptr); }

# elif defined(HAVE_IB_GCC_ATOMIC_TEST_AND_SET) @@ -861,6 +854,12 @@ for synchronization */ architecture. Disable memory barrier for Intel architecture for now. */ # define os_rmb do { } while(0) # define os_wmb do { } while(0) +# ifndef __WIN__ +# define os_mb __sync_synchronize() +# else + /* Windows atomic primitives imply a full memory barrier themselves. */ +# define os_mb do { } while(0) +# endif # define os_isync do { } while(0) # define IB_MEMORY_BARRIER_STARTUP_MSG \ "Memory barrier is not used" diff --git a/storage/innobase/include/sync0sync.ic b/storage/innobase/include/sync0sync.ic index 85c7f75..19474e7 100644 --- a/storage/innobase/include/sync0sync.ic +++ b/storage/innobase/include/sync0sync.ic @@ -178,6 +178,8 @@ mutex_exit_func( to wake up possible hanging threads if they are missed in mutex_signal_object. */

+ os_mb; + if (mutex_get_waiters(mutex) != 0) {

mutex_signal_object(mutex); diff --git a/storage/xtradb/include/os0sync.h b/storage/xtradb/include/os0sync.h index 0f93f3f..f06aa9d 100644 --- a/storage/xtradb/include/os0sync.h +++ b/storage/xtradb/include/os0sync.h @@ -531,19 +531,13 @@ os_atomic_test_and_set(volatile lock_word_t* ptr)

/** Do an atomic release.

-In theory __sync_lock_release should be used to release the lock. -Unfortunately, it does not work properly alone. The workaround is -that more conservative __sync_lock_test_and_set is used instead. - -Performance regression was observed at some conditions for Intel -architecture. Disable release barrier on Intel architecture for now. @param[in,out] ptr Memory location to write to @return the previous value */ inline -lock_word_t +void os_atomic_clear(volatile lock_word_t* ptr) { - return(__sync_lock_test_and_set(ptr, 0)); + __sync_lock_release(ptr); }

# elif defined(HAVE_IB_GCC_ATOMIC_TEST_AND_SET) @@ -912,6 +906,12 @@ for synchronization */ architecture. Disable memory barrier for Intel architecture for now. */ # define os_rmb do { } while(0) # define os_wmb do { } while(0) +# ifndef __WIN__ +# define os_mb __sync_synchronize() +# else + /* Windows atomic primitives imply a full memory barrier themselves. */ +# define os_mb do { } while(0) +# endif # define os_isync do { } while(0) # define IB_MEMORY_BARRIER_STARTUP_MSG \ "Memory barrier is not used" diff --git a/storage/xtradb/include/sync0sync.ic b/storage/xtradb/include/sync0sync.ic index 83f28bf..dbecc79 100644 --- a/storage/xtradb/include/sync0sync.ic +++ b/storage/xtradb/include/sync0sync.ic @@ -181,6 +181,8 @@ mutex_exit_func( to wake up possible hanging threads if they are missed in mutex_signal_object. */

+ os_mb; + if (mutex_get_waiters(mutex) != 0) {

mutex_signal_object(mutex); _______________________________________________ commits mailing list commits@mariadb.org https://lists.askmonty.org/cgi-bin/mailman/listinfo/commits