Re: [PATCH v2 1/7] locking/rwsem: check for active writer/spinner before wakeup

From: Waiman Long
Date: Mon Aug 11 2014 - 15:35:33 EST


On 08/10/2014 07:50 PM, Davidlohr Bueso wrote:
On Sun, 2014-08-10 at 17:41 -0400, Waiman Long wrote:
On 08/08/2014 03:03 PM, Davidlohr Bueso wrote:
On Fri, 2014-08-08 at 14:30 -0400, Waiman Long wrote:
I have 2 issues about this. First of all, the timing windows between
atomic_set() and mutex_has_owner() check is really small, I doubt it
will be that effective.
That is true, which is why I didn't bother showing any performance data
in the changelog. However, more important than any performance, avoiding
bogus wakeups is the _right_ thing to do when allowing lock stealing.

Secondly, I think you may need to call
mutex_release() and debug_mutex_unlock() to make the debugging code
work, but they seems to be called only under the wait_lock. So I think
there is more work that need to be done before this patch is ready.
When !DEBUG both mutex_release() and debug_mutex_unlock() should be
no-ops. So this allows us to do the mutex_has_owner() check *without*
holding the wait_lock.

When DEBUG is set, we don't even bother calling mutex_has_owner(), so
nothing changes.

I don't understand your concern.
It is true I forgot the fact that MUTEX_SPIN_ON_OWNER is disabled when
DEBUG_MUTEX is on. However, mutex_release is controlled by the LOCKDEP
config variable which is independent of DEBUG_MUTEX. So it is still a
concern.
But afaict you cannot have LOCKDEP without enabling DEBUG_MUTEX (but not
necessarily vice-versa). Both are quite intertwined within other
debugging dependencies/options.


I think you are right. This will require comment in the code to avoid this kind of confusion.

-Longman
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/