Re: [RFC PATCH 4/4] coroutine/rwlock: Wake writers in preference to readers

[David Edmondson <dme@dme.org>]

On Tuesday, 2021-03-09 at 12:06:22 +01, Paolo Bonzini wrote:

> On 09/03/21 11:21, David Edmondson wrote:
>> -        /* The critical section started in qemu_co_rwlock_wrlock.  */
>> -        qemu_co_queue_restart_all(&lock->queue);
>> +        /* The critical section started in qemu_co_rwlock_wrlock or
>> +         * qemu_co_rwlock_upgrade.
>> +         */
>> +        qemu_co_queue_restart_all(&lock->wqueue);
>> +        qemu_co_queue_restart_all(&lock->rqueue);
>
> Hmm, the devil is in the details---this is a thundering herd waiting to 
> happen.  But fortunately this can be fixed while making the unlock 
> primitive even simpler:
>
>      if (lock->reader) {
>           self->locks_held--;
>
>           /* Read-side critical sections do not keep lock->mutex.  */
>           qemu_co_mutex_lock(&lock->mutex);
>           lock->reader--;
>           assert(lock->reader >= 0);
>       }
>
>       /* If there are no remaining readers wake one waiting writer
>        * or all waiting readers.
>        */
>       if (!lock->reader && !qemu_co_queue_next(&lock->wqueue)) {
>           assert(!lock->pending_writer);
>           qemu_co_queue_restart_all(&lock->rqueue);
>       }

That's a nice improvement, I agree. I'll roll it into another revision
and work on a test.

dme.
-- 
But uh oh, I love her because, she moves in her own way.