On Fri, Dec 2, 2011 at 3:56 PM, Dan Williams <dan.j.williams@xxxxxxxxx> wrote:
[..]
> @@ -2405,7 +2434,7 @@ void drain_workqueue(struct workqueue_struct *wq)
> */
> spin_lock_irq(&workqueue_lock);
> if (!wq->nr_drainers++)
> - wq->flags |= WQ_DRAINING;
> + wq->flags |= WQ_DRAINING | flags;
> spin_unlock_irq(&workqueue_lock);
> reflush:
> flush_workqueue(wq);
> @@ -2429,9 +2458,25 @@ reflush:
> }
>
> spin_lock_irq(&workqueue_lock);
> - if (!--wq->nr_drainers)
> - wq->flags &= ~WQ_DRAINING;
> + if (!--wq->nr_drainers) {
> + wq->flags &= ~(WQ_DRAINING | WQ_NO_DEFER);
> + list_splice_init(&wq->drain_defer, &drain_defer);
> + ret = !list_empty(&drain_defer);
> + }
> spin_unlock_irq(&workqueue_lock);
> +
> + /* requeue work on this queue provided it was not being destroyed */
> + list_for_each_entry_safe(work, w, &drain_defer, entry) {
> + list_del_init(&work->entry);
> + queue_work(wq, work);
> + }
Actually, this won't work. The queuing of deferred work would need to
be under the lock and then flushed again to guarantee workqueue
semantics which just devolves into a "wait until all work stops being
submitted to the queue for one drain_workqueue() duration". Which is
more than what we need for libsas.
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
