Messages in this thread | | | Date | Wed, 6 Feb 2013 10:52:13 -0800 | From | Tejun Heo <> | Subject | Re: [PATCH 3/3] stop_machine: wake up stopper thread lazily |
| |
On Wed, Feb 06, 2013 at 08:40:32PM +0800, Hillf Danton wrote: > Ring no bell if the stopper thread is busy in handling enqueued works. > > Signed-off-by: Hillf Danton <dhillf@gmail.com> > --- > > --- a/kernel/stop_machine.c Wed Feb 6 20:05:44 2013 > +++ b/kernel/stop_machine.c Wed Feb 6 20:06:56 2013 > @@ -72,8 +72,10 @@ static void cpu_stop_queue_work(struct c > spin_lock_irqsave(&stopper->lock, flags); > > if (stopper->enabled) { > + int wakeup = list_empty(&stopper->works); > list_add_tail(&work->list, &stopper->works); > - wake_up_process(stopper->thread); > + if (wakeup) > + wake_up_process(stopper->thread);
Why does this matter? It shouldn't matter for correctness. Is it meant to be an optimization? Is it something worth optimizing? We hardly ever have contention on cpu stoppers after all and we shouldn't.
-- tejun
| |