On Wed 23-05-18 19:24:48, Tetsuo Handa wrote:
> Michal Hocko wrote:
> > > I don't understand why you are talking about PF_WQ_WORKER case.
> >
> > Because that seems to be the reason to have it there as per your
> > comment.
>
> OK. Then, I will fold below change into my patch.
>
> if (did_some_progress) {
> no_progress_loops = 0;
> + /*
> -+ * This schedule_timeout_*() serves as a guaranteed sleep for
> -+ * PF_WQ_WORKER threads when __zone_watermark_ok() == false.
> ++ * Try to give the OOM killer/reaper/victims some time for
> ++ * releasing memory.
> + */
> + if (!tsk_is_oom_victim(current))
> + schedule_timeout_uninterruptible(1);
Do you really need this? You are still fiddling with this path at all? I
see how removing the timeout might be reasonable after recent changes
but why do you insist in adding it outside of the lock.
--
Michal Hocko
SUSE Labs
