Re: [PATCH 2/2 v2] sched/wait: Introduce lock breaker in wake_up_page_bit

[Nicholas Piggin <npiggin@gmail.com>]

On Mon, 28 Aug 2017 11:16:48 +1000
Nicholas Piggin <npiggin@gmail.com> wrote:

> On Sun, 27 Aug 2017 16:12:19 -0700
> Linus Torvalds <torvalds@linux-foundation.org> wrote:
> 

> >     diff --git a/mm/filemap.c b/mm/filemap.c
> >     index baba290c276b..0b41c8cbeabc 100644
> >     --- a/mm/filemap.c
> >     +++ b/mm/filemap.c
> >     @@ -986,10 +986,6 @@ static inline int
> > wait_on_page_bit_common(wait_queue_head_t *q,
> > 
> >                 if (likely(test_bit(bit_nr, &page->flags))) {
> >                         io_schedule();
> >     -                   if (unlikely(signal_pending_state(state, current))) {
> >     -                           ret = -EINTR;
> >     -                           break;
> >     -                   }
> >                 }
> > 
> >                 if (lock) {
> >     @@ -999,6 +995,11 @@ static inline int
> > wait_on_page_bit_common(wait_queue_head_t *q,
> >                         if (!test_bit(bit_nr, &page->flags))
> >                                 break;
> >                 }
> >     +
> >     +           if (unlikely(signal_pending_state(state, current))) {
> >     +                   ret = -EINTR;
> >     +                   break;
> >     +           }
> >         }
> > 
> >         finish_wait(q, wait);
> > 
> > but maybe I'm missing something.
> > 
> > Nick, comments?  
> 
> No I don't think you're missing something. We surely could lose our only
> wakeup in this window. So an exclusive waiter has to always make sure
> they propagate the wakeup (regardless of what they do with the contended
> resources itself).
> 
> Seems like your fix should solve it. By the look of how wait_on_bit_lock
> is structured, the author probably did think about this case a little
> better than I did :\

BTW. since you are looking at this stuff, one other small problem I remember
with exclusive waiters is that losing to a concurrent locker puts them to
the back of the queue. I think that could be fixed with some small change to
the wait loops (first add to tail, then retries add to head). Thoughts?

Thanks,
Nick

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>