[Intel-gfx] [PATCH 07/10] drm/i915: Reduce spinlock hold time during notify_ring() interrupt

[Chris Wilson]

Quoting Tvrtko Ursulin (2018-01-17 10:45:16)
> 
> On 15/01/2018 21:24, Chris Wilson wrote:
> > -             if (wakeup)
> > -                     wake_up_process(wait->tsk);
> > +                     tsk = wait->tsk;
> > +             } else {
> > +                     if (engine->irq_seqno_barrier &&
> > +                         i915_seqno_passed(seqno, wait->seqno - 1))
> 
> Hm what is this about? Why -1 on platforms with coherency issues and not 
> some other number? Needs a comment as minimum but still is a behaviour 
> change which I did not immediately figure out how it goes with the 
> commit message. If it is some additional optimization it needs to be 
> split out into a separate patch.

It's a finger in the air statement that I don't expect to be more than
one seqno behind in the interrupt-vs-breadcrumb race. So far I haven't
been disappointed.
-Chris