Quoting Chris Wilson (2020-01-27 16:03:11)
> Quoting Tvrtko Ursulin (2020-01-27 15:52:51)
> >
> > On 26/01/2020 10:23, Chris Wilson wrote:
> > > diff --git a/drivers/gpu/drm/i915/gt/intel_gt_irq.c b/drivers/gpu/drm/i915/gt/intel_gt_irq.c
> > > index f796bdf1ed30..6ae64a224b02 100644
> > > --- a/drivers/gpu/drm/i915/gt/intel_gt_irq.c
> > > +++ b/drivers/gpu/drm/i915/gt/intel_gt_irq.c
> > > @@ -24,6 +24,13 @@ cs_irq_handler(struct intel_engine_cs *engine, u32 iir)
> > > {
> > > bool tasklet = false;
> > >
> > > + if (iir & GT_WAIT_SEMAPHORE_INTERRUPT) {
> > > + WRITE_ONCE(engine->execlists.yield,
> > > + ENGINE_READ_FW(engine, EXECLIST_CCID));
> > > + if (del_timer(&engine->execlists.timer))
> > > + tasklet = true;
> >
> > What if it fires before timeslice timer has been set up and when we miss
> > to yield?
>
> We only set the timer after the HW ack, and we can legitimately hit a
> semaphore in the user payload before we see the ack. That is
> demonstrated aptly by live_timeslice_preempt.
Hmm. I thought we would see the yield such that we would dequeue and
process the new timelice regardless of the previous timer.
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
