[Intel-gfx] [PATCH 2/2] drm/i915: Filter out spurious execlists context-switch interrupts

[Chris Wilson]

Quoting Mika Kuoppala (2017-10-20 14:31:14)
> Chris Wilson <chris at chris-wilson.co.uk> writes:
> 
> > Quoting Mika Kuoppala (2017-10-20 14:21:08)
> >> Chris Wilson <chris at chris-wilson.co.uk> writes:
> >> 
> >> > Back in commit a4b2b01523a8 ("drm/i915: Don't mark an execlists
> >> > context-switch when idle") we noticed the presence of late
> >> > context-switch interrupts. We were able to filter those out by looking
> >> > at whether the ELSP remained active, but in commit beecec901790
> >> > ("drm/i915/execlists: Preemption!") that became problematic as we now
> >> > anticipate receiving a context-switch event for preemption while ELSP
> >> > may be empty. To restore the spurious interrupt suppression, add a
> >> 
> >> This confuses me, how can we preempt something that was never submitted.
> >> Could you elaborate?
> >
> > The ELSP may become empty after we told the hw to switch to the preempt
> > context. So the preemption context-switch may appear as a separate
> > interrupt after !port_isset(execlists.port[0]). The joy of asynchronous
> > communications.
> 
> Let me check if I got this right: as we dont use port[0] to
> switch to preempt context and thus we might get 2 interrupts in a
> a row, first one clearing port[0], we can't assume any idleness
> by port[0] status alone?

Correct.
-Chris