[Intel-gfx] [PATCH] drm/i915: Flush pending interrupt following a GPU reset

Chris Wilson chris at chris-wilson.co.uk
Wed Mar 21 16:41:32 UTC 2018

Quoting Jeff McGee (2018-03-21 15:55:16)
> On Wed, Mar 21, 2018 at 03:00:23PM +0000, Chris Wilson wrote:
> > After resetting the GPU (or subset of engines), call synchronize_irq()
> > to flush any pending irq before proceeding with the cleanup. For a
> > device level reset, we disable the interupts around the reset, but when
> > resetting just one engine, we have to avoid such global disabling. This
> > leaves us open to an interrupt arriving for the engine as we try to
> > reset it. We already do try to flush the IIR following the reset, but we
> > have to ensure that the in-flight interrupt does not land after we start
> > cleaning up after the reset; enter synchronize_irq().
> > 
> > As it current stands, we very rarely, but fatally, see sequences such as:
> > 
> >     2.... 57964564us : execlists_reset_prepare: rcs0
> >     2.... 57964613us : execlists_reset: rcs0 seqno=424
> >     0d.h1 57964615us : gen8_cs_irq_handler: rcs0 CS active=1
> >     2d..1 57964617us : __i915_request_unsubmit: rcs0 fence 29:1056 <- global_seqno 1060
> >     2.... 57964703us : execlists_reset_finish: rcs0
> >     0..s. 57964705us : execlists_submission_tasklet: rcs0 awake?=1, active=0, irq-posted?=1
> > 
> I can repro this sequence easily with force preemption IGT. Just tried this
> patch and the issue is still there. For the moment I can just mitigate with
> https://patchwork.freedesktop.org/patch/211086/

These patch is purely to make sure that the irq_handler is not called
after execlists_reset, as that is what we rely on.

We *should* never be in a situation where CSB tail is invalid, that
implies either we screwed up or the hw is incoherent.

More information about the Intel-gfx mailing list