Quoting Chris Wilson (2020-01-22 10:53:19) > Since the introduction of preempt-to-busy, we leave the request on the > HW as we process the preemption request. This means that the request may > complete while it is on the submission queue, and once completed it may > be retired. We assumed that a single reference for the construction to > retirement lifetime would suffice to keep the request alive while it is > on the hardware, but with preempt-to-busy that is no longer the case and > we need to explicitly hold the reference while it is being managed by > execlists. > > Reported-by: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx> > Closes: https://gitlab.freedesktop.org/drm/intel/issues/997 > Fixes: 22b7a426bbe1 ("drm/i915/execlists: Preempt-to-busy") Scratch this... > References: b647c7df01b7 ("drm/i915: Fixup preempt-to-busy vs resubmission of a virtual request") > Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> > Cc: Mika Kuoppala <mika.kuoppala@xxxxxxxxxxxxxxx> > Cc: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx> > --- > drivers/gpu/drm/i915/gt/intel_lrc.c | 4 ++++ > 1 file changed, 4 insertions(+) > > diff --git a/drivers/gpu/drm/i915/gt/intel_lrc.c b/drivers/gpu/drm/i915/gt/intel_lrc.c > index 3a30767ff0c4..f47f55228fee 100644 > --- a/drivers/gpu/drm/i915/gt/intel_lrc.c > +++ b/drivers/gpu/drm/i915/gt/intel_lrc.c > @@ -984,6 +984,7 @@ __unwind_incomplete_requests(struct intel_engine_cs *engine) > } > GEM_BUG_ON(RB_EMPTY_ROOT(&engine->execlists.queue.rb_root)); > > + i915_request_get(rq); > list_move(&rq->sched.link, pl); > set_bit(I915_FENCE_FLAG_PQUEUE, &rq->fence.flags); > > @@ -2066,6 +2067,7 @@ static void execlists_dequeue(struct intel_engine_cs *engine) > submit = true; > last = rq; > } > + i915_request_put(rq); > } > > rb_erase_cached(&p->node, &execlists->queue); > @@ -2735,6 +2737,8 @@ static void execlists_submit_request(struct i915_request *request) > struct intel_engine_cs *engine = request->engine; > unsigned long flags; > > + i915_request_get(request); /* hold a reference for the pqueue */ > + > /* Will be called from irq-context when using foreign fences. */ > spin_lock_irqsave(&engine->active.lock, flags); It's not quite so simple. In fact, as we remove the link during completion we are reference safe. Ok, the problem is elsewhere... -Chris _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
