[Intel-gfx] [PATCH] drm/i915/selftests: Try to recover from a wedged GPU during reset tests

[Mika Kuoppala]

Chris Wilson <chris at chris-wilson.co.uk> writes:

> Quoting Mika Kuoppala (2017-09-26 13:48:17)
>> Chris Wilson <chris at chris-wilson.co.uk> writes:
>> 
>> > If we see the seqno stop progressing, we abandon the test for fear that
>> > the GPU died following the reset. However, during test teardown we still
>> > wait for the GPU to idle before continuing, but we have already
>> > confirmed that the GPU is dead. Furthermore, since we are inside a reset
>> > test, we have disabled the hangchecker, and so there is no safety net and
>> > we wait indefinitely. Detect the stuck GPU and declare it wedged as a
>> > state of emergency so we can escape.
>> >
>> > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
>> > Cc: Jari Tahvanainen <jari.tahvanainen at intel.com>
>> > Cc: Mika Kuoppala <mika.kuoppala at linux.intel.com>
>> > ---
>> >  drivers/gpu/drm/i915/selftests/intel_hangcheck.c | 25 +++++++++++++++++++-----
>> >  1 file changed, 20 insertions(+), 5 deletions(-)
>> >
>> > diff --git a/drivers/gpu/drm/i915/selftests/intel_hangcheck.c b/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
>> > index 02e52a146ed8..913fe752f6b4 100644
>> > --- a/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
>> > +++ b/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
>> > @@ -165,6 +165,7 @@ static int emit_recurse_batch(struct hang *h,
>> >               *batch++ = lower_32_bits(vma->node.start);
>> >       }
>> >       *batch++ = MI_BATCH_BUFFER_END; /* not reached */
>> > +     wmb();
>> >
>> 
>> Why not the big hammer with i915_gem_chipset_flush() here?
>
> It didn't cross my mind, I was just doodling :)
>
>> 
>> >       flags = 0;
>> >       if (INTEL_GEN(vm->i915) <= 5)
>> > @@ -621,7 +622,12 @@ static int igt_wait_reset(void *arg)
>> >       __i915_add_request(rq, true);
>> >  
>> >       if (!wait_for_hang(&h, rq)) {
>> > -             pr_err("Failed to start request %x\n", rq->fence.seqno);
>> > +             pr_err("Failed to start request %x, at %x\n",
>> > +                    rq->fence.seqno, hws_seqno(&h, rq));
>> > +
>> > +             i915_reset(i915, 0);
>> > +             i915_gem_set_wedged(i915);
>> > +
>> >               err = -EIO;
>> >               goto out_rq;
>> >       }
>> > @@ -708,10 +714,14 @@ static int igt_reset_queue(void *arg)
>> >                       __i915_add_request(rq, true);
>> >  
>> >                       if (!wait_for_hang(&h, prev)) {
>> > -                             pr_err("Failed to start request %x\n",
>> > -                                    prev->fence.seqno);
>> > +                             pr_err("Failed to start request %x, at %x\n",
>> > +                                    rq->fence.seqno, hws_seqno(&h, rq));
>> 
>> As you pointed out the debug in here is for wrong request.
>> 
>> Reviewed-by: Mika Kuoppala <mika.kuoppala at linux.intel.com>
>
> Happy if I drop the wmb() for a later patch and replace it with a
> chipset flush instead?

Will be happy.
-Mika

> -Chris