[Intel-gfx] [PATCH 1/4] drm/i915/gt: Disable arbitration around Braswell's pdp updates

Chris Wilson chris at chris-wilson.co.uk
Mon Jan 11 16:00:20 UTC 2021


Quoting Tvrtko Ursulin (2021-01-11 15:53:47)
> 
> On 11/01/2021 10:57, Chris Wilson wrote:
> > Braswell's pdp workaround is full of dragons, that may be being angered
> > when they are interrupted. Let's not take that risk and disable
> > arbitrartion during the update.
> > 
> > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> > ---
> >   drivers/gpu/drm/i915/gt/intel_execlists_submission.c | 11 ++++++++++-
> >   1 file changed, 10 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/gpu/drm/i915/gt/intel_execlists_submission.c b/drivers/gpu/drm/i915/gt/intel_execlists_submission.c
> > index 52c1fe62bdfe..10e9940cf3f5 100644
> > --- a/drivers/gpu/drm/i915/gt/intel_execlists_submission.c
> > +++ b/drivers/gpu/drm/i915/gt/intel_execlists_submission.c
> > @@ -2539,6 +2539,14 @@ static int emit_pdps(struct i915_request *rq)
> >        * GPU hangs to forcewake errors and machine lockups!
> >        */
> >   
> > +     cs = intel_ring_begin(rq, 2);
> > +     if (IS_ERR(cs))
> > +             return PTR_ERR(cs);
> > +
> > +     *cs++ = MI_ARB_ON_OFF | MI_ARB_DISABLE;
> > +     *cs++ = MI_NOOP;
> > +     intel_ring_advance(rq, cs);
> > +
> >       /* Flush any residual operations from the context load */
> >       err = engine->emit_flush(rq, EMIT_FLUSH);
> >       if (err)
> > @@ -2564,7 +2572,8 @@ static int emit_pdps(struct i915_request *rq)
> >               *cs++ = i915_mmio_reg_offset(GEN8_RING_PDP_LDW(base, i));
> >               *cs++ = lower_32_bits(pd_daddr);
> >       }
> > -     *cs++ = MI_NOOP;
> > +     *cs++ = MI_ARB_ON_OFF | MI_ARB_ENABLE;
> > +     intel_ring_advance(rq, cs);
> >   
> >       intel_ring_advance(rq, cs);
> >   
> > 
> 
> I had to remind myself that Gen8LP is indeed the only platform with 
> 32-bit ppgtt.
> 
> I presume you are fixing some sporadic CI failures here, anyway:

There's an odd one showing up in Braswell CI, but didn't occur locally
in several hours of running the selftest. Given the history of how
sensitive Braswell is to these PDP updates, I think this wild stab in
the dark is likely to hit something.
-Chris


More information about the Intel-gfx mailing list