[Intel-gfx] [PATCH 1/3] drm/i915/execlists: Force write serialisation into context image vs execution

Chris Wilson chris at chris-wilson.co.uk
Thu Nov 8 12:26:53 UTC 2018


Quoting Mika Kuoppala (2018-11-08 12:13:42)
> Chris Wilson <chris at chris-wilson.co.uk> writes:
> 
> > Quoting Mika Kuoppala (2018-11-08 12:00:39)
> >> Chris Wilson <chris at chris-wilson.co.uk> writes:
> >> > +     /*
> >> > +      * Make sure the context image is complete before we submit it to HW.
> >> > +      *
> >> > +      * Ostensibly, writes (including the WCB) should be flushed prior to
> >> > +      * an uncached write such as our mmio register access, the empirical
> >> > +      * evidence (esp. on Braswell) suggests that the WC write into memory
> >> > +      * may not be visible to the HW prior to the completion of the UC
> >> > +      * register write and that we may begin execution from the context
> >> > +      * before its image is complete leading to invalid PD chasing.
> >> > +      */
> >> > +     wmb();
> >> 
> >> Let's put it into use and gather more evidence.
> >> 
> >> Reviewed-by: Mika Kuoppala <mika.kuoppala at linux.intel.com>
> >
> > Aye. Sure about r-b? I'm quite happy to take an a-b since we're just
> > postulating to gather evidence.
> 
> Agreed that a-b is more accurate. r-b would indicate I know what the
> heck is going on there under the hood.

And pushed (this patch) to see if the next few months do quieten down.
-Chris


More information about the Intel-gfx mailing list