[Intel-gfx] [PATCH 4/5] drm/i915/guc: stop calling execlists_set_default_submission

Daniele Ceraolo Spurio daniele.ceraolospurio at intel.com
Wed Jan 6 17:22:26 UTC 2021

On 1/5/2021 7:09 PM, Chris Wilson wrote:
> Quoting Daniele Ceraolo Spurio (2021-01-06 02:38:09)
>> On 1/5/2021 5:02 PM, Chris Wilson wrote:
>>> Quoting Daniele Ceraolo Spurio (2021-01-05 23:19:46)
>>>> Initialize all required entries from guc_set_default_submission, instead
>>>> of calling the execlists function. The previously inherited setup has
>>>> been copied over from the execlist code and simplified by removing the
>>>> execlists submission-specific parts.
>>>> Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio at intel.com>
>>>> Cc: Matthew Brost <matthew.brost at intel.com>
>>>> Cc: John Harrison <john.c.harrison at intel.com>
>>>> ---
>>>> +       if (INTEL_GEN(engine->i915) >= 12)
>>>> +               engine->flags |= I915_ENGINE_HAS_RELATIVE_MMIO;
>>> We should probably lift this to intel_engine_setup().
>> GuC requires a more extensive usage of the relative mmio stuff, given
>> that it picks which engine to submit to when using virtual engines, so
>> I'm not sure if the support is going to look exactly the same for both
>> back-ends. There is an old patch from John H to rework the relative mmio
>> (https://patchwork.freedesktop.org/patch/332558/), which will have to
>> land in some form as part of the GuC submission re-enabling. I'd prefer
>> to wait for that to land before moving this flag.
> Whether or not LR* take the flag is independent of the submission
> backend. As to when to use the flag, I think that patch needs a lot more
> refinement.

it definitely does. I'll lift this up for now, we can reconsider later 
if needed.


> I915_ENGINE_HAS_RELATIVE_MMIO is the odd-one-out in the flags atm as it
> reflects HW capability.
> -Chris

More information about the Intel-gfx mailing list