[Intel-gfx] [RFC v2] drm/i915: Engine discovery query

Chris Wilson chris at chris-wilson.co.uk
Wed Apr 11 11:15:21 UTC 2018


Quoting Tvrtko Ursulin (2018-04-11 12:10:05)
> 
> On 11/04/2018 11:27, Chris Wilson wrote:
> > Quoting Tvrtko Ursulin (2018-04-11 10:46:31)
> >> +/**
> >> + * struct drm_i915_engine_info
> >> + *
> >> + * Describes one engine known to the driver, whether or not is physical engine
> >> + * only, or it can also be targetted from userspace, and what are its
> >> + * capabilities where applicable.
> >> + */
> >> +struct drm_i915_engine_info {
> >> +       /** Engine flags. */
> >> +       __u64 flags;
> >> +#define I915_ENGINE_FLAG_PHYSICAL      (1 << 0)
> >> +#define I915_ENGINE_FLAG_ABI           (1 << 1)
> >> +
> >> +       /** Engine class as in enum drm_i915_gem_engine_class. */
> >> +       __u16 class;
> >> +
> >> +       /** Engine instance number. */
> >> +       __u16 instance;
> >> +
> >> +       __u32 rsvd0;
> >> +
> >> +       /** Capabilities of this engine. */
> >> +       __u64 capabilities;
> >> +#define I915_VCS_CLASS_CAPABILITY_HEVC (1 << 0)
> >> +
> >> +       __u64 rsvd1[2];
> >> +};
> >> +
> >> +/**
> >> + * struct drm_i915_query_engine_info
> >> + *
> >> + * Engine info query enumerates all the engines known to the driver returning
> >> + * the array of struct drm_i915_engine_info structures.
> >> + */
> >> +struct drm_i915_query_engine_info {
> >> +       /** Number of struct drm_i915_engine_info structs following. */
> >> +       __u32 num_engines;
> >> +
> >> +       /** MBZ */
> >> +       __u32 rsvd[3];
> > 
> > Ok, I can understand having some reserved fields before the [], but
> > within struct drm_i915_engine_info isn't the size and layout determined
> > by flags? (Thinking about the tight protocol packing of the 80s and
> > slightly less tight packing for the 90s ;) Trade-off between ease of
> > decoder and running out of space?
> 
> How did that old saying go, was it "When in doubt, pad it out."? :)
> 
> But I think it is good to have some reserved fields rather than rely on 
> flags to extend the size of the structure. In any case I don't see a 
> significant downside to reserved fields.

The only downside is running and starting with an new query-id. We've
only 64b of them!
-Chris


More information about the Intel-gfx mailing list