Quoting Andi Shyti (2023-12-21 19:08:22)
> The CCS mode involves assigning CCS engines to slices depending
> on the number of slices and the number of engines the user wishes
> to set.
>
> In this patch, the default CCS setting is established during the
> initial GT settings. It involves assigning only one CCS to all
> the slices.
>
> Based on a patch by Chris Wilson <chris.p.wilson@xxxxxxxxxxxxxxx>
> and Tejas Upadhyay <tejas.upadhyay@xxxxxxxxx>.
>
> Signed-off-by: Andi Shyti <andi.shyti@xxxxxxxxxxxxxxx>
> Cc: Chris Wilson <chris.p.wilson@xxxxxxxxxxxxxxx>
> Cc: Joonas Lahtinen <joonas.lahtinen@xxxxxxxxxxxxxxx>
> Cc: Niranjana Vishwanathapura <niranjana.vishwanathapura@xxxxxxxxx>
> Cc: Tejas Upadhyay <tejas.upadhyay@xxxxxxxxx>
<SNIP>
> +++ b/drivers/gpu/drm/i915/gt/intel_gt_types.h
> @@ -207,6 +207,26 @@ struct intel_gt {
> [MAX_ENGINE_INSTANCE + 1];
> enum intel_submission_method submission_method;
>
> + /*
> + * Track fixed mapping between CCS engines and compute slices.
> + *
> + * In order to w/a HW that has the inability to dynamically load
> + * balance between CCS engines and EU in the compute slices, we have to
> + * reconfigure a static mapping on the fly. We track the current CCS
> + * configuration (determined by inspection of the user's engine
> + * selection during execbuf) and compare it against the current
> + * CCS_MODE (which maps CCS engines to compute slices). If there is
> + * only a single engine selected, we can map it to all available
> + * compute slices for maximal single task performance (fast/narrow). If
> + * there are more then one engine selected, we have to reduce the
> + * number of slices allocated to each engine (wide/slow), fairly
> + * distributing the EU between the equivalent engines.
> + */
This comment is outdated as we don't consider execbuf but the sysfs
configuration.
Regards, Joonas
> + struct {
> + struct mutex mutex;
> + u32 mode;
> + } ccs;
> +
> /*
> * Default address space (either GGTT or ppGTT depending on arch).
> *
