[Intel-gfx] [PATCH 1/2] drm/i915/guc: Support engine busy stats

[Chris Wilson]

Quoting Tvrtko Ursulin (2018-02-22 06:07:32)
> From: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
> 
> Place context in/out hooks into the GuC backend, when contexts are
> assigned to ports, and removed from them, in order to be able to
> provide engine busy stats in GuC mode.
> 
> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
> Testcase: igt/perf_pmu/busy-accuracy-*-*
> ---
>  drivers/gpu/drm/i915/intel_guc_submission.c | 13 ++++++++++---
>  1 file changed, 10 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/intel_guc_submission.c b/drivers/gpu/drm/i915/intel_guc_submission.c
> index 649113c7a3c2..8e99f8fd6da2 100644
> --- a/drivers/gpu/drm/i915/intel_guc_submission.c
> +++ b/drivers/gpu/drm/i915/intel_guc_submission.c
> @@ -673,6 +673,7 @@ static void guc_dequeue(struct intel_engine_cs *engine)
>         struct intel_engine_execlists * const execlists = &engine->execlists;
>         struct execlist_port *port = execlists->port;
>         struct i915_request *last = NULL;
> +       struct i915_gem_context *last_ctx = NULL;
>         const struct execlist_port * const last_port =
>                 &execlists->port[execlists->port_mask];
>         bool submit = false;
> @@ -718,8 +719,13 @@ static void guc_dequeue(struct intel_engine_cs *engine)
>                                         goto done;
>                                 }
>  
> -                               if (submit)
> +                               if (submit) {
>                                         port_assign(port, last);
> +                                       if (last->ctx != last_ctx) {
> +                                               intel_engine_context_in(last->engine);
> +                                               last_ctx = last->ctx;
> +                                       }
> +                               }
>                                 port++;
>                         }
>  
> @@ -741,6 +747,8 @@ static void guc_dequeue(struct intel_engine_cs *engine)
>         execlists->first = rb;
>         if (submit) {
>                 port_assign(port, last);
> +               if (last->ctx != last_ctx)
> +                       intel_engine_context_in(last->engine);
>                 execlists_set_active(execlists, EXECLISTS_ACTIVE_USER);
>                 guc_submit(engine);
>         }
> @@ -763,6 +771,7 @@ static void guc_submission_tasklet(unsigned long data)
>  
>         rq = port_request(&port[0]);
>         while (rq && i915_request_completed(rq)) {
> +               intel_engine_context_out(rq->engine);

If we only emit context_in once for 2 consecutive rq with the same
context, we would need to do the same for context_out.

Will be interesting to see if this explodes, or we may need yet another
test :)
-Chris