[Intel-gfx] [PATCH v2] drm/i915: Limit C-states when waiting for the active request

[Chris Wilson]

Quoting Chris Wilson (2018-07-30 13:14:09)
> If we are waiting for the currently executing request, we have a good
> idea that it will be completed in the very near future and so want to
> cap the CPU_DMA_LATENCY to ensure that we wake up the client quickly.
> 
> v2: Not allowed to block in kmalloc after setting TASK_INTERRUPTIBLE.
> 
> Testcase: igt/gem_sync/store-default
> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> Cc: Tvrtko Ursulin <tvrtko.ursulin at linux.intel.com>
> Cc: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
> Cc: Eero Tamminen <eero.t.tamminen at intel.com>
> Cc: Francisco Jerez <currojerez at riseup.net>
> ---
>  drivers/gpu/drm/i915/i915_request.c | 29 +++++++++++++++++++++++++++++
>  1 file changed, 29 insertions(+)
> 
> diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
> index 5c2c93cbab12..de907150664e 100644
> --- a/drivers/gpu/drm/i915/i915_request.c
> +++ b/drivers/gpu/drm/i915/i915_request.c
> @@ -1258,6 +1258,19 @@ static bool __i915_wait_request_check_and_reset(struct i915_request *request)
>         return true;
>  }
>  
> +struct pm_qos {
> +       struct pm_qos_request req;
> +       struct work_struct work;
> +};
> +
> +static void pm_qos_del(struct work_struct *work)
> +{
> +       struct pm_qos *pm_qos = container_of(work, typeof(*pm_qos), work);
> +
> +       pm_qos_remove_request(&pm_qos->req);
> +       kfree(pm_qos);
> +}
> +
>  /**
>   * i915_request_wait - wait until execution of request has finished
>   * @rq: the request to wait upon
> @@ -1286,6 +1299,7 @@ long i915_request_wait(struct i915_request *rq,
>         wait_queue_head_t *errq = &rq->i915->gpu_error.wait_queue;
>         DEFINE_WAIT_FUNC(reset, default_wake_function);
>         DEFINE_WAIT_FUNC(exec, default_wake_function);
> +       struct pm_qos *pm_qos = NULL;
>         struct intel_wait wait;
>  
>         might_sleep();
> @@ -1363,6 +1377,19 @@ long i915_request_wait(struct i915_request *rq,
>                         break;
>                 }
>  
> +               if (!pm_qos &&
> +                   i915_seqno_passed(intel_engine_get_seqno(rq->engine),
> +                                     wait.seqno - 1)) {
> +                       pm_qos = kzalloc(sizeof(*pm_qos),
> +                                        GFP_NOWAIT | __GFP_NOWARN);
> +                       if (pm_qos) {
> +                               pm_qos_add_request(&pm_qos->req,
> +                                                  PM_QOS_CPU_DMA_LATENCY,
> +                                                  50);

Ho hum, it should also then complain about the blocking notifier
here. Let's see how observant it is...
-Chris