[Intel-gfx] [PATCH 3/7] drm/i915/icl: Apply recommended rc6 idle hysteresis

Chris Wilson chris at chris-wilson.co.uk
Tue Apr 9 16:16:53 UTC 2019


Quoting Mika Kuoppala (2019-04-09 17:13:06)
> Use a recommended idle hysteresis for media and render powergates.
> 
> References: bspec#52070
> Signed-off-by: Mika Kuoppala <mika.kuoppala at linux.intel.com>
> ---
>  drivers/gpu/drm/i915/intel_pm.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c
> index 30ef507b88a4..b9be9ea5fc18 100644
> --- a/drivers/gpu/drm/i915/intel_pm.c
> +++ b/drivers/gpu/drm/i915/intel_pm.c
> @@ -7170,8 +7170,8 @@ static void gen11_enable_rc6(struct drm_i915_private *dev_priv)
>          * conservative, we have to factor in a context switch on top (due
>          * to ksoftirqd).
>          */
> -       I915_WRITE(GEN9_MEDIA_PG_IDLE_HYSTERESIS, 250);
> -       I915_WRITE(GEN9_RENDER_PG_IDLE_HYSTERESIS, 250);
> +       I915_WRITE(GEN9_MEDIA_PG_IDLE_HYSTERESIS, 25);
> +       I915_WRITE(GEN9_RENDER_PG_IDLE_HYSTERESIS, 25);

We were using higher than recommended for the simple reason of not
allowing it to powergate while signaling between engines. We are much
faster now (though be sure to disable semaphores to put us in worse
case) and since we use one value, we need to measure on the slow
platform.

Anyway, just pointing out there was a reason for a relatively large
hysteresis.
-Chris


More information about the Intel-gfx mailing list