[Intel-gfx] [PATCH 2/6] drm/i915: Don't wait for Punit after each freq change on VLV

[Jesse Barnes]

On Tue, 25 Jun 2013 19:21:02 +0300
ville.syrjala at linux.intel.com wrote:

> From: Ville Syrjälä <ville.syrjala at linux.intel.com>
> 
> It seems that even though Punit reports the frequency change to have
> been completed, it still reports the old frequency in the status
> register for some time.
> 
> So rather than polling for Punit to complete the frequency change after
> each request, poll before. This gets rid of the spurious "Punit overrode
> GPU freq" messages.
> 
> This also lets us continue working while Punit is performing the actual
> frequency change. As a result, openarena demo088-test1 timedemo average
> fps is increased by ~5 fps, and the slowest frame duration is reduced
> by ~25%.
> 
> The sysfs cur_freq file always reads the current frequency from Punit
> anyway, so having rps.cur_delay be slightly off at times doesn't matter.
> 
> Signed-off-by: Ville Syrjälä <ville.syrjala at linux.intel.com>
> ---
>  drivers/gpu/drm/i915/intel_pm.c | 53 +++++++++++++++++++++++++++--------------
>  1 file changed, 35 insertions(+), 18 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c
> index 48a3162..6dbcad7 100644
> --- a/drivers/gpu/drm/i915/intel_pm.c
> +++ b/drivers/gpu/drm/i915/intel_pm.c
> @@ -3066,17 +3066,49 @@ void gen6_set_rps(struct drm_device *dev, u8 val)
>  	trace_intel_gpu_freq_change(val * 50);
>  }
>  
> +/*
> + * Wait until the previous freq change has completed,
> + * or the timeout elapsed, and then update our notion
> + * of the current GPU frequency.
> + */
> +static void vlv_update_rps_cur_delay(struct drm_i915_private *dev_priv)
> +{
> +	unsigned long timeout = jiffies + msecs_to_jiffies(10);
> +	u32 pval;
> +
> +	WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
> +
> +	do {
> +		pval = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
> +		if (time_after(jiffies, timeout)) {
> +			DRM_DEBUG_DRIVER("timed out waiting for Punit\n");
> +			break;
> +		}
> +		udelay(10);
> +	} while (pval & 1);
> +
> +	pval >>= 8;
> +
> +	if (pval != dev_priv->rps.cur_delay)
> +		DRM_DEBUG_DRIVER("Punit overrode GPU freq: %d MHz (%u) requested, but got %d Mhz (%u)\n",
> +				 vlv_gpu_freq(dev_priv->mem_freq, dev_priv->rps.cur_delay),
> +				 dev_priv->rps.cur_delay,
> +				 vlv_gpu_freq(dev_priv->mem_freq, pval), pval);
> +
> +	dev_priv->rps.cur_delay = pval;
> +}
> +
>  void valleyview_set_rps(struct drm_device *dev, u8 val)
>  {
>  	struct drm_i915_private *dev_priv = dev->dev_private;
> -	unsigned long timeout = jiffies + msecs_to_jiffies(10);
>  	u32 limits = gen6_rps_limits(dev_priv, &val);
> -	u32 pval;
>  
>  	WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
>  	WARN_ON(val > dev_priv->rps.max_delay);
>  	WARN_ON(val < dev_priv->rps.min_delay);
>  
> +	vlv_update_rps_cur_delay(dev_priv);
> +
>  	DRM_DEBUG_DRIVER("GPU freq request from %d MHz (%u) to %d MHz (%u)\n",
>  			 vlv_gpu_freq(dev_priv->mem_freq,
>  				      dev_priv->rps.cur_delay),
> @@ -3088,27 +3120,12 @@ void valleyview_set_rps(struct drm_device *dev, u8 val)
>  
>  	vlv_punit_write(dev_priv, PUNIT_REG_GPU_FREQ_REQ, val);
>  
> -	do {
> -		pval = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
> -		if (time_after(jiffies, timeout)) {
> -			DRM_DEBUG_DRIVER("timed out waiting for Punit\n");
> -			break;
> -		}
> -		udelay(10);
> -	} while (pval & 1);
> -
> -	pval = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
> -	if ((pval >> 8) != val)
> -		DRM_DEBUG_DRIVER("Punit overrode GPU freq: %d MHz (%u) requested, but got %d Mhz (%u)\n",
> -				 vlv_gpu_freq(dev_priv->mem_freq, val), val,
> -				 vlv_gpu_freq(dev_priv->mem_freq, pval >> 8), pval >> 8);
> -
>  	/* Make sure we continue to get interrupts
>  	 * until we hit the minimum or maximum frequencies.
>  	 */
>  	I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, limits);
>  
> -	dev_priv->rps.cur_delay = pval >> 8;
> +	dev_priv->rps.cur_delay = val;
>  
>  	trace_intel_gpu_freq_change(vlv_gpu_freq(dev_priv->mem_freq, val));
>  }

Hiding some of the Punit latency is nice...

Reviewed-by: Jesse Barnes <jbarnes at virtuousgeek.org>


-- 
Jesse Barnes, Intel Open Source Technology Center