[Intel-gfx] [PATCH 2/2] drm/i915: load a ring frequency scaling table

[Eric Anholt]

On Fri, 24 Jun 2011 11:13:14 -0700, Jesse Barnes <jbarnes at virtuousgeek.org> wrote:
> The ring frequency scaling table tells the PCU to treat certain GPU
> frequencies as if they were a given CPU frequency for purposes of
> scaling the ring frequency.  Normally the PCU will scale the ring
> frequency based on the CPU P-state, but with the table present, it will
> also take the GPU frequency into account.
> 
> The main downside of keeping the ring frequency high while the CPU is
> at a low frequency (or asleep altogether) is increased power
> consumption.  But then if you're keeping your GPU busy, you probably
> want the extra performance.

This one is working nicely here.  Maybe I didn't test enough with the
smash-to-3000 patch when I told you those results, but my 3 runs on a
fresh boot were 64, 70, and 72 fps (in order, so there seems to be some
sort of warming up to the system perhaps?  Even though the delay between
runs was sufficient for the GPU to clock back down).  73 is the fps I
see with the cpu busy loop.

Tested-by: Eric Anholt <eric at anholt.net>
Comments below, then Reviewed-by: Eric Anholt <eric at anholt.net>

> ---
>  drivers/gpu/drm/i915/i915_debugfs.c  |   43 +++++++++++++++++++++++++
>  drivers/gpu/drm/i915/i915_reg.h      |    4 ++-
>  drivers/gpu/drm/i915/i915_suspend.c  |    4 ++-
>  drivers/gpu/drm/i915/intel_display.c |   57 +++++++++++++++++++++++++++++++++-
>  drivers/gpu/drm/i915/intel_drv.h     |    1 +
>  5 files changed, 106 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
> index 4d46441..79394cd 100644
> --- a/drivers/gpu/drm/i915/i915_debugfs.c
> +++ b/drivers/gpu/drm/i915/i915_debugfs.c
> @@ -1123,6 +1123,48 @@ static int i915_emon_status(struct seq_file *m, void *unused)
>  	return 0;
>  }
>  
> +static int i915_ring_freq_table(struct seq_file *m, void *unused)
> +{
> +	struct drm_info_node *node = (struct drm_info_node *) m->private;
> +	struct drm_device *dev = node->minor->dev;
> +	drm_i915_private_t *dev_priv = dev->dev_private;
> +	int ret;
> +	int gpu_freq, ia_freq;
> +
> +	if (!IS_GEN6(dev)) {
> +		seq_printf(m, "unsupported on this chipset\n");
> +		return 0;
> +	}
> +
> +	ret = mutex_lock_interruptible(&dev->struct_mutex);
> +	if (ret)
> +		return ret;
> +
> +	gen6_gt_force_wake_get(dev_priv);
> +
> +	seq_printf(m, "GPU freq\tEffective CPU freq\n");
> +
> +	for (gpu_freq = dev_priv->min_delay; gpu_freq <= dev_priv->max_delay;
> +	     gpu_freq++) {
> +		I915_WRITE(GEN6_PCODE_DATA, gpu_freq);
> +		I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY |
> +			   GEN6_PCODE_READ_MIN_FREQ_TABLE);
> +		if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) &
> +			      GEN6_PCODE_READY) == 0, 10)) {
> +			DRM_ERROR("pcode write of freq table timed out\n");
> +			continue;
> +		}

s/write/read/

Might stick a note of units on the header.

> diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
> index 86a3ec1..3b22e12 100644
> --- a/drivers/gpu/drm/i915/intel_display.c
> +++ b/drivers/gpu/drm/i915/intel_display.c
> +void gen6_update_ring_freq(struct drm_i915_private *dev_priv)
> +{
> +	int min_freq = 15;
> +	int gpu_freq, ia_freq, max_ia_freq;
> +	int scaling_factor = 180;
> +
> +	max_ia_freq = cpufreq_quick_get_max(0);
> +	/* default to 3GHz if none found, PCU will ensure we don't go over */
> +	if (!max_ia_freq)
> +		max_ia_freq = 3000000;
> +
> +	/* Convert from kHz to MHz */
> +	max_ia_freq /= 1000;
> +
> +	mutex_lock(&dev_priv->dev->struct_mutex);
> +	gen6_gt_force_wake_get(dev_priv);
> +
> +	/*
> +	 * For each potential GPU frequency, load a ring frequency we'd like
> +	 * to use for memory access.  We do this by specifying the IA frequency
> +	 * the PCU should use as a reference to determine the ring frequency.
> +	 */
> +	for (gpu_freq = dev_priv->max_delay; gpu_freq >= dev_priv->min_delay;
> +	     gpu_freq--) {
> +		int diff = dev_priv->max_delay - gpu_freq;
> +
> +		/*
> +		 * For GPU frequencies less than 750MHz, just use the lowest
> +		 * ring freq.
> +		 */
> +		if (gpu_freq < min_freq)
> +			ia_freq = 800;
> +		else
> +			ia_freq = max_ia_freq - ((diff * scaling_factor) / 2);
> +		ia_freq = DIV_ROUND_CLOSEST(ia_freq, 100);

If the GPU has a wide enough clock range (diff large) and the CPU is low
enough clocked (max_ia_freq low now), could we end up with the ia_freq <
800, and would that be a bad thing?  In other words, should
scaling_factor be non-constant?

> +		I915_WRITE(GEN6_PCODE_DATA,
> +			   (ia_freq << GEN6_PCODE_FREQ_IA_RATIO_SHIFT) |
> +			   gpu_freq);
> +		I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY |
> +			   GEN6_PCODE_WRITE_MIN_FREQ_TABLE);
> +		if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) &
> +			      GEN6_PCODE_READY) == 0, 10)) {
> +			DRM_ERROR("pcode write of freq table timed out\n");
> +			continue;
> +		}
> +	}
> +
> +	gen6_gt_force_wake_put(dev_priv);
> +	mutex_unlock(&dev_priv->dev->struct_mutex);
> +}
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: not available
URL: <http://lists.freedesktop.org/archives/intel-gfx/attachments/20110627/80723699/attachment.sig>