diff mbox

[CFT] drm/i915: Only set the down rps limit when at the loweset frequency

Message ID 1343294174-21811-1-git-send-email-daniel.vetter@ffwll.ch (mailing list archive)
State New, archived
Headers show

Commit Message

Daniel Vetter July 26, 2012, 9:16 a.m. UTC
The power docs say that when the gt leaves rc6, it is in the lowest
frequency and only about 25 usec later will switch to the frequency
selected in GEN6_RPNSWREQ. If the downclock limit expires in that
window and the down limit is set to the lowest possible frequency, the
hw will not send the down interrupt. Which leads to a too high gpu
clock and wasted power.

Chris Wilson already worked on this with

commit 7b9e0ae6da0a7eaf2680a1a788f08df123724f3b
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Sat Apr 28 08:56:39 2012 +0100

    drm/i915: Always update RPS interrupts thresholds along with
    frequency

but got the logic inverted: The current code set the down limit as
long as we haven't reached it. Instead of only once with reached the
lowest frequency.

Note that we can't always set the downclock limit to 0, because
otherwise the hw will keep on bugging us with downclock request irqs
once the lowest level is reached.

For similar reasons also always set the upclock limit, otherwise the
hw might poke us again with interrupts.

v2: Chris Wilson noticed that the limit reg is also computed in
sanitize_pm. To avoid duplication, extract the code into a common
function.

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
---
 drivers/gpu/drm/i915/intel_pm.c |   43 +++++++++++++++++++++++----------------
 1 file changed, 25 insertions(+), 18 deletions(-)

Comments

Chris Wilson July 26, 2012, 9:23 a.m. UTC | #1
On Thu, 26 Jul 2012 11:16:14 +0200, Daniel Vetter <daniel.vetter@ffwll.ch> wrote:
> The power docs say that when the gt leaves rc6, it is in the lowest
> frequency and only about 25 usec later will switch to the frequency
> selected in GEN6_RPNSWREQ. If the downclock limit expires in that
> window and the down limit is set to the lowest possible frequency, the
> hw will not send the down interrupt. Which leads to a too high gpu
> clock and wasted power.
> 
> Chris Wilson already worked on this with
> 
> commit 7b9e0ae6da0a7eaf2680a1a788f08df123724f3b
> Author: Chris Wilson <chris@chris-wilson.co.uk>
> Date:   Sat Apr 28 08:56:39 2012 +0100
> 
>     drm/i915: Always update RPS interrupts thresholds along with
>     frequency
> 
> but got the logic inverted: The current code set the down limit as
> long as we haven't reached it. Instead of only once with reached the
> lowest frequency.
> 
> Note that we can't always set the downclock limit to 0, because
> otherwise the hw will keep on bugging us with downclock request irqs
> once the lowest level is reached.
> 
> For similar reasons also always set the upclock limit, otherwise the
> hw might poke us again with interrupts.
> 
> v2: Chris Wilson noticed that the limit reg is also computed in
> sanitize_pm. To avoid duplication, extract the code into a common
> function.

Aye, that's the patch I wish I wrote.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

Though since I was clearly confused in the first place...
-Chris
Chris Wilson July 26, 2012, 9:37 a.m. UTC | #2
And then I remembered I was going to mention...

I spent yesterday evening trying to tempt the fail back to IVB, to no
avail. So I think the underlying cause was the sporadic read returning 0
which we promptly set to rplim with the original rmw sequence. Then we
falsely continued to warn during intel_sanitize_pm() due to the same
sporadic read failure. So I think we have painted all the elephants
here.
-Chris
diff mbox

Patch

diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c
index d0ce2a5..a537768 100644
--- a/drivers/gpu/drm/i915/intel_pm.c
+++ b/drivers/gpu/drm/i915/intel_pm.c
@@ -2267,21 +2267,33 @@  void ironlake_disable_drps(struct drm_device *dev)
 
 }
 
-void gen6_set_rps(struct drm_device *dev, u8 val)
+static u32 gen6_rps_limits(struct drm_i915_private *dev_priv, u8 val)
 {
-	struct drm_i915_private *dev_priv = dev->dev_private;
 	u32 limits;
 
 	limits = 0;
 	if (val >= dev_priv->max_delay)
 		val = dev_priv->max_delay;
-	else
-		limits |= dev_priv->max_delay << 24;
-
-	if (val <= dev_priv->min_delay)
+	limits |= dev_priv->max_delay << 24;
+
+	/* Only set the down limit when we've reached the lowest level to avoid
+	 * getting more interrupts, otherwise leave this clear. This prevents a
+	 * race in the hw when coming out of rc6: There's a tiny window where
+	 * the hw runs at the minimal clock before selecting the desired
+	 * frequency, if the down threshold expires in that window we will not
+	 * receive a down interrupt. */
+	if (val <= dev_priv->min_delay) {
 		val = dev_priv->min_delay;
-	else
 		limits |= dev_priv->min_delay << 16;
+	}
+
+	return limits;
+}
+
+void gen6_set_rps(struct drm_device *dev, u8 val)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	u32 limits = gen6_rps_limits(dev_priv, val);
 
 	if (val == dev_priv->cur_delay)
 		return;
@@ -3587,25 +3599,20 @@  void intel_init_clock_gating(struct drm_device *dev)
 static void gen6_sanitize_pm(struct drm_device *dev)
 {
 	struct drm_i915_private *dev_priv = dev->dev_private;
-	u32 limits, delay, old;
+	u32 limits, current_limits;
 
 	gen6_gt_force_wake_get(dev_priv);
 
-	old = limits = I915_READ(GEN6_RP_INTERRUPT_LIMITS);
+	current_limits = I915_READ(GEN6_RP_INTERRUPT_LIMITS);
 	/* Make sure we continue to get interrupts
 	 * until we hit the minimum or maximum frequencies.
 	 */
-	limits &= ~(0x3f << 16 | 0x3f << 24);
-	delay = dev_priv->cur_delay;
-	if (delay < dev_priv->max_delay)
-		limits |= (dev_priv->max_delay & 0x3f) << 24;
-	if (delay > dev_priv->min_delay)
-		limits |= (dev_priv->min_delay & 0x3f) << 16;
-
-	if (old != limits) {
+	limits = gen6_rps_limits(dev_priv, dev_priv->cur_delay);
+
+	if (current_limits != limits) {
 		/* Note that the known failure case is to read back 0. */
 		DRM_DEBUG_DRIVER("Power management discrepancy: GEN6_RP_INTERRUPT_LIMITS "
-				 "expected %08x, was %08x\n", limits, old);
+				 "expected %08x, was %08x\n", limits, current_limits);
 		I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, limits);
 	}