drm/i915: update last_vblank when disabling the power well
authorPaulo Zanoni <paulo.r.zanoni@intel.com>
Tue, 23 Jul 2013 13:48:11 +0000 (10:48 -0300)
committerDaniel Vetter <daniel.vetter@ffwll.ch>
Wed, 7 Aug 2013 09:57:06 +0000 (11:57 +0200)
The DRM layer keeps track of our vblanks and it assumes our vblank
counters only go back to zero when they overflow. The problem is that
when we disable the power well our counters also go to zero, but it
doesn't mean they did overflow. So on this patch we grab the lock and
update last_vblank so the DRM layer won't think our counters
overflowed.

This patch fixes the following intel-gpu-tools test:
./kms_flip --run-subtest blocking-absolute-wf_vblank

Regression introduced by the following commit:

commit bf51d5e2cda5d36d98e4b46ac7fca9461e512c41
Author: Paulo Zanoni <paulo.r.zanoni@intel.com>
Date:   Wed Jul 3 17:12:13 2013 -0300
    drm/i915: switch disable_power_well default value to 1

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=66808
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
[danvet: Added a comment that this might be better done in
drm_vblank_post_modeset in general.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drivers/gpu/drm/i915/intel_pm.c

index f895d1508df8fbb9646ed064f1d2ce8244d4f4d0..b0e4a0bd1313c0dafac84df1b6af56e7ccced950 100644 (file)
@@ -5063,8 +5063,26 @@ static void __intel_set_power_well(struct drm_device *dev, bool enable)
                }
        } else {
                if (enable_requested) {
+                       unsigned long irqflags;
+                       enum pipe p;
+
                        I915_WRITE(HSW_PWR_WELL_DRIVER, 0);
+                       POSTING_READ(HSW_PWR_WELL_DRIVER);
                        DRM_DEBUG_KMS("Requesting to disable the power well\n");
+
+                       /*
+                        * After this, the registers on the pipes that are part
+                        * of the power well will become zero, so we have to
+                        * adjust our counters according to that.
+                        *
+                        * FIXME: Should we do this in general in
+                        * drm_vblank_post_modeset?
+                        */
+                       spin_lock_irqsave(&dev->vbl_lock, irqflags);
+                       for_each_pipe(p)
+                               if (p != PIPE_A)
+                                       dev->last_vblank[p] = 0;
+                       spin_unlock_irqrestore(&dev->vbl_lock, irqflags);
                }
        }
 }