i915: Enable IMR passthrough of vblank events before enabling it in pipestat.
authorEric Anholt <eric@anholt.net>
Fri, 17 Oct 2008 22:41:26 +0000 (15:41 -0700)
committerDave Airlie <airlied@redhat.com>
Thu, 23 Oct 2008 03:42:27 +0000 (13:42 +1000)
Otherwise, if we lost the race, the pipestat bit would be set without being
reflected in IIR, and we would never clear the pipestat bit so the pipe
event would never be generated again, and all vblank waits would time out.

Signed-off-by: Eric Anholt <eric@anholt.net>
Acked-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
drivers/gpu/drm/i915/i915_irq.c

index 2cd853b2c5629fb133f968bf28e64e633432242a..ca55c40353a5cce744f440fcf1bdad3e3009190e 100644 (file)
@@ -605,6 +605,15 @@ int i915_enable_vblank(struct drm_device *dev, int plane)
        }
 
        spin_lock_irqsave(&dev_priv->user_irq_lock, irqflags);
+       /* Enabling vblank events in IMR comes before PIPESTAT write, or
+        * there's a race where the PIPESTAT vblank bit gets set to 1, so
+        * the OR of enabled PIPESTAT bits goes to 1, so the PIPExEVENT in
+        * ISR flashes to 1, but the IIR bit doesn't get set to 1 because
+        * IMR masks it.  It doesn't ever get set after we clear the masking
+        * in IMR because the ISR bit is edge, not level-triggered, on the
+        * OR of PIPESTAT bits.
+        */
+       i915_enable_irq(dev_priv, interrupt);
        pipestat = I915_READ(pipestat_reg);
        if (IS_I965G(dev))
                pipestat |= PIPE_START_VBLANK_INTERRUPT_ENABLE;
@@ -615,7 +624,6 @@ int i915_enable_vblank(struct drm_device *dev, int plane)
                     PIPE_VBLANK_INTERRUPT_STATUS);
        I915_WRITE(pipestat_reg, pipestat);
        (void) I915_READ(pipestat_reg); /* Posting read */
-       i915_enable_irq(dev_priv, interrupt);
        spin_unlock_irqrestore(&dev_priv->user_irq_lock, irqflags);
 
        return 0;