Commit 053d7f24 authored by Eric Anholt's avatar Eric Anholt Committed by Dave Airlie

i915: Enable IMR passthrough of vblank events before enabling it in pipestat.

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: default avatarEric Anholt <eric@anholt.net>
Acked-by: default avatarKeith Packard <keithp@keithp.com>
Signed-off-by: default avatarDave Airlie <airlied@redhat.com>
parent 35ad68c1
......@@ -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;
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment