drm/i915: Fix gen6 (SNB) missed BLT ring interrupts.
Daniel J Blueman [Fri, 17 Jun 2011 18:32:19 +0000 (11:32 -0700)]
The failure appeared in dmesg as:

[drm:i915_hangcheck_ring_idle] *ERROR* Hangcheck timer elapsed... blt
ring idle [waiting on 35064155, at 35064155], missed IRQ?

This works around that problem on by making the blitter command
streamer write interrupt state to the Hardware Status Page when a
MI_USER_INTERRUPT command is decoded, which appears to force the seqno
out to memory before the interrupt happens.

v1->v2: Moved to prior interrupt handler installation and RMW flags as
per feedback.
v2->v3: Removed RMW of flags (by anholt)

Cc: stable@kernel.org
Signed-off-by: Daniel J Blueman <daniel.blueman@gmail.com>
Signed-off-by: Eric Anholt <eric@anholt.net>
Tested-by: Chris Wilson <chris@chris-wilson.co.uk> [v1]
Tested-by: Eric Anholt <eric@anholt.net> [v1,v3]
   (incidence of the bug with a testcase went from avg 2/1000 to
   0/12651 in the latest test run (plus more for v1))
Tested-by: Kenneth Graunke <kenneth@whitecape.org> [v1]
Tested-by: Robert Hooker <robert.hooker@canonical.com> [v1]
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=33394
Signed-off-by: Dave Airlie <airlied@redhat.com>

drivers/gpu/drm/i915/i915_irq.c

index b9fafe3..9e34a1a 100644 (file)
@@ -1740,6 +1740,16 @@ void ironlake_irq_preinstall(struct drm_device *dev)
                INIT_WORK(&dev_priv->rps_work, gen6_pm_rps_work);
 
        I915_WRITE(HWSTAM, 0xeffe);
+       if (IS_GEN6(dev)) {
+               /* Workaround stalls observed on Sandy Bridge GPUs by
+                * making the blitter command streamer generate a
+                * write to the Hardware Status Page for
+                * MI_USER_INTERRUPT.  This appears to serialize the
+                * previous seqno write out before the interrupt
+                * happens.
+                */
+               I915_WRITE(GEN6_BLITTER_HWSTAM, ~GEN6_BLITTER_USER_INTERRUPT);
+       }
 
        /* XXX hotplug from PCH */