diff mbox

[3.0-rc3,v2] i915: Fix gen6 (SNB) GPU stalling

Message ID 1308328920-2068-1-git-send-email-daniel.blueman@gmail.com
State New, archived
Headers show

Commit Message

Daniel J Blueman June 17, 2011, 4:42 p.m. UTC
Workaround Sandy Bridge GPU stalls by allowing the blitter command streamer
to write interrupt state to the Hardware Status Page when a MI_USER_INTERRUPT
command is decoded.

v1->v2: Moved to prior interrupt handler installation and RMW flags as per
feedback.

Eric, if you're happy, please sign-off and send to Dave for inclusion in -rc4
if possible. Further changes may be needed for -stable.

Signed-off-by: Daniel J Blueman <daniel.blueman@gmail.com>
---
 drivers/gpu/drm/i915/i915_irq.c |    7 +++++++
 1 files changed, 7 insertions(+), 0 deletions(-)

Comments

Eric Anholt June 17, 2011, 5:44 p.m. UTC | #1
On Sat, 18 Jun 2011 00:42:00 +0800, Daniel J Blueman <daniel.blueman@gmail.com> wrote:
> Workaround Sandy Bridge GPU stalls by allowing the blitter command streamer
> to write interrupt state to the Hardware Status Page when a MI_USER_INTERRUPT
> command is decoded.
> 
> v1->v2: Moved to prior interrupt handler installation and RMW flags as per
> feedback.

We shouldn't be RMWing.  We're the driver, we're supposed to know what
we want.  I'm backing that bit out, will do another round of testing,
and send it off.
Daniel J Blueman June 17, 2011, 6:20 p.m. UTC | #2
On 18 June 2011 01:44, Eric Anholt <eric@anholt.net> wrote:
> On Sat, 18 Jun 2011 00:42:00 +0800, Daniel J Blueman <daniel.blueman@gmail.com> wrote:
>> Workaround Sandy Bridge GPU stalls by allowing the blitter command streamer
>> to write interrupt state to the Hardware Status Page when a MI_USER_INTERRUPT
>> command is decoded.
>>
>> v1->v2: Moved to prior interrupt handler installation and RMW flags as per
>> feedback.
>
> We shouldn't be RMWing.  We're the driver, we're supposed to know what
> we want.  I'm backing that bit out, will do another round of testing,
> and send it off.

I still prefer directly setting the mask due to my earlier reasons.
Unmasking in the IRQ preinstall tests good here, so I'll send a v3
patch in the morning.

Thanks,
  Daniel
Eric Anholt June 17, 2011, 8:39 p.m. UTC | #3
On Sat, 18 Jun 2011 02:20:08 +0800, Daniel J Blueman <daniel.blueman@gmail.com> wrote:
> On 18 June 2011 01:44, Eric Anholt <eric@anholt.net> wrote:
> > On Sat, 18 Jun 2011 00:42:00 +0800, Daniel J Blueman <daniel.blueman@gmail.com> wrote:
> >> Workaround Sandy Bridge GPU stalls by allowing the blitter command streamer
> >> to write interrupt state to the Hardware Status Page when a MI_USER_INTERRUPT
> >> command is decoded.
> >>
> >> v1->v2: Moved to prior interrupt handler installation and RMW flags as per
> >> feedback.
> >
> > We shouldn't be RMWing.  We're the driver, we're supposed to know what
> > we want.  I'm backing that bit out, will do another round of testing,
> > and send it off.
> 
> I still prefer directly setting the mask due to my earlier reasons.
> Unmasking in the IRQ preinstall tests good here, so I'll send a v3
> patch in the morning.

Already tested and sent v3 out to lkml.
diff mbox

Patch

diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index b9fafe3..b7f11a5 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -1741,6 +1741,13 @@  void ironlake_irq_preinstall(struct drm_device *dev)
 
 	I915_WRITE(HWSTAM, 0xeffe);
 
+	if (IS_GEN6(dev))
+		/* workaround stalls observed on Sandy Bridge GPUs by
+		   allowing the blitter command streamer to generate a
+		   write to the Hardware Status Page for MI_USER_INTERRUPT */
+		I915_WRITE(GEN6_BLITTER_HWSTAM, I915_READ(GEN6_BLITTER_HWSTAM)
+			& ~GEN6_BLITTER_USER_INTERRUPT);
+
 	/* XXX hotplug from PCH */
 
 	I915_WRITE(DEIMR, 0xffffffff);