i915 blocked for more than 120 seconds.

[Chris Wilson]

On Wed, 29 Sep 2010 11:36:22 +0200, Toralf Förster <toralf.foerster at gmx.de> wrote:
> Hello,
> 
> today I run at the latest 2.6.35.6 kernel few apps like Lotus Notes under wine, amarok and firefox
> and in parallel a "git bisect" when I got this :
> 
> n22 ~ # grep 2010-09-29T11:21 /var/log/messages | grep kernel
> 2010-09-29T11:21:36.984+02:00 n22 kernel: INFO: task i915:868 blocked for more than 120 seconds.
> 2010-09-29T11:21:36.984+02:00 n22 kernel: "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
> 2010-09-29T11:21:36.984+02:00 n22 kernel: i915          D ffa47280     0   868      2 0x00000000

It's a mysterious lost interrupt (or a touch of incoherency around the
dword seqno write). The pipe control dance is meant to ensure the write to
memory is truly flushed before the interrupt fires. [T400 is ironlake?]

cat /sys/kernel/debug/dri/0/i915_gem_interrupt

That should show whether the seqno was ever updated. We could extend that
file by recording the number of User Interrupts since waiting to see if we
miss the interrupt.

drm-intel-next contains patches to hopefully workaround the hang. So
hopefully your machine will no longer die whilst we uncover the root
cause.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre