[Intel-gfx] Reporting GPU hung bugs

[Chris Wilson]

On Tue, 9 Feb 2010 21:03:37 +0100, Geir Ove Myhr <gomyhr at gmail.com> wrote:
> So I take it that this (together with Sven's reply) to mean that dumps
> taken with kernels 2.6.32 and 2.6.33 are useless. We also have a lot
> of dumps from 2.6.31, but I suppose the kernel is too old in that
> case. Right?

Well that kernel is also likely coupled with an old userspace, and
considering our [or is it my? :(] bug flux in recent history there is a
good likelihood that such bugs are no longer reproducible anyway.
 
> > Please test:
> >  drm/i915: Record batch buffer following GPU error
> 
> I read through the patch, but I don't speak drm/i915 very well. Where
> does it record the batchbuffer dump? I can build some Ubuntu test
> kernels with the patch so that people with freeze problems can capture
> this information if this is useful for you.

Sorry, this essentially tries to capture all relevent info for a gpu
hang in /sys/kernel/debug/dri/.../i915_error_state. My plan is to
continue to enhance the amount of GPU state captured at the time of the
error such that we can reliably deduce the cause with just this file and
the logs. But there will always be bugs that are missed, and presumably
require novel information to be reported anyhow...
-ickle

-- 
Chris Wilson, Intel Open Source Technology Centre