[Intel-gfx] Recent regression on G965 with KMS

[Jesse Barnes]

On Friday 20 February 2009 13:52:41 Jesse Barnes wrote:
> On Friday 20 February 2009 00:09:48 Kenny Graunke wrote:
> > Hello!
> >
> > I just pulled from drm-intel-next today, and have had frequent but
> > intermittent hangs on my G965 with KMS.  They occur during ordinary use
> > (like trying to write this email!)  I do have compiz running, though.
> >
> > Once it goes haywire, my dmesg is completely full of stuff like this:
> > [drm:i915_gem_execbuffer] *ERROR* Failed to pin buffers -12
> > [drm:i915_gem_evict_something] *ERROR* inactive empty 1 request empty 1
> > flushing empty 1 [drm:i915_gem_object_bind_to_gtt] *ERROR* Failed to
> > evict a buffer -12 [drm:i915_gem_object_pin] *ERROR* Failure to bind:
> > -12<3>[drm:i915_gem_evict_something] *ERROR* inactive empty 1 request
> > empty 1 flushing empty 1
> >
> > (If it's helpful, the full dmesg is at
> > http://www.whitecape.org/files/intel-gfx/kms-dmesg for now.)
> >
> > I used git bisect, and it seems to have started with this commit:
> > commit deb1542d3b69727f7e55bc247796b7716ad95941
> >     drm/i915: Keep refs on the object over the lifetime of vmas for GTT
> > mmap.
> >
> > I'd be glad to test any patches or try and grab more information - let me
> > know how I can help.
>
> Hm this should only affect the GTT mapped case... lemme look where we might
> be using it...

So are you sure that it's just this patch causing you trouble?  The refcounts 
look ok to me with the latest bits, and only EXA uses GTT mapping anyway (and 
really it won't work well with KMS, you need UXA).  I'll try running my setup 
longer to see if I can reproduce it...

-- 
Jesse Barnes, Intel Open Source Technology Center