[Intel-gfx] [drm:i915_gem_mmap_gtt_ioctl] *ERROR* Attempting to mmap a purgeable buffer

[Alexey Fisher]

Am Mittwoch, den 20.10.2010, 10:15 +0100 schrieb Chris Wilson:
> On Wed, 20 Oct 2010 11:07:20 +0200, Alexey Fisher <bug-track at fisher-privat.net> wrote:
> > Hallo all,
> > 
> > I get fallowing error in my log:
> > [ 6657.465805] [drm:i915_gem_mmap_gtt_ioctl] *ERROR* Attempting to mmap
> > a purgeable buffer
> > [ 6657.465851] [drm:i915_gem_mmap_gtt_ioctl] *ERROR* Attempting to mmap
> > a purgeable buffer
> > [ 6657.465907] [drm:i915_gem_mmap_gtt_ioctl] *ERROR* Attempting to mmap
> > a purgeable buffer
> > 
> > das it mean any thing for you? do i need to bisect it?
> > I use current 2.6.36-rc8-01141-ga5ffe0a from intel drm next.
> 
> It's warning that userspace is breaking the rules and trying to use a
> buffer it has declared no longer in use. The challenge is to find the
> scenario that is triggering this *userspace* bug.
> 
> Currently, purgeable is usually only set on the userspace bo cache, so
> this means that we are continuing to use a freed buffer.  Disabling (or
> marking the cache as NO_ACCESS) and valgrinding would be a useful step.
> 
> Or just kill the userspace cache and implement a kernel uncached page
> cache... Less code, more performance, everyone is happy...

According to kernel logs, and time of this logs, this is some where
bethwene the bug in firefox and compiz. At the time I produced this
video for bugreport:
http://videobin.org/+281/2il.html

and switched compiz off and on.
The problem, if i reproduce this steps i get other log:

[14396.675124] [drm:i915_gem_object_bind_to_gtt] *ERROR* Attempting to
bind a purgeable object
[14473.046399] [drm:i915_gem_object_bind_to_gtt] *ERROR* Attempting to
bind a purgeable object