Comment 17 for bug 899303

This is likely to be:

commit c501ae7f332cdaf42e31af30b72b4b66cbbb1604
Author: Chris Wilson <email address hidden>
Date: Wed Dec 14 13:57:23 2011 +0100

    drm/i915: Only clear the GPU domains upon a successful finish

    By clearing the GPU read domains before waiting upon the buffer, we run
    the risk of the wait being interrupted and the domains prematurely
    cleared. The next time we attempt to wait upon the buffer (after
    userspace handles the signal), we believe that the buffer is idle and so
    skip the wait.

    There are a number of bugs across all generations which show signs of an
    overly haste reuse of active buffers.

    Such as:

      https://bugs.freedesktop.org/show_bug.cgi?id=29046
      https://bugs.freedesktop.org/show_bug.cgi?id=35863
      https://bugs.freedesktop.org/show_bug.cgi?id=38952
      https://bugs.freedesktop.org/show_bug.cgi?id=40282
      https://bugs.freedesktop.org/show_bug.cgi?id=41098
      https://bugs.freedesktop.org/show_bug.cgi?id=41102
      https://bugs.freedesktop.org/show_bug.cgi?id=41284
      https://bugs.freedesktop.org/show_bug.cgi?id=42141

    A couple of those pre-date i915_gem_object_finish_gpu(), so may be
    unrelated (such as a wild write from a userspace command buffer), but
    this does look like a convincing cause for most of those bugs.

    Signed-off-by: Chris Wilson <email address hidden>
    Cc: <email address hidden>
    Reviewed-by: Daniel Vetter <email address hidden>
    Reviewed-by: Eugeni Dodonov <email address hidden>
    Signed-off-by: Daniel Vetter <email address hidden>