[Intel-gfx] Revised error detection/handling patchset (3nd pass)

[Ben Gamari]

Hey all,

Ickle's keen eye found that there could be a variety of race conditions that
could be hit by non-atomic access to the wedged flag. This latest version makes
the flag an atomic_t and restructures the i915_handle_error() interface to make
the atomic nature of the flag a little more apparent.  He also noted that I
fail to account for the case in which the request list is empty in the
hangcheck timer, which has also been fixed. Thanks ickle!

- Ben