drm: Require __GFP_NOFAIL for the legacy drm_modeset_lock_all

To acquire all modeset locks requires a ww_ctx to be allocated. As this
is the legacy path and the allocation small, to reduce the changes
required (and complex untested error handling) to the legacy drivers, we
simply assume that the allocation succeeds. At present, it relies on the
too-small-to-fail rule, but syzbot found that by injecting a failure
here we would hit the WARN. Document that this allocation must succeed
with __GFP_NOFAIL.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20171031115535.15166-1-chris@chris-wilson.co.uk

drivers/gpu/drm/drm_modeset_lock.c

@@ -93,7 +93,7 @@ void drm_modeset_lock_all(struct drm_device *dev)
 	struct drm_modeset_acquire_ctx *ctx;
 	int ret;
 
-	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
+	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL | __GFP_NOFAIL);
 	if (WARN_ON(!ctx))
 		return;