In i915_gem_idle, don't fail if we have a wedged chipset (this only

happens when waiting for the gpu), the reset function will at the very
least clean out all of our unwanted buffers, which is all we wanted to
do here ourselves. so we can continue with disabling the hardware and
cleaning up then return success.

Fixes a QUIESCE crash for krw when he had a wedged gpu (``it works, ok
krw@''). matthew@ hit a similar thing and this should fix it.

1 files changed, 5 insertions, 3 deletions

diff --git a/sys/dev/pci/drm/i915_drv.c b/sys/dev/pci/drm/i915_drv.c
index 58e6e11e0e3..737564eb6aa 100644
--- a/sys/dev/pci/drm/i915_drv.c
+++ b/sys/dev/pci/drm/i915_drv.c
@@ -3927,10 +3927,12 @@ i915_gem_idle(struct inteldrm_softc *dev_priv)
 	}
 
 	/*
-	 * If we're wedged, the workq will clear everything, else this will
-	 * empty out the lists for us.
+	 * To idle the gpu, flush anything pending then unbind the whole
+	 * shebang. If we're wedged, assume that the reset workq will clear
+	 * everything out and continue as normal.
 	 */
-	if ((ret = i915_gem_evict_everything(dev_priv, 1)) != 0 && ret != ENOSPC) {
+	if ((ret = i915_gem_evict_everything(dev_priv, 1)) != 0 &&
+	    ret != ENOSPC && ret != EIO) {
 		DRM_UNLOCK();
 		return (ret);
 	}