drm/i915: Ignore a hung GPU when flushing the framebuffer prior to a switch
authorChris Wilson <chris@chris-wilson.co.uk>
Fri, 11 Feb 2011 20:47:45 +0000 (20:47 +0000)
committerChris Wilson <chris@chris-wilson.co.uk>
Wed, 16 Feb 2011 09:25:17 +0000 (09:25 +0000)
If the gpu is hung, then whatever was inside the render cache is lost
and there is little point waiting for it. Or complaining if we see an
EIO or EAGAIN instead. So, if the GPU is indeed in its death throes when
we need to rewrite the registers for a new framebuffer, just ignore the
error and proceed with the update.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
drivers/gpu/drm/i915/intel_display.c

index 3b006536b3d23cda6c771de53f423345f35b9c89..dcb821737f381de71e89ea1b037c69056ad1d6e5 100644 (file)
@@ -1630,19 +1630,19 @@ intel_pipe_set_base(struct drm_crtc *crtc, int x, int y,
                struct drm_i915_gem_object *obj = to_intel_framebuffer(old_fb)->obj;
 
                wait_event(dev_priv->pending_flip_queue,
+                          atomic_read(&dev_priv->mm.wedged) ||
                           atomic_read(&obj->pending_flip) == 0);
 
                /* Big Hammer, we also need to ensure that any pending
                 * MI_WAIT_FOR_EVENT inside a user batch buffer on the
                 * current scanout is retired before unpinning the old
                 * framebuffer.
+                *
+                * This should only fail upon a hung GPU, in which case we
+                * can safely continue.
                 */
                ret = i915_gem_object_flush_gpu(obj, false);
-               if (ret) {
-                       i915_gem_object_unpin(to_intel_framebuffer(crtc->fb)->obj);
-                       mutex_unlock(&dev->struct_mutex);
-                       return ret;
-               }
+               (void) ret;
        }
 
        ret = intel_pipe_set_base_atomic(crtc, crtc->fb, x, y,