drm/i915: Update write_domains on active list after flush.

Before changing the status of a buffer with a pending write we will await
upon a new flush for that buffer. So we can take advantage of any flushes
posted whilst the buffer is active and pending processing by the GPU, by
clearing its write_domain and updating its last_rendering_seqno -- thus
saving a potential flush in deep queues and improves flushing behaviour
upon eviction for both GTT space and fences.

In order to reduce the time spent searching the active list for matching
write_domains, we move those to a separate list whose elements are
the buffers belong to the active/flushing list with pending writes.

Orignal patch by Chris Wilson <[e-mail]>, forward-ported
by me.

In addition to better performance, this also fixes a real bug. Before
this changes, i915_gem_evict_everything didn't work as advertised. When
the gpu was actually busy and processing request, the flush and subsequent
wait would not move active and dirty buffers to the inactive list, but
just to the flushing list. Which triggered the BUG_ON at the end of this
function. With the more tight dirty buffer tracking, all currently busy and
dirty buffers get moved to the inactive list by one i915_gem_flush operation.

I've left the BUG_ON I've used to prove this in there.

References:
Bug 25911 - 2.10.0 causes kernel oops and system hangs
http://bugs.freedesktop.org/show_bug.cgi?id=25911

Bug 26101 - [i915] xf86-video-intel 2.10.0 (and git) triggers kernel oops
within seconds after login
http://bugs.freedesktop.org/show_bug.cgi?id=26101

Signed-off-by: Daniel Vetter <[e-mail]>
Signed-off-by: Chris Wilson <[e-mail]>
Tested-by: Adam Lantos <[e-mail]>
Cc: [e-mail]
Signed-off-by: Eric Anholt <[e-mail]>

drivers/gpu/drm/i915/i915_drv.h

···
 		struct list_head flushing_list;
 
 		/**
+		 * List of objects currently pending a GPU write flush.
+		 *
+		 * All elements on this list will belong to either the
+		 * active_list or flushing_list, last_rendering_seqno can
+		 * be used to differentiate between the two elements.
+		 */
+		struct list_head gpu_write_list;
+
+		/**
 		 * LRU list of objects which are not in the ringbuffer and
 		 * are ready to unbind, but are still in the GTT.
 		 *
···
 
 	/** This object's place on the active/flushing/inactive lists */
 	struct list_head list;
+	/** This object's place on GPU write list */
+	struct list_head gpu_write_list;
 
 	/** This object's place on the fenced object LRU */
 	struct list_head fence_list;

drivers/gpu/drm/i915/i915_gem.c

···
 	else
 		list_move_tail(&obj_priv->list, &dev_priv->mm.inactive_list);
 
+	BUG_ON(!list_empty(&obj_priv->gpu_write_list));
+
 	obj_priv->last_rendering_seqno = 0;
 	if (obj_priv->active) {
 		obj_priv->active = 0;
···
 		struct drm_i915_gem_object *obj_priv, *next;
 
 		list_for_each_entry_safe(obj_priv, next,
-					 &dev_priv->mm.flushing_list, list) {
+					 &dev_priv->mm.gpu_write_list,
+					 gpu_write_list) {
 			struct drm_gem_object *obj = obj_priv->obj;
 
 			if ((obj->write_domain & flush_domains) ==
···
 				uint32_t old_write_domain = obj->write_domain;
 
 				obj->write_domain = 0;
+				list_del_init(&obj_priv->gpu_write_list);
 				i915_gem_object_move_to_active(obj, seqno);
 
 				trace_i915_gem_object_change_domain(obj,
···
 i915_gem_evict_everything(struct drm_device *dev)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
-	uint32_t seqno;
 	int ret;
+	uint32_t seqno;
 	bool lists_empty;
 
 	spin_lock(&dev_priv->mm.active_list_lock);
···
 	ret = i915_wait_request(dev, seqno);
 	if (ret)
 		return ret;
+
+	BUG_ON(!list_empty(&dev_priv->mm.flushing_list));
 
 	ret = i915_gem_evict_from_inactive_list(dev);
 	if (ret)
···
 	old_write_domain = obj->write_domain;
 	i915_gem_flush(dev, 0, obj->write_domain);
 	seqno = i915_add_request(dev, NULL, obj->write_domain);
-	obj->write_domain = 0;
+	BUG_ON(obj->write_domain);
 	i915_gem_object_move_to_active(obj, seqno);
 
 	trace_i915_gem_object_change_domain(obj,
···
 		i915_gem_flush(dev,
 			       dev->invalidate_domains,
 			       dev->flush_domains);
-		if (dev->flush_domains)
+		if (dev->flush_domains & I915_GEM_GPU_DOMAINS)
 			(void)i915_add_request(dev, file_priv,
 					       dev->flush_domains);
 	}
 
 	for (i = 0; i < args->buffer_count; i++) {
 		struct drm_gem_object *obj = object_list[i];
+		struct drm_i915_gem_object *obj_priv = obj->driver_private;
 		uint32_t old_write_domain = obj->write_domain;
 
 		obj->write_domain = obj->pending_write_domain;
+		if (obj->write_domain)
+			list_move_tail(&obj_priv->gpu_write_list,
+				       &dev_priv->mm.gpu_write_list);
+		else
+			list_del_init(&obj_priv->gpu_write_list);
+
 		trace_i915_gem_object_change_domain(obj,
 						    obj->read_domains,
 						    old_write_domain);
···
 	obj_priv->obj = obj;
 	obj_priv->fence_reg = I915_FENCE_REG_NONE;
 	INIT_LIST_HEAD(&obj_priv->list);
+	INIT_LIST_HEAD(&obj_priv->gpu_write_list);
 	INIT_LIST_HEAD(&obj_priv->fence_list);
 	obj_priv->madv = I915_MADV_WILLNEED;
 
···
 	spin_lock_init(&dev_priv->mm.active_list_lock);
 	INIT_LIST_HEAD(&dev_priv->mm.active_list);
 	INIT_LIST_HEAD(&dev_priv->mm.flushing_list);
+	INIT_LIST_HEAD(&dev_priv->mm.gpu_write_list);
 	INIT_LIST_HEAD(&dev_priv->mm.inactive_list);
 	INIT_LIST_HEAD(&dev_priv->mm.request_list);
 	INIT_LIST_HEAD(&dev_priv->mm.fence_list);