1 files changed, 648 insertions, 508 deletions

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index cf4ffbee1c0..29e97c07542 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -37,7 +37,9 @@
 #include <linux/intel-gtt.h>
 
 static uint32_t i915_gem_get_gtt_alignment(struct drm_gem_object *obj);
-static int i915_gem_object_flush_gpu_write_domain(struct drm_gem_object *obj);
+
+static int i915_gem_object_flush_gpu_write_domain(struct drm_gem_object *obj,
+						  bool pipelined);
 static void i915_gem_object_flush_gtt_write_domain(struct drm_gem_object *obj);
 static void i915_gem_object_flush_cpu_write_domain(struct drm_gem_object *obj);
 static int i915_gem_object_set_to_cpu_domain(struct drm_gem_object *obj,
@@ -46,7 +48,8 @@ static int i915_gem_object_set_cpu_read_domain_range(struct drm_gem_object *obj,
 						     uint64_t offset,
 						     uint64_t size);
 static void i915_gem_object_set_to_full_cpu_read_domain(struct drm_gem_object *obj);
-static int i915_gem_object_wait_rendering(struct drm_gem_object *obj);
+static int i915_gem_object_wait_rendering(struct drm_gem_object *obj,
+					  bool interruptible);
 static int i915_gem_object_bind_to_gtt(struct drm_gem_object *obj,
 					   unsigned alignment);
 static void i915_gem_clear_fence_reg(struct drm_gem_object *obj);
@@ -55,9 +58,111 @@ static int i915_gem_phys_pwrite(struct drm_device *dev, struct drm_gem_object *o
 				struct drm_file *file_priv);
 static void i915_gem_free_object_tail(struct drm_gem_object *obj);
 
+static int
+i915_gem_object_get_pages(struct drm_gem_object *obj,
+			  gfp_t gfpmask);
+
+static void
+i915_gem_object_put_pages(struct drm_gem_object *obj);
+
 static LIST_HEAD(shrink_list);
 static DEFINE_SPINLOCK(shrink_list_lock);
 
+/* some bookkeeping */
+static void i915_gem_info_add_obj(struct drm_i915_private *dev_priv,
+				  size_t size)
+{
+	dev_priv->mm.object_count++;
+	dev_priv->mm.object_memory += size;
+}
+
+static void i915_gem_info_remove_obj(struct drm_i915_private *dev_priv,
+				     size_t size)
+{
+	dev_priv->mm.object_count--;
+	dev_priv->mm.object_memory -= size;
+}
+
+static void i915_gem_info_add_gtt(struct drm_i915_private *dev_priv,
+				  size_t size)
+{
+	dev_priv->mm.gtt_count++;
+	dev_priv->mm.gtt_memory += size;
+}
+
+static void i915_gem_info_remove_gtt(struct drm_i915_private *dev_priv,
+				     size_t size)
+{
+	dev_priv->mm.gtt_count--;
+	dev_priv->mm.gtt_memory -= size;
+}
+
+static void i915_gem_info_add_pin(struct drm_i915_private *dev_priv,
+				  size_t size)
+{
+	dev_priv->mm.pin_count++;
+	dev_priv->mm.pin_memory += size;
+}
+
+static void i915_gem_info_remove_pin(struct drm_i915_private *dev_priv,
+				     size_t size)
+{
+	dev_priv->mm.pin_count--;
+	dev_priv->mm.pin_memory -= size;
+}
+
+int
+i915_gem_check_is_wedged(struct drm_device *dev)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	struct completion *x = &dev_priv->error_completion;
+	unsigned long flags;
+	int ret;
+
+	if (!atomic_read(&dev_priv->mm.wedged))
+		return 0;
+
+	ret = wait_for_completion_interruptible(x);
+	if (ret)
+		return ret;
+
+	/* Success, we reset the GPU! */
+	if (!atomic_read(&dev_priv->mm.wedged))
+		return 0;
+
+	/* GPU is hung, bump the completion count to account for
+	 * the token we just consumed so that we never hit zero and
+	 * end up waiting upon a subsequent completion event that
+	 * will never happen.
+	 */
+	spin_lock_irqsave(&x->wait.lock, flags);
+	x->done++;
+	spin_unlock_irqrestore(&x->wait.lock, flags);
+	return -EIO;
+}
+
+static int i915_mutex_lock_interruptible(struct drm_device *dev)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	int ret;
+
+	ret = i915_gem_check_is_wedged(dev);
+	if (ret)
+		return ret;
+
+	ret = mutex_lock_interruptible(&dev->struct_mutex);
+	if (ret)
+		return ret;
+
+	if (atomic_read(&dev_priv->mm.wedged)) {
+		mutex_unlock(&dev->struct_mutex);
+		return -EAGAIN;
+	}
+
+	WARN_ON(i915_verify_lists(dev));
+	return 0;
+}
+
 static inline bool
 i915_gem_object_is_inactive(struct drm_i915_gem_object *obj_priv)
 {
@@ -66,7 +171,8 @@ i915_gem_object_is_inactive(struct drm_i915_gem_object *obj_priv)
 		obj_priv->pin_count == 0;
 }
 
-int i915_gem_do_init(struct drm_device *dev, unsigned long start,
+int i915_gem_do_init(struct drm_device *dev,
+		     unsigned long start,
 		     unsigned long end)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
@@ -80,7 +186,7 @@ int i915_gem_do_init(struct drm_device *dev, unsigned long start,
 	drm_mm_init(&dev_priv->mm.gtt_space, start,
 		    end - start);
 
-	dev->gtt_total = (uint32_t) (end - start);
+	dev_priv->mm.gtt_total = end - start;
 
 	return 0;
 }
@@ -103,14 +209,16 @@ int
 i915_gem_get_aperture_ioctl(struct drm_device *dev, void *data,
 			    struct drm_file *file_priv)
 {
+	struct drm_i915_private *dev_priv = dev->dev_private;
 	struct drm_i915_gem_get_aperture *args = data;
 
 	if (!(dev->driver->driver_features & DRIVER_GEM))
 		return -ENODEV;
 
-	args->aper_size = dev->gtt_total;
-	args->aper_available_size = (args->aper_size -
-				     atomic_read(&dev->pin_memory));
+	mutex_lock(&dev->struct_mutex);
+	args->aper_size = dev_priv->mm.gtt_total;
+	args->aper_available_size = args->aper_size - dev_priv->mm.pin_memory;
+	mutex_unlock(&dev->struct_mutex);
 
 	return 0;
 }
@@ -265,7 +373,9 @@ i915_gem_shmem_pread_fast(struct drm_device *dev, struct drm_gem_object *obj,
 	user_data = (char __user *) (uintptr_t) args->data_ptr;
 	remain = args->size;
 
-	mutex_lock(&dev->struct_mutex);
+	ret = i915_mutex_lock_interruptible(dev);
+	if (ret)
+		return ret;
 
 	ret = i915_gem_object_get_pages(obj, 0);
 	if (ret != 0)
@@ -384,7 +494,9 @@ i915_gem_shmem_pread_slow(struct drm_device *dev, struct drm_gem_object *obj,
 
 	do_bit17_swizzling = i915_gem_object_needs_bit17_swizzle(obj);
 
-	mutex_lock(&dev->struct_mutex);
+	ret = i915_mutex_lock_interruptible(dev);
+	if (ret)
+		goto fail_put_user_pages;
 
 	ret = i915_gem_object_get_pages_or_evict(obj);
 	if (ret)
@@ -464,21 +576,27 @@ i915_gem_pread_ioctl(struct drm_device *dev, void *data,
 	struct drm_i915_gem_pread *args = data;
 	struct drm_gem_object *obj;
 	struct drm_i915_gem_object *obj_priv;
-	int ret;
+	int ret = 0;
 
 	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
 	if (obj == NULL)
 		return -ENOENT;
 	obj_priv = to_intel_bo(obj);
 
-	/* Bounds check source.
-	 *
-	 * XXX: This could use review for overflow issues...
-	 */
-	if (args->offset > obj->size || args->size > obj->size ||
-	    args->offset + args->size > obj->size) {
-		drm_gem_object_unreference_unlocked(obj);
-		return -EINVAL;
+	/* Bounds check source.  */
+	if (args->offset > obj->size || args->size > obj->size - args->offset) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	if (args->size == 0)
+		goto out;
+
+	if (!access_ok(VERIFY_WRITE,
+		       (char __user *)(uintptr_t)args->data_ptr,
+		       args->size)) {
+		ret = -EFAULT;
+		goto out;
 	}
 
 	if (i915_gem_object_needs_bit17_swizzle(obj)) {
@@ -490,8 +608,8 @@ i915_gem_pread_ioctl(struct drm_device *dev, void *data,
 							file_priv);
 	}
 
+out:
 	drm_gem_object_unreference_unlocked(obj);
-
 	return ret;
 }
 
@@ -580,11 +698,11 @@ i915_gem_gtt_pwrite_fast(struct drm_device *dev, struct drm_gem_object *obj,
 
 	user_data = (char __user *) (uintptr_t) args->data_ptr;
 	remain = args->size;
-	if (!access_ok(VERIFY_READ, user_data, remain))
-		return -EFAULT;
 
+	ret = i915_mutex_lock_interruptible(dev);
+	if (ret)
+		return ret;
 
-	mutex_lock(&dev->struct_mutex);
 	ret = i915_gem_object_pin(obj, 0);
 	if (ret) {
 		mutex_unlock(&dev->struct_mutex);
@@ -679,7 +797,10 @@ i915_gem_gtt_pwrite_slow(struct drm_device *dev, struct drm_gem_object *obj,
 		goto out_unpin_pages;
 	}
 
-	mutex_lock(&dev->struct_mutex);
+	ret = i915_mutex_lock_interruptible(dev);
+	if (ret)
+		goto out_unpin_pages;
+
 	ret = i915_gem_object_pin(obj, 0);
 	if (ret)
 		goto out_unlock;
@@ -753,7 +874,9 @@ i915_gem_shmem_pwrite_fast(struct drm_device *dev, struct drm_gem_object *obj,
 	user_data = (char __user *) (uintptr_t) args->data_ptr;
 	remain = args->size;
 
-	mutex_lock(&dev->struct_mutex);
+	ret = i915_mutex_lock_interruptible(dev);
+	if (ret)
+		return ret;
 
 	ret = i915_gem_object_get_pages(obj, 0);
 	if (ret != 0)
@@ -849,7 +972,9 @@ i915_gem_shmem_pwrite_slow(struct drm_device *dev, struct drm_gem_object *obj,
 
 	do_bit17_swizzling = i915_gem_object_needs_bit17_swizzle(obj);
 
-	mutex_lock(&dev->struct_mutex);
+	ret = i915_mutex_lock_interruptible(dev);
+	if (ret)
+		goto fail_put_user_pages;
 
 	ret = i915_gem_object_get_pages_or_evict(obj);
 	if (ret)
@@ -934,14 +1059,20 @@ i915_gem_pwrite_ioctl(struct drm_device *dev, void *data,
 		return -ENOENT;
 	obj_priv = to_intel_bo(obj);
 
-	/* Bounds check destination.
-	 *
-	 * XXX: This could use review for overflow issues...
-	 */
-	if (args->offset > obj->size || args->size > obj->size ||
-	    args->offset + args->size > obj->size) {
-		drm_gem_object_unreference_unlocked(obj);
-		return -EINVAL;
+	/* Bounds check destination. */
+	if (args->offset > obj->size || args->size > obj->size - args->offset) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	if (args->size == 0)
+		goto out;
+
+	if (!access_ok(VERIFY_READ,
+		       (char __user *)(uintptr_t)args->data_ptr,
+		       args->size)) {
+		ret = -EFAULT;
+		goto out;
 	}
 
 	/* We can only do the GTT pwrite on untiled buffers, as otherwise
@@ -953,7 +1084,7 @@ i915_gem_pwrite_ioctl(struct drm_device *dev, void *data,
 	if (obj_priv->phys_obj)
 		ret = i915_gem_phys_pwrite(dev, obj, args, file_priv);
 	else if (obj_priv->tiling_mode == I915_TILING_NONE &&
-		 dev->gtt_total != 0 &&
+		 obj_priv->gtt_space &&
 		 obj->write_domain != I915_GEM_DOMAIN_CPU) {
 		ret = i915_gem_gtt_pwrite_fast(dev, obj, args, file_priv);
 		if (ret == -EFAULT) {
@@ -975,8 +1106,8 @@ i915_gem_pwrite_ioctl(struct drm_device *dev, void *data,
 		DRM_INFO("pwrite failed %d\n", ret);
 #endif
 
+out:
 	drm_gem_object_unreference_unlocked(obj);
-
 	return ret;
 }
 
@@ -1017,14 +1148,14 @@ i915_gem_set_domain_ioctl(struct drm_device *dev, void *data,
 		return -ENOENT;
 	obj_priv = to_intel_bo(obj);
 
-	mutex_lock(&dev->struct_mutex);
+	ret = i915_mutex_lock_interruptible(dev);
+	if (ret) {
+		drm_gem_object_unreference_unlocked(obj);
+		return ret;
+	}
 
 	intel_mark_busy(dev, obj);
 
-#if WATCH_BUF
-	DRM_INFO("set_domain_ioctl %p(%zd), %08x %08x\n",
-		 obj, obj->size, read_domains, write_domain);
-#endif
 	if (read_domains & I915_GEM_DOMAIN_GTT) {
 		ret = i915_gem_object_set_to_gtt_domain(obj, write_domain != 0);
 
@@ -1048,7 +1179,6 @@ i915_gem_set_domain_ioctl(struct drm_device *dev, void *data,
 		ret = i915_gem_object_set_to_cpu_domain(obj, write_domain != 0);
 	}
 
-	
 	/* Maintain LRU order of "inactive" objects */
 	if (ret == 0 && i915_gem_object_is_inactive(obj_priv))
 		list_move_tail(&obj_priv->list, &dev_priv->mm.inactive_list);
@@ -1067,27 +1197,23 @@ i915_gem_sw_finish_ioctl(struct drm_device *dev, void *data,
 {
 	struct drm_i915_gem_sw_finish *args = data;
 	struct drm_gem_object *obj;
-	struct drm_i915_gem_object *obj_priv;
 	int ret = 0;
 
 	if (!(dev->driver->driver_features & DRIVER_GEM))
 		return -ENODEV;
 
-	mutex_lock(&dev->struct_mutex);
 	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
-	if (obj == NULL) {
-		mutex_unlock(&dev->struct_mutex);
+	if (obj == NULL)
 		return -ENOENT;
-	}
 
-#if WATCH_BUF
-	DRM_INFO("%s: sw_finish %d (%p %zd)\n",
-		 __func__, args->handle, obj, obj->size);
-#endif
-	obj_priv = to_intel_bo(obj);
+	ret = i915_mutex_lock_interruptible(dev);
+	if (ret) {
+		drm_gem_object_unreference_unlocked(obj);
+		return ret;
+	}
 
 	/* Pinned buffers may be scanout, so flush the cache */
-	if (obj_priv->pin_count)
+	if (to_intel_bo(obj)->pin_count)
 		i915_gem_object_flush_cpu_write_domain(obj);
 
 	drm_gem_object_unreference(obj);
@@ -1179,7 +1305,7 @@ int i915_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 
 	/* Need a new fence register? */
 	if (obj_priv->tiling_mode != I915_TILING_NONE) {
-		ret = i915_gem_object_get_fence_reg(obj);
+		ret = i915_gem_object_get_fence_reg(obj, true);
 		if (ret)
 			goto unlock;
 	}
@@ -1244,7 +1370,7 @@ i915_gem_create_mmap_offset(struct drm_gem_object *obj)
 						    obj->size / PAGE_SIZE, 0, 0);
 	if (!list->file_offset_node) {
 		DRM_ERROR("failed to allocate offset for bo %d\n", obj->name);
-		ret = -ENOMEM;
+		ret = -ENOSPC;
 		goto out_free_list;
 	}
 
@@ -1256,9 +1382,9 @@ i915_gem_create_mmap_offset(struct drm_gem_object *obj)
 	}
 
 	list->hash.key = list->file_offset_node->start;
-	if (drm_ht_insert_item(&mm->offset_hash, &list->hash)) {
+	ret = drm_ht_insert_item(&mm->offset_hash, &list->hash);
+	if (ret) {
 		DRM_ERROR("failed to add to map hash\n");
-		ret = -ENOMEM;
 		goto out_free_mm;
 	}
 
@@ -1343,14 +1469,14 @@ i915_gem_get_gtt_alignment(struct drm_gem_object *obj)
 	 * Minimum alignment is 4k (GTT page size), but might be greater
 	 * if a fence register is needed for the object.
 	 */
-	if (IS_I965G(dev) || obj_priv->tiling_mode == I915_TILING_NONE)
+	if (INTEL_INFO(dev)->gen >= 4 || obj_priv->tiling_mode == I915_TILING_NONE)
 		return 4096;
 
 	/*
 	 * Previous chips need to be aligned to the size of the smallest
 	 * fence register that can contain the object.
 	 */
-	if (IS_I9XX(dev))
+	if (INTEL_INFO(dev)->gen == 3)
 		start = 1024*1024;
 	else
 		start = 512*1024;
@@ -1392,7 +1518,11 @@ i915_gem_mmap_gtt_ioctl(struct drm_device *dev, void *data,
 	if (obj == NULL)
 		return -ENOENT;
 
-	mutex_lock(&dev->struct_mutex);
+	ret = i915_mutex_lock_interruptible(dev);
+	if (ret) {
+		drm_gem_object_unreference_unlocked(obj);
+		return ret;
+	}
 
 	obj_priv = to_intel_bo(obj);
 
@@ -1434,7 +1564,7 @@ i915_gem_mmap_gtt_ioctl(struct drm_device *dev, void *data,
 	return 0;
 }
 
-void
+static void
 i915_gem_object_put_pages(struct drm_gem_object *obj)
 {
 	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
@@ -1468,13 +1598,24 @@ i915_gem_object_put_pages(struct drm_gem_object *obj)
 	obj_priv->pages = NULL;
 }
 
+static uint32_t
+i915_gem_next_request_seqno(struct drm_device *dev,
+			    struct intel_ring_buffer *ring)
+{
+	drm_i915_private_t *dev_priv = dev->dev_private;
+
+	ring->outstanding_lazy_request = true;
+	return dev_priv->next_seqno;
+}
+
 static void
-i915_gem_object_move_to_active(struct drm_gem_object *obj, uint32_t seqno,
+i915_gem_object_move_to_active(struct drm_gem_object *obj,
 			       struct intel_ring_buffer *ring)
 {
 	struct drm_device *dev = obj->dev;
-	drm_i915_private_t *dev_priv = dev->dev_private;
 	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
+	uint32_t seqno = i915_gem_next_request_seqno(dev, ring);
+
 	BUG_ON(ring == NULL);
 	obj_priv->ring = ring;
 
@@ -1483,10 +1624,9 @@ i915_gem_object_move_to_active(struct drm_gem_object *obj, uint32_t seqno,
 		drm_gem_object_reference(obj);
 		obj_priv->active = 1;
 	}
+
 	/* Move from whatever list we were on to the tail of execution. */
-	spin_lock(&dev_priv->mm.active_list_lock);
 	list_move_tail(&obj_priv->list, &ring->active_list);
-	spin_unlock(&dev_priv->mm.active_list_lock);
 	obj_priv->last_rendering_seqno = seqno;
 }
 
@@ -1536,9 +1676,8 @@ i915_gem_object_move_to_inactive(struct drm_gem_object *obj)
 	drm_i915_private_t *dev_priv = dev->dev_private;
 	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
 
-	i915_verify_inactive(dev, __FILE__, __LINE__);
 	if (obj_priv->pin_count != 0)
-		list_del_init(&obj_priv->list);
+		list_move_tail(&obj_priv->list, &dev_priv->mm.pinned_list);
 	else
 		list_move_tail(&obj_priv->list, &dev_priv->mm.inactive_list);
 
@@ -1550,12 +1689,12 @@ i915_gem_object_move_to_inactive(struct drm_gem_object *obj)
 		obj_priv->active = 0;
 		drm_gem_object_unreference(obj);
 	}
-	i915_verify_inactive(dev, __FILE__, __LINE__);
+	WARN_ON(i915_verify_lists(dev));
 }
 
 static void
 i915_gem_process_flushing_list(struct drm_device *dev,
-			       uint32_t flush_domains, uint32_t seqno,
+			       uint32_t flush_domains,
 			       struct intel_ring_buffer *ring)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
@@ -1566,14 +1705,13 @@ i915_gem_process_flushing_list(struct drm_device *dev,
 				 gpu_write_list) {
 		struct drm_gem_object *obj = &obj_priv->base;
 
-		if ((obj->write_domain & flush_domains) ==
-		    obj->write_domain &&
-		    obj_priv->ring->ring_flag == ring->ring_flag) {
+		if (obj->write_domain & flush_domains &&
+		    obj_priv->ring == ring) {
 			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, ring);
+			i915_gem_object_move_to_active(obj, ring);
 
 			/* update the fence lru list */
 			if (obj_priv->fence_reg != I915_FENCE_REG_NONE) {
@@ -1591,23 +1729,27 @@ i915_gem_process_flushing_list(struct drm_device *dev,
 }
 
 uint32_t
-i915_add_request(struct drm_device *dev, struct drm_file *file_priv,
-		 uint32_t flush_domains, struct intel_ring_buffer *ring)
+i915_add_request(struct drm_device *dev,
+		 struct drm_file *file,
+		 struct drm_i915_gem_request *request,
+		 struct intel_ring_buffer *ring)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
-	struct drm_i915_file_private *i915_file_priv = NULL;
-	struct drm_i915_gem_request *request;
+	struct drm_i915_file_private *file_priv = NULL;
 	uint32_t seqno;
 	int was_empty;
 
-	if (file_priv != NULL)
-		i915_file_priv = file_priv->driver_priv;
+	if (file != NULL)
+		file_priv = file->driver_priv;
 
-	request = kzalloc(sizeof(*request), GFP_KERNEL);
-	if (request == NULL)
-		return 0;
+	if (request == NULL) {
+		request = kzalloc(sizeof(*request), GFP_KERNEL);
+		if (request == NULL)
+			return 0;
+	}
 
-	seqno = ring->add_request(dev, ring, file_priv, flush_domains);
+	seqno = ring->add_request(dev, ring, 0);
+	ring->outstanding_lazy_request = false;
 
 	request->seqno = seqno;
 	request->ring = ring;
@@ -1615,23 +1757,20 @@ i915_add_request(struct drm_device *dev, struct drm_file *file_priv,
 	was_empty = list_empty(&ring->request_list);
 	list_add_tail(&request->list, &ring->request_list);
 
-	if (i915_file_priv) {
+	if (file_priv) {
+		spin_lock(&file_priv->mm.lock);
+		request->file_priv = file_priv;
 		list_add_tail(&request->client_list,
-			      &i915_file_priv->mm.request_list);
-	} else {
-		INIT_LIST_HEAD(&request->client_list);
+			      &file_priv->mm.request_list);
+		spin_unlock(&file_priv->mm.lock);
 	}
 
-	/* Associate any objects on the flushing list matching the write
-	 * domain we're flushing with our flush.
-	 */
-	if (flush_domains != 0) 
-		i915_gem_process_flushing_list(dev, flush_domains, seqno, ring);
-
 	if (!dev_priv->mm.suspended) {
-		mod_timer(&dev_priv->hangcheck_timer, jiffies + DRM_I915_HANGCHECK_PERIOD);
+		mod_timer(&dev_priv->hangcheck_timer,
+			  jiffies + msecs_to_jiffies(DRM_I915_HANGCHECK_PERIOD));
 		if (was_empty)
-			queue_delayed_work(dev_priv->wq, &dev_priv->mm.retire_work, HZ);
+			queue_delayed_work(dev_priv->wq,
+					   &dev_priv->mm.retire_work, HZ);
 	}
 	return seqno;
 }
@@ -1642,91 +1781,105 @@ i915_add_request(struct drm_device *dev, struct drm_file *file_priv,
  * Ensures that all commands in the ring are finished
  * before signalling the CPU
  */
-static uint32_t
+static void
 i915_retire_commands(struct drm_device *dev, struct intel_ring_buffer *ring)
 {
 	uint32_t flush_domains = 0;
 
 	/* The sampler always gets flushed on i965 (sigh) */
-	if (IS_I965G(dev))
+	if (INTEL_INFO(dev)->gen >= 4)
 		flush_domains |= I915_GEM_DOMAIN_SAMPLER;
 
 	ring->flush(dev, ring,
 			I915_GEM_DOMAIN_COMMAND, flush_domains);
-	return flush_domains;
 }
 
-/**
- * Moves buffers associated only with the given active seqno from the active
- * to inactive list, potentially freeing them.
- */
-static void
-i915_gem_retire_request(struct drm_device *dev,
-			struct drm_i915_gem_request *request)
+static inline void
+i915_gem_request_remove_from_client(struct drm_i915_gem_request *request)
 {
-	drm_i915_private_t *dev_priv = dev->dev_private;
+	struct drm_i915_file_private *file_priv = request->file_priv;
 
-	trace_i915_gem_request_retire(dev, request->seqno);
+	if (!file_priv)
+		return;
 
-	/* Move any buffers on the active list that are no longer referenced
-	 * by the ringbuffer to the flushing/inactive lists as appropriate.
-	 */
-	spin_lock(&dev_priv->mm.active_list_lock);
-	while (!list_empty(&request->ring->active_list)) {
-		struct drm_gem_object *obj;
+	spin_lock(&file_priv->mm.lock);
+	list_del(&request->client_list);
+	request->file_priv = NULL;
+	spin_unlock(&file_priv->mm.lock);
+}
+
+static void i915_gem_reset_ring_lists(struct drm_i915_private *dev_priv,
+				      struct intel_ring_buffer *ring)
+{
+	while (!list_empty(&ring->request_list)) {
+		struct drm_i915_gem_request *request;
+
+		request = list_first_entry(&ring->request_list,
+					   struct drm_i915_gem_request,
+					   list);
+
+		list_del(&request->list);
+		i915_gem_request_remove_from_client(request);
+		kfree(request);
+	}
+
+	while (!list_empty(&ring->active_list)) {
 		struct drm_i915_gem_object *obj_priv;
 
-		obj_priv = list_first_entry(&request->ring->active_list,
+		obj_priv = list_first_entry(&ring->active_list,
 					    struct drm_i915_gem_object,
 					    list);
-		obj = &obj_priv->base;
-
-		/* If the seqno being retired doesn't match the oldest in the
-		 * list, then the oldest in the list must still be newer than
-		 * this seqno.
-		 */
-		if (obj_priv->last_rendering_seqno != request->seqno)
-			goto out;
 
-#if WATCH_LRU
-		DRM_INFO("%s: retire %d moves to inactive list %p\n",
-			 __func__, request->seqno, obj);
-#endif
-
-		if (obj->write_domain != 0)
-			i915_gem_object_move_to_flushing(obj);
-		else {
-			/* Take a reference on the object so it won't be
-			 * freed while the spinlock is held.  The list
-			 * protection for this spinlock is safe when breaking
-			 * the lock like this since the next thing we do
-			 * is just get the head of the list again.
-			 */
-			drm_gem_object_reference(obj);
-			i915_gem_object_move_to_inactive(obj);
-			spin_unlock(&dev_priv->mm.active_list_lock);
-			drm_gem_object_unreference(obj);
-			spin_lock(&dev_priv->mm.active_list_lock);
-		}
+		obj_priv->base.write_domain = 0;
+		list_del_init(&obj_priv->gpu_write_list);
+		i915_gem_object_move_to_inactive(&obj_priv->base);
 	}
-out:
-	spin_unlock(&dev_priv->mm.active_list_lock);
 }
 
-/**
- * Returns true if seq1 is later than seq2.
- */
-bool
-i915_seqno_passed(uint32_t seq1, uint32_t seq2)
+void i915_gem_reset(struct drm_device *dev)
 {
-	return (int32_t)(seq1 - seq2) >= 0;
-}
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	struct drm_i915_gem_object *obj_priv;
+	int i;
 
-uint32_t
-i915_get_gem_seqno(struct drm_device *dev,
-		   struct intel_ring_buffer *ring)
-{
-	return ring->get_gem_seqno(dev, ring);
+	i915_gem_reset_ring_lists(dev_priv, &dev_priv->render_ring);
+	if (HAS_BSD(dev))
+		i915_gem_reset_ring_lists(dev_priv, &dev_priv->bsd_ring);
+
+	/* Remove anything from the flushing lists. The GPU cache is likely
+	 * to be lost on reset along with the data, so simply move the
+	 * lost bo to the inactive list.
+	 */
+	while (!list_empty(&dev_priv->mm.flushing_list)) {
+		obj_priv = list_first_entry(&dev_priv->mm.flushing_list,
+					    struct drm_i915_gem_object,
+					    list);
+
+		obj_priv->base.write_domain = 0;
+		list_del_init(&obj_priv->gpu_write_list);
+		i915_gem_object_move_to_inactive(&obj_priv->base);
+	}
+
+	/* Move everything out of the GPU domains to ensure we do any
+	 * necessary invalidation upon reuse.
+	 */
+	list_for_each_entry(obj_priv,
+			    &dev_priv->mm.inactive_list,
+			    list)
+	{
+		obj_priv->base.read_domains &= ~I915_GEM_GPU_DOMAINS;
+	}
+
+	/* The fence registers are invalidated so clear them out */
+	for (i = 0; i < 16; i++) {
+		struct drm_i915_fence_reg *reg;
+
+		reg = &dev_priv->fence_regs[i];
+		if (!reg->obj)
+			continue;
+
+		i915_gem_clear_fence_reg(reg->obj);
+	}
 }
 
 /**
@@ -1739,38 +1892,58 @@ i915_gem_retire_requests_ring(struct drm_device *dev,
 	drm_i915_private_t *dev_priv = dev->dev_private;
 	uint32_t seqno;
 
-	if (!ring->status_page.page_addr
-			|| list_empty(&ring->request_list))
+	if (!ring->status_page.page_addr ||
+	    list_empty(&ring->request_list))
 		return;
 
-	seqno = i915_get_gem_seqno(dev, ring);
+	WARN_ON(i915_verify_lists(dev));
 
+	seqno = ring->get_seqno(dev, ring);
 	while (!list_empty(&ring->request_list)) {
 		struct drm_i915_gem_request *request;
-		uint32_t retiring_seqno;
 
 		request = list_first_entry(&ring->request_list,
 					   struct drm_i915_gem_request,
 					   list);
-		retiring_seqno = request->seqno;
 
-		if (i915_seqno_passed(seqno, retiring_seqno) ||
-		    atomic_read(&dev_priv->mm.wedged)) {
-			i915_gem_retire_request(dev, request);
+		if (!i915_seqno_passed(seqno, request->seqno))
+			break;
+
+		trace_i915_gem_request_retire(dev, request->seqno);
+
+		list_del(&request->list);
+		i915_gem_request_remove_from_client(request);
+		kfree(request);
+	}
+
+	/* Move any buffers on the active list that are no longer referenced
+	 * by the ringbuffer to the flushing/inactive lists as appropriate.
+	 */
+	while (!list_empty(&ring->active_list)) {
+		struct drm_gem_object *obj;
+		struct drm_i915_gem_object *obj_priv;
+
+		obj_priv = list_first_entry(&ring->active_list,
+					    struct drm_i915_gem_object,
+					    list);
 
-			list_del(&request->list);
-			list_del(&request->client_list);
-			kfree(request);
-		} else
+		if (!i915_seqno_passed(seqno, obj_priv->last_rendering_seqno))
 			break;
+
+		obj = &obj_priv->base;
+		if (obj->write_domain != 0)
+			i915_gem_object_move_to_flushing(obj);
+		else
+			i915_gem_object_move_to_inactive(obj);
 	}
 
 	if (unlikely (dev_priv->trace_irq_seqno &&
 		      i915_seqno_passed(dev_priv->trace_irq_seqno, seqno))) {
-
 		ring->user_irq_put(dev, ring);
 		dev_priv->trace_irq_seqno = 0;
 	}
+
+	WARN_ON(i915_verify_lists(dev));
 }
 
 void
@@ -1797,7 +1970,7 @@ i915_gem_retire_requests(struct drm_device *dev)
 		i915_gem_retire_requests_ring(dev, &dev_priv->bsd_ring);
 }
 
-void
+static void
 i915_gem_retire_work_handler(struct work_struct *work)
 {
 	drm_i915_private_t *dev_priv;
@@ -1807,7 +1980,12 @@ i915_gem_retire_work_handler(struct work_struct *work)
 				mm.retire_work.work);
 	dev = dev_priv->dev;
 
-	mutex_lock(&dev->struct_mutex);
+	/* Come back later if the device is busy... */
+	if (!mutex_trylock(&dev->struct_mutex)) {
+		queue_delayed_work(dev_priv->wq, &dev_priv->mm.retire_work, HZ);
+		return;
+	}
+
 	i915_gem_retire_requests(dev);
 
 	if (!dev_priv->mm.suspended &&
@@ -1820,7 +1998,7 @@ i915_gem_retire_work_handler(struct work_struct *work)
 
 int
 i915_do_wait_request(struct drm_device *dev, uint32_t seqno,
-		int interruptible, struct intel_ring_buffer *ring)
+		     bool interruptible, struct intel_ring_buffer *ring)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
 	u32 ier;
@@ -1829,9 +2007,16 @@ i915_do_wait_request(struct drm_device *dev, uint32_t seqno,
 	BUG_ON(seqno == 0);
 
 	if (atomic_read(&dev_priv->mm.wedged))
-		return -EIO;
+		return -EAGAIN;
+
+	if (ring->outstanding_lazy_request) {
+		seqno = i915_add_request(dev, NULL, NULL, ring);
+		if (seqno == 0)
+			return -ENOMEM;
+	}
+	BUG_ON(seqno == dev_priv->next_seqno);
 
-	if (!i915_seqno_passed(ring->get_gem_seqno(dev, ring), seqno)) {
+	if (!i915_seqno_passed(ring->get_seqno(dev, ring), seqno)) {
 		if (HAS_PCH_SPLIT(dev))
 			ier = I915_READ(DEIER) | I915_READ(GTIER);
 		else
@@ -1850,12 +2035,12 @@ i915_do_wait_request(struct drm_device *dev, uint32_t seqno,
 		if (interruptible)
 			ret = wait_event_interruptible(ring->irq_queue,
 				i915_seqno_passed(
-					ring->get_gem_seqno(dev, ring), seqno)
+					ring->get_seqno(dev, ring), seqno)
 				|| atomic_read(&dev_priv->mm.wedged));
 		else
 			wait_event(ring->irq_queue,
 				i915_seqno_passed(
-					ring->get_gem_seqno(dev, ring), seqno)
+					ring->get_seqno(dev, ring), seqno)
 				|| atomic_read(&dev_priv->mm.wedged));
 
 		ring->user_irq_put(dev, ring);
@@ -1864,11 +2049,12 @@ i915_do_wait_request(struct drm_device *dev, uint32_t seqno,
 		trace_i915_gem_request_wait_end(dev, seqno);
 	}
 	if (atomic_read(&dev_priv->mm.wedged))
-		ret = -EIO;
+		ret = -EAGAIN;
 
 	if (ret && ret != -ERESTARTSYS)
-		DRM_ERROR("%s returns %d (awaiting %d at %d)\n",
-			  __func__, ret, seqno, ring->get_gem_seqno(dev, ring));
+		DRM_ERROR("%s returns %d (awaiting %d at %d, next %d)\n",
+			  __func__, ret, seqno, ring->get_seqno(dev, ring),
+			  dev_priv->next_seqno);
 
 	/* Directly dispatch request retiring.  While we have the work queue
 	 * to handle this, the waiter on a request often wants an associated
@@ -1887,27 +2073,44 @@ i915_do_wait_request(struct drm_device *dev, uint32_t seqno,
  */
 static int
 i915_wait_request(struct drm_device *dev, uint32_t seqno,
-		struct intel_ring_buffer *ring)
+		  struct intel_ring_buffer *ring)
 {
 	return i915_do_wait_request(dev, seqno, 1, ring);
 }
 
 static void
+i915_gem_flush_ring(struct drm_device *dev,
+		    struct drm_file *file_priv,
+		    struct intel_ring_buffer *ring,
+		    uint32_t invalidate_domains,
+		    uint32_t flush_domains)
+{
+	ring->flush(dev, ring, invalidate_domains, flush_domains);
+	i915_gem_process_flushing_list(dev, flush_domains, ring);
+}
+
+static void
 i915_gem_flush(struct drm_device *dev,
+	       struct drm_file *file_priv,
 	       uint32_t invalidate_domains,
-	       uint32_t flush_domains)
+	       uint32_t flush_domains,
+	       uint32_t flush_rings)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
+
 	if (flush_domains & I915_GEM_DOMAIN_CPU)
 		drm_agp_chipset_flush(dev);
-	dev_priv->render_ring.flush(dev, &dev_priv->render_ring,
-			invalidate_domains,
-			flush_domains);
 
-	if (HAS_BSD(dev))
-		dev_priv->bsd_ring.flush(dev, &dev_priv->bsd_ring,
-				invalidate_domains,
-				flush_domains);
+	if ((flush_domains | invalidate_domains) & I915_GEM_GPU_DOMAINS) {
+		if (flush_rings & RING_RENDER)
+			i915_gem_flush_ring(dev, file_priv,
+					    &dev_priv->render_ring,
+					    invalidate_domains, flush_domains);
+		if (flush_rings & RING_BSD)
+			i915_gem_flush_ring(dev, file_priv,
+					    &dev_priv->bsd_ring,
+					    invalidate_domains, flush_domains);
+	}
 }
 
 /**
@@ -1915,7 +2118,8 @@ i915_gem_flush(struct drm_device *dev,
  * safe to unbind from the GTT or access from the CPU.
  */
 static int
-i915_gem_object_wait_rendering(struct drm_gem_object *obj)
+i915_gem_object_wait_rendering(struct drm_gem_object *obj,
+			       bool interruptible)
 {
 	struct drm_device *dev = obj->dev;
 	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
@@ -1930,13 +2134,11 @@ i915_gem_object_wait_rendering(struct drm_gem_object *obj)
 	 * it.
 	 */
 	if (obj_priv->active) {
-#if WATCH_BUF
-		DRM_INFO("%s: object %p wait for seqno %08x\n",
-			  __func__, obj, obj_priv->last_rendering_seqno);
-#endif
-		ret = i915_wait_request(dev,
-				obj_priv->last_rendering_seqno, obj_priv->ring);
-		if (ret != 0)
+		ret = i915_do_wait_request(dev,
+					   obj_priv->last_rendering_seqno,
+					   interruptible,
+					   obj_priv->ring);
+		if (ret)
 			return ret;
 	}
 
@@ -1950,14 +2152,10 @@ int
 i915_gem_object_unbind(struct drm_gem_object *obj)
 {
 	struct drm_device *dev = obj->dev;
-	drm_i915_private_t *dev_priv = dev->dev_private;
+	struct drm_i915_private *dev_priv = dev->dev_private;
 	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
 	int ret = 0;
 
-#if WATCH_BUF
-	DRM_INFO("%s:%d %p\n", __func__, __LINE__, obj);
-	DRM_INFO("gtt_space %p\n", obj_priv->gtt_space);
-#endif
 	if (obj_priv->gtt_space == NULL)
 		return 0;
 
@@ -1982,33 +2180,26 @@ i915_gem_object_unbind(struct drm_gem_object *obj)
 	 * should be safe and we need to cleanup or else we might
 	 * cause memory corruption through use-after-free.
 	 */
+	if (ret) {
+		i915_gem_clflush_object(obj);
+		obj->read_domains = obj->write_domain = I915_GEM_DOMAIN_CPU;
+	}
 
 	/* release the fence reg _after_ flushing */
 	if (obj_priv->fence_reg != I915_FENCE_REG_NONE)
 		i915_gem_clear_fence_reg(obj);
 
-	if (obj_priv->agp_mem != NULL) {
-		drm_unbind_agp(obj_priv->agp_mem);
-		drm_free_agp(obj_priv->agp_mem, obj->size / PAGE_SIZE);
-		obj_priv->agp_mem = NULL;
-	}
+	drm_unbind_agp(obj_priv->agp_mem);
+	drm_free_agp(obj_priv->agp_mem, obj->size / PAGE_SIZE);
 
 	i915_gem_object_put_pages(obj);
 	BUG_ON(obj_priv->pages_refcount);
 
-	if (obj_priv->gtt_space) {
-		atomic_dec(&dev->gtt_count);
-		atomic_sub(obj->size, &dev->gtt_memory);
+	i915_gem_info_remove_gtt(dev_priv, obj->size);
+	list_del_init(&obj_priv->list);
 
-		drm_mm_put_block(obj_priv->gtt_space);
-		obj_priv->gtt_space = NULL;
-	}
-
-	/* Remove ourselves from the LRU list if present. */
-	spin_lock(&dev_priv->mm.active_list_lock);
-	if (!list_empty(&obj_priv->list))
-		list_del_init(&obj_priv->list);
-	spin_unlock(&dev_priv->mm.active_list_lock);
+	drm_mm_put_block(obj_priv->gtt_space);
+	obj_priv->gtt_space = NULL;
 
 	if (i915_gem_object_is_purgeable(obj_priv))
 		i915_gem_object_truncate(obj);
@@ -2018,48 +2209,45 @@ i915_gem_object_unbind(struct drm_gem_object *obj)
 	return ret;
 }
 
+static int i915_ring_idle(struct drm_device *dev,
+			  struct intel_ring_buffer *ring)
+{
+	i915_gem_flush_ring(dev, NULL, ring,
+			    I915_GEM_GPU_DOMAINS, I915_GEM_GPU_DOMAINS);
+	return i915_wait_request(dev,
+				 i915_gem_next_request_seqno(dev, ring),
+				 ring);
+}
+
 int
 i915_gpu_idle(struct drm_device *dev)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
 	bool lists_empty;
-	uint32_t seqno1, seqno2;
 	int ret;
 
-	spin_lock(&dev_priv->mm.active_list_