diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index fe3c82339c8e..b35e6174fa7d 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -365,7 +365,7 @@ static int per_file_stats(int id, void *ptr, void *data)
 				continue;
 		}
 
-		if (obj->active) /* XXX per-vma statistic */
+		if (i915_vma_is_active(vma))
 			stats->active += vma->node.size;
 		else
 			stats->inactive += vma->node.size;
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index ce57c504aa25..51660ceae147 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -2347,7 +2347,6 @@ i915_gem_object_retire__read(struct i915_gem_active *active,
 	int idx = request->engine->id;
 	struct drm_i915_gem_object *obj =
 		container_of(active, struct drm_i915_gem_object, last_read[idx]);
-	struct i915_vma *vma;
 
 	GEM_BUG_ON((obj->active & (1 << idx)) == 0);
 
@@ -2359,12 +2358,9 @@ i915_gem_object_retire__read(struct i915_gem_active *active,
 	 * so that we don't steal from recently used but inactive objects
 	 * (unless we are forced to ofc!)
 	 */
-	list_move_tail(&obj->global_list, &request->i915->mm.bound_list);
-
-	list_for_each_entry(vma, &obj->vma_list, obj_link) {
-		if (!list_empty(&vma->vm_link))
-			list_move_tail(&vma->vm_link, &vma->vm->inactive_list);
-	}
+	if (obj->bind_count)
+		list_move_tail(&obj->global_list,
+			       &request->i915->mm.bound_list);
 
 	i915_gem_object_put(obj);
 }
@@ -2797,8 +2793,29 @@ static void __i915_vma_iounmap(struct i915_vma *vma)
 static int __i915_vma_unbind(struct i915_vma *vma, bool wait)
 {
 	struct drm_i915_gem_object *obj = vma->obj;
+	unsigned long active;
 	int ret;
 
+	/* First wait upon any activity as retiring the request may
+	 * have side-effects such as unpinning or even unbinding this vma.
+	 */
+	active = i915_vma_get_active(vma);
+	if (active && wait) {
+		int idx;
+
+		for_each_active(active, idx) {
+			ret = i915_gem_active_retire(&vma->last_read[idx],
+						   &vma->vm->dev->struct_mutex);
+			if (ret)
+				return ret;
+		}
+
+		GEM_BUG_ON(i915_vma_is_active(vma));
+	}
+
+	if (vma->pin_count)
+		return -EBUSY;
+
 	if (list_empty(&vma->obj_link))
 		return 0;
 
@@ -2807,18 +2824,9 @@ static int __i915_vma_unbind(struct i915_vma *vma, bool wait)
 		return 0;
 	}
 
-	if (vma->pin_count)
-		return -EBUSY;
-
 	GEM_BUG_ON(obj->bind_count == 0);
 	GEM_BUG_ON(!obj->pages);
 
-	if (wait) {
-		ret = i915_gem_object_wait_rendering(obj, false);
-		if (ret)
-			return ret;
-	}
-
 	if (vma->is_ggtt && vma->ggtt_view.type == I915_GGTT_VIEW_NORMAL) {
 		i915_gem_object_finish_gtt(obj);
 
@@ -3201,9 +3209,6 @@ i915_gem_object_flush_cpu_write_domain(struct drm_i915_gem_object *obj)
 int
 i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write)
 {
-	struct drm_device *dev = obj->base.dev;
-	struct drm_i915_private *dev_priv = to_i915(dev);
-	struct i915_ggtt *ggtt = &dev_priv->ggtt;
 	uint32_t old_write_domain, old_read_domains;
 	struct i915_vma *vma;
 	int ret;
@@ -3256,9 +3261,10 @@ i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write)
 
 	/* And bump the LRU for this access */
 	vma = i915_gem_obj_to_ggtt(obj);
-	if (vma && drm_mm_node_allocated(&vma->node) && !obj->active)
-		list_move_tail(&vma->vm_link,
-			       &ggtt->base.inactive_list);
+	if (vma &&
+	    drm_mm_node_allocated(&vma->node) &&
+	    !i915_vma_is_active(vma))
+		list_move_tail(&vma->vm_link, &vma->vm->inactive_list);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index 8bf20f52b79f..5e3b5054f72d 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -1154,7 +1154,13 @@ void i915_vma_move_to_active(struct i915_vma *vma,
 
 	obj->dirty = 1; /* be paranoid  */
 
-	/* Add a reference if we're newly entering the active list. */
+	/* Add a reference if we're newly entering the active list.
+	 * The order in which we add operations to the retirement queue is
+	 * vital here: mark_active adds to the start of the callback list,
+	 * such that subsequent callbacks are called first. Therefore we
+	 * add the active reference first and queue for it to be dropped
+	 * *last*.
+	 */
 	if (obj->active == 0)
 		i915_gem_object_get(obj);
 	obj->active |= 1 << idx;
@@ -1179,6 +1185,8 @@ void i915_vma_move_to_active(struct i915_vma *vma,
 		}
 	}
 
+	i915_vma_set_active(vma, idx);
+	i915_gem_active_set(&vma->last_read[idx], req);
 	list_move_tail(&vma->vm_link, &vma->vm->active_list);
 }
 
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index ad97892726ce..3e5d39dc7de0 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -3327,12 +3327,30 @@ void i915_gem_restore_gtt_mappings(struct drm_device *dev)
 	i915_ggtt_flush(dev_priv);
 }
 
+static void
+i915_vma_retire(struct i915_gem_active *active,
+		struct drm_i915_gem_request *rq)
+{
+	const unsigned int idx = rq->engine->id;
+	struct i915_vma *vma =
+		container_of(active, struct i915_vma, last_read[idx]);
+
+	GEM_BUG_ON(!i915_vma_has_active_engine(vma, idx));
+
+	i915_vma_clear_active(vma, idx);
+	if (i915_vma_is_active(vma))
+		return;
+
+	list_move_tail(&vma->vm_link, &vma->vm->inactive_list);
+}
+
 static struct i915_vma *
 __i915_gem_vma_create(struct drm_i915_gem_object *obj,
 		      struct i915_address_space *vm,
 		      const struct i915_ggtt_view *ggtt_view)
 {
 	struct i915_vma *vma;
+	int i;
 
 	if (WARN_ON(i915_is_ggtt(vm) != !!ggtt_view))
 		return ERR_PTR(-EINVAL);
@@ -3344,6 +3362,8 @@ __i915_gem_vma_create(struct drm_i915_gem_object *obj,
 	INIT_LIST_HEAD(&vma->vm_link);
 	INIT_LIST_HEAD(&vma->obj_link);
 	INIT_LIST_HEAD(&vma->exec_list);
+	for (i = 0; i < ARRAY_SIZE(vma->last_read); i++)
+		init_request_active(&vma->last_read[i], i915_vma_retire);
 	vma->vm = vm;
 	vma->obj = obj;
 	vma->is_ggtt = i915_is_ggtt(vm);
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
index cf8e3fc0692d..bfd3c112e33c 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.h
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
@@ -36,6 +36,8 @@
 
 #include <linux/io-mapping.h>
 
+#include "i915_gem_request.h"
+
 struct drm_i915_file_private;
 
 typedef uint32_t gen6_pte_t;
@@ -179,6 +181,9 @@ struct i915_vma {
 	struct i915_address_space *vm;
 	void __iomem *iomap;
 
+	unsigned int active;
+	struct i915_gem_active last_read[I915_NUM_ENGINES];
+
 	/** Flags and address space this VMA is bound to */
 #define GLOBAL_BIND	(1<<0)
 #define LOCAL_BIND	(1<<1)
@@ -222,6 +227,34 @@ struct i915_vma {
 #define DRM_I915_GEM_OBJECT_MAX_PIN_COUNT 0xf
 };
 
+static inline unsigned int i915_vma_get_active(const struct i915_vma *vma)
+{
+	return vma->active;
+}
+
+static inline bool i915_vma_is_active(const struct i915_vma *vma)
+{
+	return i915_vma_get_active(vma);
+}
+
+static inline void i915_vma_set_active(struct i915_vma *vma,
+				       unsigned int engine)
+{
+	vma->active |= BIT(engine);
+}
+
+static inline void i915_vma_clear_active(struct i915_vma *vma,
+					 unsigned int engine)
+{
+	vma->active &= ~BIT(engine);
+}
+
+static inline bool i915_vma_has_active_engine(const struct i915_vma *vma,
+					      unsigned int engine)
+{
+	return vma->active & BIT(engine);
+}
+
 struct i915_page_dma {
 	struct page *page;
 	union {