Merge branch 'for-airlied-next' of git://people.freedesktop.org/~mlankhorst/linux into drm-next

Merge the move to generic fences for TTM using drivers.

* 'for-airlied-next' of git://people.freedesktop.org/~mlankhorst/linux:
  drm/nouveau: use shared fences for readable objects
  drm/nouveau: Keep only a single list for validation.
  drm/ttm: use rcu in core ttm
  drm/vmwgfx: use rcu in vmw_user_dmabuf_synccpu_grab
  drm/radeon: use rcu waits in some ioctls
  drm/nouveau: use rcu in nouveau_gem_ioctl_cpu_prep
  drm/ttm: flip the switch, and convert to dma_fence
  drm/qxl: rework to new fence interface
  drm/nouveau: rework to new fence interface
  drm/vmwgfx: rework to new fence interface, v2
  drm/vmwgfx: get rid of different types of fence_flags entirely
  drm/radeon: use common fence implementation for fences, v4
  drm/ttm: kill off some members to ttm_validate_buffer
  drm/ttm: add interruptible parameter to ttm_eu_reserve_buffers
  drm/ttm: kill fence_lock
  drm/ttm: call ttm_bo_wait while inside a reservation
  drm/nouveau: require reservations for nouveau_fence_sync and nouveau_bo_fence
  drm/nouveau: add reservation to nouveau_gem_ioctl_cpu_prep

47 files changed, 1347 insertions, 1224 deletions

diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index 0591ca0734e3..eea74b127b03 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -88,13 +88,13 @@ nv10_bo_get_tile_region(struct drm_device *dev, int i)
 
 static void
 nv10_bo_put_tile_region(struct drm_device *dev, struct nouveau_drm_tile *tile,
-			struct nouveau_fence *fence)
+			struct fence *fence)
 {
 	struct nouveau_drm *drm = nouveau_drm(dev);
 
 	if (tile) {
 		spin_lock(&drm->tile.lock);
-		tile->fence = nouveau_fence_ref(fence);
+		tile->fence = (struct nouveau_fence *)fence_get(fence);
 		tile->used = false;
 		spin_unlock(&drm->tile.lock);
 	}
@@ -970,13 +970,14 @@ nouveau_bo_move_m2mf(struct ttm_buffer_object *bo, int evict, bool intr,
 	}
 
 	mutex_lock_nested(&cli->mutex, SINGLE_DEPTH_NESTING);
-	ret = nouveau_fence_sync(bo->sync_obj, chan);
+	ret = nouveau_fence_sync(nouveau_bo(bo), chan, true);
 	if (ret == 0) {
 		ret = drm->ttm.move(chan, bo, &bo->mem, new_mem);
 		if (ret == 0) {
 			ret = nouveau_fence_new(chan, false, &fence);
 			if (ret == 0) {
-				ret = ttm_bo_move_accel_cleanup(bo, fence,
+				ret = ttm_bo_move_accel_cleanup(bo,
+								&fence->base,
 								evict,
 								no_wait_gpu,
 								new_mem);
@@ -1167,8 +1168,9 @@ nouveau_bo_vm_cleanup(struct ttm_buffer_object *bo,
 {
 	struct nouveau_drm *drm = nouveau_bdev(bo->bdev);
 	struct drm_device *dev = drm->dev;
+	struct fence *fence = reservation_object_get_excl(bo->resv);
 
-	nv10_bo_put_tile_region(dev, *old_tile, bo->sync_obj);
+	nv10_bo_put_tile_region(dev, *old_tile, fence);
 	*old_tile = new_tile;
 }
 
@@ -1212,9 +1214,7 @@ nouveau_bo_move(struct ttm_buffer_object *bo, bool evict, bool intr,
 	}
 
 	/* Fallback to software copy. */
-	spin_lock(&bo->bdev->fence_lock);
 	ret = ttm_bo_wait(bo, true, intr, no_wait_gpu);
-	spin_unlock(&bo->bdev->fence_lock);
 	if (ret == 0)
 		ret = ttm_bo_move_memcpy(bo, evict, no_wait_gpu, new_mem);
 
@@ -1458,47 +1458,14 @@ nouveau_ttm_tt_unpopulate(struct ttm_tt *ttm)
 }
 
 void
-nouveau_bo_fence(struct nouveau_bo *nvbo, struct nouveau_fence *fence)
+nouveau_bo_fence(struct nouveau_bo *nvbo, struct nouveau_fence *fence, bool exclusive)
 {
-	struct nouveau_fence *new_fence = nouveau_fence_ref(fence);
-	struct nouveau_fence *old_fence = NULL;
+	struct reservation_object *resv = nvbo->bo.resv;
 
-	spin_lock(&nvbo->bo.bdev->fence_lock);
-	old_fence = nvbo->bo.sync_obj;
-	nvbo->bo.sync_obj = new_fence;
-	spin_unlock(&nvbo->bo.bdev->fence_lock);
-
-	nouveau_fence_unref(&old_fence);
-}
-
-static void
-nouveau_bo_fence_unref(void **sync_obj)
-{
-	nouveau_fence_unref((struct nouveau_fence **)sync_obj);
-}
-
-static void *
-nouveau_bo_fence_ref(void *sync_obj)
-{
-	return nouveau_fence_ref(sync_obj);
-}
-
-static bool
-nouveau_bo_fence_signalled(void *sync_obj)
-{
-	return nouveau_fence_done(sync_obj);
-}
-
-static int
-nouveau_bo_fence_wait(void *sync_obj, bool lazy, bool intr)
-{
-	return nouveau_fence_wait(sync_obj, lazy, intr);
-}
-
-static int
-nouveau_bo_fence_flush(void *sync_obj)
-{
-	return 0;
+	if (exclusive)
+		reservation_object_add_excl_fence(resv, &fence->base);
+	else if (fence)
+		reservation_object_add_shared_fence(resv, &fence->base);
 }
 
 struct ttm_bo_driver nouveau_bo_driver = {
@@ -1511,11 +1478,6 @@ struct ttm_bo_driver nouveau_bo_driver = {
 	.move_notify = nouveau_bo_move_ntfy,
 	.move = nouveau_bo_move,
 	.verify_access = nouveau_bo_verify_access,
-	.sync_obj_signaled = nouveau_bo_fence_signalled,
-	.sync_obj_wait = nouveau_bo_fence_wait,
-	.sync_obj_flush = nouveau_bo_fence_flush,
-	.sync_obj_unref = nouveau_bo_fence_unref,
-	.sync_obj_ref = nouveau_bo_fence_ref,
 	.fault_reserve_notify = &nouveau_ttm_fault_reserve_notify,
 	.io_mem_reserve = &nouveau_ttm_io_mem_reserve,
 	.io_mem_free = &nouveau_ttm_io_mem_free,
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.h b/drivers/gpu/drm/nouveau/nouveau_bo.h
index 4ef88e84a694..ae95b2d43b36 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.h
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.h
@@ -78,7 +78,7 @@ u16  nouveau_bo_rd16(struct nouveau_bo *, unsigned index);
 void nouveau_bo_wr16(struct nouveau_bo *, unsigned index, u16 val);
 u32  nouveau_bo_rd32(struct nouveau_bo *, unsigned index);
 void nouveau_bo_wr32(struct nouveau_bo *, unsigned index, u32 val);
-void nouveau_bo_fence(struct nouveau_bo *, struct nouveau_fence *);
+void nouveau_bo_fence(struct nouveau_bo *, struct nouveau_fence *, bool exclusive);
 int  nouveau_bo_validate(struct nouveau_bo *, bool interruptible,
 			 bool no_wait_gpu);
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_display.c b/drivers/gpu/drm/nouveau/nouveau_display.c
index 65b4fd53dd4e..a9ec525c0994 100644
--- a/drivers/gpu/drm/nouveau/nouveau_display.c
+++ b/drivers/gpu/drm/nouveau/nouveau_display.c
@@ -658,7 +658,7 @@ nouveau_page_flip_emit(struct nouveau_channel *chan,
 	spin_unlock_irqrestore(&dev->event_lock, flags);
 
 	/* Synchronize with the old framebuffer */
-	ret = nouveau_fence_sync(old_bo->bo.sync_obj, chan);
+	ret = nouveau_fence_sync(old_bo, chan, false);
 	if (ret)
 		goto fail;
 
@@ -717,19 +717,24 @@ nouveau_crtc_page_flip(struct drm_crtc *crtc, struct drm_framebuffer *fb,
 	}
 
 	mutex_lock(&cli->mutex);
-
-	/* synchronise rendering channel with the kernel's channel */
-	spin_lock(&new_bo->bo.bdev->fence_lock);
-	fence = nouveau_fence_ref(new_bo->bo.sync_obj);
-	spin_unlock(&new_bo->bo.bdev->fence_lock);
-	ret = nouveau_fence_sync(fence, chan);
-	nouveau_fence_unref(&fence);
+	ret = ttm_bo_reserve(&new_bo->bo, true, false, false, NULL);
 	if (ret)
 		goto fail_unpin;
 
-	ret = ttm_bo_reserve(&old_bo->bo, true, false, false, NULL);
-	if (ret)
+	/* synchronise rendering channel with the kernel's channel */
+	ret = nouveau_fence_sync(new_bo, chan, false);
+	if (ret) {
+		ttm_bo_unreserve(&new_bo->bo);
 		goto fail_unpin;
+	}
+
+	if (new_bo != old_bo) {
+		ttm_bo_unreserve(&new_bo->bo);
+
+		ret = ttm_bo_reserve(&old_bo->bo, true, false, false, NULL);
+		if (ret)
+			goto fail_unpin;
+	}
 
 	/* Initialize a page flip struct */
 	*s = (struct nouveau_page_flip_state)
@@ -775,7 +780,7 @@ nouveau_crtc_page_flip(struct drm_crtc *crtc, struct drm_framebuffer *fb,
 	/* Update the crtc struct and cleanup */
 	crtc->primary->fb = fb;
 
-	nouveau_bo_fence(old_bo, fence);
+	nouveau_bo_fence(old_bo, fence, false);
 	ttm_bo_unreserve(&old_bo->bo);
 	if (old_bo != new_bo)
 		nouveau_bo_unpin(old_bo);
diff --git a/drivers/gpu/drm/nouveau/nouveau_fence.c b/drivers/gpu/drm/nouveau/nouveau_fence.c
index 0a93114158cd..decfe6c4ac07 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fence.c
+++ b/drivers/gpu/drm/nouveau/nouveau_fence.c
@@ -28,6 +28,7 @@
 
 #include <linux/ktime.h>
 #include <linux/hrtimer.h>
+#include <trace/events/fence.h>
 
 #include <nvif/notify.h>
 #include <nvif/event.h>
@@ -36,123 +37,210 @@
 #include "nouveau_dma.h"
 #include "nouveau_fence.h"
 
-struct fence_work {
-	struct work_struct base;
-	struct list_head head;
-	void (*func)(void *);
-	void *data;
-};
+static const struct fence_ops nouveau_fence_ops_uevent;
+static const struct fence_ops nouveau_fence_ops_legacy;
+
+static inline struct nouveau_fence *
+from_fence(struct fence *fence)
+{
+	return container_of(fence, struct nouveau_fence, base);
+}
+
+static inline struct nouveau_fence_chan *
+nouveau_fctx(struct nouveau_fence *fence)
+{
+	return container_of(fence->base.lock, struct nouveau_fence_chan, lock);
+}
 
 static void
 nouveau_fence_signal(struct nouveau_fence *fence)
 {
-	struct fence_work *work, *temp;
+	fence_signal_locked(&fence->base);
+	list_del(&fence->head);
+
+	if (test_bit(FENCE_FLAG_USER_BITS, &fence->base.flags)) {
+		struct nouveau_fence_chan *fctx = nouveau_fctx(fence);
 
-	list_for_each_entry_safe(work, temp, &fence->work, head) {
-		schedule_work(&work->base);
-		list_del(&work->head);
+		if (!--fctx->notify_ref)
+			nvif_notify_put(&fctx->notify);
 	}
 
-	fence->channel = NULL;
-	list_del(&fence->head);
+	fence_put(&fence->base);
+}
+
+static struct nouveau_fence *
+nouveau_local_fence(struct fence *fence, struct nouveau_drm *drm) {
+	struct nouveau_fence_priv *priv = (void*)drm->fence;
+
+	if (fence->ops != &nouveau_fence_ops_legacy &&
+	    fence->ops != &nouveau_fence_ops_uevent)
+		return NULL;
+
+	if (fence->context < priv->context_base ||
+	    fence->context >= priv->context_base + priv->contexts)
+		return NULL;
+
+	return from_fence(fence);
 }
 
 void
 nouveau_fence_context_del(struct nouveau_fence_chan *fctx)
 {
-	struct nouveau_fence *fence, *fnext;
-	spin_lock(&fctx->lock);
-	list_for_each_entry_safe(fence, fnext, &fctx->pending, head) {
+	struct nouveau_fence *fence;
+
+	nvif_notify_fini(&fctx->notify);
+
+	spin_lock_irq(&fctx->lock);
+	while (!list_empty(&fctx->pending)) {
+		fence = list_entry(fctx->pending.next, typeof(*fence), head);
+
+		nouveau_fence_signal(fence);
+		fence->channel = NULL;
+	}
+	spin_unlock_irq(&fctx->lock);
+}
+
+static void
+nouveau_fence_update(struct nouveau_channel *chan, struct nouveau_fence_chan *fctx)
+{
+	struct nouveau_fence *fence;
+
+	u32 seq = fctx->read(chan);
+
+	while (!list_empty(&fctx->pending)) {
+		fence = list_entry(fctx->pending.next, typeof(*fence), head);
+
+		if ((int)(seq - fence->base.seqno) < 0)
+			return;
+
 		nouveau_fence_signal(fence);
 	}
-	spin_unlock(&fctx->lock);
+}
+
+static int
+nouveau_fence_wait_uevent_handler(struct nvif_notify *notify)
+{
+	struct nouveau_fence_chan *fctx =
+		container_of(notify, typeof(*fctx), notify);
+	unsigned long flags;
+
+	spin_lock_irqsave(&fctx->lock, flags);
+	if (!list_empty(&fctx->pending)) {
+		struct nouveau_fence *fence;
+
+		fence = list_entry(fctx->pending.next, typeof(*fence), head);
+		nouveau_fence_update(fence->channel, fctx);
+	}
+	spin_unlock_irqrestore(&fctx->lock, flags);
+
+	/* Always return keep here. NVIF refcount is handled with nouveau_fence_update */
+	return NVIF_NOTIFY_KEEP;
 }
 
 void
-nouveau_fence_context_new(struct nouveau_fence_chan *fctx)
+nouveau_fence_context_new(struct nouveau_channel *chan, struct nouveau_fence_chan *fctx)
 {
+	struct nouveau_fence_priv *priv = (void*)chan->drm->fence;
+	int ret;
+
 	INIT_LIST_HEAD(&fctx->flip);
 	INIT_LIST_HEAD(&fctx->pending);
 	spin_lock_init(&fctx->lock);
+	fctx->context = priv->context_base + chan->chid;
+
+	if (!priv->uevent)
+		return;
+
+	ret = nvif_notify_init(chan->object, NULL,
+			 nouveau_fence_wait_uevent_handler, false,
+			 G82_CHANNEL_DMA_V0_NTFY_UEVENT,
+			 &(struct nvif_notify_uevent_req) { },
+			 sizeof(struct nvif_notify_uevent_req),
+			 sizeof(struct nvif_notify_uevent_rep),
+			 &fctx->notify);
+
+	WARN_ON(ret);
 }
 
+struct nouveau_fence_work {
+	struct work_struct work;
+	struct fence_cb cb;
+	void (*func)(void *);
+	void *data;
+};
+
 static void
 nouveau_fence_work_handler(struct work_struct *kwork)
 {
-	struct fence_work *work = container_of(kwork, typeof(*work), base);
+	struct nouveau_fence_work *work = container_of(kwork, typeof(*work), work);
 	work->func(work->data);
 	kfree(work);
 }
 
+static void nouveau_fence_work_cb(struct fence *fence, struct fence_cb *cb)
+{
+	struct nouveau_fence_work *work = container_of(cb, typeof(*work), cb);
+
+	schedule_work(&work->work);
+}
+
 void
-nouveau_fence_work(struct nouveau_fence *fence,
+nouveau_fence_work(struct fence *fence,
 		   void (*func)(void *), void *data)
 {
-	struct nouveau_channel *chan = fence->channel;
-	struct nouveau_fence_chan *fctx;
-	struct fence_work *work = NULL;
+	struct nouveau_fence_work *work;
 
-	if (nouveau_fence_done(fence)) {
-		func(data);
-		return;
-	}
+	if (fence_is_signaled(fence))
+		goto err;
 
-	fctx = chan->fence;
 	work = kmalloc(sizeof(*work), GFP_KERNEL);
 	if (!work) {
-		WARN_ON(nouveau_fence_wait(fence, false, false));
-		func(data);
-		return;
+		WARN_ON(nouveau_fence_wait((struct nouveau_fence *)fence,
+					   false, false));
+		goto err;
 	}
 
-	spin_lock(&fctx->lock);
-	if (!fence->channel) {
-		spin_unlock(&fctx->lock);
-		kfree(work);
-		func(data);
-		return;
-	}
-
-	INIT_WORK(&work->base, nouveau_fence_work_handler);
+	INIT_WORK(&work->work, nouveau_fence_work_handler);
 	work->func = func;
 	work->data = data;
-	list_add(&work->head, &fence->work);
-	spin_unlock(&fctx->lock);
-}
-
-static void
-nouveau_fence_update(struct nouveau_channel *chan)
-{
-	struct nouveau_fence_chan *fctx = chan->fence;
-	struct nouveau_fence *fence, *fnext;
 
-	spin_lock(&fctx->lock);
-	list_for_each_entry_safe(fence, fnext, &fctx->pending, head) {
-		if (fctx->read(chan) < fence->sequence)
-			break;
+	if (fence_add_callback(fence, &work->cb, nouveau_fence_work_cb) < 0)
+		goto err_free;
+	return;
 
-		nouveau_fence_signal(fence);
-		nouveau_fence_unref(&fence);
-	}
-	spin_unlock(&fctx->lock);
+err_free:
+	kfree(work);
+err:
+	func(data);
 }
 
 int
 nouveau_fence_emit(struct nouveau_fence *fence, struct nouveau_channel *chan)
 {
 	struct nouveau_fence_chan *fctx = chan->fence;
+	struct nouveau_fence_priv *priv = (void*)chan->drm->fence;
 	int ret;
 
 	fence->channel  = chan;
 	fence->timeout  = jiffies + (15 * HZ);
-	fence->sequence = ++fctx->sequence;
 
+	if (priv->uevent)
+		fence_init(&fence->base, &nouveau_fence_ops_uevent,
+			   &fctx->lock,
+			   priv->context_base + chan->chid, ++fctx->sequence);
+	else
+		fence_init(&fence->base, &nouveau_fence_ops_legacy,
+			   &fctx->lock,
+			   priv->context_base + chan->chid, ++fctx->sequence);
+
+	trace_fence_emit(&fence->base);
 	ret = fctx->emit(fence);
 	if (!ret) {
-		kref_get(&fence->kref);
-		spin_lock(&fctx->lock);
+		fence_get(&fence->base);
+		spin_lock_irq(&fctx->lock);
+		nouveau_fence_update(chan, fctx);
 		list_add_tail(&fence->head, &fctx->pending);
-		spin_unlock(&fctx->lock);
+		spin_unlock_irq(&fctx->lock);
 	}
 
 	return ret;
@@ -161,114 +249,70 @@ nouveau_fence_emit(struct nouveau_fence *fence, struct nouveau_channel *chan)
 bool
 nouveau_fence_done(struct nouveau_fence *fence)
 {
-	if (fence->channel)
-		nouveau_fence_update(fence->channel);
-	return !fence->channel;
-}
+	if (fence->base.ops == &nouveau_fence_ops_legacy ||
+	    fence->base.ops == &nouveau_fence_ops_uevent) {
+		struct nouveau_fence_chan *fctx = nouveau_fctx(fence);
+		unsigned long flags;
 
-struct nouveau_fence_wait {
-	struct nouveau_fence_priv *priv;
-	struct nvif_notify notify;
-};
+		if (test_bit(FENCE_FLAG_SIGNALED_BIT, &fence->base.flags))
+			return true;
 
-static int
-nouveau_fence_wait_uevent_handler(struct nvif_notify *notify)
-{
-	struct nouveau_fence_wait *wait =
-		container_of(notify, typeof(*wait), notify);
-	wake_up_all(&wait->priv->waiting);
-	return NVIF_NOTIFY_KEEP;
+		spin_lock_irqsave(&fctx->lock, flags);
+		nouveau_fence_update(fence->channel, fctx);
+		spin_unlock_irqrestore(&fctx->lock, flags);
+	}
+	return fence_is_signaled(&fence->base);
 }
 
-static int
-nouveau_fence_wait_uevent(struct nouveau_fence *fence, bool intr)
-
+static long
+nouveau_fence_wait_legacy(struct fence *f, bool intr, long wait)
 {
-	struct nouveau_channel *chan = fence->channel;
-	struct nouveau_fence_priv *priv = chan->drm->fence;
-	struct nouveau_fence_wait wait = { .priv = priv };
-	int ret = 0;
+	struct nouveau_fence *fence = from_fence(f);
+	unsigned long sleep_time = NSEC_PER_MSEC / 1000;
+	unsigned long t = jiffies, timeout = t + wait;
 
-	ret = nvif_notify_init(chan->object, NULL,
-			       nouveau_fence_wait_uevent_handler, false,
-			       G82_CHANNEL_DMA_V0_NTFY_UEVENT,
-			       &(struct nvif_notify_uevent_req) {
-			       },
-			       sizeof(struct nvif_notify_uevent_req),
-			       sizeof(struct nvif_notify_uevent_rep),
-			       &wait.notify);
-	if (ret)
-		return ret;
+	while (!nouveau_fence_done(fence)) {
+		ktime_t kt;
 
-	nvif_notify_get(&wait.notify);
-
-	if (fence->timeout) {
-		unsigned long timeout = fence->timeout - jiffies;
-
-		if (time_before(jiffies, fence->timeout)) {
-			if (intr) {
-				ret = wait_event_interruptible_timeout(
-						priv->waiting,
-						nouveau_fence_done(fence),
-						timeout);
-			} else {
-				ret = wait_event_timeout(priv->waiting,
-						nouveau_fence_done(fence),
-						timeout);
-			}
-		}
+		t = jiffies;
 
-		if (ret >= 0) {
-			fence->timeout = jiffies + ret;
-			if (time_after_eq(jiffies, fence->timeout))
-				ret = -EBUSY;
-		}
-	} else {
-		if (intr) {
-			ret = wait_event_interruptible(priv->waiting,
-					nouveau_fence_done(fence));
-		} else {
-			wait_event(priv->waiting, nouveau_fence_done(fence));
+		if (wait != MAX_SCHEDULE_TIMEOUT && time_after_eq(t, timeout)) {
+			__set_current_state(TASK_RUNNING);
+			return 0;
 		}
+
+		__set_current_state(intr ? TASK_INTERRUPTIBLE :
+					   TASK_UNINTERRUPTIBLE);
+
+		kt = ktime_set(0, sleep_time);
+		schedule_hrtimeout(&kt, HRTIMER_MODE_REL);
+		sleep_time *= 2;
+		if (sleep_time > NSEC_PER_MSEC)
+			sleep_time = NSEC_PER_MSEC;
+
+		if (intr && signal_pending(current))
+			return -ERESTARTSYS;
 	}
 
-	nvif_notify_fini(&wait.notify);
-	if (unlikely(ret < 0))
-		return ret;
+	__set_current_state(TASK_RUNNING);
 
-	return 0;
+	return timeout - t;
 }
 
-int
-nouveau_fence_wait(struct nouveau_fence *fence, bool lazy, bool intr)
+static int
+nouveau_fence_wait_busy(struct nouveau_fence *fence, bool intr)
 {
-	struct nouveau_channel *chan = fence->channel;
-	struct nouveau_fence_priv *priv = chan ? chan->drm->fence : NULL;
-	unsigned long sleep_time = NSEC_PER_MSEC / 1000;
-	ktime_t t;
 	int ret = 0;
 
-	while (priv && priv->uevent && lazy && !nouveau_fence_done(fence)) {
-		ret = nouveau_fence_wait_uevent(fence, intr);
-		if (ret < 0)
-			return ret;
-	}
-
 	while (!nouveau_fence_done(fence)) {
-		if (fence->timeout && time_after_eq(jiffies, fence->timeout)) {
+		if (time_after_eq(jiffies, fence->timeout)) {
 			ret = -EBUSY;
 			break;
 		}
 
-		__set_current_state(intr ? TASK_INTERRUPTIBLE :
-					   TASK_UNINTERRUPTIBLE);
-		if (lazy) {
-			t = ktime_set(0, sleep_time);
-			schedule_hrtimeout(&t, HRTIMER_MODE_REL);
-			sleep_time *= 2;
-			if (sleep_time > NSEC_PER_MSEC)
-				sleep_time = NSEC_PER_MSEC;
-		}
+		__set_current_state(intr ?
+				    TASK_INTERRUPTIBLE :
+				    TASK_UNINTERRUPTIBLE);
 
 		if (intr && signal_pending(current)) {
 			ret = -ERESTARTSYS;
@@ -281,47 +325,86 @@ nouveau_fence_wait(struct nouveau_fence *fence, bool lazy, bool intr)
 }
 
 int
-nouveau_fence_sync(struct nouveau_fence *fence, struct nouveau_channel *chan)
+nouveau_fence_wait(struct nouveau_fence *fence, bool lazy, bool intr)
 {
-	struct nouveau_fence_chan *fctx = chan->fence;
-	struct nouveau_channel *prev;
-	int ret = 0;
+	long ret;
 
-	prev = fence ? fence->channel : NULL;
-	if (prev) {
-		if (unlikely(prev != chan && !nouveau_fence_done(fence))) {
-			ret = fctx->sync(fence, prev, chan);
-			if (unlikely(ret))
-				ret = nouveau_fence_wait(fence, true, false);
-		}
-	}
+	if (!lazy)
+		return nouveau_fence_wait_busy(fence, intr);
 
-	return ret;
+	ret = fence_wait_timeout(&fence->base, intr, 15 * HZ);
+	if (ret < 0)
+		return ret;
+	else if (!ret)
+		return -EBUSY;
+	else
+		return 0;
 }
 
-static void
-nouveau_fence_del(struct kref *kref)
+int
+nouveau_fence_sync(struct nouveau_bo *nvbo, struct nouveau_channel *chan, bool exclusive)
 {
-	struct nouveau_fence *fence = container_of(kref, typeof(*fence), kref);
-	kfree(fence);
+	struct nouveau_fence_chan *fctx = chan->fence;
+	struct fence *fence;
+	struct reservation_object *resv = nvbo->bo.resv;
+	struct reservation_object_list *fobj;
+	struct nouveau_fence *f;
+	int ret = 0, i;
+
+	if (!exclusive) {
+		ret = reservation_object_reserve_shared(resv);
+
+		if (ret)
+			return ret;
+	}
+
+	fobj = reservation_object_get_list(resv);
+	fence = reservation_object_get_excl(resv);
+
+	if (fence && (!exclusive || !fobj || !fobj->shared_count)) {
+		struct nouveau_channel *prev = NULL;
+
+		f = nouveau_local_fence(fence, chan->drm);
+		if (f)
+			prev = f->channel;
+
+		if (!prev || (prev != chan && (ret = fctx->sync(f, prev, chan))))
+			ret = fence_wait(fence, true);
+
+		return ret;
+	}
+
+	if (!exclusive || !fobj)
+		return ret;
+
+	for (i = 0; i < fobj->shared_count && !ret; ++i) {
+		struct nouveau_channel *prev = NULL;
+
+		fence = rcu_dereference_protected(fobj->shared[i],
+						reservation_object_held(resv));
+
+		f = nouveau_local_fence(fence, chan->drm);
+		if (f)
+			prev = f->channel;
+
+		if (!prev || (ret = fctx->sync(f, prev, chan)))
+			ret = fence_wait(fence, true);
+
+		if (ret)
+			break;
+	}
+
+	return ret;
 }
 
 void
 nouveau_fence_unref(struct nouveau_fence **pfence)
 {
 	if (*pfence)
-		kref_put(&(*pfence)->kref, nouveau_fence_del);
+		fence_put(&(*pfence)->base);
 	*pfence = NULL;
 }
 
-struct nouveau_fence *
-nouveau_fence_ref(struct nouveau_fence *fence)
-{
-	if (fence)
-		kref_get(&fence->kref);
-	return fence;
-}
-
 int
 nouveau_fence_new(struct nouveau_channel *chan, bool sysmem,
 		  struct nouveau_fence **pfence)
@@ -336,9 +419,7 @@ nouveau_fence_new(struct nouveau_channel *chan, bool sysmem,
 	if (!fence)
 		return -ENOMEM;
 
-	INIT_LIST_HEAD(&fence->work);
 	fence->sysmem = sysmem;
-	kref_init(&fence->kref);
 
 	ret = nouveau_fence_emit(fence, chan);
 	if (ret)
@@ -347,3 +428,92 @@ nouveau_fence_new(struct nouveau_channel *chan, bool sysmem,
 	*pfence = fence;
 	return ret;
 }
+
+static const char *nouveau_fence_get_get_driver_name(struct fence *fence)
+{
+	return "nouveau";
+}
+
+static const char *nouveau_fence_get_timeline_name(struct fence *f)
+{
+	struct nouveau_fence *fence = from_fence(f);
+	struct nouveau_fence_chan *fctx = nouveau_fctx(fence);
+
+	return fence->channel ? fctx->name : "dead channel";
+}
+
+/*
+ * In an ideal world, read would not assume the channel context is still alive.
+ * This function may be called from another device, running into free memory as a
+ * result. The drm node should still be there, so we can derive the index from
+ * the fence context.
+ */
+static bool nouveau_fence_is_signaled(struct fence *f)
+{
+	struct nouveau_fence *fence = from_fence(f);
+	struct nouveau_fence_chan *fctx = nouveau_fctx(fence);
+	struct nouveau_channel *chan = fence->channel;
+
+	return (int)(fctx->read(chan) - fence->base.seqno) >= 0;
+}
+
+static bool nouveau_fence_no_signaling(struct fence *f)
+{
+	struct nouveau_fence *fence = from_fence(f);
+
+	/*
+	 * caller should have a reference on the fence,
+	 * else fence could get freed here
+	 */
+	WARN_ON(atomic_read(&fence->base.refcount.refcount) <= 1);
+
+	/*
+	 * This needs uevents to work correctly, but fence_add_callback relies on
+	 * being able to enable signaling. It will still get signaled eventually,
+	 * just not right away.
+	 */
+	if (nouveau_fence_is_signaled(f)) {
+		list_del(&fence->head);
+
+		fence_put(&fence->base);
+		return false;
+	}
+
+	return true;
+}
+
+static const struct fence_ops nouveau_fence_ops_legacy = {
+	.get_driver_name = nouveau_fence_get_get_driver_name,
+	.get_timeline_name = nouveau_fence_get_timeline_name,
+	.enable_signaling = nouveau_fence_no_signaling,
+	.signaled = nouveau_fence_is_signaled,
+	.wait = nouveau_fence_wait_legacy,
+	.release = NULL
+};
+
+static bool nouveau_fence_enable_signaling(struct fence *f)
+{
+	struct nouveau_fence *fence = from_fence(f);
+	struct nouveau_fence_chan *fctx = nouveau_fctx(fence);
+	bool ret;
+
+	if (!fctx->notify_ref++)
+		nvif_notify_get(&fctx->notify);
+
+	ret = nouveau_fence_no_signaling(f);
+	if (ret)
+		set_bit(FENCE_FLAG_USER_BITS, &fence->base.flags);
+	else if (!--fctx->notify_ref)
+		nvif_notify_put(&fctx->notify);
+
+	return ret;
+}
+
+static const struct fence_ops nouveau_fence_ops_uevent = {
+	.get_driver_name = nouveau_fence_get_get_driver_name,
+	.get_timeline_name = nouveau_fence_get_timeline_name,
+	.enable_signaling = nouveau_fence_enable_signaling,
+	.signaled = nouveau_fence_is_signaled,
+	.wait = fence_default_wait,
+	.release = NULL
+};
diff --git a/drivers/gpu/drm/nouveau/nouveau_fence.h b/drivers/gpu/drm/nouveau/nouveau_fence.h
index c57bb61da58c..986c8135e564 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fence.h
+++ b/drivers/gpu/drm/nouveau/nouveau_fence.h
@@ -1,33 +1,35 @@
 #ifndef __NOUVEAU_FENCE_H__
 #define __NOUVEAU_FENCE_H__
 
+#include <linux/fence.h>
+#include <nvif/notify.h>
+
 struct nouveau_drm;
+struct nouveau_bo;
 
 struct nouveau_fence {
+	struct fence base;
+
 	struct list_head head;
-	struct list_head work;
-	struct kref kref;
 
 	bool sysmem;
 
 	struct nouveau_channel *channel;
 	unsigned long timeout;
-	u32 sequence;
 };
 
 int  nouveau_fence_new(struct nouveau_channel *, bool sysmem,
 		       struct nouveau_fence **);
-struct nouveau_fence *
-nouveau_fence_ref(struct nouveau_fence *);
 void nouveau_fence_unref(struct nouveau_fence **);
 
 int  nouveau_fence_emit(struct nouveau_fence *, struct nouveau_channel *);
 bool nouveau_fence_done(struct nouveau_fence *);
-void nouveau_fence_work(struct nouveau_fence *, void (*)(void *), void *);
+void nouveau_fence_work(struct fence *, void (*)(void *), void *);
 int  nouveau_fence_wait(struct nouveau_fence *, bool lazy, bool intr);
-int  nouveau_fence_sync(struct nouveau_fence *, struct nouveau_channel *);
+int  nouveau_fence_sync(struct nouveau_bo *, struct nouveau_channel *, bool exclusive);
 
 struct nouveau_fence_chan {
+	spinlock_t lock;
 	struct list_head pending;
 	struct list_head flip;
 
@@ -38,8 +40,12 @@ struct nouveau_fence_chan {
 	int  (*emit32)(struct nouveau_channel *, u64, u32);
 	int  (*sync32)(struct nouveau_channel *, u64, u32);
 
-	spinlock_t lock;
 	u32 sequence;
+	u32 context;
+	char name[24];
+
+	struct nvif_notify notify;
+	int notify_ref;
 };
 
 struct nouveau_fence_priv {
@@ -49,13 +55,13 @@ struct nouveau_fence_priv {
 	int  (*context_new)(struct nouveau_channel *);
 	void (*context_del)(struct nouveau_channel *);
 
-	wait_queue_head_t waiting;
+	u32 contexts, context_base;
 	bool uevent;
 };
 
 #define nouveau_fence(drm) ((struct nouveau_fence_priv *)(drm)->fence)
 
-void nouveau_fence_context_new(struct nouveau_fence_chan *);
+void nouveau_fence_context_new(struct nouveau_channel *, struct nouveau_fence_chan *);
 void nouveau_fence_context_del(struct nouveau_fence_chan *);
 
 int nv04_fence_create(struct nouveau_drm *);
diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c b/drivers/gpu/drm/nouveau/nouveau_gem.c
index 292a677bfed4..b7dbd16904e0 100644
--- a/drivers/gpu/drm/nouveau/nouveau_gem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_gem.c
@@ -98,17 +98,23 @@ static void
 nouveau_gem_object_unmap(struct nouveau_bo *nvbo, struct nouveau_vma *vma)
 {
 	const bool mapped = nvbo->bo.mem.mem_type != TTM_PL_SYSTEM;
-	struct nouveau_fence *fence = NULL;
+	struct reservation_object *resv = nvbo->bo.resv;
+	struct reservation_object_list *fobj;
+	struct fence *fence = NULL;
+
+	fobj = reservation_object_get_list(resv);
 
 	list_del(&vma->head);
 
-	if (mapped) {
-		spin_lock(&nvbo->bo.bdev->fence_lock);
-		fence = nouveau_fence_ref(nvbo->bo.sync_obj);
-		spin_unlock(&nvbo->bo.bdev->fence_lock);
-	}
+	if (fobj && fobj->shared_count > 1)
+		ttm_bo_wait(&nvbo->bo, true, false, false);
+	else if (fobj && fobj->shared_count == 1)
+		fence = rcu_dereference_protected(fobj->shared[0],
+						reservation_object_held(resv));
+	else
+		fence = reservation_object_get_excl(nvbo->bo.resv);
 
-	if (fence) {
+	if (fence && mapped) {
 		nouveau_fence_work(fence, nouveau_gem_object_delete, vma);
 	} else {
 		if (mapped)
@@ -116,7 +122,6 @@ nouveau_gem_object_unmap(struct nouveau_bo *nvbo, struct nouveau_vma *vma)
 		nouveau_vm_put(vma);
 		kfree(vma);
 	}
-	nouveau_fence_unref(&fence);
 }
 
 void
@@ -288,24 +293,23 @@ nouveau_gem_set_domain(struct drm_gem_object *gem, uint32_t read_domains,
 }
 
 struct validate_op {
-	struct list_head vram_list;
-	struct list_head gart_list;
-	struct list_head both_list;
+	struct list_head list;
 	struct ww_acquire_ctx ticket;
 };
 
 static void
-validate_fini_list(struct list_head *list, struct nouveau_fence *fence,
-		   struct ww_acquire_ctx *ticket)
+validate_fini_no_ticket(struct validate_op *op, struct nouveau_fence *fence,
+			struct drm_nouveau_gem_pushbuf_bo *pbbo)
 {
-	struct list_head *entry, *tmp;
 	struct nouveau_bo *nvbo;
+	struct drm_nouveau_gem_pushbuf_bo *b;
 
-	list_for_each_safe(entry, tmp, list) {
-		nvbo = list_entry(entry, struct nouveau_bo, entry);
+	while (!list_empty(&op->list)) {
+		nvbo = list_entry(op->list.next, struct nouveau_bo, entry);
+		b = &pbbo[nvbo->pbbo_index];
 
 		if (likely(fence))
-			nouveau_bo_fence(nvbo, fence);
+			nouveau_bo_fence(nvbo, fence, !!b->write_domains);
 
 		if (unlikely(nvbo->validate_mapped)) {
 			ttm_bo_kunmap(&nvbo->kmap);
@@ -314,23 +318,16 @@ validate_fini_list(struct list_head *list, struct nouveau_fence *fence,
 
 		list_del(&nvbo->entry);
 		nvbo->reserved_by = NULL;
-		ttm_bo_unreserve_ticket(&nvbo->bo, ticket);
+		ttm_bo_unreserve_ticket(&nvbo->bo, &op->ticket);
 		drm_gem_object_unreference_unlocked(&nvbo->gem);
 	}
 }
 
 static void
-validate_fini_no_ticket(struct validate_op *op, struct nouveau_fence *fence)
+validate_fini(struct validate_op *op, struct nouveau_fence *fence,
+	      struct drm_nouveau_gem_pushbuf_bo *pbbo)
 {
-	validate_fini_list(&op->vram_list, fence, &op->ticket);
-	validate_fini_list(&op->gart_list, fence, &op->ticket);
-	validate_fini_list(&op->both_list, fence, &op->ticket);
-}
-
-static void
-validate_fini(struct validate_op *op, struct nouveau_fence *fence)
-{
-	validate_fini_no_ticket(op, fence);
+	validate_fini_no_ticket(op, fence, pbbo);
 	ww_acquire_fini(&op->ticket);
 }
 
@@ -344,6 +341,9 @@ validate_init(struct nouveau_channel *chan, struct drm_file *file_priv,
 	int trycnt = 0;
 	int ret, i;
 	struct nouveau_bo *res_bo = NULL;
+	LIST_HEAD(gart_list);
+	LIST_HEAD(vram_list);
+	LIST_HEAD(both_list);
 
 	ww_acquire_init(&op->ticket, &reservation_ww_class);
 retry:
@@ -360,9 +360,8 @@ retry:
 		gem = drm_gem_object_lookup(dev, file_priv, b->handle);
 		if (!gem) {
 			NV_PRINTK(error, cli, "Unknown handle 0x%08x\n", b->handle);
-			ww_acquire_done(&op->ticket);
-			validate_fini(op, NULL);
-			return -ENOENT;
+			ret = -ENOENT;
+			break;
 		}
 		nvbo = nouveau_gem_object(gem);
 		if (nvbo == res_bo) {
@@ -375,14 +374,16 @@ retry:
 			NV_PRINTK(error, cli, "multiple instances of buffer %d on "
 				      "validation list\n", b->handle);
 			drm_gem_object_unreference_unlocked(gem);
-			ww_acquire_done(&op->ticket);
-			validate_fini(op, NULL);
-			return -EINVAL;
+			ret = -EINVAL;
+			break;
 		}
 
 		ret = ttm_bo_reserve(&nvbo->bo, true, false, true, &op->ticket);
 		if (ret) {
-			validate_fini_no_ticket(op, NULL);
+			list_splice_tail_init(&vram_list, &op->list);
+			list_splice_tail_init(&gart_list, &op->list);
+			list_splice_tail_init(&both_list, &op->list);
+			validate_fini_no_ticket(op, NULL, NULL);
 			if (unlikely(ret == -EDEADLK)) {
 				ret = ttm_bo_reserve_slowpath(&nvbo->bo, true,
 							      &op->ticket);
@@ -390,12 +391,9 @@ retry:
 					res_bo = nvbo;
 			}
 			if (unlikely(ret)) {
-				ww_acquire_done(&op->ticket);
-				ww_acquire_fini(&op->ticket);
-				drm_gem_object_unreference_unlocked(gem);
 				if (ret != -ERESTARTSYS)
 					NV_PRINTK(error, cli, "fail reserve\n");
-				return ret;
+				break;
 			}
 		}
 
@@ -404,45 +402,32 @@ retry:
 		nvbo->pbbo_index = i;
 		if ((b->valid_domains & NOUVEAU_GEM_DOMAIN_VRAM) &&
 		    (b->valid_domains & NOUVEAU_GEM_DOMAIN_GART))
-			list_add_tail(&nvbo->entry, &op->both_list);
+			list_add_tail(&nvbo->entry, &both_list);
 		else
 		if (b->valid_domains & NOUVEAU_GEM_DOMAIN_VRAM)
-			list_add_tail(&nvbo->entry, &op->vram_list);
+			list_add_tail(&nvbo->entry, &vram_list);
 		else
 		if (b->valid_domains & NOUVEAU_GEM_DOMAIN_GART)
-			list_add_tail(&nvbo->entry, &op->gart_list);
+			list_add_tail(&nvbo->entry, &gart_list);
 		else {
 			NV_PRINTK(error, cli, "invalid valid domains: 0x%08x\n",
 				 b->valid_domains);
-			list_add_tail(&nvbo->entry, &op->both_list);
-			ww_acquire_done(&op->ticket);
-			validate_fini(op, NULL);
-			return -EINVAL;
+			list_add_tail(&nvbo->entry, &both_list);
+			ret = -EINVAL;
+			break;
 		}
 		if (nvbo == res_bo)
 			goto retry;
 	}
 
 	ww_acquire_done(&op->ticket);
-	return 0;
-}
-
-static int
-validate_sync(struct nouveau_channel *chan, struct nouveau_bo *nvbo)
-{
-	struct nouveau_fence *fence = NULL;
-	int ret = 0;
-
-	spin_lock(&nvbo->bo.bdev->fence_lock);
-	fence = nouveau_fence_ref(nvbo->bo.sync_obj);
-	spin_unlock(&nvbo->bo.bdev->fence_lock);
-
-	if (fence) {
-		ret = nouveau_fence_sync(fence, chan);
-		nouveau_fence_unref(&fence);
-	}
-
+	list_splice_tail(&vram_list, &op->list);
+	list_splice_tail(&gart_list, &op->list);
+	list_splice_tail(&both_list, &op->list);
+	if (ret)
+		validate_fini(op, NULL, NULL);
 	return ret;
+
 }
 
 static int
@@ -474,9 +459,10 @@ validate_list(struct nouveau_channel *chan, struct nouveau_cli *cli,
 			return ret;
 		}
 
-		ret = validate_sync(chan, nvbo);
+		ret = nouveau_fence_sync(nvbo, chan, !!b->write_domains);
 		if (unlikely(ret)) {
-			NV_PRINTK(error, cli, "fail post-validate sync\n");
+			if (ret != -ERESTARTSYS)
+				NV_PRINTK(error, cli, "fail post-validate sync\n");
 			return ret;
 		}
 
@@ -513,11 +499,9 @@ nouveau_gem_pushbuf_validate(struct nouveau_channel *chan,
 			     struct validate_op *op, int *apply_relocs)
 {
 	struct nouveau_cli *cli = nouveau_cli(file_priv);
-	int ret, relocs = 0;
+	int ret;
 
-	INIT_LIST_HEAD(&op->vram_list);
-	INIT_LIST_HEAD(&op->gart_list);
-	INIT_LIST_HEAD(&op->both_list);
+	INIT_LIST_HEAD(&op->list);
 
 	if (nr_buffers == 0)
 		return 0;
@@ -529,34 +513,14 @@ nouveau_gem_pushbuf_validate(struct nouveau_channel *chan,
 		return ret;
 	}
 
-	ret = validate_list(chan, cli, &op->vram_list, pbbo, user_buffers);
-	if (unlikely(ret < 0)) {
-		if (ret != -ERESTARTSYS)
-			NV_PRINTK(error, cli, "validate vram_list\n");
-		validate_fini(op, NULL);
-		return ret;
-	}
-	relocs += ret;
-
-	ret = validate_list(chan, cli, &op->gart_list, pbbo, user_buffers);
-	if (unlikely(ret < 0)) {
-		if (ret != -ERESTARTSYS)
-			NV_PRINTK(error, cli, "validate gart_list\n");
-		validate_fini(op, NULL);
-		return ret;
-	}
-	relocs += ret;
-
-	ret = validate_list(chan, cli, &op->both_list, pbbo, user_buffers);
+	ret = validate_list(chan, cli, &op->list, pbbo, user_buffers);
 	if (unlikely(ret < 0)) {
 		if (ret != -ERESTARTSYS)
-			NV_PRINTK(error, cli, "validate both_list\n");
-		validate_fini(op, NULL);
+			NV_PRINTK(error, cli, "validating bo list\n");
+		validate_fini(op, NULL, NULL);
 		return ret;
 	}
-	relocs += ret;
-
-	*apply_relocs = relocs;
+	*apply_relocs = ret;
 	return 0;
 }
 
@@ -659,9 +623,7 @@ nouveau_gem_pushbuf_reloc_apply(struct nouveau_cli *cli,
 				data |= r->vor;
 		}
 
-		spin_lock(&nvbo->bo.bdev->fence_lock);
-		ret = ttm_bo_wait(&nvbo->bo, false, false, false);
-		spin_unlock(&nvbo->bo.bdev->fence_lock);
+		ret = ttm_bo_wait(&nvbo->bo, true, false, false);
 		if (ret) {
 			NV_PRINTK(error, cli, "reloc wait_idle failed: %d\n", ret);
 			break;
@@ -839,7 +801,7 @@ nouveau_gem_ioctl_pushbuf(struct drm_device *dev, void *data,
 	}
 
 out:
-	validate_fini(&op, fence);
+	validate_fini(&op, fence, bo);
 	nouveau_fence_unref(&fence);
 
 out_prevalid:
@@ -884,17 +846,29 @@ nouveau_gem_ioctl_cpu_prep(struct drm_device *dev, void *data,
 	struct drm_gem_object *gem;
 	struct nouveau_bo *nvbo;
 	bool no_wait = !!(req->flags & NOUVEAU_GEM_CPU_PREP_NOWAIT);
-	int ret = -EINVAL;
+	bool write = !!(req->flags & NOUVEAU_GEM_CPU_PREP_WRITE);
+	int ret;
 
 	gem = drm_gem_object_lookup(dev, file_priv, req->handle);
 	if (!gem)
 		return -ENOENT;
 	nvbo = nouveau_gem_object(gem);
 
-	spin_lock(&nvbo->bo.bdev->fence_lock);
-	ret = ttm_bo_wait(&nvbo->bo, true, true, no_wait);
-	spin_unlock(&nvbo->bo.bdev->fence_lock);
+	if (no_wait)
+		ret = reservation_object_test_signaled_rcu(nvbo->bo.resv, write) ? 0 : -EBUSY;
+	else {
+		long lret;
+
+		lret = reservation_object_wait_timeout_rcu(nvbo->bo.resv, write, true, 30 * HZ);
+		if (!lret)
+			ret = -EBUSY;
+		else if (lret > 0)
+			ret = 0;
+		else
+			ret = lret;
+	}
 	drm_gem_object_unreference_unlocked(gem);
+
 	return ret;
 }
 
diff --git a/drivers/gpu/drm/nouveau/nv04_fence.c b/drivers/gpu/drm/nouveau/nv04_fence.c
index 239c2c5a9615..4484131d826a 100644
--- a/drivers/gpu/drm/nouveau/nv04_fence.c
+++ b/drivers/gpu/drm/nouveau/nv04_fence.c
@@ -41,7 +41,7 @@ nv04_fence_emit(struct nouveau_fence *fence)
 	int ret = RING_SPACE(chan, 2);
 	if (ret == 0) {
 		BEGIN_NV04(chan, NvSubSw, 0x0150, 1);
-		OUT_RING  (chan, fence->sequence);
+		OUT_RING  (chan, fence->base.seqno);
 		FIRE_RING (chan);
 	}
 	return ret;
@@ -75,7 +75,7 @@ nv04_fence_context_new(struct nouveau_channel *chan)
 {
 	struct nv04_fence_chan *fctx = kzalloc(sizeof(*fctx), GFP_KERNEL);
 	if (fctx) {
-		nouveau_fence_context_new(&fctx->base);
+		nouveau_fence_context_new(chan, &fctx->base);
 		fctx->base.emit = nv04_fence_emit;
 		fctx->base.sync = nv04_fence_sync;
 		fctx->base.read = nv04_fence_read;
@@ -105,5 +105,7 @@ nv04_fence_create(struct nouveau_drm *drm)
 	priv->base.dtor = nv04_fence_destroy;
 	priv->base.context_new = nv04_fence_context_new;
 	priv->base.context_del = nv04_fence_context_del;
+	priv->base.contexts = 15;
+	priv->base.context_base = fence_context_alloc(priv->base.contexts);
 	return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nv10_fence.c b/drivers/gpu/drm/nouveau/nv10_fence.c
index 4faaf0acf5d7..737d066ffc60 100644
--- a/drivers/gpu/drm/nouveau/nv10_fence.c
+++ b/drivers/gpu/drm/nouveau/nv10_fence.c
@@ -33,7 +33,7 @@ nv10_fence_emit(struct nouveau_fence *fence)
 	int ret = RING_SPACE(chan, 2);
 	if (ret == 0) {
 		BEGIN_NV04(chan, 0, NV10_SUBCHAN_REF_CNT, 1);
-		OUT_RING  (chan, fence->sequence);
+		OUT_RING  (chan, fence->base.seqno);
 		FIRE_RING (chan);
 	}
 	return ret;
@@ -75,7 +75,7 @@ nv10_fence_context_new(struct nouveau_channel *chan)
 	if (!fctx)
 		return -ENOMEM;
 
-	nouveau_fence_context_new(&fctx->base);
+	nouveau_fence_context_new(chan, &fctx->base);
 	fctx->base.emit = nv10_fence_emit;
 	fctx->base.read = nv10_fence_read;
 	fctx->base.sync = nv10_fence_sync;
@@ -106,6 +106,8 @@ nv10_fence_create(struct nouveau_drm *drm)
 	priv->base.dtor = nv10_fence_destroy;
 	priv->base.context_new = nv10_fence_context_new;
 	priv->base.context_del = nv10_fence_context_del;
+	priv->base.contexts = 31;
+	priv->base.context_base = fence_context_alloc(priv->base.contexts);
 	spin_lock_init(&priv->lock);
 	return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nv17_fence.c b/drivers/gpu/drm/nouveau/nv17_fence.c
index ca907479f92f..6f9a1f8e2d0f 100644
--- a/drivers/gpu/drm/nouveau/nv17_fence.c
+++ b/drivers/gpu/drm/nouveau/nv17_fence.c
@@ -84,7 +84,7 @@ nv17_fence_context_new(struct nouveau_channel *chan)
 	if (!fctx)
 		return -ENOMEM;
 
-	nouveau_fence_context_new(&fctx->base);
+	nouveau_fence_context_new(chan, &fctx->base);
 	fctx->base.emit = nv10_fence_emit;
 	fctx->base.read = nv10_fence_read;
 	fctx->base.sync = nv17_fence_sync;
@@ -124,6 +124,8 @@ nv17_fence_create(struct nouveau_drm *drm)
 	priv->base.resume = nv17_fence_resume;
 	priv->base.context_new = nv17_fence_context_new;
 	priv->base.context_del = nv10_fence_context_del;
+	priv->base.contexts = 31;
+	priv->base.context_base = fence_context_alloc(priv->base.contexts);
 	spin_lock_init(&priv->lock);
 
 	ret = nouveau_bo_new(drm->dev, 4096, 0x1000, TTM_PL_FLAG_VRAM,
diff --git a/drivers/gpu/drm/nouveau/nv50_fence.c b/drivers/gpu/drm/nouveau/nv50_fence.c
index 195cf51a7c31..08fad3668a1c 100644
--- a/drivers/gpu/drm/nouveau/nv50_fence.c
+++ b/drivers/gpu/drm/nouveau/nv50_fence.c
@@ -46,7 +46,7 @@ nv50_fence_context_new(struct nouveau_channel *chan)
 	if (!fctx)
 		return -ENOMEM;
 
-	nouveau_fence_context_new(&fctx->base);
+	nouveau_fence_context_new(chan, &fctx->base);
 	fctx->base.emit = nv10_fence_emit;
 	fctx->base.read = nv10_fence_read;
 	fctx->base.sync = nv17_fence_sync;
@@ -95,6 +95,8 @@ nv50_fence_create(struct nouveau_drm *drm)
 	priv->base.resume = nv17_fence_resume;
 	priv->base.context_new = nv50_fence_context_new;
 	priv->base.context_del = nv10_fence_context_del;
+	priv->base.contexts = 127;
+	priv->base.context_base = fence_context_alloc(priv->base.contexts);
 	spin_lock_init(&priv->lock);
 
 	ret = nouveau_bo_new(drm->dev, 4096, 0x1000, TTM_PL_FLAG_VRAM,
diff --git a/drivers/gpu/drm/nouveau/nv84_fence.c b/drivers/gpu/drm/nouveau/nv84_fence.c
index 933a779c93ab..a2f28082c272 100644
--- a/drivers/gpu/drm/nouveau/nv84_fence.c
+++ b/drivers/gpu/drm/nouveau/nv84_fence.c
@@ -82,7 +82,7 @@ nv84_fence_emit(struct nouveau_fence *fence)
 	else
 		addr += fctx->vma.offset;
 
-	return fctx->base.emit32(chan, addr, fence->sequence);
+	return fctx->base.emit32(chan, addr, fence->base.seqno);
 }
 
 static int
@@ -97,7 +97,7 @@ nv84_fence_sync(struct nouveau_fence *fence,
 	else
 		addr += fctx->vma.offset;
 
-	return fctx->base.sync32(chan, addr, fence->sequence);
+	return fctx->base.sync32(chan, addr, fence->base.seqno);
 }
 
 static u32
@@ -139,12 +139,13 @@ nv84_fence_context_new(struct nouveau_channel *chan)
 	if (!fctx)
 		return -ENOMEM;
 
-	nouveau_fence_context_new(&fctx->base);
+	nouveau_fence_context_new(chan, &fctx->base);
 	fctx->base.emit = nv84_fence_emit;
 	fctx->base.sync = nv84_fence_sync;
 	fctx->base.read = nv84_fence_read;
 	fctx->base.emit32 = nv84_fence_emit32;
 	fctx->base.sync32 = nv84_fence_sync32;
+	fctx->base.sequence = nv84_fence_read(chan);
 
 	ret = nouveau_bo_vma_add(priv->bo, cli->vm, &fctx->vma);
 	if (ret == 0) {
@@ -168,13 +169,12 @@ nv84_fence_context_new(struct nouveau_channel *chan)
 static bool
 nv84_fence_suspend(struct nouveau_drm *drm)
 {
-	struct nouveau_fifo *pfifo = nvkm_fifo(&drm->device);
 	struct nv84_fence_priv *priv = drm->fence;
 	int i;
 
-	priv->suspend = vmalloc((pfifo->max + 1) * sizeof(u32));
+	priv->suspend = vmalloc(priv->base.contexts * sizeof(u32));
 	if (priv->suspend) {
-		for (i = 0; i <= pfifo->max; i++)
+		for (i = 0; i < priv->base.contexts; i++)
 			priv->suspend[i] = nouveau_bo_rd32(priv->bo, i*4);
 	}
 
@@ -184,12 +184,11 @@ nv84_fence_suspend(struct nouveau_drm *drm)
 static void
 nv84_fence_resume(struct nouveau_drm *drm)
 {
-	struct nouveau_fifo *pfifo = nvkm_fifo(&drm->device);
 	struct nv84_fence_priv *priv = drm->fence;
 	int i;
 
 	if (priv->suspend) {
-		for (i = 0; i <= pfifo->max; i++)
+		for (i = 0; i < priv->base.contexts; i++)
 			nouveau_bo_wr32(priv->bo, i*4, priv->suspend[i]);
 		vfree(priv->suspend);
 		priv->suspend = NULL;
@@ -229,10 +228,11 @@ nv84_fence_create(struct nouveau_drm *drm)
 	priv->base.context_new = nv84_fence_context_new;
 	priv->base.context_del = nv84_fence_context_del;
 
-	init_waitqueue_head(&priv->base.waiting);
+	priv->base.contexts = pfifo->max + 1;
+	priv->base.context_base = fence_context_alloc(priv->base.contexts);
 	priv->base.uevent = true;
 
-	ret = nouveau_bo_new(drm->dev, 16 * (pfifo->max + 1), 0,
+	ret = nouveau_bo_new(drm->dev, 16 * priv->base.contexts, 0,
 			     TTM_PL_FLAG_VRAM, 0, 0, NULL, &priv->bo);
 	if (ret == 0) {
 		ret = nouveau_bo_pin(priv->bo, TTM_PL_FLAG_VRAM);
@@ -246,7 +246,7 @@ nv84_fence_create(struct nouveau_drm *drm)
 	}
 
 	if (ret == 0)
-		ret = nouveau_bo_new(drm->dev, 16 * (pfifo->max + 1), 0,
+		ret = nouveau_bo_new(drm->dev, 16 * priv->base.contexts, 0,
 				     TTM_PL_FLAG_TT, 0, 0, NULL,
 				     &priv->bo_gart);
 	if (ret == 0) {
diff --git a/drivers/gpu/drm/qxl/Makefile b/drivers/gpu/drm/qxl/Makefile
index ea046ba691d2..ac0d74852e11 100644
--- a/drivers/gpu/drm/qxl/Makefile
+++ b/drivers/gpu/drm/qxl/Makefile
@@ -4,6 +4,6 @@
 
 ccflags-y := -Iinclude/drm
 
-qxl-y := qxl_drv.o qxl_kms.o qxl_display.o qxl_ttm.o qxl_fb.o qxl_object.o qxl_gem.o qxl_cmd.o qxl_image.o qxl_draw.o qxl_debugfs.o qxl_irq.o qxl_dumb.o qxl_ioctl.o qxl_fence.o qxl_release.o
+qxl-y := qxl_drv.o qxl_kms.o qxl_display.o qxl_ttm.o qxl_fb.o qxl_object.o qxl_gem.o qxl_cmd.o qxl_image.o qxl_draw.o qxl_debugfs.o qxl_irq.o qxl_dumb.o qxl_ioctl.o qxl_release.o
 
 obj-$(CONFIG_DRM_QXL)+= qxl.o
diff --git a/drivers/gpu/drm/qxl/qxl_cmd.c b/drivers/gpu/drm/qxl/qxl_cmd.c
index eb89653a7a17..97823644d347 100644
--- a/drivers/gpu/drm/qxl/qxl_cmd.c
+++ b/drivers/gpu/drm/qxl/qxl_cmd.c
@@ -620,17 +620,10 @@ static int qxl_reap_surf(struct qxl_device *qdev, struct qxl_bo *surf, bool stal
 	if (ret == -EBUSY)
 		return -EBUSY;
 
-	if (surf->fence.num_active_releases > 0 && stall == false) {
-		qxl_bo_unreserve(surf);
-		return -EBUSY;
-	}
-
 	if (stall)
 		mutex_unlock(&qdev->surf_evict_mutex);
 
-	spin_lock(&surf->tbo.bdev->fence_lock);
 	ret = ttm_bo_wait(&surf->tbo, true, true, !stall);
-	spin_unlock(&surf->tbo.bdev->fence_lock);
 
 	if (stall)
 		mutex_lock(&qdev->surf_evict_mutex);
diff --git a/drivers/gpu/drm/qxl/qxl_debugfs.c b/drivers/gpu/drm/qxl/qxl_debugfs.c
index c3c2bbdc6674..a4a63fd84803 100644
--- a/drivers/gpu/drm/qxl/qxl_debugfs.c
+++ b/drivers/gpu/drm/qxl/qxl_debugfs.c
@@ -57,11 +57,21 @@ qxl_debugfs_buffers_info(struct seq_file *m, void *data)
 	struct qxl_device *qdev = node->minor->dev->dev_private;
 	struct qxl_bo *bo;
 
+	spin_lock(&qdev->release_lock);
 	list_for_each_entry(bo, &qdev->gem.objects, list) {
-		seq_printf(m, "size %ld, pc %d, sync obj %p, num releases %d\n",
-			   (unsigned long)bo->gem_base.size, bo->pin_count,
-			   bo->tbo.sync_obj, bo->fence.num_active_releases);
+		struct reservation_object_list *fobj;
+		int rel;
+
+		rcu_read_lock();
+		fobj = rcu_dereference(bo->tbo.resv->fence);
+		rel = fobj ? fobj->shared_count : 0;
+		rcu_read_unlock();
+
+		seq_printf(m, "size %ld, pc %d, num releases %d\n",
+			   (unsigned long)bo->gem_base.size,
+			   bo->pin_count, rel);
 	}
+	spin_unlock(&qdev->release_lock);
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/qxl/qxl_drv.h b/drivers/gpu/drm/qxl/qxl_drv.h
index f6022b703645..a8be87632cae 100644
--- a/drivers/gpu/drm/qxl/qxl_drv.h
+++ b/drivers/gpu/drm/qxl/qxl_drv.h
@@ -31,6 +31,7 @@
  * Definitions taken from spice-protocol, plus kernel driver specific bits.
  */
 
+#include <linux/fence.h>
 #include <linux/workqueue.h>
 #include <linux/firmware.h>
 #include <linux/platform_device.h>
@@ -95,13 +96,6 @@ enum {
 	QXL_INTERRUPT_IO_CMD |\
 	QXL_INTERRUPT_CLIENT_MONITORS_CONFIG)
 
-struct qxl_fence {
-	struct qxl_device *qdev;
-	uint32_t num_active_releases;
-	uint32_t *release_ids;
-	struct radix_tree_root tree;
-};
-
 struct qxl_bo {
 	/* Protected by gem.mutex */
 	struct list_head		list;
@@ -113,13 +107,13 @@ struct qxl_bo {
 	unsigned			pin_count;
 	void				*kptr;
 	int                             type;
+
 	/* Constant after initialization */
 	struct drm_gem_object		gem_base;
 	bool is_primary; /* is this now a primary surface */
 	bool hw_surf_alloc;
 	struct qxl_surface surf;
 	uint32_t surface_id;
-	struct qxl_fence fence; /* per bo fence  - list of releases */
 	struct qxl_release *surf_create;
 };
 #define gem_to_qxl_bo(gobj) container_of((gobj), struct qxl_bo, gem_base)
@@ -191,6 +185,8 @@ enum {
  * spice-protocol/qxl_dev.h */
 #define QXL_MAX_RES 96
 struct qxl_release {
+	struct fence base;
+
 	int id;
 	int type;
 	uint32_t release_offset;
@@ -284,7 +280,9 @@ struct qxl_device {
 	uint8_t		slot_gen_bits;
 	uint64_t	va_slot_mask;
 
+	spinlock_t	release_lock;
 	struct idr	release_idr;
+	uint32_t	release_seqno;
 	spinlock_t release_idr_lock;
 	struct mutex	async_io_mutex;
 	unsigned int last_sent_io_cmd;
@@ -561,10 +559,4 @@ qxl_surface_lookup(struct drm_device *dev, int surface_id);
 void qxl_surface_evict(struct qxl_device *qdev, struct qxl_bo *surf, bool freeing);
 int qxl_update_surface(struct qxl_device *qdev, struct qxl_bo *surf);
 
-/* qxl_fence.c */
-void qxl_fence_add_release_locked(struct qxl_fence *qfence, uint32_t rel_id);
-int qxl_fence_remove_release(struct qxl_fence *qfence, uint32_t rel_id);
-int qxl_fence_init(struct qxl_device *qdev, struct qxl_fence *qfence);
-void qxl_fence_fini(struct qxl_fence *qfence);
-
 #endif
diff --git a/drivers/gpu/drm/qxl/qxl_fence.c b/drivers/gpu/drm/qxl/qxl_fence.c
deleted file mode 100644
index ae59e91cfb9a..000000000000
--- a/drivers/gpu/drm/qxl/qxl_fence.c
+++ /dev/null
@@ -1,91 +0,0 @@
-/*
- * Copyright 2013 Red Hat Inc.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
- * OTHER DEALINGS IN THE SOFTWARE.
- *
- * Authors: Dave Airlie
- *          Alon Levy
- */
-
-
-#include "qxl_drv.h"
-
-/* QXL fencing-
-
-   When we submit operations to the GPU we pass a release reference to the GPU
-   with them, the release reference is then added to the release ring when
-   the GPU is finished with that particular operation and has removed it from
-   its tree.
-
-   So we have can have multiple outstanding non linear fences per object.
-
-   From a TTM POV we only care if the object has any outstanding releases on
-   it.
-
-   we wait until all outstanding releases are processeed.
-
-   sync object is just a list of release ids that represent that fence on
-   that buffer.
-
-   we just add new releases onto the sync object attached to the object.
-
-   This currently uses a radix tree to store the list of release ids.
-
-   For some reason every so often qxl hw fails to release, things go wrong.
-*/
-/* must be called with the fence lock held */
-void qxl_fence_add_release_locked(struct qxl_fence *qfence, uint32_t rel_id)
-{
-	radix_tree_insert(&qfence->tree, rel_id, qfence);
-	qfence->num_active_releases++;
-}
-
-int qxl_fence_remove_release(struct qxl_fence *qfence, uint32_t rel_id)
-{
-	void *ret;
-	int retval = 0;
-	struct qxl_bo *bo = container_of(qfence, struct qxl_bo, fence);
-
-	spin_lock(&bo->tbo.bdev->fence_lock);
-
-	ret = radix_tree_delete(&qfence->tree, rel_id);
-	if (ret == qfence)
-		qfence->num_active_releases--;
-	else {
-		DRM_DEBUG("didn't find fence in radix tree for %d\n", rel_id);
-		retval = -ENOENT;
-	}
-	spin_unlock(&bo->tbo.bdev->fence_lock);
-	return retval;
-}
-
-
-int qxl_fence_init(struct qxl_device *qdev, struct qxl_fence *qfence)
-{
-	qfence->qdev = qdev;
-	qfence->num_active_releases = 0;
-	INIT_RADIX_TREE(&qfence->tree, GFP_ATOMIC);
-	return 0;
-}
-
-void qxl_fence_fini(struct qxl_fence *qfence)
-{
-	kfree(qfence->release_ids);
-	qfence->num_active_releases = 0;
-}
diff --git a/drivers/gpu/drm/qxl/qxl_kms.c b/drivers/gpu/drm/qxl/qxl_kms.c
index fd88eb4a3f79..7234561e09d9 100644
--- a/drivers/gpu/drm/qxl/qxl_kms.c
+++ b/drivers/gpu/drm/qxl/qxl_kms.c
@@ -223,6 +223,7 @@ static int qxl_device_init(struct qxl_device *qdev,
 
 	idr_init(&qdev->release_idr);
 	spin_lock_init(&qdev->release_idr_lock);
+	spin_lock_init(&qdev->release_lock);
 
 	idr_init(&qdev->surf_id_idr);
 	spin_lock_init(&qdev->surf_id_idr_lock);
diff --git a/drivers/gpu/drm/qxl/qxl_object.c b/drivers/gpu/drm/qxl/qxl_object.c
index adad12d30372..69c104c3240f 100644
--- a/drivers/gpu/drm/qxl/qxl_object.c
+++ b/drivers/gpu/drm/qxl/qxl_object.c
@@ -36,7 +36,6 @@ static void qxl_ttm_bo_destroy(struct ttm_buffer_object *tbo)
 	qdev = (struct qxl_device *)bo->gem_base.dev->dev_private;
 
 	qxl_surface_evict(qdev, bo, false);
-	qxl_fence_fini(&bo->fence);
 	mutex_lock(&qdev->gem.mutex);
 	list_del_init(&bo->list);
 	mutex_unlock(&qdev->gem.mutex);
@@ -102,7 +101,6 @@ int qxl_bo_create(struct qxl_device *qdev,
 	bo->type = domain;
 	bo->pin_count = pinned ? 1 : 0;
 	bo->surface_id = 0;
-	qxl_fence_init(qdev, &bo->fence);
 	INIT_LIST_HEAD(&bo->list);
 
 	if (surf)
diff --git a/drivers/gpu/drm/qxl/qxl_object.h b/drivers/gpu/drm/qxl/qxl_object.h
index 83a423293afd..37af1bc0dd00 100644
--- a/drivers/gpu/drm/qxl/qxl_object.h
+++ b/drivers/gpu/drm/qxl/qxl_object.h
@@ -76,12 +76,10 @@ static inline int qxl_bo_wait(struct qxl_bo *bo, u32 *mem_type,
 		}
 		return r;
 	}
-	spin_lock(&bo->tbo.bdev->fence_lock);
 	if (mem_type)
 		*mem_type = bo->tbo.mem.mem_type;
-	if (bo->tbo.sync_obj)
-		r = ttm_bo_wait(&bo->tbo, true, true, no_wait);
-	spin_unlock(&bo->tbo.bdev->fence_lock);
+
+	r = ttm_bo_wait(&bo->tbo, true, true, no_wait);
 	ttm_bo_unreserve(&bo->tbo);
 	return r;
 }
diff --git a/drivers/gpu/drm/qxl/qxl_release.c b/drivers/gpu/drm/qxl/qxl_release.c
index 14e776f1d14e..15158c5a5b3a 100644
--- a/drivers/gpu/drm/qxl/qxl_release.c
+++ b/drivers/gpu/drm/qxl/qxl_release.c
@@ -21,6 +21,7 @@
  */
 #include "qxl_drv.h"
 #include "qxl_object.h"
+#include <trace/events/fence.h>
 
 /*
  * drawable cmd cache - allocate a bunch of VRAM pages, suballocate
@@ -39,6 +40,88 @@
 static const int release_size_per_bo[] = { RELEASE_SIZE, SURFACE_RELEASE_SIZE, RELEASE_SIZE };
 static const int releases_per_bo[] = { RELEASES_PER_BO, SURFACE_RELEASES_PER_BO, RELEASES_PER_BO };
 
+static const char *qxl_get_driver_name(struct fence *fence)
+{
+	return "qxl";
+}
+
+static const char *qxl_get_timeline_name(struct fence *fence)
+{
+	return "release";
+}
+
+static bool qxl_nop_signaling(struct fence *fence)
+{
+	/* fences are always automatically signaled, so just pretend we did this.. */
+	return true;
+}
+
+static long qxl_fence_wait(struct fence *fence, bool intr, signed long timeout)
+{
+	struct qxl_device *qdev;
+	struct qxl_release *release;
+	int count = 0, sc = 0;
+	bool have_drawable_releases;
+	unsigned long cur, end = jiffies + timeout;
+
+	qdev = container_of(fence->lock, struct qxl_device, release_lock);
+	release = container_of(fence, struct qxl_release, base);
+	have_drawable_releases = release->type == QXL_RELEASE_DRAWABLE;
+
+retry:
+	sc++;
+
+	if (fence_is_signaled_locked(fence))
+		goto signaled;
+
+	qxl_io_notify_oom(qdev);
+
+	for (count = 0; count < 11; count++) {
+		if (!qxl_queue_garbage_collect(qdev, true))
+			break;
+
+		if (fence_is_signaled_locked(fence))
+			goto signaled;
+	}
+
+	if (fence_is_signaled_locked(fence))
+		goto signaled;
+
+	if (have_drawable_releases || sc < 4) {
+		if (sc > 2)
+			/* back off */
+			usleep_range(500, 1000);
+
+		if (time_after(jiffies, end))
+			return 0;
+
+		if (have_drawable_releases && sc > 300) {
+			FENCE_WARN(fence, "failed to wait on release %d "
+					  "after spincount %d\n",
+					  fence->context & ~0xf0000000, sc);
+			goto signaled;
+		}
+		goto retry;
+	}
+	/*
+	 * yeah, original sync_obj_wait gave up after 3 spins when
+	 * have_drawable_releases is not set.
+	 */
+
+signaled:
+	cur = jiffies;
+	if (time_after(cur, end))
+		return 0;
+	return end - cur;
+}
+
+static const struct fence_ops qxl_fence_ops = {
+	.get_driver_name = qxl_get_driver_name,
+	.get_timeline_name = qxl_get_timeline_name,
+	.enable_signaling = qxl_nop_signaling,
+	.wait = qxl_fence_wait,
+};
+
 static uint64_t
 qxl_release_alloc(struct qxl_device *qdev, int type,
 		  struct qxl_release **ret)
@@ -46,13 +129,13 @@ qxl_release_alloc(struct qxl_device *qdev, int type,
 	struct qxl_release *release;
 	int handle;
 	size_t size = sizeof(*release);
-	int idr_ret;
 
 	release = kmalloc(size, GFP_KERNEL);
 	if (!release) {
 		DRM_ERROR("Out of memory\n");
 		return 0;
 	}
+	release->base.ops = NULL;
 	release->type = type;
 	release->release_offset = 0;
 	release->surface_release_id = 0;
@@ -60,44 +143,59 @@ qxl_release_alloc(struct qxl_device *qdev, int type,
 
 	idr_preload(GFP_KERNEL);
 	spin_lock(&qdev->release_idr_lock);
-	idr_ret = idr_alloc(&qdev->release_idr, release, 1, 0, GFP_NOWAIT);
+	handle = idr_alloc(&qdev->release_idr, release, 1, 0, GFP_NOWAIT);
+	release->base.seqno = ++qdev->release_seqno;
 	spin_unlock(&qdev->release_idr_lock);
 	idr_preload_end();
-	handle = idr_ret;
-	if (idr_ret < 0)
-		goto release_fail;
+	if (handle < 0) {
+		kfree(release);
+		*ret = NULL;
+		return handle;
+	}
 	*ret = release;
 	QXL_INFO(qdev, "allocated release %lld\n", handle);
 	release->id = handle;
-release_fail:
-
 	return handle;
 }
 
+static void
+qxl_release_free_list(struct qxl_release *release)
+{
+	while (!list_empty(&release->bos)) {
+		struct ttm_validate_buffer *entry;
+
+		entry = container_of(release->bos.next,
+				     struct ttm_validate_buffer, head);
+
+		list_del(&entry->head);
+		kfree(entry);
+	}
+}
+
 void
 qxl_release_free(struct qxl_device *qdev,
 		 struct qxl_release *release)
 {
-	struct qxl_bo_list *entry, *tmp;
 	QXL_INFO(qdev, "release %d, type %d\n", release->id,
 		 release->type);
 
 	if (release->surface_release_id)
 		qxl_surface_id_dealloc(qdev, release->surface_release_id);
 
-	list_for_each_entry_safe(entry, tmp, &release->bos, tv.head) {
-		struct qxl_bo *bo = to_qxl_bo(entry->tv.bo);
-		QXL_INFO(qdev, "release %llx\n",
-			drm_vma_node_offset_addr(&entry->tv.bo->vma_node)
-						- DRM_FILE_OFFSET);
-		qxl_fence_remove_release(&bo->fence, release->id);
-		qxl_bo_unref(&bo);
-		kfree(entry);
-	}
 	spin_lock(&qdev->release_idr_lock);
 	idr_remove(&qdev->release_idr, release->id);
 	spin_unlock(&qdev->release_idr_lock);
-	kfree(release);
+
+	if (release->base.ops) {
+		WARN_ON(list_empty(&release->bos));
+		qxl_release_free_list(release);
+
+		fence_signal(&release->base);
+		fence_put(&release->base);
+	} else {
+		qxl_release_free_list(release);
+		kfree(release);
+	}
 }
 
 static int qxl_release_bo_alloc(struct qxl_device *qdev,
@@ -142,6 +240,10 @@ static int qxl_release_validate_bo(struct qxl_bo *bo)
 			return ret;
 	}
 
+	ret = reservation_object_reserve_shared(bo->tbo.resv);
+	if (ret)
+		return ret;
+
 	/* allocate a surface for reserved + validated buffers */
 	ret = qxl_bo_check_id(bo->gem_base.dev->dev_private, bo);
 	if (ret)
@@ -159,7 +261,7 @@ int qxl_release_reserve_list(struct qxl_release *release, bool no_intr)
 	if (list_is_singular(&release->bos))
 		return 0;
 
-	ret = ttm_eu_reserve_buffers(&release->ticket, &release->bos);
+	ret = ttm_eu_reserve_buffers(&release->ticket, &release->bos, !no_intr);
 	if (ret)
 		return ret;
 
@@ -199,6 +301,8 @@ int qxl_alloc_surface_release_reserved(struct qxl_device *qdev,
 
 		/* stash the release after the create command */
 		idr_ret = qxl_release_alloc(qdev, QXL_RELEASE_SURFACE_CMD, release);
+		if (idr_ret < 0)
+			return idr_ret;
 		bo = qxl_bo_ref(to_qxl_bo(entry->tv.bo));
 
 		(*release)->release_offset = create_rel->release_offset + 64;
@@ -239,6 +343,11 @@ int qxl_alloc_release_reserved(struct qxl_device *qdev, unsigned long size,
 	}
 
 	idr_ret = qxl_release_alloc(qdev, type, release);
+	if (idr_ret < 0) {
+		if (rbo)
+			*rbo = NULL;
+		return idr_ret;
+	}
 
 	mutex_lock(&qdev->release_mutex);
 	if (qdev->current_release_bo_offset[cur_idx] + 1 >= releases_per_bo[cur_idx]) {
@@ -319,12 +428,13 @@ void qxl_release_unmap(struct qxl_device *qdev,
 
 void qxl_release_fence_buffer_objects(struct qxl_release *release)
 {
-	struct ttm_validate_buffer *entry;
 	struct ttm_buffer_object *bo;
 	struct ttm_bo_global *glob;
 	struct ttm_bo_device *bdev;
 	struct ttm_bo_driver *driver;
 	struct qxl_bo *qbo;
+	struct ttm_validate_buffer *entry;
+	struct qxl_device *qdev;
 
 	/* if only one object on the release its the release itself
 	   since these objects are pinned no need to reserve */
@@ -333,26 +443,32 @@ void qxl_release_fence_buffer_objects(struct qxl_release *release)
 
 	bo = list_first_entry(&release->bos, struct ttm_validate_buffer, head)->bo;
 	bdev = bo->bdev;
+	qdev = container_of(bdev, struct qxl_device, mman.bdev);
+
+	/*
+	 * Since we never really allocated a context and we don't want to conflict,
+	 * set the highest bits. This will break if we really allow exporting of dma-bufs.
+	 */
+	fence_init(&release->base, &qxl_fence_ops, &qdev->release_lock,
+		   release->id | 0xf0000000, release->base.seqno);
+	trace_fence_emit(&release->base);
+
 	driver = bdev->driver;
 	glob = bo->glob;
 
 	spin_lock(&glob->lru_lock);
-	spin_lock(&bdev->fence_lock);
+	/* acquire release_lock to protect bo->resv->fence and its contents */
+	spin_lock(&qdev->release_lock);
 
 	list_for_each_entry(entry, &release->bos, head) {
 		bo = entry->bo;
 		qbo = to_qxl_bo(bo);
 
-		if (!entry->bo->sync_obj)
-			entry->bo->sync_obj = &qbo->fence;
-
-		qxl_fence_add_release_locked(&qbo->fence, release->id);
-
+		reservation_object_add_shared_fence(bo->resv, &release->base);
 		ttm_bo_add_to_lru(bo);
 		__ttm_bo_unreserve(bo);
-		entry->reserved = false;
 	}
-	spin_unlock(&bdev->fence_lock);
+	spin_unlock(&qdev->release_lock);
 	spin_unlock(&glob->lru_lock);
 	ww_acquire_fini(&release->ticket);
 }
diff --git a/drivers/gpu/drm/qxl/qxl_ttm.c b/drivers/gpu/drm/qxl/qxl_ttm.c
index f66c59b222f1..abe945a04fd4 100644
--- a/drivers/gpu/drm/qxl/qxl_ttm.c
+++ b/drivers/gpu/drm/qxl/qxl_ttm.c
@@ -357,92 +357,6 @@ static int qxl_bo_move(struct ttm_buffer_object *bo,
 	return ttm_bo_move_memcpy(bo, evict, no_wait_gpu, new_mem);
 }
 
-
-static int qxl_sync_obj_wait(void *sync_obj,
-			     bool lazy, bool interruptible)
-{
-	struct qxl_fence *qfence = (struct qxl_fence *)sync_obj;
-	int count = 0, sc = 0;
-	struct qxl_bo *bo = container_of(qfence, struct qxl_bo, fence);
-
-	if (qfence->num_active_releases == 0)
-		return 0;
-
-retry:
-	if (sc == 0) {
-		if (bo->type == QXL_GEM_DOMAIN_SURFACE)
-			qxl_update_surface(qfence->qdev, bo);
-	} else if (sc >= 1) {
-		qxl_io_notify_oom(qfence->qdev);
-	}
-
-	sc++;
-
-	for (count = 0; count < 10; count++) {
-		bool ret;
-		ret = qxl_queue_garbage_collect(qfence->qdev, true);
-		if (ret == false)
-			break;
-
-		if (qfence->num_active_releases == 0)
-			return 0;
-	}
-
-	if (qfence->num_active_releases) {
-		bool have_drawable_releases = false;
-		void **slot;
-		struct radix_tree_iter iter;
-		int release_id;
-
-		radix_tree_for_each_slot(slot, &qfence->tree, &iter, 0) {
-			struct qxl_release *release;
-
-			release_id = iter.index;
-			release = qxl_release_from_id_locked(qfence->qdev, release_id);
-			if (release == NULL)
-				continue;
-
-			if (release->type == QXL_RELEASE_DRAWABLE)
-				have_drawable_releases = true;
-		}
-
-		qxl_queue_garbage_collect(qfence->qdev, true);
-
-		if (have_drawable_releases || sc < 4) {
-			if (sc > 2)
-				/* back off */
-				usleep_range(500, 1000);
-			if (have_drawable_releases && sc > 300) {
-				WARN(1, "sync obj %d still has outstanding releases %d %d %d %ld %d\n", sc, bo->surface_id, bo->is_primary, bo->pin_count, (unsigned long)bo->gem_base.size, qfence->num_active_releases);
-				return -EBUSY;
-			}
-			goto retry;
-		}
-	}
-	return 0;
-}
-
-static int qxl_sync_obj_flush(void *sync_obj)
-{
-	return 0;
-}
-
-static void qxl_sync_obj_unref(void **sync_obj)
-{
-	*sync_obj = NULL;
-}
-
-static void *qxl_sync_obj_ref(void *sync_obj)
-{
-	return sync_obj;
-}
-
-static bool qxl_sync_obj_signaled(void *sync_obj)
-{
-	struct qxl_fence *qfence = (struct qxl_fence *)sync_obj;
-	return (qfence->num_active_releases == 0);
-}
-
 static void qxl_bo_move_notify(struct ttm_buffer_object *bo,
 			       struct ttm_mem_reg *new_mem)
 {
@@ -469,16 +383,9 @@ static struct ttm_bo_driver qxl_bo_driver = {
 	.verify_access = &qxl_verify_access,
 	.io_mem_reserve = &qxl_ttm_io_mem_reserve,
 	.io_mem_free = &qxl_ttm_io_mem_free,
-	.sync_obj_signaled = &qxl_sync_obj_signaled,
-	.sync_obj_wait = &qxl_sync_obj_wait,
-	.sync_obj_flush = &qxl_sync_obj_flush,
-	.sync_obj_unref = &qxl_sync_obj_unref,
-	.sync_obj_ref = &qxl_sync_obj_ref,
 	.move_notify = &qxl_bo_move_notify,
 };
 
-
-
 int qxl_ttm_init(struct qxl_device *qdev)
 {
 	int r;
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 83a24614138a..d80dc547a105 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -66,6 +66,7 @@
 #include <linux/kref.h>
 #include <linux/interval_tree.h>
 #include <linux/hashtable.h>
+#include <linux/fence.h>
 
 #include <ttm/ttm_bo_api.h>
 #include <ttm/ttm_bo_driver.h>
@@ -354,17 +355,19 @@ struct radeon_fence_driver {
 	/* sync_seq is protected by ring emission lock */
 	uint64_t			sync_seq[RADEON_NUM_RINGS];
 	atomic64_t			last_seq;
-	bool				initialized;
+	bool				initialized, delayed_irq;
 	struct delayed_work		lockup_work;
 };
 
 struct radeon_fence {
+	struct fence base;
+
 	struct radeon_device		*rdev;
-	struct kref			kref;
-	/* protected by radeon_fence.lock */
 	uint64_t			seq;
 	/* RB, DMA, etc. */
 	unsigned			ring;
+
+	wait_queue_t			fence_wake;
 };
 
 int radeon_fence_driver_start_ring(struct radeon_device *rdev, int ring);
@@ -782,6 +785,7 @@ struct radeon_irq {
 int radeon_irq_kms_init(struct radeon_device *rdev);
 void radeon_irq_kms_fini(struct radeon_device *rdev);
 void radeon_irq_kms_sw_irq_get(struct radeon_device *rdev, int ring);
+bool radeon_irq_kms_sw_irq_get_delayed(struct radeon_device *rdev, int ring);
 void radeon_irq_kms_sw_irq_put(struct radeon_device *rdev, int ring);
 void radeon_irq_kms_pflip_irq_get(struct radeon_device *rdev, int crtc);
 void radeon_irq_kms_pflip_irq_put(struct radeon_device *rdev, int crtc);
@@ -2308,6 +2312,7 @@ struct radeon_device {
 	struct radeon_mman		mman;
 	struct radeon_fence_driver	fence_drv[RADEON_NUM_RINGS];
 	wait_queue_head_t		fence_queue;
+	unsigned			fence_context;
 	struct mutex			ring_lock;
 	struct radeon_ring		ring[RADEON_NUM_RINGS];
 	bool				ib_pool_ready;
@@ -2441,7 +2446,17 @@ void cik_mm_wdoorbell(struct radeon_device *rdev, u32 index, u32 v);
 /*
  * Cast helper
  */
-#define to_radeon_fence(p) ((struct radeon_fence *)(p))
+extern const struct fence_ops radeon_fence_ops;
+
+static inline struct radeon_fence *to_radeon_fence(struct fence *f)
+{
+	struct radeon_fence *__f = container_of(f, struct radeon_fence, base);
+
+	if (__f->base.ops == &radeon_fence_ops)
+		return __f;
+
+	return NULL;
+}
 
 /*
  * Registers read & write functions.
diff --git a/drivers/gpu/drm/radeon/radeon_cs.c b/drivers/gpu/drm/radeon/radeon_cs.c
index bd328cb6fa61..6e3d1c8f3483 100644
--- a/drivers/gpu/drm/radeon/radeon_cs.c
+++ b/drivers/gpu/drm/radeon/radeon_cs.c
@@ -253,11 +253,17 @@ static void radeon_cs_sync_rings(struct radeon_cs_parser *p)
 	int i;
 
 	for (i = 0; i < p->nrelocs; i++) {
+		struct reservation_object *resv;
+		struct fence *fence;
+
 		if (!p->relocs[i].robj)
 			continue;
 
+		resv = p->relocs[i].robj->tbo.resv;
+		fence = reservation_object_get_excl(resv);
+
 		radeon_semaphore_sync_to(p->ib.semaphore,
-					 p->relocs[i].robj->tbo.sync_obj);
+					 (struct radeon_fence *)fence);
 	}
 }
 
@@ -427,7 +433,7 @@ static void radeon_cs_parser_fini(struct radeon_cs_parser *parser, int error, bo
 
 		ttm_eu_fence_buffer_objects(&parser->ticket,
 					    &parser->validated,
-					    parser->ib.fence);
+					    &parser->ib.fence->base);
 	} else if (backoff) {
 		ttm_eu_backoff_reservation(&parser->ticket,
 					   &parser->validated);
diff --git a/drivers/gpu/drm/radeon/radeon_device.c b/drivers/gpu/drm/radeon/radeon_device.c
index d30f1cc1aa12..e84a76e6656a 100644
--- a/drivers/gpu/drm/radeon/radeon_device.c
+++ b/drivers/gpu/drm/radeon/radeon_device.c
@@ -1253,6 +1253,7 @@ int radeon_device_init(struct radeon_device *rdev,
 	for (i = 0; i < RADEON_NUM_RINGS; i++) {
 		rdev->ring[i].idx = i;
 	}
+	rdev->fence_context = fence_context_alloc(RADEON_NUM_RINGS);
 
 	DRM_INFO("initializing kernel modesetting (%s 0x%04X:0x%04X 0x%04X:0x%04X).\n",
 		radeon_family_name[rdev->family], pdev->vendor, pdev->device,
diff --git a/drivers/gpu/drm/radeon/radeon_display.c b/drivers/gpu/drm/radeon/radeon_display.c
index bd0d687379ee..bc894c17b2f9 100644
--- a/drivers/gpu/drm/radeon/radeon_display.c
+++ b/drivers/gpu/drm/radeon/radeon_display.c
@@ -476,11 +476,6 @@ static int radeon_crtc_page_flip(struct drm_crtc *crtc,
 	obj = new_radeon_fb->obj;
 	new_rbo = gem_to_radeon_bo(obj);
 
-	spin_lock(&new_rbo->tbo.bdev->fence_lock);
-	if (new_rbo->tbo.sync_obj)
-		work->fence = radeon_fence_ref(new_rbo->tbo.sync_obj);
-	spin_unlock(&new_rbo->tbo.bdev->fence_lock);
-
 	/* pin the new buffer */
 	DRM_DEBUG_DRIVER("flip-ioctl() cur_rbo = %p, new_rbo = %p\n",
 			 work->old_rbo, new_rbo);
@@ -499,6 +494,7 @@ static int radeon_crtc_page_flip(struct drm_crtc *crtc,
 		DRM_ERROR("failed to pin new rbo buffer before flip\n");
 		goto cleanup;
 	}
+	work->fence = (struct radeon_fence *)fence_get(reservation_object_get_excl(new_rbo->tbo.resv));
 	radeon_bo_get_tiling_flags(new_rbo, &tiling_flags, NULL);
 	radeon_bo_unreserve(new_rbo);
 
@@ -582,7 +578,6 @@ cleanup:
 	drm_gem_object_unreference_unlocked(&work->old_rbo->gem_base);
 	radeon_fence_unref(&work->fence);
 	kfree(work);
-
 	return r;
 }
 
diff --git a/drivers/gpu/drm/radeon/radeon_fence.c b/drivers/gpu/drm/radeon/radeon_fence.c
index ecdba3afa2c3..af9f2d6bd7d0 100644
--- a/drivers/gpu/drm/radeon/radeon_fence.c
+++ b/drivers/gpu/drm/radeon/radeon_fence.c
@@ -130,15 +130,18 @@ int radeon_fence_emit(struct radeon_device *rdev,
 		      struct radeon_fence **fence,
 		      int ring)
 {
+	u64 seq = ++rdev->fence_drv[ring].sync_seq[ring];
+
 	/* we are protected by the ring emission mutex */
 	*fence = kmalloc(sizeof(struct radeon_fence), GFP_KERNEL);
 	if ((*fence) == NULL) {
 		return -ENOMEM;
 	}
-	kref_init(&((*fence)->kref));
 	(*fence)->rdev = rdev;
-	(*fence)->seq = ++rdev->fence_drv[ring].sync_seq[ring];
+	(*fence)->seq = seq;
 	(*fence)->ring = ring;
+	fence_init(&(*fence)->base, &radeon_fence_ops,
+		   &rdev->fence_queue.lock, rdev->fence_context + ring, seq);
 	radeon_fence_ring_emit(rdev, ring, *fence);
 	trace_radeon_fence_emit(rdev->ddev, ring, (*fence)->seq);
 	radeon_fence_schedule_check(rdev, ring);
@@ -146,6 +149,41 @@ int radeon_fence_emit(struct radeon_device *rdev,
 }
 
 /**
+ * radeon_fence_check_signaled - callback from fence_queue
+ *
+ * this function is called with fence_queue lock held, which is also used
+ * for the fence locking itself, so unlocked variants are used for
+ * fence_signal, and remove_wait_queue.
+ */
+static int radeon_fence_check_signaled(wait_queue_t *wait, unsigned mode, int flags, void *key)
+{
+	struct radeon_fence *fence;
+	u64 seq;
+
+	fence = container_of(wait, struct radeon_fence, fence_wake);
+
+	/*
+	 * We cannot use radeon_fence_process here because we're already
+	 * in the waitqueue, in a call from wake_up_all.
+	 */
+	seq = atomic64_read(&fence->rdev->fence_drv[fence->ring].last_seq);
+	if (seq >= fence->seq) {
+		int ret = fence_signal_locked(&fence->base);
+
+		if (!ret)
+			FENCE_TRACE(&fence->base, "signaled from irq context\n");
+		else
+			FENCE_TRACE(&fence->base, "was already signaled\n");
+
+		radeon_irq_kms_sw_irq_put(fence->rdev, fence->ring);
+		__remove_wait_queue(&fence->rdev->fence_queue, &fence->fence_wake);
+		fence_put(&fence->base);
+	} else
+		FENCE_TRACE(&fence->base, "pending\n");
+	return 0;
+}
+
+/**
  * radeon_fence_activity - check for fence activity
  *
  * @rdev: radeon_device pointer
@@ -242,6 +280,15 @@ static void radeon_fence_check_lockup(struct work_struct *work)
 		return;
 	}
 
+	if (fence_drv->delayed_irq && rdev->ddev->irq_enabled) {
+		unsigned long irqflags;
+
+		fence_drv->delayed_irq = false;
+		spin_lock_irqsave(&rdev->irq.lock, irqflags);
+		radeon_irq_set(rdev);
+		spin_unlock_irqrestore(&rdev->irq.lock, irqflags);
+	}
+
 	if (radeon_fence_activity(rdev, ring))
 		wake_up_all(&rdev->fence_queue);
 
@@ -276,21 +323,6 @@ void radeon_fence_process(struct radeon_device *rdev, int ring)
 }
 
 /**
- * radeon_fence_destroy - destroy a fence
- *
- * @kref: fence kref
- *
- * Frees the fence object (all asics).
- */
-static void radeon_fence_destroy(struct kref *kref)
-{
-	struct radeon_fence *fence;
-
-	fence = container_of(kref, struct radeon_fence, kref);
-	kfree(fence);
-}
-
-/**
  * radeon_fence_seq_signaled - check if a fence sequence number has signaled
  *
  * @rdev: radeon device pointer
@@ -318,6 +350,75 @@ static bool radeon_fence_seq_signaled(struct radeon_device *rdev,
 	return false;
 }
 
+static bool radeon_fence_is_signaled(struct fence *f)
+{
+	struct radeon_fence *fence = to_radeon_fence(f);
+	struct radeon_device *rdev = fence->rdev;
+	unsigned ring = fence->ring;
+	u64 seq = fence->seq;
+
+	if (atomic64_read(&rdev->fence_drv[ring].last_seq) >= seq) {
+		return true;
+	}
+
+	if (down_read_trylock(&rdev->exclusive_lock)) {
+		radeon_fence_process(rdev, ring);
+		up_read(&rdev->exclusive_lock);
+
+		if (atomic64_read(&rdev->fence_drv[ring].last_seq) >= seq) {
+			return true;
+		}
+	}
+	return false;
+}
+
+/**
+ * radeon_fence_enable_signaling - enable signalling on fence
+ * @fence: fence
+ *
+ * This function is called with fence_queue lock held, and adds a callback
+ * to fence_queue that checks if this fence is signaled, and if so it
+ * signals the fence and removes itself.
+ */
+static bool radeon_fence_enable_signaling(struct fence *f)
+{
+	struct radeon_fence *fence = to_radeon_fence(f);
+	struct radeon_device *rdev = fence->rdev;
+
+	if (atomic64_read(&rdev->fence_drv[fence->ring].last_seq) >= fence->seq)
+		return false;
+
+	if (down_read_trylock(&rdev->exclusive_lock)) {
+		radeon_irq_kms_sw_irq_get(rdev, fence->ring);
+
+		if (radeon_fence_activity(rdev, fence->ring))
+			wake_up_all_locked(&rdev->fence_queue);
+
+		/* did fence get signaled after we enabled the sw irq? */
+		if (atomic64_read(&rdev->fence_drv[fence->ring].last_seq) >= fence->seq) {
+			radeon_irq_kms_sw_irq_put(rdev, fence->ring);
+			up_read(&rdev->exclusive_lock);
+			return false;
+		}
+
+		up_read(&rdev->exclusive_lock);
+	} else {
+		/* we're probably in a lockup, lets not fiddle too much */
+		if (radeon_irq_kms_sw_irq_get_delayed(rdev, fence->ring))
+			rdev->fence_drv[fence->ring].delayed_irq = true;
+		radeon_fence_schedule_check(rdev, fence->ring);
+	}
+
+	fence->fence_wake.flags = 0;
+	fence->fence_wake.private = NULL;
+	fence->fence_wake.func = radeon_fence_check_signaled;
+	__add_wait_queue(&rdev->fence_queue, &fence->fence_wake);
+	fence_get(f);
+
+	FENCE_TRACE(&fence->base, "armed on ring %i!\n", fence->ring);
+	return true;
+}
+
 /**
  * radeon_fence_signaled - check if a fence has signaled
  *
@@ -330,8 +431,15 @@ bool radeon_fence_signaled(struct radeon_fence *fence)
 {
 	if (!fence)
 		return true;
-	if (radeon_fence_seq_signaled(fence->rdev, fence->seq, fence->ring))
+
+	if (radeon_fence_seq_signaled(fence->rdev, fence->seq, fence->ring)) {
+		int ret;
+
+		ret = fence_signal(&fence->base);
+		if (!ret)
+			FENCE_TRACE(&fence->base, "signaled from radeon_fence_signaled\n");
 		return true;
+	}
 	return false;
 }
 
@@ -433,17 +541,15 @@ int radeon_fence_wait(struct radeon_fence *fence, bool intr)
 	uint64_t seq[RADEON_NUM_RINGS] = {};
 	long r;
 
-	if (fence == NULL) {
-		WARN(1, "Querying an invalid fence : %p !\n", fence);
-		return -EINVAL;
-	}
-
 	seq[fence->ring] = fence->seq;
 	r = radeon_fence_wait_seq_timeout(fence->rdev, seq, intr, MAX_SCHEDULE_TIMEOUT);
 	if (r < 0) {
 		return r;
 	}
 
+	r = fence_signal(&fence->base);
+	if (!r)
+		FENCE_TRACE(&fence->base, "signaled from fence_wait\n");
 	return 0;
 }
 
@@ -557,7 +663,7 @@ int radeon_fence_wait_empty(struct radeon_device *rdev, int ring)
  */
 struct radeon_fence *radeon_fence_ref(struct radeon_fence *fence)
 {
-	kref_get(&fence->kref);
+	fence_get(&fence->base);
 	return fence;
 }
 
@@ -574,7 +680,7 @@ void radeon_fence_unref(struct radeon_fence **fence)
 
 	*fence = NULL;
 	if (tmp) {
-		kref_put(&tmp->kref, radeon_fence_destroy);
+		fence_put(&tmp->base);
 	}
 }
 
@@ -887,3 +993,72 @@ int radeon_debugfs_fence_init(struct radeon_device *rdev)
 	return 0;
 #endif
 }
+
+static const char *radeon_fence_get_driver_name(struct fence *fence)
+{
+	return "radeon";
+}
+
+static const char *radeon_fence_get_timeline_name(struct fence *f)
+{
+	struct radeon_fence *fence = to_radeon_fence(f);
+	switch (fence->ring) {
+	case RADEON_RING_TYPE_GFX_INDEX: return "radeon.gfx";
+	case CAYMAN_RING_TYPE_CP1_INDEX: return "radeon.cp1";
+	case CAYMAN_RING_TYPE_CP2_INDEX: return "radeon.cp2";
+	case R600_RING_TYPE_DMA_INDEX: return "radeon.dma";
+	case CAYMAN_RING_TYPE_DMA1_INDEX: return "radeon.dma1";
+	case R600_RING_TYPE_UVD_INDEX: return "radeon.uvd";
+	case TN_RING_TYPE_VCE1_INDEX: return "radeon.vce1";
+	case TN_RING_TYPE_VCE2_INDEX: return "radeon.vce2";
+	default: WARN_ON_ONCE(1); return "radeon.unk";
+	}
+}
+
+static inline bool radeon_test_signaled(struct radeon_fence *fence)
+{
+	return test_bit(FENCE_FLAG_SIGNALED_BIT, &fence->base.flags);
+}
+
+static signed long radeon_fence_default_wait(struct fence *f, bool intr,
+					     signed long t)
+{
+	struct radeon_fence *fence = to_radeon_fence(f);
+	struct radeon_device *rdev = fence->rdev;
+	bool signaled;
+
+	fence_enable_sw_signaling(&fence->base);
+
+	/*
+	 * This function has to return -EDEADLK, but cannot hold
+	 * exclusive_lock during the wait because some callers
+	 * may already hold it. This means checking needs_reset without
+	 * lock, and not fiddling with any gpu internals.
+	 *
+	 * The callback installed with fence_enable_sw_signaling will
+	 * run before our wait_event_*timeout call, so we will see
+	 * both the signaled fence and the changes to needs_reset.
+	 */
+
+	if (intr)
+		t = wait_event_interruptible_timeout(rdev->fence_queue,
+			((signaled = radeon_test_signaled(fence)) ||
+			 rdev->needs_reset), t);
+	else
+		t = wait_event_timeout(rdev->fence_queue,
+			((signaled = radeon_test_signaled(fence)) ||
+			 rdev->needs_reset), t);
+
+	if (t > 0 && !signaled)
+		return -EDEADLK;
+	return t;
+}
+
+const struct fence_ops radeon_fence_ops = {
+	.get_driver_name = radeon_fence_get_driver_name,
+	.get_timeline_name = radeon_fence_get_timeline_name,
+	.enable_signaling = radeon_fence_enable_signaling,
+	.signaled = radeon_fence_is_signaled,
+	.wait = radeon_fence_default_wait,
+	.release = NULL,
+};
diff --git a/drivers/gpu/drm/radeon/radeon_gem.c b/drivers/gpu/drm/radeon/radeon_gem.c
index 01b58941acd4..4b7c8ec36c2f 100644
--- a/drivers/gpu/drm/radeon/radeon_gem.c
+++ b/drivers/gpu/drm/radeon/radeon_gem.c
@@ -94,7 +94,7 @@ static int radeon_gem_set_domain(struct drm_gem_object *gobj,
 {
 	struct radeon_bo *robj;
 	uint32_t domain;
-	int r;
+	long r;
 
 	/* FIXME: reeimplement */
 	robj = gem_to_radeon_bo(gobj);
@@ -110,9 +110,12 @@ static int radeon_gem_set_domain(struct drm_gem_object *gobj,
 	}
 	if (domain == RADEON_GEM_DOMAIN_CPU) {
 		/* Asking for cpu access wait for object idle */
-		r = radeon_bo_wait(robj, NULL, false);
-		if (r) {
-			printk(KERN_ERR "Failed to wait for object !\n");
+		r = reservation_object_wait_timeout_rcu(robj->tbo.resv, true, true, 30 * HZ);
+		if (!r)
+			r = -EBUSY;
+
+		if (r < 0 && r != -EINTR) {
+			printk(KERN_ERR "Failed to wait for object: %li\n", r);
 			return r;
 		}
 	}
@@ -449,15 +452,22 @@ int radeon_gem_wait_idle_ioctl(struct drm_device *dev, void *data,
 	struct drm_radeon_gem_wait_idle *args = data;
 	struct drm_gem_object *gobj;
 	struct radeon_bo *robj;
-	int r;
+	int r = 0;
 	uint32_t cur_placement = 0;
+	long ret;
 
 	gobj = drm_gem_object_lookup(dev, filp, args->handle);
 	if (gobj == NULL) {
 		return -ENOENT;
 	}
 	robj = gem_to_radeon_bo(gobj);
-	r = radeon_bo_wait(robj, &cur_placement, false);
+
+	ret = reservation_object_wait_timeout_rcu(robj->tbo.resv, true, true, 30 * HZ);
+	if (ret == 0)
+		r = -EBUSY;
+	else if (ret < 0)
+		r = ret;
+
 	/* Flush HDP cache via MMIO if necessary */
 	if (rdev->asic->mmio_hdp_flush &&
 	    radeon_mem_type_to_domain(cur_placement) == RADEON_GEM_DOMAIN_VRAM)
diff --git a/drivers/gpu/drm/radeon/radeon_irq_kms.c b/drivers/gpu/drm/radeon/radeon_irq_kms.c
index f0bff4be67f1..7784911d78ef 100644
--- a/drivers/gpu/drm/radeon/radeon_irq_kms.c
+++ b/drivers/gpu/drm/radeon/radeon_irq_kms.c
@@ -324,6 +324,21 @@ void radeon_irq_kms_sw_irq_get(struct radeon_device *rdev, int ring)
 }
 
 /**
+ * radeon_irq_kms_sw_irq_get_delayed - enable software interrupt
+ *
+ * @rdev: radeon device pointer
+ * @ring: ring whose interrupt you want to enable
+ *
+ * Enables the software interrupt for a specific ring (all asics).
+ * The software interrupt is generally used to signal a fence on
+ * a particular ring.
+ */
+bool radeon_irq_kms_sw_irq_get_delayed(struct radeon_device *rdev, int ring)
+{
+	return atomic_inc_return(&rdev->irq.ring_int[ring]) == 1;
+}
+
+/**
  * radeon_irq_kms_sw_irq_put - disable software interrupt
  *
  * @rdev: radeon device pointer
diff --git a/drivers/gpu/drm/radeon/radeon_mn.c b/drivers/gpu/drm/radeon/radeon_mn.c
index 0157bc2f11f8..a69bd441dd2d 100644
--- a/drivers/gpu/drm/radeon/radeon_mn.c
+++ b/drivers/gpu/drm/radeon/radeon_mn.c
@@ -122,6 +122,7 @@ static void radeon_mn_invalidate_range_start(struct mmu_notifier *mn,
 	it = interval_tree_iter_first(&rmn->objects, start, end);
 	while (it) {
 		struct radeon_bo *bo;
+		struct fence *fence;
 		int r;
 
 		bo = container_of(it, struct radeon_bo, mn_it);
@@ -133,8 +134,9 @@ static void radeon_mn_invalidate_range_start(struct mmu_notifier *mn,
 			continue;
 		}
 
-		if (bo->tbo.sync_obj) {
-			r = radeon_fence_wait(bo->tbo.sync_obj, false);
+		fence = reservation_object_get_excl(bo->tbo.resv);
+		if (fence) {
+			r = radeon_fence_wait((struct radeon_fence *)fence, false);
 			if (r)
 				DRM_ERROR("(%d) failed to wait for user bo\n", r);
 		}
diff --git a/drivers/gpu/drm/radeon/radeon_object.c b/drivers/gpu/drm/radeon/radeon_object.c
index c97a42432e2b..aadbd36e64b9 100644
--- a/drivers/gpu/drm/radeon/radeon_object.c
+++ b/drivers/gpu/drm/radeon/radeon_object.c
@@ -482,7 +482,7 @@ int radeon_bo_list_validate(struct radeon_device *rdev,
 	u64 bytes_moved = 0, initial_bytes_moved;
 	u64 bytes_moved_threshold = radeon_bo_get_threshold_for_moves(rdev);
 
-	r = ttm_eu_reserve_buffers(ticket, head);
+	r = ttm_eu_reserve_buffers(ticket, head, true);
 	if (unlikely(r != 0)) {
 		return r;
 	}
@@ -779,12 +779,10 @@ int radeon_bo_wait(struct radeon_bo *bo, u32 *mem_type, bool no_wait)
 	r = ttm_bo_reserve(&bo->tbo, true, no_wait, false, NULL);
 	if (unlikely(r != 0))
 		return r;
-	spin_lock(&bo->tbo.bdev->fence_lock);
 	if (mem_type)
 		*mem_type = bo->tbo.mem.mem_type;
-	if (bo->tbo.sync_obj)
-		r = ttm_bo_wait(&bo->tbo, true, true, no_wait);
-	spin_unlock(&bo->tbo.bdev->fence_lock);
+
+	r = ttm_bo_wait(&bo->tbo, true, true, no_wait);
 	ttm_bo_unreserve(&bo->tbo);
 	return r;
 }
diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c
index 822eb3630045..62d1f4d730a2 100644
--- a/drivers/gpu/drm/radeon/radeon_ttm.c
+++ b/drivers/gpu/drm/radeon/radeon_ttm.c
@@ -270,12 +270,12 @@ static int radeon_move_blit(struct ttm_buffer_object *bo,
 	BUILD_BUG_ON((PAGE_SIZE % RADEON_GPU_PAGE_SIZE) != 0);
 
 	/* sync other rings */
-	fence = bo->sync_obj;
+	fence = (struct radeon_fence *)reservation_object_get_excl(bo->resv);
 	r = radeon_copy(rdev, old_start, new_start,
 			new_mem->num_pages * (PAGE_SIZE / RADEON_GPU_PAGE_SIZE), /* GPU pages */
 			&fence);
 	/* FIXME: handle copy error */
-	r = ttm_bo_move_accel_cleanup(bo, (void *)fence,
+	r = ttm_bo_move_accel_cleanup(bo, &fence->base,
 				      evict, no_wait_gpu, new_mem);
 	radeon_fence_unref(&fence);
 	return r;
@@ -488,31 +488,6 @@ static void radeon_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_re
 {
 }
 
-static int radeon_sync_obj_wait(void *sync_obj, bool lazy, bool interruptible)
-{
-	return radeon_fence_wait((struct radeon_fence *)sync_obj, interruptible);
-}
-
-static int radeon_sync_obj_flush(void *sync_obj)
-{
-	return 0;
-}
-
-static void radeon_sync_obj_unref(void **sync_obj)
-{
-	radeon_fence_unref((struct radeon_fence **)sync_obj);
-}
-
-static void *radeon_sync_obj_ref(void *sync_obj)
-{
-	return radeon_fence_ref((struct radeon_fence *)sync_obj);
-}
-
-static bool radeon_sync_obj_signaled(void *sync_obj)
-{
-	return radeon_fence_signaled((struct radeon_fence *)sync_obj);
-}
-
 /*
  * TTM backend functions.
  */
@@ -847,11 +822,6 @@ static struct ttm_bo_driver radeon_bo_driver = {
 	.evict_flags = &radeon_evict_flags,
 	.move = &radeon_bo_move,
 	.verify_access = &radeon_verify_access,
-	.sync_obj_signaled = &radeon_sync_obj_signaled,
-	.sync_obj_wait = &radeon_sync_obj_wait,
-	.sync_obj_flush = &radeon_sync_obj_flush,
-	.sync_obj_unref = &radeon_sync_obj_unref,
-	.sync_obj_ref = &radeon_sync_obj_ref,
 	.move_notify = &radeon_bo_move_notify,
 	.fault_reserve_notify = &radeon_bo_fault_reserve_notify,
 	.io_mem_reserve = &radeon_ttm_io_mem_reserve,
diff --git a/drivers/gpu/drm/radeon/radeon_uvd.c b/drivers/gpu/drm/radeon/radeon_uvd.c
index 5729e9bebd9d..ba4f38916026 100644
--- a/drivers/gpu/drm/radeon/radeon_uvd.c
+++ b/drivers/gpu/drm/radeon/radeon_uvd.c
@@ -400,6 +400,7 @@ static int radeon_uvd_cs_msg(struct radeon_cs_parser *p, struct radeon_bo *bo,
 {
 	int32_t *msg, msg_type, handle;
 	unsigned img_size = 0;
+	struct fence *f;
 	void *ptr;
 
 	int i, r;
@@ -409,8 +410,9 @@ static int radeon_uvd_cs_msg(struct radeon_cs_parser *p, struct radeon_bo *bo,
 		return -EINVAL;
 	}
 
-	if (bo->tbo.sync_obj) {
-		r = radeon_fence_wait(bo->tbo.sync_obj, false);
+	f = reservation_object_get_excl(bo->tbo.resv);
+	if (f) {
+		r = radeon_fence_wait((struct radeon_fence *)f, false);
 		if (r) {
 			DRM_ERROR("Failed waiting for UVD message (%d)!\n", r);
 			return r;
diff --git a/drivers/gpu/drm/radeon/radeon_vm.c b/drivers/gpu/drm/radeon/radeon_vm.c
index 4751c6728fe9..671ee566aa51 100644
--- a/drivers/gpu/drm/radeon/radeon_vm.c
+++ b/drivers/gpu/drm/radeon/radeon_vm.c
@@ -399,7 +399,7 @@ static int radeon_vm_clear_bo(struct radeon_device *rdev,
         INIT_LIST_HEAD(&head);
         list_add(&tv.head, &head);
 
-        r = ttm_eu_reserve_buffers(&ticket, &head);
+        r = ttm_eu_reserve_buffers(&ticket, &head, true);
         if (r)
 		return r;
 
@@ -424,7 +424,7 @@ static int radeon_vm_clear_bo(struct radeon_device *rdev,
 	if (r)
                 goto error;
 
-	ttm_eu_fence_buffer_objects(&ticket, &head, ib.fence);
+	ttm_eu_fence_buffer_objects(&ticket, &head, &ib.fence->base);
 	radeon_ib_free(rdev, &ib);
 
 	return 0;
@@ -693,8 +693,14 @@ int radeon_vm_update_page_directory(struct radeon_device *rdev,
 				    incr, R600_PTE_VALID);
 
 	if (ib.length_dw != 0) {
+		struct fence *fence;
+
 		radeon_asic_vm_pad_ib(rdev, &ib);
-		radeon_semaphore_sync_to(ib.semaphore, pd->tbo.sync_obj);
+
+		fence = reservation_object_get_excl(pd->tbo.resv);
+		radeon_semaphore_sync_to(ib.semaphore,
+					 (struct radeon_fence *)fence);
+
 		radeon_semaphore_sync_to(ib.semaphore, vm->last_id_use);
 		WARN_ON(ib.length_dw > ndw);
 		r = radeon_ib_schedule(rdev, &ib, NULL, false);
@@ -820,8 +826,11 @@ static void radeon_vm_update_ptes(struct radeon_device *rdev,
 		struct radeon_bo *pt = vm->page_tables[pt_idx].bo;
 		unsigned nptes;
 		uint64_t pte;
+		struct fence *fence;
 
-		radeon_semaphore_sync_to(ib->semaphore, pt->tbo.sync_obj);
+		fence = reservation_object_get_excl(pt->tbo.resv);
+		radeon_semaphore_sync_to(ib->semaphore,
+					 (struct radeon_fence *)fence);
 
 		if ((addr & ~mask) == (end & ~mask))
 			nptes = end - addr;
diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index b992ec3c318a..a11969acfea5 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -40,6 +40,7 @@
 #include <linux/file.h>
 #include <linux/module.h>
 #include <linux/atomic.h>
+#include <linux/reservation.h>
 
 #define TTM_ASSERT_LOCKED(param)
 #define TTM_DEBUG(fmt, arg...)
@@ -142,7 +143,6 @@ static void ttm_bo_release_list(struct kref *list_kref)
 	BUG_ON(atomic_read(&bo->list_kref.refcount));
 	BUG_ON(atomic_read(&bo->kref.refcount));
 	BUG_ON(atomic_read(&bo->cpu_writers));
-	BUG_ON(bo->sync_obj != NULL);
 	BUG_ON(bo->mem.mm_node != NULL);
 	BUG_ON(!list_empty(&bo->lru));
 	BUG_ON(!list_empty(&bo->ddestroy));
@@ -403,36 +403,48 @@ static void ttm_bo_cleanup_memtype_use(struct ttm_buffer_object *bo)
 	ww_mutex_unlock (&bo->resv->lock);
 }
 
+static void ttm_bo_flush_all_fences(struct ttm_buffer_object *bo)
+{
+	struct reservation_object_list *fobj;
+	struct fence *fence;
+	int i;
+
+	fobj = reservation_object_get_list(bo->resv);
+	fence = reservation_object_get_excl(bo->resv);
+	if (fence && !fence->ops->signaled)
+		fence_enable_sw_signaling(fence);
+
+	for (i = 0; fobj && i < fobj->shared_count; ++i) {
+		fence = rcu_dereference_protected(fobj->shared[i],
+					reservation_object_held(bo->resv));
+
+		if (!fence->ops->signaled)
+			fence_enable_sw_signaling(fence);
+	}
+}
+
 static void ttm_bo_cleanup_refs_or_queue(struct ttm_buffer_object *bo)
 {
 	struct ttm_bo_device *bdev = bo->bdev;
 	struct ttm_bo_global *glob = bo->glob;
-	struct ttm_bo_driver *driver = bdev->driver;
-	void *sync_obj = NULL;
 	int put_count;
 	int ret;
 
 	spin_lock(&glob->lru_lock);
 	ret = __ttm_bo_reserve(bo, false, true, false, NULL);
 
-	spin_lock(&bdev->fence_lock);
-	(void) ttm_bo_wait(bo, false, false, true);
-	if (!ret && !bo->sync_obj) {
-		spin_unlock(&bdev->fence_lock);
-		put_count = ttm_bo_del_from_lru(bo);
+	if (!ret) {
+		if (!ttm_bo_wait(bo, false, false, true)) {
+			put_count = ttm_bo_del_from_lru(bo);
 
-		spin_unlock(&glob->lru_lock);
-		ttm_bo_cleanup_memtype_use(bo);
+			spin_unlock(&glob->lru_lock);
+			ttm_bo_cleanup_memtype_use(bo);
 
-		ttm_bo_list_ref_sub(bo, put_count, true);
+			ttm_bo_list_ref_sub(bo, put_count, true);
 
-		return;
-	}
-	if (bo->sync_obj)
-		sync_obj = driver->sync_obj_ref(bo->sync_obj);
-	spin_unlock(&bdev->fence_lock);
-
-	if (!ret) {
+			return;
+		} else
+			ttm_bo_flush_all_fences(bo);
 
 		/*
 		 * Make NO_EVICT bos immediately available to
@@ -451,10 +463,6 @@ static void ttm_bo_cleanup_refs_or_queue(struct ttm_buffer_object *bo)
 	list_add_tail(&bo->ddestroy, &bdev->ddestroy);
 	spin_unlock(&glob->lru_lock);
 
-	if (sync_obj) {
-		driver->sync_obj_flush(sync_obj);
-		driver->sync_obj_unref(&sync_obj);
-	}
 	schedule_delayed_work(&bdev->wq,
 			      ((HZ / 100) < 1) ? 1 : HZ / 100);
 }
@@ -475,44 +483,26 @@ static int ttm_bo_cleanup_refs_and_unlock(struct ttm_buffer_object *bo,
 					  bool interruptible,
 					  bool no_wait_gpu)
 {
-	struct ttm_bo_device *bdev = bo->bdev;
-	struct ttm_bo_driver *driver = bdev->driver;
 	struct ttm_bo_global *glob = bo->glob;
 	int put_count;
 	int ret;
 
-	spin_lock(&bdev->fence_lock);
 	ret = ttm_bo_wait(bo, false, false, true);
 
 	if (ret && !no_wait_gpu) {
-		void *sync_obj;
-
-		/*
-		 * Take a reference to the fence and unreserve,
-		 * at this point the buffer should be dead, so
-		 * no new sync objects can be attached.
-		 */
-		sync_obj = driver->sync_obj_ref(bo->sync_obj);
-		spin_unlock(&bdev->fence_lock);
-
-		__ttm_bo_unreserve(bo);
+		long lret;
+		ww_mutex_unlock(&bo->resv->lock);
 		spin_unlock(&glob->lru_lock);
 
-		ret = driver->sync_obj_wait(sync_obj, false, interruptible);
-		driver->sync_obj_unref(&sync_obj);
-		if (ret)
-			return ret;
+		lret = reservation_object_wait_timeout_rcu(bo->resv,
+							   true,
+							   interruptible,
+							   30 * HZ);
 
-		/*
-		 * remove sync_obj with ttm_bo_wait, the wait should be
-		 * finished, and no new wait object should have been added.
-		 */
-		spin_lock(&bdev->fence_lock);
-		ret = ttm_bo_wait(bo, false, false, true);
-		WARN_ON(ret);
-		spin_unlock(&bdev->fence_lock);
-		if (ret)
-			return ret;
+		if (lret < 0)
+			return lret;
+		else if (lret == 0)
+			return -EBUSY;
 
 		spin_lock(&glob->lru_lock);
 		ret = __ttm_bo_reserve(bo, false, true, false, NULL);
@@ -529,8 +519,14 @@ static int ttm_bo_cleanup_refs_and_unlock(struct ttm_buffer_object *bo,
 			spin_unlock(&glob->lru_lock);
 			return 0;
 		}
-	} else
-		spin_unlock(&bdev->fence_lock);
+
+		/*
+		 * remove sync_obj with ttm_bo_wait, the wait should be
+		 * finished, and no new wait object should have been added.
+		 */
+		ret = ttm_bo_wait(bo, false, false, true);
+		WARN_ON(ret);
+	}
 
 	if (ret || unlikely(list_empty(&bo->ddestroy))) {
 		__ttm_bo_unreserve(bo);
@@ -668,9 +664,7 @@ static int ttm_bo_evict(struct ttm_buffer_object *bo, bool interruptible,
 	struct ttm_placement placement;
 	int ret = 0;
 
-	spin_lock(&bdev->fence_lock);
 	ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu);
-	spin_unlock(&bdev->fence_lock);
 
 	if (unlikely(ret != 0)) {
 		if (ret != -ERESTARTSYS) {
@@ -961,7 +955,6 @@ static int ttm_bo_move_buffer(struct ttm_buffer_object *bo,
 {
 	int ret = 0;
 	struct ttm_mem_reg mem;
-	struct ttm_bo_device *bdev = bo->bdev;
 
 	lockdep_assert_held(&bo->resv->lock.base);
 
@@ -970,9 +963,7 @@ static int ttm_bo_move_buffer(struct ttm_buffer_object *bo,
 	 * Have the driver move function wait for idle when necessary,
 	 * instead of doing it here.
 	 */
-	spin_lock(&bdev->fence_lock);
 	ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu);
-	spin_unlock(&bdev->fence_lock);
 	if (ret)
 		return ret;
 	mem.num_pages = bo->num_pages;
@@ -1462,7 +1453,6 @@ int ttm_bo_device_init(struct ttm_bo_device *bdev,
 	bdev->glob = glob;
 	bdev->need_dma32 = need_dma32;
 	bdev->val_seq = 0;
-	spin_lock_init(&bdev->fence_lock);
 	mutex_lock(&glob->device_list_mutex);
 	list_add_tail(&bdev->device_list, &glob->device_list);
 	mutex_unlock(&glob->device_list_mutex);
@@ -1515,65 +1505,56 @@ void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo)
 
 EXPORT_SYMBOL(ttm_bo_unmap_virtual);
 
-
 int ttm_bo_wait(struct ttm_buffer_object *bo,
 		bool lazy, bool interruptible, bool no_wait)
 {
-	struct ttm_bo_driver *driver = bo->bdev->driver;
-	struct ttm_bo_device *bdev = bo->bdev;
-	void *sync_obj;
-	int ret = 0;
-
-	if (likely(bo->sync_obj == NULL))
-		return 0;
+	struct reservation_object_list *fobj;
+	struct reservation_object *resv;
+	struct fence *excl;
+	long timeout = 15 * HZ;
+	int i;
 
-	while (bo->sync_obj) {
+	resv = bo->resv;
+	fobj = reservation_object_get_list(resv);
+	excl = reservation_object_get_excl(resv);
+	if (excl) {
+		if (!fence_is_signaled(excl)) {
+			if (no_wait)
+				return -EBUSY;
 
-		if (driver->sync_obj_signaled(bo->sync_obj)) {
-			void *tmp_obj = bo->sync_obj;
-			bo->sync_obj = NULL;
-			clear_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags);
-			spin_unlock(&bdev->fence_lock);
-			driver->sync_obj_unref(&tmp_obj);
-			spin_lock(&bdev->fence_lock);
-			continue;
+			timeout = fence_wait_timeout(excl,
+						     interruptible, timeout);
 		}
+	}
 
-		if (no_wait)
-			return -EBUSY;
+	for (i = 0; fobj && timeout > 0 && i < fobj->shared_count; ++i) {
+		struct fence *fence;
+		fence = rcu_dereference_protected(fobj->shared[i],
+						reservation_object_held(resv));
 
-		sync_obj = driver->sync_obj_ref(bo->sync_obj);
-		spin_unlock(&bdev->fence_lock);
-		ret = driver->sync_obj_wait(sync_obj,
-					    lazy, interruptible);
-		if (unlikely(ret != 0)) {
-			driver->sync_obj_unref(&sync_obj);
-			spin_lock(&bdev->fence_lock);
-			return ret;
-		}
-		spin_lock(&bdev->fence_lock);
-		if (likely(bo->sync_obj == sync_obj)) {
-			void *tmp_obj = bo->sync_obj;
-			bo->sync_obj = NULL;
-			clear_bit(TTM_BO_PRIV_FLAG_MOVING,
-				  &bo->priv_flags);
-			spin_unlock(&bdev->fence_lock);
-			driver->sync_obj_unref(&sync_obj);
-			driver->sync_obj_unref(&tmp_obj);
-			spin_lock(&bdev->fence_lock);
-		} else {
-			spin_unlock(&bdev->fence_lock);
-			driver->sync_obj_unref(&sync_obj);
-			spin_lock(&bdev->fence_lock);
+		if (!fence_is_signaled(fence)) {
+			if (no_wait)
+				return -EBUSY;
+
+			timeout = fence_wait_timeout(fence,
+						     interruptible, timeout);
 		}
 	}
+
+	if (timeout < 0)
+		return timeout;
+
+	if (timeout == 0)
+		return -EBUSY;
+
+	reservation_object_add_excl_fence(resv, NULL);
+	clear_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags);
 	return 0;
 }
 EXPORT_SYMBOL(ttm_bo_wait);
 
 int ttm_bo_synccpu_write_grab(struct ttm_buffer_object *bo, bool no_wait)
 {
-	struct ttm_bo_device *bdev = bo->bdev;
 	int ret = 0;
 
 	/*
@@ -1583,9 +1564,7 @@ int ttm_bo_synccpu_write_grab(struct ttm_buffer_object *bo, bool no_wait)
 	ret = ttm_bo_reserve(bo, true, no_wait, false, NULL);
 	if (unlikely(ret != 0))
 		return ret;
-	spin_lock(&bdev->fence_lock);
 	ret = ttm_bo_wait(bo, false, true, no_wait);
-	spin_unlock(&bdev->fence_lock);
 	if (likely(ret == 0))
 		atomic_inc(&bo->cpu_writers);
 	ttm_bo_unreserve(bo);
@@ -1642,9 +1621,7 @@ static int ttm_bo_swapout(struct ttm_mem_shrink *shrink)
 	 * Wait for GPU, then move to system cached.
 	 */
 
-	spin_lock(&bo->bdev->fence_lock);
 	ret = ttm_bo_wait(bo, false, false, false);
-	spin_unlock(&bo->bdev->fence_lock);
 
 	if (unlikely(ret != 0))
 		goto out;
diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c
index 30e5d90cb7bc..824af90cbe31 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_util.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_util.c
@@ -37,6 +37,7 @@
 #include <linux/slab.h>
 #include <linux/vmalloc.h>
 #include <linux/module.h>
+#include <linux/reservation.h>
 
 void ttm_bo_free_old_node(struct ttm_buffer_object *bo)
 {
@@ -444,8 +445,6 @@ static int ttm_buffer_object_transfer(struct ttm_buffer_object *bo,
 				      struct ttm_buffer_object **new_obj)
 {
 	struct ttm_buffer_object *fbo;
-	struct ttm_bo_device *bdev = bo->bdev;
-	struct ttm_bo_driver *driver = bdev->driver;
 	int ret;
 
 	fbo = kmalloc(sizeof(*fbo), GFP_KERNEL);
@@ -466,12 +465,6 @@ static int ttm_buffer_object_transfer(struct ttm_buffer_object *bo,
 	drm_vma_node_reset(&fbo->vma_node);
 	atomic_set(&fbo->cpu_writers, 0);
 
-	spin_lock(&bdev->fence_lock);
-	if (bo->sync_obj)
-		fbo->sync_obj = driver->sync_obj_ref(bo->sync_obj);
-	else
-		fbo->sync_obj = NULL;
-	spin_unlock(&bdev->fence_lock);
 	kref_init(&fbo->list_kref);
 	kref_init(&fbo->kref);
 	fbo->destroy = &ttm_transfered_destroy;
@@ -644,30 +637,20 @@ void ttm_bo_kunmap(struct ttm_bo_kmap_obj *map)
 EXPORT_SYMBOL(ttm_bo_kunmap);
 
 int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo,
-			      void *sync_obj,
+			      struct fence *fence,
 			      bool evict,
 			      bool no_wait_gpu,
 			      struct ttm_mem_reg *new_mem)
 {
 	struct ttm_bo_device *bdev = bo->bdev;
-	struct ttm_bo_driver *driver = bdev->driver;
 	struct ttm_mem_type_manager *man = &bdev->man[new_mem->mem_type];
 	struct ttm_mem_reg *old_mem = &bo->mem;
 	int ret;
 	struct ttm_buffer_object *ghost_obj;
-	void *tmp_obj = NULL;
 
-	spin_lock(&bdev->fence_lock);
-	if (bo->sync_obj) {
-		tmp_obj = bo->sync_obj;
-		bo->sync_obj = NULL;
-	}
-	bo->sync_obj = driver->sync_obj_ref(sync_obj);
+	reservation_object_add_excl_fence(bo->resv, fence);
 	if (evict) {
 		ret = ttm_bo_wait(bo, false, false, false);
-		spin_unlock(&bdev->fence_lock);
-		if (tmp_obj)
-			driver->sync_obj_unref(&tmp_obj);
 		if (ret)
 			return ret;
 
@@ -688,14 +671,13 @@ int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo,
 		 */
 
 		set_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags);
-		spin_unlock(&bdev->fence_lock);
-		if (tmp_obj)
-			driver->sync_obj_unref(&tmp_obj);
 
 		ret = ttm_buffer_object_transfer(bo, &ghost_obj);
 		if (ret)
 			return ret;
 
+		reservation_object_add_excl_fence(ghost_obj->resv, fence);
+
 		/**
 		 * If we're not moving to fixed memory, the TTM object
 		 * needs to stay alive. Otherwhise hang it on the ghost
diff --git a/drivers/gpu/drm/ttm/ttm_bo_vm.c b/drivers/gpu/drm/ttm/ttm_bo_vm.c
index 0ce48e5a9cb4..d05437f219e9 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_vm.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c
@@ -45,10 +45,8 @@ static int ttm_bo_vm_fault_idle(struct ttm_buffer_object *bo,
 				struct vm_area_struct *vma,
 				struct vm_fault *vmf)
 {
-	struct ttm_bo_device *bdev = bo->bdev;
 	int ret = 0;
 
-	spin_lock(&bdev->fence_lock);
 	if (likely(!test_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags)))
 		goto out_unlock;
 
@@ -82,7 +80,6 @@ static int ttm_bo_vm_fault_idle(struct ttm_buffer_object *bo,
 			VM_FAULT_NOPAGE;
 
 out_unlock:
-	spin_unlock(&bdev->fence_lock);
 	return ret;
 }
 
diff --git a/drivers/gpu/drm/ttm/ttm_execbuf_util.c b/drivers/gpu/drm/ttm/ttm_execbuf_util.c
index e8dac8758528..adafc0f8ec06 100644
--- a/drivers/gpu/drm/ttm/ttm_execbuf_util.c
+++ b/drivers/gpu/drm/ttm/ttm_execbuf_util.c
@@ -32,20 +32,12 @@
 #include <linux/sched.h>
 #include <linux/module.h>
 
-static void ttm_eu_backoff_reservation_locked(struct list_head *list)
+static void ttm_eu_backoff_reservation_reverse(struct list_head *list,
+					      struct ttm_validate_buffer *entry)
 {
-	struct ttm_validate_buffer *entry;
-
-	list_for_each_entry(entry, list, head) {
+	list_for_each_entry_continue_reverse(entry, list, head) {
 		struct ttm_buffer_object *bo = entry->bo;
-		if (!entry->reserved)
-			continue;
 
-		entry->reserved = false;
-		if (entry->removed) {
-			ttm_bo_add_to_lru(bo);
-			entry->removed = false;
-		}
 		__ttm_bo_unreserve(bo);
 	}
 }
@@ -56,27 +48,9 @@ static void ttm_eu_del_from_lru_locked(struct list_head *list)
 
 	list_for_each_entry(entry, list, head) {
 		struct ttm_buffer_object *bo = entry->bo;
-		if (!entry->reserved)
-			continue;
-
-		if (!entry->removed) {
-			entry->put_count = ttm_bo_del_from_lru(bo);
-			entry->removed = true;
-		}
-	}
-}
-
-static void ttm_eu_list_ref_sub(struct list_head *list)
-{
-	struct ttm_validate_buffer *entry;
-
-	list_for_each_entry(entry, list, head) {
-		struct ttm_buffer_object *bo = entry->bo;
+		unsigned put_count = ttm_bo_del_from_lru(bo);
 
-		if (entry->put_count) {
-			ttm_bo_list_ref_sub(bo, entry->put_count, true);
-			entry->put_count = 0;
-		}
+		ttm_bo_list_ref_sub(bo, put_count, true);
 	}
 }
 
@@ -91,11 +65,18 @@ void ttm_eu_backoff_reservation(struct ww_acquire_ctx *ticket,
 
 	entry = list_first_entry(list, struct ttm_validate_buffer, head);
 	glob = entry->bo->glob;
+
 	spin_lock(&glob->lru_lock);
-	ttm_eu_backoff_reservation_locked(list);
+	list_for_each_entry(entry, list, head) {
+		struct ttm_buffer_object *bo = entry->bo;
+
+		ttm_bo_add_to_lru(bo);
+		__ttm_bo_unreserve(bo);
+	}
+	spin_unlock(&glob->lru_lock);
+
 	if (ticket)
 		ww_acquire_fini(ticket);
-	spin_unlock(&glob->lru_lock);
 }
 EXPORT_SYMBOL(ttm_eu_backoff_reservation);
 
@@ -112,7 +93,7 @@ EXPORT_SYMBOL(ttm_eu_backoff_reservation);
  */
 
 int ttm_eu_reserve_buffers(struct ww_acquire_ctx *ticket,
-			   struct list_head *list)
+			   struct list_head *list, bool intr)
 {
 	struct ttm_bo_global *glob;
 	struct ttm_validate_buffer *entry;
@@ -121,60 +102,55 @@ int ttm_eu_reserve_buffers(struct ww_acquire_ctx *ticket,
 	if (list_empty(list))
 		return 0;
 
-	list_for_each_entry(entry, list, head) {
-		entry->reserved = false;
-		entry->put_count = 0;
-		entry->removed = false;
-	}
-
 	entry = list_first_entry(list, struct ttm_validate_buffer, head);
 	glob = entry->bo->glob;
 
 	if (ticket)
 		ww_acquire_init(ticket, &reservation_ww_class);
-retry:
+
 	list_for_each_entry(entry, list, head) {
 		struct ttm_buffer_object *bo = entry->bo;
 
-		/* already slowpath reserved? */
-		if (entry->reserved)
+		ret = __ttm_bo_reserve(bo, intr, (ticket == NULL), true,
+				       ticket);
+		if (!ret && unlikely(atomic_read(&bo->cpu_writers) > 0)) {
+			__ttm_bo_unreserve(bo);
+
+			ret = -EBUSY;
+		}
+
+		if (!ret)
 			continue;
 
-		ret = __ttm_bo_reserve(bo, true, (ticket == NULL), true,
-				       ticket);
+		/* uh oh, we lost out, drop every reservation and try
+		 * to only reserve this buffer, then start over if
+		 * this succeeds.
+		 */
+		ttm_eu_backoff_reservation_reverse(list, entry);
 
-		if (ret == -EDEADLK) {
-			/* uh oh, we lost out, drop every reservation and try
-			 * to only reserve this buffer, then start over if
-			 * this succeeds.
-			 */
-			BUG_ON(ticket == NULL);
-			spin_lock(&glob->lru_lock);
-			ttm_eu_backoff_reservation_locked(list);
-			spin_unlock(&glob->lru_lock);
-			ttm_eu_list_ref_sub(list);
+		if (ret == -EDEADLK && intr) {
 			ret = ww_mutex_lock_slow_interruptible(&bo->resv->lock,
 							       ticket);
-			if (unlikely(ret != 0)) {
-				if (ret == -EINTR)
-					ret = -ERESTARTSYS;
-				goto err_fini;
-			}
+		} else if (ret == -EDEADLK) {
+			ww_mutex_lock_slow(&bo->resv->lock, ticket);
+			ret = 0;
+		}
 
-			entry->reserved = true;
-			if (unlikely(atomic_read(&bo->cpu_writers) > 0)) {
-				ret = -EBUSY;
-				goto err;
+		if (unlikely(ret != 0)) {
+			if (ret == -EINTR)
+				ret = -ERESTARTSYS;
+			if (ticket) {
+				ww_acquire_done(ticket);
+				ww_acquire_fini(ticket);
 			}
-			goto retry;
-		} else if (ret)
-			goto err;
-
-		entry->reserved = true;
-		if (unlikely(atomic_read(&bo->cpu_writers) > 0)) {
-			ret = -EBUSY;
-			goto err;
+			return ret;
 		}
+
+		/* move this item to the front of the list,
+		 * forces correct iteration of the loop without keeping track
+		 */
+		list_del(&entry->head);
+		list_add(&entry->head, list);
 	}
 
 	if (ticket)
@@ -182,25 +158,12 @@ retry:
 	spin_lock(&glob->lru_lock);
 	ttm_eu_del_from_lru_locked(list);
 	spin_unlock(&glob->lru_lock);
-	ttm_eu_list_ref_sub(list);
 	return 0;
-
-err:
-	spin_lock(&glob->lru_lock);
-	ttm_eu_backoff_reservation_locked(list);
-	spin_unlock(&glob->lru_lock);
-	ttm_eu_list_ref_sub(list);
-err_fini:
-	if (ticket) {
-		ww_acquire_done(ticket);
-		ww_acquire_fini(ticket);
-	}
-	return ret;
 }
 EXPORT_SYMBOL(ttm_eu_reserve_buffers);
 
 void ttm_eu_fence_buffer_objects(struct ww_acquire_ctx *ticket,
-				 struct list_head *list, void *sync_obj)
+				 struct list_head *list, struct fence *fence)
 {
 	struct ttm_validate_buffer *entry;
 	struct ttm_buffer_object *bo;
@@ -217,24 +180,15 @@ void ttm_eu_fence_buffer_objects(struct ww_acquire_ctx *ticket,
 	glob = bo->glob;
 
 	spin_lock(&glob->lru_lock);
-	spin_lock(&bdev->fence_lock);
 
 	list_for_each_entry(entry, list, head) {
 		bo = entry->bo;
-		entry->old_sync_obj = bo->sync_obj;
-		bo->sync_obj = driver->sync_obj_ref(sync_obj);
+		reservation_object_add_excl_fence(bo->resv, fence);
 		ttm_bo_add_to_lru(bo);
 		__ttm_bo_unreserve(bo);
-		entry->reserved = false;
 	}
-	spin_unlock(&bdev->fence_lock);
 	spin_unlock(&glob->lru_lock);
 	if (ticket)
 		ww_acquire_fini(ticket);
-
-	list_for_each_entry(entry, list, head) {
-		if (entry->old_sync_obj)
-			driver->sync_obj_unref(&entry->old_sync_obj);
-	}
 }
 EXPORT_SYMBOL(ttm_eu_fence_buffer_objects);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
index 37c093c0c7b8..cff2bf9db9d2 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
@@ -802,44 +802,6 @@ static int vmw_ttm_fault_reserve_notify(struct ttm_buffer_object *bo)
 }
 
 /**
- * FIXME: We're using the old vmware polling method to sync.
- * Do this with fences instead.
- */
-
-static void *vmw_sync_obj_ref(void *sync_obj)
-{
-
-	return (void *)
-		vmw_fence_obj_reference((struct vmw_fence_obj *) sync_obj);
-}
-
-static void vmw_sync_obj_unref(void **sync_obj)
-{
-	vmw_fence_obj_unreference((struct vmw_fence_obj **) sync_obj);
-}
-
-static int vmw_sync_obj_flush(void *sync_obj)
-{
-	vmw_fence_obj_flush((struct vmw_fence_obj *) sync_obj);
-	return 0;
-}
-
-static bool vmw_sync_obj_signaled(void *sync_obj)
-{
-	return	vmw_fence_obj_signaled((struct vmw_fence_obj *) sync_obj,
-				       DRM_VMW_FENCE_FLAG_EXEC);
-
-}
-
-static int vmw_sync_obj_wait(void *sync_obj, bool lazy, bool interruptible)
-{
-	return vmw_fence_obj_wait((struct vmw_fence_obj *) sync_obj,
-				  DRM_VMW_FENCE_FLAG_EXEC,
-				  lazy, interruptible,
-				  VMW_FENCE_WAIT_TIMEOUT);
-}
-
-/**
  * vmw_move_notify - TTM move_notify_callback
  *
  * @bo:             The TTM buffer object about to move.
@@ -863,11 +825,7 @@ static void vmw_move_notify(struct ttm_buffer_object *bo,
  */
 static void vmw_swap_notify(struct ttm_buffer_object *bo)
 {
-	struct ttm_bo_device *bdev = bo->bdev;
-
-	spin_lock(&bdev->fence_lock);
 	ttm_bo_wait(bo, false, false, false);
-	spin_unlock(&bdev->fence_lock);
 }
 
 
@@ -880,11 +838,6 @@ struct ttm_bo_driver vmw_bo_driver = {
 	.evict_flags = vmw_evict_flags,
 	.move = NULL,
 	.verify_access = vmw_verify_access,
-	.sync_obj_signaled = vmw_sync_obj_signaled,
-	.sync_obj_wait = vmw_sync_obj_wait,
-	.sync_obj_flush = vmw_sync_obj_flush,
-	.sync_obj_unref = vmw_sync_obj_unref,
-	.sync_obj_ref = vmw_sync_obj_ref,
 	.move_notify = vmw_move_notify,
 	.swap_notify = vmw_swap_notify,
 	.fault_reserve_notify = &vmw_ttm_fault_reserve_notify,
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index 99f731757c4b..4ee799b43d5d 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -342,7 +342,6 @@ struct vmw_sw_context{
 	uint32_t *cmd_bounce;
 	uint32_t cmd_bounce_size;
 	struct list_head resource_list;
-	uint32_t fence_flags;
 	struct ttm_buffer_object *cur_query_bo;
 	struct list_head res_relocations;
 	uint32_t *buf_start;
@@ -704,6 +703,7 @@ extern void *vmw_fifo_reserve(struct vmw_private *dev_priv, uint32_t bytes);
 extern void vmw_fifo_commit(struct vmw_private *dev_priv, uint32_t bytes);
 extern int vmw_fifo_send_fence(struct vmw_private *dev_priv,
 			       uint32_t *seqno);
+extern void vmw_fifo_ping_host_locked(struct vmw_private *, uint32_t reason);
 extern void vmw_fifo_ping_host(struct vmw_private *dev_priv, uint32_t reason);
 extern bool vmw_fifo_have_3d(struct vmw_private *dev_priv);
 extern bool vmw_fifo_have_pitchlock(struct vmw_private *dev_priv);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
index 7bfdaa163a33..0ceaddc8e4f7 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
@@ -346,13 +346,10 @@ static int vmw_bo_to_validate_list(struct vmw_sw_context *sw_context,
 		++sw_context->cur_val_buf;
 		val_buf = &vval_buf->base;
 		val_buf->bo = ttm_bo_reference(bo);
-		val_buf->reserved = false;
 		list_add_tail(&val_buf->head, &sw_context->validate_nodes);
 		vval_buf->validate_as_mob = validate_as_mob;
 	}
 
-	sw_context->fence_flags |= DRM_VMW_FENCE_FLAG_EXEC;
-
 	if (p_val_node)
 		*p_val_node = val_node;
 
@@ -2338,13 +2335,9 @@ int vmw_execbuf_fence_commands(struct drm_file *file_priv,
 
 	if (p_handle != NULL)
 		ret = vmw_user_fence_create(file_priv, dev_priv->fman,
-					    sequence,
-					    DRM_VMW_FENCE_FLAG_EXEC,
-					    p_fence, p_handle);
+					    sequence, p_fence, p_handle);
 	else
-		ret = vmw_fence_create(dev_priv->fman, sequence,
-				       DRM_VMW_FENCE_FLAG_EXEC,
-				       p_fence);
+		ret = vmw_fence_create(dev_priv->fman, sequence, p_fence);
 
 	if (unlikely(ret != 0 && !synced)) {
 		(void) vmw_fallback_wait(dev_priv, false, false,
@@ -2396,7 +2389,7 @@ vmw_execbuf_copy_fence_user(struct vmw_private *dev_priv,
 		BUG_ON(fence == NULL);
 
 		fence_rep.handle = fence_handle;
-		fence_rep.seqno = fence->seqno;
+		fence_rep.seqno = fence->base.seqno;
 		vmw_update_seqno(dev_priv, &dev_priv->fifo);
 		fence_rep.passed_seqno = dev_priv->last_read_seqno;
 	}
@@ -2417,8 +2410,7 @@ vmw_execbuf_copy_fence_user(struct vmw_private *dev_priv,
 		ttm_ref_object_base_unref(vmw_fp->tfile,
 					  fence_handle, TTM_REF_USAGE);
 		DRM_ERROR("Fence copy error. Syncing.\n");
-		(void) vmw_fence_obj_wait(fence, fence->signal_mask,
-					  false, false,
+		(void) vmw_fence_obj_wait(fence, false, false,
 					  VMW_FENCE_WAIT_TIMEOUT);
 	}
 }
@@ -2470,7 +2462,6 @@ int vmw_execbuf_process(struct drm_file *file_priv,
 	sw_context->fp = vmw_fpriv(file_priv);
 	sw_context->cur_reloc = 0;
 	sw_context->cur_val_buf = 0;
-	sw_context->fence_flags = 0;
 	INIT_LIST_HEAD(&sw_context->resource_list);
 	sw_context->cur_query_bo = dev_priv->pinned_bo;
 	sw_context->last_query_ctx = NULL;
@@ -2496,7 +2487,7 @@ int vmw_execbuf_process(struct drm_file *file_priv,
 	if (unlikely(ret != 0))
 		goto out_err_nores;
 
-	ret = ttm_eu_reserve_buffers(&ticket, &sw_context->validate_nodes);
+	ret = ttm_eu_reserve_buffers(&ticket, &sw_context->validate_nodes, true);
 	if (unlikely(ret != 0))
 		goto out_err;
 
@@ -2684,10 +2675,7 @@ void __vmw_execbuf_release_pinned_bo(struct vmw_private *dev_priv,
 	query_val.bo = ttm_bo_reference(dev_priv->dummy_query_bo);
 	list_add_tail(&query_val.head, &validate_list);
 
-	do {
-		ret = ttm_eu_reserve_buffers(&ticket, &validate_list);
-	} while (ret == -ERESTARTSYS);
-
+	ret = ttm_eu_reserve_buffers(&ticket, &validate_list, false);
 	if (unlikely(ret != 0)) {
 		vmw_execbuf_unpin_panic(dev_priv);
 		goto out_no_reserve;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
index 436b013b4231..197164fd7803 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
@@ -35,7 +35,7 @@ struct vmw_fence_manager {
 	struct vmw_private *dev_priv;
 	spinlock_t lock;
 	struct list_head fence_list;
-	struct work_struct work;
+	struct work_struct work, ping_work;
 	u32 user_fence_size;
 	u32 fence_size;
 	u32 event_fence_action_size;
@@ -46,6 +46,7 @@ struct vmw_fence_manager {
 	bool goal_irq_on; /* Protected by @goal_irq_mutex */
 	bool seqno_valid; /* Protected by @lock, and may not be set to true
 			     without the @goal_irq_mutex held. */
+	unsigned ctx;
 };
 
 struct vmw_user_fence {
@@ -80,6 +81,12 @@ struct vmw_event_fence_action {
 	uint32_t *tv_usec;
 };
 
+static struct vmw_fence_manager *
+fman_from_fence(struct vmw_fence_obj *fence)
+{
+	return container_of(fence->base.lock, struct vmw_fence_manager, lock);
+}
+
 /**
  * Note on fencing subsystem usage of irqs:
  * Typically the vmw_fences_update function is called
@@ -102,25 +109,143 @@ struct vmw_event_fence_action {
  * objects with actions attached to them.
  */
 
-static void vmw_fence_obj_destroy_locked(struct kref *kref)
+static void vmw_fence_obj_destroy(struct fence *f)
 {
 	struct vmw_fence_obj *fence =
-		container_of(kref, struct vmw_fence_obj, kref);
+		container_of(f, struct vmw_fence_obj, base);
 
-	struct vmw_fence_manager *fman = fence->fman;
-	unsigned int num_fences;
+	struct vmw_fence_manager *fman = fman_from_fence(fence);
+	unsigned long irq_flags;
 
+	spin_lock_irqsave(&fman->lock, irq_flags);
 	list_del_init(&fence->head);
-	num_fences = --fman->num_fence_objects;
-	spin_unlock_irq(&fman->lock);
-	if (fence->destroy)
-		fence->destroy(fence);
-	else
-		kfree(fence);
+	--fman->num_fence_objects;
+	spin_unlock_irqrestore(&fman->lock, irq_flags);
+	fence->destroy(fence);
+}
 
-	spin_lock_irq(&fman->lock);
+static const char *vmw_fence_get_driver_name(struct fence *f)
+{
+	return "vmwgfx";
+}
+
+static const char *vmw_fence_get_timeline_name(struct fence *f)
+{
+	return "svga";
+}
+
+static void vmw_fence_ping_func(struct work_struct *work)
+{
+	struct vmw_fence_manager *fman =
+		container_of(work, struct vmw_fence_manager, ping_work);
+
+	vmw_fifo_ping_host(fman->dev_priv, SVGA_SYNC_GENERIC);
+}
+
+static bool vmw_fence_enable_signaling(struct fence *f)
+{
+	struct vmw_fence_obj *fence =
+		container_of(f, struct vmw_fence_obj, base);
+
+	struct vmw_fence_manager *fman = fman_from_fence(fence);
+	struct vmw_private *dev_priv = fman->dev_priv;
+
+	__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
+	u32 seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE);
+	if (seqno - fence->base.seqno < VMW_FENCE_WRAP)
+		return false;
+
+	if (mutex_trylock(&dev_priv->hw_mutex)) {
+		vmw_fifo_ping_host_locked(dev_priv, SVGA_SYNC_GENERIC);
+		mutex_unlock(&dev_priv->hw_mutex);
+	} else
+		schedule_work(&fman->ping_work);
+
+	return true;
+}
+
+struct vmwgfx_wait_cb {
+	struct fence_cb base;
+	struct task_struct *task;
+};
+
+static void
+vmwgfx_wait_cb(struct fence *fence, struct fence_cb *cb)
+{
+	struct vmwgfx_wait_cb *wait =
+		container_of(cb, struct vmwgfx_wait_cb, base);
+
+	wake_up_process(wait->task);
+}
+
+static void __vmw_fences_update(struct vmw_fence_manager *fman);
+
+static long vmw_fence_wait(struct fence *f, bool intr, signed long timeout)
+{
+	struct vmw_fence_obj *fence =
+		container_of(f, struct vmw_fence_obj, base);
+
+	struct vmw_fence_manager *fman = fman_from_fence(fence);
+	struct vmw_private *dev_priv = fman->dev_priv;
+	struct vmwgfx_wait_cb cb;
+	long ret = timeout;
+	unsigned long irq_flags;
+
+	if (likely(vmw_fence_obj_signaled(fence)))
+		return timeout;
+
+	vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC);
+	vmw_seqno_waiter_add(dev_priv);
+
+	spin_lock_irqsave(f->lock, irq_flags);
+
+	if (intr && signal_pending(current)) {
+		ret = -ERESTARTSYS;
+		goto out;
+	}
+
+	cb.base.func = vmwgfx_wait_cb;
+	cb.task = current;
+	list_add(&cb.base.node, &f->cb_list);
+
+	while (ret > 0) {
+		__vmw_fences_update(fman);
+		if (test_bit(FENCE_FLAG_SIGNALED_BIT, &f->flags))
+			break;
+
+		if (intr)
+			__set_current_state(TASK_INTERRUPTIBLE);
+		else
+			__set_current_state(TASK_UNINTERRUPTIBLE);
+		spin_unlock_irqrestore(f->lock, irq_flags);
+
+		ret = schedule_timeout(ret);
+
+		spin_lock_irqsave(f->lock, irq_flags);
+		if (ret > 0 && intr && signal_pending(current))
+			ret = -ERESTARTSYS;
+	}
+
+	if (!list_empty(&cb.base.node))
+		list_del(&cb.base.node);
+	__set_current_state(TASK_RUNNING);
+
+out:
+	spin_unlock_irqrestore(f->lock, irq_flags);
+
+	vmw_seqno_waiter_remove(dev_priv);
+
+	return ret;
 }
 
+static struct fence_ops vmw_fence_ops = {
+	.get_driver_name = vmw_fence_get_driver_name,
+	.get_timeline_name = vmw_fence_get_timeline_name,
+	.enable_signaling = vmw_fence_enable_signaling,
+	.wait = vmw_fence_wait,
+	.release = vmw_fence_obj_destroy,
+};
+
 
 /**
  * Execute signal actions on fences recently signaled.
@@ -180,12 +305,14 @@ struct vmw_fence_manager *vmw_fence_manager_init(struct vmw_private *dev_priv)
 	INIT_LIST_HEAD(&fman->fence_list);
 	INIT_LIST_HEAD(&fman->cleanup_list);
 	INIT_WORK(&fman->work, &vmw_fence_work_func);
+	INIT_WORK(&fman->ping_work, &vmw_fence_ping_func);
 	fman->fifo_down = true;
 	fman->user_fence_size = ttm_round_pot(sizeof(struct vmw_user_fence));
 	fman->fence_size = ttm_round_pot(sizeof(struct vmw_fence_obj));
 	fman->event_fence_action_size =
 		ttm_round_pot(sizeof(struct vmw_event_fence_action));
 	mutex_init(&fman->goal_irq_mutex);
+	fman->ctx = fence_context_alloc(1);
 
 	return fman;
 }
@@ -196,6 +323,7 @@ void vmw_fence_manager_takedown(struct vmw_fence_manager *fman)
 	bool lists_empty;
 
 	(void) cancel_work_sync(&fman->work);
+	(void) cancel_work_sync(&fman->ping_work);
 
 	spin_lock_irqsave(&fman->lock, irq_flags);
 	lists_empty = list_empty(&fman->fence_list) &&
@@ -207,23 +335,16 @@ void vmw_fence_manager_takedown(struct vmw_fence_manager *fman)
 }
 
 static int vmw_fence_obj_init(struct vmw_fence_manager *fman,
-			      struct vmw_fence_obj *fence,
-			      u32 seqno,
-			      uint32_t mask,
+			      struct vmw_fence_obj *fence, u32 seqno,
 			      void (*destroy) (struct vmw_fence_obj *fence))
 {
 	unsigned long irq_flags;
-	unsigned int num_fences;
 	int ret = 0;
 
-	fence->seqno = seqno;
+	fence_init(&fence->base, &vmw_fence_ops, &fman->lock,
+		   fman->ctx, seqno);
 	INIT_LIST_HEAD(&fence->seq_passed_actions);
-	fence->fman = fman;
-	fence->signaled = 0;
-	fence->signal_mask = mask;
-	kref_init(&fence->kref);
 	fence->destroy = destroy;
-	init_waitqueue_head(&fence->queue);
 
 	spin_lock_irqsave(&fman->lock, irq_flags);
 	if (unlikely(fman->fifo_down)) {
@@ -231,7 +352,7 @@ static int vmw_fence_obj_init(struct vmw_fence_manager *fman,
 		goto out_unlock;
 	}
 	list_add_tail(&fence->head, &fman->fence_list);
-	num_fences = ++fman->num_fence_objects;
+	++fman->num_fence_objects;
 
 out_unlock:
 	spin_unlock_irqrestore(&fman->lock, irq_flags);
@@ -239,38 +360,6 @@ out_unlock:
 
 }
 
-struct vmw_fence_obj *vmw_fence_obj_reference(struct vmw_fence_obj *fence)
-{
-	if (unlikely(fence == NULL))
-		return NULL;
-
-	kref_get(&fence->kref);
-	return fence;
-}
-
-/**
- * vmw_fence_obj_unreference
- *
- * Note that this function may not be entered with disabled irqs since
- * it may re-enable them in the destroy function.
- *
- */
-void vmw_fence_obj_unreference(struct vmw_fence_obj **fence_p)
-{
-	struct vmw_fence_obj *fence = *fence_p;
-	struct vmw_fence_manager *fman;
-
-	if (unlikely(fence == NULL))
-		return;
-
-	fman = fence->fman;
-	*fence_p = NULL;
-	spin_lock_irq(&fman->lock);
-	BUG_ON(atomic_read(&fence->kref.refcount) == 0);
-	kref_put(&fence->kref, vmw_fence_obj_destroy_locked);
-	spin_unlock_irq(&fman->lock);
-}
-
 static void vmw_fences_perform_actions(struct vmw_fence_manager *fman,
 				struct list_head *list)
 {
@@ -326,7 +415,7 @@ static bool vmw_fence_goal_new_locked(struct vmw_fence_manager *fman,
 	list_for_each_entry(fence, &fman->fence_list, head) {
 		if (!list_empty(&fence->seq_passed_actions)) {
 			fman->seqno_valid = true;
-			iowrite32(fence->seqno,
+			iowrite32(fence->base.seqno,
 				  fifo_mem + SVGA_FIFO_FENCE_GOAL);
 			break;
 		}
@@ -353,27 +442,27 @@ static bool vmw_fence_goal_new_locked(struct vmw_fence_manager *fman,
  */
 static bool vmw_fence_goal_check_locked(struct vmw_fence_obj *fence)
 {
+	struct vmw_fence_manager *fman = fman_from_fence(fence);
 	u32 goal_seqno;
 	__le32 __iomem *fifo_mem;
 
-	if (fence->signaled & DRM_VMW_FENCE_FLAG_EXEC)
+	if (fence_is_signaled_locked(&fence->base))
 		return false;
 
-	fifo_mem = fence->fman->dev_priv->mmio_virt;
+	fifo_mem = fman->dev_priv->mmio_virt;
 	goal_seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE_GOAL);
-	if (likely(fence->fman->seqno_valid &&
-		   goal_seqno - fence->seqno < VMW_FENCE_WRAP))
+	if (likely(fman->seqno_valid &&
+		   goal_seqno - fence->base.seqno < VMW_FENCE_WRAP))
 		return false;
 
-	iowrite32(fence->seqno, fifo_mem + SVGA_FIFO_FENCE_GOAL);
-	fence->fman->seqno_valid = true;
+	iowrite32(fence->base.seqno, fifo_mem + SVGA_FIFO_FENCE_GOAL);
+	fman->seqno_valid = true;
 
 	return true;
 }
 
-void vmw_fences_update(struct vmw_fence_manager *fman)
+static void __vmw_fences_update(struct vmw_fence_manager *fman)
 {
-	unsigned long flags;
 	struct vmw_fence_obj *fence, *next_fence;
 	struct list_head action_list;
 	bool needs_rerun;
@@ -382,32 +471,25 @@ void vmw_fences_update(struct vmw_fence_manager *fman)
 
 	seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE);
 rerun:
-	spin_lock_irqsave(&fman->lock, flags);
 	list_for_each_entry_safe(fence, next_fence, &fman->fence_list, head) {
-		if (seqno - fence->seqno < VMW_FENCE_WRAP) {
+		if (seqno - fence->base.seqno < VMW_FENCE_WRAP) {
 			list_del_init(&fence->head);
-			fence->signaled |= DRM_VMW_FENCE_FLAG_EXEC;
+			fence_signal_locked(&fence->base);
 			INIT_LIST_HEAD(&action_list);
 			list_splice_init(&fence->seq_passed_actions,
 					 &action_list);
 			vmw_fences_perform_actions(fman, &action_list);
-			wake_up_all(&fence->queue);
 		} else
 			break;
 	}
 
-	needs_rerun = vmw_fence_goal_new_locked(fman, seqno);
-
-	if (!list_empty(&fman->cleanup_list))
-		(void) schedule_work(&fman->work);
-	spin_unlock_irqrestore(&fman->lock, flags);
-
 	/*
 	 * Rerun if the fence goal seqno was updated, and the
 	 * hardware might have raced with that update, so that
 	 * we missed a fence_goal irq.
 	 */
 
+	needs_rerun = vmw_fence_goal_new_locked(fman, seqno);
 	if (unlikely(needs_rerun)) {
 		new_seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE);
 		if (new_seqno != seqno) {
@@ -415,79 +497,58 @@ rerun:
 			goto rerun;
 		}
 	}
+
+	if (!list_empty(&fman->cleanup_list))
+		(void) schedule_work(&fman->work);
 }
 
-bool vmw_fence_obj_signaled(struct vmw_fence_obj *fence,
-			    uint32_t flags)
+void vmw_fences_update(struct vmw_fence_manager *fman)
 {
-	struct vmw_fence_manager *fman = fence->fman;
 	unsigned long irq_flags;
-	uint32_t signaled;
 
 	spin_lock_irqsave(&fman->lock, irq_flags);
-	signaled = fence->signaled;
+	__vmw_fences_update(fman);
 	spin_unlock_irqrestore(&fman->lock, irq_flags);
+}
 
-	flags &= fence->signal_mask;
-	if ((signaled & flags) == flags)
-		return 1;
+bool vmw_fence_obj_signaled(struct vmw_fence_obj *fence)
+{
+	struct vmw_fence_manager *fman = fman_from_fence(fence);
 
-	if ((signaled & DRM_VMW_FENCE_FLAG_EXEC) == 0)
-		vmw_fences_update(fman);
+	if (test_bit(FENCE_FLAG_SIGNALED_BIT, &fence->base.flags))
+		return 1;
 
-	spin_lock_irqsave(&fman->lock, irq_flags);
-	signaled = fence->signaled;
-	spin_unlock_irqrestore(&fman->lock, irq_flags);
+	vmw_fences_update(fman);
 
-	return ((signaled & flags) == flags);
+	return fence_is_signaled(&fence->base);
 }
 
-int vmw_fence_obj_wait(struct vmw_fence_obj *fence,
-		       uint32_t flags, bool lazy,
+int vmw_fence_obj_wait(struct vmw_fence_obj *fence, bool lazy,
 		       bool interruptible, unsigned long timeout)
 {
-	struct vmw_private *dev_priv = fence->fman->dev_priv;
-	long ret;
+	long ret = fence_wait_timeout(&fence->base, interruptible, timeout);
 
-	if (likely(vmw_fence_obj_signaled(fence, flags)))
+	if (likely(ret > 0))
 		return 0;
-
-	vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC);
-	vmw_seqno_waiter_add(dev_priv);
-
-	if (interruptible)
-		ret = wait_event_interruptible_timeout
-			(fence->queue,
-			 vmw_fence_obj_signaled(fence, flags),
-			 timeout);
+	else if (ret == 0)
+		return -EBUSY;
 	else
-		ret = wait_event_timeout
-			(fence->queue,
-			 vmw_fence_obj_signaled(fence, flags),
-			 timeout);
-
-	vmw_seqno_waiter_remove(dev_priv);
-
-	if (unlikely(ret == 0))
-		ret = -EBUSY;
-	else if (likely(ret > 0))
-		ret = 0;
-
-	return ret;
+		return ret;
 }
 
 void vmw_fence_obj_flush(struct vmw_fence_obj *fence)
 {
-	struct vmw_private *dev_priv = fence->fman->dev_priv;
+	struct vmw_private *dev_priv = fman_from_fence(fence)->dev_priv;
 
 	vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC);
 }
 
 static void vmw_fence_destroy(struct vmw_fence_obj *fence)
 {
-	struct vmw_fence_manager *fman = fence->fman;
+	struct vmw_fence_manager *fman = fman_from_fence(fence);
+
+	fence_free(&fence->base);
 
-	kfree(fence);
 	/*
 	 * Free kernel space accounting.
 	 */
@@ -497,7 +558,6 @@ static void vmw_fence_destroy(struct vmw_fence_obj *fence)
 
 int vmw_fence_create(struct vmw_fence_manager *fman,
 		     uint32_t seqno,
-		     uint32_t mask,
 		     struct vmw_fence_obj **p_fence)
 {
 	struct ttm_mem_global *mem_glob = vmw_mem_glob(fman->dev_priv);
@@ -515,7 +575,7 @@ int vmw_fence_create(struct vmw_fence_manager *fman,
 		goto out_no_object;
 	}
 
-	ret = vmw_fence_obj_init(fman, fence, seqno, mask,
+	ret = vmw_fence_obj_init(fman, fence, seqno,
 				 vmw_fence_destroy);
 	if (unlikely(ret != 0))
 		goto out_err_init;
@@ -535,7 +595,7 @@ static void vmw_user_fence_destroy(struct vmw_fence_obj *fence)
 {
 	struct vmw_user_fence *ufence =
 		container_of(fence, struct vmw_user_fence, fence);
-	struct vmw_fence_manager *fman = fence->fman;
+	struct vmw_fence_manager *fman = fman_from_fence(fence);
 
 	ttm_base_object_kfree(ufence, base);
 	/*
@@ -559,7 +619,6 @@ static void vmw_user_fence_base_release(struct ttm_base_object **p_base)
 int vmw_user_fence_create(struct drm_file *file_priv,
 			  struct vmw_fence_manager *fman,
 			  uint32_t seqno,
-			  uint32_t mask,
 			  struct vmw_fence_obj **p_fence,
 			  uint32_t *p_handle)
 {
@@ -586,7 +645,7 @@ int vmw_user_fence_create(struct drm_file *file_priv,
 	}
 
 	ret = vmw_fence_obj_init(fman, &ufence->fence, seqno,
-				 mask, vmw_user_fence_destroy);
+				 vmw_user_fence_destroy);
 	if (unlikely(ret != 0)) {
 		kfree(ufence);
 		goto out_no_object;
@@ -629,7 +688,6 @@ out_no_object:
 
 void vmw_fence_fifo_down(struct vmw_fence_manager *fman)
 {
-	unsigned long irq_flags;
 	struct list_head action_list;
 	int ret;
 
@@ -638,35 +696,32 @@ void vmw_fence_fifo_down(struct vmw_fence_manager *fman)
 	 * restart when we've released the fman->lock.
 	 */
 
-	spin_lock_irqsave(&fman->lock, irq_flags);
+	spin_lock_irq(&fman->lock);
 	fman->fifo_down = true;
 	while (!list_empty(&fman->fence_list)) {
 		struct vmw_fence_obj *fence =
 			list_entry(fman->fence_list.prev, struct vmw_fence_obj,
 				   head);
-		kref_get(&fence->kref);
+		fence_get(&fence->base);
 		spin_unlock_irq(&fman->lock);
 
-		ret = vmw_fence_obj_wait(fence, fence->signal_mask,
-					 false, false,
+		ret = vmw_fence_obj_wait(fence, false, false,
 					 VMW_FENCE_WAIT_TIMEOUT);
 
 		if (unlikely(ret != 0)) {
 			list_del_init(&fence->head);
-			fence->signaled |= DRM_VMW_FENCE_FLAG_EXEC;
+			fence_signal(&fence->base);
 			INIT_LIST_HEAD(&action_list);
 			list_splice_init(&fence->seq_passed_actions,
 					 &action_list);
 			vmw_fences_perform_actions(fman, &action_list);
-			wake_up_all(&fence->queue);
 		}
 
-		spin_lock_irq(&fman->lock);
-
 		BUG_ON(!list_empty(&fence->head));
-		kref_put(&fence->kref, vmw_fence_obj_destroy_locked);
+		fence_put(&fence->base);
+		spin_lock_irq(&fman->lock);
 	}
-	spin_unlock_irqrestore(&fman->lock, irq_flags);
+	spin_unlock_irq(&fman->lock);
 }
 
 void vmw_fence_fifo_up(struct vmw_fence_manager *fman)
@@ -716,14 +771,14 @@ int vmw_fence_obj_wait_ioctl(struct drm_device *dev, void *data,
 
 	timeout = jiffies;
 	if (time_after_eq(timeout, (unsigned long)arg->kernel_cookie)) {
-		ret = ((vmw_fence_obj_signaled(fence, arg->flags)) ?
+		ret = ((vmw_fence_obj_signaled(fence)) ?
 		       0 : -EBUSY);
 		goto out;
 	}
 
 	timeout = (unsigned long)arg->kernel_cookie - timeout;
 
-	ret = vmw_fence_obj_wait(fence, arg->flags, arg->lazy, true, timeout);
+	ret = vmw_fence_obj_wait(fence, arg->lazy, true, timeout);
 
 out:
 	ttm_base_object_unref(&base);
@@ -758,12 +813,12 @@ int vmw_fence_obj_signaled_ioctl(struct drm_device *dev, void *data,
 	}
 
 	fence = &(container_of(base, struct vmw_user_fence, base)->fence);
-	fman = fence->fman;
+	fman = fman_from_fence(fence);
 
-	arg->signaled = vmw_fence_obj_signaled(fence, arg->flags);
-	spin_lock_irq(&fman->lock);
+	arg->signaled = vmw_fence_obj_signaled(fence);
 
-	arg->signaled_flags = fence->signaled;
+	arg->signaled_flags = arg->flags;
+	spin_lock_irq(&fman->lock);
 	arg->passed_seqno = dev_priv->last_read_seqno;
 	spin_unlock_irq(&fman->lock);
 
@@ -876,7 +931,7 @@ static void vmw_event_fence_action_cleanup(struct vmw_fence_action *action)
 {
 	struct vmw_event_fence_action *eaction =
 		container_of(action, struct vmw_event_fence_action, action);
-	struct vmw_fence_manager *fman = eaction->fence->fman;
+	struct vmw_fence_manager *fman = fman_from_fence(eaction->fence);
 	unsigned long irq_flags;
 
 	spin_lock_irqsave(&fman->lock, irq_flags);
@@ -900,7 +955,7 @@ static void vmw_event_fence_action_cleanup(struct vmw_fence_action *action)
 static void vmw_fence_obj_add_action(struct vmw_fence_obj *fence,
 			      struct vmw_fence_action *action)
 {
-	struct vmw_fence_manager *fman = fence->fman;
+	struct vmw_fence_manager *fman = fman_from_fence(fence);
 	unsigned long irq_flags;
 	bool run_update = false;
 
@@ -908,7 +963,7 @@ static void vmw_fence_obj_add_action(struct vmw_fence_obj *fence,
 	spin_lock_irqsave(&fman->lock, irq_flags);
 
 	fman->pending_actions[action->type]++;
-	if (fence->signaled & DRM_VMW_FENCE_FLAG_EXEC) {
+	if (fence_is_signaled_locked(&fence->base)) {
 		struct list_head action_list;
 
 		INIT_LIST_HEAD(&action_list);
@@ -960,7 +1015,7 @@ int vmw_event_fence_action_queue(struct drm_file *file_priv,
 				 bool interruptible)
 {
 	struct vmw_event_fence_action *eaction;
-	struct vmw_fence_manager *fman = fence->fman;
+	struct vmw_fence_manager *fman = fman_from_fence(fence);
 	struct vmw_fpriv *vmw_fp = vmw_fpriv(file_priv);
 	unsigned long irq_flags;
 
@@ -1000,7 +1055,8 @@ static int vmw_event_fence_action_create(struct drm_file *file_priv,
 				  bool interruptible)
 {
 	struct vmw_event_fence_pending *event;
-	struct drm_device *dev = fence->fman->dev_priv->dev;
+	struct vmw_fence_manager *fman = fman_from_fence(fence);
+	struct drm_device *dev = fman->dev_priv->dev;
 	unsigned long irq_flags;
 	int ret;
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.h b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.h
index faf2e7873860..26a4add39208 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.h
@@ -27,6 +27,8 @@
 
 #ifndef _VMWGFX_FENCE_H_
 
+#include <linux/fence.h>
+
 #define VMW_FENCE_WAIT_TIMEOUT (5*HZ)
 
 struct vmw_private;
@@ -50,16 +52,11 @@ struct vmw_fence_action {
 };
 
 struct vmw_fence_obj {
-	struct kref kref;
-	u32 seqno;
+	struct fence base;
 
-	struct vmw_fence_manager *fman;
 	struct list_head head;
-	uint32_t signaled;
-	uint32_t signal_mask;
 	struct list_head seq_passed_actions;
 	void (*destroy)(struct vmw_fence_obj *fence);
-	wait_queue_head_t queue;
 };
 
 extern struct vmw_fence_manager *
@@ -67,17 +64,29 @@ vmw_fence_manager_init(struct vmw_private *dev_priv);
 
 extern void vmw_fence_manager_takedown(struct vmw_fence_manager *fman);
 
-extern void vmw_fence_obj_unreference(struct vmw_fence_obj **fence_p);
+static inline void
+vmw_fence_obj_unreference(struct vmw_fence_obj **fence_p)
+{
+	struct vmw_fence_obj *fence = *fence_p;
+
+	*fence_p = NULL;
+	if (fence)
+		fence_put(&fence->base);
+}
 
-extern struct vmw_fence_obj *
-vmw_fence_obj_reference(struct vmw_fence_obj *fence);
+static inline struct vmw_fence_obj *
+vmw_fence_obj_reference(struct vmw_fence_obj *fence)
+{
+	if (fence)
+		fence_get(&fence->base);
+	return fence;
+}
 
 extern void vmw_fences_update(struct vmw_fence_manager *fman);
 
-extern bool vmw_fence_obj_signaled(struct vmw_fence_obj *fence,
-				   uint32_t flags);
+extern bool vmw_fence_obj_signaled(struct vmw_fence_obj *fence);
 
-extern int vmw_fence_obj_wait(struct vmw_fence_obj *fence, uint32_t flags,
+extern int vmw_fence_obj_wait(struct vmw_fence_obj *fence,
 			      bool lazy,
 			      bool interruptible, unsigned long timeout);
 
@@ -85,13 +94,11 @@ extern void vmw_fence_obj_flush(struct vmw_fence_obj *fence);
 
 extern int vmw_fence_create(struct vmw_fence_manager *fman,
 			    uint32_t seqno,
-			    uint32_t mask,
 			    struct vmw_fence_obj **p_fence);
 
 extern int vmw_user_fence_create(struct drm_file *file_priv,
 				 struct vmw_fence_manager *fman,
 				 uint32_t sequence,
-				 uint32_t mask,
 				 struct vmw_fence_obj **p_fence,
 				 uint32_t *p_handle);
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
index 6ccd993e26bf..d9b4e6959750 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
@@ -160,16 +160,21 @@ int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 	return vmw_fifo_send_fence(dev_priv, &dummy);
 }
 
-void vmw_fifo_ping_host(struct vmw_private *dev_priv, uint32_t reason)
+void vmw_fifo_ping_host_locked(struct vmw_private *dev_priv, uint32_t reason)
 {
 	__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
 
-	mutex_lock(&dev_priv->hw_mutex);
-
 	if (unlikely(ioread32(fifo_mem + SVGA_FIFO_BUSY) == 0)) {
 		iowrite32(1, fifo_mem + SVGA_FIFO_BUSY);
 		vmw_write(dev_priv, SVGA_REG_SYNC, reason);
 	}
+}
+
+void vmw_fifo_ping_host(struct vmw_private *dev_priv, uint32_t reason)
+{
+	mutex_lock(&dev_priv->hw_mutex);
+
+	vmw_fifo_ping_host_locked(dev_priv, reason);
 
 	mutex_unlock(&dev_priv->hw_mutex);
 }
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
index a432c0db257c..ff0e03b97753 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
@@ -567,13 +567,18 @@ static int vmw_user_dmabuf_synccpu_grab(struct vmw_user_dma_buffer *user_bo,
 	int ret;
 
 	if (flags & drm_vmw_synccpu_allow_cs) {
-		struct ttm_bo_device *bdev = bo->bdev;
+		bool nonblock = !!(flags & drm_vmw_synccpu_dontblock);
+		long lret;
 
-		spin_lock(&bdev->fence_lock);
-		ret = ttm_bo_wait(bo, false, true,
-				  !!(flags & drm_vmw_synccpu_dontblock));
-		spin_unlock(&bdev->fence_lock);
-		return ret;
+		if (nonblock)
+			return reservation_object_test_signaled_rcu(bo->resv, true) ? 0 : -EBUSY;
+
+		lret = reservation_object_wait_timeout_rcu(bo->resv, true, true, MAX_SCHEDULE_TIMEOUT);
+		if (!lret)
+			return -EBUSY;
+		else if (lret < 0)
+			return lret;
+		return 0;
 	}
 
 	ret = ttm_bo_synccpu_write_grab
@@ -1215,7 +1220,7 @@ vmw_resource_check_buffer(struct vmw_resource *res,
 	INIT_LIST_HEAD(&val_list);
 	val_buf->bo = ttm_bo_reference(&res->backup->base);
 	list_add_tail(&val_buf->head, &val_list);
-	ret = ttm_eu_reserve_buffers(NULL, &val_list);
+	ret = ttm_eu_reserve_buffers(NULL, &val_list, interruptible);
 	if (unlikely(ret != 0))
 		goto out_no_reserve;
 
@@ -1419,25 +1424,16 @@ void vmw_fence_single_bo(struct ttm_buffer_object *bo,
 			 struct vmw_fence_obj *fence)
 {
 	struct ttm_bo_device *bdev = bo->bdev;
-	struct ttm_bo_driver *driver = bdev->driver;
-	struct vmw_fence_obj *old_fence_obj;
+
 	struct vmw_private *dev_priv =
 		container_of(bdev, struct vmw_private, bdev);
 
-	if (fence == NULL)
+	if (fence == NULL) {
 		vmw_execbuf_fence_commands(NULL, dev_priv, &fence, NULL);
-	else
-		driver->sync_obj_ref(fence);
-
-	spin_lock(&bdev->fence_lock);
-
-	old_fence_obj = bo->sync_obj;
-	bo->sync_obj = fence;
-
-	spin_unlock(&bdev->fence_lock);
-
-	if (old_fence_obj)
-		vmw_fence_obj_unreference(&old_fence_obj);
+		reservation_object_add_excl_fence(bo->resv, &fence->base);
+		fence_put(&fence->base);
+	} else
+		reservation_object_add_excl_fence(bo->resv, &fence->base);
 }
 
 /**
@@ -1475,7 +1471,6 @@ void vmw_resource_move_notify(struct ttm_buffer_object *bo,
 
 	if (mem->mem_type != VMW_PL_MOB) {
 		struct vmw_resource *res, *n;
-		struct ttm_bo_device *bdev = bo->bdev;
 		struct ttm_validate_buffer val_buf;
 
 		val_buf.bo = bo;
@@ -1491,9 +1486,7 @@ void vmw_resource_move_notify(struct ttm_buffer_object *bo,
 			list_del_init(&res->mob_head);
 		}
 
-		spin_lock(&bdev->fence_lock);
 		(void) ttm_bo_wait(bo, false, false, false);
-		spin_unlock(&bdev->fence_lock);
 	}
 }
 
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
index e3d39c80a091..70b44917c368 100644
--- a/include/drm/ttm/ttm_bo_api.h
+++ b/include/drm/ttm/ttm_bo_api.h
@@ -173,7 +173,6 @@ struct ttm_tt;
  * @lru: List head for the lru list.
  * @ddestroy: List head for the delayed destroy list.
  * @swap: List head for swap LRU list.
- * @sync_obj: Pointer to a synchronization object.
  * @priv_flags: Flags describing buffer object internal state.
  * @vma_node: Address space manager node.
  * @offset: The current GPU offset, which can have different meanings
@@ -237,13 +236,9 @@ struct ttm_buffer_object {
 	struct list_head io_reserve_lru;
 
 	/**
-	 * Members protected by struct buffer_object_device::fence_lock
-	 * In addition, setting sync_obj to anything else
-	 * than NULL requires bo::reserved to be held. This allows for
-	 * checking NULL while reserved but not holding the mentioned lock.
+	 * Members protected by a bo reservation.
 	 */
 
-	void *sync_obj;
 	unsigned long priv_flags;
 
 	struct drm_vma_offset_node vma_node;
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index 5c8bb5699a6f..142d752fc450 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -312,11 +312,6 @@ struct ttm_mem_type_manager {
  * @move: Callback for a driver to hook in accelerated functions to
  * move a buffer.
  * If set to NULL, a potentially slow memcpy() move is used.
- * @sync_obj_signaled: See ttm_fence_api.h
- * @sync_obj_wait: See ttm_fence_api.h
- * @sync_obj_flush: See ttm_fence_api.h
- * @sync_obj_unref: See ttm_fence_api.h
- * @sync_obj_ref: See ttm_fence_api.h
  */
 
 struct ttm_bo_driver {
@@ -418,23 +413,6 @@ struct ttm_bo_driver {
 	int (*verify_access) (struct ttm_buffer_object *bo,
 			      struct file *filp);
 
-	/**
-	 * In case a driver writer dislikes the TTM fence objects,
-	 * the driver writer can replace those with sync objects of
-	 * his / her own. If it turns out that no driver writer is
-	 * using these. I suggest we remove these hooks and plug in
-	 * fences directly. The bo driver needs the following functionality:
-	 * See the corresponding functions in the fence object API
-	 * documentation.
-	 */
-
-	bool (*sync_obj_signaled) (void *sync_obj);
-	int (*sync_obj_wait) (void *sync_obj,
-			      bool lazy, bool interruptible);
-	int (*sync_obj_flush) (void *sync_obj);
-	void (*sync_obj_unref) (void **sync_obj);
-	void *(*sync_obj_ref) (void *sync_obj);
-
 	/* hook to notify driver about a driver move so it
 	 * can do tiling things */
 	void (*move_notify)(struct ttm_buffer_object *bo,
@@ -521,8 +499,6 @@ struct ttm_bo_global {
  *
  * @driver: Pointer to a struct ttm_bo_driver struct setup by the driver.
  * @man: An array of mem_type_managers.
- * @fence_lock: Protects the synchronizing members on *all* bos belonging
- * to this device.
  * @vma_manager: Address space manager
  * lru_lock: Spinlock that protects the buffer+device lru lists and
  * ddestroy lists.
@@ -542,7 +518,6 @@ struct ttm_bo_device {
 	struct ttm_bo_global *glob;
 	struct ttm_bo_driver *driver;
 	struct ttm_mem_type_manager man[TTM_NUM_MEM_TYPES];
-	spinlock_t fence_lock;
 
 	/*
 	 * Protected by internal locks.
@@ -1025,7 +1000,7 @@ extern void ttm_bo_free_old_node(struct ttm_buffer_object *bo);
  * ttm_bo_move_accel_cleanup.
  *
  * @bo: A pointer to a struct ttm_buffer_object.
- * @sync_obj: A sync object that signals when moving is complete.
+ * @fence: A fence object that signals when moving is complete.
  * @evict: This is an evict move. Don't return until the buffer is idle.
  * @no_wait_gpu: Return immediately if the GPU is busy.
  * @new_mem: struct ttm_mem_reg indicating where to move.
@@ -1039,7 +1014,7 @@ extern void ttm_bo_free_old_node(struct ttm_buffer_object *bo);
  */
 
 extern int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo,
-				     void *sync_obj,
+				     struct fence *fence,
 				     bool evict, bool no_wait_gpu,
 				     struct ttm_mem_reg *new_mem);
 /**
diff --git a/include/drm/ttm/ttm_execbuf_util.h b/include/drm/ttm/ttm_execbuf_util.h
index 16db7d01a336..ff11a424f752 100644
--- a/include/drm/ttm/ttm_execbuf_util.h
+++ b/include/drm/ttm/ttm_execbuf_util.h
@@ -39,19 +39,11 @@
  *
  * @head:           list head for thread-private list.
  * @bo:             refcounted buffer object pointer.
- * @reserved:       Indicates whether @bo has been reserved for validation.
- * @removed:        Indicates whether @bo has been removed from lru lists.
- * @put_count:      Number of outstanding references on bo::list_kref.
- * @old_sync_obj:   Pointer to a sync object about to be unreferenced
  */
 
 struct ttm_validate_buffer {
 	struct list_head head;
 	struct ttm_buffer_object *bo;
-	bool reserved;
-	bool removed;
-	int put_count;
-	void *old_sync_obj;
 };
 
 /**
@@ -73,6 +65,7 @@ extern void ttm_eu_backoff_reservation(struct ww_acquire_ctx *ticket,
  * @ticket:  [out] ww_acquire_ctx filled in by call, or NULL if only
  *           non-blocking reserves should be tried.
  * @list:    thread private list of ttm_validate_buffer structs.
+ * @intr:    should the wait be interruptible
  *
  * Tries to reserve bos pointed to by the list entries for validation.
  * If the function returns 0, all buffers are marked as "unfenced",
@@ -84,9 +77,9 @@ extern void ttm_eu_backoff_reservation(struct ww_acquire_ctx *ticket,
  * CPU write reservations to be cleared, and for other threads to
  * unreserve their buffers.
  *
- * This function may return -ERESTART or -EAGAIN if the calling process
- * receives a signal while waiting. In that case, no buffers on the list
- * will be reserved upon return.
+ * If intr is set to true, this function may return -ERESTARTSYS if the
+ * calling process receives a signal while waiting. In that case, no
+ * buffers on the list will be reserved upon return.
  *
  * Buffers reserved by this function should be unreserved by
  * a call to either ttm_eu_backoff_reservation() or
@@ -95,14 +88,14 @@ extern void ttm_eu_backoff_reservation(struct ww_acquire_ctx *ticket,
  */
 
 extern int ttm_eu_reserve_buffers(struct ww_acquire_ctx *ticket,
-				  struct list_head *list);
+				  struct list_head *list, bool intr);
 
 /**
  * function ttm_eu_fence_buffer_objects.
  *
  * @ticket:      ww_acquire_ctx from reserve call
  * @list:        thread private list of ttm_validate_buffer structs.
- * @sync_obj:    The new sync object for the buffers.
+ * @fence:       The new exclusive fence for the buffers.
  *
  * This function should be called when command submission is complete, and
  * it will add a new sync object to bos pointed to by entries on @list.
@@ -111,6 +104,7 @@ extern int ttm_eu_reserve_buffers(struct ww_acquire_ctx *ticket,
  */
 
 extern void ttm_eu_fence_buffer_objects(struct ww_acquire_ctx *ticket,
-					struct list_head *list, void *sync_obj);
+					struct list_head *list,
+					struct fence *fence);
 
 #endif