Merge branches 'clockevents/4.4-fixes' and 'clockevents/4.5-fixes' of http://git.linaro.org/people/daniel.lezcano/linux into timers/urgent

Pull in fixes from Daniel Lezcano:

 - Fix the vt8500 timer leading to a system lock up when dealing with too
   small delta (Roman Volkov)

 - Select the CLKSRC_MMIO when the fsl_ftm_timer is enabled with COMPILE_TEST
   (Daniel Lezcano)

 - Prevent to compile timers using the 'iomem' API when the architecture has
   not HAS_IOMEM set (Richard Weinberger)

19 files changed, 379 insertions, 142 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 306f75700bf8..048cfe073dae 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -496,6 +496,7 @@ struct amdgpu_bo_va_mapping {
 
 /* bo virtual addresses in a specific vm */
 struct amdgpu_bo_va {
+	struct mutex		        mutex;
 	/* protected by bo being reserved */
 	struct list_head		bo_list;
 	struct fence		        *last_pt_update;
@@ -538,6 +539,7 @@ struct amdgpu_bo {
 	/* Constant after initialization */
 	struct amdgpu_device		*adev;
 	struct drm_gem_object		gem_base;
+	struct amdgpu_bo		*parent;
 
 	struct ttm_bo_kmap_obj		dma_buf_vmap;
 	pid_t				pid;
@@ -928,8 +930,6 @@ struct amdgpu_vm_id {
 };
 
 struct amdgpu_vm {
-	struct mutex		mutex;
-
 	struct rb_root		va;
 
 	/* protecting invalidated */
@@ -956,6 +956,8 @@ struct amdgpu_vm {
 	struct amdgpu_vm_id	ids[AMDGPU_MAX_RINGS];
 	/* for interval tree */
 	spinlock_t		it_lock;
+	/* protecting freed */
+	spinlock_t		freed_lock;
 };
 
 struct amdgpu_vm_manager {
@@ -1262,7 +1264,8 @@ struct amdgpu_cs_parser {
 	struct ww_acquire_ctx	ticket;
 
 	/* user fence */
-	struct amdgpu_user_fence uf;
+	struct amdgpu_user_fence	uf;
+	struct amdgpu_bo_list_entry	uf_entry;
 };
 
 struct amdgpu_job {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 3afcf0237c25..25a3e2485cc2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -127,6 +127,37 @@ int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type,
 	return 0;
 }
 
+static int amdgpu_cs_user_fence_chunk(struct amdgpu_cs_parser *p,
+				      struct drm_amdgpu_cs_chunk_fence *fence_data)
+{
+	struct drm_gem_object *gobj;
+	uint32_t handle;
+
+	handle = fence_data->handle;
+	gobj = drm_gem_object_lookup(p->adev->ddev, p->filp,
+				     fence_data->handle);
+	if (gobj == NULL)
+		return -EINVAL;
+
+	p->uf.bo = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj));
+	p->uf.offset = fence_data->offset;
+
+	if (amdgpu_ttm_tt_has_userptr(p->uf.bo->tbo.ttm)) {
+		drm_gem_object_unreference_unlocked(gobj);
+		return -EINVAL;
+	}
+
+	p->uf_entry.robj = amdgpu_bo_ref(p->uf.bo);
+	p->uf_entry.prefered_domains = AMDGPU_GEM_DOMAIN_GTT;
+	p->uf_entry.allowed_domains = AMDGPU_GEM_DOMAIN_GTT;
+	p->uf_entry.priority = 0;
+	p->uf_entry.tv.bo = &p->uf_entry.robj->tbo;
+	p->uf_entry.tv.shared = true;
+
+	drm_gem_object_unreference_unlocked(gobj);
+	return 0;
+}
+
 int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 {
 	union drm_amdgpu_cs *cs = data;
@@ -207,26 +238,15 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 
 		case AMDGPU_CHUNK_ID_FENCE:
 			size = sizeof(struct drm_amdgpu_cs_chunk_fence);
-			if (p->chunks[i].length_dw * sizeof(uint32_t) >= size) {
-				uint32_t handle;
-				struct drm_gem_object *gobj;
-				struct drm_amdgpu_cs_chunk_fence *fence_data;
-
-				fence_data = (void *)p->chunks[i].kdata;
-				handle = fence_data->handle;
-				gobj = drm_gem_object_lookup(p->adev->ddev,
-							     p->filp, handle);
-				if (gobj == NULL) {
-					ret = -EINVAL;
-					goto free_partial_kdata;
-				}
-
-				p->uf.bo = gem_to_amdgpu_bo(gobj);
-				p->uf.offset = fence_data->offset;
-			} else {
+			if (p->chunks[i].length_dw * sizeof(uint32_t) < size) {
 				ret = -EINVAL;
 				goto free_partial_kdata;
 			}
+
+			ret = amdgpu_cs_user_fence_chunk(p, (void *)p->chunks[i].kdata);
+			if (ret)
+				goto free_partial_kdata;
+
 			break;
 
 		case AMDGPU_CHUNK_ID_DEPENDENCIES:
@@ -389,6 +409,9 @@ static int amdgpu_cs_parser_relocs(struct amdgpu_cs_parser *p)
 	p->vm_bos = amdgpu_vm_get_bos(p->adev, &fpriv->vm,
 				      &p->validated);
 
+	if (p->uf.bo)
+		list_add(&p->uf_entry.tv.head, &p->validated);
+
 	if (need_mmap_lock)
 		down_read(&current->mm->mmap_sem);
 
@@ -486,8 +509,8 @@ static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bo
 		for (i = 0; i < parser->num_ibs; i++)
 			amdgpu_ib_free(parser->adev, &parser->ibs[i]);
 	kfree(parser->ibs);
-	if (parser->uf.bo)
-		drm_gem_object_unreference_unlocked(&parser->uf.bo->gem_base);
+	amdgpu_bo_unref(&parser->uf.bo);
+	amdgpu_bo_unref(&parser->uf_entry.robj);
 }
 
 static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
@@ -776,7 +799,7 @@ static int amdgpu_cs_free_job(struct amdgpu_job *job)
 			amdgpu_ib_free(job->adev, &job->ibs[i]);
 	kfree(job->ibs);
 	if (job->uf.bo)
-		drm_gem_object_unreference_unlocked(&job->uf.bo->gem_base);
+		amdgpu_bo_unref(&job->uf.bo);
 	return 0;
 }
 
@@ -784,8 +807,6 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 {
 	struct amdgpu_device *adev = dev->dev_private;
 	union drm_amdgpu_cs *cs = data;
-	struct amdgpu_fpriv *fpriv = filp->driver_priv;
-	struct amdgpu_vm *vm = &fpriv->vm;
 	struct amdgpu_cs_parser parser = {};
 	bool reserved_buffers = false;
 	int i, r;
@@ -803,7 +824,6 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 		r = amdgpu_cs_handle_lockup(adev, r);
 		return r;
 	}
-	mutex_lock(&vm->mutex);
 	r = amdgpu_cs_parser_relocs(&parser);
 	if (r == -ENOMEM)
 		DRM_ERROR("Not enough memory for command submission!\n");
@@ -888,7 +908,6 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 
 out:
 	amdgpu_cs_parser_fini(&parser, r, reserved_buffers);
-	mutex_unlock(&vm->mutex);
 	r = amdgpu_cs_handle_lockup(adev, r);
 	return r;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c
index e173a5a02f0d..5580d3420c3a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c
@@ -73,6 +73,8 @@ static void amdgpu_flip_work_func(struct work_struct *__work)
 	struct drm_crtc *crtc = &amdgpuCrtc->base;
 	unsigned long flags;
 	unsigned i;
+	int vpos, hpos, stat, min_udelay;
+	struct drm_vblank_crtc *vblank = &crtc->dev->vblank[work->crtc_id];
 
 	amdgpu_flip_wait_fence(adev, &work->excl);
 	for (i = 0; i < work->shared_count; ++i)
@@ -81,6 +83,41 @@ static void amdgpu_flip_work_func(struct work_struct *__work)
 	/* We borrow the event spin lock for protecting flip_status */
 	spin_lock_irqsave(&crtc->dev->event_lock, flags);
 
+	/* If this happens to execute within the "virtually extended" vblank
+	 * interval before the start of the real vblank interval then it needs
+	 * to delay programming the mmio flip until the real vblank is entered.
+	 * This prevents completing a flip too early due to the way we fudge
+	 * our vblank counter and vblank timestamps in order to work around the
+	 * problem that the hw fires vblank interrupts before actual start of
+	 * vblank (when line buffer refilling is done for a frame). It
+	 * complements the fudging logic in amdgpu_get_crtc_scanoutpos() for
+	 * timestamping and amdgpu_get_vblank_counter_kms() for vblank counts.
+	 *
+	 * In practice this won't execute very often unless on very fast
+	 * machines because the time window for this to happen is very small.
+	 */
+	for (;;) {
+		/* GET_DISTANCE_TO_VBLANKSTART returns distance to real vblank
+		 * start in hpos, and to the "fudged earlier" vblank start in
+		 * vpos.
+		 */
+		stat = amdgpu_get_crtc_scanoutpos(adev->ddev, work->crtc_id,
+						  GET_DISTANCE_TO_VBLANKSTART,
+						  &vpos, &hpos, NULL, NULL,
+						  &crtc->hwmode);
+
+		if ((stat & (DRM_SCANOUTPOS_VALID | DRM_SCANOUTPOS_ACCURATE)) !=
+		    (DRM_SCANOUTPOS_VALID | DRM_SCANOUTPOS_ACCURATE) ||
+		    !(vpos >= 0 && hpos <= 0))
+			break;
+
+		/* Sleep at least until estimated real start of hw vblank */
+		spin_unlock_irqrestore(&crtc->dev->event_lock, flags);
+		min_udelay = (-hpos + 1) * max(vblank->linedur_ns / 1000, 5);
+		usleep_range(min_udelay, 2 * min_udelay);
+		spin_lock_irqsave(&crtc->dev->event_lock, flags);
+	};
+
 	/* do the flip (mmio) */
 	adev->mode_info.funcs->page_flip(adev, work->crtc_id, work->base);
 	/* set the flip status */
@@ -109,7 +146,7 @@ static void amdgpu_unpin_work_func(struct work_struct *__work)
 	} else
 		DRM_ERROR("failed to reserve buffer after flip\n");
 
-	drm_gem_object_unreference_unlocked(&work->old_rbo->gem_base);
+	amdgpu_bo_unref(&work->old_rbo);
 	kfree(work->shared);
 	kfree(work);
 }
@@ -148,8 +185,8 @@ int amdgpu_crtc_page_flip(struct drm_crtc *crtc,
 	obj = old_amdgpu_fb->obj;
 
 	/* take a reference to the old object */
-	drm_gem_object_reference(obj);
 	work->old_rbo = gem_to_amdgpu_bo(obj);
+	amdgpu_bo_ref(work->old_rbo);
 
 	new_amdgpu_fb = to_amdgpu_framebuffer(fb);
 	obj = new_amdgpu_fb->obj;
@@ -222,7 +259,7 @@ pflip_cleanup:
 	amdgpu_bo_unreserve(new_rbo);
 
 cleanup:
-	drm_gem_object_unreference_unlocked(&work->old_rbo->gem_base);
+	amdgpu_bo_unref(&work->old_rbo);
 	fence_put(work->excl);
 	for (i = 0; i < work->shared_count; ++i)
 		fence_put(work->shared[i]);
@@ -712,6 +749,15 @@ bool amdgpu_crtc_scaling_mode_fixup(struct drm_crtc *crtc,
  * \param dev Device to query.
  * \param pipe Crtc to query.
  * \param flags Flags from caller (DRM_CALLED_FROM_VBLIRQ or 0).
+ *              For driver internal use only also supports these flags:
+ *
+ *              USE_REAL_VBLANKSTART to use the real start of vblank instead
+ *              of a fudged earlier start of vblank.
+ *
+ *              GET_DISTANCE_TO_VBLANKSTART to return distance to the
+ *              fudged earlier start of vblank in *vpos and the distance
+ *              to true start of vblank in *hpos.
+ *
  * \param *vpos Location where vertical scanout position should be stored.
  * \param *hpos Location where horizontal scanout position should go.
  * \param *stime Target location for timestamp taken immediately before
@@ -776,10 +822,40 @@ int amdgpu_get_crtc_scanoutpos(struct drm_device *dev, unsigned int pipe,
 		vbl_end = 0;
 	}
 
+	/* Called from driver internal vblank counter query code? */
+	if (flags & GET_DISTANCE_TO_VBLANKSTART) {
+	    /* Caller wants distance from real vbl_start in *hpos */
+	    *hpos = *vpos - vbl_start;
+	}
+
+	/* Fudge vblank to start a few scanlines earlier to handle the
+	 * problem that vblank irqs fire a few scanlines before start
+	 * of vblank. Some driver internal callers need the true vblank
+	 * start to be used and signal this via the USE_REAL_VBLANKSTART flag.
+	 *
+	 * The cause of the "early" vblank irq is that the irq is triggered
+	 * by the line buffer logic when the line buffer read position enters
+	 * the vblank, whereas our crtc scanout position naturally lags the
+	 * line buffer read position.
+	 */
+	if (!(flags & USE_REAL_VBLANKSTART))
+		vbl_start -= adev->mode_info.crtcs[pipe]->lb_vblank_lead_lines;
+
 	/* Test scanout position against vblank region. */
 	if ((*vpos < vbl_start) && (*vpos >= vbl_end))
 		in_vbl = false;
 
+	/* In vblank? */
+	if (in_vbl)
+	    ret |= DRM_SCANOUTPOS_IN_VBLANK;
+
+	/* Called from driver internal vblank counter query code? */
+	if (flags & GET_DISTANCE_TO_VBLANKSTART) {
+		/* Caller wants distance from fudged earlier vbl_start */
+		*vpos -= vbl_start;
+		return ret;
+	}
+
 	/* Check if inside vblank area and apply corrective offsets:
 	 * vpos will then be >=0 in video scanout area, but negative
 	 * within vblank area, counting down the number of lines until
@@ -795,32 +871,6 @@ int amdgpu_get_crtc_scanoutpos(struct drm_device *dev, unsigned int pipe,
 	/* Correct for shifted end of vbl at vbl_end. */
 	*vpos = *vpos - vbl_end;
 
-	/* In vblank? */
-	if (in_vbl)
-		ret |= DRM_SCANOUTPOS_IN_VBLANK;
-
-	/* Is vpos outside nominal vblank area, but less than
-	 * 1/100 of a frame height away from start of vblank?
-	 * If so, assume this isn't a massively delayed vblank
-	 * interrupt, but a vblank interrupt that fired a few
-	 * microseconds before true start of vblank. Compensate
-	 * by adding a full frame duration to the final timestamp.
-	 * Happens, e.g., on ATI R500, R600.
-	 *
-	 * We only do this if DRM_CALLED_FROM_VBLIRQ.
-	 */
-	if ((flags & DRM_CALLED_FROM_VBLIRQ) && !in_vbl) {
-		vbl_start = mode->crtc_vdisplay;
-		vtotal = mode->crtc_vtotal;
-
-		if (vbl_start - *vpos < vtotal / 100) {
-			*vpos -= vtotal;
-
-			/* Signal this correction as "applied". */
-			ret |= 0x8;
-		}
-	}
-
 	return ret;
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index 00c5b580f56c..9c253c535d26 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -115,12 +115,9 @@ int amdgpu_gem_object_open(struct drm_gem_object *obj, struct drm_file *file_pri
 	struct amdgpu_vm *vm = &fpriv->vm;
 	struct amdgpu_bo_va *bo_va;
 	int r;
-	mutex_lock(&vm->mutex);
 	r = amdgpu_bo_reserve(rbo, false);
-	if (r) {
-		mutex_unlock(&vm->mutex);
+	if (r)
 		return r;
-	}
 
 	bo_va = amdgpu_vm_bo_find(vm, rbo);
 	if (!bo_va) {
@@ -129,7 +126,6 @@ int amdgpu_gem_object_open(struct drm_gem_object *obj, struct drm_file *file_pri
 		++bo_va->ref_count;
 	}
 	amdgpu_bo_unreserve(rbo);
-	mutex_unlock(&vm->mutex);
 	return 0;
 }
 
@@ -142,10 +138,8 @@ void amdgpu_gem_object_close(struct drm_gem_object *obj,
 	struct amdgpu_vm *vm = &fpriv->vm;
 	struct amdgpu_bo_va *bo_va;
 	int r;
-	mutex_lock(&vm->mutex);
 	r = amdgpu_bo_reserve(rbo, true);
 	if (r) {
-		mutex_unlock(&vm->mutex);
 		dev_err(adev->dev, "leaking bo va because "
 			"we fail to reserve bo (%d)\n", r);
 		return;
@@ -157,7 +151,6 @@ void amdgpu_gem_object_close(struct drm_gem_object *obj,
 		}
 	}
 	amdgpu_bo_unreserve(rbo);
-	mutex_unlock(&vm->mutex);
 }
 
 static int amdgpu_gem_handle_lockup(struct amdgpu_device *adev, int r)
@@ -242,8 +235,9 @@ int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
 	    AMDGPU_GEM_USERPTR_REGISTER))
 		return -EINVAL;
 
-	if (!(args->flags & AMDGPU_GEM_USERPTR_ANONONLY) ||
-		   !(args->flags & AMDGPU_GEM_USERPTR_REGISTER)) {
+	if (!(args->flags & AMDGPU_GEM_USERPTR_READONLY) && (
+	     !(args->flags & AMDGPU_GEM_USERPTR_ANONONLY) ||
+	     !(args->flags & AMDGPU_GEM_USERPTR_REGISTER))) {
 
 		/* if we want to write to it we must require anonymous
 		   memory and install a MMU notifier */
@@ -483,6 +477,14 @@ static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev,
 		if (domain == AMDGPU_GEM_DOMAIN_CPU)
 			goto error_unreserve;
 	}
+	list_for_each_entry(entry, &duplicates, head) {
+		domain = amdgpu_mem_type_to_domain(entry->bo->mem.mem_type);
+		/* if anything is swapped out don't swap it in here,
+		   just abort and wait for the next CS */
+		if (domain == AMDGPU_GEM_DOMAIN_CPU)
+			goto error_unreserve;
+	}
+
 	r = amdgpu_vm_update_page_directory(adev, bo_va->vm);
 	if (r)
 		goto error_unreserve;
@@ -553,7 +555,6 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 	gobj = drm_gem_object_lookup(dev, filp, args->handle);
 	if (gobj == NULL)
 		return -ENOENT;
-	mutex_lock(&fpriv->vm.mutex);
 	rbo = gem_to_amdgpu_bo(gobj);
 	INIT_LIST_HEAD(&list);
 	INIT_LIST_HEAD(&duplicates);
@@ -568,7 +569,6 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 	}
 	r = ttm_eu_reserve_buffers(&ticket, &list, true, &duplicates);
 	if (r) {
-		mutex_unlock(&fpriv->vm.mutex);
 		drm_gem_object_unreference_unlocked(gobj);
 		return r;
 	}
@@ -577,7 +577,6 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 	if (!bo_va) {
 		ttm_eu_backoff_reservation(&ticket, &list);
 		drm_gem_object_unreference_unlocked(gobj);
-		mutex_unlock(&fpriv->vm.mutex);
 		return -ENOENT;
 	}
 
@@ -602,7 +601,7 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 	ttm_eu_backoff_reservation(&ticket, &list);
 	if (!r && !(args->flags & AMDGPU_VM_DELAY_UPDATE))
 		amdgpu_gem_va_update_vm(adev, bo_va, args->operation);
-	mutex_unlock(&fpriv->vm.mutex);
+
 	drm_gem_object_unreference_unlocked(gobj);
 	return r;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index 1618e2294a16..e23843f4d877 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -611,13 +611,59 @@ void amdgpu_driver_preclose_kms(struct drm_device *dev,
 u32 amdgpu_get_vblank_counter_kms(struct drm_device *dev, unsigned int pipe)
 {
 	struct amdgpu_device *adev = dev->dev_private;
+	int vpos, hpos, stat;
+	u32 count;
 
 	if (pipe >= adev->mode_info.num_crtc) {
 		DRM_ERROR("Invalid crtc %u\n", pipe);
 		return -EINVAL;
 	}
 
-	return amdgpu_display_vblank_get_counter(adev, pipe);
+	/* The hw increments its frame counter at start of vsync, not at start
+	 * of vblank, as is required by DRM core vblank counter handling.
+	 * Cook the hw count here to make it appear to the caller as if it
+	 * incremented at start of vblank. We measure distance to start of
+	 * vblank in vpos. vpos therefore will be >= 0 between start of vblank
+	 * and start of vsync, so vpos >= 0 means to bump the hw frame counter
+	 * result by 1 to give the proper appearance to caller.
+	 */
+	if (adev->mode_info.crtcs[pipe]) {
+		/* Repeat readout if needed to provide stable result if
+		 * we cross start of vsync during the queries.
+		 */
+		do {
+			count = amdgpu_display_vblank_get_counter(adev, pipe);
+			/* Ask amdgpu_get_crtc_scanoutpos to return vpos as
+			 * distance to start of vblank, instead of regular
+			 * vertical scanout pos.
+			 */
+			stat = amdgpu_get_crtc_scanoutpos(
+				dev, pipe, GET_DISTANCE_TO_VBLANKSTART,
+				&vpos, &hpos, NULL, NULL,
+				&adev->mode_info.crtcs[pipe]->base.hwmode);
+		} while (count != amdgpu_display_vblank_get_counter(adev, pipe));
+
+		if (((stat & (DRM_SCANOUTPOS_VALID | DRM_SCANOUTPOS_ACCURATE)) !=
+		    (DRM_SCANOUTPOS_VALID | DRM_SCANOUTPOS_ACCURATE))) {
+			DRM_DEBUG_VBL("Query failed! stat %d\n", stat);
+		} else {
+			DRM_DEBUG_VBL("crtc %d: dist from vblank start %d\n",
+				      pipe, vpos);
+
+			/* Bump counter if we are at >= leading edge of vblank,
+			 * but before vsync where vpos would turn negative and
+			 * the hw counter really increments.
+			 */
+			if (vpos >= 0)
+				count++;
+		}
+	} else {
+		/* Fallback to use value as is. */
+		count = amdgpu_display_vblank_get_counter(adev, pipe);
+		DRM_DEBUG_VBL("NULL mode info! Returned count may be wrong.\n");
+	}
+
+	return count;
 }
 
 /**
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h
index b62c1710cab6..064ebb347074 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h
@@ -407,6 +407,7 @@ struct amdgpu_crtc {
 	u32 line_time;
 	u32 wm_low;
 	u32 wm_high;
+	u32 lb_vblank_lead_lines;
 	struct drm_display_mode hw_mode;
 };
 
@@ -528,6 +529,10 @@ struct amdgpu_framebuffer {
 #define ENCODER_MODE_IS_DP(em) (((em) == ATOM_ENCODER_MODE_DP) || \
 				((em) == ATOM_ENCODER_MODE_DP_MST))
 
+/* Driver internal use only flags of amdgpu_get_crtc_scanoutpos() */
+#define USE_REAL_VBLANKSTART 		(1 << 30)
+#define GET_DISTANCE_TO_VBLANKSTART	(1 << 31)
+
 void amdgpu_link_encoder_connector(struct drm_device *dev);
 
 struct drm_connector *
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 0d524384ff79..c3ce103b6a33 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -100,6 +100,7 @@ static void amdgpu_ttm_bo_destroy(struct ttm_buffer_object *tbo)
 	list_del_init(&bo->list);
 	mutex_unlock(&bo->adev->gem.mutex);
 	drm_gem_object_release(&bo->gem_base);
+	amdgpu_bo_unref(&bo->parent);
 	kfree(bo->metadata);
 	kfree(bo);
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index d4bac5f49939..8a1752ff3d8e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -587,9 +587,13 @@ static int amdgpu_ttm_backend_bind(struct ttm_tt *ttm,
 	uint32_t flags = amdgpu_ttm_tt_pte_flags(gtt->adev, ttm, bo_mem);
 	int r;
 
-	if (gtt->userptr)
-		amdgpu_ttm_tt_pin_userptr(ttm);
-
+	if (gtt->userptr) {
+		r = amdgpu_ttm_tt_pin_userptr(ttm);
+		if (r) {
+			DRM_ERROR("failed to pin userptr\n");
+			return r;
+		}
+	}
 	gtt->offset = (unsigned long)(bo_mem->start << PAGE_SHIFT);
 	if (!ttm->num_pages) {
 		WARN(1, "nothing to bind %lu pages for mreg %p back %p!\n",
@@ -797,11 +801,12 @@ uint32_t amdgpu_ttm_tt_pte_flags(struct amdgpu_device *adev, struct ttm_tt *ttm,
 	if (mem && mem->mem_type != TTM_PL_SYSTEM)
 		flags |= AMDGPU_PTE_VALID;
 
-	if (mem && mem->mem_type == TTM_PL_TT)
+	if (mem && mem->mem_type == TTM_PL_TT) {
 		flags |= AMDGPU_PTE_SYSTEM;
 
-	if (!ttm || ttm->caching_state == tt_cached)
-		flags |= AMDGPU_PTE_SNOOPED;
+		if (ttm->caching_state == tt_cached)
+			flags |= AMDGPU_PTE_SNOOPED;
+	}
 
 	if (adev->asic_type >= CHIP_TOPAZ)
 		flags |= AMDGPU_PTE_EXECUTABLE;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
index 03f0c3bae516..a745eeeb5d82 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
@@ -392,7 +392,10 @@ int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
 	ib->ptr[ib->length_dw++] = 0x00000001; /* session cmd */
 	ib->ptr[ib->length_dw++] = handle;
 
-	ib->ptr[ib->length_dw++] = 0x00000030; /* len */
+	if ((ring->adev->vce.fw_version >> 24) >= 52)
+		ib->ptr[ib->length_dw++] = 0x00000040; /* len */
+	else
+		ib->ptr[ib->length_dw++] = 0x00000030; /* len */
 	ib->ptr[ib->length_dw++] = 0x01000001; /* create cmd */
 	ib->ptr[ib->length_dw++] = 0x00000000;
 	ib->ptr[ib->length_dw++] = 0x00000042;
@@ -404,6 +407,12 @@ int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
 	ib->ptr[ib->length_dw++] = 0x00000100;
 	ib->ptr[ib->length_dw++] = 0x0000000c;
 	ib->ptr[ib->length_dw++] = 0x00000000;
+	if ((ring->adev->vce.fw_version >> 24) >= 52) {
+		ib->ptr[ib->length_dw++] = 0x00000000;
+		ib->ptr[ib->length_dw++] = 0x00000000;
+		ib->ptr[ib->length_dw++] = 0x00000000;
+		ib->ptr[ib->length_dw++] = 0x00000000;
+	}
 
 	ib->ptr[ib->length_dw++] = 0x00000014; /* len */
 	ib->ptr[ib->length_dw++] = 0x05000005; /* feedback buffer */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 159ce54bbd8d..b53d273eb7a1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -885,17 +885,21 @@ int amdgpu_vm_clear_freed(struct amdgpu_device *adev,
 	struct amdgpu_bo_va_mapping *mapping;
 	int r;
 
+	spin_lock(&vm->freed_lock);
 	while (!list_empty(&vm->freed)) {
 		mapping = list_first_entry(&vm->freed,
 			struct amdgpu_bo_va_mapping, list);
 		list_del(&mapping->list);
-
+		spin_unlock(&vm->freed_lock);
 		r = amdgpu_vm_bo_update_mapping(adev, vm, mapping, 0, 0, NULL);
 		kfree(mapping);
 		if (r)
 			return r;
 
+		spin_lock(&vm->freed_lock);
 	}
+	spin_unlock(&vm->freed_lock);
+
 	return 0;
 
 }
@@ -922,8 +926,9 @@ int amdgpu_vm_clear_invalids(struct amdgpu_device *adev,
 		bo_va = list_first_entry(&vm->invalidated,
 			struct amdgpu_bo_va, vm_status);
 		spin_unlock(&vm->status_lock);
-
+		mutex_lock(&bo_va->mutex);
 		r = amdgpu_vm_bo_update(adev, bo_va, NULL);
+		mutex_unlock(&bo_va->mutex);
 		if (r)
 			return r;
 
@@ -967,7 +972,7 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,
 	INIT_LIST_HEAD(&bo_va->valids);
 	INIT_LIST_HEAD(&bo_va->invalids);
 	INIT_LIST_HEAD(&bo_va->vm_status);
-
+	mutex_init(&bo_va->mutex);
 	list_add_tail(&bo_va->bo_list, &bo->va);
 
 	return bo_va;
@@ -1045,7 +1050,9 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 	mapping->offset = offset;
 	mapping->flags = flags;
 
+	mutex_lock(&bo_va->mutex);
 	list_add(&mapping->list, &bo_va->invalids);
+	mutex_unlock(&bo_va->mutex);
 	spin_lock(&vm->it_lock);
 	interval_tree_insert(&mapping->it, &vm->va);
 	spin_unlock(&vm->it_lock);
@@ -1076,6 +1083,11 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 		if (r)
 			goto error_free;
 
+		/* Keep a reference to the page table to avoid freeing
+		 * them up in the wrong order.
+		 */
+		pt->parent = amdgpu_bo_ref(vm->page_directory);
+
 		r = amdgpu_vm_clear_bo(adev, pt);
 		if (r) {
 			amdgpu_bo_unref(&pt);
@@ -1121,7 +1133,7 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
 	bool valid = true;
 
 	saddr /= AMDGPU_GPU_PAGE_SIZE;
-
+	mutex_lock(&bo_va->mutex);
 	list_for_each_entry(mapping, &bo_va->valids, list) {
 		if (mapping->it.start == saddr)
 			break;
@@ -1135,20 +1147,25 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
 				break;
 		}
 
-		if (&mapping->list == &bo_va->invalids)
+		if (&mapping->list == &bo_va->invalids) {
+			mutex_unlock(&bo_va->mutex);
 			return -ENOENT;
+		}
 	}
-
+	mutex_unlock(&bo_va->mutex);
 	list_del(&mapping->list);
 	spin_lock(&vm->it_lock);
 	interval_tree_remove(&mapping->it, &vm->va);
 	spin_unlock(&vm->it_lock);
 	trace_amdgpu_vm_bo_unmap(bo_va, mapping);
 
-	if (valid)
+	if (valid) {
+		spin_lock(&vm->freed_lock);
 		list_add(&mapping->list, &vm->freed);
-	else
+		spin_unlock(&vm->freed_lock);
+	} else {
 		kfree(mapping);
+	}
 
 	return 0;
 }
@@ -1181,7 +1198,9 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
 		interval_tree_remove(&mapping->it, &vm->va);
 		spin_unlock(&vm->it_lock);
 		trace_amdgpu_vm_bo_unmap(bo_va, mapping);
+		spin_lock(&vm->freed_lock);
 		list_add(&mapping->list, &vm->freed);
+		spin_unlock(&vm->freed_lock);
 	}
 	list_for_each_entry_safe(mapping, next, &bo_va->invalids, list) {
 		list_del(&mapping->list);
@@ -1190,8 +1209,8 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
 		spin_unlock(&vm->it_lock);
 		kfree(mapping);
 	}
-
 	fence_put(bo_va->last_pt_update);
+	mutex_destroy(&bo_va->mutex);
 	kfree(bo_va);
 }
 
@@ -1236,13 +1255,13 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 		vm->ids[i].id = 0;
 		vm->ids[i].flushed_updates = NULL;
 	}
-	mutex_init(&vm->mutex);
 	vm->va = RB_ROOT;
 	spin_lock_init(&vm->status_lock);
 	INIT_LIST_HEAD(&vm->invalidated);
 	INIT_LIST_HEAD(&vm->cleared);
 	INIT_LIST_HEAD(&vm->freed);
 	spin_lock_init(&vm->it_lock);
+	spin_lock_init(&vm->freed_lock);
 	pd_size = amdgpu_vm_directory_size(adev);
 	pd_entries = amdgpu_vm_num_pdes(adev);
 
@@ -1320,7 +1339,6 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 		fence_put(vm->ids[i].flushed_updates);
 	}
 
-	mutex_destroy(&vm->mutex);
 }
 
 /**
diff --git a/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c b/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c
index cb0f7747e3dc..4dcc8fba5792 100644
--- a/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c
@@ -1250,7 +1250,7 @@ static void dce_v10_0_program_watermarks(struct amdgpu_device *adev,
 	u32 pixel_period;
 	u32 line_time = 0;
 	u32 latency_watermark_a = 0, latency_watermark_b = 0;
-	u32 tmp, wm_mask;
+	u32 tmp, wm_mask, lb_vblank_lead_lines = 0;
 
 	if (amdgpu_crtc->base.enabled && num_heads && mode) {
 		pixel_period = 1000000 / (u32)mode->clock;
@@ -1333,6 +1333,7 @@ static void dce_v10_0_program_watermarks(struct amdgpu_device *adev,
 		    (adev->mode_info.disp_priority == 2)) {
 			DRM_DEBUG_KMS("force priority to high\n");
 		}
+		lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode->crtc_hdisplay);
 	}
 
 	/* select wm A */
@@ -1357,6 +1358,8 @@ static void dce_v10_0_program_watermarks(struct amdgpu_device *adev,
 	amdgpu_crtc->line_time = line_time;
 	amdgpu_crtc->wm_high = latency_watermark_a;
 	amdgpu_crtc->wm_low = latency_watermark_b;
+	/* Save number of lines the linebuffer leads before the scanout */
+	amdgpu_crtc->lb_vblank_lead_lines = lb_vblank_lead_lines;
 }
 
 /**
diff --git a/drivers/gpu/drm/amd/amdgpu/dce_v11_0.c b/drivers/gpu/drm/amd/amdgpu/dce_v11_0.c
index 5af3721851d6..8f1e51128b33 100644
--- a/drivers/gpu/drm/amd/amdgpu/dce_v11_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/dce_v11_0.c
@@ -1238,7 +1238,7 @@ static void dce_v11_0_program_watermarks(struct amdgpu_device *adev,
 	u32 pixel_period;
 	u32 line_time = 0;
 	u32 latency_watermark_a = 0, latency_watermark_b = 0;
-	u32 tmp, wm_mask;
+	u32 tmp, wm_mask, lb_vblank_lead_lines = 0;
 
 	if (amdgpu_crtc->base.enabled && num_heads && mode) {
 		pixel_period = 1000000 / (u32)mode->clock;
@@ -1321,6 +1321,7 @@ static void dce_v11_0_program_watermarks(struct amdgpu_device *adev,
 		    (adev->mode_info.disp_priority == 2)) {
 			DRM_DEBUG_KMS("force priority to high\n");
 		}
+		lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode->crtc_hdisplay);
 	}
 
 	/* select wm A */
@@ -1345,6 +1346,8 @@ static void dce_v11_0_program_watermarks(struct amdgpu_device *adev,
 	amdgpu_crtc->line_time = line_time;
 	amdgpu_crtc->wm_high = latency_watermark_a;
 	amdgpu_crtc->wm_low = latency_watermark_b;
+	/* Save number of lines the linebuffer leads before the scanout */
+	amdgpu_crtc->lb_vblank_lead_lines = lb_vblank_lead_lines;
 }
 
 /**
diff --git a/drivers/gpu/drm/amd/amdgpu/dce_v8_0.c b/drivers/gpu/drm/amd/amdgpu/dce_v8_0.c
index 4f7b49a6dc50..42d954dc436d 100644
--- a/drivers/gpu/drm/amd/amdgpu/dce_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/dce_v8_0.c
@@ -1193,7 +1193,7 @@ static void dce_v8_0_program_watermarks(struct amdgpu_device *adev,
 	u32 pixel_period;
 	u32 line_time = 0;
 	u32 latency_watermark_a = 0, latency_watermark_b = 0;
-	u32 tmp, wm_mask;
+	u32 tmp, wm_mask, lb_vblank_lead_lines = 0;
 
 	if (amdgpu_crtc->base.enabled && num_heads && mode) {
 		pixel_period = 1000000 / (u32)mode->clock;
@@ -1276,6 +1276,7 @@ static void dce_v8_0_program_watermarks(struct amdgpu_device *adev,
 		    (adev->mode_info.disp_priority == 2)) {
 			DRM_DEBUG_KMS("force priority to high\n");
 		}
+		lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode->crtc_hdisplay);
 	}
 
 	/* select wm A */
@@ -1302,6 +1303,8 @@ static void dce_v8_0_program_watermarks(struct amdgpu_device *adev,
 	amdgpu_crtc->line_time = line_time;
 	amdgpu_crtc->wm_high = latency_watermark_a;
 	amdgpu_crtc->wm_low = latency_watermark_b;
+	/* Save number of lines the linebuffer leads before the scanout */
+	amdgpu_crtc->lb_vblank_lead_lines = lb_vblank_lead_lines;
 }
 
 /**
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
index 7427d8cd4c43..ed8abb58a785 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
@@ -513,7 +513,7 @@ static int gmc_v7_0_gart_enable(struct amdgpu_device *adev)
 	WREG32(mmVM_L2_CNTL3, tmp);
 	/* setup context0 */
 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->mc.gtt_start >> 12);
-	WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, (adev->mc.gtt_end >> 12) - 1);
+	WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->mc.gtt_end >> 12);
 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, adev->gart.table_addr >> 12);
 	WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
 			(u32)(adev->dummy_page.addr >> 12));
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
index cb0e50ebb528..d39028440814 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
@@ -657,7 +657,7 @@ static int gmc_v8_0_gart_enable(struct amdgpu_device *adev)
 	WREG32(mmVM_L2_CNTL4, tmp);
 	/* setup context0 */
 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->mc.gtt_start >> 12);
-	WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, (adev->mc.gtt_end >> 12) - 1);
+	WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->mc.gtt_end >> 12);
 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, adev->gart.table_addr >> 12);
 	WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
 			(u32)(adev->dummy_page.addr >> 12));
diff --git a/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c b/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c
index 6a52db6ad8d7..370c6c9d81c2 100644
--- a/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c
@@ -40,6 +40,9 @@
 
 #define GRBM_GFX_INDEX__VCE_INSTANCE__SHIFT	0x04
 #define GRBM_GFX_INDEX__VCE_INSTANCE_MASK	0x10
+#define mmVCE_LMI_VCPU_CACHE_40BIT_BAR0 	0x8616
+#define mmVCE_LMI_VCPU_CACHE_40BIT_BAR1 	0x8617
+#define mmVCE_LMI_VCPU_CACHE_40BIT_BAR2 	0x8618
 
 #define VCE_V3_0_FW_SIZE	(384 * 1024)
 #define VCE_V3_0_STACK_SIZE	(64 * 1024)
@@ -130,9 +133,11 @@ static int vce_v3_0_start(struct amdgpu_device *adev)
 
 		/* set BUSY flag */
 		WREG32_P(mmVCE_STATUS, 1, ~1);
-
-		WREG32_P(mmVCE_VCPU_CNTL, VCE_VCPU_CNTL__CLK_EN_MASK,
-			~VCE_VCPU_CNTL__CLK_EN_MASK);
+		if (adev->asic_type >= CHIP_STONEY)
+			WREG32_P(mmVCE_VCPU_CNTL, 1, ~0x200001);
+		else
+			WREG32_P(mmVCE_VCPU_CNTL, VCE_VCPU_CNTL__CLK_EN_MASK,
+				~VCE_VCPU_CNTL__CLK_EN_MASK);
 
 		WREG32_P(mmVCE_SOFT_RESET,
 			 VCE_SOFT_RESET__ECPU_SOFT_RESET_MASK,
@@ -391,8 +396,12 @@ static void vce_v3_0_mc_resume(struct amdgpu_device *adev, int idx)
 	WREG32(mmVCE_LMI_SWAP_CNTL, 0);
 	WREG32(mmVCE_LMI_SWAP_CNTL1, 0);
 	WREG32(mmVCE_LMI_VM_CTRL, 0);
-
-	WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR, (adev->vce.gpu_addr >> 8));
+	if (adev->asic_type >= CHIP_STONEY) {
+		WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR0, (adev->vce.gpu_addr >> 8));
+		WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR1, (adev->vce.gpu_addr >> 8));
+		WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR2, (adev->vce.gpu_addr >> 8));
+	} else
+		WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR, (adev->vce.gpu_addr >> 8));
 	offset = AMDGPU_VCE_FIRMWARE_OFFSET;
 	size = VCE_V3_0_FW_SIZE;
 	WREG32(mmVCE_VCPU_CACHE_OFFSET0, offset & 0x7fffffff);
@@ -576,6 +585,11 @@ static int vce_v3_0_process_interrupt(struct amdgpu_device *adev,
 				      struct amdgpu_iv_entry *entry)
 {
 	DRM_DEBUG("IH: VCE\n");
+
+	WREG32_P(mmVCE_SYS_INT_STATUS,
+		VCE_SYS_INT_STATUS__VCE_SYS_INT_TRAP_INTERRUPT_INT_MASK,
+		~VCE_SYS_INT_STATUS__VCE_SYS_INT_TRAP_INTERRUPT_INT_MASK);
+
 	switch (entry->src_data) {
 	case 0:
 		amdgpu_fence_process(&adev->vce.ring[0]);
diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
index ea30d6ad4c13..3a4820e863ec 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
+++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
@@ -30,8 +30,7 @@
 #define CREATE_TRACE_POINTS
 #include "gpu_sched_trace.h"
 
-static struct amd_sched_job *
-amd_sched_entity_pop_job(struct amd_sched_entity *entity);
+static bool amd_sched_entity_is_ready(struct amd_sched_entity *entity);
 static void amd_sched_wakeup(struct amd_gpu_scheduler *sched);
 
 struct kmem_cache *sched_fence_slab;
@@ -64,36 +63,36 @@ static void amd_sched_rq_remove_entity(struct amd_sched_rq *rq,
 }
 
 /**
- * Select next job from a specified run queue with round robin policy.
- * Return NULL if nothing available.
+ * Select an entity which could provide a job to run
+ *
+ * @rq		The run queue to check.
+ *
+ * Try to find a ready entity, returns NULL if none found.
  */
-static struct amd_sched_job *
-amd_sched_rq_select_job(struct amd_sched_rq *rq)
+static struct amd_sched_entity *
+amd_sched_rq_select_entity(struct amd_sched_rq *rq)
 {
 	struct amd_sched_entity *entity;
-	struct amd_sched_job *sched_job;
 
 	spin_lock(&rq->lock);
 
 	entity = rq->current_entity;
 	if (entity) {
 		list_for_each_entry_continue(entity, &rq->entities, list) {
-			sched_job = amd_sched_entity_pop_job(entity);
-			if (sched_job) {
+			if (amd_sched_entity_is_ready(entity)) {
 				rq->current_entity = entity;
 				spin_unlock(&rq->lock);
-				return sched_job;
+				return entity;
 			}
 		}
 	}
 
 	list_for_each_entry(entity, &rq->entities, list) {
 
-		sched_job = amd_sched_entity_pop_job(entity);
-		if (sched_job) {
+		if (amd_sched_entity_is_ready(entity)) {
 			rq->current_entity = entity;
 			spin_unlock(&rq->lock);
-			return sched_job;
+			return entity;
 		}
 
 		if (entity == rq->current_entity)
@@ -177,6 +176,24 @@ static bool amd_sched_entity_is_idle(struct amd_sched_entity *entity)
 }
 
 /**
+ * Check if entity is ready
+ *
+ * @entity	The pointer to a valid scheduler entity
+ *
+ * Return true if entity could provide a job.
+ */
+static bool amd_sched_entity_is_ready(struct amd_sched_entity *entity)
+{
+	if (kfifo_is_empty(&entity->job_queue))
+		return false;
+
+	if (ACCESS_ONCE(entity->dependency))
+		return false;
+
+	return true;
+}
+
+/**
  * Destroy a context entity
  *
  * @sched       Pointer to scheduler instance
@@ -211,32 +228,53 @@ static void amd_sched_entity_wakeup(struct fence *f, struct fence_cb *cb)
 	amd_sched_wakeup(entity->sched);
 }
 
+static bool amd_sched_entity_add_dependency_cb(struct amd_sched_entity *entity)
+{
+	struct amd_gpu_scheduler *sched = entity->sched;
+	struct fence * fence = entity->dependency;
+	struct amd_sched_fence *s_fence;
+
+	if (fence->context == entity->fence_context) {
+		/* We can ignore fences from ourself */
+		fence_put(entity->dependency);
+		return false;
+	}
+
+	s_fence = to_amd_sched_fence(fence);
+	if (s_fence && s_fence->sched == sched) {
+		/* Fence is from the same scheduler */
+		if (test_bit(AMD_SCHED_FENCE_SCHEDULED_BIT, &fence->flags)) {
+			/* Ignore it when it is already scheduled */
+			fence_put(entity->dependency);
+			return false;
+		}
+
+		/* Wait for fence to be scheduled */
+		entity->cb.func = amd_sched_entity_wakeup;
+		list_add_tail(&entity->cb.node, &s_fence->scheduled_cb);
+		return true;
+	}
+
+	if (!fence_add_callback(entity->dependency, &entity->cb,
+				amd_sched_entity_wakeup))
+		return true;
+
+	fence_put(entity->dependency);
+	return false;
+}
+
 static struct amd_sched_job *
 amd_sched_entity_pop_job(struct amd_sched_entity *entity)
 {
 	struct amd_gpu_scheduler *sched = entity->sched;
 	struct amd_sched_job *sched_job;
 
-	if (ACCESS_ONCE(entity->dependency))
-		return NULL;
-
 	if (!kfifo_out_peek(&entity->job_queue, &sched_job, sizeof(sched_job)))
 		return NULL;
 
-	while ((entity->dependency = sched->ops->dependency(sched_job))) {
-
-		if (entity->dependency->context == entity->fence_context) {
-			/* We can ignore fences from ourself */
-			fence_put(entity->dependency);
-			continue;
-		}
-
-		if (fence_add_callback(entity->dependency, &entity->cb,
-				       amd_sched_entity_wakeup))
-			fence_put(entity->dependency);
-		else
+	while ((entity->dependency = sched->ops->dependency(sched_job)))
+		if (amd_sched_entity_add_dependency_cb(entity))
 			return NULL;
-	}
 
 	return sched_job;
 }
@@ -250,6 +288,7 @@ amd_sched_entity_pop_job(struct amd_sched_entity *entity)
  */
 static bool amd_sched_entity_in(struct amd_sched_job *sched_job)
 {
+	struct amd_gpu_scheduler *sched = sched_job->sched;
 	struct amd_sched_entity *entity = sched_job->s_entity;
 	bool added, first = false;
 
@@ -264,7 +303,7 @@ static bool amd_sched_entity_in(struct amd_sched_job *sched_job)
 
 	/* first job wakes up scheduler */
 	if (first)
-		amd_sched_wakeup(sched_job->sched);
+		amd_sched_wakeup(sched);
 
 	return added;
 }
@@ -280,9 +319,9 @@ void amd_sched_entity_push_job(struct amd_sched_job *sched_job)
 {
 	struct amd_sched_entity *entity = sched_job->s_entity;
 
+	trace_amd_sched_job(sched_job);
 	wait_event(entity->sched->job_scheduled,
 		   amd_sched_entity_in(sched_job));
-	trace_amd_sched_job(sched_job);
 }
 
 /**
@@ -304,22 +343,22 @@ static void amd_sched_wakeup(struct amd_gpu_scheduler *sched)
 }
 
 /**
- * Select next to run
+ * Select next entity to process
 */
-static struct amd_sched_job *
-amd_sched_select_job(struct amd_gpu_scheduler *sched)
+static struct amd_sched_entity *
+amd_sched_select_entity(struct amd_gpu_scheduler *sched)
 {
-	struct amd_sched_job *sched_job;
+	struct amd_sched_entity *entity;
 
 	if (!amd_sched_ready(sched))
 		return NULL;
 
 	/* Kernel run queue has higher priority than normal run queue*/
-	sched_job = amd_sched_rq_select_job(&sched->kernel_rq);
-	if (sched_job == NULL)
-		sched_job = amd_sched_rq_select_job(&sched->sched_rq);
+	entity = amd_sched_rq_select_entity(&sched->kernel_rq);
+	if (entity == NULL)
+		entity = amd_sched_rq_select_entity(&sched->sched_rq);
 
-	return sched_job;
+	return entity;
 }
 
 static void amd_sched_process_job(struct fence *f, struct fence_cb *cb)
@@ -381,13 +420,16 @@ static int amd_sched_main(void *param)
 		unsigned long flags;
 
 		wait_event_interruptible(sched->wake_up_worker,
-			kthread_should_stop() ||
-			(sched_job = amd_sched_select_job(sched)));
+			(entity = amd_sched_select_entity(sched)) ||
+			kthread_should_stop());
 
+		if (!entity)
+			continue;
+
+		sched_job = amd_sched_entity_pop_job(entity);
 		if (!sched_job)
 			continue;
 
-		entity = sched_job->s_entity;
 		s_fence = sched_job->s_fence;
 
 		if (sched->timeout != MAX_SCHEDULE_TIMEOUT) {
@@ -400,6 +442,7 @@ static int amd_sched_main(void *param)
 
 		atomic_inc(&sched->hw_rq_count);
 		fence = sched->ops->run_job(sched_job);
+		amd_sched_fence_scheduled(s_fence);
 		if (fence) {
 			r = fence_add_callback(fence, &s_fence->cb,
 					       amd_sched_process_job);
diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
index 939692b14f4b..a0f0ae53aacd 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
+++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
@@ -27,6 +27,8 @@
 #include <linux/kfifo.h>
 #include <linux/fence.h>
 
+#define AMD_SCHED_FENCE_SCHEDULED_BIT	FENCE_FLAG_USER_BITS
+
 struct amd_gpu_scheduler;
 struct amd_sched_rq;
 
@@ -68,6 +70,7 @@ struct amd_sched_rq {
 struct amd_sched_fence {
 	struct fence                    base;
 	struct fence_cb                 cb;
+	struct list_head		scheduled_cb;
 	struct amd_gpu_scheduler	*sched;
 	spinlock_t			lock;
 	void                            *owner;
@@ -134,7 +137,7 @@ void amd_sched_entity_push_job(struct amd_sched_job *sched_job);
 
 struct amd_sched_fence *amd_sched_fence_create(
 	struct amd_sched_entity *s_entity, void *owner);
+void amd_sched_fence_scheduled(struct amd_sched_fence *fence);
 void amd_sched_fence_signal(struct amd_sched_fence *fence);
 
-
 #endif
diff --git a/drivers/gpu/drm/amd/scheduler/sched_fence.c b/drivers/gpu/drm/amd/scheduler/sched_fence.c
index 8d2130b9ff05..87c78eecea64 100644
--- a/drivers/gpu/drm/amd/scheduler/sched_fence.c
+++ b/drivers/gpu/drm/amd/scheduler/sched_fence.c
@@ -35,6 +35,8 @@ struct amd_sched_fence *amd_sched_fence_create(struct amd_sched_entity *s_entity
 	fence = kmem_cache_zalloc(sched_fence_slab, GFP_KERNEL);
 	if (fence == NULL)
 		return NULL;
+
+	INIT_LIST_HEAD(&fence->scheduled_cb);
 	fence->owner = owner;
 	fence->sched = s_entity->sched;
 	spin_lock_init(&fence->lock);
@@ -55,6 +57,17 @@ void amd_sched_fence_signal(struct amd_sched_fence *fence)
 		FENCE_TRACE(&fence->base, "was already signaled\n");
 }
 
+void amd_sched_fence_scheduled(struct amd_sched_fence *s_fence)
+{
+	struct fence_cb *cur, *tmp;
+
+	set_bit(AMD_SCHED_FENCE_SCHEDULED_BIT, &s_fence->base.flags);
+	list_for_each_entry_safe(cur, tmp, &s_fence->scheduled_cb, node) {
+		list_del_init(&cur->node);
+		cur->func(&s_fence->base, cur);
+	}
+}
+
 static const char *amd_sched_fence_get_driver_name(struct fence *fence)
 {
 	return "amd_sched";