1 files changed, 327 insertions, 132 deletions

diff --git a/drivers/gpu/drm/radeon/radeon_object.c b/drivers/gpu/drm/radeon/radeon_object.c
index 8eb18346601..6c717b257d6 100644
--- a/drivers/gpu/drm/radeon/radeon_object.c
+++ b/drivers/gpu/drm/radeon/radeon_object.c
@@ -32,8 +32,9 @@
 #include <linux/list.h>
 #include <linux/slab.h>
 #include <drm/drmP.h>
-#include "radeon_drm.h"
+#include <drm/radeon_drm.h>
 #include "radeon.h"
+#include "radeon_trace.h"
 
 
 int radeon_ttm_init(struct radeon_device *rdev);
@@ -45,15 +46,52 @@ static void radeon_bo_clear_surface_reg(struct radeon_bo *bo);
  * function are calling it.
  */
 
+static void radeon_bo_clear_va(struct radeon_bo *bo)
+{
+	struct radeon_bo_va *bo_va, *tmp;
+
+	list_for_each_entry_safe(bo_va, tmp, &bo->va, bo_list) {
+		/* remove from all vm address space */
+		radeon_vm_bo_rmv(bo->rdev, bo_va);
+	}
+}
+
+static void radeon_update_memory_usage(struct radeon_bo *bo,
+				       unsigned mem_type, int sign)
+{
+	struct radeon_device *rdev = bo->rdev;
+	u64 size = (u64)bo->tbo.num_pages << PAGE_SHIFT;
+
+	switch (mem_type) {
+	case TTM_PL_TT:
+		if (sign > 0)
+			atomic64_add(size, &rdev->gtt_usage);
+		else
+			atomic64_sub(size, &rdev->gtt_usage);
+		break;
+	case TTM_PL_VRAM:
+		if (sign > 0)
+			atomic64_add(size, &rdev->vram_usage);
+		else
+			atomic64_sub(size, &rdev->vram_usage);
+		break;
+	}
+}
+
 static void radeon_ttm_bo_destroy(struct ttm_buffer_object *tbo)
 {
 	struct radeon_bo *bo;
 
 	bo = container_of(tbo, struct radeon_bo, tbo);
+
+	radeon_update_memory_usage(bo, bo->tbo.mem.mem_type, -1);
+
 	mutex_lock(&bo->rdev->gem.mutex);
 	list_del_init(&bo->list);
 	mutex_unlock(&bo->rdev->gem.mutex);
 	radeon_bo_clear_surface_reg(bo);
+	radeon_bo_clear_va(bo);
+	drm_gem_object_release(&bo->gem_base);
 	kfree(bo);
 }
 
@@ -66,76 +104,99 @@ bool radeon_ttm_bo_is_radeon_bo(struct ttm_buffer_object *bo)
 
 void radeon_ttm_placement_from_domain(struct radeon_bo *rbo, u32 domain)
 {
-	u32 c = 0;
+	u32 c = 0, i;
 
 	rbo->placement.fpfn = 0;
-	rbo->placement.lpfn = rbo->rdev->mc.active_vram_size >> PAGE_SHIFT;
+	rbo->placement.lpfn = 0;
 	rbo->placement.placement = rbo->placements;
 	rbo->placement.busy_placement = rbo->placements;
 	if (domain & RADEON_GEM_DOMAIN_VRAM)
 		rbo->placements[c++] = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED |
 					TTM_PL_FLAG_VRAM;
-	if (domain & RADEON_GEM_DOMAIN_GTT)
-		rbo->placements[c++] = TTM_PL_MASK_CACHING | TTM_PL_FLAG_TT;
-	if (domain & RADEON_GEM_DOMAIN_CPU)
-		rbo->placements[c++] = TTM_PL_MASK_CACHING | TTM_PL_FLAG_SYSTEM;
+	if (domain & RADEON_GEM_DOMAIN_GTT) {
+		if (rbo->rdev->flags & RADEON_IS_AGP) {
+			rbo->placements[c++] = TTM_PL_FLAG_WC | TTM_PL_FLAG_TT;
+		} else {
+			rbo->placements[c++] = TTM_PL_FLAG_CACHED | TTM_PL_FLAG_TT;
+		}
+	}
+	if (domain & RADEON_GEM_DOMAIN_CPU) {
+		if (rbo->rdev->flags & RADEON_IS_AGP) {
+			rbo->placements[c++] = TTM_PL_FLAG_WC | TTM_PL_FLAG_SYSTEM;
+		} else {
+			rbo->placements[c++] = TTM_PL_FLAG_CACHED | TTM_PL_FLAG_SYSTEM;
+		}
+	}
 	if (!c)
 		rbo->placements[c++] = TTM_PL_MASK_CACHING | TTM_PL_FLAG_SYSTEM;
 	rbo->placement.num_placement = c;
 	rbo->placement.num_busy_placement = c;
+
+	/*
+	 * Use two-ended allocation depending on the buffer size to
+	 * improve fragmentation quality.
+	 * 512kb was measured as the most optimal number.
+	 */
+	if (rbo->tbo.mem.size > 512 * 1024) {
+		for (i = 0; i < c; i++) {
+			rbo->placements[i] |= TTM_PL_FLAG_TOPDOWN;
+		}
+	}
 }
 
-int radeon_bo_create(struct radeon_device *rdev, struct drm_gem_object *gobj,
-			unsigned long size, bool kernel, u32 domain,
-			struct radeon_bo **bo_ptr)
+int radeon_bo_create(struct radeon_device *rdev,
+		     unsigned long size, int byte_align, bool kernel, u32 domain,
+		     struct sg_table *sg, struct radeon_bo **bo_ptr)
 {
 	struct radeon_bo *bo;
 	enum ttm_bo_type type;
+	unsigned long page_align = roundup(byte_align, PAGE_SIZE) >> PAGE_SHIFT;
+	size_t acc_size;
 	int r;
 
-	if (unlikely(rdev->mman.bdev.dev_mapping == NULL)) {
-		rdev->mman.bdev.dev_mapping = rdev->ddev->dev_mapping;
-	}
+	size = ALIGN(size, PAGE_SIZE);
+
 	if (kernel) {
 		type = ttm_bo_type_kernel;
+	} else if (sg) {
+		type = ttm_bo_type_sg;
 	} else {
 		type = ttm_bo_type_device;
 	}
 	*bo_ptr = NULL;
 
-retry:
+	acc_size = ttm_bo_dma_acc_size(&rdev->mman.bdev, size,
+				       sizeof(struct radeon_bo));
+
 	bo = kzalloc(sizeof(struct radeon_bo), GFP_KERNEL);
 	if (bo == NULL)
 		return -ENOMEM;
+	r = drm_gem_object_init(rdev->ddev, &bo->gem_base, size);
+	if (unlikely(r)) {
+		kfree(bo);
+		return r;
+	}
 	bo->rdev = rdev;
-	bo->gobj = gobj;
 	bo->surface_reg = -1;
 	INIT_LIST_HEAD(&bo->list);
+	INIT_LIST_HEAD(&bo->va);
+	bo->initial_domain = domain & (RADEON_GEM_DOMAIN_VRAM |
+	                               RADEON_GEM_DOMAIN_GTT |
+	                               RADEON_GEM_DOMAIN_CPU);
 	radeon_ttm_placement_from_domain(bo, domain);
 	/* Kernel allocation are uninterruptible */
-	mutex_lock(&rdev->vram_mutex);
+	down_read(&rdev->pm.mclk_lock);
 	r = ttm_bo_init(&rdev->mman.bdev, &bo->tbo, size, type,
-			&bo->placement, 0, 0, !kernel, NULL, size,
-			&radeon_ttm_bo_destroy);
-	mutex_unlock(&rdev->vram_mutex);
+			&bo->placement, page_align, !kernel, NULL,
+			acc_size, sg, &radeon_ttm_bo_destroy);
+	up_read(&rdev->pm.mclk_lock);
 	if (unlikely(r != 0)) {
-		if (r != -ERESTARTSYS) {
-			if (domain == RADEON_GEM_DOMAIN_VRAM) {
-				domain |= RADEON_GEM_DOMAIN_GTT;
-				goto retry;
-			}
-			dev_err(rdev->dev,
-				"object_init failed for (%lu, 0x%08X)\n",
-				size, domain);
-		}
 		return r;
 	}
 	*bo_ptr = bo;
-	if (gobj) {
-		mutex_lock(&bo->rdev->gem.mutex);
-		list_add_tail(&bo->list, &rdev->gem.objects);
-		mutex_unlock(&bo->rdev->gem.mutex);
-	}
+
+	trace_radeon_bo_create(bo);
+
 	return 0;
 }
 
@@ -180,14 +241,15 @@ void radeon_bo_unref(struct radeon_bo **bo)
 		return;
 	rdev = (*bo)->rdev;
 	tbo = &((*bo)->tbo);
-	mutex_lock(&rdev->vram_mutex);
+	down_read(&rdev->pm.mclk_lock);
 	ttm_bo_unref(&tbo);
-	mutex_unlock(&rdev->vram_mutex);
+	up_read(&rdev->pm.mclk_lock);
 	if (tbo == NULL)
 		*bo = NULL;
 }
 
-int radeon_bo_pin(struct radeon_bo *bo, u32 domain, u64 *gpu_addr)
+int radeon_bo_pin_restricted(struct radeon_bo *bo, u32 domain, u64 max_offset,
+			     u64 *gpu_addr)
 {
 	int r, i;
 
@@ -195,6 +257,18 @@ int radeon_bo_pin(struct radeon_bo *bo, u32 domain, u64 *gpu_addr)
 		bo->pin_count++;
 		if (gpu_addr)
 			*gpu_addr = radeon_bo_gpu_offset(bo);
+
+		if (max_offset != 0) {
+			u64 domain_start;
+
+			if (domain == RADEON_GEM_DOMAIN_VRAM)
+				domain_start = bo->rdev->mc.vram_start;
+			else
+				domain_start = bo->rdev->mc.gtt_start;
+			WARN_ON_ONCE(max_offset <
+				     (radeon_bo_gpu_offset(bo) - domain_start));
+		}
+
 		return 0;
 	}
 	radeon_ttm_placement_from_domain(bo, domain);
@@ -202,9 +276,18 @@ int radeon_bo_pin(struct radeon_bo *bo, u32 domain, u64 *gpu_addr)
 		/* force to pin into visible video ram */
 		bo->placement.lpfn = bo->rdev->mc.visible_vram_size >> PAGE_SHIFT;
 	}
+	if (max_offset) {
+		u64 lpfn = max_offset >> PAGE_SHIFT;
+
+		if (!bo->placement.lpfn)
+			bo->placement.lpfn = bo->rdev->mc.gtt_size >> PAGE_SHIFT;
+
+		if (lpfn < bo->placement.lpfn)
+			bo->placement.lpfn = lpfn;
+	}
 	for (i = 0; i < bo->placement.num_placement; i++)
 		bo->placements[i] |= TTM_PL_FLAG_NO_EVICT;
-	r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false, false);
+	r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
 	if (likely(r == 0)) {
 		bo->pin_count = 1;
 		if (gpu_addr != NULL)
@@ -215,6 +298,11 @@ int radeon_bo_pin(struct radeon_bo *bo, u32 domain, u64 *gpu_addr)
 	return r;
 }
 
+int radeon_bo_pin(struct radeon_bo *bo, u32 domain, u64 *gpu_addr)
+{
+	return radeon_bo_pin_restricted(bo, domain, 0, gpu_addr);
+}
+
 int radeon_bo_unpin(struct radeon_bo *bo)
 {
 	int r, i;
@@ -228,7 +316,7 @@ int radeon_bo_unpin(struct radeon_bo *bo)
 		return 0;
 	for (i = 0; i < bo->placement.num_placement; i++)
 		bo->placements[i] &= ~TTM_PL_FLAG_NO_EVICT;
-	r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false, false);
+	r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
 	if (unlikely(r != 0))
 		dev_err(bo->rdev->dev, "%p validate failed for unpin\n", bo);
 	return r;
@@ -248,7 +336,6 @@ int radeon_bo_evict_vram(struct radeon_device *rdev)
 void radeon_bo_force_delete(struct radeon_device *rdev)
 {
 	struct radeon_bo *bo, *n;
-	struct drm_gem_object *gobj;
 
 	if (list_empty(&rdev->gem.objects)) {
 		return;
@@ -256,16 +343,14 @@ void radeon_bo_force_delete(struct radeon_device *rdev)
 	dev_err(rdev->dev, "Userspace still has active objects !\n");
 	list_for_each_entry_safe(bo, n, &rdev->gem.objects, list) {
 		mutex_lock(&rdev->ddev->struct_mutex);
-		gobj = bo->gobj;
 		dev_err(rdev->dev, "%p %p %lu %lu force free\n",
-			gobj, bo, (unsigned long)gobj->size,
-			*((unsigned long *)&gobj->refcount));
+			&bo->gem_base, bo, (unsigned long)bo->gem_base.size,
+			*((unsigned long *)&bo->gem_base.refcount));
 		mutex_lock(&bo->rdev->gem.mutex);
 		list_del_init(&bo->list);
 		mutex_unlock(&bo->rdev->gem.mutex);
-		radeon_bo_unref(&bo);
-		gobj->driver_private = NULL;
-		drm_gem_object_unreference(gobj);
+		/* this should unref the ttm bo */
+		drm_gem_object_unreference(&bo->gem_base);
 		mutex_unlock(&rdev->ddev->struct_mutex);
 	}
 }
@@ -273,8 +358,10 @@ void radeon_bo_force_delete(struct radeon_device *rdev)
 int radeon_bo_init(struct radeon_device *rdev)
 {
 	/* Add an MTRR for the VRAM */
-	rdev->mc.vram_mtrr = mtrr_add(rdev->mc.aper_base, rdev->mc.aper_size,
-			MTRR_TYPE_WRCOMB, 1);
+	if (!rdev->fastfb_working) {
+		rdev->mc.vram_mtrr = arch_phys_wc_add(rdev->mc.aper_base,
+						      rdev->mc.aper_size);
+	}
 	DRM_INFO("Detected VRAM RAM=%lluM, BAR=%lluM\n",
 		rdev->mc.mc_vram_size >> 20,
 		(unsigned long long)rdev->mc.aper_size >> 20);
@@ -286,71 +373,115 @@ int radeon_bo_init(struct radeon_device *rdev)
 void radeon_bo_fini(struct radeon_device *rdev)
 {
 	radeon_ttm_fini(rdev);
+	arch_phys_wc_del(rdev->mc.vram_mtrr);
 }
 
-void radeon_bo_list_add_object(struct radeon_bo_list *lobj,
-				struct list_head *head)
-{
-	if (lobj->wdomain) {
-		list_add(&lobj->list, head);
-	} else {
-		list_add_tail(&lobj->list, head);
-	}
-}
-
-int radeon_bo_list_reserve(struct list_head *head)
-{
-	struct radeon_bo_list *lobj;
-	int r;
-
-	list_for_each_entry(lobj, head, list){
-		r = radeon_bo_reserve(lobj->bo, false);
-		if (unlikely(r != 0))
-			return r;
-		lobj->reserved = true;
-	}
-	return 0;
-}
-
-void radeon_bo_list_unreserve(struct list_head *head)
+/* Returns how many bytes TTM can move per IB.
+ */
+static u64 radeon_bo_get_threshold_for_moves(struct radeon_device *rdev)
 {
-	struct radeon_bo_list *lobj;
-
-	list_for_each_entry(lobj, head, list) {
-		/* only unreserve object we successfully reserved */
-		if (lobj->reserved && radeon_bo_is_reserved(lobj->bo))
-			radeon_bo_unreserve(lobj->bo);
-	}
+	u64 real_vram_size = rdev->mc.real_vram_size;
+	u64 vram_usage = atomic64_read(&rdev->vram_usage);
+
+	/* This function is based on the current VRAM usage.
+	 *
+	 * - If all of VRAM is free, allow relocating the number of bytes that
+	 *   is equal to 1/4 of the size of VRAM for this IB.
+
+	 * - If more than one half of VRAM is occupied, only allow relocating
+	 *   1 MB of data for this IB.
+	 *
+	 * - From 0 to one half of used VRAM, the threshold decreases
+	 *   linearly.
+	 *         __________________
+	 * 1/4 of -|\               |
+	 * VRAM    | \              |
+	 *         |  \             |
+	 *         |   \            |
+	 *         |    \           |
+	 *         |     \          |
+	 *         |      \         |
+	 *         |       \________|1 MB
+	 *         |----------------|
+	 *    VRAM 0 %             100 %
+	 *         used            used
+	 *
+	 * Note: It's a threshold, not a limit. The threshold must be crossed
+	 * for buffer relocations to stop, so any buffer of an arbitrary size
+	 * can be moved as long as the threshold isn't crossed before
+	 * the relocation takes place. We don't want to disable buffer
+	 * relocations completely.
+	 *
+	 * The idea is that buffers should be placed in VRAM at creation time
+	 * and TTM should only do a minimum number of relocations during
+	 * command submission. In practice, you need to submit at least
+	 * a dozen IBs to move all buffers to VRAM if they are in GTT.
+	 *
+	 * Also, things can get pretty crazy under memory pressure and actual
+	 * VRAM usage can change a lot, so playing safe even at 50% does
+	 * consistently increase performance.
+	 */
+
+	u64 half_vram = real_vram_size >> 1;
+	u64 half_free_vram = vram_usage >= half_vram ? 0 : half_vram - vram_usage;
+	u64 bytes_moved_threshold = half_free_vram >> 1;
+	return max(bytes_moved_threshold, 1024*1024ull);
 }
 
-int radeon_bo_list_validate(struct list_head *head)
+int radeon_bo_list_validate(struct radeon_device *rdev,
+			    struct ww_acquire_ctx *ticket,
+			    struct list_head *head, int ring)
 {
-	struct radeon_bo_list *lobj;
+	struct radeon_cs_reloc *lobj;
 	struct radeon_bo *bo;
-	u32 domain;
 	int r;
+	u64 bytes_moved = 0, initial_bytes_moved;
+	u64 bytes_moved_threshold = radeon_bo_get_threshold_for_moves(rdev);
 
-	list_for_each_entry(lobj, head, list) {
-		lobj->reserved = false;
-	}
-	r = radeon_bo_list_reserve(head);
+	r = ttm_eu_reserve_buffers(ticket, head);
 	if (unlikely(r != 0)) {
 		return r;
 	}
-	list_for_each_entry(lobj, head, list) {
-		bo = lobj->bo;
+
+	list_for_each_entry(lobj, head, tv.head) {
+		bo = lobj->robj;
 		if (!bo->pin_count) {
-			domain = lobj->wdomain ? lobj->wdomain : lobj->rdomain;
-			
+			u32 domain = lobj->prefered_domains;
+			u32 current_domain =
+				radeon_mem_type_to_domain(bo->tbo.mem.mem_type);
+
+			/* Check if this buffer will be moved and don't move it
+			 * if we have moved too many buffers for this IB already.
+			 *
+			 * Note that this allows moving at least one buffer of
+			 * any size, because it doesn't take the current "bo"
+			 * into account. We don't want to disallow buffer moves
+			 * completely.
+			 */
+			if ((lobj->allowed_domains & current_domain) != 0 &&
+			    (domain & current_domain) == 0 && /* will be moved */
+			    bytes_moved > bytes_moved_threshold) {
+				/* don't move it */
+				domain = current_domain;
+			}
+
 		retry:
 			radeon_ttm_placement_from_domain(bo, domain);
-			r = ttm_bo_validate(&bo->tbo, &bo->placement,
-						true, false, false);
+			if (ring == R600_RING_TYPE_UVD_INDEX)
+				radeon_uvd_force_into_uvd_segment(bo);
+
+			initial_bytes_moved = atomic64_read(&rdev->num_bytes_moved);
+			r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false);
+			bytes_moved += atomic64_read(&rdev->num_bytes_moved) -
+				       initial_bytes_moved;
+
 			if (unlikely(r)) {
-				if (r != -ERESTARTSYS && domain == RADEON_GEM_DOMAIN_VRAM) {
-					domain |= RADEON_GEM_DOMAIN_GTT;
+				if (r != -ERESTARTSYS &&
+				    domain != lobj->allowed_domains) {
+					domain = lobj->allowed_domains;
 					goto retry;
 				}
+				ttm_eu_backoff_reservation(ticket, head);
 				return r;
 			}
 		}
@@ -360,25 +491,6 @@ int radeon_bo_list_validate(struct list_head *head)
 	return 0;
 }
 
-void radeon_bo_list_fence(struct list_head *head, void *fence)
-{
-	struct radeon_bo_list *lobj;
-	struct radeon_bo *bo;
-	struct radeon_fence *old_fence = NULL;
-
-	list_for_each_entry(lobj, head, list) {
-		bo = lobj->bo;
-		spin_lock(&bo->tbo.lock);
-		old_fence = (struct radeon_fence *)bo->tbo.sync_obj;
-		bo->tbo.sync_obj = radeon_fence_ref(fence);
-		bo->tbo.sync_obj_arg = NULL;
-		spin_unlock(&bo->tbo.lock);
-		if (old_fence) {
-			radeon_fence_unref(&old_fence);
-		}
-	}
-}
-
 int radeon_bo_fbdev_mmap(struct radeon_bo *bo,
 			     struct vm_area_struct *vma)
 {
@@ -393,7 +505,7 @@ int radeon_bo_get_surface_reg(struct radeon_bo *bo)
 	int steal;
 	int i;
 
-	BUG_ON(!atomic_read(&bo->tbo.reserved));
+	lockdep_assert_held(&bo->tbo.resv->lock.base);
 
 	if (!bo->tiling_flags)
 		return 0;
@@ -458,8 +570,54 @@ static void radeon_bo_clear_surface_reg(struct radeon_bo *bo)
 int radeon_bo_set_tiling_flags(struct radeon_bo *bo,
 				uint32_t tiling_flags, uint32_t pitch)
 {
+	struct radeon_device *rdev = bo->rdev;
 	int r;
 
+	if (rdev->family >= CHIP_CEDAR) {
+		unsigned bankw, bankh, mtaspect, tilesplit, stilesplit;
+
+		bankw = (tiling_flags >> RADEON_TILING_EG_BANKW_SHIFT) & RADEON_TILING_EG_BANKW_MASK;
+		bankh = (tiling_flags >> RADEON_TILING_EG_BANKH_SHIFT) & RADEON_TILING_EG_BANKH_MASK;
+		mtaspect = (tiling_flags >> RADEON_TILING_EG_MACRO_TILE_ASPECT_SHIFT) & RADEON_TILING_EG_MACRO_TILE_ASPECT_MASK;
+		tilesplit = (tiling_flags >> RADEON_TILING_EG_TILE_SPLIT_SHIFT) & RADEON_TILING_EG_TILE_SPLIT_MASK;
+		stilesplit = (tiling_flags >> RADEON_TILING_EG_STENCIL_TILE_SPLIT_SHIFT) & RADEON_TILING_EG_STENCIL_TILE_SPLIT_MASK;
+		switch (bankw) {
+		case 0:
+		case 1:
+		case 2:
+		case 4:
+		case 8:
+			break;
+		default:
+			return -EINVAL;
+		}
+		switch (bankh) {
+		case 0:
+		case 1:
+		case 2:
+		case 4:
+		case 8:
+			break;
+		default:
+			return -EINVAL;
+		}
+		switch (mtaspect) {
+		case 0:
+		case 1:
+		case 2:
+		case 4:
+		case 8:
+			break;
+		default:
+			return -EINVAL;
+		}
+		if (tilesplit > 6) {
+			return -EINVAL;
+		}
+		if (stilesplit > 6) {
+			return -EINVAL;
+		}
+	}
 	r = radeon_bo_reserve(bo, false);
 	if (unlikely(r != 0))
 		return r;
@@ -473,7 +631,8 @@ void radeon_bo_get_tiling_flags(struct radeon_bo *bo,
 				uint32_t *tiling_flags,
 				uint32_t *pitch)
 {
-	BUG_ON(!atomic_read(&bo->tbo.reserved));
+	lockdep_assert_held(&bo->tbo.resv->lock.base);
+
 	if (tiling_flags)
 		*tiling_flags = bo->tiling_flags;
 	if (pitch)
@@ -483,7 +642,8 @@ void radeon_bo_get_tiling_flags(struct radeon_bo *bo,
 int radeon_bo_check_tiling(struct radeon_bo *bo, bool has_moved,
 				bool force_drop)
 {
-	BUG_ON(!atomic_read(&bo->tbo.reserved));
+	if (!force_drop)
+		lockdep_assert_held(&bo->tbo.resv->lock.base);
 
 	if (!(bo->tiling_flags & RADEON_TILING_SURFACE))
 		return 0;
@@ -509,13 +669,23 @@ int radeon_bo_check_tiling(struct radeon_bo *bo, bool has_moved,
 }
 
 void radeon_bo_move_notify(struct ttm_buffer_object *bo,
-			   struct ttm_mem_reg *mem)
+			   struct ttm_mem_reg *new_mem)
 {
 	struct radeon_bo *rbo;
+
 	if (!radeon_ttm_bo_is_radeon_bo(bo))
 		return;
+
 	rbo = container_of(bo, struct radeon_bo, tbo);
 	radeon_bo_check_tiling(rbo, 0, 1);
+	radeon_vm_bo_invalidate(rbo->rdev, rbo);
+
+	/* update statistics */
+	if (!new_mem)
+		return;
+
+	radeon_update_memory_usage(rbo, bo->mem.mem_type, -1);
+	radeon_update_memory_usage(rbo, new_mem->mem_type, 1);
 }
 
 int radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
@@ -530,21 +700,46 @@ int radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
 	rbo = container_of(bo, struct radeon_bo, tbo);
 	radeon_bo_check_tiling(rbo, 0, 0);
 	rdev = rbo->rdev;
-	if (bo->mem.mem_type == TTM_PL_VRAM) {
-		size = bo->mem.num_pages << PAGE_SHIFT;
-		offset = bo->mem.start << PAGE_SHIFT;
-		if ((offset + size) > rdev->mc.visible_vram_size) {
-			/* hurrah the memory is not visible ! */
-			radeon_ttm_placement_from_domain(rbo, RADEON_GEM_DOMAIN_VRAM);
-			rbo->placement.lpfn = rdev->mc.visible_vram_size >> PAGE_SHIFT;
-			r = ttm_bo_validate(bo, &rbo->placement, false, true, false);
-			if (unlikely(r != 0))
-				return r;
-			offset = bo->mem.start << PAGE_SHIFT;
-			/* this should not happen */
-			if ((offset + size) > rdev->mc.visible_vram_size)
-				return -EINVAL;
-		}
+	if (bo->mem.mem_type != TTM_PL_VRAM)
+		return 0;
+
+	size = bo->mem.num_pages << PAGE_SHIFT;
+	offset = bo->mem.start << PAGE_SHIFT;
+	if ((offset + size) <= rdev->mc.visible_vram_size)
+		return 0;
+
+	/* hurrah the memory is not visible ! */
+	radeon_ttm_placement_from_domain(rbo, RADEON_GEM_DOMAIN_VRAM);
+	rbo->placement.lpfn = rdev->mc.visible_vram_size >> PAGE_SHIFT;
+	r = ttm_bo_validate(bo, &rbo->placement, false, false);
+	if (unlikely(r == -ENOMEM)) {
+		radeon_ttm_placement_from_domain(rbo, RADEON_GEM_DOMAIN_GTT);
+		return ttm_bo_validate(bo, &rbo->placement, false, false);
+	} else if (unlikely(r != 0)) {
+		return r;
 	}
+
+	offset = bo->mem.start << PAGE_SHIFT;
+	/* this should never happen */
+	if ((offset + size) > rdev->mc.visible_vram_size)
+		return -EINVAL;
+
 	return 0;
 }
+
+int radeon_bo_wait(struct radeon_bo *bo, u32 *mem_type, bool no_wait)
+{
+	int r;
+
+	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);
+	ttm_bo_unreserve(&bo->tbo);
+	return r;
+}