Merge branch 'drm-next-3.15' of git://people.freedesktop.org/~deathsimple/linux into drm-next

this is the second pull request for 3.15 radeon changes. Highlights this time:
- Better VRAM usage
- VM page table rework
- Enabling different UVD clocks again
- Some general cleanups and improvements

* 'drm-next-3.15' of git://people.freedesktop.org/~deathsimple/linux:
  drm/radeon: remove struct radeon_bo_list
  drm/radeon: drop non blocking allocations from sub allocator
  drm/radeon: remove global vm lock
  drm/radeon: use normal BOs for the page tables v4
  drm/radeon: further cleanup vm flushing & fencing
  drm/radeon: separate gart and vm functions
  drm/radeon: fix VCE suspend/resume
  drm/radeon: fix missing bo reservation
  drm/radeon: limit how much memory TTM can move per IB according to VRAM usage
  drm/radeon: validate relocations in the order determined by userspace v3
  drm/radeon: add buffers to the LRU list from smallest to largest
  drm/radeon: deduplicate code in radeon_gem_busy_ioctl
  drm/radeon: track memory statistics about VRAM and GTT usage and buffer moves v2
  drm/radeon: add a way to get and set initial buffer domains v2
  drm/radeon: use variable UVD clocks
  drm/radeon: cleanup the fence ring locking code
  drm/radeon: improve ring lockup detection code v2

drivers/gpu/drm/radeon/Makefile

@@ -80,7 +80,7 @@ radeon-y += radeon_device.o radeon_asic.o radeon_kms.o \
 	r600_dpm.o rs780_dpm.o rv6xx_dpm.o rv770_dpm.o rv730_dpm.o rv740_dpm.o \
 	rv770_smc.o cypress_dpm.o btc_dpm.o sumo_dpm.o sumo_smc.o trinity_dpm.o \
 	trinity_smc.o ni_dpm.o si_smc.o si_dpm.o kv_smc.o kv_dpm.o ci_smc.o \
-	ci_dpm.o dce6_afmt.o
+	ci_dpm.o dce6_afmt.o radeon_vm.o
 
 # add async DMA block
 radeon-y += \

drivers/gpu/drm/radeon/r100.c

@@ -1274,12 +1274,12 @@ int r100_reloc_pitch_offset(struct radeon_cs_parser *p,
 
 	value = radeon_get_ib_value(p, idx);
 	tmp = value & 0x003fffff;
-	tmp += (((u32)reloc->lobj.gpu_offset) >> 10);
+	tmp += (((u32)reloc->gpu_offset) >> 10);
 
 	if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
-		if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
+		if (reloc->tiling_flags & RADEON_TILING_MACRO)
 			tile_flags |= RADEON_DST_TILE_MACRO;
-		if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO) {
+		if (reloc->tiling_flags & RADEON_TILING_MICRO) {
 			if (reg == RADEON_SRC_PITCH_OFFSET) {
 				DRM_ERROR("Cannot src blit from microtiled surface\n");
 				radeon_cs_dump_packet(p, pkt);
@@ -1325,7 +1325,7 @@ int r100_packet3_load_vbpntr(struct radeon_cs_parser *p,
 			return r;
 		}
 		idx_value = radeon_get_ib_value(p, idx);
-		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->lobj.gpu_offset);
+		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->gpu_offset);
 
 		track->arrays[i + 0].esize = idx_value >> 8;
 		track->arrays[i + 0].robj = reloc->robj;
@@ -1337,7 +1337,7 @@ int r100_packet3_load_vbpntr(struct radeon_cs_parser *p,
 			radeon_cs_dump_packet(p, pkt);
 			return r;
 		}
-		ib[idx+2] = radeon_get_ib_value(p, idx + 2) + ((u32)reloc->lobj.gpu_offset);
+		ib[idx+2] = radeon_get_ib_value(p, idx + 2) + ((u32)reloc->gpu_offset);
 		track->arrays[i + 1].robj = reloc->robj;
 		track->arrays[i + 1].esize = idx_value >> 24;
 		track->arrays[i + 1].esize &= 0x7F;
@@ -1351,7 +1351,7 @@ int r100_packet3_load_vbpntr(struct radeon_cs_parser *p,
 			return r;
 		}
 		idx_value = radeon_get_ib_value(p, idx);
-		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->lobj.gpu_offset);
+		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->gpu_offset);
 		track->arrays[i + 0].robj = reloc->robj;
 		track->arrays[i + 0].esize = idx_value >> 8;
 		track->arrays[i + 0].esize &= 0x7F;
@@ -1594,7 +1594,7 @@ static int r100_packet0_check(struct radeon_cs_parser *p,
 		track->zb.robj = reloc->robj;
 		track->zb.offset = idx_value;
 		track->zb_dirty = true;
-		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
+		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
 		break;
 	case RADEON_RB3D_COLOROFFSET:
 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
@@ -1607,7 +1607,7 @@ static int r100_packet0_check(struct radeon_cs_parser *p,
 		track->cb[0].robj = reloc->robj;
 		track->cb[0].offset = idx_value;
 		track->cb_dirty = true;
-		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
+		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
 		break;
 	case RADEON_PP_TXOFFSET_0:
 	case RADEON_PP_TXOFFSET_1:
@@ -1621,16 +1621,16 @@ static int r100_packet0_check(struct radeon_cs_parser *p,
 			return r;
 		}
 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
-			if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
+			if (reloc->tiling_flags & RADEON_TILING_MACRO)
 				tile_flags |= RADEON_TXO_MACRO_TILE;
-			if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
+			if (reloc->tiling_flags & RADEON_TILING_MICRO)
 				tile_flags |= RADEON_TXO_MICRO_TILE_X2;
 
 			tmp = idx_value & ~(0x7 << 2);
 			tmp |= tile_flags;
-			ib[idx] = tmp + ((u32)reloc->lobj.gpu_offset);
+			ib[idx] = tmp + ((u32)reloc->gpu_offset);
 		} else
-			ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
+			ib[idx] = idx_value + ((u32)reloc->gpu_offset);
 		track->textures[i].robj = reloc->robj;
 		track->tex_dirty = true;
 		break;
@@ -1648,7 +1648,7 @@ static int r100_packet0_check(struct radeon_cs_parser *p,
 			return r;
 		}
 		track->textures[0].cube_info[i].offset = idx_value;
-		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
+		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
 		track->textures[0].cube_info[i].robj = reloc->robj;
 		track->tex_dirty = true;
 		break;
@@ -1666,7 +1666,7 @@ static int r100_packet0_check(struct radeon_cs_parser *p,
 			return r;
 		}
 		track->textures[1].cube_info[i].offset = idx_value;
-		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
+		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
 		track->textures[1].cube_info[i].robj = reloc->robj;
 		track->tex_dirty = true;
 		break;
@@ -1684,7 +1684,7 @@ static int r100_packet0_check(struct radeon_cs_parser *p,
 			return r;
 		}
 		track->textures[2].cube_info[i].offset = idx_value;
-		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
+		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
 		track->textures[2].cube_info[i].robj = reloc->robj;
 		track->tex_dirty = true;
 		break;
@@ -1702,9 +1702,9 @@ static int r100_packet0_check(struct radeon_cs_parser *p,
 			return r;
 		}
 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
-			if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
+			if (reloc->tiling_flags & RADEON_TILING_MACRO)
 				tile_flags |= RADEON_COLOR_TILE_ENABLE;
-			if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
+			if (reloc->tiling_flags & RADEON_TILING_MICRO)
 				tile_flags |= RADEON_COLOR_MICROTILE_ENABLE;
 
 			tmp = idx_value & ~(0x7 << 16);
@@ -1772,7 +1772,7 @@ static int r100_packet0_check(struct radeon_cs_parser *p,
 			radeon_cs_dump_packet(p, pkt);
 			return r;
 		}
-		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
+		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
 		break;
 	case RADEON_PP_CNTL:
 		{
@@ -1932,7 +1932,7 @@ static int r100_packet3_check(struct radeon_cs_parser *p,
 			radeon_cs_dump_packet(p, pkt);
 			return r;
 		}
-		ib[idx+1] = radeon_get_ib_value(p, idx+1) + ((u32)reloc->lobj.gpu_offset);
+		ib[idx+1] = radeon_get_ib_value(p, idx+1) + ((u32)reloc->gpu_offset);
 		r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
 		if (r) {
 			return r;
@@ -1946,7 +1946,7 @@ static int r100_packet3_check(struct radeon_cs_parser *p,
 			radeon_cs_dump_packet(p, pkt);
 			return r;
 		}
-		ib[idx] = radeon_get_ib_value(p, idx) + ((u32)reloc->lobj.gpu_offset);
+		ib[idx] = radeon_get_ib_value(p, idx) + ((u32)reloc->gpu_offset);
 		track->num_arrays = 1;
 		track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 2));
 

drivers/gpu/drm/radeon/r200.c

@@ -185,7 +185,7 @@ int r200_packet0_check(struct radeon_cs_parser *p,
 		track->zb.robj = reloc->robj;
 		track->zb.offset = idx_value;
 		track->zb_dirty = true;
-		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
+		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
 		break;
 	case RADEON_RB3D_COLOROFFSET:
 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
@@ -198,7 +198,7 @@ int r200_packet0_check(struct radeon_cs_parser *p,
 		track->cb[0].robj = reloc->robj;
 		track->cb[0].offset = idx_value;
 		track->cb_dirty = true;
-		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
+		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
 		break;
 	case R200_PP_TXOFFSET_0:
 	case R200_PP_TXOFFSET_1:
@@ -215,16 +215,16 @@ int r200_packet0_check(struct radeon_cs_parser *p,
 			return r;
 		}
 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
-			if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
+			if (reloc->tiling_flags & RADEON_TILING_MACRO)
 				tile_flags |= R200_TXO_MACRO_TILE;
-			if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
+			if (reloc->tiling_flags & RADEON_TILING_MICRO)
 				tile_flags |= R200_TXO_MICRO_TILE;
 
 			tmp = idx_value & ~(0x7 << 2);
 			tmp |= tile_flags;
-			ib[idx] = tmp + ((u32)reloc->lobj.gpu_offset);
+			ib[idx] = tmp + ((u32)reloc->gpu_offset);
 		} else
-			ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
+			ib[idx] = idx_value + ((u32)reloc->gpu_offset);
 		track->textures[i].robj = reloc->robj;
 		track->tex_dirty = true;
 		break;
@@ -268,7 +268,7 @@ int r200_packet0_check(struct radeon_cs_parser *p,
 			return r;
 		}
 		track->textures[i].cube_info[face - 1].offset = idx_value;
-		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
+		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
 		track->textures[i].cube_info[face - 1].robj = reloc->robj;
 		track->tex_dirty = true;
 		break;
@@ -287,9 +287,9 @@ int r200_packet0_check(struct radeon_cs_parser *p,
 		}
 
 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
-			if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
+			if (reloc->tiling_flags & RADEON_TILING_MACRO)
 				tile_flags |= RADEON_COLOR_TILE_ENABLE;
-			if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
+			if (reloc->tiling_flags & RADEON_TILING_MICRO)
 				tile_flags |= RADEON_COLOR_MICROTILE_ENABLE;
 
 			tmp = idx_value & ~(0x7 << 16);
@@ -362,7 +362,7 @@ int r200_packet0_check(struct radeon_cs_parser *p,
 			radeon_cs_dump_packet(p, pkt);
 			return r;
 		}
-		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
+		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
 		break;
 	case RADEON_PP_CNTL:
 		{

drivers/gpu/drm/radeon/r300.c

@@ -640,7 +640,7 @@ static int r300_packet0_check(struct radeon_cs_parser *p,
 		track->cb[i].robj = reloc->robj;
 		track->cb[i].offset = idx_value;
 		track->cb_dirty = true;
-		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
+		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
 		break;
 	case R300_ZB_DEPTHOFFSET:
 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
@@ -653,7 +653,7 @@ static int r300_packet0_check(struct radeon_cs_parser *p,
 		track->zb.robj = reloc->robj;
 		track->zb.offset = idx_value;
 		track->zb_dirty = true;
-		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
+		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
 		break;
 	case R300_TX_OFFSET_0:
 	case R300_TX_OFFSET_0+4:
@@ -682,16 +682,16 @@ static int r300_packet0_check(struct radeon_cs_parser *p,
 
 		if (p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS) {
 			ib[idx] = (idx_value & 31) | /* keep the 1st 5 bits */
-				  ((idx_value & ~31) + (u32)reloc->lobj.gpu_offset);
+				  ((idx_value & ~31) + (u32)reloc->gpu_offset);
 		} else {
-			if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
+			if (reloc->tiling_flags & RADEON_TILING_MACRO)
 				tile_flags |= R300_TXO_MACRO_TILE;
-			if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
+			if (reloc->tiling_flags & RADEON_TILING_MICRO)
 				tile_flags |= R300_TXO_MICRO_TILE;
-			else if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO_SQUARE)
+			else if (reloc->tiling_flags & RADEON_TILING_MICRO_SQUARE)
 				tile_flags |= R300_TXO_MICRO_TILE_SQUARE;
 
-			tmp = idx_value + ((u32)reloc->lobj.gpu_offset);
+			tmp = idx_value + ((u32)reloc->gpu_offset);
 			tmp |= tile_flags;
 			ib[idx] = tmp;
 		}
@@ -753,11 +753,11 @@ static int r300_packet0_check(struct radeon_cs_parser *p,
 				return r;
 			}
 
-			if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
+			if (reloc->tiling_flags & RADEON_TILING_MACRO)
 				tile_flags |= R300_COLOR_TILE_ENABLE;
-			if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
+			if (reloc->tiling_flags & RADEON_TILING_MICRO)
 				tile_flags |= R300_COLOR_MICROTILE_ENABLE;
-			else if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO_SQUARE)
+			else if (reloc->tiling_flags & RADEON_TILING_MICRO_SQUARE)
 				tile_flags |= R300_COLOR_MICROTILE_SQUARE_ENABLE;
 
 			tmp = idx_value & ~(0x7 << 16);
@@ -838,11 +838,11 @@ static int r300_packet0_check(struct radeon_cs_parser *p,
 				return r;
 			}
 
-			if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
+			if (reloc->tiling_flags & RADEON_TILING_MACRO)
 				tile_flags |= R300_DEPTHMACROTILE_ENABLE;
-			if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
+			if (reloc->tiling_flags & RADEON_TILING_MICRO)
 				tile_flags |= R300_DEPTHMICROTILE_TILED;
-			else if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO_SQUARE)
+			else if (reloc->tiling_flags & RADEON_TILING_MICRO_SQUARE)
 				tile_flags |= R300_DEPTHMICROTILE_TILED_SQUARE;
 
 			tmp = idx_value & ~(0x7 << 16);
@@ -1052,7 +1052,7 @@ static int r300_packet0_check(struct radeon_cs_parser *p,
 			radeon_cs_dump_packet(p, pkt);
 			return r;
 		}
-		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
+		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
 		break;
 	case 0x4e0c:
 		/* RB3D_COLOR_CHANNEL_MASK */
@@ -1097,7 +1097,7 @@ static int r300_packet0_check(struct radeon_cs_parser *p,
 		track->aa.robj = reloc->robj;
 		track->aa.offset = idx_value;
 		track->aa_dirty = true;
-		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
+		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
 		break;
 	case R300_RB3D_AARESOLVE_PITCH:
 		track->aa.pitch = idx_value & 0x3FFE;
@@ -1162,7 +1162,7 @@ static int r300_packet3_check(struct radeon_cs_parser *p,
 			radeon_cs_dump_packet(p, pkt);
 			return r;
 		}
-		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->lobj.gpu_offset);
+		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->gpu_offset);
 		r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
 		if (r) {
 			return r;

drivers/gpu/drm/radeon/radeon.h

@@ -363,9 +363,8 @@ int radeon_fence_emit(struct radeon_device *rdev, struct radeon_fence **fence, i
 void radeon_fence_process(struct radeon_device *rdev, int ring);
 bool radeon_fence_signaled(struct radeon_fence *fence);
 int radeon_fence_wait(struct radeon_fence *fence, bool interruptible);
-int radeon_fence_wait_locked(struct radeon_fence *fence);
-int radeon_fence_wait_next_locked(struct radeon_device *rdev, int ring);
-int radeon_fence_wait_empty_locked(struct radeon_device *rdev, int ring);
+int radeon_fence_wait_next(struct radeon_device *rdev, int ring);
+int radeon_fence_wait_empty(struct radeon_device *rdev, int ring);
 int radeon_fence_wait_any(struct radeon_device *rdev,
 			  struct radeon_fence **fences,
 			  bool intr);
@@ -457,6 +456,7 @@ struct radeon_bo {
 	/* Protected by gem.mutex */
 	struct list_head		list;
 	/* Protected by tbo.reserved */
+	u32				initial_domain;
 	u32				placements[3];
 	struct ttm_placement		placement;
 	struct ttm_buffer_object	tbo;
@@ -479,16 +479,6 @@ struct radeon_bo {
 };
 #define gem_to_radeon_bo(gobj) container_of((gobj), struct radeon_bo, gem_base)
 
-struct radeon_bo_list {
-	struct ttm_validate_buffer tv;
-	struct radeon_bo	*bo;
-	uint64_t		gpu_offset;
-	bool			written;
-	unsigned		domain;
-	unsigned		alt_domain;
-	u32			tiling_flags;
-};
-
 int radeon_gem_debugfs_init(struct radeon_device *rdev);
 
 /* sub-allocation manager, it has to be protected by another lock.
@@ -805,8 +795,8 @@ struct radeon_ring {
 	unsigned		ring_size;
 	unsigned		ring_free_dw;
 	int			count_dw;
-	unsigned long		last_activity;
-	unsigned		last_rptr;
+	atomic_t		last_rptr;
+	atomic64_t		last_activity;
 	uint64_t		gpu_addr;
 	uint32_t		align_mask;
 	uint32_t		ptr_mask;
@@ -858,17 +848,22 @@ struct radeon_mec {
 #define R600_PTE_READABLE	(1 << 5)
 #define R600_PTE_WRITEABLE	(1 << 6)
 
+struct radeon_vm_pt {
+	struct radeon_bo		*bo;
+	uint64_t			addr;
+};
+
 struct radeon_vm {
-	struct list_head		list;
 	struct list_head		va;
 	unsigned			id;
 
 	/* contains the page directory */
-	struct radeon_sa_bo		*page_directory;
+	struct radeon_bo		*page_directory;
 	uint64_t			pd_gpu_addr;
+	unsigned			max_pde_used;
 
 	/* array of page tables, one for each page directory entry */
-	struct radeon_sa_bo		**page_tables;
+	struct radeon_vm_pt		*page_tables;
 
 	struct mutex			mutex;
 	/* last fence for cs using this vm */
@@ -880,10 +875,7 @@ struct radeon_vm {
 };
 
 struct radeon_vm_manager {
-	struct mutex			lock;
-	struct list_head		lru_vm;
 	struct radeon_fence		*active[RADEON_NUM_VM];
-	struct radeon_sa_manager	sa_manager;
 	uint32_t			max_pfn;
 	/* number of VMIDs */
 	unsigned			nvm;
@@ -986,9 +978,12 @@ void cayman_dma_fini(struct radeon_device *rdev);
 struct radeon_cs_reloc {
 	struct drm_gem_object		*gobj;
 	struct radeon_bo		*robj;
-	struct radeon_bo_list		lobj;
+	struct ttm_validate_buffer	tv;
+	uint64_t			gpu_offset;
+	unsigned			domain;
+	unsigned			alt_domain;
+	uint32_t			tiling_flags;
 	uint32_t			handle;
-	uint32_t			flags;
 };
 
 struct radeon_cs_chunk {
@@ -1012,6 +1007,7 @@ struct radeon_cs_parser {
 	unsigned		nrelocs;
 	struct radeon_cs_reloc	*relocs;
 	struct radeon_cs_reloc	**relocs_ptr;
+	struct radeon_cs_reloc	*vm_bos;
 	struct list_head	validated;
 	unsigned		dma_reloc_idx;
 	/* indices of various chunks */
@@ -1635,7 +1631,6 @@ int radeon_uvd_send_upll_ctlreq(struct radeon_device *rdev,
 
 struct radeon_vce {
 	struct radeon_bo	*vcpu_bo;
-	void			*cpu_addr;
 	uint64_t		gpu_addr;
 	unsigned		fw_version;
 	unsigned		fb_version;
@@ -2117,6 +2112,8 @@ int radeon_gem_wait_idle_ioctl(struct drm_device *dev, void *data,
 			      struct drm_file *filp);
 int radeon_gem_va_ioctl(struct drm_device *dev, void *data,
 			  struct drm_file *filp);
+int radeon_gem_op_ioctl(struct drm_device *dev, void *data,
+			struct drm_file *filp);
 int radeon_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp);
 int radeon_gem_set_tiling_ioctl(struct drm_device *dev, void *data,
 				struct drm_file *filp);
@@ -2307,6 +2304,10 @@ struct radeon_device {
 	/* virtual memory */
 	struct radeon_vm_manager	vm_manager;
 	struct mutex			gpu_clock_mutex;
+	/* memory stats */
+	atomic64_t			vram_usage;
+	atomic64_t			gtt_usage;
+	atomic64_t			num_bytes_moved;
 	/* ACPI interface */
 	struct radeon_atif		atif;
 	struct radeon_atcs		atcs;
@@ -2794,16 +2795,22 @@ extern void radeon_program_register_sequence(struct radeon_device *rdev,
  */
 int radeon_vm_manager_init(struct radeon_device *rdev);
 void radeon_vm_manager_fini(struct radeon_device *rdev);
-void radeon_vm_init(struct radeon_device *rdev, struct radeon_vm *vm);
+int radeon_vm_init(struct radeon_device *rdev, struct radeon_vm *vm);
 void radeon_vm_fini(struct radeon_device *rdev, struct radeon_vm *vm);
-int radeon_vm_alloc_pt(struct radeon_device *rdev, struct radeon_vm *vm);
-void radeon_vm_add_to_lru(struct radeon_device *rdev, struct radeon_vm *vm);
+struct radeon_cs_reloc *radeon_vm_get_bos(struct radeon_device *rdev,
+					  struct radeon_vm *vm,
+                                          struct list_head *head);
 struct radeon_fence *radeon_vm_grab_id(struct radeon_device *rdev,
 				       struct radeon_vm *vm, int ring);
+void radeon_vm_flush(struct radeon_device *rdev,
+                     struct radeon_vm *vm,
+                     int ring);
 void radeon_vm_fence(struct radeon_device *rdev,
 		     struct radeon_vm *vm,
 		     struct radeon_fence *fence);
 uint64_t radeon_vm_map_gart(struct radeon_device *rdev, uint64_t addr);
+int radeon_vm_update_page_directory(struct radeon_device *rdev,
+				    struct radeon_vm *vm);
 int radeon_vm_bo_update(struct radeon_device *rdev,
 			struct radeon_vm *vm,
 			struct radeon_bo *bo,

drivers/gpu/drm/radeon/radeon_cs.c

@@ -24,16 +24,59 @@
  * Authors:
  *    Jerome Glisse <glisse@freedesktop.org>
  */
+#include <linux/list_sort.h>
 #include <drm/drmP.h>
 #include <drm/radeon_drm.h>
 #include "radeon_reg.h"
 #include "radeon.h"
 #include "radeon_trace.h"
 
+#define RADEON_CS_MAX_PRIORITY		32u
+#define RADEON_CS_NUM_BUCKETS		(RADEON_CS_MAX_PRIORITY + 1)
+
+/* This is based on the bucket sort with O(n) time complexity.
+ * An item with priority "i" is added to bucket[i]. The lists are then
+ * concatenated in descending order.
+ */
+struct radeon_cs_buckets {
+	struct list_head bucket[RADEON_CS_NUM_BUCKETS];
+};
+
+static void radeon_cs_buckets_init(struct radeon_cs_buckets *b)
+{
+	unsigned i;
+
+	for (i = 0; i < RADEON_CS_NUM_BUCKETS; i++)
+		INIT_LIST_HEAD(&b->bucket[i]);
+}
+
+static void radeon_cs_buckets_add(struct radeon_cs_buckets *b,
+				  struct list_head *item, unsigned priority)
+{
+	/* Since buffers which appear sooner in the relocation list are
+	 * likely to be used more often than buffers which appear later
+	 * in the list, the sort mustn't change the ordering of buffers
+	 * with the same priority, i.e. it must be stable.
+	 */
+	list_add_tail(item, &b->bucket[min(priority, RADEON_CS_MAX_PRIORITY)]);
+}
+
+static void radeon_cs_buckets_get_list(struct radeon_cs_buckets *b,
+				       struct list_head *out_list)
+{
+	unsigned i;
+
+	/* Connect the sorted buckets in the output list. */
+	for (i = 0; i < RADEON_CS_NUM_BUCKETS; i++) {
+		list_splice(&b->bucket[i], out_list);
+	}
+}
+
 static int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
 {
 	struct drm_device *ddev = p->rdev->ddev;
 	struct radeon_cs_chunk *chunk;
+	struct radeon_cs_buckets buckets;
 	unsigned i, j;
 	bool duplicate;
 
@@ -52,8 +95,12 @@ static int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
 	if (p->relocs == NULL) {
 		return -ENOMEM;
 	}
+
+	radeon_cs_buckets_init(&buckets);
+
 	for (i = 0; i < p->nrelocs; i++) {
 		struct drm_radeon_cs_reloc *r;
+		unsigned priority;
 
 		duplicate = false;
 		r = (struct drm_radeon_cs_reloc *)&chunk->kdata[i*4];
@@ -78,8 +125,14 @@ static int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
 		}
 		p->relocs_ptr[i] = &p->relocs[i];
 		p->relocs[i].robj = gem_to_radeon_bo(p->relocs[i].gobj);
-		p->relocs[i].lobj.bo = p->relocs[i].robj;
-		p->relocs[i].lobj.written = !!r->write_domain;
+
+		/* The userspace buffer priorities are from 0 to 15. A higher
+		 * number means the buffer is more important.
+		 * Also, the buffers used for write have a higher priority than
+		 * the buffers used for read only, which doubles the range
+		 * to 0 to 31. 32 is reserved for the kernel driver.
+		 */
+		priority = (r->flags & 0xf) * 2 + !!r->write_domain;
 
 		/* the first reloc of an UVD job is the msg and that must be in
 		   VRAM, also but everything into VRAM on AGP cards to avoid
@@ -87,29 +140,38 @@ static int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
 		if (p->ring == R600_RING_TYPE_UVD_INDEX &&
 		    (i == 0 || drm_pci_device_is_agp(p->rdev->ddev))) {
 			/* TODO: is this still needed for NI+ ? */
-			p->relocs[i].lobj.domain =
+			p->relocs[i].domain =
 				RADEON_GEM_DOMAIN_VRAM;
 
-			p->relocs[i].lobj.alt_domain =
+			p->relocs[i].alt_domain =
 				RADEON_GEM_DOMAIN_VRAM;
 
+			/* prioritize this over any other relocation */
+			priority = RADEON_CS_MAX_PRIORITY;
 		} else {
 			uint32_t domain = r->write_domain ?
 				r->write_domain : r->read_domains;
 
-			p->relocs[i].lobj.domain = domain;
+			p->relocs[i].domain = domain;
 			if (domain == RADEON_GEM_DOMAIN_VRAM)
 				domain |= RADEON_GEM_DOMAIN_GTT;
-			p->relocs[i].lobj.alt_domain = domain;
+			p->relocs[i].alt_domain = domain;
 		}
 
-		p->relocs[i].lobj.tv.bo = &p->relocs[i].robj->tbo;
+		p->relocs[i].tv.bo = &p->relocs[i].robj->tbo;
 		p->relocs[i].handle = r->handle;
 
-		radeon_bo_list_add_object(&p->relocs[i].lobj,
-					  &p->validated);
+		radeon_cs_buckets_add(&buckets, &p->relocs[i].tv.head,
+				      priority);
 	}
-	return radeon_bo_list_validate(&p->ticket, &p->validated, p->ring);
+
+	radeon_cs_buckets_get_list(&buckets, &p->validated);
+
+	if (p->cs_flags & RADEON_CS_USE_VM)
+		p->vm_bos = radeon_vm_get_bos(p->rdev, p->ib.vm,
+					      &p->validated);
+
+	return radeon_bo_list_validate(p->rdev, &p->ticket, &p->validated, p->ring);
 }
 
 static int radeon_cs_get_ring(struct radeon_cs_parser *p, u32 ring, s32 priority)
@@ -290,6 +352,16 @@ int radeon_cs_parser_init(struct radeon_cs_parser *p, void *data)
 	return 0;
 }
 
+static int cmp_size_smaller_first(void *priv, struct list_head *a,
+				  struct list_head *b)
+{
+	struct radeon_cs_reloc *la = list_entry(a, struct radeon_cs_reloc, tv.head);
+	struct radeon_cs_reloc *lb = list_entry(b, struct radeon_cs_reloc, tv.head);
+
+	/* Sort A before B if A is smaller. */
+	return (int)la->robj->tbo.num_pages - (int)lb->robj->tbo.num_pages;
+}
+
 /**
  * cs_parser_fini() - clean parser states
  * @parser:	parser structure holding parsing context.
@@ -303,6 +375,18 @@ static void radeon_cs_parser_fini(struct radeon_cs_parser *parser, int error, bo
 	unsigned i;
 
 	if (!error) {
+		/* Sort the buffer list from the smallest to largest buffer,
+		 * which affects the order of buffers in the LRU list.
+		 * This assures that the smallest buffers are added first
+		 * to the LRU list, so they are likely to be later evicted
+		 * first, instead of large buffers whose eviction is more
+		 * expensive.
+		 *
+		 * This slightly lowers the number of bytes moved by TTM
+		 * per frame under memory pressure.
+		 */
+		list_sort(NULL, &parser->validated, cmp_size_smaller_first);
+
 		ttm_eu_fence_buffer_objects(&parser->ticket,
 					    &parser->validated,
 					    parser->ib.fence);
@@ -320,6 +404,7 @@ static void radeon_cs_parser_fini(struct radeon_cs_parser *parser, int error, bo
 	kfree(parser->track);
 	kfree(parser->relocs);
 	kfree(parser->relocs_ptr);
+	kfree(parser->vm_bos);
 	for (i = 0; i < parser->nchunks; i++)
 		drm_free_large(parser->chunks[i].kdata);
 	kfree(parser->chunks);
@@ -359,24 +444,32 @@ static int radeon_cs_ib_chunk(struct radeon_device *rdev,
 	return r;
 }
 
-static int radeon_bo_vm_update_pte(struct radeon_cs_parser *parser,
+static int radeon_bo_vm_update_pte(struct radeon_cs_parser *p,
 				   struct radeon_vm *vm)
 {
-	struct radeon_device *rdev = parser->rdev;
-	struct radeon_bo_list *lobj;
-	struct radeon_bo *bo;
-	int r;
+	struct radeon_device *rdev = p->rdev;
+	int i, r;
 
-	r = radeon_vm_bo_update(rdev, vm, rdev->ring_tmp_bo.bo, &rdev->ring_tmp_bo.bo->tbo.mem);
-	if (r) {
+	r = radeon_vm_update_page_directory(rdev, vm);
+	if (r)
 		return r;
-	}
-	list_for_each_entry(lobj, &parser->validated, tv.head) {
-		bo = lobj->bo;
-		r = radeon_vm_bo_update(parser->rdev, vm, bo, &bo->tbo.mem);
-		if (r) {
+
+	r = radeon_vm_bo_update(rdev, vm, rdev->ring_tmp_bo.bo,
+				&rdev->ring_tmp_bo.bo->tbo.mem);
+	if (r)
+		return r;
+
+	for (i = 0; i < p->nrelocs; i++) {
+		struct radeon_bo *bo;
+
+		/* ignore duplicates */
+		if (p->relocs_ptr[i] != &p->relocs[i])
+			continue;
+
+		bo = p->relocs[i].robj;
+		r = radeon_vm_bo_update(rdev, vm, bo, &bo->tbo.mem);
+		if (r)
 			return r;
-		}
 	}
 	return 0;
 }
@@ -408,20 +501,13 @@ static int radeon_cs_ib_vm_chunk(struct radeon_device *rdev,
 	if (parser->ring == R600_RING_TYPE_UVD_INDEX)
 		radeon_uvd_note_usage(rdev);
 
-	mutex_lock(&rdev->vm_manager.lock);
 	mutex_lock(&vm->mutex);
-	r = radeon_vm_alloc_pt(rdev, vm);
-	if (r) {
-		goto out;
-	}
 	r = radeon_bo_vm_update_pte(parser, vm);
 	if (r) {
 		goto out;
 	}
 	radeon_cs_sync_rings(parser);
 	radeon_semaphore_sync_to(parser->ib.semaphore, vm->fence);
-	radeon_semaphore_sync_to(parser->ib.semaphore,
-				 radeon_vm_grab_id(rdev, vm, parser->ring));
 
 	if ((rdev->family >= CHIP_TAHITI) &&
 	    (parser->chunk_const_ib_idx != -1)) {
@@ -430,14 +516,8 @@ static int radeon_cs_ib_vm_chunk(struct radeon_device *rdev,
 		r = radeon_ib_schedule(rdev, &parser->ib, NULL);
 	}
 
-	if (!r) {
-		radeon_vm_fence(rdev, vm, parser->ib.fence);
-	}
-
 out:
-	radeon_vm_add_to_lru(rdev, vm);
 	mutex_unlock(&vm->mutex);
-	mutex_unlock(&rdev->vm_manager.lock);
 	return r;
 }
 
@@ -705,9 +785,9 @@ int radeon_cs_packet_next_reloc(struct radeon_cs_parser *p,
 	/* FIXME: we assume reloc size is 4 dwords */
 	if (nomm) {
 		*cs_reloc = p->relocs;
-		(*cs_reloc)->lobj.gpu_offset =
+		(*cs_reloc)->gpu_offset =
 			(u64)relocs_chunk->kdata[idx + 3] << 32;
-		(*cs_reloc)->lobj.gpu_offset |= relocs_chunk->kdata[idx + 0];
+		(*cs_reloc)->gpu_offset |= relocs_chunk->kdata[idx + 0];
 	} else
 		*cs_reloc = p->relocs_ptr[(idx / 4)];
 	return 0;

drivers/gpu/drm/radeon/radeon_device.c

@@ -1191,14 +1191,12 @@ int radeon_device_init(struct radeon_device *rdev,
 	r = radeon_gem_init(rdev);
 	if (r)
 		return r;
-	/* initialize vm here */
-	mutex_init(&rdev->vm_manager.lock);
+
 	/* Adjust VM size here.
 	 * Currently set to 4GB ((1 << 20) 4k pages).
 	 * Max GPUVM size for cayman and SI is 40 bits.
 	 */
 	rdev->vm_manager.max_pfn = 1 << 20;
-	INIT_LIST_HEAD(&rdev->vm_manager.lru_vm);
 
 	/* Set asic functions */
 	r = radeon_asic_init(rdev);
@@ -1445,10 +1443,9 @@ int radeon_suspend_kms(struct drm_device *dev, bool suspend, bool fbcon)
 	/* evict vram memory */
 	radeon_bo_evict_vram(rdev);
 
-	mutex_lock(&rdev->ring_lock);
 	/* wait for gpu to finish processing current batch */
 	for (i = 0; i < RADEON_NUM_RINGS; i++) {
-		r = radeon_fence_wait_empty_locked(rdev, i);
+		r = radeon_fence_wait_empty(rdev, i);
 		if (r) {
 			/* delay GPU reset to resume */
 			force_completion = true;
@@ -1457,7 +1454,6 @@ int radeon_suspend_kms(struct drm_device *dev, bool suspend, bool fbcon)
 	if (force_completion) {
 		radeon_fence_driver_force_completion(rdev);
 	}
-	mutex_unlock(&rdev->ring_lock);
 
 	radeon_save_bios_scratch_regs(rdev);
 

drivers/gpu/drm/radeon/radeon_drv.c

@@ -79,9 +79,10 @@
  *   2.35.0 - Add CIK macrotile mode array query
  *   2.36.0 - Fix CIK DCE tiling setup
  *   2.37.0 - allow GS ring setup on r6xx/r7xx
+ *   2.38.0 - RADEON_GEM_OP (GET_INITIAL_DOMAIN, SET_INITIAL_DOMAIN)
  */
 #define KMS_DRIVER_MAJOR	2
-#define KMS_DRIVER_MINOR	37
+#define KMS_DRIVER_MINOR	38
 #define KMS_DRIVER_PATCHLEVEL	0
 int radeon_driver_load_kms(struct drm_device *dev, unsigned long flags);
 int radeon_driver_unload_kms(struct drm_device *dev);

drivers/gpu/drm/radeon/radeon_fence.c

@@ -288,7 +288,6 @@ static bool radeon_fence_any_seq_signaled(struct radeon_device *rdev, u64 *seq)
  * @rdev: radeon device pointer
  * @target_seq: sequence number(s) we want to wait for
  * @intr: use interruptable sleep
- * @lock_ring: whether the ring should be locked or not
  *
  * Wait for the requested sequence number(s) to be written by any ring
  * (all asics).  Sequnce number array is indexed by ring id.
@@ -299,7 +298,7 @@ static bool radeon_fence_any_seq_signaled(struct radeon_device *rdev, u64 *seq)
  * -EDEADLK is returned when a GPU lockup has been detected.
  */
 static int radeon_fence_wait_seq(struct radeon_device *rdev, u64 *target_seq,
-				 bool intr, bool lock_ring)
+				 bool intr)
 {
 	uint64_t last_seq[RADEON_NUM_RINGS];
 	bool signaled;
@@ -358,9 +357,6 @@ static int radeon_fence_wait_seq(struct radeon_device *rdev, u64 *target_seq,
 			if (i != RADEON_NUM_RINGS)
 				continue;
 
-			if (lock_ring)
-				mutex_lock(&rdev->ring_lock);
-
 			for (i = 0; i < RADEON_NUM_RINGS; ++i) {
 				if (!target_seq[i])
 					continue;
@@ -378,14 +374,9 @@ static int radeon_fence_wait_seq(struct radeon_device *rdev, u64 *target_seq,
 
 				/* remember that we need an reset */
 				rdev->needs_reset = true;
-				if (lock_ring)
-					mutex_unlock(&rdev->ring_lock);
 				wake_up_all(&rdev->fence_queue);
 				return -EDEADLK;
 			}
-
-			if (lock_ring)
-				mutex_unlock(&rdev->ring_lock);
 		}
 	}
 	return 0;
@@ -416,7 +407,7 @@ int radeon_fence_wait(struct radeon_fence *fence, bool intr)
 	if (seq[fence->ring] == RADEON_FENCE_SIGNALED_SEQ)
 		return 0;
 
-	r = radeon_fence_wait_seq(fence->rdev, seq, intr, true);
+	r = radeon_fence_wait_seq(fence->rdev, seq, intr);
 	if (r)
 		return r;
 
@@ -464,7 +455,7 @@ int radeon_fence_wait_any(struct radeon_device *rdev,
 	if (num_rings == 0)
 		return -ENOENT;
 
-	r = radeon_fence_wait_seq(rdev, seq, intr, true);
+	r = radeon_fence_wait_seq(rdev, seq, intr);
 	if (r) {
 		return r;
 	}
@@ -472,37 +463,7 @@ int radeon_fence_wait_any(struct radeon_device *rdev,
 }
 
 /**
- * radeon_fence_wait_locked - wait for a fence to signal
- *
- * @fence: radeon fence object
- *
- * Wait for the requested fence to signal (all asics).
- * Returns 0 if the fence has passed, error for all other cases.
- */
-int radeon_fence_wait_locked(struct radeon_fence *fence)
-{
-	uint64_t seq[RADEON_NUM_RINGS] = {};
-	int r;
-
-	if (fence == NULL) {
-		WARN(1, "Querying an invalid fence : %p !\n", fence);
-		return -EINVAL;
-	}
-
-	seq[fence->ring] = fence->seq;
-	if (seq[fence->ring] == RADEON_FENCE_SIGNALED_SEQ)
-		return 0;
-
-	r = radeon_fence_wait_seq(fence->rdev, seq, false, false);
-	if (r)
-		return r;
-
-	fence->seq = RADEON_FENCE_SIGNALED_SEQ;
-	return 0;
-}
-
-/**
- * radeon_fence_wait_next_locked - wait for the next fence to signal
+ * radeon_fence_wait_next - wait for the next fence to signal
  *
  * @rdev: radeon device pointer
  * @ring: ring index the fence is associated with
@@ -511,7 +472,7 @@ int radeon_fence_wait_locked(struct radeon_fence *fence)
  * Returns 0 if the next fence has passed, error for all other cases.
  * Caller must hold ring lock.
  */
-int radeon_fence_wait_next_locked(struct radeon_device *rdev, int ring)
+int radeon_fence_wait_next(struct radeon_device *rdev, int ring)
 {
 	uint64_t seq[RADEON_NUM_RINGS] = {};
 
@@ -521,11 +482,11 @@ int radeon_fence_wait_next_locked(struct radeon_device *rdev, int ring)
 		   already the last emited fence */
 		return -ENOENT;
 	}
-	return radeon_fence_wait_seq(rdev, seq, false, false);
+	return radeon_fence_wait_seq(rdev, seq, false);
 }
 
 /**
- * radeon_fence_wait_empty_locked - wait for all fences to signal
+ * radeon_fence_wait_empty - wait for all fences to signal
  *
  * @rdev: radeon device pointer
  * @ring: ring index the fence is associated with
@@ -534,7 +495,7 @@ int radeon_fence_wait_next_locked(struct radeon_device *rdev, int ring)
  * Returns 0 if the fences have passed, error for all other cases.
  * Caller must hold ring lock.
  */
-int radeon_fence_wait_empty_locked(struct radeon_device *rdev, int ring)
+int radeon_fence_wait_empty(struct radeon_device *rdev, int ring)
 {
 	uint64_t seq[RADEON_NUM_RINGS] = {};
 	int r;
@@ -543,7 +504,7 @@ int radeon_fence_wait_empty_locked(struct radeon_device *rdev, int ring)
 	if (!seq[ring])
 		return 0;
 
-	r = radeon_fence_wait_seq(rdev, seq, false, false);
+	r = radeon_fence_wait_seq(rdev, seq, false);
 	if (r) {
 		if (r == -EDEADLK)
 			return -EDEADLK;
@@ -794,7 +755,7 @@ void radeon_fence_driver_fini(struct radeon_device *rdev)
 	for (ring = 0; ring < RADEON_NUM_RINGS; ring++) {
 		if (!rdev->fence_drv[ring].initialized)
 			continue;
-		r = radeon_fence_wait_empty_locked(rdev, ring);
+		r = radeon_fence_wait_empty(rdev, ring);
 		if (r) {
 			/* no need to trigger GPU reset as we are unloading */
 			radeon_fence_driver_force_completion(rdev);

drivers/gpu/drm/radeon/radeon_gem.c

@@ -344,18 +344,7 @@ int radeon_gem_busy_ioctl(struct drm_device *dev, void *data,
 	}
 	robj = gem_to_radeon_bo(gobj);
 	r = radeon_bo_wait(robj, &cur_placement, true);
-	switch (cur_placement) {
-	case TTM_PL_VRAM:
-		args->domain = RADEON_GEM_DOMAIN_VRAM;
-		break;
-	case TTM_PL_TT:
-		args->domain = RADEON_GEM_DOMAIN_GTT;
-		break;
-	case TTM_PL_SYSTEM:
-		args->domain = RADEON_GEM_DOMAIN_CPU;
-	default:
-		break;
-	}
+	args->domain = radeon_mem_type_to_domain(cur_placement);
 	drm_gem_object_unreference_unlocked(gobj);
 	r = radeon_gem_handle_lockup(rdev, r);
 	return r;
@@ -533,6 +522,42 @@ int radeon_gem_va_ioctl(struct drm_device *dev, void *data,
 	return r;
 }
 
+int radeon_gem_op_ioctl(struct drm_device *dev, void *data,
+			struct drm_file *filp)
+{
+	struct drm_radeon_gem_op *args = data;
+	struct drm_gem_object *gobj;
+	struct radeon_bo *robj;
+	int r;
+
+	gobj = drm_gem_object_lookup(dev, filp, args->handle);
+	if (gobj == NULL) {
+		return -ENOENT;
+	}
+	robj = gem_to_radeon_bo(gobj);
+	r = radeon_bo_reserve(robj, false);
+	if (unlikely(r))
+		goto out;
+
+	switch (args->op) {
+	case RADEON_GEM_OP_GET_INITIAL_DOMAIN:
+		args->value = robj->initial_domain;
+		break;
+	case RADEON_GEM_OP_SET_INITIAL_DOMAIN:
+		robj->initial_domain = args->value & (RADEON_GEM_DOMAIN_VRAM |
+						      RADEON_GEM_DOMAIN_GTT |
+						      RADEON_GEM_DOMAIN_CPU);
+		break;
+	default:
+		r = -EINVAL;
+	}
+
+	radeon_bo_unreserve(robj);
+out:
+	drm_gem_object_unreference_unlocked(gobj);
+	return r;
+}
+
 int radeon_mode_dumb_create(struct drm_file *file_priv,
 			    struct drm_device *dev,
 			    struct drm_mode_create_dumb *args)

drivers/gpu/drm/radeon/radeon_kms.c

@@ -486,6 +486,21 @@ static int radeon_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 	case RADEON_INFO_VCE_FB_VERSION:
 		*value = rdev->vce.fb_version;
 		break;
+	case RADEON_INFO_NUM_BYTES_MOVED:
+		value = (uint32_t*)&value64;
+		value_size = sizeof(uint64_t);
+		value64 = atomic64_read(&rdev->num_bytes_moved);
+		break;
+	case RADEON_INFO_VRAM_USAGE:
+		value = (uint32_t*)&value64;
+		value_size = sizeof(uint64_t);
+		value64 = atomic64_read(&rdev->vram_usage);
+		break;
+	case RADEON_INFO_GTT_USAGE:
+		value = (uint32_t*)&value64;
+		value_size = sizeof(uint64_t);
+		value64 = atomic64_read(&rdev->gtt_usage);
+		break;
 	default:
 		DRM_DEBUG_KMS("Invalid request %d\n", info->request);
 		return -EINVAL;
@@ -544,7 +559,13 @@ int radeon_driver_open_kms(struct drm_device *dev, struct drm_file *file_priv)
 			return -ENOMEM;
 		}
 
-		radeon_vm_init(rdev, &fpriv->vm);
+		r = radeon_vm_init(rdev, &fpriv->vm);
+		if (r)
+			return r;
+
+		r = radeon_bo_reserve(rdev->ring_tmp_bo.bo, false);
+		if (r)
+			return r;
 
 		/* map the ib pool buffer read only into
 		 * virtual address space */
@@ -553,6 +574,8 @@ int radeon_driver_open_kms(struct drm_device *dev, struct drm_file *file_priv)
 		r = radeon_vm_bo_set_addr(rdev, bo_va, RADEON_VA_IB_OFFSET,
 					  RADEON_VM_PAGE_READABLE |
 					  RADEON_VM_PAGE_SNOOPED);
+
+		radeon_bo_unreserve(rdev->ring_tmp_bo.bo);
 		if (r) {
 			radeon_vm_fini(rdev, &fpriv->vm);
 			kfree(fpriv);
@@ -814,5 +837,6 @@ const struct drm_ioctl_desc radeon_ioctls_kms[] = {
 	DRM_IOCTL_DEF_DRV(RADEON_GEM_GET_TILING, radeon_gem_get_tiling_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW),
 	DRM_IOCTL_DEF_DRV(RADEON_GEM_BUSY, radeon_gem_busy_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW),
 	DRM_IOCTL_DEF_DRV(RADEON_GEM_VA, radeon_gem_va_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(RADEON_GEM_OP, radeon_gem_op_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW),
 };
 int radeon_max_kms_ioctl = DRM_ARRAY_SIZE(radeon_ioctls_kms);

drivers/gpu/drm/radeon/radeon_object.c

@@ -56,11 +56,36 @@ static void radeon_bo_clear_va(struct radeon_bo *bo)
 	}
 }
 
+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);
@@ -145,6 +170,9 @@ int radeon_bo_create(struct radeon_device *rdev,
 	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 */
 	down_read(&rdev->pm.mclk_lock);
@@ -338,39 +366,105 @@ void radeon_bo_fini(struct radeon_device *rdev)
 	arch_phys_wc_del(rdev->mc.vram_mtrr);
 }
 
-void radeon_bo_list_add_object(struct radeon_bo_list *lobj,
-				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)
 {
-	if (lobj->written) {
-		list_add(&lobj->tv.head, head);
-	} else {
-		list_add_tail(&lobj->tv.head, head);
-	}
+	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 ww_acquire_ctx *ticket,
+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);
 
 	r = ttm_eu_reserve_buffers(ticket, head);
 	if (unlikely(r != 0)) {
 		return r;
 	}
+
 	list_for_each_entry(lobj, head, tv.head) {
-		bo = lobj->bo;
+		bo = lobj->robj;
 		if (!bo->pin_count) {
-			domain = lobj->domain;
+			u32 domain = lobj->domain;
+			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 (current_domain != RADEON_GEM_DOMAIN_CPU &&
+			    (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);
 			if (ring == R600_RING_TYPE_UVD_INDEX)
 				radeon_uvd_force_into_uvd_segment(bo);
-			r = ttm_bo_validate(&bo->tbo, &bo->placement,
-						true, false);
+
+			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 != lobj->alt_domain) {
 					domain = lobj->alt_domain;
@@ -564,14 +658,23 @@ 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)

drivers/gpu/drm/radeon/radeon_object.h

@@ -138,9 +138,8 @@ extern int radeon_bo_evict_vram(struct radeon_device *rdev);
 extern void radeon_bo_force_delete(struct radeon_device *rdev);
 extern int radeon_bo_init(struct radeon_device *rdev);
 extern void radeon_bo_fini(struct radeon_device *rdev);
-extern void radeon_bo_list_add_object(struct radeon_bo_list *lobj,
-				struct list_head *head);
-extern int radeon_bo_list_validate(struct ww_acquire_ctx *ticket,
+extern int radeon_bo_list_validate(struct radeon_device *rdev,
+				   struct ww_acquire_ctx *ticket,
 				   struct list_head *head, int ring);
 extern int radeon_bo_fbdev_mmap(struct radeon_bo *bo,
 				struct vm_area_struct *vma);
@@ -151,7 +150,7 @@ extern void radeon_bo_get_tiling_flags(struct radeon_bo *bo,
 extern int radeon_bo_check_tiling(struct radeon_bo *bo, bool has_moved,
 				bool force_drop);
 extern void radeon_bo_move_notify(struct ttm_buffer_object *bo,
-					struct ttm_mem_reg *mem);
+				  struct ttm_mem_reg *new_mem);
 extern int radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
 extern int radeon_bo_get_surface_reg(struct radeon_bo *bo);
 
@@ -181,7 +180,7 @@ extern int radeon_sa_bo_manager_suspend(struct radeon_device *rdev,
 extern int radeon_sa_bo_new(struct radeon_device *rdev,
 			    struct radeon_sa_manager *sa_manager,
 			    struct radeon_sa_bo **sa_bo,
-			    unsigned size, unsigned align, bool block);
+			    unsigned size, unsigned align);
 extern void radeon_sa_bo_free(struct radeon_device *rdev,
 			      struct radeon_sa_bo **sa_bo,
 			      struct radeon_fence *fence);

drivers/gpu/drm/radeon/radeon_pm.c

@@ -260,7 +260,7 @@ static void radeon_pm_set_clocks(struct radeon_device *rdev)
 		if (!ring->ready) {
 			continue;
 		}
-		r = radeon_fence_wait_empty_locked(rdev, i);
+		r = radeon_fence_wait_empty(rdev, i);
 		if (r) {
 			/* needs a GPU reset dont reset here */
 			mutex_unlock(&rdev->ring_lock);
@@ -896,7 +896,7 @@ static void radeon_dpm_change_power_state_locked(struct radeon_device *rdev)
 	for (i = 0; i < RADEON_NUM_RINGS; i++) {
 		struct radeon_ring *ring = &rdev->ring[i];
 		if (ring->ready)
-			radeon_fence_wait_empty_locked(rdev, i);
+			radeon_fence_wait_empty(rdev, i);
 	}
 
 	/* program the new power state */
@@ -943,8 +943,6 @@ void radeon_dpm_enable_uvd(struct radeon_device *rdev, bool enable)
 		if (enable) {
 			mutex_lock(&rdev->pm.mutex);
 			rdev->pm.dpm.uvd_active = true;
-			/* disable this for now */
-#if 0
 			if ((rdev->pm.dpm.sd == 1) && (rdev->pm.dpm.hd == 0))
 				dpm_state = POWER_STATE_TYPE_INTERNAL_UVD_SD;
 			else if ((rdev->pm.dpm.sd == 2) && (rdev->pm.dpm.hd == 0))
@@ -954,7 +952,6 @@ void radeon_dpm_enable_uvd(struct radeon_device *rdev, bool enable)
 			else if ((rdev->pm.dpm.sd == 0) && (rdev->pm.dpm.hd == 2))
 				dpm_state = POWER_STATE_TYPE_INTERNAL_UVD_HD2;
 			else
-#endif
 				dpm_state = POWER_STATE_TYPE_INTERNAL_UVD;
 			rdev->pm.dpm.state = dpm_state;
 			mutex_unlock(&rdev->pm.mutex);

drivers/gpu/drm/radeon/radeon_ring.c

@@ -63,7 +63,7 @@ int radeon_ib_get(struct radeon_device *rdev, int ring,
 {
 	int r;
 
-	r = radeon_sa_bo_new(rdev, &rdev->ring_tmp_bo, &ib->sa_bo, size, 256, true);
+	r = radeon_sa_bo_new(rdev, &rdev->ring_tmp_bo, &ib->sa_bo, size, 256);
 	if (r) {
 		dev_err(rdev->dev, "failed to get a new IB (%d)\n", r);
 		return r;
@@ -145,6 +145,13 @@ int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib,
 		return r;
 	}
 
+	/* grab a vm id if necessary */
+	if (ib->vm) {
+		struct radeon_fence *vm_id_fence;
+		vm_id_fence = radeon_vm_grab_id(rdev, ib->vm, ib->ring);
+        	radeon_semaphore_sync_to(ib->semaphore, vm_id_fence);
+	}
+
 	/* sync with other rings */
 	r = radeon_semaphore_sync_rings(rdev, ib->semaphore, ib->ring);
 	if (r) {
@@ -153,11 +160,9 @@ int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib,
 		return r;
 	}
 
-	/* if we can't remember our last VM flush then flush now! */
-	/* XXX figure out why we have to flush for every IB */
-	if (ib->vm /*&& !ib->vm->last_flush*/) {
-		radeon_ring_vm_flush(rdev, ib->ring, ib->vm);
-	}
+	if (ib->vm)
+		radeon_vm_flush(rdev, ib->vm, ib->ring);
+
 	if (const_ib) {
 		radeon_ring_ib_execute(rdev, const_ib->ring, const_ib);
 		radeon_semaphore_free(rdev, &const_ib->semaphore, NULL);
@@ -172,10 +177,10 @@ int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib,
 	if (const_ib) {
 		const_ib->fence = radeon_fence_ref(ib->fence);
 	}
-	/* we just flushed the VM, remember that */
-	if (ib->vm && !ib->vm->last_flush) {
-		ib->vm->last_flush = radeon_fence_ref(ib->fence);
-	}
+
+	if (ib->vm)
+		radeon_vm_fence(rdev, ib->vm, ib->fence);
+
 	radeon_ring_unlock_commit(rdev, ring);
 	return 0;
 }
@@ -382,7 +387,7 @@ int radeon_ring_alloc(struct radeon_device *rdev, struct radeon_ring *ring, unsi
 		if (ndw < ring->ring_free_dw) {
 			break;
 		}
-		r = radeon_fence_wait_next_locked(rdev, ring->idx);
+		r = radeon_fence_wait_next(rdev, ring->idx);
 		if (r)
 			return r;
 	}
@@ -485,8 +490,8 @@ void radeon_ring_unlock_undo(struct radeon_device *rdev, struct radeon_ring *rin
 void radeon_ring_lockup_update(struct radeon_device *rdev,
 			       struct radeon_ring *ring)
 {
-	ring->last_rptr = radeon_ring_get_rptr(rdev, ring);
-	ring->last_activity = jiffies;
+	atomic_set(&ring->last_rptr, radeon_ring_get_rptr(rdev, ring));
+	atomic64_set(&ring->last_activity, jiffies_64);
 }
 
 /**
@@ -498,22 +503,19 @@ void radeon_ring_lockup_update(struct radeon_device *rdev,
 bool radeon_ring_test_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
 {
 	uint32_t rptr = radeon_ring_get_rptr(rdev, ring);
-	unsigned long cjiffies, elapsed;
+	uint64_t last = atomic64_read(&ring->last_activity);
+	uint64_t elapsed;
 
-	cjiffies = jiffies;
-	if (!time_after(cjiffies, ring->last_activity)) {
-		/* likely a wrap around */
+	if (rptr != atomic_read(&ring->last_rptr)) {
+		/* ring is still working, no lockup */
 		radeon_ring_lockup_update(rdev, ring);
 		return false;
 	}
-	if (rptr != ring->last_rptr) {
-		/* CP is still working no lockup */
-		radeon_ring_lockup_update(rdev, ring);
-		return false;
-	}
-	elapsed = jiffies_to_msecs(cjiffies - ring->last_activity);
+
+	elapsed = jiffies_to_msecs(jiffies_64 - last);
 	if (radeon_lockup_timeout && elapsed >= radeon_lockup_timeout) {
-		dev_err(rdev->dev, "GPU lockup CP stall for more than %lumsec\n", elapsed);
+		dev_err(rdev->dev, "ring %d stalled for more than %llumsec\n",
+			ring->idx, elapsed);
 		return true;
 	}
 	/* give a chance to the GPU ... */

drivers/gpu/drm/radeon/radeon_sa.c

@@ -312,7 +312,7 @@ static bool radeon_sa_bo_next_hole(struct radeon_sa_manager *sa_manager,
 int radeon_sa_bo_new(struct radeon_device *rdev,
 		     struct radeon_sa_manager *sa_manager,
 		     struct radeon_sa_bo **sa_bo,
-		     unsigned size, unsigned align, bool block)
+		     unsigned size, unsigned align)
 {
 	struct radeon_fence *fences[RADEON_NUM_RINGS];
 	unsigned tries[RADEON_NUM_RINGS];
@@ -353,14 +353,11 @@ int radeon_sa_bo_new(struct radeon_device *rdev,
 		r = radeon_fence_wait_any(rdev, fences, false);
 		spin_lock(&sa_manager->wq.lock);
 		/* if we have nothing to wait for block */
-		if (r == -ENOENT && block) {
+		if (r == -ENOENT) {
 			r = wait_event_interruptible_locked(
 				sa_manager->wq, 
 				radeon_sa_event(sa_manager, size, align)
 			);
-
-		} else if (r == -ENOENT) {
-			r = -ENOMEM;
 		}
 
 	} while (!r);

drivers/gpu/drm/radeon/radeon_semaphore.c

@@ -42,7 +42,7 @@ int radeon_semaphore_create(struct radeon_device *rdev,
 		return -ENOMEM;
 	}
 	r = radeon_sa_bo_new(rdev, &rdev->ring_tmp_bo, &(*semaphore)->sa_bo,
-			     8 * RADEON_NUM_SYNCS, 8, true);
+			     8 * RADEON_NUM_SYNCS, 8);
 	if (r) {
 		kfree(*semaphore);
 		*semaphore = NULL;
@@ -147,7 +147,9 @@ int radeon_semaphore_sync_rings(struct radeon_device *rdev,
 
 		if (++count > RADEON_NUM_SYNCS) {
 			/* not enough room, wait manually */
-			radeon_fence_wait_locked(fence);
+			r = radeon_fence_wait(fence, false);
+			if (r)
+				return r;
 			continue;
 		}
 
@@ -161,7 +163,9 @@ int radeon_semaphore_sync_rings(struct radeon_device *rdev,
 		if (!radeon_semaphore_emit_signal(rdev, i, semaphore)) {
 			/* signaling wasn't successful wait manually */
 			radeon_ring_undo(&rdev->ring[i]);
-			radeon_fence_wait_locked(fence);
+			r = radeon_fence_wait(fence, false);
+			if (r)
+				return r;
 			continue;
 		}
 
@@ -169,7 +173,9 @@ int radeon_semaphore_sync_rings(struct radeon_device *rdev,
 		if (!radeon_semaphore_emit_wait(rdev, ring, semaphore)) {
 			/* waiting wasn't successful wait manually */
 			radeon_ring_undo(&rdev->ring[i]);
-			radeon_fence_wait_locked(fence);
+			r = radeon_fence_wait(fence, false);
+			if (r)
+				return r;
 			continue;
 		}
 

drivers/gpu/drm/radeon/radeon_ttm.c

@@ -406,8 +406,14 @@ static int radeon_bo_move(struct ttm_buffer_object *bo,
 	if (r) {
 memcpy:
 		r = ttm_bo_move_memcpy(bo, evict, no_wait_gpu, new_mem);
-	}
+		if (r) {
 			return r;
+		}
+	}
+
+	/* update statistics */
+	atomic64_add((u64)bo->num_pages << PAGE_SHIFT, &rdev->num_bytes_moved);
+	return 0;
 }
 
 static int radeon_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)

drivers/gpu/drm/radeon/radeon_uvd.c

@@ -453,7 +453,7 @@ static int radeon_uvd_cs_reloc(struct radeon_cs_parser *p,
 	}
 
 	reloc = p->relocs_ptr[(idx / 4)];
-	start = reloc->lobj.gpu_offset;
+	start = reloc->gpu_offset;
 	end = start + radeon_bo_size(reloc->robj);
 	start += offset;
 
@@ -805,8 +805,7 @@ void radeon_uvd_note_usage(struct radeon_device *rdev)
 		    (rdev->pm.dpm.hd != hd)) {
 			rdev->pm.dpm.sd = sd;
 			rdev->pm.dpm.hd = hd;
-			/* disable this for now */
-			/*streams_changed = true;*/
+			streams_changed = true;
 		}
 	}
 

drivers/gpu/drm/radeon/radeon_vce.c

@@ -119,7 +119,7 @@ int radeon_vce_init(struct radeon_device *rdev)
 	if (rdev->vce.fw_version != ((40 << 24) | (2 << 16) | (2 << 8)))
 		return -EINVAL;
 
-	/* load firmware into VRAM */
+	/* allocate firmware, stack and heap BO */
 
 	size = RADEON_GPU_PAGE_ALIGN(rdev->vce_fw->size) +
 	       RADEON_VCE_STACK_SIZE + RADEON_VCE_HEAP_SIZE;
@@ -130,16 +130,21 @@ int radeon_vce_init(struct radeon_device *rdev)
 		return r;
 	}
 
-	r = radeon_vce_resume(rdev);
-	if (r)
+	r = radeon_bo_reserve(rdev->vce.vcpu_bo, false);
+	if (r) {
+		radeon_bo_unref(&rdev->vce.vcpu_bo);
+		dev_err(rdev->dev, "(%d) failed to reserve VCE bo\n", r);
 		return r;
+	}
 
-	memset(rdev->vce.cpu_addr, 0, size);
-	memcpy(rdev->vce.cpu_addr, rdev->vce_fw->data, rdev->vce_fw->size);
-
-	r = radeon_vce_suspend(rdev);
-	if (r)
+	r = radeon_bo_pin(rdev->vce.vcpu_bo, RADEON_GEM_DOMAIN_VRAM,
+			  &rdev->vce.gpu_addr);
+	radeon_bo_unreserve(rdev->vce.vcpu_bo);
+	if (r) {
+		radeon_bo_unref(&rdev->vce.vcpu_bo);
+		dev_err(rdev->dev, "(%d) VCE bo pin failed\n", r);
 		return r;
+	}
 
 	for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) {
 		atomic_set(&rdev->vce.handles[i], 0);
@@ -158,8 +163,12 @@ int radeon_vce_init(struct radeon_device *rdev)
  */
 void radeon_vce_fini(struct radeon_device *rdev)
 {
-	radeon_vce_suspend(rdev);
+	if (rdev->vce.vcpu_bo == NULL)
+		return;
+
 	radeon_bo_unref(&rdev->vce.vcpu_bo);
+
+	release_firmware(rdev->vce_fw);
 }
 
 /**
@@ -167,22 +176,23 @@ void radeon_vce_fini(struct radeon_device *rdev)
  *
  * @rdev: radeon_device pointer
  *
- * TODO: Test VCE suspend/resume
  */
 int radeon_vce_suspend(struct radeon_device *rdev)
 {
-	int r;
+	int i;
 
 	if (rdev->vce.vcpu_bo == NULL)
 		return 0;
 
-	r = radeon_bo_reserve(rdev->vce.vcpu_bo, false);
-	if (!r) {
-		radeon_bo_kunmap(rdev->vce.vcpu_bo);
-		radeon_bo_unpin(rdev->vce.vcpu_bo);
-		radeon_bo_unreserve(rdev->vce.vcpu_bo);
-	}
-	return r;
+	for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i)
+		if (atomic_read(&rdev->vce.handles[i]))
+			break;
+
+	if (i == RADEON_MAX_VCE_HANDLES)
+		return 0;
+
+	/* TODO: suspending running encoding sessions isn't supported */
+	return -EINVAL;
 }
 
 /**
@@ -190,10 +200,10 @@ int radeon_vce_suspend(struct radeon_device *rdev)
  *
  * @rdev: radeon_device pointer
  *
- * TODO: Test VCE suspend/resume
  */
 int radeon_vce_resume(struct radeon_device *rdev)
 {
+	void *cpu_addr;
 	int r;
 
 	if (rdev->vce.vcpu_bo == NULL)
@@ -201,26 +211,21 @@ int radeon_vce_resume(struct radeon_device *rdev)
 
 	r = radeon_bo_reserve(rdev->vce.vcpu_bo, false);
 	if (r) {
-		radeon_bo_unref(&rdev->vce.vcpu_bo);
 		dev_err(rdev->dev, "(%d) failed to reserve VCE bo\n", r);
 		return r;
 	}
 
-	r = radeon_bo_pin(rdev->vce.vcpu_bo, RADEON_GEM_DOMAIN_VRAM,
-			  &rdev->vce.gpu_addr);
+	r = radeon_bo_kmap(rdev->vce.vcpu_bo, &cpu_addr);
 	if (r) {
 		radeon_bo_unreserve(rdev->vce.vcpu_bo);
-		radeon_bo_unref(&rdev->vce.vcpu_bo);
-		dev_err(rdev->dev, "(%d) VCE bo pin failed\n", r);
-		return r;
-	}
-
-	r = radeon_bo_kmap(rdev->vce.vcpu_bo, &rdev->vce.cpu_addr);
-	if (r) {
 		dev_err(rdev->dev, "(%d) VCE map failed\n", r);
 		return r;
 	}
 
+	memcpy(cpu_addr, rdev->vce_fw->data, rdev->vce_fw->size);
+
+	radeon_bo_kunmap(rdev->vce.vcpu_bo);
+
 	radeon_bo_unreserve(rdev->vce.vcpu_bo);
 
 	return 0;
@@ -456,7 +461,7 @@ int radeon_vce_cs_reloc(struct radeon_cs_parser *p, int lo, int hi)
 		return -EINVAL;
 	}
 
-	offset += p->relocs_ptr[(idx / 4)]->lobj.gpu_offset;
+	offset += p->relocs_ptr[(idx / 4)]->gpu_offset;
 
         p->ib.ptr[lo] = offset & 0xFFFFFFFF;
         p->ib.ptr[hi] = offset >> 32;

include/uapi/drm/radeon_drm.h

@@ -510,6 +510,7 @@ typedef struct {
 #define DRM_RADEON_GEM_GET_TILING	0x29
 #define DRM_RADEON_GEM_BUSY		0x2a
 #define DRM_RADEON_GEM_VA		0x2b
+#define DRM_RADEON_GEM_OP		0x2c
 
 #define DRM_IOCTL_RADEON_CP_INIT    DRM_IOW( DRM_COMMAND_BASE + DRM_RADEON_CP_INIT, drm_radeon_init_t)
 #define DRM_IOCTL_RADEON_CP_START   DRM_IO(  DRM_COMMAND_BASE + DRM_RADEON_CP_START)
@@ -552,6 +553,7 @@ typedef struct {
 #define DRM_IOCTL_RADEON_GEM_GET_TILING	DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_GET_TILING, struct drm_radeon_gem_get_tiling)
 #define DRM_IOCTL_RADEON_GEM_BUSY	DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_BUSY, struct drm_radeon_gem_busy)
 #define DRM_IOCTL_RADEON_GEM_VA		DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_VA, struct drm_radeon_gem_va)
+#define DRM_IOCTL_RADEON_GEM_OP		DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_OP, struct drm_radeon_gem_op)
 
 typedef struct drm_radeon_init {
 	enum {
@@ -884,6 +886,16 @@ struct drm_radeon_gem_pwrite {
 	uint64_t data_ptr;
 };
 
+/* Sets or returns a value associated with a buffer. */
+struct drm_radeon_gem_op {
+	uint32_t	handle; /* buffer */
+	uint32_t	op;     /* RADEON_GEM_OP_* */
+	uint64_t	value;  /* input or return value */
+};
+
+#define RADEON_GEM_OP_GET_INITIAL_DOMAIN	0
+#define RADEON_GEM_OP_SET_INITIAL_DOMAIN	1
+
 #define RADEON_VA_MAP			1
 #define RADEON_VA_UNMAP			2
 
@@ -992,6 +1004,9 @@ struct drm_radeon_cs {
 #define RADEON_INFO_VCE_FW_VERSION	0x1b
 /* version of VCE feedback */
 #define RADEON_INFO_VCE_FB_VERSION	0x1c
+#define RADEON_INFO_NUM_BYTES_MOVED	0x1d
+#define RADEON_INFO_VRAM_USAGE		0x1e
+#define RADEON_INFO_GTT_USAGE		0x1f
 
 
 struct drm_radeon_info {