r600g: attempt to abstract kernel bos from pipe driver.

introduce an abstraction layer between kernel bos and the winsys BOs.

this is to allow plugging in pb manager with minimal disruption to pipe driver.

19 files changed, 272 insertions, 199 deletions

diff --git a/src/gallium/drivers/r600/eg_hw_states.c b/src/gallium/drivers/r600/eg_hw_states.c
index a58adc6bab..e4b5b316b8 100644
--- a/src/gallium/drivers/r600/eg_hw_states.c
+++ b/src/gallium/drivers/r600/eg_hw_states.c
@@ -120,7 +120,7 @@ static void eg_cb(struct r600_context *rctx, struct radeon_state *rstate,
 	radeon_state_init(rstate, rscreen->rw, R600_STATE_CB0, cb, 0);
 	rtex = (struct r600_resource_texture*)state->cbufs[cb]->texture;
 	rbuffer = &rtex->resource;
-	rstate->bo[0] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
+	radeon_ws_bo_reference(rscreen->rw, &rstate->bo[0], rbuffer->bo);
 	rstate->placement[0] = RADEON_GEM_DOMAIN_VRAM;
 	rstate->nbo = 1;
 	pitch = (rtex->pitch[level] / rtex->bpt) / 8 - 1;
@@ -170,7 +170,7 @@ static void eg_db(struct r600_context *rctx, struct radeon_state *rstate,
 	rtex->depth = 1;
 	rbuffer = &rtex->resource;
 
-	rstate->bo[0] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
+	radeon_ws_bo_reference(rscreen->rw, &rstate->bo[0], rbuffer->bo);
 	rstate->nbo = 1;
 	rstate->placement[0] = RADEON_GEM_DOMAIN_VRAM;
 	level = state->zsbuf->level;
@@ -544,11 +544,11 @@ static void eg_resource(struct pipe_context *ctx, struct radeon_state *rstate,
 		if (r) {
 			return;
 		}
-		rstate->bo[0] = radeon_bo_incref(rscreen->rw, tmp->uncompressed);
-		rstate->bo[1] = radeon_bo_incref(rscreen->rw, tmp->uncompressed);
+		radeon_ws_bo_reference(rscreen->rw, &rstate->bo[0], tmp->uncompressed);
+		radeon_ws_bo_reference(rscreen->rw, &rstate->bo[1], tmp->uncompressed);
 	} else {
-		rstate->bo[0] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
-		rstate->bo[1] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
+		radeon_ws_bo_reference(rscreen->rw, &rstate->bo[0], rbuffer->bo);
+		radeon_ws_bo_reference(rscreen->rw, &rstate->bo[1], rbuffer->bo);
 	}
 	rstate->nbo = 2;
 	rstate->placement[0] = RADEON_GEM_DOMAIN_GTT;
@@ -860,10 +860,11 @@ static int eg_vs_resource(struct r600_context *rctx, int id, struct r600_resourc
 	struct r600_screen *rscreen = rctx->screen;
 
 	radeon_state_init(vs_resource, rscreen->rw, R600_STATE_RESOURCE, id, R600_SHADER_VS);
-	vs_resource->bo[0] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
+
+	radeon_ws_bo_reference(rscreen->rw, &vs_resource->bo[0], rbuffer->bo);
 	vs_resource->nbo = 1;
 	vs_resource->states[EG_PS_RESOURCE__RESOURCE0_WORD0] = offset;
-	vs_resource->states[EG_PS_RESOURCE__RESOURCE0_WORD1] = rbuffer->bo->size - offset - 1;
+	vs_resource->states[EG_PS_RESOURCE__RESOURCE0_WORD1] = rbuffer->size - offset - 1;
 	vs_resource->states[EG_PS_RESOURCE__RESOURCE0_WORD2] = S_030008_STRIDE(stride) |
 		S_030008_DATA_FORMAT(format);
 	vs_resource->states[EG_PS_RESOURCE__RESOURCE0_WORD3] = S_03000C_DST_SEL_X(V_03000C_SQ_SEL_X) |
@@ -891,7 +892,7 @@ static int eg_draw_vgt_init(struct r600_draw *draw,
 	draw->draw.states[EG_DRAW__VGT_DRAW_INITIATOR] = vgt_draw_initiator;
 	draw->draw.states[EG_DRAW__VGT_DMA_BASE] = draw->index_buffer_offset;
 	if (rbuffer) {
-		draw->draw.bo[0] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
+		radeon_ws_bo_reference(rscreen->rw, &draw->draw.bo[0], rbuffer->bo);
 		draw->draw.placement[0] = RADEON_GEM_DOMAIN_GTT;
 		draw->draw.placement[1] = RADEON_GEM_DOMAIN_GTT;
 		draw->draw.nbo = 1;
@@ -973,7 +974,7 @@ static int eg_ps_shader(struct r600_context *rctx, struct r600_context_state *rp
 	state->states[EG_PS_SHADER__SQ_PGM_EXPORTS_PS] = exports_ps;
 	state->states[EG_PS_SHADER__SPI_BARYC_CNTL] = S_0286E0_PERSP_CENTROID_ENA(1) |
 	  S_0286E0_LINEAR_CENTROID_ENA(1); 
-	state->bo[0] = radeon_bo_incref(rscreen->rw, rpshader->bo);
+	radeon_ws_bo_reference(rscreen->rw, &state->bo[0], rpshader->bo);
 	state->nbo = 1;
 	state->placement[0] = RADEON_GEM_DOMAIN_GTT;
 	return radeon_state_pm4(state);
@@ -998,8 +999,8 @@ static int eg_vs_shader(struct r600_context *rctx, struct r600_context_state *rp
 	state->states[EG_VS_SHADER__SPI_VS_OUT_CONFIG] = S_0286C4_VS_EXPORT_COUNT(rshader->noutput - 2);
 	state->states[EG_VS_SHADER__SQ_PGM_RESOURCES_VS] = S_028860_NUM_GPRS(rshader->bc.ngpr) |
 		S_028860_STACK_SIZE(rshader->bc.nstack);
-	state->bo[0] = radeon_bo_incref(rscreen->rw, rpshader->bo);
-	state->bo[1] = radeon_bo_incref(rscreen->rw, rpshader->bo);
+	radeon_ws_bo_reference(rscreen->rw, &state->bo[0], rpshader->bo);
+	radeon_ws_bo_reference(rscreen->rw, &state->bo[1], rpshader->bo);
 	state->nbo = 2;
 	state->placement[0] = RADEON_GEM_DOMAIN_GTT;
 	state->placement[2] = RADEON_GEM_DOMAIN_GTT;
@@ -1064,12 +1065,12 @@ static void eg_texture_state_cb(struct r600_screen *rscreen, struct r600_resourc
 	format = r600_translate_colorformat(rtexture->resource.base.b.format);
 	swap = r600_translate_colorswap(rtexture->resource.base.b.format);
 	if (desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS) {
-		rstate->bo[0] = radeon_bo_incref(rscreen->rw, rtexture->uncompressed);
+		radeon_ws_bo_reference(rscreen->rw, &rstate->bo[0], rtexture->uncompressed);
 		rstate->placement[0] = RADEON_GEM_DOMAIN_GTT;
 		rstate->nbo = 1;
 		color_info = 0;
 	} else {
-		rstate->bo[0] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
+		radeon_ws_bo_reference(rscreen->rw, &rstate->bo[0], rbuffer->bo);
 		rstate->placement[0] = RADEON_GEM_DOMAIN_GTT;
 		rstate->nbo = 1;
 		color_info = S_028C70_SOURCE_FORMAT(1);
@@ -1113,7 +1114,7 @@ static void eg_texture_state_db(struct r600_screen *rscreen, struct r600_resourc
 	rstate->states[EG_DB__DB_DEPTH_VIEW] = 0x00000000;
 	rstate->states[EG_DB__DB_DEPTH_SIZE] = S_028058_PITCH_TILE_MAX(pitch);
 	rstate->states[EG_DB__DB_DEPTH_SLICE] = S_02805C_SLICE_TILE_MAX(slice);
-	rstate->bo[0] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
+	radeon_ws_bo_reference(rscreen->rw, &rstate->bo[0], rbuffer->bo);
 	rstate->placement[0] = RADEON_GEM_DOMAIN_GTT;
 	rstate->nbo = 1;
 
@@ -1202,7 +1203,7 @@ void eg_set_constant_buffer(struct pipe_context *ctx,
 	rstate->states[EG_VS_CBUF__ALU_CONST_BUFFER_SIZE_VS_0] = size;
 	rstate->states[EG_VS_CBUF__ALU_CONST_CACHE_VS_0] = 0;
 
-	rstate->bo[0] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
+	radeon_ws_bo_reference(rscreen->rw, &rstate->bo[0], rbuffer->bo);
 	rstate->nbo = 1;
 	rstate->placement[0] = RADEON_GEM_DOMAIN_VRAM;
 	if (radeon_state_pm4(rstate))
diff --git a/src/gallium/drivers/r600/r600_blit.c b/src/gallium/drivers/r600/r600_blit.c
index 2c22adb62a..0e061c25f7 100644
--- a/src/gallium/drivers/r600/r600_blit.c
+++ b/src/gallium/drivers/r600/r600_blit.c
@@ -165,7 +165,8 @@ struct r600_blit_states {
 static int r600_blit_state_vs_resources(struct r600_screen *rscreen, struct r600_blit_states *bstates)
 {
 	struct radeon_state *rstate;
-	struct radeon_bo *bo;
+	struct radeon_ws_bo *bo;
+	void *data;
 	u32 vbo[] = {
 		0xBF800000, 0xBF800000, 0x3F800000, 0x3F800000,
 		0x3F000000, 0x3F000000, 0x3F000000, 0x00000000,
@@ -178,16 +179,17 @@ static int r600_blit_state_vs_resources(struct r600_screen *rscreen, struct r600
 	};
 
 	/* simple shader */
-	bo = radeon_bo(rscreen->rw, 0, 128, 4096, NULL);
+	bo = radeon_ws_bo(rscreen->rw, 128, 4096);
 	if (bo == NULL) {
 		return -ENOMEM;
 	}
-	if (radeon_bo_map(rscreen->rw, bo)) {
-		radeon_bo_decref(rscreen->rw, bo);
+	data = radeon_ws_bo_map(rscreen->rw, bo);
+	if (!data) {
+		radeon_ws_bo_reference(rscreen->rw, &bo, NULL);
 		return -ENOMEM;
 	}
-	memcpy(bo->data, vbo, 128);
-	radeon_bo_unmap(rscreen->rw, bo);
+	memcpy(data, vbo, 128);
+	radeon_ws_bo_unmap(rscreen->rw, bo);
 
 	rstate = &bstates->vs_resource0;
 	radeon_state_init(rstate, rscreen->rw, R600_STATE_RESOURCE, 0, R600_SHADER_VS);
@@ -219,7 +221,7 @@ static int r600_blit_state_vs_resources(struct r600_screen *rscreen, struct r600
 	rstate->states[R600_VS_RESOURCE__RESOURCE160_WORD4] = 0x00000000;
 	rstate->states[R600_VS_RESOURCE__RESOURCE160_WORD5] = 0x00000000;
 	rstate->states[R600_VS_RESOURCE__RESOURCE160_WORD6] = 0xC0000000;
-	rstate->bo[0] = radeon_bo_incref(rscreen->rw, bo);
+	radeon_ws_bo_reference(rscreen->rw, &rstate->bo[0], bo);
 	rstate->nbo = 1;
 	rstate->placement[0] = RADEON_GEM_DOMAIN_GTT;
 	if (radeon_state_pm4(rstate)) {
@@ -232,7 +234,8 @@ static int r600_blit_state_vs_resources(struct r600_screen *rscreen, struct r600
 
 static void r600_blit_state_vs_shader(struct r600_screen *rscreen, struct radeon_state *rstate)
 {
-	struct radeon_bo *bo;
+	struct radeon_ws_bo *bo;
+	void *data;
 	u32 shader_bc_r600[] = {
 		0x00000004, 0x81000400,
 		0x00000008, 0xA01C0000,
@@ -271,28 +274,29 @@ static void r600_blit_state_vs_shader(struct r600_screen *rscreen, struct radeon
 	};
 
 	/* simple shader */
-	bo = radeon_bo(rscreen->rw, 0, 128, 4096, NULL);
+	bo = radeon_ws_bo(rscreen->rw, 128, 4096);
 	if (bo == NULL) {
 		return;
 	}
-	if (radeon_bo_map(rscreen->rw, bo)) {
-		radeon_bo_decref(rscreen->rw, bo);
+	data = radeon_ws_bo_map(rscreen->rw, bo);
+	if (!data) {
+		radeon_ws_bo_reference(rscreen->rw, &bo, NULL);
 		return;
 	}
 	switch (rscreen->chip_class) {
 	case R600:
-		memcpy(bo->data, shader_bc_r600, 128);
+		memcpy(data, shader_bc_r600, 128);
 		break;
 	case R700:
-		memcpy(bo->data, shader_bc_r700, 128);
+		memcpy(data, shader_bc_r700, 128);
 		break;
 	default:
 		R600_ERR("unsupported chip family\n");
-		radeon_bo_unmap(rscreen->rw, bo);
-		radeon_bo_decref(rscreen->rw, bo);
+		radeon_ws_bo_unmap(rscreen->rw, bo);
+		radeon_ws_bo_reference(rscreen->rw, &bo, NULL);
 		return;
 	}
-	radeon_bo_unmap(rscreen->rw, bo);
+	radeon_ws_bo_unmap(rscreen->rw, bo);
 
 	radeon_state_init(rstate, rscreen->rw, R600_STATE_SHADER, 0, R600_SHADER_VS);
 
@@ -304,7 +308,7 @@ static void r600_blit_state_vs_shader(struct r600_screen *rscreen, struct radeon
 	rstate->states[R600_VS_SHADER__SQ_PGM_RESOURCES_VS] = 0x00000005;
 
 	rstate->bo[0] = bo;
-	rstate->bo[1] = radeon_bo_incref(rscreen->rw, bo);
+	radeon_ws_bo_reference(rscreen->rw, &rstate->bo[1], bo);
 	rstate->nbo = 2;
 	rstate->placement[0] = RADEON_GEM_DOMAIN_GTT;
 	rstate->placement[2] = RADEON_GEM_DOMAIN_GTT;
@@ -314,7 +318,8 @@ static void r600_blit_state_vs_shader(struct r600_screen *rscreen, struct radeon
 
 static void r600_blit_state_ps_shader(struct r600_screen *rscreen, struct radeon_state *rstate)
 {
-	struct radeon_bo *bo;
+	struct radeon_ws_bo *bo;
+	void *data;
 	u32 shader_bc_r600[] = {
 		0x00000002, 0xA00C0000,
 		0xC0008000, 0x94200688,
@@ -333,28 +338,29 @@ static void r600_blit_state_ps_shader(struct r600_screen *rscreen, struct radeon
 	};
 
 	/* simple shader */
-	bo = radeon_bo(rscreen->rw, 0, 128, 4096, NULL);
+	bo = radeon_ws_bo(rscreen->rw, 128, 4096);
 	if (bo == NULL) {
 		return;
 	}
-	if (radeon_bo_map(rscreen->rw, bo)) {
-		radeon_bo_decref(rscreen->rw, bo);
+	data = radeon_ws_bo_map(rscreen->rw, bo);
+	if (!data) {
+		radeon_ws_bo_reference(rscreen->rw, &bo, NULL);
 		return;
 	}
 	switch (rscreen->chip_class) {
 	case R600:
-		memcpy(bo->data, shader_bc_r600, 48);
+		memcpy(data, shader_bc_r600, 48);
 		break;
 	case R700:
-		memcpy(bo->data, shader_bc_r700, 48);
+		memcpy(data, shader_bc_r700, 48);
 		break;
 	default:
 		R600_ERR("unsupported chip family\n");
-		radeon_bo_unmap(rscreen->rw, bo);
-		radeon_bo_decref(rscreen->rw, bo);
+		radeon_ws_bo_unmap(rscreen->rw, bo);
+		radeon_ws_bo_reference(rscreen->rw, &bo, NULL);
 		return;
 	}
-	radeon_bo_unmap(rscreen->rw, bo);
+	radeon_ws_bo_unmap(rscreen->rw, bo);
 
 	radeon_state_init(rstate, rscreen->rw, R600_STATE_SHADER, 0, R600_SHADER_PS);
 
diff --git a/src/gallium/drivers/r600/r600_buffer.c b/src/gallium/drivers/r600/r600_buffer.c
index 06197d3d7d..37abf42d34 100644
--- a/src/gallium/drivers/r600/r600_buffer.c
+++ b/src/gallium/drivers/r600/r600_buffer.c
@@ -68,7 +68,7 @@ struct pipe_resource *r600_buffer_create(struct pipe_screen *screen,
 {
 	struct r600_screen *rscreen = r600_screen(screen);
 	struct r600_resource *rbuffer;
-	struct radeon_bo *bo;
+	struct radeon_ws_bo *bo;
 	struct pb_desc desc;
 	/* XXX We probably want a different alignment for buffers and textures. */
 	unsigned alignment = 4096;
@@ -81,7 +81,7 @@ struct pipe_resource *r600_buffer_create(struct pipe_screen *screen,
 	pipe_reference_init(&rbuffer->base.b.reference, 1);
 	rbuffer->base.b.screen = screen;
 	rbuffer->base.vtbl = &r600_buffer_vtbl;
-
+	rbuffer->size = rbuffer->base.b.width0;
 	if ((rscreen->use_mem_constant == FALSE) && (rbuffer->base.b.bind & PIPE_BIND_CONSTANT_BUFFER)) {
 		desc.alignment = alignment;
 		desc.usage = rbuffer->base.b.bind;
@@ -94,7 +94,7 @@ struct pipe_resource *r600_buffer_create(struct pipe_screen *screen,
 		return &rbuffer->base.b;
 	}
 	rbuffer->domain = r600_domain_from_usage(rbuffer->base.b.bind);
-	bo = radeon_bo(rscreen->rw, 0, rbuffer->base.b.width0, alignment, NULL);
+	bo = radeon_ws_bo(rscreen->rw, rbuffer->base.b.width0, alignment);
 	if (bo == NULL) {
 		FREE(rbuffer);
 		return NULL;
@@ -110,6 +110,7 @@ struct pipe_resource *r600_user_buffer_create(struct pipe_screen *screen,
 	struct r600_resource *rbuffer;
 	struct r600_screen *rscreen = r600_screen(screen);
 	struct pipe_resource templ;
+	void *data;
 
 	memset(&templ, 0, sizeof(struct pipe_resource));
 	templ.target = PIPE_BUFFER;
@@ -124,9 +125,9 @@ struct pipe_resource *r600_user_buffer_create(struct pipe_screen *screen,
 	if (rbuffer == NULL) {
 		return NULL;
 	}
-	radeon_bo_map(rscreen->rw, rbuffer->bo);
-	memcpy(rbuffer->bo->data, ptr, bytes);
-	radeon_bo_unmap(rscreen->rw, rbuffer->bo);
+	data = radeon_ws_bo_map(rscreen->rw, rbuffer->bo);
+	memcpy(data, ptr, bytes);
+	radeon_ws_bo_unmap(rscreen->rw, rbuffer->bo);
 	return &rbuffer->base.b;
 }
 
@@ -142,7 +143,7 @@ static void r600_buffer_destroy(struct pipe_screen *screen,
 		rbuffer->pb = NULL;
 	}
 	if (rbuffer->bo) {
-		radeon_bo_decref(rscreen->rw, rbuffer->bo);
+		radeon_ws_bo_reference(rscreen->rw, &rbuffer->bo, NULL);
 	}
 	memset(rbuffer, 0, sizeof(struct r600_resource));
 	FREE(rbuffer);
@@ -154,6 +155,7 @@ static void *r600_buffer_transfer_map(struct pipe_context *pipe,
 	struct r600_resource *rbuffer = (struct r600_resource*)transfer->resource;
 	struct r600_screen *rscreen = r600_screen(pipe->screen);
 	int write = 0;
+	uint8_t *data;
 
 	if (rbuffer->pb) {
 		return (uint8_t*)pb_map(rbuffer->pb, transfer->usage, NULL) + transfer->box.x;
@@ -164,10 +166,11 @@ static void *r600_buffer_transfer_map(struct pipe_context *pipe,
 	if (transfer->usage & PIPE_TRANSFER_WRITE) {
 		write = 1;
 	}
-	if (radeon_bo_map(rscreen->rw, rbuffer->bo)) {
+	data = radeon_ws_bo_map(rscreen->rw, rbuffer->bo);
+	if (!data)
 		return NULL;
-	}
-	return (uint8_t*)rbuffer->bo->data + transfer->box.x;
+
+	return (uint8_t*)data + transfer->box.x;
 }
 
 static void r600_buffer_transfer_unmap(struct pipe_context *pipe,
@@ -179,7 +182,7 @@ static void r600_buffer_transfer_unmap(struct pipe_context *pipe,
 	if (rbuffer->pb) {
 		pb_unmap(rbuffer->pb);
 	} else {
-		radeon_bo_unmap(rscreen->rw, rbuffer->bo);
+		radeon_ws_bo_unmap(rscreen->rw, rbuffer->bo);
 	}
 }
 
@@ -202,16 +205,16 @@ struct pipe_resource *r600_buffer_from_handle(struct pipe_screen *screen,
 {
 	struct radeon *rw = (struct radeon*)screen->winsys;
 	struct r600_resource *rbuffer;
-	struct radeon_bo *bo = NULL;
+	struct radeon_ws_bo *bo = NULL;
 
-	bo = radeon_bo(rw, whandle->handle, 0, 0, NULL);
+	bo = radeon_ws_bo_handle(rw, whandle->handle);
 	if (bo == NULL) {
 		return NULL;
 	}
 
 	rbuffer = CALLOC_STRUCT(r600_resource);
 	if (rbuffer == NULL) {
-		radeon_bo_decref(rw, bo);
+		radeon_ws_bo_reference(rw, &bo, NULL);
 		return NULL;
 	}
 
diff --git a/src/gallium/drivers/r600/r600_context.h b/src/gallium/drivers/r600/r600_context.h
index 73037fdb1b..f82c8f82fe 100644
--- a/src/gallium/drivers/r600/r600_context.h
+++ b/src/gallium/drivers/r600/r600_context.h
@@ -49,7 +49,7 @@ struct r600_query {
 	boolean					flushed;
 	unsigned				state;
 	/* The buffer where query results are stored. */
-	struct radeon_bo			*buffer;
+	struct radeon_ws_bo			*buffer;
 	unsigned				buffer_size;
 	/* linked list of queries */
 	struct list_head			list;
@@ -103,7 +103,7 @@ struct r600_context_state {
 	unsigned			type;
 	struct radeon_state		rstate[R600_MAX_RSTATE];
 	struct r600_shader		shader;
-	struct radeon_bo		*bo;
+	struct radeon_ws_bo		*bo;
 	unsigned			nrstate;
 };
 
diff --git a/src/gallium/drivers/r600/r600_hw_states.c b/src/gallium/drivers/r600/r600_hw_states.c
index b7070d7d6e..1974b20d86 100644
--- a/src/gallium/drivers/r600/r600_hw_states.c
+++ b/src/gallium/drivers/r600/r600_hw_states.c
@@ -120,7 +120,7 @@ static void r600_cb(struct r600_context *rctx, struct radeon_state *rstate,
 	radeon_state_init(rstate, rscreen->rw, R600_STATE_CB0 + cb, 0, 0);
 	rtex = (struct r600_resource_texture*)state->cbufs[cb]->texture;
 	rbuffer = &rtex->resource;
-	rstate->bo[0] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
+	radeon_ws_bo_reference(rscreen->rw, &rstate->bo[0], rbuffer->bo);
 	rstate->placement[0] = RADEON_GEM_DOMAIN_GTT;
 	rstate->nbo = 1;
 	pitch = (rtex->pitch[level] / rtex->bpt) / 8 - 1;
@@ -171,7 +171,7 @@ static void r600_db(struct r600_context *rctx, struct radeon_state *rstate,
 	rtex->depth = 1;
 	rbuffer = &rtex->resource;
 
-	rstate->bo[0] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
+	radeon_ws_bo_reference(rscreen->rw, &rstate->bo[0], rbuffer->bo);
 	rstate->nbo = 1;
 	rstate->placement[0] = RADEON_GEM_DOMAIN_VRAM;
 	level = state->zsbuf->level;
@@ -537,11 +537,11 @@ static void r600_resource(struct pipe_context *ctx, struct radeon_state *rstate,
 		if (r) {
 			return;
 		}
-		rstate->bo[0] = radeon_bo_incref(rscreen->rw, tmp->uncompressed);
-		rstate->bo[1] = radeon_bo_incref(rscreen->rw, tmp->uncompressed);
+		radeon_ws_bo_reference(rscreen->rw, &rstate->bo[0], tmp->uncompressed);
+		radeon_ws_bo_reference(rscreen->rw, &rstate->bo[1], tmp->uncompressed);
 	} else {
-		rstate->bo[0] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
-		rstate->bo[1] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
+		radeon_ws_bo_reference(rscreen->rw, &rstate->bo[0], rbuffer->bo);
+		radeon_ws_bo_reference(rscreen->rw, &rstate->bo[1], rbuffer->bo);
 	}
 	rstate->nbo = 2;
 	rstate->placement[0] = RADEON_GEM_DOMAIN_GTT;
@@ -873,10 +873,10 @@ static int r600_vs_resource(struct r600_context *rctx, int id, struct r600_resou
 	struct r600_screen *rscreen = rctx->screen;
 
 	radeon_state_init(vs_resource, rscreen->rw, R600_STATE_RESOURCE, id, R600_SHADER_VS);
-	vs_resource->bo[0] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
+	radeon_ws_bo_reference(rscreen->rw, &vs_resource->bo[0], rbuffer->bo);
 	vs_resource->nbo = 1;
 	vs_resource->states[R600_PS_RESOURCE__RESOURCE0_WORD0] = offset;
-	vs_resource->states[R600_PS_RESOURCE__RESOURCE0_WORD1] = rbuffer->bo->size - offset - 1;
+	vs_resource->states[R600_PS_RESOURCE__RESOURCE0_WORD1] = rbuffer->size - offset - 1;
 	vs_resource->states[R600_PS_RESOURCE__RESOURCE0_WORD2] = S_038008_STRIDE(stride) |
 		S_038008_DATA_FORMAT(format);
 	vs_resource->states[R600_PS_RESOURCE__RESOURCE0_WORD3] = 0x00000000;
@@ -899,7 +899,7 @@ static int r600_draw_vgt_init(struct r600_draw *draw,
 	draw->draw.states[R600_DRAW__VGT_DRAW_INITIATOR] = vgt_draw_initiator;
 	draw->draw.states[R600_DRAW__VGT_DMA_BASE] = draw->index_buffer_offset;
 	if (rbuffer) {
-		draw->draw.bo[0] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
+		radeon_ws_bo_reference(rscreen->rw, &draw->draw.bo[0], rbuffer->bo);
 		draw->draw.placement[0] = RADEON_GEM_DOMAIN_GTT;
 		draw->draw.placement[1] = RADEON_GEM_DOMAIN_GTT;
 		draw->draw.nbo = 1;
@@ -980,7 +980,7 @@ static int r600_ps_shader(struct r600_context *rctx, struct r600_context_state *
 	state->states[R600_PS_SHADER__SQ_PGM_RESOURCES_PS] = S_028868_NUM_GPRS(rshader->bc.ngpr) |
 		S_028868_STACK_SIZE(rshader->bc.nstack);
 	state->states[R600_PS_SHADER__SQ_PGM_EXPORTS_PS] = exports_ps;
-	state->bo[0] = radeon_bo_incref(rscreen->rw, rpshader->bo);
+	radeon_ws_bo_reference(rscreen->rw, &state->bo[0], rpshader->bo);
 	state->nbo = 1;
 	state->placement[0] = RADEON_GEM_DOMAIN_GTT;
 	return radeon_state_pm4(state);
@@ -1005,8 +1005,8 @@ static int r600_vs_shader(struct r600_context *rctx, struct r600_context_state *
 	state->states[R600_VS_SHADER__SPI_VS_OUT_CONFIG] = S_0286C4_VS_EXPORT_COUNT(rshader->noutput - 2);
 	state->states[R600_VS_SHADER__SQ_PGM_RESOURCES_VS] = S_028868_NUM_GPRS(rshader->bc.ngpr) |
 		S_028868_STACK_SIZE(rshader->bc.nstack);
-	state->bo[0] = radeon_bo_incref(rscreen->rw, rpshader->bo);
-	state->bo[1] = radeon_bo_incref(rscreen->rw, rpshader->bo);
+	radeon_ws_bo_reference(rscreen->rw, &state->bo[0], rpshader->bo);
+	radeon_ws_bo_reference(rscreen->rw, &state->bo[1], rpshader->bo);
 	state->nbo = 2;
 	state->placement[0] = RADEON_GEM_DOMAIN_GTT;
 	state->placement[2] = RADEON_GEM_DOMAIN_GTT;
@@ -1070,18 +1070,18 @@ static void r600_texture_state_cb(struct r600_screen *rscreen, struct r600_resou
 	format = r600_translate_colorformat(rtexture->resource.base.b.format);
 	swap = r600_translate_colorswap(rtexture->resource.base.b.format);
 	if (desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS) {
-		rstate->bo[0] = radeon_bo_incref(rscreen->rw, rtexture->uncompressed);
-		rstate->bo[1] = radeon_bo_incref(rscreen->rw, rtexture->uncompressed);
-		rstate->bo[2] = radeon_bo_incref(rscreen->rw, rtexture->uncompressed);
+		radeon_ws_bo_reference(rscreen->rw, &rstate->bo[0], rtexture->uncompressed);
+		radeon_ws_bo_reference(rscreen->rw, &rstate->bo[1], rtexture->uncompressed);
+		radeon_ws_bo_reference(rscreen->rw, &rstate->bo[2], rtexture->uncompressed);
 		rstate->placement[0] = RADEON_GEM_DOMAIN_GTT;
 		rstate->placement[2] = RADEON_GEM_DOMAIN_GTT;
 		rstate->placement[4] = RADEON_GEM_DOMAIN_GTT;
 		rstate->nbo = 3;
 		color_info = 0;
 	} else {
-		rstate->bo[0] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
-		rstate->bo[1] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
-		rstate->bo[2] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
+		radeon_ws_bo_reference(rscreen->rw, &rstate->bo[0], rbuffer->bo);
+		radeon_ws_bo_reference(rscreen->rw, &rstate->bo[1], rbuffer->bo);
+		radeon_ws_bo_reference(rscreen->rw, &rstate->bo[2], rbuffer->bo);
 		rstate->placement[0] = RADEON_GEM_DOMAIN_GTT;
 		rstate->placement[2] = RADEON_GEM_DOMAIN_GTT;
 		rstate->placement[4] = RADEON_GEM_DOMAIN_GTT;
@@ -1126,7 +1126,7 @@ static void r600_texture_state_db(struct r600_screen *rscreen, struct r600_resou
 	rstate->states[R600_DB__DB_PREFETCH_LIMIT] = (rtexture->height[level] / 8) -1;
 	rstate->states[R600_DB__DB_DEPTH_SIZE] = S_028000_PITCH_TILE_MAX(pitch) |
 						S_028000_SLICE_TILE_MAX(slice);
-	rstate->bo[0] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
+	radeon_ws_bo_reference(rscreen->rw, &rstate->bo[0], rbuffer->bo);
 	rstate->placement[0] = RADEON_GEM_DOMAIN_GTT;
 	rstate->nbo = 1;
 
@@ -1262,7 +1262,7 @@ void r600_set_constant_buffer_mem(struct pipe_context *ctx,
 	rstate->states[R600_VS_CBUF__ALU_CONST_BUFFER_SIZE_VS_0] = size;
 	rstate->states[R600_VS_CBUF__ALU_CONST_CACHE_VS_0] = 0;
 
-	rstate->bo[0] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
+	radeon_ws_bo_reference(rscreen->rw, &rstate->bo[0], rbuffer->bo);
 	rstate->nbo = 1;
 	rstate->placement[0] = RADEON_GEM_DOMAIN_VRAM;
 	if (radeon_state_pm4(rstate))
diff --git a/src/gallium/drivers/r600/r600_query.c b/src/gallium/drivers/r600/r600_query.c
index 68358f9dd7..922d7ace13 100644
--- a/src/gallium/drivers/r600/r600_query.c
+++ b/src/gallium/drivers/r600/r600_query.c
@@ -39,7 +39,7 @@ static void r600_query_begin(struct r600_context *rctx, struct r600_query *rquer
 	radeon_state_fini(rstate);
 	radeon_state_init(rstate, rscreen->rw, R600_STATE_QUERY_BEGIN, 0, 0);
 	rstate->states[R600_QUERY__OFFSET] = rquery->num_results;
-	rstate->bo[0] = radeon_bo_incref(rscreen->rw, rquery->buffer);
+	radeon_ws_bo_reference(rscreen->rw, &rstate->bo[0], rquery->buffer);
 	rstate->nbo = 1;
 	rstate->placement[0] = RADEON_GEM_DOMAIN_GTT;
 	if (radeon_state_pm4(rstate)) {
@@ -55,7 +55,7 @@ static void r600_query_end(struct r600_context *rctx, struct r600_query *rquery)
 	radeon_state_fini(rstate);
 	radeon_state_init(rstate, rscreen->rw, R600_STATE_QUERY_END, 0, 0);
 	rstate->states[R600_QUERY__OFFSET] = rquery->num_results + 8;
-	rstate->bo[0] = radeon_bo_incref(rscreen->rw, rquery->buffer);
+	radeon_ws_bo_reference(rscreen->rw, &rstate->bo[0], rquery->buffer);
 	rstate->nbo = 1;
 	rstate->placement[0] = RADEON_GEM_DOMAIN_GTT;
 	if (radeon_state_pm4(rstate)) {
@@ -79,7 +79,7 @@ static struct pipe_query *r600_create_query(struct pipe_context *ctx, unsigned q
 	q->type = query_type;
 	q->buffer_size = 4096;
 
-	q->buffer = radeon_bo(rscreen->rw, 0, q->buffer_size, 1, NULL);
+	q->buffer = radeon_ws_bo(rscreen->rw, q->buffer_size, 1);
 	if (!q->buffer) {
 		FREE(q);
 		return NULL;
@@ -96,7 +96,7 @@ static void r600_destroy_query(struct pipe_context *ctx,
 	struct r600_screen *rscreen = r600_screen(ctx->screen);
 	struct r600_query *q = r600_query(query);
 
-	radeon_bo_decref(rscreen->rw, q->buffer);
+	radeon_ws_bo_reference(rscreen->rw, &q->buffer, NULL);
 	LIST_DEL(&q->list);
 	FREE(query);
 }
@@ -108,9 +108,8 @@ static void r600_query_result(struct pipe_context *ctx, struct r600_query *rquer
 	u32 *results;
 	int i;
 
-	radeon_bo_wait(rscreen->rw, rquery->buffer);
-	radeon_bo_map(rscreen->rw, rquery->buffer);
-	results = rquery->buffer->data;
+	radeon_ws_bo_wait(rscreen->rw, rquery->buffer);
+	results = radeon_ws_bo_map(rscreen->rw, rquery->buffer);
 	for (i = 0; i < rquery->num_results; i += 4) {
 		start = (u64)results[i] | (u64)results[i + 1] << 32;
 		end = (u64)results[i + 2] | (u64)results[i + 3] << 32;
@@ -118,7 +117,7 @@ static void r600_query_result(struct pipe_context *ctx, struct r600_query *rquer
 			rquery->result += end - start;
 		}
 	}
-	radeon_bo_unmap(rscreen->rw, rquery->buffer);
+	radeon_ws_bo_unmap(rscreen->rw, rquery->buffer);
 	rquery->num_results = 0;
 }
 
diff --git a/src/gallium/drivers/r600/r600_resource.h b/src/gallium/drivers/r600/r600_resource.h
index 129667ad20..8a110b1b72 100644
--- a/src/gallium/drivers/r600/r600_resource.h
+++ b/src/gallium/drivers/r600/r600_resource.h
@@ -34,10 +34,11 @@ struct r600_screen;
  */
 struct r600_resource {
 	struct u_resource		base;
-	struct radeon_bo		*bo;
+	struct radeon_ws_bo		*bo;
 	u32				domain;
 	u32				flink;
 	struct pb_buffer		*pb;
+	u32				size;
 };
 
 struct r600_resource_texture {
@@ -55,7 +56,7 @@ struct r600_resource_texture {
 	unsigned			tile_type;
 	unsigned			depth;
 	unsigned			dirty;
-	struct radeon_bo		*uncompressed;
+	struct radeon_ws_bo		*uncompressed;
 	struct radeon_state		scissor[PIPE_MAX_TEXTURE_LEVELS];
 	struct radeon_state		cb[8][PIPE_MAX_TEXTURE_LEVELS];
 	struct radeon_state		db[PIPE_MAX_TEXTURE_LEVELS];
diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index ad19238697..10f6d016a3 100644
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -157,18 +157,19 @@ static int r600_pipe_shader(struct pipe_context *ctx, struct r600_context_state
 	struct r600_context *rctx = r600_context(ctx);
 	struct r600_shader *rshader = &rpshader->shader;
 	int r;
+	void *data;
 
 	/* copy new shader */
-	radeon_bo_decref(rscreen->rw, rpshader->bo);
+	radeon_ws_bo_reference(rscreen->rw, &rpshader->bo, NULL);
 	rpshader->bo = NULL;
-	rpshader->bo = radeon_bo(rscreen->rw, 0, rshader->bc.ndw * 4,
-				4096, NULL);
+	rpshader->bo = radeon_ws_bo(rscreen->rw, rshader->bc.ndw * 4,
+				    4096);
 	if (rpshader->bo == NULL) {
 		return -ENOMEM;
 	}
-	radeon_bo_map(rscreen->rw, rpshader->bo);
-	memcpy(rpshader->bo->data, rshader->bc.bytecode, rshader->bc.ndw * 4);
-	radeon_bo_unmap(rscreen->rw, rpshader->bo);
+	data = radeon_ws_bo_map(rscreen->rw, rpshader->bo);
+	memcpy(data, rshader->bc.bytecode, rshader->bc.ndw * 4);
+	radeon_ws_bo_unmap(rscreen->rw, rpshader->bo);
 	/* build state */
 	rshader->flat_shade = rctx->flat_shade;
 	switch (rshader->processor_type) {
diff --git a/src/gallium/drivers/r600/r600_state.c b/src/gallium/drivers/r600/r600_state.c
index 5a4a72d64f..5d6236206f 100644
--- a/src/gallium/drivers/r600/r600_state.c
+++ b/src/gallium/drivers/r600/r600_state.c
@@ -613,7 +613,7 @@ static int setup_cb_flush(struct r600_context *rctx, struct radeon_state *flush)
 		rtex = (struct r600_resource_texture*)surf->texture;
 		rbuffer = &rtex->resource;
 		/* just need to the bo to the flush list */
-		flush->bo[i] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
+		radeon_ws_bo_reference(rscreen->rw, &flush->bo[i], rbuffer->bo);
 		flush->placement[i] = RADEON_GEM_DOMAIN_VRAM;
 	}
 	flush->nbo = rctx->framebuffer->state.framebuffer.nr_cbufs;
@@ -636,7 +636,7 @@ static int setup_db_flush(struct r600_context *rctx, struct radeon_state *flush)
 	rtex = (struct r600_resource_texture*)surf->texture;
 	rbuffer = &rtex->resource;
 	/* just need to the bo to the flush list */
-	flush->bo[0] = radeon_bo_incref(rscreen->rw, rbuffer->bo);
+	radeon_ws_bo_reference(rscreen->rw, &flush->bo[0], rbuffer->bo);
 	flush->placement[0] = RADEON_GEM_DOMAIN_VRAM;
 
 	flush->nbo = 1;
diff --git a/src/gallium/drivers/r600/r600_texture.c b/src/gallium/drivers/r600/r600_texture.c
index 80cfa36ac0..4fa8cf4709 100644
--- a/src/gallium/drivers/r600/r600_texture.c
+++ b/src/gallium/drivers/r600/r600_texture.c
@@ -120,7 +120,8 @@ struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
 
 	/* FIXME alignment 4096 enought ? too much ? */
 	resource->domain = r600_domain_from_usage(resource->base.b.bind);
-	resource->bo = radeon_bo(radeon, 0, rtex->size, 4096, NULL);
+	resource->size = rtex->size;
+	resource->bo = radeon_ws_bo(radeon, rtex->size, 4096);
 	if (resource->bo == NULL) {
 		FREE(rtex);
 		return NULL;
@@ -149,10 +150,10 @@ static void r600_texture_destroy(struct pipe_screen *screen,
 	struct radeon *radeon = (struct radeon *)screen->winsys;
 
 	if (resource->bo) {
-		radeon_bo_decref(radeon, resource->bo);
+		radeon_ws_bo_reference(radeon, &resource->bo, NULL);
 	}
 	if (rtex->uncompressed) {
-		radeon_bo_decref(radeon, rtex->uncompressed);
+		radeon_ws_bo_reference(radeon, &rtex->uncompressed, NULL);
 	}
 	r600_texture_destroy_state(ptex);
 	FREE(rtex);
@@ -197,7 +198,7 @@ struct pipe_resource *r600_texture_from_handle(struct pipe_screen *screen,
 	struct radeon *rw = (struct radeon*)screen->winsys;
 	struct r600_resource_texture *rtex;
 	struct r600_resource *resource;
-	struct radeon_bo *bo = NULL;
+	struct radeon_ws_bo *bo = NULL;
 
 	/* Support only 2D textures without mipmaps */
 	if ((templ->target != PIPE_TEXTURE_2D && templ->target != PIPE_TEXTURE_RECT) ||
@@ -208,7 +209,7 @@ struct pipe_resource *r600_texture_from_handle(struct pipe_screen *screen,
 	if (rtex == NULL)
 		return NULL;
 
-	bo = radeon_bo(rw, whandle->handle, 0, 0, NULL);
+	bo = radeon_ws_bo_handle(rw, whandle->handle);
 	if (bo == NULL) {
 		FREE(rtex);
 		return NULL;
@@ -316,7 +317,7 @@ void* r600_texture_transfer_map(struct pipe_context *ctx,
 {
 	struct r600_screen *rscreen = r600_screen(ctx->screen);
 	struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
-	struct radeon_bo *bo;
+	struct radeon_ws_bo *bo;
 	enum pipe_format format = transfer->resource->format;
 	struct radeon *radeon = (struct radeon *)ctx->screen->winsys;
 	struct r600_resource_texture *rtex;
@@ -343,12 +344,12 @@ void* r600_texture_transfer_map(struct pipe_context *ctx,
 			transfer->box.y / util_format_get_blockheight(format) * transfer->stride +
 			transfer->box.x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
 	}
-	if (radeon_bo_map(radeon, bo)) {
+	map = radeon_ws_bo_map(radeon, bo);
+	if (!map) {
 		return NULL;
 	}
-	radeon_bo_wait(radeon, bo);
+	radeon_ws_bo_wait(radeon, bo);
 
-	map = bo->data;
 	return map + offset;
 }
 
@@ -358,7 +359,7 @@ void r600_texture_transfer_unmap(struct pipe_context *ctx,
 	struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
 	struct radeon *radeon = (struct radeon *)ctx->screen->winsys;
 	struct r600_resource_texture *rtex;
-	struct radeon_bo *bo;
+	struct radeon_ws_bo *bo;
 
 	if (rtransfer->linear_texture) {
 		bo = ((struct r600_resource *)rtransfer->linear_texture)->bo;
@@ -370,7 +371,7 @@ void r600_texture_transfer_unmap(struct pipe_context *ctx,
 			bo = ((struct r600_resource *)transfer->resource)->bo;
 		}
 	}
-	radeon_bo_unmap(radeon, bo);
+	radeon_ws_bo_unmap(radeon, bo);
 }
 
 struct u_resource_vtbl r600_texture_vtbl =
@@ -654,7 +655,7 @@ int r600_texture_from_depth(struct pipe_context *ctx, struct r600_resource_textu
 
 	/* allocate uncompressed texture */
 	if (rtexture->uncompressed == NULL) {
-		rtexture->uncompressed = radeon_bo(rscreen->rw, 0, rtexture->size, 4096, NULL);
+		rtexture->uncompressed = radeon_ws_bo(rscreen->rw, rtexture->size, 4096);
 		if (rtexture->uncompressed == NULL) {
 			return -ENOMEM;
 		}
diff --git a/src/gallium/drivers/r600/radeon.h b/src/gallium/drivers/r600/radeon.h
index 12e8b993c8..be28ad19ff 100644
--- a/src/gallium/drivers/r600/radeon.h
+++ b/src/gallium/drivers/r600/radeon.h
@@ -87,24 +87,17 @@ enum {
 
 enum radeon_family radeon_get_family(struct radeon *rw);
 
-/*
- * radeon object functions
- */
-struct radeon_bo {
-	unsigned			refcount;
-	unsigned			handle;
-	unsigned			size;
-	unsigned			alignment;
-	unsigned			map_count;
-	void				*data;
-};
-struct radeon_bo *radeon_bo(struct radeon *radeon, unsigned handle,
-			unsigned size, unsigned alignment, void *ptr);
-int radeon_bo_map(struct radeon *radeon, struct radeon_bo *bo);
-void radeon_bo_unmap(struct radeon *radeon, struct radeon_bo *bo);
-struct radeon_bo *radeon_bo_incref(struct radeon *radeon, struct radeon_bo *bo);
-struct radeon_bo *radeon_bo_decref(struct radeon *radeon, struct radeon_bo *bo);
-int radeon_bo_wait(struct radeon *radeon, struct radeon_bo *bo);
+/* lowlevel WS bo */
+struct radeon_ws_bo;
+struct radeon_ws_bo *radeon_ws_bo(struct radeon *radeon,
+				  unsigned size, unsigned alignment);
+struct radeon_ws_bo *radeon_ws_bo_handle(struct radeon *radeon,
+					 unsigned handle);
+void *radeon_ws_bo_map(struct radeon *radeon, struct radeon_ws_bo *bo);
+void radeon_ws_bo_unmap(struct radeon *radeon, struct radeon_ws_bo *bo);
+void radeon_ws_bo_reference(struct radeon *radeon, struct radeon_ws_bo **dst,
+			    struct radeon_ws_bo *src);
+int radeon_ws_bo_wait(struct radeon *radeon, struct radeon_ws_bo *bo);
 
 struct radeon_stype_info;
 /*
@@ -124,7 +117,7 @@ struct radeon_state {
 	u32				pm4_crc;
 	u32				pm4[128];
 	unsigned			nbo;
-	struct radeon_bo		*bo[4];
+	struct radeon_ws_bo		*bo[4];
 	unsigned			nreloc;
 	unsigned			reloc_pm4_id[8];
 	unsigned			reloc_bo_id[8];
diff --git a/src/gallium/winsys/r600/drm/Makefile b/src/gallium/winsys/r600/drm/Makefile
index e3b943c4d4..86688a213c 100644
--- a/src/gallium/winsys/r600/drm/Makefile
+++ b/src/gallium/winsys/r600/drm/Makefile
@@ -13,7 +13,8 @@ C_SOURCES = \
 	radeon_bo.c \
 	radeon_pciid.c \
 	radeon.c \
-	r600_drm.c
+	r600_drm.c \
+	radeon_ws_bo.c
 
 LIBRARY_INCLUDES = -I$(TOP)/src/gallium/drivers/r600 \
 		   $(shell pkg-config libdrm --cflags-only-I)
diff --git a/src/gallium/winsys/r600/drm/r600_drm.c b/src/gallium/winsys/r600/drm/r600_drm.c
index c76e7f5fa5..bbf53fcbdc 100644
--- a/src/gallium/winsys/r600/drm/r600_drm.c
+++ b/src/gallium/winsys/r600/drm/r600_drm.c
@@ -39,26 +39,3 @@ struct radeon *r600_drm_winsys_create(int drmfd)
 	return radeon_new(drmfd, 0);
 }
 
-boolean r600_buffer_get_handle(struct radeon *rw,
-			       struct pipe_resource *buf,
-			       struct winsys_handle *whandle)
-{
-	struct drm_gem_flink flink;
-	struct r600_resource* rbuffer = (struct r600_resource*)buf;
-
-	if (whandle->type == DRM_API_HANDLE_TYPE_SHARED) {
-		if (!rbuffer->flink) {
-			flink.handle = rbuffer->bo->handle;
-
-			if (ioctl(rw->fd, DRM_IOCTL_GEM_FLINK, &flink)) {
-				return FALSE;
-			}
-
-			rbuffer->flink = flink.name;
-		}
-		whandle->handle = rbuffer->flink;
-	} else if (whandle->type == DRM_API_HANDLE_TYPE_KMS) {
-		whandle->handle = rbuffer->bo->handle;
-	}
-	return TRUE;
-}
diff --git a/src/gallium/winsys/r600/drm/r600_state.c b/src/gallium/winsys/r600/drm/r600_state.c
index b8ef89dfd6..82eb0cc863 100644
--- a/src/gallium/winsys/r600/drm/r600_state.c
+++ b/src/gallium/winsys/r600/drm/r600_state.c
@@ -155,7 +155,7 @@ static int r600_state_pm4_bytecode(struct radeon_state *state, unsigned offset,
 				r = radeon_state_reloc(state, state->cpm4, regs[id + i].bo_id);
 				if (r)
 					return r;
-				state->pm4[state->cpm4++] = state->bo[regs[id + i].bo_id]->handle;
+				state->pm4[state->cpm4++] = state->bo[regs[id + i].bo_id]->bo->handle;
 			}
 		}
 		return 0;
@@ -172,7 +172,7 @@ static int r600_state_pm4_bytecode(struct radeon_state *state, unsigned offset,
 				r = radeon_state_reloc(state, state->cpm4, regs[id + i].bo_id);
 				if (r)
 					return r;
-				state->pm4[state->cpm4++] = state->bo[regs[id + i].bo_id]->handle;
+				state->pm4[state->cpm4++] = state->bo[regs[id + i].bo_id]->bo->handle;
 			}
 		}
 		return 0;
@@ -220,7 +220,7 @@ static int eg_state_pm4_bytecode(struct radeon_state *state, unsigned offset, un
 				r = radeon_state_reloc(state, state->cpm4, regs[id + i].bo_id);
 				if (r)
 					return r;
-				state->pm4[state->cpm4++] = state->bo[regs[id + i].bo_id]->handle;
+				state->pm4[state->cpm4++] = state->bo[regs[id + i].bo_id]->bo->handle;
 			}
 		}
 		return 0;
@@ -237,7 +237,7 @@ static int eg_state_pm4_bytecode(struct radeon_state *state, unsigned offset, un
 				r = radeon_state_reloc(state, state->cpm4, regs[id + i].bo_id);
 				if (r)
 					return r;
-				state->pm4[state->cpm4++] = state->bo[regs[id + i].bo_id]->handle;
+				state->pm4[state->cpm4++] = state->bo[regs[id + i].bo_id]->bo->handle;
 			}
 		}
 		return 0;
@@ -318,7 +318,7 @@ static void r600_state_pm4_with_flush(struct radeon_state *state, u32 flags, int
 		}
 	}
 	for (i = 0; i < state->nreloc; i++) {
-		size = (state->bo[state->reloc_bo_id[i]]->size + 255) >> 8;
+		size = (state->bo[state->reloc_bo_id[i]]->bo->size + 255) >> 8;
 		state->pm4[state->cpm4++] = PKT3(PKT3_SURFACE_SYNC, 3);
 		if (bufs_are_cbs)
 			flags |= S_0085F0_CB0_DEST_BASE_ENA(1 << i);
@@ -328,7 +328,7 @@ static void r600_state_pm4_with_flush(struct radeon_state *state, u32 flags, int
 		state->pm4[state->cpm4++] = 0x0000000A;
 		state->pm4[state->cpm4++] = PKT3(PKT3_NOP, 0);
 		state->reloc_pm4_id[i] = state->cpm4;
-		state->pm4[state->cpm4++] = state->bo[state->reloc_bo_id[i]]->handle;
+		state->pm4[state->cpm4++] = state->bo[state->reloc_bo_id[i]]->bo->handle;
 	}
 }
 
@@ -384,7 +384,7 @@ static int r600_state_pm4_query_begin(struct radeon_state *state)
 	r = radeon_state_reloc(state, state->cpm4, 0);
 	if (r)
 		return r;
-	state->pm4[state->cpm4++] = state->bo[0]->handle;
+	state->pm4[state->cpm4++] = state->bo[0]->bo->handle;
 	return 0;
 }
 
@@ -401,7 +401,7 @@ static int r600_state_pm4_query_end(struct radeon_state *state)
 	r = radeon_state_reloc(state, state->cpm4, 0);
 	if (r)
 		return r;
-	state->pm4[state->cpm4++] = state->bo[0]->handle;
+	state->pm4[state->cpm4++] = state->bo[0]->bo->handle;
 	return 0;
 }
 
@@ -486,7 +486,7 @@ static int r600_state_pm4_draw(struct radeon_state *state)
 		r = radeon_state_reloc(state, state->cpm4, 0);
 		if (r)
 			return r;
-		state->pm4[state->cpm4++] = state->bo[0]->handle;
+		state->pm4[state->cpm4++] = state->bo[0]->bo->handle;
 	} else {
 		state->pm4[state->cpm4++] = PKT3(PKT3_DRAW_INDEX_AUTO, 1);
 		state->pm4[state->cpm4++] = state->states[R600_DRAW__VGT_NUM_INDICES];
@@ -572,13 +572,13 @@ static int r600_state_pm4_resource(struct radeon_state *state)
 	r = radeon_state_reloc(state, state->cpm4, 0);
 	if (r)
 		return r;
-	state->pm4[state->cpm4++] = state->bo[0]->handle;
+	state->pm4[state->cpm4++] = state->bo[0]->bo->handle;
 	if (type == 2) {
 		state->pm4[state->cpm4++] = PKT3(PKT3_NOP, 0);
 		r = radeon_state_reloc(state, state->cpm4, 1);
 		if (r)
 			return r;
-		state->pm4[state->cpm4++] = state->bo[1]->handle;
+		state->pm4[state->cpm4++] = state->bo[1]->bo->handle;
 	}
 	return 0;
 }
diff --git a/src/gallium/winsys/r600/drm/radeon_bo.c b/src/gallium/winsys/r600/drm/radeon_bo.c
index f79135bfa1..e8604a82db 100644
--- a/src/gallium/winsys/r600/drm/radeon_bo.c
+++ b/src/gallium/winsys/r600/drm/radeon_bo.c
@@ -45,7 +45,7 @@ struct radeon_bo *radeon_bo(struct radeon *radeon, unsigned handle,
 	}
 	bo->size = size;
 	bo->handle = handle;
-	bo->refcount = 1;
+	pipe_reference_init(&bo->reference, 1);
 	bo->alignment = alignment;
 
 	if (handle) {
@@ -82,7 +82,8 @@ struct radeon_bo *radeon_bo(struct radeon *radeon, unsigned handle,
 	if (ptr) {
 		if (radeon_bo_map(radeon, bo)) {
 			fprintf(stderr, "%s failed to copy data into bo\n", __func__);
-			return radeon_bo_decref(radeon, bo);
+			radeon_bo_reference(radeon, &bo, NULL);
+			return bo;
 		}
 		memcpy(bo->data, ptr, size);
 		radeon_bo_unmap(radeon, bo);
@@ -133,31 +134,26 @@ void radeon_bo_unmap(struct radeon *radeon, struct radeon_bo *bo)
 	bo->data = NULL;
 }
 
-struct radeon_bo *radeon_bo_incref(struct radeon *radeon, struct radeon_bo *bo)
-{
-	bo->refcount++;
-	return bo;
-}
-
-struct radeon_bo *radeon_bo_decref(struct radeon *radeon, struct radeon_bo *bo)
+static void radeon_bo_destroy(struct radeon *radeon, struct radeon_bo *bo)
 {
 	struct drm_gem_close args;
 
-	if (bo == NULL)
-		return NULL;
-	if (--bo->refcount > 0) {
-		return NULL;
-	}
-
-	if (bo->map_count) {
-		munmap(bo->data, bo->size);
-	}
 	memset(&args, 0, sizeof(args));
 	args.handle = bo->handle;
 	drmIoctl(radeon->fd, DRM_IOCTL_GEM_CLOSE, &args);
 	memset(bo, 0, sizeof(struct radeon_bo));
 	free(bo);
-	return NULL;
+}
+
+void radeon_bo_reference(struct radeon *radeon,
+			 struct radeon_bo **dst,
+			 struct radeon_bo *src)
+{
+	struct radeon_bo *old = *dst;
+	if (pipe_reference(&(*dst)->reference, &src->reference)) {
+		radeon_bo_destroy(radeon, old);
+	}
+	*dst = src;
 }
 
 int radeon_bo_wait(struct radeon *radeon, struct radeon_bo *bo)
diff --git a/src/gallium/winsys/r600/drm/radeon_ctx.c b/src/gallium/winsys/r600/drm/radeon_ctx.c
index 9f166b209a..d6e58451b8 100644
--- a/src/gallium/winsys/r600/drm/radeon_ctx.c
+++ b/src/gallium/winsys/r600/drm/radeon_ctx.c
@@ -30,24 +30,26 @@
 #include "radeon_drm.h"
 #include "bof.h"
 
-static int radeon_ctx_set_bo_new(struct radeon_ctx *ctx, struct radeon_bo *bo)
+static int radeon_ctx_set_bo_new(struct radeon_ctx *ctx, struct radeon_ws_bo *bo)
 {
 	if (ctx->nbo >= RADEON_CTX_MAX_PM4)
 		return -EBUSY;
-	ctx->bo[ctx->nbo] = radeon_bo_incref(ctx->radeon, bo);
+	radeon_ws_bo_reference(ctx->radeon, &ctx->bo[ctx->nbo], bo);
 	ctx->nbo++;
 	return 0;
 }
 
-static struct radeon_bo *radeon_ctx_get_bo(struct radeon_ctx *ctx, unsigned reloc)
+static struct radeon_ws_bo *radeon_ctx_get_bo(struct radeon_ctx *ctx, unsigned reloc)
 {
 	struct radeon_cs_reloc *greloc;
 	unsigned i;
+	struct radeon_ws_bo *bo;
 
 	greloc = (void *)(((u8 *)ctx->reloc) + reloc * 4);
 	for (i = 0; i < ctx->nbo; i++) {
-		if (ctx->bo[i]->handle == greloc->handle) {
-			return radeon_bo_incref(ctx->radeon, ctx->bo[i]);
+		if (ctx->bo[i]->bo->handle == greloc->handle) {
+			radeon_ws_bo_reference(ctx->radeon, &bo, ctx->bo[i]);
+			return bo;
 		}
 	}
 	fprintf(stderr, "%s no bo for reloc[%d 0x%08X] %d\n", __func__, reloc, greloc->handle, ctx->nbo);
@@ -63,7 +65,7 @@ static void radeon_ctx_get_placement(struct radeon_ctx *ctx, unsigned reloc, u32
 	placement[1] = 0;
 	greloc = (void *)(((u8 *)ctx->reloc) + reloc * 4);
 	for (i = 0; i < ctx->nbo; i++) {
-		if (ctx->bo[i]->handle == greloc->handle) {
+		if (ctx->bo[i]->bo->handle == greloc->handle) {
 			placement[0] = greloc->read_domain | greloc->write_domain;
 			placement[1] = placement[0];
 			return;
@@ -74,7 +76,7 @@ static void radeon_ctx_get_placement(struct radeon_ctx *ctx, unsigned reloc, u32
 void radeon_ctx_clear(struct radeon_ctx *ctx)
 {
 	for (int i = 0; i < ctx->nbo; i++) {
-		ctx->bo[i] = radeon_bo_decref(ctx->radeon, ctx->bo[i]);
+		radeon_ws_bo_reference(ctx->radeon, &ctx->bo[i], NULL);
 	}
 	ctx->ndwords = RADEON_CTX_MAX_PM4;
 	ctx->cdwords = 0;
@@ -116,7 +118,7 @@ void radeon_ctx_fini(struct radeon_ctx *ctx)
 		return;
 
 	for (i = 0; i < ctx->nbo; i++) {
-		ctx->bo[i] = radeon_bo_decref(ctx->radeon, ctx->bo[i]);
+		radeon_ws_bo_reference(ctx->radeon, &ctx->bo[i], NULL);
 	}
 	ctx->radeon = radeon_decref(ctx->radeon);
 	free(ctx->bo);
@@ -178,13 +180,13 @@ int radeon_ctx_submit(struct radeon_ctx *ctx)
 	return r;
 }
 
-static int radeon_ctx_reloc(struct radeon_ctx *ctx, struct radeon_bo *bo,
+static int radeon_ctx_reloc(struct radeon_ctx *ctx, struct radeon_ws_bo *bo,
 			unsigned id, unsigned *placement)
 {
 	unsigned i;
 
 	for (i = 0; i < ctx->nreloc; i++) {
-		if (ctx->reloc[i].handle == bo->handle) {
+		if (ctx->reloc[i].handle == bo->bo->handle) {
 			ctx->pm4[id] = i * sizeof(struct radeon_cs_reloc) / 4;
 			return 0;
 		}
@@ -192,7 +194,7 @@ static int radeon_ctx_reloc(struct radeon_ctx *ctx, struct radeon_bo *bo,
 	if (ctx->nreloc >= RADEON_CTX_MAX_PM4) {
 		return -EBUSY;
 	}
-	ctx->reloc[ctx->nreloc].handle = bo->handle;
+	ctx->reloc[ctx->nreloc].handle = bo->bo->handle;
 	ctx->reloc[ctx->nreloc].read_domain = placement[0] | placement [1];
 	ctx->reloc[ctx->nreloc].write_domain = placement[0] | placement [1];
 	ctx->reloc[ctx->nreloc].flags = 0;
@@ -307,6 +309,7 @@ void radeon_ctx_dump_bof(struct radeon_ctx *ctx, const char *file)
 {
 	bof_t *bcs, *blob, *array, *bo, *size, *handle, *device_id, *root;
 	unsigned i;
+	void *data;
 
 	root = device_id = bcs = blob = array = bo = size = handle = NULL;
 	root = bof_object();
@@ -343,23 +346,23 @@ void radeon_ctx_dump_bof(struct radeon_ctx *ctx, const char *file)
 		bo = bof_object();
 		if (bo == NULL)
 			goto out_err;
-		size = bof_int32(ctx->bo[i]->size);
+		size = bof_int32(ctx->bo[i]->bo->size);
 		if (size == NULL)
 			goto out_err;
 		if (bof_object_set(bo, "size", size))
 			goto out_err;
 		bof_decref(size);
 		size = NULL;
-		handle = bof_int32(ctx->bo[i]->handle);
+		handle = bof_int32(ctx->bo[i]->bo->handle);
 		if (handle == NULL)
 			goto out_err;
 		if (bof_object_set(bo, "handle", handle))
 			goto out_err;
 		bof_decref(handle);
 		handle = NULL;
-		radeon_bo_map(ctx->radeon, ctx->bo[i]);
-		blob = bof_blob(ctx->bo[i]->size, ctx->bo[i]->data);
-		radeon_bo_unmap(ctx->radeon, ctx->bo[i]);
+		data = radeon_ws_bo_map(ctx->radeon, ctx->bo[i]);
+		blob = bof_blob(ctx->bo[i]->bo->size, data);
+		radeon_ws_bo_unmap(ctx->radeon, ctx->bo[i]);
 		if (blob == NULL)
 			goto out_err;
 		if (bof_object_set(bo, "data", blob))
diff --git a/src/gallium/winsys/r600/drm/radeon_priv.h b/src/gallium/winsys/r600/drm/radeon_priv.h
index b5a4eeae6b..6bd8d9850f 100644
--- a/src/gallium/winsys/r600/drm/radeon_priv.h
+++ b/src/gallium/winsys/r600/drm/radeon_priv.h
@@ -23,9 +23,14 @@
 #include <errno.h>
 #include "radeon.h"
 
+#include "pipe/p_compiler.h"
+#include "util/u_inlines.h"
+#include "pipe/p_defines.h"
+
 struct radeon;
 struct radeon_ctx;
 
+
 /*
  * radeon functions
  */
@@ -37,6 +42,15 @@ struct radeon_register {
 	char				name[64];
 };
 
+struct radeon_bo {
+	struct pipe_reference           reference;
+	unsigned			handle;
+	unsigned			size;
+	unsigned			alignment;
+	unsigned			map_count;
+	void				*data;
+};
+
 struct radeon_sub_type {
 	int				shader_type;
 	const struct radeon_register	*regs;
@@ -61,7 +75,7 @@ struct radeon_ctx {
 	unsigned			nreloc;
 	struct radeon_cs_reloc		*reloc;
 	unsigned			nbo;
-	struct radeon_bo		**bo;
+	struct radeon_ws_bo		**bo;
 };
 
 struct radeon {
@@ -74,6 +88,11 @@ struct radeon {
 	unsigned max_states;
 };
 
+struct radeon_ws_bo {
+	struct pipe_reference reference;
+	struct radeon_bo *bo;
+};
+
 extern struct radeon *radeon_new(int fd, unsigned device);
 extern struct radeon *radeon_incref(struct radeon *radeon);
 extern struct radeon *radeon_decref(struct radeon *radeon);
@@ -102,4 +121,13 @@ extern int radeon_state_reloc(struct radeon_state *state, unsigned id, unsigned
  */
 extern int radeon_draw_pm4(struct radeon_draw *draw);
 
+/* bo */
+struct radeon_bo *radeon_bo(struct radeon *radeon, unsigned handle,
+			    unsigned size, unsigned alignment, void *ptr);
+int radeon_bo_map(struct radeon *radeon, struct radeon_bo *bo);
+void radeon_bo_unmap(struct radeon *radeon, struct radeon_bo *bo);
+void radeon_bo_reference(struct radeon *radeon, struct radeon_bo **dst,
+			 struct radeon_bo *src);
+int radeon_bo_wait(struct radeon *radeon, struct radeon_bo *bo);
+
 #endif
diff --git a/src/gallium/winsys/r600/drm/radeon_state.c b/src/gallium/winsys/r600/drm/radeon_state.c
index e37e714533..b237b39c2b 100644
--- a/src/gallium/winsys/r600/drm/radeon_state.c
+++ b/src/gallium/winsys/r600/drm/radeon_state.c
@@ -139,7 +139,7 @@ void radeon_state_fini(struct radeon_state *state)
 	if (state == NULL)
 		return NULL;
 	for (i = 0; i < state->nbo; i++) {
-		state->bo[i] = radeon_bo_decref(state->radeon, state->bo[i]);
+		radeon_ws_bo_reference(state->radeon, &state->bo[i], NULL);
 	}
 	memset(state, 0, sizeof(struct radeon_state));
 }
diff --git a/src/gallium/winsys/r600/drm/radeon_ws_bo.c b/src/gallium/winsys/r600/drm/radeon_ws_bo.c
new file mode 100644
index 0000000000..c789f8780f
--- /dev/null
+++ b/src/gallium/winsys/r600/drm/radeon_ws_bo.c
@@ -0,0 +1,63 @@
+#include <malloc.h>
+#include "radeon_priv.h"
+
+struct radeon_ws_bo *radeon_ws_bo(struct radeon *radeon,
+				  unsigned size, unsigned alignment)
+{
+	struct radeon_ws_bo *ws_bo = calloc(1, sizeof(struct radeon_ws_bo));
+
+	ws_bo->bo = radeon_bo(radeon, 0, size, alignment, NULL);
+	if (!ws_bo->bo) {
+		free(ws_bo);
+		return NULL;
+	}
+
+	pipe_reference_init(&ws_bo->reference, 1);
+	return ws_bo;
+}
+
+struct radeon_ws_bo *radeon_ws_bo_handle(struct radeon *radeon,
+					 unsigned handle)
+{
+	struct radeon_ws_bo *ws_bo = calloc(1, sizeof(struct radeon_ws_bo));
+
+	ws_bo->bo = radeon_bo(radeon, handle, 0, 0, NULL);
+	if (!ws_bo->bo) {
+		free(ws_bo);
+		return NULL;
+	}
+	pipe_reference_init(&ws_bo->reference, 1);
+	return ws_bo;
+}
+
+void *radeon_ws_bo_map(struct radeon *radeon, struct radeon_ws_bo *bo)
+{
+	radeon_bo_map(radeon, bo->bo);
+	return bo->bo->data;
+}
+
+void radeon_ws_bo_unmap(struct radeon *radeon, struct radeon_ws_bo *bo)
+{
+	radeon_bo_unmap(radeon, bo->bo);
+}
+
+static void radeon_ws_bo_destroy(struct radeon *radeon, struct radeon_ws_bo *bo)
+{
+	radeon_bo_reference(radeon, &bo->bo, NULL);
+	free(bo);
+}
+
+void radeon_ws_bo_reference(struct radeon *radeon, struct radeon_ws_bo **dst,
+			    struct radeon_ws_bo *src)
+{
+	struct radeon_ws_bo *old = *dst;
+	if (pipe_reference(&(*dst)->reference, &src->reference)) {
+		radeon_ws_bo_destroy(radeon, old);
+	}
+	*dst = src;
+}
+
+int radeon_ws_bo_wait(struct radeon *radeon, struct radeon_ws_bo *bo)
+{
+	return radeon_bo_wait(radeon, bo->bo);
+}