Merge branch 'master' into gallium-texture-transfer

Conflicts:

	src/mesa/state_tracker/st_cb_accum.c
	src/mesa/state_tracker/st_cb_drawpixels.c

12 files changed, 309 insertions, 88 deletions

diff --git a/src/gallium/drivers/nv50/nv50_clear.c b/src/gallium/drivers/nv50/nv50_clear.c
index 6380f397ea6..f9bc3b53caa 100644
--- a/src/gallium/drivers/nv50/nv50_clear.c
+++ b/src/gallium/drivers/nv50/nv50_clear.c
@@ -31,6 +31,8 @@ nv50_clear(struct pipe_context *pipe, struct pipe_surface *ps,
 	   unsigned clearValue)
 {
 	struct nv50_context *nv50 = nv50_context(pipe);
+	struct nouveau_channel *chan = nv50->screen->nvws->channel;
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
 	struct pipe_framebuffer_state fb, s_fb = nv50->framebuffer;
 	struct pipe_scissor_state sc, s_sc = nv50->scissor;
 	unsigned dirty = nv50->dirty;
@@ -59,21 +61,21 @@ nv50_clear(struct pipe_context *pipe, struct pipe_surface *ps,
 
 	switch (ps->format) {
 	case PIPE_FORMAT_A8R8G8B8_UNORM:
-		BEGIN_RING(tesla, 0x0d80, 4);
-		OUT_RINGf (ubyte_to_float((clearValue >> 16) & 0xff));
-		OUT_RINGf (ubyte_to_float((clearValue >>  8) & 0xff));
-		OUT_RINGf (ubyte_to_float((clearValue >>  0) & 0xff));
-		OUT_RINGf (ubyte_to_float((clearValue >> 24) & 0xff));
-		BEGIN_RING(tesla, 0x19d0, 1);
-		OUT_RING  (0x3c);
+		BEGIN_RING(chan, tesla, 0x0d80, 4);
+		OUT_RINGf (chan, ubyte_to_float((clearValue >> 16) & 0xff));
+		OUT_RINGf (chan, ubyte_to_float((clearValue >>  8) & 0xff));
+		OUT_RINGf (chan, ubyte_to_float((clearValue >>  0) & 0xff));
+		OUT_RINGf (chan, ubyte_to_float((clearValue >> 24) & 0xff));
+		BEGIN_RING(chan, tesla, 0x19d0, 1);
+		OUT_RING  (chan, 0x3c);
 		break;
 	case PIPE_FORMAT_Z24S8_UNORM:
-		BEGIN_RING(tesla, 0x0d90, 1);
-		OUT_RINGf ((float)(clearValue >> 8) * (1.0 / 16777215.0));
-		BEGIN_RING(tesla, 0x0da0, 1);
-		OUT_RING  (clearValue & 0xff);
-		BEGIN_RING(tesla, 0x19d0, 1);
-		OUT_RING  (0x03);
+		BEGIN_RING(chan, tesla, 0x0d90, 1);
+		OUT_RINGf (chan, (float)(clearValue >> 8) * (1.0 / 16777215.0));
+		BEGIN_RING(chan, tesla, 0x0da0, 1);
+		OUT_RING  (chan, clearValue & 0xff);
+		BEGIN_RING(chan, tesla, 0x19d0, 1);
+		OUT_RING  (chan, 0x03);
 		break;
 	default:
 		pipe->surface_fill(pipe, ps, 0, 0, ps->width, ps->height,
diff --git a/src/gallium/drivers/nv50/nv50_context.c b/src/gallium/drivers/nv50/nv50_context.c
index 99776239d2b..565a5da668c 100644
--- a/src/gallium/drivers/nv50/nv50_context.c
+++ b/src/gallium/drivers/nv50/nv50_context.c
@@ -33,7 +33,7 @@ nv50_flush(struct pipe_context *pipe, unsigned flags,
 {
 	struct nv50_context *nv50 = (struct nv50_context *)pipe;
 	
-	FIRE_RING(fence);
+	FIRE_RING(nv50->screen->nvws->channel);
 }
 
 static void
diff --git a/src/gallium/drivers/nv50/nv50_context.h b/src/gallium/drivers/nv50/nv50_context.h
index 6c9e18429ac..1e9d45cb340 100644
--- a/src/gallium/drivers/nv50/nv50_context.h
+++ b/src/gallium/drivers/nv50/nv50_context.h
@@ -15,10 +15,6 @@
 #include "nouveau/nouveau_gldefs.h"
 #include "nouveau/nouveau_stateobj.h"
 
-#define NOUVEAU_PUSH_CONTEXT(ctx)                                              \
-	struct nv50_screen *ctx = nv50->screen
-#include "nouveau/nouveau_push.h"
-
 #include "nv50_screen.h"
 #include "nv50_program.h"
 
@@ -171,6 +167,11 @@ extern void nv50_init_query_functions(struct nv50_context *nv50);
 
 extern void nv50_screen_init_miptree_functions(struct pipe_screen *pscreen);
 
+extern int
+nv50_surface_do_copy(struct nv50_screen *screen, struct pipe_surface *dst,
+		     int dx, int dy, struct pipe_surface *src, int sx, int sy,
+		     int w, int h);
+
 /* nv50_draw.c */
 extern struct draw_stage *nv50_draw_render_stage(struct nv50_context *nv50);
 
diff --git a/src/gallium/drivers/nv50/nv50_miptree.c b/src/gallium/drivers/nv50/nv50_miptree.c
index c6e65c98160..91091d53f57 100644
--- a/src/gallium/drivers/nv50/nv50_miptree.c
+++ b/src/gallium/drivers/nv50/nv50_miptree.c
@@ -104,6 +104,31 @@ nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp)
 	return &mt->base;
 }
 
+static struct pipe_texture *
+nv50_miptree_blanket(struct pipe_screen *pscreen, const struct pipe_texture *pt,
+		     const unsigned *stride, struct pipe_buffer *pb)
+{
+	struct nv50_miptree *mt;
+
+	/* Only supports 2D, non-mipmapped textures for the moment */
+	if (pt->target != PIPE_TEXTURE_2D || pt->last_level != 0 ||
+	    pt->depth[0] != 1)
+		return NULL;
+
+	mt = CALLOC_STRUCT(nv50_miptree);
+	if (!mt)
+		return NULL;
+
+	mt->base = *pt;
+	mt->base.refcount = 1;
+	mt->base.screen = pscreen;
+	mt->image_nr = 1;
+	mt->level[0].image_offset = CALLOC(1, sizeof(unsigned));
+
+	pipe_buffer_reference(pscreen, &mt->buffer, pb);
+	return &mt->base;
+}
+
 static INLINE void
 mark_dirty(uint32_t *flags, unsigned image)
 {
@@ -141,7 +166,7 @@ void
 nv50_miptree_sync(struct pipe_screen *pscreen, struct nv50_miptree *mt,
 		  unsigned level, unsigned image)
 {
-	struct nouveau_winsys *nvws = nv50_screen(pscreen)->nvws;
+	struct nv50_screen *nvscreen = nv50_screen(pscreen);
 	struct nv50_miptree_level *lvl = &mt->level[level];
 	struct pipe_surface *dst, *src;
 	unsigned face = 0, zslice = 0;
@@ -172,7 +197,7 @@ nv50_miptree_sync(struct pipe_screen *pscreen, struct nv50_miptree *mt,
 	dst = pscreen->get_tex_surface(pscreen, &mt->base, face, level, zslice,
 				       PIPE_BUFFER_USAGE_GPU_READ);
 
-	nvws->surface_copy(nvws, dst, 0, 0, src, 0, 0, dst->width, dst->height);
+	nv50_surface_do_copy(nvscreen, dst, 0, 0, src, 0, 0, dst->width, dst->height);
 
 	pscreen->tex_surface_release(pscreen, &dst);
 	pscreen->tex_surface_release(pscreen, &src);
@@ -183,7 +208,7 @@ static void
 nv50_miptree_sync_cpu(struct pipe_screen *pscreen, struct nv50_miptree *mt,
 		      unsigned level, unsigned image)
 {
-	struct nouveau_winsys *nvws = nv50_screen(pscreen)->nvws;
+	struct nv50_screen *nvscreen = nv50_screen(pscreen);
 	struct nv50_miptree_level *lvl = &mt->level[level];
 	struct pipe_surface *dst, *src;
 	unsigned face = 0, zslice = 0;
@@ -204,7 +229,7 @@ nv50_miptree_sync_cpu(struct pipe_screen *pscreen, struct nv50_miptree *mt,
 	dst = pscreen->get_tex_surface(pscreen, &mt->base, face, level, zslice,
 				       PIPE_BUFFER_USAGE_CPU_READ);
 
-	nvws->surface_copy(nvws, dst, 0, 0, src, 0, 0, dst->width, dst->height);
+	nv50_surface_do_copy(nvscreen, dst, 0, 0, src, 0, 0, dst->width, dst->height);
 
 	pscreen->tex_surface_release(pscreen, &dst);
 	pscreen->tex_surface_release(pscreen, &src);
@@ -287,6 +312,7 @@ void
 nv50_screen_init_miptree_functions(struct pipe_screen *pscreen)
 {
 	pscreen->texture_create = nv50_miptree_create;
+	pscreen->texture_blanket = nv50_miptree_blanket;
 	pscreen->texture_release = nv50_miptree_release;
 	pscreen->get_tex_surface = nv50_miptree_surface_new;
 	pscreen->tex_surface_release = nv50_miptree_surface_del;
diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c
index b902c8cf53f..14c5d47e790 100644
--- a/src/gallium/drivers/nv50/nv50_program.c
+++ b/src/gallium/drivers/nv50/nv50_program.c
@@ -1545,13 +1545,16 @@ static void
 nv50_program_upload_data(struct nv50_context *nv50, float *map,
 			 unsigned start, unsigned count)
 {
+	struct nouveau_channel *chan = nv50->screen->nvws->channel;
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+
 	while (count) {
 		unsigned nr = count > 2047 ? 2047 : count;
 
-		BEGIN_RING(tesla, 0x00000f00, 1);
-		OUT_RING  ((NV50_CB_PMISC << 0) | (start << 8));
-		BEGIN_RING(tesla, 0x40000f04, nr);
-		OUT_RINGp (map, nr);
+		BEGIN_RING(chan, tesla, 0x00000f00, 1);
+		OUT_RING  (chan, (NV50_CB_PMISC << 0) | (start << 8));
+		BEGIN_RING(chan, tesla, 0x40000f04, nr);
+		OUT_RINGp (chan, map, nr);
 
 		map += nr;
 		start += nr;
@@ -1598,6 +1601,8 @@ nv50_program_validate_data(struct nv50_context *nv50, struct nv50_program *p)
 static void
 nv50_program_validate_code(struct nv50_context *nv50, struct nv50_program *p)
 {
+	struct nouveau_channel *chan = nv50->screen->nvws->channel;
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
 	struct pipe_winsys *ws = nv50->pipe.winsys;
 	struct nv50_program_exec *e;
 	struct nouveau_stateobj *so;
@@ -1664,14 +1669,14 @@ nv50_program_validate_code(struct nv50_context *nv50, struct nv50_program *p)
 		nr = MIN2(count, 2047);
 		nr = MIN2(nvws->channel->pushbuf->remaining, nr);
 		if (nvws->channel->pushbuf->remaining < (nr + 3)) {
-			FIRE_RING(NULL);
+			FIRE_RING(chan);
 			continue;
 		}
 
-		BEGIN_RING(tesla, 0x0f00, 1);
-		OUT_RING  ((start << 8) | NV50_CB_PUPLOAD);
-		BEGIN_RING(tesla, 0x40000f04, nr);	
-		OUT_RINGp (up + start, nr);
+		BEGIN_RING(chan, tesla, 0x0f00, 1);
+		OUT_RING  (chan, (start << 8) | NV50_CB_PUPLOAD);
+		BEGIN_RING(chan, tesla, 0x40000f04, nr);	
+		OUT_RINGp (chan, up + start, nr);
 
 		start += nr;
 		count -= nr;
diff --git a/src/gallium/drivers/nv50/nv50_query.c b/src/gallium/drivers/nv50/nv50_query.c
index 1b3a41340a9..20745ceab8b 100644
--- a/src/gallium/drivers/nv50/nv50_query.c
+++ b/src/gallium/drivers/nv50/nv50_query.c
@@ -71,12 +71,14 @@ static void
 nv50_query_begin(struct pipe_context *pipe, struct pipe_query *pq)
 {
 	struct nv50_context *nv50 = nv50_context(pipe);
+	struct nouveau_channel *chan = nv50->screen->nvws->channel;
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
 	struct nv50_query *q = nv50_query(pq);
 
-	BEGIN_RING(tesla, 0x1530, 1);
-	OUT_RING  (1);
-	BEGIN_RING(tesla, 0x1514, 1);
-	OUT_RING  (1);
+	BEGIN_RING(chan, tesla, 0x1530, 1);
+	OUT_RING  (chan, 1);
+	BEGIN_RING(chan, tesla, 0x1514, 1);
+	OUT_RING  (chan, 1);
 
 	q->ready = FALSE;
 }
@@ -85,14 +87,17 @@ static void
 nv50_query_end(struct pipe_context *pipe, struct pipe_query *pq)
 {
 	struct nv50_context *nv50 = nv50_context(pipe);
+	struct nouveau_channel *chan = nv50->screen->nvws->channel;
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
 	struct nv50_query *q = nv50_query(pq);
 
-	BEGIN_RING(tesla, 0x1b00, 4);
-	OUT_RELOCh(q->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
-	OUT_RELOCl(q->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
-	OUT_RING  (0x00000000);
-	OUT_RING  (0x0100f002);
-	FIRE_RING (NULL);
+	WAIT_RING (chan, 5);
+	BEGIN_RING(chan, tesla, 0x1b00, 4);
+	OUT_RELOCh(chan, q->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+	OUT_RELOCl(chan, q->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+	OUT_RING  (chan, 0x00000000);
+	OUT_RING  (chan, 0x0100f002);
+	FIRE_RING (chan);
 }
 
 static boolean
diff --git a/src/gallium/drivers/nv50/nv50_screen.c b/src/gallium/drivers/nv50/nv50_screen.c
index 6cddddacd5b..58d7a621a80 100644
--- a/src/gallium/drivers/nv50/nv50_screen.c
+++ b/src/gallium/drivers/nv50/nv50_screen.c
@@ -173,6 +173,14 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 		return NULL;
 	screen->nvws = nvws;
 
+	/* 2D object */
+	ret = nvws->grobj_alloc(nvws, NV50_2D, &screen->eng2d);
+	if (ret) {
+		NOUVEAU_ERR("Error creating 2D object: %d\n", ret);
+		nv50_screen_destroy(&screen->pipe);
+		return NULL;
+	}
+
 	/* 3D object */
 	if ((chipset & 0xf0) != 0x50 && (chipset & 0xf0) != 0x80) {
 		NOUVEAU_ERR("Not a G8x chipset\n");
@@ -218,6 +226,22 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 		return NULL;
 	}
 
+	/* Static 2D init */
+	so = so_new(64, 0);
+	so_method(so, screen->eng2d, NV50_2D_DMA_NOTIFY, 4);
+	so_data  (so, screen->sync->handle);
+	so_data  (so, screen->nvws->channel->vram->handle);
+	so_data  (so, screen->nvws->channel->vram->handle);
+	so_data  (so, screen->nvws->channel->vram->handle);
+	so_method(so, screen->eng2d, NV50_2D_OPERATION, 1);
+	so_data  (so, NV50_2D_OPERATION_SRCCOPY);
+	so_method(so, screen->eng2d, 0x0290, 1);
+	so_data  (so, 0);
+	so_method(so, screen->eng2d, 0x0888, 1);
+	so_data  (so, 1);
+	so_emit(nvws, so);
+	so_ref(NULL, &so);
+
 	/* Static tesla init */
 	so = so_new(256, 20);
 
diff --git a/src/gallium/drivers/nv50/nv50_screen.h b/src/gallium/drivers/nv50/nv50_screen.h
index 400ddcef06d..c888ca071c8 100644
--- a/src/gallium/drivers/nv50/nv50_screen.h
+++ b/src/gallium/drivers/nv50/nv50_screen.h
@@ -11,6 +11,7 @@ struct nv50_screen {
 	unsigned cur_pctx;
 
 	struct nouveau_grobj *tesla;
+	struct nouveau_grobj *eng2d;
 	struct nouveau_notifier *sync;
 
 	struct pipe_buffer *constbuf;
diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c
index 602d76ac743..948112ffa9f 100644
--- a/src/gallium/drivers/nv50/nv50_state_validate.c
+++ b/src/gallium/drivers/nv50/nv50_state_validate.c
@@ -47,9 +47,11 @@ nv50_state_validate_fb(struct nv50_context *nv50)
 
 		so_method(so, tesla, NV50TCL_RT_ADDRESS_HIGH(i), 5);
 		so_reloc (so, nv50_surface_buffer(fb->cbufs[i]), fb->cbufs[i]->offset,
-			  NOUVEAU_BO_VRAM | NOUVEAU_BO_HIGH, 0, 0);
+			  NOUVEAU_BO_VRAM | NOUVEAU_BO_HIGH |
+			  NOUVEAU_BO_RDWR, 0, 0);
 		so_reloc (so, nv50_surface_buffer(fb->cbufs[i]), fb->cbufs[i]->offset,
-			  NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW, 0, 0);
+			  NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW |
+			  NOUVEAU_BO_RDWR, 0, 0);
 		switch (fb->cbufs[i]->format) {
 		case PIPE_FORMAT_A8R8G8B8_UNORM:
 			so_data(so, 0xcf);
@@ -82,9 +84,11 @@ nv50_state_validate_fb(struct nv50_context *nv50)
 
 		so_method(so, tesla, NV50TCL_ZETA_ADDRESS_HIGH, 5);
 		so_reloc (so, nv50_surface_buffer(fb->zsbuf), fb->zsbuf->offset,
-			  NOUVEAU_BO_VRAM | NOUVEAU_BO_HIGH, 0, 0);
+			  NOUVEAU_BO_VRAM | NOUVEAU_BO_HIGH |
+			  NOUVEAU_BO_RDWR, 0, 0);
 		so_reloc (so, nv50_surface_buffer(fb->zsbuf), fb->zsbuf->offset,
-			  NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW, 0, 0);
+			  NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW |
+			  NOUVEAU_BO_RDWR, 0, 0);
 		switch (fb->zsbuf->format) {
 		case PIPE_FORMAT_Z24S8_UNORM:
 			so_data(so, 0x16);
diff --git a/src/gallium/drivers/nv50/nv50_surface.c b/src/gallium/drivers/nv50/nv50_surface.c
index 8ebbc84817e..f2dd2eb30be 100644
--- a/src/gallium/drivers/nv50/nv50_surface.c
+++ b/src/gallium/drivers/nv50/nv50_surface.c
@@ -20,6 +20,9 @@
  * SOFTWARE.
  */
 
+#define __NOUVEAU_PUSH_H__
+#include <stdint.h>
+#include "nouveau/nouveau_pushbuf.h"
 #include "nv50_context.h"
 #include "pipe/p_defines.h"
 #include "pipe/internal/p_winsys_screen.h"
@@ -27,6 +30,118 @@
 
 #include "util/u_tile.h"
 
+static INLINE int
+nv50_format(enum pipe_format format)
+{
+	switch (format) {
+	case PIPE_FORMAT_A8R8G8B8_UNORM:
+	case PIPE_FORMAT_Z24S8_UNORM:
+		return NV50_2D_DST_FORMAT_32BPP;
+	case PIPE_FORMAT_X8R8G8B8_UNORM:
+		return NV50_2D_DST_FORMAT_24BPP;
+	case PIPE_FORMAT_R5G6B5_UNORM:
+		return NV50_2D_DST_FORMAT_16BPP;
+	case PIPE_FORMAT_A8_UNORM:
+		return NV50_2D_DST_FORMAT_8BPP;
+	default:
+		return -1;
+	}
+}
+
+static int
+nv50_surface_set(struct nv50_screen *screen, struct pipe_surface *ps, int dst)
+{
+	struct nouveau_channel *chan = screen->nvws->channel;
+	struct nouveau_grobj *eng2d = screen->eng2d;
+	struct nouveau_bo *bo;
+ 	int format, mthd = dst ? NV50_2D_DST_FORMAT : NV50_2D_SRC_FORMAT;
+ 	int flags = NOUVEAU_BO_VRAM | (dst ? NOUVEAU_BO_WR : NOUVEAU_BO_RD);
+ 
+	bo = screen->nvws->get_bo(nv50_miptree(ps->texture)->buffer);
+	if (!bo)
+		return 1;
+
+ 	format = nv50_format(ps->format);
+ 	if (format < 0)
+ 		return 1;
+  
+ 	if (!bo->tiled) {
+ 		BEGIN_RING(chan, eng2d, mthd, 2);
+ 		OUT_RING  (chan, format);
+ 		OUT_RING  (chan, 1);
+ 		BEGIN_RING(chan, eng2d, mthd + 0x14, 5);
+ 		OUT_RING  (chan, ps->stride);
+ 		OUT_RING  (chan, ps->width);
+ 		OUT_RING  (chan, ps->height);
+ 		OUT_RELOCh(chan, bo, ps->offset, flags);
+ 		OUT_RELOCl(chan, bo, ps->offset, flags);
+ 	} else {
+ 		BEGIN_RING(chan, eng2d, mthd, 5);
+ 		OUT_RING  (chan, format);
+ 		OUT_RING  (chan, 0);
+ 		OUT_RING  (chan, 0);
+ 		OUT_RING  (chan, 1);
+ 		OUT_RING  (chan, 0);
+ 		BEGIN_RING(chan, eng2d, mthd + 0x18, 4);
+ 		OUT_RING  (chan, ps->width);
+ 		OUT_RING  (chan, ps->height);
+ 		OUT_RELOCh(chan, bo, ps->offset, flags);
+ 		OUT_RELOCl(chan, bo, ps->offset, flags);
+ 	}
+ 
+#if 0
+ 	if (dst) {
+ 		BEGIN_RING(chan, eng2d, NV50_2D_CLIP_X, 4);
+ 		OUT_RING  (chan, 0);
+ 		OUT_RING  (chan, 0);
+ 		OUT_RING  (chan, surf->width);
+ 		OUT_RING  (chan, surf->height);
+ 	}
+#endif
+  
+ 	return 0;
+}
+
+int
+nv50_surface_do_copy(struct nv50_screen *screen, struct pipe_surface *dst,
+		     int dx, int dy, struct pipe_surface *src, int sx, int sy,
+		     int w, int h)
+{
+	struct nouveau_channel *chan = screen->nvws->channel;
+	struct nouveau_grobj *eng2d = screen->eng2d;
+	int ret;
+
+	WAIT_RING (chan, 32);
+
+	ret = nv50_surface_set(screen, dst, 1);
+	if (ret)
+		return ret;
+
+	ret = nv50_surface_set(screen, src, 0);
+	if (ret)
+		return ret;
+
+	BEGIN_RING(chan, eng2d, 0x088c, 1);
+	OUT_RING  (chan, 0);
+	BEGIN_RING(chan, eng2d, NV50_2D_BLIT_DST_X, 4);
+	OUT_RING  (chan, dx);
+	OUT_RING  (chan, dy);
+	OUT_RING  (chan, w);
+	OUT_RING  (chan, h);
+	BEGIN_RING(chan, eng2d, 0x08c0, 4);
+	OUT_RING  (chan, 0);
+	OUT_RING  (chan, 1);
+	OUT_RING  (chan, 0);
+	OUT_RING  (chan, 1);
+	BEGIN_RING(chan, eng2d, 0x08d0, 4);
+	OUT_RING  (chan, 0);
+	OUT_RING  (chan, sx);
+	OUT_RING  (chan, 0);
+	OUT_RING  (chan, sy);
+
+	return 0;
+}
+
 static void
 nv50_surface_copy(struct pipe_context *pipe, boolean flip,
 		  struct pipe_surface *dest, unsigned destx, unsigned desty,
@@ -34,17 +149,19 @@ nv50_surface_copy(struct pipe_context *pipe, boolean flip,
 		  unsigned width, unsigned height)
 {
 	struct nv50_context *nv50 = (struct nv50_context *)pipe;
-	struct nouveau_winsys *nvws = nv50->screen->nvws;
+	struct nv50_screen *screen = nv50->screen;
+
+	assert(src->format == dest->format);
 
 	if (flip) {
 		desty += height;
 		while (height--) {
-			nvws->surface_copy(nvws, dest, destx, desty--, src,
-					   srcx, srcy++, width, 1);
+			nv50_surface_do_copy(screen, dest, destx, desty--, src,
+					     srcx, srcy++, width, 1);
 		}
 	} else {
-		nvws->surface_copy(nvws, dest, destx, desty, src, srcx, srcy,
-				   width, height);
+		nv50_surface_do_copy(screen, dest, destx, desty, src, srcx,
+				     srcy, width, height);
 	}
 }
 
@@ -54,9 +171,30 @@ nv50_surface_fill(struct pipe_context *pipe, struct pipe_surface *dest,
 		  unsigned height, unsigned value)
 {
 	struct nv50_context *nv50 = (struct nv50_context *)pipe;
-	struct nouveau_winsys *nvws = nv50->screen->nvws;
+	struct nv50_screen *screen = nv50->screen;
+	struct nouveau_channel *chan = screen->nvws->channel;
+	struct nouveau_grobj *eng2d = screen->eng2d;
+	int format, ret;
+
+	format = nv50_format(dest->format);
+	if (format < 0)
+		return;
+
+	WAIT_RING (chan, 32);
+
+	ret = nv50_surface_set(screen, dest, 1);
+	if (ret)
+		return;
 
-	nvws->surface_fill(nvws, dest, destx, desty, width, height, value);
+	BEGIN_RING(chan, eng2d, 0x0580, 3);
+	OUT_RING  (chan, 4);
+	OUT_RING  (chan, format);
+	OUT_RING  (chan, value);
+	BEGIN_RING(chan, eng2d, NV50_2D_RECT_X1, 4);
+	OUT_RING  (chan, destx);
+	OUT_RING  (chan, desty);
+	OUT_RING  (chan, width);
+	OUT_RING  (chan, height);
 }
 
 static void *
diff --git a/src/gallium/drivers/nv50/nv50_tex.c b/src/gallium/drivers/nv50/nv50_tex.c
index 239407c92bb..675f9b20cbc 100644
--- a/src/gallium/drivers/nv50/nv50_tex.c
+++ b/src/gallium/drivers/nv50/nv50_tex.c
@@ -117,13 +117,15 @@ nv50_tex_construct(struct nouveau_stateobj *so, struct nv50_miptree *mt)
 		return 1;
 	}
 
-	so_reloc(so, mt->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW, 0, 0);
+	so_reloc(so, mt->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW |
+		     NOUVEAU_BO_RD, 0, 0);
 	so_data (so, 0xd0005000);
 	so_data (so, 0x00300000);
 	so_data (so, mt->base.width[0]);
 	so_data (so, (mt->base.depth[0] << 16) | mt->base.height[0]);
 	so_data (so, 0x03000000);
-	so_reloc(so, mt->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_HIGH, 0, 0);
+	so_reloc(so, mt->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_HIGH |
+		     NOUVEAU_BO_RD, 0, 0);
 
 	return 0;
 }
diff --git a/src/gallium/drivers/nv50/nv50_vbo.c b/src/gallium/drivers/nv50/nv50_vbo.c
index c482a4c241e..0c970adb03a 100644
--- a/src/gallium/drivers/nv50/nv50_vbo.c
+++ b/src/gallium/drivers/nv50/nv50_vbo.c
@@ -53,25 +53,27 @@ nv50_draw_arrays(struct pipe_context *pipe, unsigned mode, unsigned start,
 		 unsigned count)
 {
 	struct nv50_context *nv50 = nv50_context(pipe);
+	struct nouveau_channel *chan = nv50->screen->nvws->channel;
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
 
 	nv50_state_validate(nv50);
 
-	BEGIN_RING(tesla, 0x142c, 1);
-	OUT_RING  (0);
-	BEGIN_RING(tesla, 0x142c, 1);
-	OUT_RING  (0);
-	BEGIN_RING(tesla, 0x1440, 1);
-	OUT_RING  (0);
-	BEGIN_RING(tesla, 0x1334, 1);
-	OUT_RING  (0);
-
-	BEGIN_RING(tesla, NV50TCL_VERTEX_BEGIN, 1);
-	OUT_RING  (nv50_prim(mode));
-	BEGIN_RING(tesla, NV50TCL_VERTEX_BUFFER_FIRST, 2);
-	OUT_RING  (start);
-	OUT_RING  (count);
-	BEGIN_RING(tesla, NV50TCL_VERTEX_END, 1);
-	OUT_RING  (0);
+	BEGIN_RING(chan, tesla, 0x142c, 1);
+	OUT_RING  (chan, 0);
+	BEGIN_RING(chan, tesla, 0x142c, 1);
+	OUT_RING  (chan, 0);
+	BEGIN_RING(chan, tesla, 0x1440, 1);
+	OUT_RING  (chan, 0);
+	BEGIN_RING(chan, tesla, 0x1334, 1);
+	OUT_RING  (chan, 0);
+
+	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
+	OUT_RING  (chan, nv50_prim(mode));
+	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BUFFER_FIRST, 2);
+	OUT_RING  (chan, start);
+	OUT_RING  (chan, count);
+	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
+	OUT_RING  (chan, 0);
 
 	pipe->flush(pipe, 0, NULL);
 	return TRUE;
@@ -81,11 +83,14 @@ static INLINE void
 nv50_draw_elements_inline_u08(struct nv50_context *nv50, uint8_t *map,
 			      unsigned start, unsigned count)
 {
+	struct nouveau_channel *chan = nv50->screen->nvws->channel;
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+
 	map += start;
 
 	if (count & 1) {
-		BEGIN_RING(tesla, 0x15e8, 1);
-		OUT_RING  (map[0]);
+		BEGIN_RING(chan, tesla, 0x15e8, 1);
+		OUT_RING  (chan, map[0]);
 		map++;
 		count--;
 	}
@@ -94,9 +99,9 @@ nv50_draw_elements_inline_u08(struct nv50_context *nv50, uint8_t *map,
 		unsigned nr = count > 2046 ? 2046 : count;
 		int i;
 
-		BEGIN_RING(tesla, 0x400015f0, nr >> 1);
+		BEGIN_RING(chan, tesla, 0x400015f0, nr >> 1);
 		for (i = 0; i < nr; i += 2)
-			OUT_RING  ((map[1] << 16) | map[0]);
+			OUT_RING  (chan, (map[1] << 16) | map[0]);
 
 		count -= nr;
 		map += nr;
@@ -107,11 +112,14 @@ static INLINE void
 nv50_draw_elements_inline_u16(struct nv50_context *nv50, uint16_t *map,
 			      unsigned start, unsigned count)
 {
+	struct nouveau_channel *chan = nv50->screen->nvws->channel;
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+
 	map += start;
 
 	if (count & 1) {
-		BEGIN_RING(tesla, 0x15e8, 1);
-		OUT_RING  (map[0]);
+		BEGIN_RING(chan, tesla, 0x15e8, 1);
+		OUT_RING  (chan, map[0]);
 		map++;
 		count--;
 	}
@@ -120,9 +128,9 @@ nv50_draw_elements_inline_u16(struct nv50_context *nv50, uint16_t *map,
 		unsigned nr = count > 2046 ? 2046 : count;
 		int i;
 
-		BEGIN_RING(tesla, 0x400015f0, nr >> 1);
+		BEGIN_RING(chan, tesla, 0x400015f0, nr >> 1);
 		for (i = 0; i < nr; i += 2)
-			OUT_RING  ((map[1] << 16) | map[0]);
+			OUT_RING  (chan, (map[1] << 16) | map[0]);
 
 		count -= nr;
 		map += nr;
@@ -133,13 +141,16 @@ static INLINE void
 nv50_draw_elements_inline_u32(struct nv50_context *nv50, uint8_t *map,
 			      unsigned start, unsigned count)
 {
+	struct nouveau_channel *chan = nv50->screen->nvws->channel;
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+
 	map += start;
 
 	while (count) {
 		unsigned nr = count > 2047 ? 2047 : count;
 
-		BEGIN_RING(tesla, 0x400015e8, nr);
-		OUT_RINGp (map, nr);
+		BEGIN_RING(chan, tesla, 0x400015e8, nr);
+		OUT_RINGp (chan, map, nr);
 
 		count -= nr;
 		map += nr;
@@ -152,18 +163,20 @@ nv50_draw_elements(struct pipe_context *pipe,
 		   unsigned mode, unsigned start, unsigned count)
 {
 	struct nv50_context *nv50 = nv50_context(pipe);
+	struct nouveau_channel *chan = nv50->screen->nvws->channel;
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
 	struct pipe_winsys *ws = pipe->winsys;
 	void *map = ws->buffer_map(ws, indexBuffer, PIPE_BUFFER_USAGE_CPU_READ);
 
 	nv50_state_validate(nv50);
 
-	BEGIN_RING(tesla, 0x142c, 1);
-	OUT_RING  (0);
-	BEGIN_RING(tesla, 0x142c, 1);
-	OUT_RING  (0);
+	BEGIN_RING(chan, tesla, 0x142c, 1);
+	OUT_RING  (chan, 0);
+	BEGIN_RING(chan, tesla, 0x142c, 1);
+	OUT_RING  (chan, 0);
 
-	BEGIN_RING(tesla, NV50TCL_VERTEX_BEGIN, 1);
-	OUT_RING  (nv50_prim(mode));
+	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
+	OUT_RING  (chan, nv50_prim(mode));
 	switch (indexSize) {
 	case 1:
 		nv50_draw_elements_inline_u08(nv50, map, start, count);
@@ -177,8 +190,8 @@ nv50_draw_elements(struct pipe_context *pipe,
 	default:
 		assert(0);
 	}
-	BEGIN_RING(tesla, NV50TCL_VERTEX_END, 1);
-	OUT_RING  (0);
+	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
+	OUT_RING  (chan, 0);
 
 	pipe->flush(pipe, 0, NULL);
 	return TRUE;