X-Git-Url: http://ftp.safe.ca/?a=blobdiff_plain;f=drivers%2Fgpu%2Fdrm%2Fradeon%2Fr300_cmdbuf.c;h=c5c2742e41405514364b34bfdbc39c725bc195b7;hb=3374cd1abd478f767aaedf2c21d109596ff0fe72;hp=702df45320f7bcc897439e8b38a1f8d74200c5c4;hpb=c0e09200dc0813972442e550a5905a132768e56c;p=safe%2Fjmp%2Flinux-2.6

diff --git a/drivers/gpu/drm/radeon/r300_cmdbuf.c b/drivers/gpu/drm/radeon/r300_cmdbuf.c
index 702df45..c5c2742 100644
--- a/drivers/gpu/drm/radeon/r300_cmdbuf.c
+++ b/drivers/gpu/drm/radeon/r300_cmdbuf.c
@@ -33,10 +33,13 @@
 
 #include "drmP.h"
 #include "drm.h"
+#include "drm_buffer.h"
 #include "radeon_drm.h"
 #include "radeon_drv.h"
 #include "r300_reg.h"
 
+#include <asm/unaligned.h>
+
 #define R300_SIMULTANEOUS_CLIPRECTS		4
 
 /* Values for R300_RE_CLIPRECT_CNTL depending on the number of cliprects
@@ -77,6 +80,9 @@ static int r300_emit_cliprects(drm_radeon_private_t *dev_priv,
 				return -EFAULT;
 			}
 
+			box.x2--; /* Hardware expects inclusive bottom-right corner */
+			box.y2--;
+
 			if ((dev_priv->flags & RADEON_FAMILY_MASK) >= CHIP_RV515) {
 				box.x1 = (box.x1) &
 					R300_CLIPRECT_MASK;
@@ -95,8 +101,8 @@ static int r300_emit_cliprects(drm_radeon_private_t *dev_priv,
 					R300_CLIPRECT_MASK;
 				box.y2 = (box.y2 + R300_CLIPRECT_OFFSET) &
 					R300_CLIPRECT_MASK;
-
 			}
+
 			OUT_RING((box.x1 << R300_CLIPRECT_X_SHIFT) |
 				 (box.y1 << R300_CLIPRECT_Y_SHIFT));
 			OUT_RING((box.x2 << R300_CLIPRECT_X_SHIFT) |
@@ -136,6 +142,18 @@ static int r300_emit_cliprects(drm_radeon_private_t *dev_priv,
 		ADVANCE_RING();
 	}
 
+	/* flus cache and wait idle clean after cliprect change */
+	BEGIN_RING(2);
+	OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
+	OUT_RING(R300_RB3D_DC_FLUSH);
+	ADVANCE_RING();
+	BEGIN_RING(2);
+	OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
+	OUT_RING(RADEON_WAIT_3D_IDLECLEAN);
+	ADVANCE_RING();
+	/* set flush flag */
+	dev_priv->track_flush |= RADEON_FLUSH_EMITED;
+
 	return 0;
 }
 
@@ -166,13 +184,13 @@ void r300_init_reg_flags(struct drm_device *dev)
 	ADD_RANGE(0x21DC, 1);
 	ADD_RANGE(R300_VAP_UNKNOWN_221C, 1);
 	ADD_RANGE(R300_VAP_CLIP_X_0, 4);
-	ADD_RANGE(R300_VAP_PVS_WAITIDLE, 1);
+	ADD_RANGE(R300_VAP_PVS_STATE_FLUSH_REG, 1);
 	ADD_RANGE(R300_VAP_UNKNOWN_2288, 1);
 	ADD_RANGE(R300_VAP_OUTPUT_VTX_FMT_0, 2);
 	ADD_RANGE(R300_VAP_PVS_CNTL_1, 3);
 	ADD_RANGE(R300_GB_ENABLE, 1);
 	ADD_RANGE(R300_GB_MSPOS0, 5);
-	ADD_RANGE(R300_TX_CNTL, 1);
+	ADD_RANGE(R300_TX_INVALTAGS, 1);
 	ADD_RANGE(R300_TX_ENABLE, 1);
 	ADD_RANGE(0x4200, 4);
 	ADD_RANGE(0x4214, 1);
@@ -190,6 +208,10 @@ void r300_init_reg_flags(struct drm_device *dev)
 	ADD_RANGE(0x42C0, 2);
 	ADD_RANGE(R300_RS_CNTL_0, 2);
 
+	ADD_RANGE(R300_SU_REG_DEST, 1);
+	if ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV530)
+		ADD_RANGE(RV530_FG_ZBREG_DEST, 1);
+
 	ADD_RANGE(R300_SC_HYPERZ, 2);
 	ADD_RANGE(0x43E8, 1);
 
@@ -215,6 +237,7 @@ void r300_init_reg_flags(struct drm_device *dev)
 	ADD_RANGE(R300_ZB_DEPTHPITCH, 1);
 	ADD_RANGE(R300_ZB_DEPTHCLEARVALUE, 1);
 	ADD_RANGE(R300_ZB_ZMASK_OFFSET, 13);
+	ADD_RANGE(R300_ZB_ZPASS_DATA, 2); /* ZB_ZPASS_DATA, ZB_ZPASS_ADDR */
 
 	ADD_RANGE(R300_TX_FILTER_0, 16);
 	ADD_RANGE(R300_TX_FILTER1_0, 16);
@@ -277,46 +300,42 @@ static __inline__ int r300_emit_carefully_checked_packet0(drm_radeon_private_t *
 	int reg;
 	int sz;
 	int i;
-	int values[64];
+	u32 *value;
 	RING_LOCALS;
 
 	sz = header.packet0.count;
 	reg = (header.packet0.reghi << 8) | header.packet0.reglo;
 
 	if ((sz > 64) || (sz < 0)) {
-		DRM_ERROR
-		    ("Cannot emit more than 64 values at a time (reg=%04x sz=%d)\n",
-		     reg, sz);
+		DRM_ERROR("Cannot emit more than 64 values at a time (reg=%04x sz=%d)\n",
+			 reg, sz);
 		return -EINVAL;
 	}
+
 	for (i = 0; i < sz; i++) {
-		values[i] = ((int *)cmdbuf->buf)[i];
 		switch (r300_reg_flags[(reg >> 2) + i]) {
 		case MARK_SAFE:
 			break;
 		case MARK_CHECK_OFFSET:
-			if (!radeon_check_offset(dev_priv, (u32) values[i])) {
-				DRM_ERROR
-				    ("Offset failed range check (reg=%04x sz=%d)\n",
-				     reg, sz);
+			value = drm_buffer_pointer_to_dword(cmdbuf->buffer, i);
+			if (!radeon_check_offset(dev_priv, *value)) {
+				DRM_ERROR("Offset failed range check (reg=%04x sz=%d)\n",
+					 reg, sz);
 				return -EINVAL;
 			}
 			break;
 		default:
 			DRM_ERROR("Register %04x failed check as flag=%02x\n",
-				  reg + i * 4, r300_reg_flags[(reg >> 2) + i]);
+				reg + i * 4, r300_reg_flags[(reg >> 2) + i]);
 			return -EINVAL;
 		}
 	}
 
 	BEGIN_RING(1 + sz);
 	OUT_RING(CP_PACKET0(reg, sz - 1));
-	OUT_RING_TABLE(values, sz);
+	OUT_RING_DRM_BUFFER(cmdbuf->buffer, sz);
 	ADVANCE_RING();
 
-	cmdbuf->buf += sz * 4;
-	cmdbuf->bufsz -= sz * 4;
-
 	return 0;
 }
 
@@ -340,7 +359,7 @@ static __inline__ int r300_emit_packet0(drm_radeon_private_t *dev_priv,
 	if (!sz)
 		return 0;
 
-	if (sz * 4 > cmdbuf->bufsz)
+	if (sz * 4 > drm_buffer_unprocessed(cmdbuf->buffer))
 		return -EINVAL;
 
 	if (reg + sz * 4 >= 0x10000) {
@@ -358,12 +377,9 @@ static __inline__ int r300_emit_packet0(drm_radeon_private_t *dev_priv,
 
 	BEGIN_RING(1 + sz);
 	OUT_RING(CP_PACKET0(reg, sz - 1));
-	OUT_RING_TABLE((int *)cmdbuf->buf, sz);
+	OUT_RING_DRM_BUFFER(cmdbuf->buffer, sz);
 	ADVANCE_RING();
 
-	cmdbuf->buf += sz * 4;
-	cmdbuf->bufsz -= sz * 4;
-
 	return 0;
 }
 
@@ -385,22 +401,32 @@ static __inline__ int r300_emit_vpu(drm_radeon_private_t *dev_priv,
 
 	if (!sz)
 		return 0;
-	if (sz * 16 > cmdbuf->bufsz)
+	if (sz * 16 > drm_buffer_unprocessed(cmdbuf->buffer))
 		return -EINVAL;
 
-	BEGIN_RING(5 + sz * 4);
-	/* Wait for VAP to come to senses.. */
-	/* there is no need to emit it multiple times, (only once before VAP is programmed,
-	   but this optimization is for later */
-	OUT_RING_REG(R300_VAP_PVS_WAITIDLE, 0);
+	/* VAP is very sensitive so we purge cache before we program it
+	 * and we also flush its state before & after */
+	BEGIN_RING(6);
+	OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
+	OUT_RING(R300_RB3D_DC_FLUSH);
+	OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
+	OUT_RING(RADEON_WAIT_3D_IDLECLEAN);
+	OUT_RING(CP_PACKET0(R300_VAP_PVS_STATE_FLUSH_REG, 0));
+	OUT_RING(0);
+	ADVANCE_RING();
+	/* set flush flag */
+	dev_priv->track_flush |= RADEON_FLUSH_EMITED;
+
+	BEGIN_RING(3 + sz * 4);
 	OUT_RING_REG(R300_VAP_PVS_UPLOAD_ADDRESS, addr);
 	OUT_RING(CP_PACKET0_TABLE(R300_VAP_PVS_UPLOAD_DATA, sz * 4 - 1));
-	OUT_RING_TABLE((int *)cmdbuf->buf, sz * 4);
-
+	OUT_RING_DRM_BUFFER(cmdbuf->buffer, sz * 4);
 	ADVANCE_RING();
 
-	cmdbuf->buf += sz * 16;
-	cmdbuf->bufsz -= sz * 16;
+	BEGIN_RING(2);
+	OUT_RING(CP_PACKET0(R300_VAP_PVS_STATE_FLUSH_REG, 0));
+	OUT_RING(0);
+	ADVANCE_RING();
 
 	return 0;
 }
@@ -414,18 +440,24 @@ static __inline__ int r300_emit_clear(drm_radeon_private_t *dev_priv,
 {
 	RING_LOCALS;
 
-	if (8 * 4 > cmdbuf->bufsz)
+	if (8 * 4 > drm_buffer_unprocessed(cmdbuf->buffer))
 		return -EINVAL;
 
 	BEGIN_RING(10);
 	OUT_RING(CP_PACKET3(R200_3D_DRAW_IMMD_2, 8));
 	OUT_RING(R300_PRIM_TYPE_POINT | R300_PRIM_WALK_RING |
 		 (1 << R300_PRIM_NUM_VERTICES_SHIFT));
-	OUT_RING_TABLE((int *)cmdbuf->buf, 8);
+	OUT_RING_DRM_BUFFER(cmdbuf->buffer, 8);
 	ADVANCE_RING();
 
-	cmdbuf->buf += 8 * 4;
-	cmdbuf->bufsz -= 8 * 4;
+	BEGIN_RING(4);
+	OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
+	OUT_RING(R300_RB3D_DC_FLUSH);
+	OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
+	OUT_RING(RADEON_WAIT_3D_IDLECLEAN);
+	ADVANCE_RING();
+	/* set flush flag */
+	dev_priv->track_flush |= RADEON_FLUSH_EMITED;
 
 	return 0;
 }
@@ -436,28 +468,29 @@ static __inline__ int r300_emit_3d_load_vbpntr(drm_radeon_private_t *dev_priv,
 {
 	int count, i, k;
 #define MAX_ARRAY_PACKET  64
-	u32 payload[MAX_ARRAY_PACKET];
+	u32 *data;
 	u32 narrays;
 	RING_LOCALS;
 
-	count = (header >> 16) & 0x3fff;
+	count = (header & RADEON_CP_PACKET_COUNT_MASK) >> 16;
 
 	if ((count + 1) > MAX_ARRAY_PACKET) {
 		DRM_ERROR("Too large payload in 3D_LOAD_VBPNTR (count=%d)\n",
 			  count);
 		return -EINVAL;
 	}
-	memset(payload, 0, MAX_ARRAY_PACKET * 4);
-	memcpy(payload, cmdbuf->buf + 4, (count + 1) * 4);
-
 	/* carefully check packet contents */
 
-	narrays = payload[0];
+	/* We have already read the header so advance the buffer. */
+	drm_buffer_advance(cmdbuf->buffer, 4);
+
+	narrays = *(u32 *)drm_buffer_pointer_to_dword(cmdbuf->buffer, 0);
 	k = 0;
 	i = 1;
 	while ((k < narrays) && (i < (count + 1))) {
 		i++;		/* skip attribute field */
-		if (!radeon_check_offset(dev_priv, payload[i])) {
+		data = drm_buffer_pointer_to_dword(cmdbuf->buffer, i);
+		if (!radeon_check_offset(dev_priv, *data)) {
 			DRM_ERROR
 			    ("Offset failed range check (k=%d i=%d) while processing 3D_LOAD_VBPNTR packet.\n",
 			     k, i);
@@ -468,7 +501,8 @@ static __inline__ int r300_emit_3d_load_vbpntr(drm_radeon_private_t *dev_priv,
 		if (k == narrays)
 			break;
 		/* have one more to process, they come in pairs */
-		if (!radeon_check_offset(dev_priv, payload[i])) {
+		data = drm_buffer_pointer_to_dword(cmdbuf->buffer, i);
+		if (!radeon_check_offset(dev_priv, *data)) {
 			DRM_ERROR
 			    ("Offset failed range check (k=%d i=%d) while processing 3D_LOAD_VBPNTR packet.\n",
 			     k, i);
@@ -489,30 +523,30 @@ static __inline__ int r300_emit_3d_load_vbpntr(drm_radeon_private_t *dev_priv,
 
 	BEGIN_RING(count + 2);
 	OUT_RING(header);
-	OUT_RING_TABLE(payload, count + 1);
+	OUT_RING_DRM_BUFFER(cmdbuf->buffer, count + 1);
 	ADVANCE_RING();
 
-	cmdbuf->buf += (count + 2) * 4;
-	cmdbuf->bufsz -= (count + 2) * 4;
-
 	return 0;
 }
 
 static __inline__ int r300_emit_bitblt_multi(drm_radeon_private_t *dev_priv,
 					     drm_radeon_kcmd_buffer_t *cmdbuf)
 {
-	u32 *cmd = (u32 *) cmdbuf->buf;
+	u32 *cmd = drm_buffer_pointer_to_dword(cmdbuf->buffer, 0);
 	int count, ret;
 	RING_LOCALS;
 
-	count=(cmd[0]>>16) & 0x3fff;
 
-	if (cmd[0] & 0x8000) {
-		u32 offset;
+	count = (*cmd & RADEON_CP_PACKET_COUNT_MASK) >> 16;
 
-		if (cmd[1] & (RADEON_GMC_SRC_PITCH_OFFSET_CNTL
+	if (*cmd & 0x8000) {
+		u32 offset;
+		u32 *cmd1 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 1);
+		if (*cmd1 & (RADEON_GMC_SRC_PITCH_OFFSET_CNTL
 			      | RADEON_GMC_DST_PITCH_OFFSET_CNTL)) {
-			offset = cmd[2] << 10;
+
+			u32 *cmd2 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 2);
+			offset = *cmd2 << 10;
 			ret = !radeon_check_offset(dev_priv, offset);
 			if (ret) {
 				DRM_ERROR("Invalid bitblt first offset is %08X\n", offset);
@@ -520,9 +554,10 @@ static __inline__ int r300_emit_bitblt_multi(drm_radeon_private_t *dev_priv,
 			}
 		}
 
-		if ((cmd[1] & RADEON_GMC_SRC_PITCH_OFFSET_CNTL) &&
-		    (cmd[1] & RADEON_GMC_DST_PITCH_OFFSET_CNTL)) {
-			offset = cmd[3] << 10;
+		if ((*cmd1 & RADEON_GMC_SRC_PITCH_OFFSET_CNTL) &&
+		    (*cmd1 & RADEON_GMC_DST_PITCH_OFFSET_CNTL)) {
+			u32 *cmd3 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 3);
+			offset = *cmd3 << 10;
 			ret = !radeon_check_offset(dev_priv, offset);
 			if (ret) {
 				DRM_ERROR("Invalid bitblt second offset is %08X\n", offset);
@@ -533,42 +568,82 @@ static __inline__ int r300_emit_bitblt_multi(drm_radeon_private_t *dev_priv,
 	}
 
 	BEGIN_RING(count+2);
-	OUT_RING(cmd[0]);
-	OUT_RING_TABLE((int *)(cmdbuf->buf + 4), count + 1);
+	OUT_RING_DRM_BUFFER(cmdbuf->buffer, count + 2);
 	ADVANCE_RING();
 
-	cmdbuf->buf += (count+2)*4;
-	cmdbuf->bufsz -= (count+2)*4;
-
 	return 0;
 }
 
-static __inline__ int r300_emit_indx_buffer(drm_radeon_private_t *dev_priv,
-					     drm_radeon_kcmd_buffer_t *cmdbuf)
+static __inline__ int r300_emit_draw_indx_2(drm_radeon_private_t *dev_priv,
+					    drm_radeon_kcmd_buffer_t *cmdbuf)
 {
-	u32 *cmd = (u32 *) cmdbuf->buf;
-	int count, ret;
+	u32 *cmd = drm_buffer_pointer_to_dword(cmdbuf->buffer, 0);
+	u32 *cmd1 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 1);
+	int count;
+	int expected_count;
 	RING_LOCALS;
 
-	count=(cmd[0]>>16) & 0x3fff;
+	count = (*cmd & RADEON_CP_PACKET_COUNT_MASK) >> 16;
 
-	if ((cmd[1] & 0x8000ffff) != 0x80000810) {
-		DRM_ERROR("Invalid indx_buffer reg address %08X\n", cmd[1]);
-		return -EINVAL;
-	}
-	ret = !radeon_check_offset(dev_priv, cmd[2]);
-	if (ret) {
-		DRM_ERROR("Invalid indx_buffer offset is %08X\n", cmd[2]);
+	expected_count = *cmd1 >> 16;
+	if (!(*cmd1 & R300_VAP_VF_CNTL__INDEX_SIZE_32bit))
+		expected_count = (expected_count+1)/2;
+
+	if (count && count != expected_count) {
+		DRM_ERROR("3D_DRAW_INDX_2: packet size %i, expected %i\n",
+			count, expected_count);
 		return -EINVAL;
 	}
 
 	BEGIN_RING(count+2);
-	OUT_RING(cmd[0]);
-	OUT_RING_TABLE((int *)(cmdbuf->buf + 4), count + 1);
+	OUT_RING_DRM_BUFFER(cmdbuf->buffer, count + 2);
 	ADVANCE_RING();
 
-	cmdbuf->buf += (count+2)*4;
-	cmdbuf->bufsz -= (count+2)*4;
+	if (!count) {
+		drm_r300_cmd_header_t stack_header, *header;
+		u32 *cmd1, *cmd2, *cmd3;
+
+		if (drm_buffer_unprocessed(cmdbuf->buffer)
+				< 4*4 + sizeof(stack_header)) {
+			DRM_ERROR("3D_DRAW_INDX_2: expect subsequent INDX_BUFFER, but stream is too short.\n");
+			return -EINVAL;
+		}
+
+		header = drm_buffer_read_object(cmdbuf->buffer,
+				sizeof(stack_header), &stack_header);
+
+		cmd = drm_buffer_pointer_to_dword(cmdbuf->buffer, 0);
+		cmd1 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 1);
+		cmd2 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 2);
+		cmd3 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 3);
+
+		if (header->header.cmd_type != R300_CMD_PACKET3 ||
+		    header->packet3.packet != R300_CMD_PACKET3_RAW ||
+		    *cmd != CP_PACKET3(RADEON_CP_INDX_BUFFER, 2)) {
+			DRM_ERROR("3D_DRAW_INDX_2: expect subsequent INDX_BUFFER.\n");
+			return -EINVAL;
+		}
+
+		if ((*cmd1 & 0x8000ffff) != 0x80000810) {
+			DRM_ERROR("Invalid indx_buffer reg address %08X\n",
+					*cmd1);
+			return -EINVAL;
+		}
+		if (!radeon_check_offset(dev_priv, *cmd2)) {
+			DRM_ERROR("Invalid indx_buffer offset is %08X\n",
+					*cmd2);
+			return -EINVAL;
+		}
+		if (*cmd3 != expected_count) {
+			DRM_ERROR("INDX_BUFFER: buffer size %i, expected %i\n",
+				*cmd3, expected_count);
+			return -EINVAL;
+		}
+
+		BEGIN_RING(4);
+		OUT_RING_DRM_BUFFER(cmdbuf->buffer, 4);
+		ADVANCE_RING();
+	}
 
 	return 0;
 }
@@ -576,65 +651,72 @@ static __inline__ int r300_emit_indx_buffer(drm_radeon_private_t *dev_priv,
 static __inline__ int r300_emit_raw_packet3(drm_radeon_private_t *dev_priv,
 					    drm_radeon_kcmd_buffer_t *cmdbuf)
 {
-	u32 header;
+	u32 *header;
 	int count;
 	RING_LOCALS;
 
-	if (4 > cmdbuf->bufsz)
+	if (4 > drm_buffer_unprocessed(cmdbuf->buffer))
 		return -EINVAL;
 
 	/* Fixme !! This simply emits a packet without much checking.
 	   We need to be smarter. */
 
 	/* obtain first word - actual packet3 header */
-	header = *(u32 *) cmdbuf->buf;
+	header = drm_buffer_pointer_to_dword(cmdbuf->buffer, 0);
 
 	/* Is it packet 3 ? */
-	if ((header >> 30) != 0x3) {
-		DRM_ERROR("Not a packet3 header (0x%08x)\n", header);
+	if ((*header >> 30) != 0x3) {
+		DRM_ERROR("Not a packet3 header (0x%08x)\n", *header);
 		return -EINVAL;
 	}
 
-	count = (header >> 16) & 0x3fff;
+	count = (*header >> 16) & 0x3fff;
 
 	/* Check again now that we know how much data to expect */
-	if ((count + 2) * 4 > cmdbuf->bufsz) {
+	if ((count + 2) * 4 > drm_buffer_unprocessed(cmdbuf->buffer)) {
 		DRM_ERROR
 		    ("Expected packet3 of length %d but have only %d bytes left\n",
-		     (count + 2) * 4, cmdbuf->bufsz);
+		     (count + 2) * 4, drm_buffer_unprocessed(cmdbuf->buffer));
 		return -EINVAL;
 	}
 
 	/* Is it a packet type we know about ? */
-	switch (header & 0xff00) {
+	switch (*header & 0xff00) {
 	case RADEON_3D_LOAD_VBPNTR:	/* load vertex array pointers */
-		return r300_emit_3d_load_vbpntr(dev_priv, cmdbuf, header);
+		return r300_emit_3d_load_vbpntr(dev_priv, cmdbuf, *header);
 
 	case RADEON_CNTL_BITBLT_MULTI:
 		return r300_emit_bitblt_multi(dev_priv, cmdbuf);
 
-	case RADEON_CP_INDX_BUFFER:	/* DRAW_INDX_2 without INDX_BUFFER seems to lock up the gpu */
-		return r300_emit_indx_buffer(dev_priv, cmdbuf);
-	case RADEON_CP_3D_DRAW_IMMD_2:	/* triggers drawing using in-packet vertex data */
-	case RADEON_CP_3D_DRAW_VBUF_2:	/* triggers drawing of vertex buffers setup elsewhere */
-	case RADEON_CP_3D_DRAW_INDX_2:	/* triggers drawing using indices to vertex buffer */
+	case RADEON_CP_INDX_BUFFER:
+		DRM_ERROR("packet3 INDX_BUFFER without preceding 3D_DRAW_INDX_2 is illegal.\n");
+		return -EINVAL;
+	case RADEON_CP_3D_DRAW_IMMD_2:
+		/* triggers drawing using in-packet vertex data */
+	case RADEON_CP_3D_DRAW_VBUF_2:
+		/* triggers drawing of vertex buffers setup elsewhere */
+		dev_priv->track_flush &= ~(RADEON_FLUSH_EMITED |
+					   RADEON_PURGE_EMITED);
+		break;
+	case RADEON_CP_3D_DRAW_INDX_2:
+		/* triggers drawing using indices to vertex buffer */
+		/* whenever we send vertex we clear flush & purge */
+		dev_priv->track_flush &= ~(RADEON_FLUSH_EMITED |
+					   RADEON_PURGE_EMITED);
+		return r300_emit_draw_indx_2(dev_priv, cmdbuf);
 	case RADEON_WAIT_FOR_IDLE:
 	case RADEON_CP_NOP:
 		/* these packets are safe */
 		break;
 	default:
-		DRM_ERROR("Unknown packet3 header (0x%08x)\n", header);
+		DRM_ERROR("Unknown packet3 header (0x%08x)\n", *header);
 		return -EINVAL;
 	}
 
 	BEGIN_RING(count + 2);
-	OUT_RING(header);
-	OUT_RING_TABLE((int *)(cmdbuf->buf + 4), count + 1);
+	OUT_RING_DRM_BUFFER(cmdbuf->buffer, count + 2);
 	ADVANCE_RING();
 
-	cmdbuf->buf += (count + 2) * 4;
-	cmdbuf->bufsz -= (count + 2) * 4;
-
 	return 0;
 }
 
@@ -648,8 +730,7 @@ static __inline__ int r300_emit_packet3(drm_radeon_private_t *dev_priv,
 {
 	int n;
 	int ret;
-	char *orig_buf = cmdbuf->buf;
-	int orig_bufsz = cmdbuf->bufsz;
+	int orig_iter = cmdbuf->buffer->iterator;
 
 	/* This is a do-while-loop so that we run the interior at least once,
 	 * even if cmdbuf->nbox is 0. Compare r300_emit_cliprects for rationale.
@@ -661,8 +742,7 @@ static __inline__ int r300_emit_packet3(drm_radeon_private_t *dev_priv,
 			if (ret)
 				return ret;
 
-			cmdbuf->buf = orig_buf;
-			cmdbuf->bufsz = orig_bufsz;
+			cmdbuf->buffer->iterator = orig_iter;
 		}
 
 		switch (header.packet3.packet) {
@@ -685,9 +765,9 @@ static __inline__ int r300_emit_packet3(drm_radeon_private_t *dev_priv,
 			break;
 
 		default:
-			DRM_ERROR("bad packet3 type %i at %p\n",
+			DRM_ERROR("bad packet3 type %i at byte %d\n",
 				  header.packet3.packet,
-				  cmdbuf->buf - sizeof(header));
+				  cmdbuf->buffer->iterator - (int)sizeof(header));
 			return -EINVAL;
 		}
 
@@ -713,17 +793,53 @@ static __inline__ int r300_emit_packet3(drm_radeon_private_t *dev_priv,
  */
 static __inline__ void r300_pacify(drm_radeon_private_t *dev_priv)
 {
+	uint32_t cache_z, cache_3d, cache_2d;
 	RING_LOCALS;
 
-	BEGIN_RING(6);
-	OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
-	OUT_RING(R300_RB3D_DSTCACHE_UNKNOWN_0A);
+	cache_z = R300_ZC_FLUSH;
+	cache_2d = R300_RB2D_DC_FLUSH;
+	cache_3d = R300_RB3D_DC_FLUSH;
+	if (!(dev_priv->track_flush & RADEON_PURGE_EMITED)) {
+		/* we can purge, primitive where draw since last purge */
+		cache_z |= R300_ZC_FREE;
+		cache_2d |= R300_RB2D_DC_FREE;
+		cache_3d |= R300_RB3D_DC_FREE;
+	}
+
+	/* flush & purge zbuffer */
+	BEGIN_RING(2);
 	OUT_RING(CP_PACKET0(R300_ZB_ZCACHE_CTLSTAT, 0));
-	OUT_RING(R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE|
-		 R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
-	OUT_RING(CP_PACKET3(RADEON_CP_NOP, 0));
-	OUT_RING(0x0);
+	OUT_RING(cache_z);
+	ADVANCE_RING();
+	/* flush & purge 3d */
+	BEGIN_RING(2);
+	OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
+	OUT_RING(cache_3d);
+	ADVANCE_RING();
+	/* flush & purge texture */
+	BEGIN_RING(2);
+	OUT_RING(CP_PACKET0(R300_TX_INVALTAGS, 0));
+	OUT_RING(0);
+	ADVANCE_RING();
+	/* FIXME: is this one really needed ? */
+	BEGIN_RING(2);
+	OUT_RING(CP_PACKET0(R300_RB3D_AARESOLVE_CTL, 0));
+	OUT_RING(0);
+	ADVANCE_RING();
+	BEGIN_RING(2);
+	OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
+	OUT_RING(RADEON_WAIT_3D_IDLECLEAN);
+	ADVANCE_RING();
+	/* flush & purge 2d through E2 as RB2D will trigger lockup */
+	BEGIN_RING(4);
+	OUT_RING(CP_PACKET0(R300_DSTCACHE_CTLSTAT, 0));
+	OUT_RING(cache_2d);
+	OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
+	OUT_RING(RADEON_WAIT_2D_IDLECLEAN |
+		 RADEON_WAIT_HOST_IDLECLEAN);
 	ADVANCE_RING();
+	/* set flush & purge flags */
+	dev_priv->track_flush |= RADEON_FLUSH_EMITED | RADEON_PURGE_EMITED;
 }
 
 /**
@@ -731,12 +847,12 @@ static __inline__ void r300_pacify(drm_radeon_private_t *dev_priv)
  * The actual age emit is done by r300_do_cp_cmdbuf, which is why you must
  * be careful about how this function is called.
  */
-static void r300_discard_buffer(struct drm_device * dev, struct drm_buf * buf)
+static void r300_discard_buffer(struct drm_device *dev, struct drm_master *master, struct drm_buf *buf)
 {
-	drm_radeon_private_t *dev_priv = dev->dev_private;
 	drm_radeon_buf_priv_t *buf_priv = buf->dev_private;
+	struct drm_radeon_master_private *master_priv = master->driver_priv;
 
-	buf_priv->age = ++dev_priv->sarea_priv->last_dispatch;
+	buf_priv->age = ++master_priv->sarea_priv->last_dispatch;
 	buf->pending = 1;
 	buf->used = 0;
 }
@@ -787,11 +903,13 @@ static int r300_scratch(drm_radeon_private_t *dev_priv,
 			drm_r300_cmd_header_t header)
 {
 	u32 *ref_age_base;
-	u32 i, buf_idx, h_pending;
+	u32 i, *buf_idx, h_pending;
+	u64 *ptr_addr;
+	u64 stack_ptr_addr;
 	RING_LOCALS;
 
-	if (cmdbuf->bufsz <
-	    (sizeof(u64) + header.scratch.n_bufs * sizeof(buf_idx))) {
+	if (drm_buffer_unprocessed(cmdbuf->buffer) <
+	    (sizeof(u64) + header.scratch.n_bufs * sizeof(*buf_idx))) {
 		return -EINVAL;
 	}
 
@@ -801,35 +919,35 @@ static int r300_scratch(drm_radeon_private_t *dev_priv,
 
 	dev_priv->scratch_ages[header.scratch.reg]++;
 
-	ref_age_base =  (u32 *)(unsigned long)*((uint64_t *)cmdbuf->buf);
-
-	cmdbuf->buf += sizeof(u64);
-	cmdbuf->bufsz -= sizeof(u64);
+	ptr_addr = drm_buffer_read_object(cmdbuf->buffer,
+			sizeof(stack_ptr_addr), &stack_ptr_addr);
+	ref_age_base = (u32 *)(unsigned long)get_unaligned(ptr_addr);
 
 	for (i=0; i < header.scratch.n_bufs; i++) {
-		buf_idx = *(u32 *)cmdbuf->buf;
-		buf_idx *= 2; /* 8 bytes per buf */
+		buf_idx = drm_buffer_pointer_to_dword(cmdbuf->buffer, 0);
+		*buf_idx *= 2; /* 8 bytes per buf */
 
-		if (DRM_COPY_TO_USER(ref_age_base + buf_idx, &dev_priv->scratch_ages[header.scratch.reg], sizeof(u32))) {
+		if (DRM_COPY_TO_USER(ref_age_base + *buf_idx,
+				&dev_priv->scratch_ages[header.scratch.reg],
+				sizeof(u32)))
 			return -EINVAL;
-		}
 
-		if (DRM_COPY_FROM_USER(&h_pending, ref_age_base + buf_idx + 1, sizeof(u32))) {
+		if (DRM_COPY_FROM_USER(&h_pending,
+				ref_age_base + *buf_idx + 1,
+				sizeof(u32)))
 			return -EINVAL;
-		}
 
-		if (h_pending == 0) {
+		if (h_pending == 0)
 			return -EINVAL;
-		}
 
 		h_pending--;
 
-		if (DRM_COPY_TO_USER(ref_age_base + buf_idx + 1, &h_pending, sizeof(u32))) {
+		if (DRM_COPY_TO_USER(ref_age_base + *buf_idx + 1,
+					&h_pending,
+					sizeof(u32)))
 			return -EINVAL;
-		}
 
-		cmdbuf->buf += sizeof(buf_idx);
-		cmdbuf->bufsz -= sizeof(buf_idx);
+		drm_buffer_advance(cmdbuf->buffer, sizeof(*buf_idx));
 	}
 
 	BEGIN_RING(2);
@@ -852,7 +970,7 @@ static inline int r300_emit_r500fp(drm_radeon_private_t *dev_priv,
 	int sz;
 	int addr;
 	int type;
-	int clamp;
+	int isclamp;
 	int stride;
 	RING_LOCALS;
 
@@ -861,29 +979,26 @@ static inline int r300_emit_r500fp(drm_radeon_private_t *dev_priv,
 	addr = ((header.r500fp.adrhi_flags & 1) << 8) | header.r500fp.adrlo;
 
 	type = !!(header.r500fp.adrhi_flags & R500FP_CONSTANT_TYPE);
-	clamp = !!(header.r500fp.adrhi_flags & R500FP_CONSTANT_CLAMP);
+	isclamp = !!(header.r500fp.adrhi_flags & R500FP_CONSTANT_CLAMP);
 
 	addr |= (type << 16);
-	addr |= (clamp << 17);
+	addr |= (isclamp << 17);
 
 	stride = type ? 4 : 6;
 
 	DRM_DEBUG("r500fp %d %d type: %d\n", sz, addr, type);
 	if (!sz)
 		return 0;
-	if (sz * stride * 4 > cmdbuf->bufsz)
+	if (sz * stride * 4 > drm_buffer_unprocessed(cmdbuf->buffer))
 		return -EINVAL;
 
 	BEGIN_RING(3 + sz * stride);
 	OUT_RING_REG(R500_GA_US_VECTOR_INDEX, addr);
 	OUT_RING(CP_PACKET0_TABLE(R500_GA_US_VECTOR_DATA, sz * stride - 1));
-	OUT_RING_TABLE((int *)cmdbuf->buf, sz * stride);
+	OUT_RING_DRM_BUFFER(cmdbuf->buffer, sz * stride);
 
 	ADVANCE_RING();
 
-	cmdbuf->buf += sz * stride * 4;
-	cmdbuf->bufsz -= sz * stride * 4;
-
 	return 0;
 }
 
@@ -898,6 +1013,7 @@ int r300_do_cp_cmdbuf(struct drm_device *dev,
 		      drm_radeon_kcmd_buffer_t *cmdbuf)
 {
 	drm_radeon_private_t *dev_priv = dev->dev_private;
+	struct drm_radeon_master_private *master_priv = file_priv->master->driver_priv;
 	struct drm_device_dma *dma = dev->dma;
 	struct drm_buf *buf = NULL;
 	int emit_dispatch_age = 0;
@@ -905,8 +1021,7 @@ int r300_do_cp_cmdbuf(struct drm_device *dev,
 
 	DRM_DEBUG("\n");
 
-	/* See the comment above r300_emit_begin3d for why this call must be here,
-	 * and what the cleanup gotos are for. */
+	/* pacify */
 	r300_pacify(dev_priv);
 
 	if (cmdbuf->nbox <= R300_SIMULTANEOUS_CLIPRECTS) {
@@ -915,19 +1030,18 @@ int r300_do_cp_cmdbuf(struct drm_device *dev,
 			goto cleanup;
 	}
 
-	while (cmdbuf->bufsz >= sizeof(drm_r300_cmd_header_t)) {
+	while (drm_buffer_unprocessed(cmdbuf->buffer)
+			>= sizeof(drm_r300_cmd_header_t)) {
 		int idx;
-		drm_r300_cmd_header_t header;
-
-		header.u = *(unsigned int *)cmdbuf->buf;
+		drm_r300_cmd_header_t *header, stack_header;
 
-		cmdbuf->buf += sizeof(header);
-		cmdbuf->bufsz -= sizeof(header);
+		header = drm_buffer_read_object(cmdbuf->buffer,
+				sizeof(stack_header), &stack_header);
 
-		switch (header.header.cmd_type) {
+		switch (header->header.cmd_type) {
 		case R300_CMD_PACKET0:
 			DRM_DEBUG("R300_CMD_PACKET0\n");
-			ret = r300_emit_packet0(dev_priv, cmdbuf, header);
+			ret = r300_emit_packet0(dev_priv, cmdbuf, *header);
 			if (ret) {
 				DRM_ERROR("r300_emit_packet0 failed\n");
 				goto cleanup;
@@ -936,7 +1050,7 @@ int r300_do_cp_cmdbuf(struct drm_device *dev,
 
 		case R300_CMD_VPU:
 			DRM_DEBUG("R300_CMD_VPU\n");
-			ret = r300_emit_vpu(dev_priv, cmdbuf, header);
+			ret = r300_emit_vpu(dev_priv, cmdbuf, *header);
 			if (ret) {
 				DRM_ERROR("r300_emit_vpu failed\n");
 				goto cleanup;
@@ -945,7 +1059,7 @@ int r300_do_cp_cmdbuf(struct drm_device *dev,
 
 		case R300_CMD_PACKET3:
 			DRM_DEBUG("R300_CMD_PACKET3\n");
-			ret = r300_emit_packet3(dev_priv, cmdbuf, header);
+			ret = r300_emit_packet3(dev_priv, cmdbuf, *header);
 			if (ret) {
 				DRM_ERROR("r300_emit_packet3 failed\n");
 				goto cleanup;
@@ -979,8 +1093,8 @@ int r300_do_cp_cmdbuf(struct drm_device *dev,
 				int i;
 				RING_LOCALS;
 
-				BEGIN_RING(header.delay.count);
-				for (i = 0; i < header.delay.count; i++)
+				BEGIN_RING(header->delay.count);
+				for (i = 0; i < header->delay.count; i++)
 					OUT_RING(RADEON_CP_PACKET2);
 				ADVANCE_RING();
 			}
@@ -988,7 +1102,7 @@ int r300_do_cp_cmdbuf(struct drm_device *dev,
 
 		case R300_CMD_DMA_DISCARD:
 			DRM_DEBUG("RADEON_CMD_DMA_DISCARD\n");
-			idx = header.dma.buf_idx;
+			idx = header->dma.buf_idx;
 			if (idx < 0 || idx >= dma->buf_count) {
 				DRM_ERROR("buffer index %d (of %d max)\n",
 					  idx, dma->buf_count - 1);
@@ -1006,17 +1120,17 @@ int r300_do_cp_cmdbuf(struct drm_device *dev,
 			}
 
 			emit_dispatch_age = 1;
-			r300_discard_buffer(dev, buf);
+			r300_discard_buffer(dev, file_priv->master, buf);
 			break;
 
 		case R300_CMD_WAIT:
 			DRM_DEBUG("R300_CMD_WAIT\n");
-			r300_cmd_wait(dev_priv, header);
+			r300_cmd_wait(dev_priv, *header);
 			break;
 
 		case R300_CMD_SCRATCH:
 			DRM_DEBUG("R300_CMD_SCRATCH\n");
-			ret = r300_scratch(dev_priv, cmdbuf, header);
+			ret = r300_scratch(dev_priv, cmdbuf, *header);
 			if (ret) {
 				DRM_ERROR("r300_scratch failed\n");
 				goto cleanup;
@@ -1030,16 +1144,16 @@ int r300_do_cp_cmdbuf(struct drm_device *dev,
 				goto cleanup;
 			}
 			DRM_DEBUG("R300_CMD_R500FP\n");
-			ret = r300_emit_r500fp(dev_priv, cmdbuf, header);
+			ret = r300_emit_r500fp(dev_priv, cmdbuf, *header);
 			if (ret) {
 				DRM_ERROR("r300_emit_r500fp failed\n");
 				goto cleanup;
 			}
 			break;
 		default:
-			DRM_ERROR("bad cmd_type %i at %p\n",
-				  header.header.cmd_type,
-				  cmdbuf->buf - sizeof(header));
+			DRM_ERROR("bad cmd_type %i at byte %d\n",
+				  header->header.cmd_type,
+				  cmdbuf->buffer->iterator - (int)sizeof(*header));
 			ret = -EINVAL;
 			goto cleanup;
 		}
@@ -1061,7 +1175,7 @@ int r300_do_cp_cmdbuf(struct drm_device *dev,
 
 		/* Emit the vertex buffer age */
 		BEGIN_RING(2);
-		RADEON_DISPATCH_AGE(dev_priv->sarea_priv->last_dispatch);
+		RADEON_DISPATCH_AGE(master_priv->sarea_priv->last_dispatch);
 		ADVANCE_RING();
 	}