drm/radeon: Fix memory allocation failures in the preKMS command stream checking.

Allocation of single large block of memory may fail under memory
presure. drm_buffer object can hold one large block of data in
multiple independ pages which preents alloation failures.

This patch converts all access to command stream to use drm_buffer
interface. All direct access to array has to go tough drm_buffer
functions to get correct pointer.

Outputting the command stream to ring buffer needs to be awear of
the split nature of drm_buffer. The output operation requires the
new OUT_RING_DRM_BUFFER.

Signed-off-by: Pauli Nieminen <suokkos@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

3 files changed, 262 insertions, 247 deletions

diff --git a/drivers/gpu/drm/radeon/r300_cmdbuf.c b/drivers/gpu/drm/radeon/r300_cmdbuf.c
index 34bffa0e4b7..7f59352cd63 100644
--- a/drivers/gpu/drm/radeon/r300_cmdbuf.c
+++ b/drivers/gpu/drm/radeon/r300_cmdbuf.c
@@ -33,6 +33,7 @@
 
 #include "drmP.h"
 #include "drm.h"
+#include "drm_buffer.h"
 #include "radeon_drm.h"
 #include "radeon_drv.h"
 #include "r300_reg.h"
@@ -299,46 +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;
 }
 
@@ -362,7 +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) {
@@ -380,12 +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;
 }
 
@@ -407,7 +401,7 @@ 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;
 
 	/* VAP is very sensitive so we purge cache before we program it
@@ -426,7 +420,7 @@ static __inline__ int r300_emit_vpu(drm_radeon_private_t *dev_priv,
 	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();
 
 	BEGIN_RING(2);
@@ -434,9 +428,6 @@ static __inline__ int r300_emit_vpu(drm_radeon_private_t *dev_priv,
 	OUT_RING(0);
 	ADVANCE_RING();
 
-	cmdbuf->buf += sz * 16;
-	cmdbuf->bufsz -= sz * 16;
-
 	return 0;
 }
 
@@ -449,14 +440,14 @@ 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();
 
 	BEGIN_RING(4);
@@ -468,9 +459,6 @@ static __inline__ int r300_emit_clear(drm_radeon_private_t *dev_priv,
 	/* set flush flag */
 	dev_priv->track_flush |= RADEON_FLUSH_EMITED;
 
-	cmdbuf->buf += 8 * 4;
-	cmdbuf->bufsz -= 8 * 4;
-
 	return 0;
 }
 
@@ -480,28 +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);
@@ -512,7 +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);
@@ -533,30 +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);
@@ -564,9 +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);
@@ -577,28 +568,25 @@ 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_draw_indx_2(drm_radeon_private_t *dev_priv,
 					    drm_radeon_kcmd_buffer_t *cmdbuf)
 {
-	u32 *cmd;
+	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;
 
-	cmd = (u32 *) cmdbuf->buf;
-	count = (cmd[0]>>16) & 0x3fff;
-	expected_count = cmd[1] >> 16;
-	if (!(cmd[1] & R300_VAP_VF_CNTL__INDEX_SIZE_32bit))
+	count = (*cmd & RADEON_CP_PACKET_COUNT_MASK) >> 16;
+
+	expected_count = *cmd1 >> 16;
+	if (!(*cmd1 & R300_VAP_VF_CNTL__INDEX_SIZE_32bit))
 		expected_count = (expected_count+1)/2;
 
 	if (count && count != expected_count) {
@@ -608,55 +596,53 @@ static __inline__ int r300_emit_draw_indx_2(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;
-
 	if (!count) {
-		drm_r300_cmd_header_t header;
+		drm_r300_cmd_header_t stack_header, *header;
+		u32 *cmd1, *cmd2, *cmd3;
 
-		if (cmdbuf->bufsz < 4*4 + sizeof(header)) {
+		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.u = *(unsigned int *)cmdbuf->buf;
+		header = drm_buffer_read_object(cmdbuf->buffer,
+				sizeof(stack_header), &stack_header);
 
-		cmdbuf->buf += sizeof(header);
-		cmdbuf->bufsz -= sizeof(header);
-		cmd = (u32 *) cmdbuf->buf;
+		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[0] != CP_PACKET3(RADEON_CP_INDX_BUFFER, 2)) {
+		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 ((cmd[1] & 0x8000ffff) != 0x80000810) {
-			DRM_ERROR("Invalid indx_buffer reg address %08X\n", cmd[1]);
+		if ((*cmd1 & 0x8000ffff) != 0x80000810) {
+			DRM_ERROR("Invalid indx_buffer reg address %08X\n",
+					*cmd1);
 			return -EINVAL;
 		}
-		if (!radeon_check_offset(dev_priv, cmd[2])) {
-			DRM_ERROR("Invalid indx_buffer offset is %08X\n", cmd[2]);
+		if (!radeon_check_offset(dev_priv, *cmd2)) {
+			DRM_ERROR("Invalid indx_buffer offset is %08X\n",
+					*cmd2);
 			return -EINVAL;
 		}
-		if (cmd[3] != expected_count) {
+		if (*cmd3 != expected_count) {
 			DRM_ERROR("INDX_BUFFER: buffer size %i, expected %i\n",
-				cmd[3], expected_count);
+				*cmd3, expected_count);
 			return -EINVAL;
 		}
 
 		BEGIN_RING(4);
-		OUT_RING(cmd[0]);
-		OUT_RING_TABLE((int *)(cmdbuf->buf + 4), 3);
+		OUT_RING_DRM_BUFFER(cmdbuf->buffer, 4);
 		ADVANCE_RING();
-
-		cmdbuf->buf += 4*4;
-		cmdbuf->bufsz -= 4*4;
 	}
 
 	return 0;
@@ -665,39 +651,39 @@ static __inline__ int r300_emit_draw_indx_2(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);
@@ -723,18 +709,14 @@ static __inline__ int r300_emit_raw_packet3(drm_radeon_private_t *dev_priv,
 		/* 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;
 }
 
@@ -748,8 +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.
@@ -761,8 +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) {
@@ -785,9 +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 - sizeof(header));
 			return -EINVAL;
 		}
 
@@ -923,12 +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;
-	u64 ptr_addr;
+	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;
 	}
 
@@ -938,36 +919,35 @@ static int r300_scratch(drm_radeon_private_t *dev_priv,
 
 	dev_priv->scratch_ages[header.scratch.reg]++;
 
-	ptr_addr = get_unaligned((u64 *)cmdbuf->buf);
-	ref_age_base = (u32 *)(unsigned long)ptr_addr;
-
-	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)*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);
@@ -1009,19 +989,16 @@ static inline int r300_emit_r500fp(drm_radeon_private_t *dev_priv,
 	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;
 }
 
@@ -1053,19 +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;
@@ -1074,7 +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;
@@ -1083,7 +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;
@@ -1117,8 +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();
 			}
@@ -1126,7 +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);
@@ -1149,12 +1125,12 @@ int r300_do_cp_cmdbuf(struct drm_device *dev,
 
 		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;
@@ -1168,16 +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 - sizeof(*header));
 			ret = -EINVAL;
 			goto cleanup;
 		}
diff --git a/drivers/gpu/drm/radeon/radeon_drv.h b/drivers/gpu/drm/radeon/radeon_drv.h
index b058316e311..f6d20cee570 100644
--- a/drivers/gpu/drm/radeon/radeon_drv.h
+++ b/drivers/gpu/drm/radeon/radeon_drv.h
@@ -312,9 +312,11 @@ typedef struct drm_radeon_buf_priv {
 	u32 age;
 } drm_radeon_buf_priv_t;
 
+struct drm_buffer;
+
 typedef struct drm_radeon_kcmd_buffer {
 	int bufsz;
-	char *buf;
+	struct drm_buffer *buffer;
 	int nbox;
 	struct drm_clip_rect __user *boxes;
 } drm_radeon_kcmd_buffer_t;
@@ -2124,4 +2126,32 @@ extern void radeon_commit_ring(drm_radeon_private_t *dev_priv);
 	write &= mask;						\
 } while (0)
 
+/**
+ * Copy given number of dwords from drm buffer to the ring buffer.
+ */
+#define OUT_RING_DRM_BUFFER(buf, sz) do {				\
+	int _size = (sz) * 4;						\
+	struct drm_buffer *_buf = (buf);				\
+	int _part_size;							\
+	while (_size > 0) {						\
+		_part_size = _size;					\
+									\
+		if (write + _part_size/4 > mask)			\
+			_part_size = ((mask + 1) - write)*4;		\
+									\
+		if (drm_buffer_index(_buf) + _part_size > PAGE_SIZE)	\
+			_part_size = PAGE_SIZE - drm_buffer_index(_buf);\
+									\
+									\
+									\
+		memcpy(ring + write, &_buf->data[drm_buffer_page(_buf)]	\
+			[drm_buffer_index(_buf)], _part_size);		\
+									\
+		_size -= _part_size;					\
+		write = (write + _part_size/4) & mask;			\
+		drm_buffer_advance(_buf, _part_size);			\
+	}								\
+} while (0)
+
+
 #endif				/* __RADEON_DRV_H__ */
diff --git a/drivers/gpu/drm/radeon/radeon_state.c b/drivers/gpu/drm/radeon/radeon_state.c
index 38537d971a3..44b6d66b0ab 100644
--- a/drivers/gpu/drm/radeon/radeon_state.c
+++ b/drivers/gpu/drm/radeon/radeon_state.c
@@ -29,6 +29,7 @@
 
 #include "drmP.h"
 #include "drm.h"
+#include "drm_buffer.h"
 #include "drm_sarea.h"
 #include "radeon_drm.h"
 #include "radeon_drv.h"
@@ -91,21 +92,26 @@ static __inline__ int radeon_check_and_fixup_offset(drm_radeon_private_t *
 static __inline__ int radeon_check_and_fixup_packets(drm_radeon_private_t *
 						     dev_priv,
 						     struct drm_file *file_priv,
-						     int id, u32 *data)
+						     int id, struct drm_buffer *buf)
 {
+	u32 *data;
 	switch (id) {
 
 	case RADEON_EMIT_PP_MISC:
-		if (radeon_check_and_fixup_offset(dev_priv, file_priv,
-		    &data[(RADEON_RB3D_DEPTHOFFSET - RADEON_PP_MISC) / 4])) {
+		data = drm_buffer_pointer_to_dword(buf,
+			(RADEON_RB3D_DEPTHOFFSET - RADEON_PP_MISC) / 4);
+
+		if (radeon_check_and_fixup_offset(dev_priv, file_priv, data)) {
 			DRM_ERROR("Invalid depth buffer offset\n");
 			return -EINVAL;
 		}
 		break;
 
 	case RADEON_EMIT_PP_CNTL:
-		if (radeon_check_and_fixup_offset(dev_priv, file_priv,
-		    &data[(RADEON_RB3D_COLOROFFSET - RADEON_PP_CNTL) / 4])) {
+		data = drm_buffer_pointer_to_dword(buf,
+			(RADEON_RB3D_COLOROFFSET - RADEON_PP_CNTL) / 4);
+
+		if (radeon_check_and_fixup_offset(dev_priv, file_priv, data)) {
 			DRM_ERROR("Invalid colour buffer offset\n");
 			return -EINVAL;
 		}
@@ -117,8 +123,8 @@ static __inline__ int radeon_check_and_fixup_packets(drm_radeon_private_t *
 	case R200_EMIT_PP_TXOFFSET_3:
 	case R200_EMIT_PP_TXOFFSET_4:
 	case R200_EMIT_PP_TXOFFSET_5:
-		if (radeon_check_and_fixup_offset(dev_priv, file_priv,
-						  &data[0])) {
+		data = drm_buffer_pointer_to_dword(buf, 0);
+		if (radeon_check_and_fixup_offset(dev_priv, file_priv, data)) {
 			DRM_ERROR("Invalid R200 texture offset\n");
 			return -EINVAL;
 		}
@@ -127,8 +133,9 @@ static __inline__ int radeon_check_and_fixup_packets(drm_radeon_private_t *
 	case RADEON_EMIT_PP_TXFILTER_0:
 	case RADEON_EMIT_PP_TXFILTER_1:
 	case RADEON_EMIT_PP_TXFILTER_2:
-		if (radeon_check_and_fixup_offset(dev_priv, file_priv,
-		    &data[(RADEON_PP_TXOFFSET_0 - RADEON_PP_TXFILTER_0) / 4])) {
+		data = drm_buffer_pointer_to_dword(buf,
+			(RADEON_PP_TXOFFSET_0 - RADEON_PP_TXFILTER_0) / 4);
+		if (radeon_check_and_fixup_offset(dev_priv, file_priv, data)) {
 			DRM_ERROR("Invalid R100 texture offset\n");
 			return -EINVAL;
 		}
@@ -142,9 +149,10 @@ static __inline__ int radeon_check_and_fixup_packets(drm_radeon_private_t *
 	case R200_EMIT_PP_CUBIC_OFFSETS_5:{
 			int i;
 			for (i = 0; i < 5; i++) {
+				data = drm_buffer_pointer_to_dword(buf, i);
 				if (radeon_check_and_fixup_offset(dev_priv,
 								  file_priv,
-								  &data[i])) {
+								  data)) {
 					DRM_ERROR
 					    ("Invalid R200 cubic texture offset\n");
 					return -EINVAL;
@@ -158,9 +166,10 @@ static __inline__ int radeon_check_and_fixup_packets(drm_radeon_private_t *
 	case RADEON_EMIT_PP_CUBIC_OFFSETS_T2:{
 			int i;
 			for (i = 0; i < 5; i++) {
+				data = drm_buffer_pointer_to_dword(buf, i);
 				if (radeon_check_and_fixup_offset(dev_priv,
 								  file_priv,
-								  &data[i])) {
+								  data)) {
 					DRM_ERROR
 					    ("Invalid R100 cubic texture offset\n");
 					return -EINVAL;
@@ -269,23 +278,24 @@ static __inline__ int radeon_check_and_fixup_packet3(drm_radeon_private_t *
 						     cmdbuf,
 						     unsigned int *cmdsz)
 {
-	u32 *cmd = (u32 *) cmdbuf->buf;
+	u32 *cmd = drm_buffer_pointer_to_dword(cmdbuf->buffer, 0);
 	u32 offset, narrays;
 	int count, i, k;
 
-	*cmdsz = 2 + ((cmd[0] & RADEON_CP_PACKET_COUNT_MASK) >> 16);
+	count = ((*cmd & RADEON_CP_PACKET_COUNT_MASK) >> 16);
+	*cmdsz = 2 + count;
 
-	if ((cmd[0] & 0xc0000000) != RADEON_CP_PACKET3) {
+	if ((*cmd & 0xc0000000) != RADEON_CP_PACKET3) {
 		DRM_ERROR("Not a type 3 packet\n");
 		return -EINVAL;
 	}
 
-	if (4 * *cmdsz > cmdbuf->bufsz) {
+	if (4 * *cmdsz > drm_buffer_unprocessed(cmdbuf->buffer)) {
 		DRM_ERROR("Packet size larger than size of data provided\n");
 		return -EINVAL;
 	}
 
-	switch(cmd[0] & 0xff00) {
+	switch (*cmd & 0xff00) {
 	/* XXX Are there old drivers needing other packets? */
 
 	case RADEON_3D_DRAW_IMMD:
@@ -312,7 +322,6 @@ static __inline__ int radeon_check_and_fixup_packet3(drm_radeon_private_t *
 		break;
 
 	case RADEON_3D_LOAD_VBPNTR:
-		count = (cmd[0] >> 16) & 0x3fff;
 
 		if (count > 18) { /* 12 arrays max */
 			DRM_ERROR("Too large payload in 3D_LOAD_VBPNTR (count=%d)\n",
@@ -321,13 +330,16 @@ static __inline__ int radeon_check_and_fixup_packet3(drm_radeon_private_t *
 		}
 
 		/* carefully check packet contents */
-		narrays = cmd[1] & ~0xc000;
+		cmd = drm_buffer_pointer_to_dword(cmdbuf->buffer, 1);
+
+		narrays = *cmd & ~0xc000;
 		k = 0;
 		i = 2;
 		while ((k < narrays) && (i < (count + 2))) {
 			i++;		/* skip attribute field */
+			cmd = drm_buffer_pointer_to_dword(cmdbuf->buffer, i);
 			if (radeon_check_and_fixup_offset(dev_priv, file_priv,
-							  &cmd[i])) {
+							  cmd)) {
 				DRM_ERROR
 				    ("Invalid offset (k=%d i=%d) in 3D_LOAD_VBPNTR packet.\n",
 				     k, i);
@@ -338,8 +350,10 @@ static __inline__ int radeon_check_and_fixup_packet3(drm_radeon_private_t *
 			if (k == narrays)
 				break;
 			/* have one more to process, they come in pairs */
+			cmd = drm_buffer_pointer_to_dword(cmdbuf->buffer, i);
+
 			if (radeon_check_and_fixup_offset(dev_priv,
-							  file_priv, &cmd[i]))
+							  file_priv, cmd))
 			{
 				DRM_ERROR
 				    ("Invalid offset (k=%d i=%d) in 3D_LOAD_VBPNTR packet.\n",
@@ -363,7 +377,9 @@ static __inline__ int radeon_check_and_fixup_packet3(drm_radeon_private_t *
 			DRM_ERROR("Invalid 3d packet for r200-class chip\n");
 			return -EINVAL;
 		}
-		if (radeon_check_and_fixup_offset(dev_priv, file_priv, &cmd[1])) {
+
+		cmd = drm_buffer_pointer_to_dword(cmdbuf->buffer, 1);
+		if (radeon_check_and_fixup_offset(dev_priv, file_priv, cmd)) {
 				DRM_ERROR("Invalid rndr_gen_indx offset\n");
 				return -EINVAL;
 		}
@@ -374,12 +390,15 @@ static __inline__ int radeon_check_and_fixup_packet3(drm_radeon_private_t *
 			DRM_ERROR("Invalid 3d packet for r100-class chip\n");
 			return -EINVAL;
 		}
-		if ((cmd[1] & 0x8000ffff) != 0x80000810) {
-			DRM_ERROR("Invalid indx_buffer reg address %08X\n", cmd[1]);
+
+		cmd = drm_buffer_pointer_to_dword(cmdbuf->buffer, 1);
+		if ((*cmd & 0x8000ffff) != 0x80000810) {
+			DRM_ERROR("Invalid indx_buffer reg address %08X\n", *cmd);
 			return -EINVAL;
 		}
-		if (radeon_check_and_fixup_offset(dev_priv, file_priv, &cmd[2])) {
-			DRM_ERROR("Invalid indx_buffer offset is %08X\n", cmd[2]);
+		cmd = drm_buffer_pointer_to_dword(cmdbuf->buffer, 2);
+		if (radeon_check_and_fixup_offset(dev_priv, file_priv, cmd)) {
+			DRM_ERROR("Invalid indx_buffer offset is %08X\n", *cmd);
 			return -EINVAL;
 		}
 		break;
@@ -388,31 +407,34 @@ static __inline__ int radeon_check_and_fixup_packet3(drm_radeon_private_t *
 	case RADEON_CNTL_PAINT_MULTI:
 	case RADEON_CNTL_BITBLT_MULTI:
 		/* MSB of opcode: next DWORD GUI_CNTL */
-		if (cmd[1] & (RADEON_GMC_SRC_PITCH_OFFSET_CNTL
+		cmd = drm_buffer_pointer_to_dword(cmdbuf->buffer, 1);
+		if (*cmd & (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;
 			if (radeon_check_and_fixup_offset
 			    (dev_priv, file_priv, &offset)) {
 				DRM_ERROR("Invalid first packet offset\n");
 				return -EINVAL;
 			}
-			cmd[2] = (cmd[2] & 0xffc00000) | offset >> 10;
+			*cmd2 = (*cmd2 & 0xffc00000) | offset >> 10;
 		}
 
-		if ((cmd[1] & RADEON_GMC_SRC_PITCH_OFFSET_CNTL) &&
-		    (cmd[1] & RADEON_GMC_DST_PITCH_OFFSET_CNTL)) {
-			offset = cmd[3] << 10;
+		if ((*cmd & RADEON_GMC_SRC_PITCH_OFFSET_CNTL) &&
+		    (*cmd & RADEON_GMC_DST_PITCH_OFFSET_CNTL)) {
+			u32 *cmd3 = drm_buffer_pointer_to_dword(cmdbuf->buffer, 3);
+			offset = *cmd << 10;
 			if (radeon_check_and_fixup_offset
 			    (dev_priv, file_priv, &offset)) {
 				DRM_ERROR("Invalid second packet offset\n");
 				return -EINVAL;
 			}
-			cmd[3] = (cmd[3] & 0xffc00000) | offset >> 10;
+			*cmd3 = (*cmd3 & 0xffc00000) | offset >> 10;
 		}
 		break;
 
 	default:
-		DRM_ERROR("Invalid packet type %x\n", cmd[0] & 0xff00);
+		DRM_ERROR("Invalid packet type %x\n", *cmd & 0xff00);
 		return -EINVAL;
 	}
 
@@ -2611,7 +2633,6 @@ static int radeon_emit_packets(drm_radeon_private_t * dev_priv,
 {
 	int id = (int)header.packet.packet_id;
 	int sz, reg;
-	int *data = (int *)cmdbuf->buf;
 	RING_LOCALS;
 
 	if (id >= RADEON_MAX_STATE_PACKETS)
@@ -2620,23 +2641,22 @@ static int radeon_emit_packets(drm_radeon_private_t * dev_priv,
 	sz = packet[id].len;
 	reg = packet[id].start;
 
-	if (sz * sizeof(int) > cmdbuf->bufsz) {
+	if (sz * sizeof(u32) > drm_buffer_unprocessed(cmdbuf->buffer)) {
 		DRM_ERROR("Packet size provided larger than data provided\n");
 		return -EINVAL;
 	}
 
-	if (radeon_check_and_fixup_packets(dev_priv, file_priv, id, data)) {
+	if (radeon_check_and_fixup_packets(dev_priv, file_priv, id,
+				cmdbuf->buffer)) {
 		DRM_ERROR("Packet verification failed\n");
 		return -EINVAL;
 	}
 
 	BEGIN_RING(sz + 1);
 	OUT_RING(CP_PACKET0(reg, (sz - 1)));
-	OUT_RING_TABLE(data, sz);
+	OUT_RING_DRM_BUFFER(cmdbuf->buffer, sz);
 	ADVANCE_RING();
 
-	cmdbuf->buf += sz * sizeof(int);
-	cmdbuf->bufsz -= sz * sizeof(int);
 	return 0;
 }
 
@@ -2653,10 +2673,8 @@ static __inline__ int radeon_emit_scalars(drm_radeon_private_t *dev_priv,
 	OUT_RING(CP_PACKET0(RADEON_SE_TCL_SCALAR_INDX_REG, 0));
 	OUT_RING(start | (stride << RADEON_SCAL_INDX_DWORD_STRIDE_SHIFT));
 	OUT_RING(CP_PACKET0_TABLE(RADEON_SE_TCL_SCALAR_DATA_REG, sz - 1));
-	OUT_RING_TABLE(cmdbuf->buf, sz);
+	OUT_RING_DRM_BUFFER(cmdbuf->buffer, sz);
 	ADVANCE_RING();
-	cmdbuf->buf += sz * sizeof(int);
-	cmdbuf->bufsz -= sz * sizeof(int);
 	return 0;
 }
 
@@ -2675,10 +2693,8 @@ static __inline__ int radeon_emit_scalars2(drm_radeon_private_t *dev_priv,
 	OUT_RING(CP_PACKET0(RADEON_SE_TCL_SCALAR_INDX_REG, 0));
 	OUT_RING(start | (stride << RADEON_SCAL_INDX_DWORD_STRIDE_SHIFT));
 	OUT_RING(CP_PACKET0_TABLE(RADEON_SE_TCL_SCALAR_DATA_REG, sz - 1));
-	OUT_RING_TABLE(cmdbuf->buf, sz);
+	OUT_RING_DRM_BUFFER(cmdbuf->buffer, sz);
 	ADVANCE_RING();
-	cmdbuf->buf += sz * sizeof(int);
-	cmdbuf->bufsz -= sz * sizeof(int);
 	return 0;
 }
 
@@ -2696,11 +2712,9 @@ static __inline__ int radeon_emit_vectors(drm_radeon_private_t *dev_priv,
 	OUT_RING(CP_PACKET0(RADEON_SE_TCL_VECTOR_INDX_REG, 0));
 	OUT_RING(start | (stride << RADEON_VEC_INDX_OCTWORD_STRIDE_SHIFT));
 	OUT_RING(CP_PACKET0_TABLE(RADEON_SE_TCL_VECTOR_DATA_REG, (sz - 1)));
-	OUT_RING_TABLE(cmdbuf->buf, sz);
+	OUT_RING_DRM_BUFFER(cmdbuf->buffer, sz);
 	ADVANCE_RING();
 
-	cmdbuf->buf += sz * sizeof(int);
-	cmdbuf->bufsz -= sz * sizeof(int);
 	return 0;
 }
 
@@ -2714,7 +2728,7 @@ static __inline__ int radeon_emit_veclinear(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;
 
 	BEGIN_RING(5 + sz);
@@ -2722,11 +2736,9 @@ static __inline__ int radeon_emit_veclinear(drm_radeon_private_t *dev_priv,
 	OUT_RING(CP_PACKET0(RADEON_SE_TCL_VECTOR_INDX_REG, 0));
 	OUT_RING(start | (1 << RADEON_VEC_INDX_OCTWORD_STRIDE_SHIFT));
 	OUT_RING(CP_PACKET0_TABLE(RADEON_SE_TCL_VECTOR_DATA_REG, (sz - 1)));
-	OUT_RING_TABLE(cmdbuf->buf, sz);
+	OUT_RING_DRM_BUFFER(cmdbuf->buffer, sz);
 	ADVANCE_RING();
 
-	cmdbuf->buf += sz * sizeof(int);
-	cmdbuf->bufsz -= sz * sizeof(int);
 	return 0;
 }
 
@@ -2748,11 +2760,9 @@ static int radeon_emit_packet3(struct drm_device * dev,
 	}
 
 	BEGIN_RING(cmdsz);
-	OUT_RING_TABLE(cmdbuf->buf, cmdsz);
+	OUT_RING_DRM_BUFFER(cmdbuf->buffer, cmdsz);
 	ADVANCE_RING();
 
-	cmdbuf->buf += cmdsz * 4;
-	cmdbuf->bufsz -= cmdsz * 4;
 	return 0;
 }
 
@@ -2805,16 +2815,16 @@ static int radeon_emit_packet3_cliprect(struct drm_device *dev,
 		}
 
 		BEGIN_RING(cmdsz);
-		OUT_RING_TABLE(cmdbuf->buf, cmdsz);
+		OUT_RING_DRM_BUFFER(cmdbuf->buffer, cmdsz);
 		ADVANCE_RING();
 
 	} while (++i < cmdbuf->nbox);
 	if (cmdbuf->nbox == 1)
 		cmdbuf->nbox = 0;
 
+	return 0;
       out:
-	cmdbuf->buf += cmdsz * 4;
-	cmdbuf->bufsz -= cmdsz * 4;
+	drm_buffer_advance(cmdbuf->buffer, cmdsz * 4);
 	return 0;
 }
 
@@ -2847,16 +2857,16 @@ static int radeon_emit_wait(struct drm_device * dev, int flags)
 	return 0;
 }
 
-static int radeon_cp_cmdbuf(struct drm_device *dev, void *data, struct drm_file *file_priv)
+static int radeon_cp_cmdbuf(struct drm_device *dev, void *data,
+		struct drm_file *file_priv)
 {
 	drm_radeon_private_t *dev_priv = dev->dev_private;
 	struct drm_device_dma *dma = dev->dma;
 	struct drm_buf *buf = NULL;
+	drm_radeon_cmd_header_t stack_header;
 	int idx;
 	drm_radeon_kcmd_buffer_t *cmdbuf = data;
-	drm_radeon_cmd_header_t header;
-	int orig_nbox, orig_bufsz;
-	char *kbuf = NULL;
+	int orig_nbox;
 
 	LOCK_TEST_WITH_RETURN(dev, file_priv);
 
@@ -2871,17 +2881,16 @@ static int radeon_cp_cmdbuf(struct drm_device *dev, void *data, struct drm_file
 	 * races between checking values and using those values in other code,
 	 * and simply to avoid a lot of function calls to copy in data.
 	 */
-	orig_bufsz = cmdbuf->bufsz;
-	if (orig_bufsz != 0) {
-		kbuf = kmalloc(cmdbuf->bufsz, GFP_KERNEL);
-		if (kbuf == NULL)
-			return -ENOMEM;
-		if (DRM_COPY_FROM_USER(kbuf, (void __user *)cmdbuf->buf,
-				       cmdbuf->bufsz)) {
-			kfree(kbuf);
-			return -EFAULT;
-		}
-		cmdbuf->buf = kbuf;
+	if (cmdbuf->bufsz != 0) {
+		int rv;
+		void __user *buffer = cmdbuf->buffer;
+		rv = drm_buffer_alloc(&cmdbuf->buffer, cmdbuf->bufsz);
+		if (rv)
+			return rv;
+		rv = drm_buffer_copy_from_user(cmdbuf->buffer, buffer,
+						cmdbuf->bufsz);
+		if (rv)
+			return rv;
 	}
 
 	orig_nbox = cmdbuf->nbox;
@@ -2890,24 +2899,24 @@ static int radeon_cp_cmdbuf(struct drm_device *dev, void *data, struct drm_file
 		int temp;
 		temp = r300_do_cp_cmdbuf(dev, file_priv, cmdbuf);
 
-		if (orig_bufsz != 0)
-			kfree(kbuf);
+		if (cmdbuf->bufsz != 0)
+			drm_buffer_free(cmdbuf->buffer);
 
 		return temp;
 	}
 
 	/* microcode_version != r300 */
-	while (cmdbuf->bufsz >= sizeof(header)) {
+	while (drm_buffer_unprocessed(cmdbuf->buffer) >= sizeof(stack_header)) {
 
-		header.i = *(int *)cmdbuf->buf;
-		cmdbuf->buf += sizeof(header);
-		cmdbuf->bufsz -= sizeof(header);
+		drm_radeon_cmd_header_t *header;
+		header = drm_buffer_read_object(cmdbuf->buffer,
+				sizeof(stack_header), &stack_header);
 
-		switch (header.header.cmd_type) {
+		switch (header->header.cmd_type) {
 		case RADEON_CMD_PACKET:
 			DRM_DEBUG("RADEON_CMD_PACKET\n");
 			if (radeon_emit_packets
-			    (dev_priv, file_priv, header, cmdbuf)) {
+			    (dev_priv, file_priv, *header, cmdbuf)) {
 				DRM_ERROR("radeon_emit_packets failed\n");
 				goto err;
 			}
@@ -2915,7 +2924,7 @@ static int radeon_cp_cmdbuf(struct drm_device *dev, void *data, struct drm_file
 
 		case RADEON_CMD_SCALARS:
 			DRM_DEBUG("RADEON_CMD_SCALARS\n");
-			if (radeon_emit_scalars(dev_priv, header, cmdbuf)) {
+			if (radeon_emit_scalars(dev_priv, *header, cmdbuf)) {
 				DRM_ERROR("radeon_emit_scalars failed\n");
 				goto err;
 			}
@@ -2923,7 +2932,7 @@ static int radeon_cp_cmdbuf(struct drm_device *dev, void *data, struct drm_file
 
 		case RADEON_CMD_VECTORS:
 			DRM_DEBUG("RADEON_CMD_VECTORS\n");
-			if (radeon_emit_vectors(dev_priv, header, cmdbuf)) {
+			if (radeon_emit_vectors(dev_priv, *header, cmdbuf)) {
 				DRM_ERROR("radeon_emit_vectors failed\n");
 				goto err;
 			}
@@ -2931,7 +2940,7 @@ static int radeon_cp_cmdbuf(struct drm_device *dev, void *data, struct drm_file
 
 		case RADEON_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);
@@ -2968,7 +2977,7 @@ static int radeon_cp_cmdbuf(struct drm_device *dev, void *data, struct drm_file
 
 		case RADEON_CMD_SCALARS2:
 			DRM_DEBUG("RADEON_CMD_SCALARS2\n");
-			if (radeon_emit_scalars2(dev_priv, header, cmdbuf)) {
+			if (radeon_emit_scalars2(dev_priv, *header, cmdbuf)) {
 				DRM_ERROR("radeon_emit_scalars2 failed\n");
 				goto err;
 			}
@@ -2976,37 +2985,37 @@ static int radeon_cp_cmdbuf(struct drm_device *dev, void *data, struct drm_file
 
 		case RADEON_CMD_WAIT:
 			DRM_DEBUG("RADEON_CMD_WAIT\n");
-			if (radeon_emit_wait(dev, header.wait.flags)) {
+			if (radeon_emit_wait(dev, header->wait.flags)) {
 				DRM_ERROR("radeon_emit_wait failed\n");
 				goto err;
 			}
 			break;
 		case RADEON_CMD_VECLINEAR:
 			DRM_DEBUG("RADEON_CMD_VECLINEAR\n");
-			if (radeon_emit_veclinear(dev_priv, header, cmdbuf)) {
+			if (radeon_emit_veclinear(dev_priv, *header, cmdbuf)) {
 				DRM_ERROR("radeon_emit_veclinear failed\n");
 				goto err;
 			}
 			break;
 
 		default:
-			DRM_ERROR("bad cmd_type %d at %p\n",
-				  header.header.cmd_type,
-				  cmdbuf->buf - sizeof(header));
+			DRM_ERROR("bad cmd_type %d at byte %d\n",
+				  header->header.cmd_type,
+				  cmdbuf->buffer->iterator);
 			goto err;
 		}
 	}
 
-	if (orig_bufsz != 0)
-		kfree(kbuf);
+	if (cmdbuf->bufsz != 0)
+		drm_buffer_free(cmdbuf->buffer);
 
 	DRM_DEBUG("DONE\n");
 	COMMIT_RING();
 	return 0;
 
       err:
-	if (orig_bufsz != 0)
-		kfree(kbuf);
+	if (cmdbuf->bufsz != 0)
+		drm_buffer_free(cmdbuf->buffer);
 	return -EINVAL;
 }