Merge branch 'drm-next-3.8' of git://people.freedesktop.org/~agd5f/linux into drm-next

Fix regression, and some locking races, also as CS support
for the DMA engines.

* 'drm-next-3.8' of git://people.freedesktop.org/~agd5f/linux:
  radeon: fix regression with eviction since evict caching changes
  drm/radeon: add more pedantic checks in the CP DMA checker
  drm/radeon: bump version for CS ioctl support for async DMA
  drm/radeon: enable the async DMA rings in the CS ioctl
  drm/radeon: add VM CS parser support for async DMA on cayman/TN/SI
  drm/radeon/kms: add evergreen/cayman CS parser for async DMA (v2)
  drm/radeon/kms: add 6xx/7xx CS parser for async DMA (v2)
  drm/radeon: fix htile buffer size computation for command stream checker
  drm/radeon: fix fence locking in the pageflip callback
  drm/radeon: make indirect register access concurrency-safe
  drm/radeon: add W|RREG32_IDX for MM_INDEX|DATA based mmio accesss

drivers/gpu/drm/radeon/evergreen_cs.c

@@ -34,6 +34,8 @@
 #define MAX(a,b)                   (((a)>(b))?(a):(b))
 #define MIN(a,b)                   (((a)<(b))?(a):(b))
 
+int r600_dma_cs_next_reloc(struct radeon_cs_parser *p,
+			   struct radeon_cs_reloc **cs_reloc);
 static int evergreen_cs_packet_next_reloc(struct radeon_cs_parser *p,
 					  struct radeon_cs_reloc **cs_reloc);
 
@@ -507,20 +509,28 @@ static int evergreen_cs_track_validate_htile(struct radeon_cs_parser *p,
 		/* height is npipes htiles aligned == npipes * 8 pixel aligned */
 		nby = round_up(nby, track->npipes * 8);
 	} else {
+		/* always assume 8x8 htile */
+		/* align is htile align * 8, htile align vary according to
+		 * number of pipe and tile width and nby
+		 */
 		switch (track->npipes) {
 		case 8:
+			/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
 			nbx = round_up(nbx, 64 * 8);
 			nby = round_up(nby, 64 * 8);
 			break;
 		case 4:
+			/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
 			nbx = round_up(nbx, 64 * 8);
 			nby = round_up(nby, 32 * 8);
 			break;
 		case 2:
+			/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
 			nbx = round_up(nbx, 32 * 8);
 			nby = round_up(nby, 32 * 8);
 			break;
 		case 1:
+			/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
 			nbx = round_up(nbx, 32 * 8);
 			nby = round_up(nby, 16 * 8);
 			break;
@@ -531,9 +541,10 @@ static int evergreen_cs_track_validate_htile(struct radeon_cs_parser *p,
 		}
 	}
 	/* compute number of htile */
-	nbx = nbx / 8;
-	nby = nby / 8;
-	size = nbx * nby * 4;
+	nbx = nbx >> 3;
+	nby = nby >> 3;
+	/* size must be aligned on npipes * 2K boundary */
+	size = roundup(nbx * nby * 4, track->npipes * (2 << 10));
 	size += track->htile_offset;
 
 	if (size > radeon_bo_size(track->htile_bo)) {
@@ -1790,6 +1801,8 @@ static int evergreen_cs_check_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
 	case DB_HTILE_SURFACE:
 		/* 8x8 only */
 		track->htile_surface = radeon_get_ib_value(p, idx);
+		/* force 8x8 htile width and height */
+		ib[idx] |= 3;
 		track->db_dirty = true;
 		break;
 	case CB_IMMED0_BASE:
@@ -2243,6 +2256,18 @@ static int evergreen_packet3_check(struct radeon_cs_parser *p,
 		command = radeon_get_ib_value(p, idx+4);
 		size = command & 0x1fffff;
 		info = radeon_get_ib_value(p, idx+1);
+		if ((((info & 0x60000000) >> 29) != 0) || /* src = GDS or DATA */
+		    (((info & 0x00300000) >> 20) != 0) || /* dst = GDS */
+		    ((((info & 0x00300000) >> 20) == 0) &&
+		     (command & PACKET3_CP_DMA_CMD_DAS)) || /* dst = register */
+		    ((((info & 0x60000000) >> 29) == 0) &&
+		     (command & PACKET3_CP_DMA_CMD_SAS))) { /* src = register */
+			/* non mem to mem copies requires dw aligned count */
+			if (size % 4) {
+				DRM_ERROR("CP DMA command requires dw count alignment\n");
+				return -EINVAL;
+			}
+		}
 		if (command & PACKET3_CP_DMA_CMD_SAS) {
 			/* src address space is register */
 			/* GDS is ok */
@@ -2804,6 +2829,455 @@ int evergreen_cs_parse(struct radeon_cs_parser *p)
 	return 0;
 }
 
+/*
+ *  DMA
+ */
+
+#define GET_DMA_CMD(h) (((h) & 0xf0000000) >> 28)
+#define GET_DMA_COUNT(h) ((h) & 0x000fffff)
+#define GET_DMA_T(h) (((h) & 0x00800000) >> 23)
+#define GET_DMA_NEW(h) (((h) & 0x04000000) >> 26)
+#define GET_DMA_MISC(h) (((h) & 0x0700000) >> 20)
+
+/**
+ * evergreen_dma_cs_parse() - parse the DMA IB
+ * @p:		parser structure holding parsing context.
+ *
+ * Parses the DMA IB from the CS ioctl and updates
+ * the GPU addresses based on the reloc information and
+ * checks for errors. (Evergreen-Cayman)
+ * Returns 0 for success and an error on failure.
+ **/
+int evergreen_dma_cs_parse(struct radeon_cs_parser *p)
+{
+	struct radeon_cs_chunk *ib_chunk = &p->chunks[p->chunk_ib_idx];
+	struct radeon_cs_reloc *src_reloc, *dst_reloc, *dst2_reloc;
+	u32 header, cmd, count, tiled, new_cmd, misc;
+	volatile u32 *ib = p->ib.ptr;
+	u32 idx, idx_value;
+	u64 src_offset, dst_offset, dst2_offset;
+	int r;
+
+	do {
+		if (p->idx >= ib_chunk->length_dw) {
+			DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
+				  p->idx, ib_chunk->length_dw);
+			return -EINVAL;
+		}
+		idx = p->idx;
+		header = radeon_get_ib_value(p, idx);
+		cmd = GET_DMA_CMD(header);
+		count = GET_DMA_COUNT(header);
+		tiled = GET_DMA_T(header);
+		new_cmd = GET_DMA_NEW(header);
+		misc = GET_DMA_MISC(header);
+
+		switch (cmd) {
+		case DMA_PACKET_WRITE:
+			r = r600_dma_cs_next_reloc(p, &dst_reloc);
+			if (r) {
+				DRM_ERROR("bad DMA_PACKET_WRITE\n");
+				return -EINVAL;
+			}
+			if (tiled) {
+				dst_offset = ib[idx+1];
+				dst_offset <<= 8;
+
+				ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset >> 8);
+				p->idx += count + 7;
+			} else {
+				dst_offset = ib[idx+1];
+				dst_offset |= ((u64)(ib[idx+2] & 0xff)) << 32;
+
+				ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset & 0xfffffffc);
+				ib[idx+2] += upper_32_bits(dst_reloc->lobj.gpu_offset) & 0xff;
+				p->idx += count + 3;
+			}
+			if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
+				dev_warn(p->dev, "DMA write buffer too small (%llu %lu)\n",
+					 dst_offset, radeon_bo_size(dst_reloc->robj));
+				return -EINVAL;
+			}
+			break;
+		case DMA_PACKET_COPY:
+			r = r600_dma_cs_next_reloc(p, &src_reloc);
+			if (r) {
+				DRM_ERROR("bad DMA_PACKET_COPY\n");
+				return -EINVAL;
+			}
+			r = r600_dma_cs_next_reloc(p, &dst_reloc);
+			if (r) {
+				DRM_ERROR("bad DMA_PACKET_COPY\n");
+				return -EINVAL;
+			}
+			if (tiled) {
+				idx_value = radeon_get_ib_value(p, idx + 2);
+				if (new_cmd) {
+					switch (misc) {
+					case 0:
+						/* L2T, frame to fields */
+						if (idx_value & (1 << 31)) {
+							DRM_ERROR("bad L2T, frame to fields DMA_PACKET_COPY\n");
+							return -EINVAL;
+						}
+						r = r600_dma_cs_next_reloc(p, &dst2_reloc);
+						if (r) {
+							DRM_ERROR("bad L2T, frame to fields DMA_PACKET_COPY\n");
+							return -EINVAL;
+						}
+						dst_offset = ib[idx+1];
+						dst_offset <<= 8;
+						dst2_offset = ib[idx+2];
+						dst2_offset <<= 8;
+						src_offset = ib[idx+8];
+						src_offset |= ((u64)(ib[idx+9] & 0xff)) << 32;
+						if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
+							dev_warn(p->dev, "DMA L2T, frame to fields src buffer too small (%llu %lu)\n",
+								 src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
+							return -EINVAL;
+						}
+						if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
+							dev_warn(p->dev, "DMA L2T, frame to fields buffer too small (%llu %lu)\n",
+								 dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
+							return -EINVAL;
+						}
+						if ((dst2_offset + (count * 4)) > radeon_bo_size(dst2_reloc->robj)) {
+							dev_warn(p->dev, "DMA L2T, frame to fields buffer too small (%llu %lu)\n",
+								 dst2_offset + (count * 4), radeon_bo_size(dst2_reloc->robj));
+							return -EINVAL;
+						}
+						ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset >> 8);
+						ib[idx+2] += (u32)(dst2_reloc->lobj.gpu_offset >> 8);
+						ib[idx+8] += (u32)(src_reloc->lobj.gpu_offset & 0xfffffffc);
+						ib[idx+9] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
+						p->idx += 10;
+						break;
+					case 1:
+						/* L2T, T2L partial */
+						if (p->family < CHIP_CAYMAN) {
+							DRM_ERROR("L2T, T2L Partial is cayman only !\n");
+							return -EINVAL;
+						}
+						/* detile bit */
+						if (idx_value & (1 << 31)) {
+							/* tiled src, linear dst */
+							ib[idx+1] += (u32)(src_reloc->lobj.gpu_offset >> 8);
+
+							ib[idx+7] += (u32)(dst_reloc->lobj.gpu_offset & 0xfffffffc);
+							ib[idx+8] += upper_32_bits(dst_reloc->lobj.gpu_offset) & 0xff;
+						} else {
+							/* linear src, tiled dst */
+							ib[idx+7] += (u32)(src_reloc->lobj.gpu_offset & 0xfffffffc);
+							ib[idx+8] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
+
+							ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset >> 8);
+						}
+						p->idx += 12;
+						break;
+					case 3:
+						/* L2T, broadcast */
+						if (idx_value & (1 << 31)) {
+							DRM_ERROR("bad L2T, broadcast DMA_PACKET_COPY\n");
+							return -EINVAL;
+						}
+						r = r600_dma_cs_next_reloc(p, &dst2_reloc);
+						if (r) {
+							DRM_ERROR("bad L2T, broadcast DMA_PACKET_COPY\n");
+							return -EINVAL;
+						}
+						dst_offset = ib[idx+1];
+						dst_offset <<= 8;
+						dst2_offset = ib[idx+2];
+						dst2_offset <<= 8;
+						src_offset = ib[idx+8];
+						src_offset |= ((u64)(ib[idx+9] & 0xff)) << 32;
+						if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
+							dev_warn(p->dev, "DMA L2T, broadcast src buffer too small (%llu %lu)\n",
+								 src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
+							return -EINVAL;
+						}
+						if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
+							dev_warn(p->dev, "DMA L2T, broadcast dst buffer too small (%llu %lu)\n",
+								 dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
+							return -EINVAL;
+						}
+						if ((dst2_offset + (count * 4)) > radeon_bo_size(dst2_reloc->robj)) {
+							dev_warn(p->dev, "DMA L2T, broadcast dst2 buffer too small (%llu %lu)\n",
+								 dst2_offset + (count * 4), radeon_bo_size(dst2_reloc->robj));
+							return -EINVAL;
+						}
+						ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset >> 8);
+						ib[idx+2] += (u32)(dst2_reloc->lobj.gpu_offset >> 8);
+						ib[idx+8] += (u32)(src_reloc->lobj.gpu_offset & 0xfffffffc);
+						ib[idx+9] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
+						p->idx += 10;
+						break;
+					case 4:
+						/* L2T, T2L */
+						/* detile bit */
+						if (idx_value & (1 << 31)) {
+							/* tiled src, linear dst */
+							src_offset = ib[idx+1];
+							src_offset <<= 8;
+							ib[idx+1] += (u32)(src_reloc->lobj.gpu_offset >> 8);
+
+							dst_offset = ib[idx+7];
+							dst_offset |= ((u64)(ib[idx+8] & 0xff)) << 32;
+							ib[idx+7] += (u32)(dst_reloc->lobj.gpu_offset & 0xfffffffc);
+							ib[idx+8] += upper_32_bits(dst_reloc->lobj.gpu_offset) & 0xff;
+						} else {
+							/* linear src, tiled dst */
+							src_offset = ib[idx+7];
+							src_offset |= ((u64)(ib[idx+8] & 0xff)) << 32;
+							ib[idx+7] += (u32)(src_reloc->lobj.gpu_offset & 0xfffffffc);
+							ib[idx+8] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
+
+							dst_offset = ib[idx+1];
+							dst_offset <<= 8;
+							ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset >> 8);
+						}
+						if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
+							dev_warn(p->dev, "DMA L2T, T2L src buffer too small (%llu %lu)\n",
+								 src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
+							return -EINVAL;
+						}
+						if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
+							dev_warn(p->dev, "DMA L2T, T2L dst buffer too small (%llu %lu)\n",
+								 dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
+							return -EINVAL;
+						}
+						p->idx += 9;
+						break;
+					case 5:
+						/* T2T partial */
+						if (p->family < CHIP_CAYMAN) {
+							DRM_ERROR("L2T, T2L Partial is cayman only !\n");
+							return -EINVAL;
+						}
+						ib[idx+1] += (u32)(src_reloc->lobj.gpu_offset >> 8);
+						ib[idx+4] += (u32)(dst_reloc->lobj.gpu_offset >> 8);
+						p->idx += 13;
+						break;
+					case 7:
+						/* L2T, broadcast */
+						if (idx_value & (1 << 31)) {
+							DRM_ERROR("bad L2T, broadcast DMA_PACKET_COPY\n");
+							return -EINVAL;
+						}
+						r = r600_dma_cs_next_reloc(p, &dst2_reloc);
+						if (r) {
+							DRM_ERROR("bad L2T, broadcast DMA_PACKET_COPY\n");
+							return -EINVAL;
+						}
+						dst_offset = ib[idx+1];
+						dst_offset <<= 8;
+						dst2_offset = ib[idx+2];
+						dst2_offset <<= 8;
+						src_offset = ib[idx+8];
+						src_offset |= ((u64)(ib[idx+9] & 0xff)) << 32;
+						if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
+							dev_warn(p->dev, "DMA L2T, broadcast src buffer too small (%llu %lu)\n",
+								 src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
+							return -EINVAL;
+						}
+						if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
+							dev_warn(p->dev, "DMA L2T, broadcast dst buffer too small (%llu %lu)\n",
+								 dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
+							return -EINVAL;
+						}
+						if ((dst2_offset + (count * 4)) > radeon_bo_size(dst2_reloc->robj)) {
+							dev_warn(p->dev, "DMA L2T, broadcast dst2 buffer too small (%llu %lu)\n",
+								 dst2_offset + (count * 4), radeon_bo_size(dst2_reloc->robj));
+							return -EINVAL;
+						}
+						ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset >> 8);
+						ib[idx+2] += (u32)(dst2_reloc->lobj.gpu_offset >> 8);
+						ib[idx+8] += (u32)(src_reloc->lobj.gpu_offset & 0xfffffffc);
+						ib[idx+9] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
+						p->idx += 10;
+						break;
+					default:
+						DRM_ERROR("bad DMA_PACKET_COPY misc %u\n", misc);
+						return -EINVAL;
+					}
+				} else {
+					switch (misc) {
+					case 0:
+						/* detile bit */
+						if (idx_value & (1 << 31)) {
+							/* tiled src, linear dst */
+							src_offset = ib[idx+1];
+							src_offset <<= 8;
+							ib[idx+1] += (u32)(src_reloc->lobj.gpu_offset >> 8);
+
+							dst_offset = ib[idx+7];
+							dst_offset |= ((u64)(ib[idx+8] & 0xff)) << 32;
+							ib[idx+7] += (u32)(dst_reloc->lobj.gpu_offset & 0xfffffffc);
+							ib[idx+8] += upper_32_bits(dst_reloc->lobj.gpu_offset) & 0xff;
+						} else {
+							/* linear src, tiled dst */
+							src_offset = ib[idx+7];
+							src_offset |= ((u64)(ib[idx+8] & 0xff)) << 32;
+							ib[idx+7] += (u32)(src_reloc->lobj.gpu_offset & 0xfffffffc);
+							ib[idx+8] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
+
+							dst_offset = ib[idx+1];
+							dst_offset <<= 8;
+							ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset >> 8);
+						}
+						if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
+							dev_warn(p->dev, "DMA L2T, broadcast src buffer too small (%llu %lu)\n",
+								 src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
+							return -EINVAL;
+						}
+						if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
+							dev_warn(p->dev, "DMA L2T, broadcast dst buffer too small (%llu %lu)\n",
+								 dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
+							return -EINVAL;
+						}
+						p->idx += 9;
+						break;
+					default:
+						DRM_ERROR("bad DMA_PACKET_COPY misc %u\n", misc);
+						return -EINVAL;
+					}
+				}
+			} else {
+				if (new_cmd) {
+					switch (misc) {
+					case 0:
+						/* L2L, byte */
+						src_offset = ib[idx+2];
+						src_offset |= ((u64)(ib[idx+4] & 0xff)) << 32;
+						dst_offset = ib[idx+1];
+						dst_offset |= ((u64)(ib[idx+3] & 0xff)) << 32;
+						if ((src_offset + count) > radeon_bo_size(src_reloc->robj)) {
+							dev_warn(p->dev, "DMA L2L, byte src buffer too small (%llu %lu)\n",
+								 src_offset + count, radeon_bo_size(src_reloc->robj));
+							return -EINVAL;
+						}
+						if ((dst_offset + count) > radeon_bo_size(dst_reloc->robj)) {
+							dev_warn(p->dev, "DMA L2L, byte dst buffer too small (%llu %lu)\n",
+								 dst_offset + count, radeon_bo_size(dst_reloc->robj));
+							return -EINVAL;
+						}
+						ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset & 0xffffffff);
+						ib[idx+2] += (u32)(src_reloc->lobj.gpu_offset & 0xffffffff);
+						ib[idx+3] += upper_32_bits(dst_reloc->lobj.gpu_offset) & 0xff;
+						ib[idx+4] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
+						p->idx += 5;
+						break;
+					case 1:
+						/* L2L, partial */
+						if (p->family < CHIP_CAYMAN) {
+							DRM_ERROR("L2L Partial is cayman only !\n");
+							return -EINVAL;
+						}
+						ib[idx+1] += (u32)(src_reloc->lobj.gpu_offset & 0xffffffff);
+						ib[idx+2] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
+						ib[idx+4] += (u32)(dst_reloc->lobj.gpu_offset & 0xffffffff);
+						ib[idx+5] += upper_32_bits(dst_reloc->lobj.gpu_offset) & 0xff;
+
+						p->idx += 9;
+						break;
+					case 4:
+						/* L2L, dw, broadcast */
+						r = r600_dma_cs_next_reloc(p, &dst2_reloc);
+						if (r) {
+							DRM_ERROR("bad L2L, dw, broadcast DMA_PACKET_COPY\n");
+							return -EINVAL;
+						}
+						dst_offset = ib[idx+1];
+						dst_offset |= ((u64)(ib[idx+4] & 0xff)) << 32;
+						dst2_offset = ib[idx+2];
+						dst2_offset |= ((u64)(ib[idx+5] & 0xff)) << 32;
+						src_offset = ib[idx+3];
+						src_offset |= ((u64)(ib[idx+6] & 0xff)) << 32;
+						if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
+							dev_warn(p->dev, "DMA L2L, dw, broadcast src buffer too small (%llu %lu)\n",
+								 src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
+							return -EINVAL;
+						}
+						if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
+							dev_warn(p->dev, "DMA L2L, dw, broadcast dst buffer too small (%llu %lu)\n",
+								 dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
+							return -EINVAL;
+						}
+						if ((dst2_offset + (count * 4)) > radeon_bo_size(dst2_reloc->robj)) {
+							dev_warn(p->dev, "DMA L2L, dw, broadcast dst2 buffer too small (%llu %lu)\n",
+								 dst2_offset + (count * 4), radeon_bo_size(dst2_reloc->robj));
+							return -EINVAL;
+						}
+						ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset & 0xfffffffc);
+						ib[idx+2] += (u32)(dst2_reloc->lobj.gpu_offset & 0xfffffffc);
+						ib[idx+3] += (u32)(src_reloc->lobj.gpu_offset & 0xfffffffc);
+						ib[idx+4] += upper_32_bits(dst_reloc->lobj.gpu_offset) & 0xff;
+						ib[idx+5] += upper_32_bits(dst2_reloc->lobj.gpu_offset) & 0xff;
+						ib[idx+6] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
+						p->idx += 7;
+						break;
+					default:
+						DRM_ERROR("bad DMA_PACKET_COPY misc %u\n", misc);
+						return -EINVAL;
+					}
+				} else {
+					/* L2L, dw */
+					src_offset = ib[idx+2];
+					src_offset |= ((u64)(ib[idx+4] & 0xff)) << 32;
+					dst_offset = ib[idx+1];
+					dst_offset |= ((u64)(ib[idx+3] & 0xff)) << 32;
+					if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
+						dev_warn(p->dev, "DMA L2L, dw src buffer too small (%llu %lu)\n",
+							 src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
+						return -EINVAL;
+					}
+					if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
+						dev_warn(p->dev, "DMA L2L, dw dst buffer too small (%llu %lu)\n",
+							 dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
+						return -EINVAL;
+					}
+					ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset & 0xfffffffc);
+					ib[idx+2] += (u32)(src_reloc->lobj.gpu_offset & 0xfffffffc);
+					ib[idx+3] += upper_32_bits(dst_reloc->lobj.gpu_offset) & 0xff;
+					ib[idx+4] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
+					p->idx += 5;
+				}
+			}
+			break;
+		case DMA_PACKET_CONSTANT_FILL:
+			r = r600_dma_cs_next_reloc(p, &dst_reloc);
+			if (r) {
+				DRM_ERROR("bad DMA_PACKET_CONSTANT_FILL\n");
+				return -EINVAL;
+			}
+			dst_offset = ib[idx+1];
+			dst_offset |= ((u64)(ib[idx+3] & 0x00ff0000)) << 16;
+			if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
+				dev_warn(p->dev, "DMA constant fill buffer too small (%llu %lu)\n",
+					 dst_offset, radeon_bo_size(dst_reloc->robj));
+				return -EINVAL;
+			}
+			ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset & 0xfffffffc);
+			ib[idx+3] += (upper_32_bits(dst_reloc->lobj.gpu_offset) << 16) & 0x00ff0000;
+			p->idx += 4;
+			break;
+		case DMA_PACKET_NOP:
+			p->idx += 1;
+			break;
+		default:
+			DRM_ERROR("Unknown packet type %d at %d !\n", cmd, idx);
+			return -EINVAL;
+		}
+	} while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
+#if 0
+	for (r = 0; r < p->ib->length_dw; r++) {
+		printk(KERN_INFO "%05d  0x%08X\n", r, p->ib.ptr[r]);
+		mdelay(1);
+	}
+#endif
+	return 0;
+}
+
 /* vm parser */
 static bool evergreen_vm_reg_valid(u32 reg)
 {
@@ -3010,6 +3484,18 @@ static int evergreen_vm_packet3_check(struct radeon_device *rdev,
 	case PACKET3_CP_DMA:
 		command = ib[idx + 4];
 		info = ib[idx + 1];
+		if ((((info & 0x60000000) >> 29) != 0) || /* src = GDS or DATA */
+		    (((info & 0x00300000) >> 20) != 0) || /* dst = GDS */
+		    ((((info & 0x00300000) >> 20) == 0) &&
+		     (command & PACKET3_CP_DMA_CMD_DAS)) || /* dst = register */
+		    ((((info & 0x60000000) >> 29) == 0) &&
+		     (command & PACKET3_CP_DMA_CMD_SAS))) { /* src = register */
+			/* non mem to mem copies requires dw aligned count */
+			if ((command & 0x1fffff) % 4) {
+				DRM_ERROR("CP DMA command requires dw count alignment\n");
+				return -EINVAL;
+			}
+		}
 		if (command & PACKET3_CP_DMA_CMD_SAS) {
 			/* src address space is register */
 			if (((info & 0x60000000) >> 29) == 0) {
@@ -3094,3 +3580,114 @@ int evergreen_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib)
 
 	return ret;
 }
+
+/**
+ * evergreen_dma_ib_parse() - parse the DMA IB for VM
+ * @rdev: radeon_device pointer
+ * @ib:	radeon_ib pointer
+ *
+ * Parses the DMA IB from the VM CS ioctl
+ * checks for errors. (Cayman-SI)
+ * Returns 0 for success and an error on failure.
+ **/
+int evergreen_dma_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib)
+{
+	u32 idx = 0;
+	u32 header, cmd, count, tiled, new_cmd, misc;
+
+	do {
+		header = ib->ptr[idx];
+		cmd = GET_DMA_CMD(header);
+		count = GET_DMA_COUNT(header);
+		tiled = GET_DMA_T(header);
+		new_cmd = GET_DMA_NEW(header);
+		misc = GET_DMA_MISC(header);
+
+		switch (cmd) {
+		case DMA_PACKET_WRITE:
+			if (tiled)
+				idx += count + 7;
+			else
+				idx += count + 3;
+			break;
+		case DMA_PACKET_COPY:
+			if (tiled) {
+				if (new_cmd) {
+					switch (misc) {
+					case 0:
+						/* L2T, frame to fields */
+						idx += 10;
+						break;
+					case 1:
+						/* L2T, T2L partial */
+						idx += 12;
+						break;
+					case 3:
+						/* L2T, broadcast */
+						idx += 10;
+						break;
+					case 4:
+						/* L2T, T2L */
+						idx += 9;
+						break;
+					case 5:
+						/* T2T partial */
+						idx += 13;
+						break;
+					case 7:
+						/* L2T, broadcast */
+						idx += 10;
+						break;
+					default:
+						DRM_ERROR("bad DMA_PACKET_COPY misc %u\n", misc);
+						return -EINVAL;
+					}
+				} else {
+					switch (misc) {
+					case 0:
+						idx += 9;
+						break;
+					default:
+						DRM_ERROR("bad DMA_PACKET_COPY misc %u\n", misc);
+						return -EINVAL;
+					}
+				}
+			} else {
+				if (new_cmd) {
+					switch (misc) {
+					case 0:
+						/* L2L, byte */
+						idx += 5;
+						break;
+					case 1:
+						/* L2L, partial */
+						idx += 9;
+						break;
+					case 4:
+						/* L2L, dw, broadcast */
+						idx += 7;
+						break;
+					default:
+						DRM_ERROR("bad DMA_PACKET_COPY misc %u\n", misc);
+						return -EINVAL;
+					}
+				} else {
+					/* L2L, dw */
+					idx += 5;
+				}
+			}
+			break;
+		case DMA_PACKET_CONSTANT_FILL:
+			idx += 4;
+			break;
+		case DMA_PACKET_NOP:
+			idx += 1;
+			break;
+		default:
+			DRM_ERROR("Unknown packet type %d at %d !\n", cmd, idx);
+			return -EINVAL;
+		}
+	} while (idx < ib->length_dw);
+
+	return 0;
+}

drivers/gpu/drm/radeon/r100.c

@@ -4135,23 +4135,36 @@ int r100_init(struct radeon_device *rdev)
 	return 0;
 }
 
-uint32_t r100_mm_rreg(struct radeon_device *rdev, uint32_t reg)
+uint32_t r100_mm_rreg(struct radeon_device *rdev, uint32_t reg,
+		      bool always_indirect)
 {
-	if (reg < rdev->rmmio_size)
+	if (reg < rdev->rmmio_size && !always_indirect)
 		return readl(((void __iomem *)rdev->rmmio) + reg);
 	else {
+		unsigned long flags;
+		uint32_t ret;
+
+		spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
 		writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
-		return readl(((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
+		ret = readl(((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
+		spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
+
+		return ret;
 	}
 }
 
-void r100_mm_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
+void r100_mm_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v,
+		  bool always_indirect)
 {
-	if (reg < rdev->rmmio_size)
+	if (reg < rdev->rmmio_size && !always_indirect)
 		writel(v, ((void __iomem *)rdev->rmmio) + reg);
 	else {
+		unsigned long flags;
+
+		spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
 		writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
 		writel(v, ((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
+		spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
 	}
 }
 

drivers/gpu/drm/radeon/r600_cs.c

@@ -657,87 +657,30 @@ static int r600_cs_track_validate_db(struct radeon_cs_parser *p)
 			/* nby is npipes htiles aligned == npipes * 8 pixel aligned */
 			nby = round_up(nby, track->npipes * 8);
 		} else {
-			/* htile widht & nby (8 or 4) make 2 bits number */
-			tmp = track->htile_surface & 3;
+			/* always assume 8x8 htile */
 			/* align is htile align * 8, htile align vary according to
 			 * number of pipe and tile width and nby
 			 */
 			switch (track->npipes) {
 			case 8:
-				switch (tmp) {
-				case 3:	/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
-					nbx = round_up(nbx, 64 * 8);
-					nby = round_up(nby, 64 * 8);
-					break;
-				case 2:	/* HTILE_WIDTH = 4 & HTILE_HEIGHT = 8*/
-				case 1:	/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 4*/
-					nbx = round_up(nbx, 64 * 8);
-					nby = round_up(nby, 32 * 8);
-					break;
-				case 0:	/* HTILE_WIDTH = 4 & HTILE_HEIGHT = 4*/
-					nbx = round_up(nbx, 32 * 8);
-					nby = round_up(nby, 32 * 8);
-					break;
-				default:
-					return -EINVAL;
-				}
+				/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
+				nbx = round_up(nbx, 64 * 8);
+				nby = round_up(nby, 64 * 8);
 				break;
 			case 4:
-				switch (tmp) {
-				case 3:	/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
-					nbx = round_up(nbx, 64 * 8);
-					nby = round_up(nby, 32 * 8);
-					break;
-				case 2:	/* HTILE_WIDTH = 4 & HTILE_HEIGHT = 8*/
-				case 1:	/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 4*/
-					nbx = round_up(nbx, 32 * 8);
-					nby = round_up(nby, 32 * 8);
-					break;
-				case 0:	/* HTILE_WIDTH = 4 & HTILE_HEIGHT = 4*/
-					nbx = round_up(nbx, 32 * 8);
-					nby = round_up(nby, 16 * 8);
-					break;
-				default:
-					return -EINVAL;
-				}
+				/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
+				nbx = round_up(nbx, 64 * 8);
+				nby = round_up(nby, 32 * 8);
 				break;
 			case 2:
-				switch (tmp) {
-				case 3:	/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
-					nbx = round_up(nbx, 32 * 8);
-					nby = round_up(nby, 32 * 8);
-					break;
-				case 2:	/* HTILE_WIDTH = 4 & HTILE_HEIGHT = 8*/
-				case 1:	/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 4*/
-					nbx = round_up(nbx, 32 * 8);
-					nby = round_up(nby, 16 * 8);
-					break;
-				case 0:	/* HTILE_WIDTH = 4 & HTILE_HEIGHT = 4*/
-					nbx = round_up(nbx, 16 * 8);
-					nby = round_up(nby, 16 * 8);
-					break;
-				default:
-					return -EINVAL;
-				}
+				/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
+				nbx = round_up(nbx, 32 * 8);
+				nby = round_up(nby, 32 * 8);
 				break;
 			case 1:
-				switch (tmp) {
-				case 3:	/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
-					nbx = round_up(nbx, 32 * 8);
-					nby = round_up(nby, 16 * 8);
-					break;
-				case 2:	/* HTILE_WIDTH = 4 & HTILE_HEIGHT = 8*/
-				case 1:	/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 4*/
-					nbx = round_up(nbx, 16 * 8);
-					nby = round_up(nby, 16 * 8);
-					break;
-				case 0:	/* HTILE_WIDTH = 4 & HTILE_HEIGHT = 4*/
-					nbx = round_up(nbx, 16 * 8);
-					nby = round_up(nby, 8 * 8);
-					break;
-				default:
-					return -EINVAL;
-				}
+				/* HTILE_WIDTH = 8 & HTILE_HEIGHT = 8*/
+				nbx = round_up(nbx, 32 * 8);
+				nby = round_up(nby, 16 * 8);
 				break;
 			default:
 				dev_warn(p->dev, "%s:%d invalid num pipes %d\n",
@@ -746,9 +689,10 @@ static int r600_cs_track_validate_db(struct radeon_cs_parser *p)
 			}
 		}
 		/* compute number of htile */
-		nbx = G_028D24_HTILE_WIDTH(track->htile_surface) ? nbx / 8 : nbx / 4;
-		nby = G_028D24_HTILE_HEIGHT(track->htile_surface) ? nby / 8 : nby / 4;
-		size = nbx * nby * 4;
+		nbx = nbx >> 3;
+		nby = nby >> 3;
+		/* size must be aligned on npipes * 2K boundary */
+		size = roundup(nbx * nby * 4, track->npipes * (2 << 10));
 		size += track->htile_offset;
 
 		if (size > radeon_bo_size(track->htile_bo)) {
@@ -1492,6 +1436,8 @@ static int r600_cs_check_reg(struct radeon_cs_parser *p, u32 reg, u32 idx)
 		break;
 	case DB_HTILE_SURFACE:
 		track->htile_surface = radeon_get_ib_value(p, idx);
+		/* force 8x8 htile width and height */
+		ib[idx] |= 3;
 		track->db_dirty = true;
 		break;
 	case SQ_PGM_START_FS:
@@ -2568,3 +2514,196 @@ void r600_cs_legacy_init(void)
 {
 	r600_cs_packet_next_reloc = &r600_cs_packet_next_reloc_nomm;
 }
+
+/*
+ *  DMA
+ */
+/**
+ * r600_dma_cs_next_reloc() - parse next reloc
+ * @p:		parser structure holding parsing context.
+ * @cs_reloc:		reloc informations
+ *
+ * Return the next reloc, do bo validation and compute
+ * GPU offset using the provided start.
+ **/
+int r600_dma_cs_next_reloc(struct radeon_cs_parser *p,
+			   struct radeon_cs_reloc **cs_reloc)
+{
+	struct radeon_cs_chunk *relocs_chunk;
+	unsigned idx;
+
+	if (p->chunk_relocs_idx == -1) {
+		DRM_ERROR("No relocation chunk !\n");
+		return -EINVAL;
+	}
+	*cs_reloc = NULL;
+	relocs_chunk = &p->chunks[p->chunk_relocs_idx];
+	idx = p->dma_reloc_idx;
+	if (idx >= relocs_chunk->length_dw) {
+		DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
+			  idx, relocs_chunk->length_dw);
+		return -EINVAL;
+	}
+	*cs_reloc = p->relocs_ptr[idx];
+	p->dma_reloc_idx++;
+	return 0;
+}
+
+#define GET_DMA_CMD(h) (((h) & 0xf0000000) >> 28)
+#define GET_DMA_COUNT(h) ((h) & 0x0000ffff)
+#define GET_DMA_T(h) (((h) & 0x00800000) >> 23)
+
+/**
+ * r600_dma_cs_parse() - parse the DMA IB
+ * @p:		parser structure holding parsing context.
+ *
+ * Parses the DMA IB from the CS ioctl and updates
+ * the GPU addresses based on the reloc information and
+ * checks for errors. (R6xx-R7xx)
+ * Returns 0 for success and an error on failure.
+ **/
+int r600_dma_cs_parse(struct radeon_cs_parser *p)
+{
+	struct radeon_cs_chunk *ib_chunk = &p->chunks[p->chunk_ib_idx];
+	struct radeon_cs_reloc *src_reloc, *dst_reloc;
+	u32 header, cmd, count, tiled;
+	volatile u32 *ib = p->ib.ptr;
+	u32 idx, idx_value;
+	u64 src_offset, dst_offset;
+	int r;
+
+	do {
+		if (p->idx >= ib_chunk->length_dw) {
+			DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
+				  p->idx, ib_chunk->length_dw);
+			return -EINVAL;
+		}
+		idx = p->idx;
+		header = radeon_get_ib_value(p, idx);
+		cmd = GET_DMA_CMD(header);
+		count = GET_DMA_COUNT(header);
+		tiled = GET_DMA_T(header);
+
+		switch (cmd) {
+		case DMA_PACKET_WRITE:
+			r = r600_dma_cs_next_reloc(p, &dst_reloc);
+			if (r) {
+				DRM_ERROR("bad DMA_PACKET_WRITE\n");
+				return -EINVAL;
+			}
+			if (tiled) {
+				dst_offset = ib[idx+1];
+				dst_offset <<= 8;
+
+				ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset >> 8);
+				p->idx += count + 5;
+			} else {
+				dst_offset = ib[idx+1];
+				dst_offset |= ((u64)(ib[idx+2] & 0xff)) << 32;
+
+				ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset & 0xfffffffc);
+				ib[idx+2] += upper_32_bits(dst_reloc->lobj.gpu_offset) & 0xff;
+				p->idx += count + 3;
+			}
+			if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
+				dev_warn(p->dev, "DMA write buffer too small (%llu %lu)\n",
+					 dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
+				return -EINVAL;
+			}
+			break;
+		case DMA_PACKET_COPY:
+			r = r600_dma_cs_next_reloc(p, &src_reloc);
+			if (r) {
+				DRM_ERROR("bad DMA_PACKET_COPY\n");
+				return -EINVAL;
+			}
+			r = r600_dma_cs_next_reloc(p, &dst_reloc);
+			if (r) {
+				DRM_ERROR("bad DMA_PACKET_COPY\n");
+				return -EINVAL;
+			}
+			if (tiled) {
+				idx_value = radeon_get_ib_value(p, idx + 2);
+				/* detile bit */
+				if (idx_value & (1 << 31)) {
+					/* tiled src, linear dst */
+					src_offset = ib[idx+1];
+					src_offset <<= 8;
+					ib[idx+1] += (u32)(src_reloc->lobj.gpu_offset >> 8);
+
+					dst_offset = ib[idx+5];
+					dst_offset |= ((u64)(ib[idx+6] & 0xff)) << 32;
+					ib[idx+5] += (u32)(dst_reloc->lobj.gpu_offset & 0xfffffffc);
+					ib[idx+6] += upper_32_bits(dst_reloc->lobj.gpu_offset) & 0xff;
+				} else {
+					/* linear src, tiled dst */
+					src_offset = ib[idx+5];
+					src_offset |= ((u64)(ib[idx+6] & 0xff)) << 32;
+					ib[idx+5] += (u32)(src_reloc->lobj.gpu_offset & 0xfffffffc);
+					ib[idx+6] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
+
+					dst_offset = ib[idx+1];
+					dst_offset <<= 8;
+					ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset >> 8);
+				}
+				p->idx += 7;
+			} else {
+				src_offset = ib[idx+2];
+				src_offset |= ((u64)(ib[idx+4] & 0xff)) << 32;
+				dst_offset = ib[idx+1];
+				dst_offset |= ((u64)(ib[idx+3] & 0xff)) << 32;
+
+				ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset & 0xfffffffc);
+				ib[idx+2] += (u32)(src_reloc->lobj.gpu_offset & 0xfffffffc);
+				ib[idx+3] += upper_32_bits(dst_reloc->lobj.gpu_offset) & 0xff;
+				ib[idx+4] += upper_32_bits(src_reloc->lobj.gpu_offset) & 0xff;
+				p->idx += 5;
+			}
+			if ((src_offset + (count * 4)) > radeon_bo_size(src_reloc->robj)) {
+				dev_warn(p->dev, "DMA copy src buffer too small (%llu %lu)\n",
+					 src_offset + (count * 4), radeon_bo_size(src_reloc->robj));
+				return -EINVAL;
+			}
+			if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
+				dev_warn(p->dev, "DMA write dst buffer too small (%llu %lu)\n",
+					 dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
+				return -EINVAL;
+			}
+			break;
+		case DMA_PACKET_CONSTANT_FILL:
+			if (p->family < CHIP_RV770) {
+				DRM_ERROR("Constant Fill is 7xx only !\n");
+				return -EINVAL;
+			}
+			r = r600_dma_cs_next_reloc(p, &dst_reloc);
+			if (r) {
+				DRM_ERROR("bad DMA_PACKET_WRITE\n");
+				return -EINVAL;
+			}
+			dst_offset = ib[idx+1];
+			dst_offset |= ((u64)(ib[idx+3] & 0x00ff0000)) << 16;
+			if ((dst_offset + (count * 4)) > radeon_bo_size(dst_reloc->robj)) {
+				dev_warn(p->dev, "DMA constant fill buffer too small (%llu %lu)\n",
+					 dst_offset + (count * 4), radeon_bo_size(dst_reloc->robj));
+				return -EINVAL;
+			}
+			ib[idx+1] += (u32)(dst_reloc->lobj.gpu_offset & 0xfffffffc);
+			ib[idx+3] += (upper_32_bits(dst_reloc->lobj.gpu_offset) << 16) & 0x00ff0000;
+			p->idx += 4;
+			break;
+		case DMA_PACKET_NOP:
+			p->idx += 1;
+			break;
+		default:
+			DRM_ERROR("Unknown packet type %d at %d !\n", cmd, idx);
+			return -EINVAL;
+		}
+	} while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
+#if 0
+	for (r = 0; r < p->ib->length_dw; r++) {
+		printk(KERN_INFO "%05d  0x%08X\n", r, p->ib.ptr[r]);
+		mdelay(1);
+	}
+#endif
+	return 0;
+}

drivers/gpu/drm/radeon/radeon.h

@@ -839,6 +839,7 @@ struct radeon_cs_parser {
 	struct radeon_cs_reloc	*relocs;
 	struct radeon_cs_reloc	**relocs_ptr;
 	struct list_head	validated;
+	unsigned		dma_reloc_idx;
 	/* indices of various chunks */
 	int			chunk_ib_idx;
 	int			chunk_relocs_idx;
@@ -1556,6 +1557,8 @@ struct radeon_device {
 	/* Register mmio */
 	resource_size_t			rmmio_base;
 	resource_size_t			rmmio_size;
+	/* protects concurrent MM_INDEX/DATA based register access */
+	spinlock_t mmio_idx_lock;
 	void __iomem			*rmmio;
 	radeon_rreg_t			mc_rreg;
 	radeon_wreg_t			mc_wreg;
@@ -1631,8 +1634,10 @@ int radeon_device_init(struct radeon_device *rdev,
 void radeon_device_fini(struct radeon_device *rdev);
 int radeon_gpu_wait_for_idle(struct radeon_device *rdev);
 
-uint32_t r100_mm_rreg(struct radeon_device *rdev, uint32_t reg);
-void r100_mm_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v);
+uint32_t r100_mm_rreg(struct radeon_device *rdev, uint32_t reg,
+		      bool always_indirect);
+void r100_mm_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v,
+		  bool always_indirect);
 u32 r100_io_rreg(struct radeon_device *rdev, u32 reg);
 void r100_io_wreg(struct radeon_device *rdev, u32 reg, u32 v);
 
@@ -1648,9 +1653,11 @@ void r100_io_wreg(struct radeon_device *rdev, u32 reg, u32 v);
 #define WREG8(reg, v) writeb(v, (rdev->rmmio) + (reg))
 #define RREG16(reg) readw((rdev->rmmio) + (reg))
 #define WREG16(reg, v) writew(v, (rdev->rmmio) + (reg))
-#define RREG32(reg) r100_mm_rreg(rdev, (reg))
-#define DREG32(reg) printk(KERN_INFO "REGISTER: " #reg " : 0x%08X\n", r100_mm_rreg(rdev, (reg)))
-#define WREG32(reg, v) r100_mm_wreg(rdev, (reg), (v))
+#define RREG32(reg) r100_mm_rreg(rdev, (reg), false)
+#define RREG32_IDX(reg) r100_mm_rreg(rdev, (reg), true)
+#define DREG32(reg) printk(KERN_INFO "REGISTER: " #reg " : 0x%08X\n", r100_mm_rreg(rdev, (reg), false))
+#define WREG32(reg, v) r100_mm_wreg(rdev, (reg), (v), false)
+#define WREG32_IDX(reg, v) r100_mm_wreg(rdev, (reg), (v), true)
 #define REG_SET(FIELD, v) (((v) << FIELD##_SHIFT) & FIELD##_MASK)
 #define REG_GET(FIELD, v) (((v) << FIELD##_SHIFT) & FIELD##_MASK)
 #define RREG32_PLL(reg) rdev->pll_rreg(rdev, (reg))
@@ -1675,7 +1682,7 @@ void r100_io_wreg(struct radeon_device *rdev, u32 reg, u32 v);
 		tmp_ |= ((val) & ~(mask));			\
 		WREG32_PLL(reg, tmp_);				\
 	} while (0)
-#define DREG32_SYS(sqf, rdev, reg) seq_printf((sqf), #reg " : 0x%08X\n", r100_mm_rreg((rdev), (reg)))
+#define DREG32_SYS(sqf, rdev, reg) seq_printf((sqf), #reg " : 0x%08X\n", r100_mm_rreg((rdev), (reg), false))
 #define RREG32_IO(reg) r100_io_rreg(rdev, (reg))
 #define WREG32_IO(reg, v) r100_io_wreg(rdev, (reg), (v))
 

drivers/gpu/drm/radeon/radeon_asic.c

@@ -952,7 +952,7 @@ static struct radeon_asic r600_asic = {
 			.ib_execute = &r600_dma_ring_ib_execute,
 			.emit_fence = &r600_dma_fence_ring_emit,
 			.emit_semaphore = &r600_dma_semaphore_ring_emit,
-			.cs_parse = NULL,
+			.cs_parse = &r600_dma_cs_parse,
 			.ring_test = &r600_dma_ring_test,
 			.ib_test = &r600_dma_ib_test,
 			.is_lockup = &r600_dma_is_lockup,
@@ -1036,7 +1036,7 @@ static struct radeon_asic rs780_asic = {
 			.ib_execute = &r600_dma_ring_ib_execute,
 			.emit_fence = &r600_dma_fence_ring_emit,
 			.emit_semaphore = &r600_dma_semaphore_ring_emit,
-			.cs_parse = NULL,
+			.cs_parse = &r600_dma_cs_parse,
 			.ring_test = &r600_dma_ring_test,
 			.ib_test = &r600_dma_ib_test,
 			.is_lockup = &r600_dma_is_lockup,
@@ -1120,7 +1120,7 @@ static struct radeon_asic rv770_asic = {
 			.ib_execute = &r600_dma_ring_ib_execute,
 			.emit_fence = &r600_dma_fence_ring_emit,
 			.emit_semaphore = &r600_dma_semaphore_ring_emit,
-			.cs_parse = NULL,
+			.cs_parse = &r600_dma_cs_parse,
 			.ring_test = &r600_dma_ring_test,
 			.ib_test = &r600_dma_ib_test,
 			.is_lockup = &r600_dma_is_lockup,
@@ -1204,7 +1204,7 @@ static struct radeon_asic evergreen_asic = {
 			.ib_execute = &evergreen_dma_ring_ib_execute,
 			.emit_fence = &evergreen_dma_fence_ring_emit,
 			.emit_semaphore = &r600_dma_semaphore_ring_emit,
-			.cs_parse = NULL,
+			.cs_parse = &evergreen_dma_cs_parse,
 			.ring_test = &r600_dma_ring_test,
 			.ib_test = &r600_dma_ib_test,
 			.is_lockup = &r600_dma_is_lockup,
@@ -1288,7 +1288,7 @@ static struct radeon_asic sumo_asic = {
 			.ib_execute = &evergreen_dma_ring_ib_execute,
 			.emit_fence = &evergreen_dma_fence_ring_emit,
 			.emit_semaphore = &r600_dma_semaphore_ring_emit,
-			.cs_parse = NULL,
+			.cs_parse = &evergreen_dma_cs_parse,
 			.ring_test = &r600_dma_ring_test,
 			.ib_test = &r600_dma_ib_test,
 			.is_lockup = &r600_dma_is_lockup,
@@ -1372,7 +1372,7 @@ static struct radeon_asic btc_asic = {
 			.ib_execute = &evergreen_dma_ring_ib_execute,
 			.emit_fence = &evergreen_dma_fence_ring_emit,
 			.emit_semaphore = &r600_dma_semaphore_ring_emit,
-			.cs_parse = NULL,
+			.cs_parse = &evergreen_dma_cs_parse,
 			.ring_test = &r600_dma_ring_test,
 			.ib_test = &r600_dma_ib_test,
 			.is_lockup = &r600_dma_is_lockup,
@@ -1484,9 +1484,10 @@ static struct radeon_asic cayman_asic = {
 		},
 		[R600_RING_TYPE_DMA_INDEX] = {
 			.ib_execute = &cayman_dma_ring_ib_execute,
+			.ib_parse = &evergreen_dma_ib_parse,
 			.emit_fence = &evergreen_dma_fence_ring_emit,
 			.emit_semaphore = &r600_dma_semaphore_ring_emit,
-			.cs_parse = NULL,
+			.cs_parse = &evergreen_dma_cs_parse,
 			.ring_test = &r600_dma_ring_test,
 			.ib_test = &r600_dma_ib_test,
 			.is_lockup = &cayman_dma_is_lockup,
@@ -1494,9 +1495,10 @@ static struct radeon_asic cayman_asic = {
 		},
 		[CAYMAN_RING_TYPE_DMA1_INDEX] = {
 			.ib_execute = &cayman_dma_ring_ib_execute,
+			.ib_parse = &evergreen_dma_ib_parse,
 			.emit_fence = &evergreen_dma_fence_ring_emit,
 			.emit_semaphore = &r600_dma_semaphore_ring_emit,
-			.cs_parse = NULL,
+			.cs_parse = &evergreen_dma_cs_parse,
 			.ring_test = &r600_dma_ring_test,
 			.ib_test = &r600_dma_ib_test,
 			.is_lockup = &cayman_dma_is_lockup,
@@ -1609,9 +1611,10 @@ static struct radeon_asic trinity_asic = {
 		},
 		[R600_RING_TYPE_DMA_INDEX] = {
 			.ib_execute = &cayman_dma_ring_ib_execute,
+			.ib_parse = &evergreen_dma_ib_parse,
 			.emit_fence = &evergreen_dma_fence_ring_emit,
 			.emit_semaphore = &r600_dma_semaphore_ring_emit,
-			.cs_parse = NULL,
+			.cs_parse = &evergreen_dma_cs_parse,
 			.ring_test = &r600_dma_ring_test,
 			.ib_test = &r600_dma_ib_test,
 			.is_lockup = &cayman_dma_is_lockup,
@@ -1619,9 +1622,10 @@ static struct radeon_asic trinity_asic = {
 		},
 		[CAYMAN_RING_TYPE_DMA1_INDEX] = {
 			.ib_execute = &cayman_dma_ring_ib_execute,
+			.ib_parse = &evergreen_dma_ib_parse,
 			.emit_fence = &evergreen_dma_fence_ring_emit,
 			.emit_semaphore = &r600_dma_semaphore_ring_emit,
-			.cs_parse = NULL,
+			.cs_parse = &evergreen_dma_cs_parse,
 			.ring_test = &r600_dma_ring_test,
 			.ib_test = &r600_dma_ib_test,
 			.is_lockup = &cayman_dma_is_lockup,
@@ -1734,6 +1738,7 @@ static struct radeon_asic si_asic = {
 		},
 		[R600_RING_TYPE_DMA_INDEX] = {
 			.ib_execute = &cayman_dma_ring_ib_execute,
+			.ib_parse = &evergreen_dma_ib_parse,
 			.emit_fence = &evergreen_dma_fence_ring_emit,
 			.emit_semaphore = &r600_dma_semaphore_ring_emit,
 			.cs_parse = NULL,
@@ -1744,6 +1749,7 @@ static struct radeon_asic si_asic = {
 		},
 		[CAYMAN_RING_TYPE_DMA1_INDEX] = {
 			.ib_execute = &cayman_dma_ring_ib_execute,
+			.ib_parse = &evergreen_dma_ib_parse,
 			.emit_fence = &evergreen_dma_fence_ring_emit,
 			.emit_semaphore = &r600_dma_semaphore_ring_emit,
 			.cs_parse = NULL,

drivers/gpu/drm/radeon/radeon_asic.h

@@ -304,6 +304,7 @@ void r600_pcie_gart_tlb_flush(struct radeon_device *rdev);
 uint32_t r600_pciep_rreg(struct radeon_device *rdev, uint32_t reg);
 void r600_pciep_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v);
 int r600_cs_parse(struct radeon_cs_parser *p);
+int r600_dma_cs_parse(struct radeon_cs_parser *p);
 void r600_fence_ring_emit(struct radeon_device *rdev,
 			  struct radeon_fence *fence);
 void r600_semaphore_ring_emit(struct radeon_device *rdev,
@@ -430,6 +431,7 @@ u32 evergreen_get_vblank_counter(struct radeon_device *rdev, int crtc);
 int evergreen_irq_set(struct radeon_device *rdev);
 int evergreen_irq_process(struct radeon_device *rdev);
 extern int evergreen_cs_parse(struct radeon_cs_parser *p);
+extern int evergreen_dma_cs_parse(struct radeon_cs_parser *p);
 extern void evergreen_pm_misc(struct radeon_device *rdev);
 extern void evergreen_pm_prepare(struct radeon_device *rdev);
 extern void evergreen_pm_finish(struct radeon_device *rdev);
@@ -471,6 +473,7 @@ void cayman_vm_set_page(struct radeon_device *rdev, uint64_t pe,
 			uint64_t addr, unsigned count,
 			uint32_t incr, uint32_t flags);
 int evergreen_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib);
+int evergreen_dma_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib);
 void cayman_dma_ring_ib_execute(struct radeon_device *rdev,
 				struct radeon_ib *ib);
 bool cayman_dma_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring);

drivers/gpu/drm/radeon/radeon_combios.c

@@ -3246,11 +3246,9 @@ static uint32_t combios_detect_ram(struct drm_device *dev, int ram,
 	while (ram--) {
 		addr = ram * 1024 * 1024;
 		/* write to each page */
-		WREG32(RADEON_MM_INDEX, (addr) | RADEON_MM_APER);
-		WREG32(RADEON_MM_DATA, 0xdeadbeef);
+		WREG32_IDX((addr) | RADEON_MM_APER, 0xdeadbeef);
 		/* read back and verify */
-		WREG32(RADEON_MM_INDEX, (addr) | RADEON_MM_APER);
-		if (RREG32(RADEON_MM_DATA) != 0xdeadbeef)
+		if (RREG32_IDX((addr) | RADEON_MM_APER) != 0xdeadbeef)
 			return 0;
 	}
 

drivers/gpu/drm/radeon/radeon_cp.c

@@ -116,20 +116,6 @@ u32 radeon_get_scratch(drm_radeon_private_t *dev_priv, int index)
 	}
 }
 
-u32 RADEON_READ_MM(drm_radeon_private_t *dev_priv, int addr)
-{
-	u32 ret;
-
-	if (addr < 0x10000)
-		ret = DRM_READ32(dev_priv->mmio, addr);
-	else {
-		DRM_WRITE32(dev_priv->mmio, RADEON_MM_INDEX, addr);
-		ret = DRM_READ32(dev_priv->mmio, RADEON_MM_DATA);
-	}
-
-	return ret;
-}
-
 static u32 R500_READ_MCIND(drm_radeon_private_t *dev_priv, int addr)
 {
 	u32 ret;

drivers/gpu/drm/radeon/radeon_cs.c

@@ -43,6 +43,7 @@ static int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
 		return 0;
 	}
 	chunk = &p->chunks[p->chunk_relocs_idx];
+	p->dma_reloc_idx = 0;
 	/* FIXME: we assume that each relocs use 4 dwords */
 	p->nrelocs = chunk->length_dw / 4;
 	p->relocs_ptr = kcalloc(p->nrelocs, sizeof(void *), GFP_KERNEL);
@@ -111,6 +112,18 @@ static int radeon_cs_get_ring(struct radeon_cs_parser *p, u32 ring, s32 priority
 		} else
 			p->ring = RADEON_RING_TYPE_GFX_INDEX;
 		break;
+	case RADEON_CS_RING_DMA:
+		if (p->rdev->family >= CHIP_CAYMAN) {
+			if (p->priority > 0)
+				p->ring = R600_RING_TYPE_DMA_INDEX;
+			else
+				p->ring = CAYMAN_RING_TYPE_DMA1_INDEX;
+		} else if (p->rdev->family >= CHIP_R600) {
+			p->ring = R600_RING_TYPE_DMA_INDEX;
+		} else {
+			return -EINVAL;
+		}
+		break;
 	}
 	return 0;
 }

drivers/gpu/drm/radeon/radeon_cursor.c

@@ -66,24 +66,25 @@ static void radeon_hide_cursor(struct drm_crtc *crtc)
 	struct radeon_device *rdev = crtc->dev->dev_private;
 
 	if (ASIC_IS_DCE4(rdev)) {
-		WREG32(RADEON_MM_INDEX, EVERGREEN_CUR_CONTROL + radeon_crtc->crtc_offset);
-		WREG32(RADEON_MM_DATA, EVERGREEN_CURSOR_MODE(EVERGREEN_CURSOR_24_8_PRE_MULT) |
-		       EVERGREEN_CURSOR_URGENT_CONTROL(EVERGREEN_CURSOR_URGENT_1_2));
+		WREG32_IDX(EVERGREEN_CUR_CONTROL + radeon_crtc->crtc_offset,
+			   EVERGREEN_CURSOR_MODE(EVERGREEN_CURSOR_24_8_PRE_MULT) |
+			   EVERGREEN_CURSOR_URGENT_CONTROL(EVERGREEN_CURSOR_URGENT_1_2));
 	} else if (ASIC_IS_AVIVO(rdev)) {
-		WREG32(RADEON_MM_INDEX, AVIVO_D1CUR_CONTROL + radeon_crtc->crtc_offset);
-		WREG32(RADEON_MM_DATA, (AVIVO_D1CURSOR_MODE_24BPP << AVIVO_D1CURSOR_MODE_SHIFT));
+		WREG32_IDX(AVIVO_D1CUR_CONTROL + radeon_crtc->crtc_offset,
+			   (AVIVO_D1CURSOR_MODE_24BPP << AVIVO_D1CURSOR_MODE_SHIFT));
 	} else {
+		u32 reg;
 		switch (radeon_crtc->crtc_id) {
 		case 0:
-			WREG32(RADEON_MM_INDEX, RADEON_CRTC_GEN_CNTL);
+			reg = RADEON_CRTC_GEN_CNTL;
 			break;
 		case 1:
-			WREG32(RADEON_MM_INDEX, RADEON_CRTC2_GEN_CNTL);
+			reg = RADEON_CRTC2_GEN_CNTL;
 			break;
 		default:
 			return;
 		}
-		WREG32_P(RADEON_MM_DATA, 0, ~RADEON_CRTC_CUR_EN);
+		WREG32_IDX(reg, RREG32_IDX(reg) & ~RADEON_CRTC_CUR_EN);
 	}
 }
 

drivers/gpu/drm/radeon/radeon_device.c

@@ -1059,6 +1059,7 @@ int radeon_device_init(struct radeon_device *rdev,
 
 	/* Registers mapping */
 	/* TODO: block userspace mapping of io register */
+	spin_lock_init(&rdev->mmio_idx_lock);
 	rdev->rmmio_base = pci_resource_start(rdev->pdev, 2);
 	rdev->rmmio_size = pci_resource_len(rdev->pdev, 2);
 	rdev->rmmio = ioremap(rdev->rmmio_base, rdev->rmmio_size);

drivers/gpu/drm/radeon/radeon_display.c

@@ -378,8 +378,12 @@ static int radeon_crtc_page_flip(struct drm_crtc *crtc,
 	work->old_rbo = rbo;
 	obj = new_radeon_fb->obj;
 	rbo = gem_to_radeon_bo(obj);
+
+	spin_lock(&rbo->tbo.bdev->fence_lock);
 	if (rbo->tbo.sync_obj)
 		work->fence = radeon_fence_ref(rbo->tbo.sync_obj);
+	spin_unlock(&rbo->tbo.bdev->fence_lock);
+
 	INIT_WORK(&work->work, radeon_unpin_work_func);
 
 	/* We borrow the event spin lock for protecting unpin_work */

drivers/gpu/drm/radeon/radeon_drv.c

@@ -66,9 +66,11 @@
  *   2.23.0 - allow STRMOUT_BASE_UPDATE on RS780 and RS880
  *   2.24.0 - eg only: allow MIP_ADDRESS=0 for MSAA textures
  *   2.25.0 - eg+: new info request for num SE and num SH
+ *   2.26.0 - r600-eg: fix htile size computation
+ *   2.27.0 - r600-SI: Add CS ioctl support for async DMA
  */
 #define KMS_DRIVER_MAJOR	2
-#define KMS_DRIVER_MINOR	25
+#define KMS_DRIVER_MINOR	27
 #define KMS_DRIVER_PATCHLEVEL	0
 int radeon_driver_load_kms(struct drm_device *dev, unsigned long flags);
 int radeon_driver_unload_kms(struct drm_device *dev);

drivers/gpu/drm/radeon/radeon_drv.h

@@ -366,7 +366,6 @@ extern int radeon_cp_buffers(struct drm_device *dev, void *data, struct drm_file
 extern u32 radeon_read_fb_location(drm_radeon_private_t *dev_priv);
 extern void radeon_write_agp_location(drm_radeon_private_t *dev_priv, u32 agp_loc);
 extern void radeon_write_agp_base(drm_radeon_private_t *dev_priv, u64 agp_base);
-extern u32 RADEON_READ_MM(drm_radeon_private_t *dev_priv, int addr);
 
 extern void radeon_freelist_reset(struct drm_device * dev);
 extern struct drm_buf *radeon_freelist_get(struct drm_device * dev);

drivers/gpu/drm/radeon/radeon_object.c

@@ -96,9 +96,9 @@ void radeon_ttm_placement_from_domain(struct radeon_bo *rbo, u32 domain)
 	}
 	if (domain & RADEON_GEM_DOMAIN_CPU) {
 		if (rbo->rdev->flags & RADEON_IS_AGP) {
-			rbo->placements[c++] = TTM_PL_FLAG_WC | TTM_PL_FLAG_TT;
+			rbo->placements[c++] = TTM_PL_FLAG_WC | TTM_PL_FLAG_SYSTEM;
 		} else {
-			rbo->placements[c++] = TTM_PL_FLAG_CACHED | TTM_PL_FLAG_TT;
+			rbo->placements[c++] = TTM_PL_FLAG_CACHED | TTM_PL_FLAG_SYSTEM;
 		}
 	}
 	if (!c)

include/uapi/drm/radeon_drm.h

@@ -917,6 +917,7 @@ struct drm_radeon_gem_va {
 /* The second dword of RADEON_CHUNK_ID_FLAGS is a uint32 that sets the ring type */
 #define RADEON_CS_RING_GFX          0
 #define RADEON_CS_RING_COMPUTE      1
+#define RADEON_CS_RING_DMA          2
 /* The third dword of RADEON_CHUNK_ID_FLAGS is a sint32 that sets the priority */
 /* 0 = normal, + = higher priority, - = lower priority */