drm/radeon: add VM CS parser support for async DMA on cayman/TN/SI

Allows us to use async DMA from userspace.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

3 files changed, 118 insertions, 0 deletions

diff --git a/drivers/gpu/drm/radeon/evergreen_cs.c b/drivers/gpu/drm/radeon/evergreen_cs.c
index 0a1ec4e6f15e..9a9d3ae6c188 100644
--- a/drivers/gpu/drm/radeon/evergreen_cs.c
+++ b/drivers/gpu/drm/radeon/evergreen_cs.c
@@ -3556,3 +3556,114 @@ 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;
+}
diff --git a/drivers/gpu/drm/radeon/radeon_asic.c b/drivers/gpu/drm/radeon/radeon_asic.c
index ac1d5702144c..596bcbe80ed0 100644
--- a/drivers/gpu/drm/radeon/radeon_asic.c
+++ b/drivers/gpu/drm/radeon/radeon_asic.c
@@ -1484,6 +1484,7 @@ 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 = &evergreen_dma_cs_parse,
@@ -1494,6 +1495,7 @@ 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 = &evergreen_dma_cs_parse,
@@ -1609,6 +1611,7 @@ 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 = &evergreen_dma_cs_parse,
@@ -1619,6 +1622,7 @@ 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 = &evergreen_dma_cs_parse,
@@ -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,
diff --git a/drivers/gpu/drm/radeon/radeon_asic.h b/drivers/gpu/drm/radeon/radeon_asic.h
index d2ac64619f37..5f4882cc2152 100644
--- a/drivers/gpu/drm/radeon/radeon_asic.h
+++ b/drivers/gpu/drm/radeon/radeon_asic.h
@@ -473,6 +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);