drm/radeon: separate DMA code

Similar to separating the UVD code, just put the DMA
functions into separate files.

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

13 files changed, 2181 insertions, 1899 deletions

diff --git a/drivers/gpu/drm/radeon/Makefile b/drivers/gpu/drm/radeon/Makefile
index 1e23b18d549a..da2a8e9e9308 100644
--- a/drivers/gpu/drm/radeon/Makefile
+++ b/drivers/gpu/drm/radeon/Makefile
@@ -82,6 +82,15 @@ radeon-y += radeon_device.o radeon_asic.o radeon_kms.o \
 	trinity_smc.o ni_dpm.o si_smc.o si_dpm.o kv_smc.o kv_dpm.o ci_smc.o \
 	ci_dpm.o
 
+# add async DMA block
+radeon-y += \
+	r600_dma.o \
+	rv770_dma.o \
+	evergreen_dma.o \
+	ni_dma.o \
+	si_dma.o \
+	cik_sdma.o \
+
 # add UVD block
 radeon-y += \
 	radeon_uvd.o \
diff --git a/drivers/gpu/drm/radeon/cik.c b/drivers/gpu/drm/radeon/cik.c
index 1400b5203db1..692e31b95d34 100644
--- a/drivers/gpu/drm/radeon/cik.c
+++ b/drivers/gpu/drm/radeon/cik.c
@@ -64,6 +64,14 @@ extern int sumo_rlc_init(struct radeon_device *rdev);
 extern void si_vram_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc);
 extern void si_rlc_reset(struct radeon_device *rdev);
 extern void si_init_uvd_internal_cg(struct radeon_device *rdev);
+extern int cik_sdma_resume(struct radeon_device *rdev);
+extern void cik_sdma_enable(struct radeon_device *rdev, bool enable);
+extern void cik_sdma_fini(struct radeon_device *rdev);
+extern void cik_sdma_vm_set_page(struct radeon_device *rdev,
+				 struct radeon_ib *ib,
+				 uint64_t pe,
+				 uint64_t addr, unsigned count,
+				 uint32_t incr, uint32_t flags);
 static void cik_rlc_stop(struct radeon_device *rdev);
 static void cik_pcie_gen3_enable(struct radeon_device *rdev);
 static void cik_program_aspm(struct radeon_device *rdev);
@@ -3987,579 +3995,6 @@ static int cik_cp_resume(struct radeon_device *rdev)
 	return 0;
 }
 
-/*
- * sDMA - System DMA
- * Starting with CIK, the GPU has new asynchronous
- * DMA engines.  These engines are used for compute
- * and gfx.  There are two DMA engines (SDMA0, SDMA1)
- * and each one supports 1 ring buffer used for gfx
- * and 2 queues used for compute.
- *
- * The programming model is very similar to the CP
- * (ring buffer, IBs, etc.), but sDMA has it's own
- * packet format that is different from the PM4 format
- * used by the CP. sDMA supports copying data, writing
- * embedded data, solid fills, and a number of other
- * things.  It also has support for tiling/detiling of
- * buffers.
- */
-/**
- * cik_sdma_ring_ib_execute - Schedule an IB on the DMA engine
- *
- * @rdev: radeon_device pointer
- * @ib: IB object to schedule
- *
- * Schedule an IB in the DMA ring (CIK).
- */
-void cik_sdma_ring_ib_execute(struct radeon_device *rdev,
-			      struct radeon_ib *ib)
-{
-	struct radeon_ring *ring = &rdev->ring[ib->ring];
-	u32 extra_bits = (ib->vm ? ib->vm->id : 0) & 0xf;
-
-	if (rdev->wb.enabled) {
-		u32 next_rptr = ring->wptr + 5;
-		while ((next_rptr & 7) != 4)
-			next_rptr++;
-		next_rptr += 4;
-		radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0));
-		radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
-		radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xffffffff);
-		radeon_ring_write(ring, 1); /* number of DWs to follow */
-		radeon_ring_write(ring, next_rptr);
-	}
-
-	/* IB packet must end on a 8 DW boundary */
-	while ((ring->wptr & 7) != 4)
-		radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0));
-	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_INDIRECT_BUFFER, 0, extra_bits));
-	radeon_ring_write(ring, ib->gpu_addr & 0xffffffe0); /* base must be 32 byte aligned */
-	radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xffffffff);
-	radeon_ring_write(ring, ib->length_dw);
-
-}
-
-/**
- * cik_sdma_fence_ring_emit - emit a fence on the DMA ring
- *
- * @rdev: radeon_device pointer
- * @fence: radeon fence object
- *
- * Add a DMA fence packet to the ring to write
- * the fence seq number and DMA trap packet to generate
- * an interrupt if needed (CIK).
- */
-void cik_sdma_fence_ring_emit(struct radeon_device *rdev,
-			      struct radeon_fence *fence)
-{
-	struct radeon_ring *ring = &rdev->ring[fence->ring];
-	u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
-	u32 extra_bits = (SDMA_POLL_REG_MEM_EXTRA_OP(1) |
-			  SDMA_POLL_REG_MEM_EXTRA_FUNC(3)); /* == */
-	u32 ref_and_mask;
-
-	if (fence->ring == R600_RING_TYPE_DMA_INDEX)
-		ref_and_mask = SDMA0;
-	else
-		ref_and_mask = SDMA1;
-
-	/* write the fence */
-	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_FENCE, 0, 0));
-	radeon_ring_write(ring, addr & 0xffffffff);
-	radeon_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
-	radeon_ring_write(ring, fence->seq);
-	/* generate an interrupt */
-	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_TRAP, 0, 0));
-	/* flush HDP */
-	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_POLL_REG_MEM, 0, extra_bits));
-	radeon_ring_write(ring, GPU_HDP_FLUSH_DONE);
-	radeon_ring_write(ring, GPU_HDP_FLUSH_REQ);
-	radeon_ring_write(ring, ref_and_mask); /* REFERENCE */
-	radeon_ring_write(ring, ref_and_mask); /* MASK */
-	radeon_ring_write(ring, (4 << 16) | 10); /* RETRY_COUNT, POLL_INTERVAL */
-}
-
-/**
- * cik_sdma_semaphore_ring_emit - emit a semaphore on the dma ring
- *
- * @rdev: radeon_device pointer
- * @ring: radeon_ring structure holding ring information
- * @semaphore: radeon semaphore object
- * @emit_wait: wait or signal semaphore
- *
- * Add a DMA semaphore packet to the ring wait on or signal
- * other rings (CIK).
- */
-void cik_sdma_semaphore_ring_emit(struct radeon_device *rdev,
-				  struct radeon_ring *ring,
-				  struct radeon_semaphore *semaphore,
-				  bool emit_wait)
-{
-	u64 addr = semaphore->gpu_addr;
-	u32 extra_bits = emit_wait ? 0 : SDMA_SEMAPHORE_EXTRA_S;
-
-	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SEMAPHORE, 0, extra_bits));
-	radeon_ring_write(ring, addr & 0xfffffff8);
-	radeon_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
-}
-
-/**
- * cik_sdma_gfx_stop - stop the gfx async dma engines
- *
- * @rdev: radeon_device pointer
- *
- * Stop the gfx async dma ring buffers (CIK).
- */
-static void cik_sdma_gfx_stop(struct radeon_device *rdev)
-{
-	u32 rb_cntl, reg_offset;
-	int i;
-
-	radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
-
-	for (i = 0; i < 2; i++) {
-		if (i == 0)
-			reg_offset = SDMA0_REGISTER_OFFSET;
-		else
-			reg_offset = SDMA1_REGISTER_OFFSET;
-		rb_cntl = RREG32(SDMA0_GFX_RB_CNTL + reg_offset);
-		rb_cntl &= ~SDMA_RB_ENABLE;
-		WREG32(SDMA0_GFX_RB_CNTL + reg_offset, rb_cntl);
-		WREG32(SDMA0_GFX_IB_CNTL + reg_offset, 0);
-	}
-}
-
-/**
- * cik_sdma_rlc_stop - stop the compute async dma engines
- *
- * @rdev: radeon_device pointer
- *
- * Stop the compute async dma queues (CIK).
- */
-static void cik_sdma_rlc_stop(struct radeon_device *rdev)
-{
-	/* XXX todo */
-}
-
-/**
- * cik_sdma_enable - stop the async dma engines
- *
- * @rdev: radeon_device pointer
- * @enable: enable/disable the DMA MEs.
- *
- * Halt or unhalt the async dma engines (CIK).
- */
-static void cik_sdma_enable(struct radeon_device *rdev, bool enable)
-{
-	u32 me_cntl, reg_offset;
-	int i;
-
-	for (i = 0; i < 2; i++) {
-		if (i == 0)
-			reg_offset = SDMA0_REGISTER_OFFSET;
-		else
-			reg_offset = SDMA1_REGISTER_OFFSET;
-		me_cntl = RREG32(SDMA0_ME_CNTL + reg_offset);
-		if (enable)
-			me_cntl &= ~SDMA_HALT;
-		else
-			me_cntl |= SDMA_HALT;
-		WREG32(SDMA0_ME_CNTL + reg_offset, me_cntl);
-	}
-}
-
-/**
- * cik_sdma_gfx_resume - setup and start the async dma engines
- *
- * @rdev: radeon_device pointer
- *
- * Set up the gfx DMA ring buffers and enable them (CIK).
- * Returns 0 for success, error for failure.
- */
-static int cik_sdma_gfx_resume(struct radeon_device *rdev)
-{
-	struct radeon_ring *ring;
-	u32 rb_cntl, ib_cntl;
-	u32 rb_bufsz;
-	u32 reg_offset, wb_offset;
-	int i, r;
-
-	for (i = 0; i < 2; i++) {
-		if (i == 0) {
-			ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
-			reg_offset = SDMA0_REGISTER_OFFSET;
-			wb_offset = R600_WB_DMA_RPTR_OFFSET;
-		} else {
-			ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
-			reg_offset = SDMA1_REGISTER_OFFSET;
-			wb_offset = CAYMAN_WB_DMA1_RPTR_OFFSET;
-		}
-
-		WREG32(SDMA0_SEM_INCOMPLETE_TIMER_CNTL + reg_offset, 0);
-		WREG32(SDMA0_SEM_WAIT_FAIL_TIMER_CNTL + reg_offset, 0);
-
-		/* Set ring buffer size in dwords */
-		rb_bufsz = drm_order(ring->ring_size / 4);
-		rb_cntl = rb_bufsz << 1;
-#ifdef __BIG_ENDIAN
-		rb_cntl |= SDMA_RB_SWAP_ENABLE | SDMA_RPTR_WRITEBACK_SWAP_ENABLE;
-#endif
-		WREG32(SDMA0_GFX_RB_CNTL + reg_offset, rb_cntl);
-
-		/* Initialize the ring buffer's read and write pointers */
-		WREG32(SDMA0_GFX_RB_RPTR + reg_offset, 0);
-		WREG32(SDMA0_GFX_RB_WPTR + reg_offset, 0);
-
-		/* set the wb address whether it's enabled or not */
-		WREG32(SDMA0_GFX_RB_RPTR_ADDR_HI + reg_offset,
-		       upper_32_bits(rdev->wb.gpu_addr + wb_offset) & 0xFFFFFFFF);
-		WREG32(SDMA0_GFX_RB_RPTR_ADDR_LO + reg_offset,
-		       ((rdev->wb.gpu_addr + wb_offset) & 0xFFFFFFFC));
-
-		if (rdev->wb.enabled)
-			rb_cntl |= SDMA_RPTR_WRITEBACK_ENABLE;
-
-		WREG32(SDMA0_GFX_RB_BASE + reg_offset, ring->gpu_addr >> 8);
-		WREG32(SDMA0_GFX_RB_BASE_HI + reg_offset, ring->gpu_addr >> 40);
-
-		ring->wptr = 0;
-		WREG32(SDMA0_GFX_RB_WPTR + reg_offset, ring->wptr << 2);
-
-		ring->rptr = RREG32(SDMA0_GFX_RB_RPTR + reg_offset) >> 2;
-
-		/* enable DMA RB */
-		WREG32(SDMA0_GFX_RB_CNTL + reg_offset, rb_cntl | SDMA_RB_ENABLE);
-
-		ib_cntl = SDMA_IB_ENABLE;
-#ifdef __BIG_ENDIAN
-		ib_cntl |= SDMA_IB_SWAP_ENABLE;
-#endif
-		/* enable DMA IBs */
-		WREG32(SDMA0_GFX_IB_CNTL + reg_offset, ib_cntl);
-
-		ring->ready = true;
-
-		r = radeon_ring_test(rdev, ring->idx, ring);
-		if (r) {
-			ring->ready = false;
-			return r;
-		}
-	}
-
-	radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
-
-	return 0;
-}
-
-/**
- * cik_sdma_rlc_resume - setup and start the async dma engines
- *
- * @rdev: radeon_device pointer
- *
- * Set up the compute DMA queues and enable them (CIK).
- * Returns 0 for success, error for failure.
- */
-static int cik_sdma_rlc_resume(struct radeon_device *rdev)
-{
-	/* XXX todo */
-	return 0;
-}
-
-/**
- * cik_sdma_load_microcode - load the sDMA ME ucode
- *
- * @rdev: radeon_device pointer
- *
- * Loads the sDMA0/1 ucode.
- * Returns 0 for success, -EINVAL if the ucode is not available.
- */
-static int cik_sdma_load_microcode(struct radeon_device *rdev)
-{
-	const __be32 *fw_data;
-	int i;
-
-	if (!rdev->sdma_fw)
-		return -EINVAL;
-
-	/* stop the gfx rings and rlc compute queues */
-	cik_sdma_gfx_stop(rdev);
-	cik_sdma_rlc_stop(rdev);
-
-	/* halt the MEs */
-	cik_sdma_enable(rdev, false);
-
-	/* sdma0 */
-	fw_data = (const __be32 *)rdev->sdma_fw->data;
-	WREG32(SDMA0_UCODE_ADDR + SDMA0_REGISTER_OFFSET, 0);
-	for (i = 0; i < CIK_SDMA_UCODE_SIZE; i++)
-		WREG32(SDMA0_UCODE_DATA + SDMA0_REGISTER_OFFSET, be32_to_cpup(fw_data++));
-	WREG32(SDMA0_UCODE_DATA + SDMA0_REGISTER_OFFSET, CIK_SDMA_UCODE_VERSION);
-
-	/* sdma1 */
-	fw_data = (const __be32 *)rdev->sdma_fw->data;
-	WREG32(SDMA0_UCODE_ADDR + SDMA1_REGISTER_OFFSET, 0);
-	for (i = 0; i < CIK_SDMA_UCODE_SIZE; i++)
-		WREG32(SDMA0_UCODE_DATA + SDMA1_REGISTER_OFFSET, be32_to_cpup(fw_data++));
-	WREG32(SDMA0_UCODE_DATA + SDMA1_REGISTER_OFFSET, CIK_SDMA_UCODE_VERSION);
-
-	WREG32(SDMA0_UCODE_ADDR + SDMA0_REGISTER_OFFSET, 0);
-	WREG32(SDMA0_UCODE_ADDR + SDMA1_REGISTER_OFFSET, 0);
-	return 0;
-}
-
-/**
- * cik_sdma_resume - setup and start the async dma engines
- *
- * @rdev: radeon_device pointer
- *
- * Set up the DMA engines and enable them (CIK).
- * Returns 0 for success, error for failure.
- */
-static int cik_sdma_resume(struct radeon_device *rdev)
-{
-	int r;
-
-	/* Reset dma */
-	WREG32(SRBM_SOFT_RESET, SOFT_RESET_SDMA | SOFT_RESET_SDMA1);
-	RREG32(SRBM_SOFT_RESET);
-	udelay(50);
-	WREG32(SRBM_SOFT_RESET, 0);
-	RREG32(SRBM_SOFT_RESET);
-
-	r = cik_sdma_load_microcode(rdev);
-	if (r)
-		return r;
-
-	/* unhalt the MEs */
-	cik_sdma_enable(rdev, true);
-
-	/* start the gfx rings and rlc compute queues */
-	r = cik_sdma_gfx_resume(rdev);
-	if (r)
-		return r;
-	r = cik_sdma_rlc_resume(rdev);
-	if (r)
-		return r;
-
-	return 0;
-}
-
-/**
- * cik_sdma_fini - tear down the async dma engines
- *
- * @rdev: radeon_device pointer
- *
- * Stop the async dma engines and free the rings (CIK).
- */
-static void cik_sdma_fini(struct radeon_device *rdev)
-{
-	/* stop the gfx rings and rlc compute queues */
-	cik_sdma_gfx_stop(rdev);
-	cik_sdma_rlc_stop(rdev);
-	/* halt the MEs */
-	cik_sdma_enable(rdev, false);
-	radeon_ring_fini(rdev, &rdev->ring[R600_RING_TYPE_DMA_INDEX]);
-	radeon_ring_fini(rdev, &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX]);
-	/* XXX - compute dma queue tear down */
-}
-
-/**
- * cik_copy_dma - copy pages using the DMA engine
- *
- * @rdev: radeon_device pointer
- * @src_offset: src GPU address
- * @dst_offset: dst GPU address
- * @num_gpu_pages: number of GPU pages to xfer
- * @fence: radeon fence object
- *
- * Copy GPU paging using the DMA engine (CIK).
- * Used by the radeon ttm implementation to move pages if
- * registered as the asic copy callback.
- */
-int cik_copy_dma(struct radeon_device *rdev,
-		 uint64_t src_offset, uint64_t dst_offset,
-		 unsigned num_gpu_pages,
-		 struct radeon_fence **fence)
-{
-	struct radeon_semaphore *sem = NULL;
-	int ring_index = rdev->asic->copy.dma_ring_index;
-	struct radeon_ring *ring = &rdev->ring[ring_index];
-	u32 size_in_bytes, cur_size_in_bytes;
-	int i, num_loops;
-	int r = 0;
-
-	r = radeon_semaphore_create(rdev, &sem);
-	if (r) {
-		DRM_ERROR("radeon: moving bo (%d).\n", r);
-		return r;
-	}
-
-	size_in_bytes = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT);
-	num_loops = DIV_ROUND_UP(size_in_bytes, 0x1fffff);
-	r = radeon_ring_lock(rdev, ring, num_loops * 7 + 14);
-	if (r) {
-		DRM_ERROR("radeon: moving bo (%d).\n", r);
-		radeon_semaphore_free(rdev, &sem, NULL);
-		return r;
-	}
-
-	if (radeon_fence_need_sync(*fence, ring->idx)) {
-		radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring,
-					    ring->idx);
-		radeon_fence_note_sync(*fence, ring->idx);
-	} else {
-		radeon_semaphore_free(rdev, &sem, NULL);
-	}
-
-	for (i = 0; i < num_loops; i++) {
-		cur_size_in_bytes = size_in_bytes;
-		if (cur_size_in_bytes > 0x1fffff)
-			cur_size_in_bytes = 0x1fffff;
-		size_in_bytes -= cur_size_in_bytes;
-		radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_COPY, SDMA_COPY_SUB_OPCODE_LINEAR, 0));
-		radeon_ring_write(ring, cur_size_in_bytes);
-		radeon_ring_write(ring, 0); /* src/dst endian swap */
-		radeon_ring_write(ring, src_offset & 0xffffffff);
-		radeon_ring_write(ring, upper_32_bits(src_offset) & 0xffffffff);
-		radeon_ring_write(ring, dst_offset & 0xfffffffc);
-		radeon_ring_write(ring, upper_32_bits(dst_offset) & 0xffffffff);
-		src_offset += cur_size_in_bytes;
-		dst_offset += cur_size_in_bytes;
-	}
-
-	r = radeon_fence_emit(rdev, fence, ring->idx);
-	if (r) {
-		radeon_ring_unlock_undo(rdev, ring);
-		return r;
-	}
-
-	radeon_ring_unlock_commit(rdev, ring);
-	radeon_semaphore_free(rdev, &sem, *fence);
-
-	return r;
-}
-
-/**
- * cik_sdma_ring_test - simple async dma engine test
- *
- * @rdev: radeon_device pointer
- * @ring: radeon_ring structure holding ring information
- *
- * Test the DMA engine by writing using it to write an
- * value to memory. (CIK).
- * Returns 0 for success, error for failure.
- */
-int cik_sdma_ring_test(struct radeon_device *rdev,
-		       struct radeon_ring *ring)
-{
-	unsigned i;
-	int r;
-	void __iomem *ptr = (void *)rdev->vram_scratch.ptr;
-	u32 tmp;
-
-	if (!ptr) {
-		DRM_ERROR("invalid vram scratch pointer\n");
-		return -EINVAL;
-	}
-
-	tmp = 0xCAFEDEAD;
-	writel(tmp, ptr);
-
-	r = radeon_ring_lock(rdev, ring, 4);
-	if (r) {
-		DRM_ERROR("radeon: dma failed to lock ring %d (%d).\n", ring->idx, r);
-		return r;
-	}
-	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0));
-	radeon_ring_write(ring, rdev->vram_scratch.gpu_addr & 0xfffffffc);
-	radeon_ring_write(ring, upper_32_bits(rdev->vram_scratch.gpu_addr) & 0xffffffff);
-	radeon_ring_write(ring, 1); /* number of DWs to follow */
-	radeon_ring_write(ring, 0xDEADBEEF);
-	radeon_ring_unlock_commit(rdev, ring);
-
-	for (i = 0; i < rdev->usec_timeout; i++) {
-		tmp = readl(ptr);
-		if (tmp == 0xDEADBEEF)
-			break;
-		DRM_UDELAY(1);
-	}
-
-	if (i < rdev->usec_timeout) {
-		DRM_INFO("ring test on %d succeeded in %d usecs\n", ring->idx, i);
-	} else {
-		DRM_ERROR("radeon: ring %d test failed (0x%08X)\n",
-			  ring->idx, tmp);
-		r = -EINVAL;
-	}
-	return r;
-}
-
-/**
- * cik_sdma_ib_test - test an IB on the DMA engine
- *
- * @rdev: radeon_device pointer
- * @ring: radeon_ring structure holding ring information
- *
- * Test a simple IB in the DMA ring (CIK).
- * Returns 0 on success, error on failure.
- */
-int cik_sdma_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
-{
-	struct radeon_ib ib;
-	unsigned i;
-	int r;
-	void __iomem *ptr = (void *)rdev->vram_scratch.ptr;
-	u32 tmp = 0;
-
-	if (!ptr) {
-		DRM_ERROR("invalid vram scratch pointer\n");
-		return -EINVAL;
-	}
-
-	tmp = 0xCAFEDEAD;
-	writel(tmp, ptr);
-
-	r = radeon_ib_get(rdev, ring->idx, &ib, NULL, 256);
-	if (r) {
-		DRM_ERROR("radeon: failed to get ib (%d).\n", r);
-		return r;
-	}
-
-	ib.ptr[0] = SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0);
-	ib.ptr[1] = rdev->vram_scratch.gpu_addr & 0xfffffffc;
-	ib.ptr[2] = upper_32_bits(rdev->vram_scratch.gpu_addr) & 0xffffffff;
-	ib.ptr[3] = 1;
-	ib.ptr[4] = 0xDEADBEEF;
-	ib.length_dw = 5;
-
-	r = radeon_ib_schedule(rdev, &ib, NULL);
-	if (r) {
-		radeon_ib_free(rdev, &ib);
-		DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
-		return r;
-	}
-	r = radeon_fence_wait(ib.fence, false);
-	if (r) {
-		DRM_ERROR("radeon: fence wait failed (%d).\n", r);
-		return r;
-	}
-	for (i = 0; i < rdev->usec_timeout; i++) {
-		tmp = readl(ptr);
-		if (tmp == 0xDEADBEEF)
-			break;
-		DRM_UDELAY(1);
-	}
-	if (i < rdev->usec_timeout) {
-		DRM_INFO("ib test on ring %d succeeded in %u usecs\n", ib.fence->ring, i);
-	} else {
-		DRM_ERROR("radeon: ib test failed (0x%08X)\n", tmp);
-		r = -EINVAL;
-	}
-	radeon_ib_free(rdev, &ib);
-	return r;
-}
-
-
 static void cik_print_gpu_status_regs(struct radeon_device *rdev)
 {
 	dev_info(rdev->dev, "  GRBM_STATUS=0x%08X\n",
@@ -4609,7 +4044,7 @@ static void cik_print_gpu_status_regs(struct radeon_device *rdev)
  * mask to be used by cik_gpu_soft_reset().
  * Returns a mask of the blocks to be reset.
  */
-static u32 cik_gpu_check_soft_reset(struct radeon_device *rdev)
+u32 cik_gpu_check_soft_reset(struct radeon_device *rdev)
 {
 	u32 reset_mask = 0;
 	u32 tmp;
@@ -4860,34 +4295,6 @@ bool cik_gfx_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
 	return radeon_ring_test_lockup(rdev, ring);
 }
 
-/**
- * cik_sdma_is_lockup - Check if the DMA engine is locked up
- *
- * @rdev: radeon_device pointer
- * @ring: radeon_ring structure holding ring information
- *
- * Check if the async DMA engine is locked up (CIK).
- * Returns true if the engine appears to be locked up, false if not.
- */
-bool cik_sdma_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
-{
-	u32 reset_mask = cik_gpu_check_soft_reset(rdev);
-	u32 mask;
-
-	if (ring->idx == R600_RING_TYPE_DMA_INDEX)
-		mask = RADEON_RESET_DMA;
-	else
-		mask = RADEON_RESET_DMA1;
-
-	if (!(reset_mask & mask)) {
-		radeon_ring_lockup_update(ring);
-		return false;
-	}
-	/* force ring activities */
-	radeon_ring_force_activity(rdev, ring);
-	return radeon_ring_test_lockup(rdev, ring);
-}
-
 /* MC */
 /**
  * cik_mc_program - program the GPU memory controller
@@ -5424,131 +4831,8 @@ void cik_vm_set_page(struct radeon_device *rdev,
 		}
 	} else {
 		/* DMA */
-		if (flags & RADEON_VM_PAGE_SYSTEM) {
-			while (count) {
-				ndw = count * 2;
-				if (ndw > 0xFFFFE)
-					ndw = 0xFFFFE;
-
-				/* for non-physically contiguous pages (system) */
-				ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0);
-				ib->ptr[ib->length_dw++] = pe;
-				ib->ptr[ib->length_dw++] = upper_32_bits(pe);
-				ib->ptr[ib->length_dw++] = ndw;
-				for (; ndw > 0; ndw -= 2, --count, pe += 8) {
-					if (flags & RADEON_VM_PAGE_SYSTEM) {
-						value = radeon_vm_map_gart(rdev, addr);
-						value &= 0xFFFFFFFFFFFFF000ULL;
-					} else if (flags & RADEON_VM_PAGE_VALID) {
-						value = addr;
-					} else {
-						value = 0;
-					}
-					addr += incr;
-					value |= r600_flags;
-					ib->ptr[ib->length_dw++] = value;
-					ib->ptr[ib->length_dw++] = upper_32_bits(value);
-				}
-			}
-		} else {
-			while (count) {
-				ndw = count;
-				if (ndw > 0x7FFFF)
-					ndw = 0x7FFFF;
-
-				if (flags & RADEON_VM_PAGE_VALID)
-					value = addr;
-				else
-					value = 0;
-				/* for physically contiguous pages (vram) */
-				ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_GENERATE_PTE_PDE, 0, 0);
-				ib->ptr[ib->length_dw++] = pe; /* dst addr */
-				ib->ptr[ib->length_dw++] = upper_32_bits(pe);
-				ib->ptr[ib->length_dw++] = r600_flags; /* mask */
-				ib->ptr[ib->length_dw++] = 0;
-				ib->ptr[ib->length_dw++] = value; /* value */
-				ib->ptr[ib->length_dw++] = upper_32_bits(value);
-				ib->ptr[ib->length_dw++] = incr; /* increment size */
-				ib->ptr[ib->length_dw++] = 0;
-				ib->ptr[ib->length_dw++] = ndw; /* number of entries */
-				pe += ndw * 8;
-				addr += ndw * incr;
-				count -= ndw;
-			}
-		}
-		while (ib->length_dw & 0x7)
-			ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0);
-	}
-}
-
-/**
- * cik_dma_vm_flush - cik vm flush using sDMA
- *
- * @rdev: radeon_device pointer
- *
- * Update the page table base and flush the VM TLB
- * using sDMA (CIK).
- */
-void cik_dma_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm)
-{
-	struct radeon_ring *ring = &rdev->ring[ridx];
-	u32 extra_bits = (SDMA_POLL_REG_MEM_EXTRA_OP(1) |
-			  SDMA_POLL_REG_MEM_EXTRA_FUNC(3)); /* == */
-	u32 ref_and_mask;
-
-	if (vm == NULL)
-		return;
-
-	if (ridx == R600_RING_TYPE_DMA_INDEX)
-		ref_and_mask = SDMA0;
-	else
-		ref_and_mask = SDMA1;
-
-	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
-	if (vm->id < 8) {
-		radeon_ring_write(ring, (VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2)) >> 2);
-	} else {
-		radeon_ring_write(ring, (VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((vm->id - 8) << 2)) >> 2);
+		cik_sdma_vm_set_page(rdev, ib, pe, addr, count, incr, flags);
 	}
-	radeon_ring_write(ring, vm->pd_gpu_addr >> 12);
-
-	/* update SH_MEM_* regs */
-	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
-	radeon_ring_write(ring, SRBM_GFX_CNTL >> 2);
-	radeon_ring_write(ring, VMID(vm->id));
-
-	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
-	radeon_ring_write(ring, SH_MEM_BASES >> 2);
-	radeon_ring_write(ring, 0);
-
-	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
-	radeon_ring_write(ring, SH_MEM_CONFIG >> 2);
-	radeon_ring_write(ring, 0);
-
-	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
-	radeon_ring_write(ring, SH_MEM_APE1_BASE >> 2);
-	radeon_ring_write(ring, 1);
-
-	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
-	radeon_ring_write(ring, SH_MEM_APE1_LIMIT >> 2);
-	radeon_ring_write(ring, 0);
-
-	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
-	radeon_ring_write(ring, SRBM_GFX_CNTL >> 2);
-	radeon_ring_write(ring, VMID(0));
-
-	/* flush HDP */
-	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_POLL_REG_MEM, 0, extra_bits));
-	radeon_ring_write(ring, GPU_HDP_FLUSH_DONE);
-	radeon_ring_write(ring, GPU_HDP_FLUSH_REQ);
-	radeon_ring_write(ring, ref_and_mask); /* REFERENCE */
-	radeon_ring_write(ring, ref_and_mask); /* MASK */
-	radeon_ring_write(ring, (4 << 16) | 10); /* RETRY_COUNT, POLL_INTERVAL */
-
-	/* flush TLB */
-	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
-	radeon_ring_write(ring, VM_INVALIDATE_REQUEST >> 2);
-	radeon_ring_write(ring, 1 << vm->id);
 }
 
 /*
diff --git a/drivers/gpu/drm/radeon/cik_sdma.c b/drivers/gpu/drm/radeon/cik_sdma.c
new file mode 100644
index 000000000000..8925185a0049
--- /dev/null
+++ b/drivers/gpu/drm/radeon/cik_sdma.c
@@ -0,0 +1,785 @@
+/*
+ * Copyright 2013 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Alex Deucher
+ */
+#include <linux/firmware.h>
+#include <drm/drmP.h>
+#include "radeon.h"
+#include "radeon_asic.h"
+#include "cikd.h"
+
+/* sdma */
+#define CIK_SDMA_UCODE_SIZE 1050
+#define CIK_SDMA_UCODE_VERSION 64
+
+u32 cik_gpu_check_soft_reset(struct radeon_device *rdev);
+
+/*
+ * sDMA - System DMA
+ * Starting with CIK, the GPU has new asynchronous
+ * DMA engines.  These engines are used for compute
+ * and gfx.  There are two DMA engines (SDMA0, SDMA1)
+ * and each one supports 1 ring buffer used for gfx
+ * and 2 queues used for compute.
+ *
+ * The programming model is very similar to the CP
+ * (ring buffer, IBs, etc.), but sDMA has it's own
+ * packet format that is different from the PM4 format
+ * used by the CP. sDMA supports copying data, writing
+ * embedded data, solid fills, and a number of other
+ * things.  It also has support for tiling/detiling of
+ * buffers.
+ */
+
+/**
+ * cik_sdma_ring_ib_execute - Schedule an IB on the DMA engine
+ *
+ * @rdev: radeon_device pointer
+ * @ib: IB object to schedule
+ *
+ * Schedule an IB in the DMA ring (CIK).
+ */
+void cik_sdma_ring_ib_execute(struct radeon_device *rdev,
+			      struct radeon_ib *ib)
+{
+	struct radeon_ring *ring = &rdev->ring[ib->ring];
+	u32 extra_bits = (ib->vm ? ib->vm->id : 0) & 0xf;
+
+	if (rdev->wb.enabled) {
+		u32 next_rptr = ring->wptr + 5;
+		while ((next_rptr & 7) != 4)
+			next_rptr++;
+		next_rptr += 4;
+		radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0));
+		radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
+		radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xffffffff);
+		radeon_ring_write(ring, 1); /* number of DWs to follow */
+		radeon_ring_write(ring, next_rptr);
+	}
+
+	/* IB packet must end on a 8 DW boundary */
+	while ((ring->wptr & 7) != 4)
+		radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0));
+	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_INDIRECT_BUFFER, 0, extra_bits));
+	radeon_ring_write(ring, ib->gpu_addr & 0xffffffe0); /* base must be 32 byte aligned */
+	radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xffffffff);
+	radeon_ring_write(ring, ib->length_dw);
+
+}
+
+/**
+ * cik_sdma_fence_ring_emit - emit a fence on the DMA ring
+ *
+ * @rdev: radeon_device pointer
+ * @fence: radeon fence object
+ *
+ * Add a DMA fence packet to the ring to write
+ * the fence seq number and DMA trap packet to generate
+ * an interrupt if needed (CIK).
+ */
+void cik_sdma_fence_ring_emit(struct radeon_device *rdev,
+			      struct radeon_fence *fence)
+{
+	struct radeon_ring *ring = &rdev->ring[fence->ring];
+	u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
+	u32 extra_bits = (SDMA_POLL_REG_MEM_EXTRA_OP(1) |
+			  SDMA_POLL_REG_MEM_EXTRA_FUNC(3)); /* == */
+	u32 ref_and_mask;
+
+	if (fence->ring == R600_RING_TYPE_DMA_INDEX)
+		ref_and_mask = SDMA0;
+	else
+		ref_and_mask = SDMA1;
+
+	/* write the fence */
+	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_FENCE, 0, 0));
+	radeon_ring_write(ring, addr & 0xffffffff);
+	radeon_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
+	radeon_ring_write(ring, fence->seq);
+	/* generate an interrupt */
+	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_TRAP, 0, 0));
+	/* flush HDP */
+	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_POLL_REG_MEM, 0, extra_bits));
+	radeon_ring_write(ring, GPU_HDP_FLUSH_DONE);
+	radeon_ring_write(ring, GPU_HDP_FLUSH_REQ);
+	radeon_ring_write(ring, ref_and_mask); /* REFERENCE */
+	radeon_ring_write(ring, ref_and_mask); /* MASK */
+	radeon_ring_write(ring, (4 << 16) | 10); /* RETRY_COUNT, POLL_INTERVAL */
+}
+
+/**
+ * cik_sdma_semaphore_ring_emit - emit a semaphore on the dma ring
+ *
+ * @rdev: radeon_device pointer
+ * @ring: radeon_ring structure holding ring information
+ * @semaphore: radeon semaphore object
+ * @emit_wait: wait or signal semaphore
+ *
+ * Add a DMA semaphore packet to the ring wait on or signal
+ * other rings (CIK).
+ */
+void cik_sdma_semaphore_ring_emit(struct radeon_device *rdev,
+				  struct radeon_ring *ring,
+				  struct radeon_semaphore *semaphore,
+				  bool emit_wait)
+{
+	u64 addr = semaphore->gpu_addr;
+	u32 extra_bits = emit_wait ? 0 : SDMA_SEMAPHORE_EXTRA_S;
+
+	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SEMAPHORE, 0, extra_bits));
+	radeon_ring_write(ring, addr & 0xfffffff8);
+	radeon_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
+}
+
+/**
+ * cik_sdma_gfx_stop - stop the gfx async dma engines
+ *
+ * @rdev: radeon_device pointer
+ *
+ * Stop the gfx async dma ring buffers (CIK).
+ */
+static void cik_sdma_gfx_stop(struct radeon_device *rdev)
+{
+	u32 rb_cntl, reg_offset;
+	int i;
+
+	radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
+
+	for (i = 0; i < 2; i++) {
+		if (i == 0)
+			reg_offset = SDMA0_REGISTER_OFFSET;
+		else
+			reg_offset = SDMA1_REGISTER_OFFSET;
+		rb_cntl = RREG32(SDMA0_GFX_RB_CNTL + reg_offset);
+		rb_cntl &= ~SDMA_RB_ENABLE;
+		WREG32(SDMA0_GFX_RB_CNTL + reg_offset, rb_cntl);
+		WREG32(SDMA0_GFX_IB_CNTL + reg_offset, 0);
+	}
+}
+
+/**
+ * cik_sdma_rlc_stop - stop the compute async dma engines
+ *
+ * @rdev: radeon_device pointer
+ *
+ * Stop the compute async dma queues (CIK).
+ */
+static void cik_sdma_rlc_stop(struct radeon_device *rdev)
+{
+	/* XXX todo */
+}
+
+/**
+ * cik_sdma_enable - stop the async dma engines
+ *
+ * @rdev: radeon_device pointer
+ * @enable: enable/disable the DMA MEs.
+ *
+ * Halt or unhalt the async dma engines (CIK).
+ */
+void cik_sdma_enable(struct radeon_device *rdev, bool enable)
+{
+	u32 me_cntl, reg_offset;
+	int i;
+
+	for (i = 0; i < 2; i++) {
+		if (i == 0)
+			reg_offset = SDMA0_REGISTER_OFFSET;
+		else
+			reg_offset = SDMA1_REGISTER_OFFSET;
+		me_cntl = RREG32(SDMA0_ME_CNTL + reg_offset);
+		if (enable)
+			me_cntl &= ~SDMA_HALT;
+		else
+			me_cntl |= SDMA_HALT;
+		WREG32(SDMA0_ME_CNTL + reg_offset, me_cntl);
+	}
+}
+
+/**
+ * cik_sdma_gfx_resume - setup and start the async dma engines
+ *
+ * @rdev: radeon_device pointer
+ *
+ * Set up the gfx DMA ring buffers and enable them (CIK).
+ * Returns 0 for success, error for failure.
+ */
+static int cik_sdma_gfx_resume(struct radeon_device *rdev)
+{
+	struct radeon_ring *ring;
+	u32 rb_cntl, ib_cntl;
+	u32 rb_bufsz;
+	u32 reg_offset, wb_offset;
+	int i, r;
+
+	for (i = 0; i < 2; i++) {
+		if (i == 0) {
+			ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
+			reg_offset = SDMA0_REGISTER_OFFSET;
+			wb_offset = R600_WB_DMA_RPTR_OFFSET;
+		} else {
+			ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
+			reg_offset = SDMA1_REGISTER_OFFSET;
+			wb_offset = CAYMAN_WB_DMA1_RPTR_OFFSET;
+		}
+
+		WREG32(SDMA0_SEM_INCOMPLETE_TIMER_CNTL + reg_offset, 0);
+		WREG32(SDMA0_SEM_WAIT_FAIL_TIMER_CNTL + reg_offset, 0);
+
+		/* Set ring buffer size in dwords */
+		rb_bufsz = drm_order(ring->ring_size / 4);
+		rb_cntl = rb_bufsz << 1;
+#ifdef __BIG_ENDIAN
+		rb_cntl |= SDMA_RB_SWAP_ENABLE | SDMA_RPTR_WRITEBACK_SWAP_ENABLE;
+#endif
+		WREG32(SDMA0_GFX_RB_CNTL + reg_offset, rb_cntl);
+
+		/* Initialize the ring buffer's read and write pointers */
+		WREG32(SDMA0_GFX_RB_RPTR + reg_offset, 0);
+		WREG32(SDMA0_GFX_RB_WPTR + reg_offset, 0);
+
+		/* set the wb address whether it's enabled or not */
+		WREG32(SDMA0_GFX_RB_RPTR_ADDR_HI + reg_offset,
+		       upper_32_bits(rdev->wb.gpu_addr + wb_offset) & 0xFFFFFFFF);
+		WREG32(SDMA0_GFX_RB_RPTR_ADDR_LO + reg_offset,
+		       ((rdev->wb.gpu_addr + wb_offset) & 0xFFFFFFFC));
+
+		if (rdev->wb.enabled)
+			rb_cntl |= SDMA_RPTR_WRITEBACK_ENABLE;
+
+		WREG32(SDMA0_GFX_RB_BASE + reg_offset, ring->gpu_addr >> 8);
+		WREG32(SDMA0_GFX_RB_BASE_HI + reg_offset, ring->gpu_addr >> 40);
+
+		ring->wptr = 0;
+		WREG32(SDMA0_GFX_RB_WPTR + reg_offset, ring->wptr << 2);
+
+		ring->rptr = RREG32(SDMA0_GFX_RB_RPTR + reg_offset) >> 2;
+
+		/* enable DMA RB */
+		WREG32(SDMA0_GFX_RB_CNTL + reg_offset, rb_cntl | SDMA_RB_ENABLE);
+
+		ib_cntl = SDMA_IB_ENABLE;
+#ifdef __BIG_ENDIAN
+		ib_cntl |= SDMA_IB_SWAP_ENABLE;
+#endif
+		/* enable DMA IBs */
+		WREG32(SDMA0_GFX_IB_CNTL + reg_offset, ib_cntl);
+
+		ring->ready = true;
+
+		r = radeon_ring_test(rdev, ring->idx, ring);
+		if (r) {
+			ring->ready = false;
+			return r;
+		}
+	}
+
+	radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
+
+	return 0;
+}
+
+/**
+ * cik_sdma_rlc_resume - setup and start the async dma engines
+ *
+ * @rdev: radeon_device pointer
+ *
+ * Set up the compute DMA queues and enable them (CIK).
+ * Returns 0 for success, error for failure.
+ */
+static int cik_sdma_rlc_resume(struct radeon_device *rdev)
+{
+	/* XXX todo */
+	return 0;
+}
+
+/**
+ * cik_sdma_load_microcode - load the sDMA ME ucode
+ *
+ * @rdev: radeon_device pointer
+ *
+ * Loads the sDMA0/1 ucode.
+ * Returns 0 for success, -EINVAL if the ucode is not available.
+ */
+static int cik_sdma_load_microcode(struct radeon_device *rdev)
+{
+	const __be32 *fw_data;
+	int i;
+
+	if (!rdev->sdma_fw)
+		return -EINVAL;
+
+	/* stop the gfx rings and rlc compute queues */
+	cik_sdma_gfx_stop(rdev);
+	cik_sdma_rlc_stop(rdev);
+
+	/* halt the MEs */
+	cik_sdma_enable(rdev, false);
+
+	/* sdma0 */
+	fw_data = (const __be32 *)rdev->sdma_fw->data;
+	WREG32(SDMA0_UCODE_ADDR + SDMA0_REGISTER_OFFSET, 0);
+	for (i = 0; i < CIK_SDMA_UCODE_SIZE; i++)
+		WREG32(SDMA0_UCODE_DATA + SDMA0_REGISTER_OFFSET, be32_to_cpup(fw_data++));
+	WREG32(SDMA0_UCODE_DATA + SDMA0_REGISTER_OFFSET, CIK_SDMA_UCODE_VERSION);
+
+	/* sdma1 */
+	fw_data = (const __be32 *)rdev->sdma_fw->data;
+	WREG32(SDMA0_UCODE_ADDR + SDMA1_REGISTER_OFFSET, 0);
+	for (i = 0; i < CIK_SDMA_UCODE_SIZE; i++)
+		WREG32(SDMA0_UCODE_DATA + SDMA1_REGISTER_OFFSET, be32_to_cpup(fw_data++));
+	WREG32(SDMA0_UCODE_DATA + SDMA1_REGISTER_OFFSET, CIK_SDMA_UCODE_VERSION);
+
+	WREG32(SDMA0_UCODE_ADDR + SDMA0_REGISTER_OFFSET, 0);
+	WREG32(SDMA0_UCODE_ADDR + SDMA1_REGISTER_OFFSET, 0);
+	return 0;
+}
+
+/**
+ * cik_sdma_resume - setup and start the async dma engines
+ *
+ * @rdev: radeon_device pointer
+ *
+ * Set up the DMA engines and enable them (CIK).
+ * Returns 0 for success, error for failure.
+ */
+int cik_sdma_resume(struct radeon_device *rdev)
+{
+	int r;
+
+	/* Reset dma */
+	WREG32(SRBM_SOFT_RESET, SOFT_RESET_SDMA | SOFT_RESET_SDMA1);
+	RREG32(SRBM_SOFT_RESET);
+	udelay(50);
+	WREG32(SRBM_SOFT_RESET, 0);
+	RREG32(SRBM_SOFT_RESET);
+
+	r = cik_sdma_load_microcode(rdev);
+	if (r)
+		return r;
+
+	/* unhalt the MEs */
+	cik_sdma_enable(rdev, true);
+
+	/* start the gfx rings and rlc compute queues */
+	r = cik_sdma_gfx_resume(rdev);
+	if (r)
+		return r;
+	r = cik_sdma_rlc_resume(rdev);
+	if (r)
+		return r;
+
+	return 0;
+}
+
+/**
+ * cik_sdma_fini - tear down the async dma engines
+ *
+ * @rdev: radeon_device pointer
+ *
+ * Stop the async dma engines and free the rings (CIK).
+ */
+void cik_sdma_fini(struct radeon_device *rdev)
+{
+	/* stop the gfx rings and rlc compute queues */
+	cik_sdma_gfx_stop(rdev);
+	cik_sdma_rlc_stop(rdev);
+	/* halt the MEs */
+	cik_sdma_enable(rdev, false);
+	radeon_ring_fini(rdev, &rdev->ring[R600_RING_TYPE_DMA_INDEX]);
+	radeon_ring_fini(rdev, &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX]);
+	/* XXX - compute dma queue tear down */
+}
+
+/**
+ * cik_copy_dma - copy pages using the DMA engine
+ *
+ * @rdev: radeon_device pointer
+ * @src_offset: src GPU address
+ * @dst_offset: dst GPU address
+ * @num_gpu_pages: number of GPU pages to xfer
+ * @fence: radeon fence object
+ *
+ * Copy GPU paging using the DMA engine (CIK).
+ * Used by the radeon ttm implementation to move pages if
+ * registered as the asic copy callback.
+ */
+int cik_copy_dma(struct radeon_device *rdev,
+		 uint64_t src_offset, uint64_t dst_offset,
+		 unsigned num_gpu_pages,
+		 struct radeon_fence **fence)
+{
+	struct radeon_semaphore *sem = NULL;
+	int ring_index = rdev->asic->copy.dma_ring_index;
+	struct radeon_ring *ring = &rdev->ring[ring_index];
+	u32 size_in_bytes, cur_size_in_bytes;
+	int i, num_loops;
+	int r = 0;
+
+	r = radeon_semaphore_create(rdev, &sem);
+	if (r) {
+		DRM_ERROR("radeon: moving bo (%d).\n", r);
+		return r;
+	}
+
+	size_in_bytes = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT);
+	num_loops = DIV_ROUND_UP(size_in_bytes, 0x1fffff);
+	r = radeon_ring_lock(rdev, ring, num_loops * 7 + 14);
+	if (r) {
+		DRM_ERROR("radeon: moving bo (%d).\n", r);
+		radeon_semaphore_free(rdev, &sem, NULL);
+		return r;
+	}
+
+	if (radeon_fence_need_sync(*fence, ring->idx)) {
+		radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring,
+					    ring->idx);
+		radeon_fence_note_sync(*fence, ring->idx);
+	} else {
+		radeon_semaphore_free(rdev, &sem, NULL);
+	}
+
+	for (i = 0; i < num_loops; i++) {
+		cur_size_in_bytes = size_in_bytes;
+		if (cur_size_in_bytes > 0x1fffff)
+			cur_size_in_bytes = 0x1fffff;
+		size_in_bytes -= cur_size_in_bytes;
+		radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_COPY, SDMA_COPY_SUB_OPCODE_LINEAR, 0));
+		radeon_ring_write(ring, cur_size_in_bytes);
+		radeon_ring_write(ring, 0); /* src/dst endian swap */
+		radeon_ring_write(ring, src_offset & 0xffffffff);
+		radeon_ring_write(ring, upper_32_bits(src_offset) & 0xffffffff);
+		radeon_ring_write(ring, dst_offset & 0xfffffffc);
+		radeon_ring_write(ring, upper_32_bits(dst_offset) & 0xffffffff);
+		src_offset += cur_size_in_bytes;
+		dst_offset += cur_size_in_bytes;
+	}
+
+	r = radeon_fence_emit(rdev, fence, ring->idx);
+	if (r) {
+		radeon_ring_unlock_undo(rdev, ring);
+		return r;
+	}
+
+	radeon_ring_unlock_commit(rdev, ring);
+	radeon_semaphore_free(rdev, &sem, *fence);
+
+	return r;
+}
+
+/**
+ * cik_sdma_ring_test - simple async dma engine test
+ *
+ * @rdev: radeon_device pointer
+ * @ring: radeon_ring structure holding ring information
+ *
+ * Test the DMA engine by writing using it to write an
+ * value to memory. (CIK).
+ * Returns 0 for success, error for failure.
+ */
+int cik_sdma_ring_test(struct radeon_device *rdev,
+		       struct radeon_ring *ring)
+{
+	unsigned i;
+	int r;
+	void __iomem *ptr = (void *)rdev->vram_scratch.ptr;
+	u32 tmp;
+
+	if (!ptr) {
+		DRM_ERROR("invalid vram scratch pointer\n");
+		return -EINVAL;
+	}
+
+	tmp = 0xCAFEDEAD;
+	writel(tmp, ptr);
+
+	r = radeon_ring_lock(rdev, ring, 4);
+	if (r) {
+		DRM_ERROR("radeon: dma failed to lock ring %d (%d).\n", ring->idx, r);
+		return r;
+	}
+	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0));
+	radeon_ring_write(ring, rdev->vram_scratch.gpu_addr & 0xfffffffc);
+	radeon_ring_write(ring, upper_32_bits(rdev->vram_scratch.gpu_addr) & 0xffffffff);
+	radeon_ring_write(ring, 1); /* number of DWs to follow */
+	radeon_ring_write(ring, 0xDEADBEEF);
+	radeon_ring_unlock_commit(rdev, ring);
+
+	for (i = 0; i < rdev->usec_timeout; i++) {
+		tmp = readl(ptr);
+		if (tmp == 0xDEADBEEF)
+			break;
+		DRM_UDELAY(1);
+	}
+
+	if (i < rdev->usec_timeout) {
+		DRM_INFO("ring test on %d succeeded in %d usecs\n", ring->idx, i);
+	} else {
+		DRM_ERROR("radeon: ring %d test failed (0x%08X)\n",
+			  ring->idx, tmp);
+		r = -EINVAL;
+	}
+	return r;
+}
+
+/**
+ * cik_sdma_ib_test - test an IB on the DMA engine
+ *
+ * @rdev: radeon_device pointer
+ * @ring: radeon_ring structure holding ring information
+ *
+ * Test a simple IB in the DMA ring (CIK).
+ * Returns 0 on success, error on failure.
+ */
+int cik_sdma_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
+{
+	struct radeon_ib ib;
+	unsigned i;
+	int r;
+	void __iomem *ptr = (void *)rdev->vram_scratch.ptr;
+	u32 tmp = 0;
+
+	if (!ptr) {
+		DRM_ERROR("invalid vram scratch pointer\n");
+		return -EINVAL;
+	}
+
+	tmp = 0xCAFEDEAD;
+	writel(tmp, ptr);
+
+	r = radeon_ib_get(rdev, ring->idx, &ib, NULL, 256);
+	if (r) {
+		DRM_ERROR("radeon: failed to get ib (%d).\n", r);
+		return r;
+	}
+
+	ib.ptr[0] = SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0);
+	ib.ptr[1] = rdev->vram_scratch.gpu_addr & 0xfffffffc;
+	ib.ptr[2] = upper_32_bits(rdev->vram_scratch.gpu_addr) & 0xffffffff;
+	ib.ptr[3] = 1;
+	ib.ptr[4] = 0xDEADBEEF;
+	ib.length_dw = 5;
+
+	r = radeon_ib_schedule(rdev, &ib, NULL);
+	if (r) {
+		radeon_ib_free(rdev, &ib);
+		DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
+		return r;
+	}
+	r = radeon_fence_wait(ib.fence, false);
+	if (r) {
+		DRM_ERROR("radeon: fence wait failed (%d).\n", r);
+		return r;
+	}
+	for (i = 0; i < rdev->usec_timeout; i++) {
+		tmp = readl(ptr);
+		if (tmp == 0xDEADBEEF)
+			break;
+		DRM_UDELAY(1);
+	}
+	if (i < rdev->usec_timeout) {
+		DRM_INFO("ib test on ring %d succeeded in %u usecs\n", ib.fence->ring, i);
+	} else {
+		DRM_ERROR("radeon: ib test failed (0x%08X)\n", tmp);
+		r = -EINVAL;
+	}
+	radeon_ib_free(rdev, &ib);
+	return r;
+}
+
+/**
+ * cik_sdma_is_lockup - Check if the DMA engine is locked up
+ *
+ * @rdev: radeon_device pointer
+ * @ring: radeon_ring structure holding ring information
+ *
+ * Check if the async DMA engine is locked up (CIK).
+ * Returns true if the engine appears to be locked up, false if not.
+ */
+bool cik_sdma_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
+{
+	u32 reset_mask = cik_gpu_check_soft_reset(rdev);
+	u32 mask;
+
+	if (ring->idx == R600_RING_TYPE_DMA_INDEX)
+		mask = RADEON_RESET_DMA;
+	else
+		mask = RADEON_RESET_DMA1;
+
+	if (!(reset_mask & mask)) {
+		radeon_ring_lockup_update(ring);
+		return false;
+	}
+	/* force ring activities */
+	radeon_ring_force_activity(rdev, ring);
+	return radeon_ring_test_lockup(rdev, ring);
+}
+
+/**
+ * cik_sdma_vm_set_page - update the page tables using sDMA
+ *
+ * @rdev: radeon_device pointer
+ * @ib: indirect buffer to fill with commands
+ * @pe: addr of the page entry
+ * @addr: dst addr to write into pe
+ * @count: number of page entries to update
+ * @incr: increase next addr by incr bytes
+ * @flags: access flags
+ *
+ * Update the page tables using sDMA (CIK).
+ */
+void cik_sdma_vm_set_page(struct radeon_device *rdev,
+			  struct radeon_ib *ib,
+			  uint64_t pe,
+			  uint64_t addr, unsigned count,
+			  uint32_t incr, uint32_t flags)
+{
+	uint32_t r600_flags = cayman_vm_page_flags(rdev, flags);
+	uint64_t value;
+	unsigned ndw;
+
+	if (flags & RADEON_VM_PAGE_SYSTEM) {
+		while (count) {
+			ndw = count * 2;
+			if (ndw > 0xFFFFE)
+				ndw = 0xFFFFE;
+
+			/* for non-physically contiguous pages (system) */
+			ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0);
+			ib->ptr[ib->length_dw++] = pe;
+			ib->ptr[ib->length_dw++] = upper_32_bits(pe);
+			ib->ptr[ib->length_dw++] = ndw;
+			for (; ndw > 0; ndw -= 2, --count, pe += 8) {
+				if (flags & RADEON_VM_PAGE_SYSTEM) {
+					value = radeon_vm_map_gart(rdev, addr);
+					value &= 0xFFFFFFFFFFFFF000ULL;
+				} else if (flags & RADEON_VM_PAGE_VALID) {
+					value = addr;
+				} else {
+					value = 0;
+				}
+				addr += incr;
+				value |= r600_flags;
+				ib->ptr[ib->length_dw++] = value;
+				ib->ptr[ib->length_dw++] = upper_32_bits(value);
+			}
+		}
+	} else {
+		while (count) {
+			ndw = count;
+			if (ndw > 0x7FFFF)
+				ndw = 0x7FFFF;
+
+			if (flags & RADEON_VM_PAGE_VALID)
+				value = addr;
+			else
+				value = 0;
+			/* for physically contiguous pages (vram) */
+			ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_GENERATE_PTE_PDE, 0, 0);
+			ib->ptr[ib->length_dw++] = pe; /* dst addr */
+			ib->ptr[ib->length_dw++] = upper_32_bits(pe);
+			ib->ptr[ib->length_dw++] = r600_flags; /* mask */
+			ib->ptr[ib->length_dw++] = 0;
+			ib->ptr[ib->length_dw++] = value; /* value */
+			ib->ptr[ib->length_dw++] = upper_32_bits(value);
+			ib->ptr[ib->length_dw++] = incr; /* increment size */
+			ib->ptr[ib->length_dw++] = 0;
+			ib->ptr[ib->length_dw++] = ndw; /* number of entries */
+			pe += ndw * 8;
+			addr += ndw * incr;
+			count -= ndw;
+		}
+	}
+	while (ib->length_dw & 0x7)
+		ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0);
+}
+
+/**
+ * cik_dma_vm_flush - cik vm flush using sDMA
+ *
+ * @rdev: radeon_device pointer
+ *
+ * Update the page table base and flush the VM TLB
+ * using sDMA (CIK).
+ */
+void cik_dma_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm)
+{
+	struct radeon_ring *ring = &rdev->ring[ridx];
+	u32 extra_bits = (SDMA_POLL_REG_MEM_EXTRA_OP(1) |
+			  SDMA_POLL_REG_MEM_EXTRA_FUNC(3)); /* == */
+	u32 ref_and_mask;
+
+	if (vm == NULL)
+		return;
+
+	if (ridx == R600_RING_TYPE_DMA_INDEX)
+		ref_and_mask = SDMA0;
+	else
+		ref_and_mask = SDMA1;
+
+	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
+	if (vm->id < 8) {
+		radeon_ring_write(ring, (VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2)) >> 2);
+	} else {
+		radeon_ring_write(ring, (VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((vm->id - 8) << 2)) >> 2);
+	}
+	radeon_ring_write(ring, vm->pd_gpu_addr >> 12);
+
+	/* update SH_MEM_* regs */
+	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
+	radeon_ring_write(ring, SRBM_GFX_CNTL >> 2);
+	radeon_ring_write(ring, VMID(vm->id));
+
+	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
+	radeon_ring_write(ring, SH_MEM_BASES >> 2);
+	radeon_ring_write(ring, 0);
+
+	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
+	radeon_ring_write(ring, SH_MEM_CONFIG >> 2);
+	radeon_ring_write(ring, 0);
+
+	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
+	radeon_ring_write(ring, SH_MEM_APE1_BASE >> 2);
+	radeon_ring_write(ring, 1);
+
+	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
+	radeon_ring_write(ring, SH_MEM_APE1_LIMIT >> 2);
+	radeon_ring_write(ring, 0);
+
+	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
+	radeon_ring_write(ring, SRBM_GFX_CNTL >> 2);
+	radeon_ring_write(ring, VMID(0));
+
+	/* flush HDP */
+	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_POLL_REG_MEM, 0, extra_bits));
+	radeon_ring_write(ring, GPU_HDP_FLUSH_DONE);
+	radeon_ring_write(ring, GPU_HDP_FLUSH_REQ);
+	radeon_ring_write(ring, ref_and_mask); /* REFERENCE */
+	radeon_ring_write(ring, ref_and_mask); /* MASK */
+	radeon_ring_write(ring, (4 << 16) | 10); /* RETRY_COUNT, POLL_INTERVAL */
+
+	/* flush TLB */
+	radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
+	radeon_ring_write(ring, VM_INVALIDATE_REQUEST >> 2);
+	radeon_ring_write(ring, 1 << vm->id);
+}
+
diff --git a/drivers/gpu/drm/radeon/evergreen.c b/drivers/gpu/drm/radeon/evergreen.c
index 52ed22333f0d..bbaa4f2056ce 100644
--- a/drivers/gpu/drm/radeon/evergreen.c
+++ b/drivers/gpu/drm/radeon/evergreen.c
@@ -3613,7 +3613,7 @@ bool evergreen_is_display_hung(struct radeon_device *rdev)
 	return true;
 }
 
-static u32 evergreen_gpu_check_soft_reset(struct radeon_device *rdev)
+u32 evergreen_gpu_check_soft_reset(struct radeon_device *rdev)
 {
 	u32 reset_mask = 0;
 	u32 tmp;
@@ -3839,28 +3839,6 @@ bool evergreen_gfx_is_lockup(struct radeon_device *rdev, struct radeon_ring *rin
 	return radeon_ring_test_lockup(rdev, ring);
 }
 
-/**
- * evergreen_dma_is_lockup - Check if the DMA engine is locked up
- *
- * @rdev: radeon_device pointer
- * @ring: radeon_ring structure holding ring information
- *
- * Check if the async DMA engine is locked up.
- * Returns true if the engine appears to be locked up, false if not.
- */
-bool evergreen_dma_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
-{
-	u32 reset_mask = evergreen_gpu_check_soft_reset(rdev);
-
-	if (!(reset_mask & RADEON_RESET_DMA)) {
-		radeon_ring_lockup_update(ring);
-		return false;
-	}
-	/* force ring activities */
-	radeon_ring_force_activity(rdev, ring);
-	return radeon_ring_test_lockup(rdev, ring);
-}
-
 /*
  * RLC
  */
@@ -5024,143 +5002,6 @@ restart_ih:
 	return IRQ_HANDLED;
 }
 
-/**
- * evergreen_dma_fence_ring_emit - emit a fence on the DMA ring
- *
- * @rdev: radeon_device pointer
- * @fence: radeon fence object
- *
- * Add a DMA fence packet to the ring to write
- * the fence seq number and DMA trap packet to generate
- * an interrupt if needed (evergreen-SI).
- */
-void evergreen_dma_fence_ring_emit(struct radeon_device *rdev,
-				   struct radeon_fence *fence)
-{
-	struct radeon_ring *ring = &rdev->ring[fence->ring];
-	u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
-	/* write the fence */
-	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_FENCE, 0, 0));
-	radeon_ring_write(ring, addr & 0xfffffffc);
-	radeon_ring_write(ring, (upper_32_bits(addr) & 0xff));
-	radeon_ring_write(ring, fence->seq);
-	/* generate an interrupt */
-	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_TRAP, 0, 0));
-	/* flush HDP */
-	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0));
-	radeon_ring_write(ring, (0xf << 16) | (HDP_MEM_COHERENCY_FLUSH_CNTL >> 2));
-	radeon_ring_write(ring, 1);
-}
-
-/**
- * evergreen_dma_ring_ib_execute - schedule an IB on the DMA engine
- *
- * @rdev: radeon_device pointer
- * @ib: IB object to schedule
- *
- * Schedule an IB in the DMA ring (evergreen).
- */
-void evergreen_dma_ring_ib_execute(struct radeon_device *rdev,
-				   struct radeon_ib *ib)
-{
-	struct radeon_ring *ring = &rdev->ring[ib->ring];
-
-	if (rdev->wb.enabled) {
-		u32 next_rptr = ring->wptr + 4;
-		while ((next_rptr & 7) != 5)
-			next_rptr++;
-		next_rptr += 3;
-		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_WRITE, 0, 1));
-		radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
-		radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xff);
-		radeon_ring_write(ring, next_rptr);
-	}
-
-	/* The indirect buffer packet must end on an 8 DW boundary in the DMA ring.
-	 * Pad as necessary with NOPs.
-	 */
-	while ((ring->wptr & 7) != 5)
-		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_NOP, 0, 0));
-	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_INDIRECT_BUFFER, 0, 0));
-	radeon_ring_write(ring, (ib->gpu_addr & 0xFFFFFFE0));
-	radeon_ring_write(ring, (ib->length_dw << 12) | (upper_32_bits(ib->gpu_addr) & 0xFF));
-
-}
-
-/**
- * evergreen_copy_dma - copy pages using the DMA engine
- *
- * @rdev: radeon_device pointer
- * @src_offset: src GPU address
- * @dst_offset: dst GPU address
- * @num_gpu_pages: number of GPU pages to xfer
- * @fence: radeon fence object
- *
- * Copy GPU paging using the DMA engine (evergreen-cayman).
- * Used by the radeon ttm implementation to move pages if
- * registered as the asic copy callback.
- */
-int evergreen_copy_dma(struct radeon_device *rdev,
-		       uint64_t src_offset, uint64_t dst_offset,
-		       unsigned num_gpu_pages,
-		       struct radeon_fence **fence)
-{
-	struct radeon_semaphore *sem = NULL;
-	int ring_index = rdev->asic->copy.dma_ring_index;
-	struct radeon_ring *ring = &rdev->ring[ring_index];
-	u32 size_in_dw, cur_size_in_dw;
-	int i, num_loops;
-	int r = 0;
-
-	r = radeon_semaphore_create(rdev, &sem);
-	if (r) {
-		DRM_ERROR("radeon: moving bo (%d).\n", r);
-		return r;
-	}
-
-	size_in_dw = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT) / 4;
-	num_loops = DIV_ROUND_UP(size_in_dw, 0xfffff);
-	r = radeon_ring_lock(rdev, ring, num_loops * 5 + 11);
-	if (r) {
-		DRM_ERROR("radeon: moving bo (%d).\n", r);
-		radeon_semaphore_free(rdev, &sem, NULL);
-		return r;
-	}
-
-	if (radeon_fence_need_sync(*fence, ring->idx)) {
-		radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring,
-					    ring->idx);
-		radeon_fence_note_sync(*fence, ring->idx);
-	} else {
-		radeon_semaphore_free(rdev, &sem, NULL);
-	}
-
-	for (i = 0; i < num_loops; i++) {
-		cur_size_in_dw = size_in_dw;
-		if (cur_size_in_dw > 0xFFFFF)
-			cur_size_in_dw = 0xFFFFF;
-		size_in_dw -= cur_size_in_dw;
-		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_COPY, 0, cur_size_in_dw));
-		radeon_ring_write(ring, dst_offset & 0xfffffffc);
-		radeon_ring_write(ring, src_offset & 0xfffffffc);
-		radeon_ring_write(ring, upper_32_bits(dst_offset) & 0xff);
-		radeon_ring_write(ring, upper_32_bits(src_offset) & 0xff);
-		src_offset += cur_size_in_dw * 4;
-		dst_offset += cur_size_in_dw * 4;
-	}
-
-	r = radeon_fence_emit(rdev, fence, ring->idx);
-	if (r) {
-		radeon_ring_unlock_undo(rdev, ring);
-		return r;
-	}
-
-	radeon_ring_unlock_commit(rdev, ring);
-	radeon_semaphore_free(rdev, &sem, *fence);
-
-	return r;
-}
-
 static int evergreen_startup(struct radeon_device *rdev)
 {
 	struct radeon_ring *ring;
diff --git a/drivers/gpu/drm/radeon/evergreen_dma.c b/drivers/gpu/drm/radeon/evergreen_dma.c
new file mode 100644
index 000000000000..6a0656d00ed0
--- /dev/null
+++ b/drivers/gpu/drm/radeon/evergreen_dma.c
@@ -0,0 +1,190 @@
+/*
+ * Copyright 2010 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Alex Deucher
+ */
+#include <drm/drmP.h>
+#include "radeon.h"
+#include "radeon_asic.h"
+#include "evergreend.h"
+
+u32 evergreen_gpu_check_soft_reset(struct radeon_device *rdev);
+
+/**
+ * evergreen_dma_fence_ring_emit - emit a fence on the DMA ring
+ *
+ * @rdev: radeon_device pointer
+ * @fence: radeon fence object
+ *
+ * Add a DMA fence packet to the ring to write
+ * the fence seq number and DMA trap packet to generate
+ * an interrupt if needed (evergreen-SI).
+ */
+void evergreen_dma_fence_ring_emit(struct radeon_device *rdev,
+				   struct radeon_fence *fence)
+{
+	struct radeon_ring *ring = &rdev->ring[fence->ring];
+	u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
+	/* write the fence */
+	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_FENCE, 0, 0));
+	radeon_ring_write(ring, addr & 0xfffffffc);
+	radeon_ring_write(ring, (upper_32_bits(addr) & 0xff));
+	radeon_ring_write(ring, fence->seq);
+	/* generate an interrupt */
+	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_TRAP, 0, 0));
+	/* flush HDP */
+	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0));
+	radeon_ring_write(ring, (0xf << 16) | (HDP_MEM_COHERENCY_FLUSH_CNTL >> 2));
+	radeon_ring_write(ring, 1);
+}
+
+/**
+ * evergreen_dma_ring_ib_execute - schedule an IB on the DMA engine
+ *
+ * @rdev: radeon_device pointer
+ * @ib: IB object to schedule
+ *
+ * Schedule an IB in the DMA ring (evergreen).
+ */
+void evergreen_dma_ring_ib_execute(struct radeon_device *rdev,
+				   struct radeon_ib *ib)
+{
+	struct radeon_ring *ring = &rdev->ring[ib->ring];
+
+	if (rdev->wb.enabled) {
+		u32 next_rptr = ring->wptr + 4;
+		while ((next_rptr & 7) != 5)
+			next_rptr++;
+		next_rptr += 3;
+		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_WRITE, 0, 1));
+		radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
+		radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xff);
+		radeon_ring_write(ring, next_rptr);
+	}
+
+	/* The indirect buffer packet must end on an 8 DW boundary in the DMA ring.
+	 * Pad as necessary with NOPs.
+	 */
+	while ((ring->wptr & 7) != 5)
+		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_NOP, 0, 0));
+	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_INDIRECT_BUFFER, 0, 0));
+	radeon_ring_write(ring, (ib->gpu_addr & 0xFFFFFFE0));
+	radeon_ring_write(ring, (ib->length_dw << 12) | (upper_32_bits(ib->gpu_addr) & 0xFF));
+
+}
+
+/**
+ * evergreen_copy_dma - copy pages using the DMA engine
+ *
+ * @rdev: radeon_device pointer
+ * @src_offset: src GPU address
+ * @dst_offset: dst GPU address
+ * @num_gpu_pages: number of GPU pages to xfer
+ * @fence: radeon fence object
+ *
+ * Copy GPU paging using the DMA engine (evergreen-cayman).
+ * Used by the radeon ttm implementation to move pages if
+ * registered as the asic copy callback.
+ */
+int evergreen_copy_dma(struct radeon_device *rdev,
+		       uint64_t src_offset, uint64_t dst_offset,
+		       unsigned num_gpu_pages,
+		       struct radeon_fence **fence)
+{
+	struct radeon_semaphore *sem = NULL;
+	int ring_index = rdev->asic->copy.dma_ring_index;
+	struct radeon_ring *ring = &rdev->ring[ring_index];
+	u32 size_in_dw, cur_size_in_dw;
+	int i, num_loops;
+	int r = 0;
+
+	r = radeon_semaphore_create(rdev, &sem);
+	if (r) {
+		DRM_ERROR("radeon: moving bo (%d).\n", r);
+		return r;
+	}
+
+	size_in_dw = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT) / 4;
+	num_loops = DIV_ROUND_UP(size_in_dw, 0xfffff);
+	r = radeon_ring_lock(rdev, ring, num_loops * 5 + 11);
+	if (r) {
+		DRM_ERROR("radeon: moving bo (%d).\n", r);
+		radeon_semaphore_free(rdev, &sem, NULL);
+		return r;
+	}
+
+	if (radeon_fence_need_sync(*fence, ring->idx)) {
+		radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring,
+					    ring->idx);
+		radeon_fence_note_sync(*fence, ring->idx);
+	} else {
+		radeon_semaphore_free(rdev, &sem, NULL);
+	}
+
+	for (i = 0; i < num_loops; i++) {
+		cur_size_in_dw = size_in_dw;
+		if (cur_size_in_dw > 0xFFFFF)
+			cur_size_in_dw = 0xFFFFF;
+		size_in_dw -= cur_size_in_dw;
+		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_COPY, 0, cur_size_in_dw));
+		radeon_ring_write(ring, dst_offset & 0xfffffffc);
+		radeon_ring_write(ring, src_offset & 0xfffffffc);
+		radeon_ring_write(ring, upper_32_bits(dst_offset) & 0xff);
+		radeon_ring_write(ring, upper_32_bits(src_offset) & 0xff);
+		src_offset += cur_size_in_dw * 4;
+		dst_offset += cur_size_in_dw * 4;
+	}
+
+	r = radeon_fence_emit(rdev, fence, ring->idx);
+	if (r) {
+		radeon_ring_unlock_undo(rdev, ring);
+		return r;
+	}
+
+	radeon_ring_unlock_commit(rdev, ring);
+	radeon_semaphore_free(rdev, &sem, *fence);
+
+	return r;
+}
+
+/**
+ * evergreen_dma_is_lockup - Check if the DMA engine is locked up
+ *
+ * @rdev: radeon_device pointer
+ * @ring: radeon_ring structure holding ring information
+ *
+ * Check if the async DMA engine is locked up.
+ * Returns true if the engine appears to be locked up, false if not.
+ */
+bool evergreen_dma_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
+{
+	u32 reset_mask = evergreen_gpu_check_soft_reset(rdev);
+
+	if (!(reset_mask & RADEON_RESET_DMA)) {
+		radeon_ring_lockup_update(ring);
+		return false;
+	}
+	/* force ring activities */
+	radeon_ring_force_activity(rdev, ring);
+	return radeon_ring_test_lockup(rdev, ring);
+}
+
+
diff --git a/drivers/gpu/drm/radeon/ni.c b/drivers/gpu/drm/radeon/ni.c
index 0205fa1594fa..2db8ce0023ac 100644
--- a/drivers/gpu/drm/radeon/ni.c
+++ b/drivers/gpu/drm/radeon/ni.c
@@ -174,6 +174,11 @@ extern void evergreen_pcie_gen2_enable(struct radeon_device *rdev);
 extern void evergreen_program_aspm(struct radeon_device *rdev);
 extern void sumo_rlc_fini(struct radeon_device *rdev);
 extern int sumo_rlc_init(struct radeon_device *rdev);
+extern void cayman_dma_vm_set_page(struct radeon_device *rdev,
+				   struct radeon_ib *ib,
+				   uint64_t pe,
+				   uint64_t addr, unsigned count,
+				   uint32_t incr, uint32_t flags);
 
 /* Firmware Names */
 MODULE_FIRMWARE("radeon/BARTS_pfp.bin");
@@ -1595,186 +1600,7 @@ static int cayman_cp_resume(struct radeon_device *rdev)
 	return 0;
 }
 
-/*
- * DMA
- * Starting with R600, the GPU has an asynchronous
- * DMA engine.  The programming model is very similar
- * to the 3D engine (ring buffer, IBs, etc.), but the
- * DMA controller has it's own packet format that is
- * different form the PM4 format used by the 3D engine.
- * It supports copying data, writing embedded data,
- * solid fills, and a number of other things.  It also
- * has support for tiling/detiling of buffers.
- * Cayman and newer support two asynchronous DMA engines.
- */
-/**
- * cayman_dma_ring_ib_execute - Schedule an IB on the DMA engine
- *
- * @rdev: radeon_device pointer
- * @ib: IB object to schedule
- *
- * Schedule an IB in the DMA ring (cayman-SI).
- */
-void cayman_dma_ring_ib_execute(struct radeon_device *rdev,
-				struct radeon_ib *ib)
-{
-	struct radeon_ring *ring = &rdev->ring[ib->ring];
-
-	if (rdev->wb.enabled) {
-		u32 next_rptr = ring->wptr + 4;
-		while ((next_rptr & 7) != 5)
-			next_rptr++;
-		next_rptr += 3;
-		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 1));
-		radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
-		radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xff);
-		radeon_ring_write(ring, next_rptr);
-	}
-
-	/* The indirect buffer packet must end on an 8 DW boundary in the DMA ring.
-	 * Pad as necessary with NOPs.
-	 */
-	while ((ring->wptr & 7) != 5)
-		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0));
-	radeon_ring_write(ring, DMA_IB_PACKET(DMA_PACKET_INDIRECT_BUFFER, ib->vm ? ib->vm->id : 0, 0));
-	radeon_ring_write(ring, (ib->gpu_addr & 0xFFFFFFE0));
-	radeon_ring_write(ring, (ib->length_dw << 12) | (upper_32_bits(ib->gpu_addr) & 0xFF));
-
-}
-
-/**
- * cayman_dma_stop - stop the async dma engines
- *
- * @rdev: radeon_device pointer
- *
- * Stop the async dma engines (cayman-SI).
- */
-void cayman_dma_stop(struct radeon_device *rdev)
-{
-	u32 rb_cntl;
-
-	radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
-
-	/* dma0 */
-	rb_cntl = RREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET);
-	rb_cntl &= ~DMA_RB_ENABLE;
-	WREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET, rb_cntl);
-
-	/* dma1 */
-	rb_cntl = RREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET);
-	rb_cntl &= ~DMA_RB_ENABLE;
-	WREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET, rb_cntl);
-
-	rdev->ring[R600_RING_TYPE_DMA_INDEX].ready = false;
-	rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX].ready = false;
-}
-
-/**
- * cayman_dma_resume - setup and start the async dma engines
- *
- * @rdev: radeon_device pointer
- *
- * Set up the DMA ring buffers and enable them. (cayman-SI).
- * Returns 0 for success, error for failure.
- */
-int cayman_dma_resume(struct radeon_device *rdev)
-{
-	struct radeon_ring *ring;
-	u32 rb_cntl, dma_cntl, ib_cntl;
-	u32 rb_bufsz;
-	u32 reg_offset, wb_offset;
-	int i, r;
-
-	/* Reset dma */
-	WREG32(SRBM_SOFT_RESET, SOFT_RESET_DMA | SOFT_RESET_DMA1);
-	RREG32(SRBM_SOFT_RESET);
-	udelay(50);
-	WREG32(SRBM_SOFT_RESET, 0);
-
-	for (i = 0; i < 2; i++) {
-		if (i == 0) {
-			ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
-			reg_offset = DMA0_REGISTER_OFFSET;
-			wb_offset = R600_WB_DMA_RPTR_OFFSET;
-		} else {
-			ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
-			reg_offset = DMA1_REGISTER_OFFSET;
-			wb_offset = CAYMAN_WB_DMA1_RPTR_OFFSET;
-		}
-
-		WREG32(DMA_SEM_INCOMPLETE_TIMER_CNTL + reg_offset, 0);
-		WREG32(DMA_SEM_WAIT_FAIL_TIMER_CNTL + reg_offset, 0);
-
-		/* Set ring buffer size in dwords */
-		rb_bufsz = drm_order(ring->ring_size / 4);
-		rb_cntl = rb_bufsz << 1;
-#ifdef __BIG_ENDIAN
-		rb_cntl |= DMA_RB_SWAP_ENABLE | DMA_RPTR_WRITEBACK_SWAP_ENABLE;
-#endif
-		WREG32(DMA_RB_CNTL + reg_offset, rb_cntl);
-
-		/* Initialize the ring buffer's read and write pointers */
-		WREG32(DMA_RB_RPTR + reg_offset, 0);
-		WREG32(DMA_RB_WPTR + reg_offset, 0);
-
-		/* set the wb address whether it's enabled or not */
-		WREG32(DMA_RB_RPTR_ADDR_HI + reg_offset,
-		       upper_32_bits(rdev->wb.gpu_addr + wb_offset) & 0xFF);
-		WREG32(DMA_RB_RPTR_ADDR_LO + reg_offset,
-		       ((rdev->wb.gpu_addr + wb_offset) & 0xFFFFFFFC));
-
-		if (rdev->wb.enabled)
-			rb_cntl |= DMA_RPTR_WRITEBACK_ENABLE;
-
-		WREG32(DMA_RB_BASE + reg_offset, ring->gpu_addr >> 8);
-
-		/* enable DMA IBs */
-		ib_cntl = DMA_IB_ENABLE | CMD_VMID_FORCE;
-#ifdef __BIG_ENDIAN
-		ib_cntl |= DMA_IB_SWAP_ENABLE;
-#endif
-		WREG32(DMA_IB_CNTL + reg_offset, ib_cntl);
-
-		dma_cntl = RREG32(DMA_CNTL + reg_offset);
-		dma_cntl &= ~CTXEMPTY_INT_ENABLE;
-		WREG32(DMA_CNTL + reg_offset, dma_cntl);
-
-		ring->wptr = 0;
-		WREG32(DMA_RB_WPTR + reg_offset, ring->wptr << 2);
-
-		ring->rptr = RREG32(DMA_RB_RPTR + reg_offset) >> 2;
-
-		WREG32(DMA_RB_CNTL + reg_offset, rb_cntl | DMA_RB_ENABLE);
-
-		ring->ready = true;
-
-		r = radeon_ring_test(rdev, ring->idx, ring);
-		if (r) {
-			ring->ready = false;
-			return r;
-		}
-	}
-
-	radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
-
-	return 0;
-}
-
-/**
- * cayman_dma_fini - tear down the async dma engines
- *
- * @rdev: radeon_device pointer
- *
- * Stop the async dma engines and free the rings (cayman-SI).
- */
-void cayman_dma_fini(struct radeon_device *rdev)
-{
-	cayman_dma_stop(rdev);
-	radeon_ring_fini(rdev, &rdev->ring[R600_RING_TYPE_DMA_INDEX]);
-	radeon_ring_fini(rdev, &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX]);
-}
-
-static u32 cayman_gpu_check_soft_reset(struct radeon_device *rdev)
+u32 cayman_gpu_check_soft_reset(struct radeon_device *rdev)
 {
 	u32 reset_mask = 0;
 	u32 tmp;
@@ -2027,34 +1853,6 @@ bool cayman_gfx_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
 	return radeon_ring_test_lockup(rdev, ring);
 }
 
-/**
- * cayman_dma_is_lockup - Check if the DMA engine is locked up
- *
- * @rdev: radeon_device pointer
- * @ring: radeon_ring structure holding ring information
- *
- * Check if the async DMA engine is locked up.
- * Returns true if the engine appears to be locked up, false if not.
- */
-bool cayman_dma_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
-{
-	u32 reset_mask = cayman_gpu_check_soft_reset(rdev);
-	u32 mask;
-
-	if (ring->idx == R600_RING_TYPE_DMA_INDEX)
-		mask = RADEON_RESET_DMA;
-	else
-		mask = RADEON_RESET_DMA1;
-
-	if (!(reset_mask & mask)) {
-		radeon_ring_lockup_update(ring);
-		return false;
-	}
-	/* force ring activities */
-	radeon_ring_force_activity(rdev, ring);
-	return radeon_ring_test_lockup(rdev, ring);
-}
-
 static int cayman_startup(struct radeon_device *rdev)
 {
 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
@@ -2658,61 +2456,7 @@ void cayman_vm_set_page(struct radeon_device *rdev,
 			}
 		}
 	} else {
-		if ((flags & RADEON_VM_PAGE_SYSTEM) ||
-		    (count == 1)) {
-			while (count) {
-				ndw = count * 2;
-				if (ndw > 0xFFFFE)
-					ndw = 0xFFFFE;
-
-				/* for non-physically contiguous pages (system) */
-				ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_WRITE, 0, 0, ndw);
-				ib->ptr[ib->length_dw++] = pe;
-				ib->ptr[ib->length_dw++] = upper_32_bits(pe) & 0xff;
-				for (; ndw > 0; ndw -= 2, --count, pe += 8) {
-					if (flags & RADEON_VM_PAGE_SYSTEM) {
-						value = radeon_vm_map_gart(rdev, addr);
-						value &= 0xFFFFFFFFFFFFF000ULL;
-					} else if (flags & RADEON_VM_PAGE_VALID) {
-						value = addr;
-					} else {
-						value = 0;
-					}
-					addr += incr;
-					value |= r600_flags;
-					ib->ptr[ib->length_dw++] = value;
-					ib->ptr[ib->length_dw++] = upper_32_bits(value);
-				}
-			}
-			while (ib->length_dw & 0x7)
-				ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0);
-		} else {
-			while (count) {
-				ndw = count * 2;
-				if (ndw > 0xFFFFE)
-					ndw = 0xFFFFE;
-
-				if (flags & RADEON_VM_PAGE_VALID)
-					value = addr;
-				else
-					value = 0;
-				/* for physically contiguous pages (vram) */
-				ib->ptr[ib->length_dw++] = DMA_PTE_PDE_PACKET(ndw);
-				ib->ptr[ib->length_dw++] = pe; /* dst addr */
-				ib->ptr[ib->length_dw++] = upper_32_bits(pe) & 0xff;
-				ib->ptr[ib->length_dw++] = r600_flags; /* mask */
-				ib->ptr[ib->length_dw++] = 0;
-				ib->ptr[ib->length_dw++] = value; /* value */
-				ib->ptr[ib->length_dw++] = upper_32_bits(value);
-				ib->ptr[ib->length_dw++] = incr; /* increment size */
-				ib->ptr[ib->length_dw++] = 0;
-				pe += ndw * 4;
-				addr += (ndw / 2) * incr;
-				count -= ndw / 2;
-			}
-		}
-		while (ib->length_dw & 0x7)
-			ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0);
+		cayman_dma_vm_set_page(rdev, ib, pe, addr, count, incr, flags);
 	}
 }
 
@@ -2746,26 +2490,3 @@ void cayman_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm)
 	radeon_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
 	radeon_ring_write(ring, 0x0);
 }
-
-void cayman_dma_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm)
-{
-	struct radeon_ring *ring = &rdev->ring[ridx];
-
-	if (vm == NULL)
-		return;
-
-	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0));
-	radeon_ring_write(ring, (0xf << 16) | ((VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2)) >> 2));
-	radeon_ring_write(ring, vm->pd_gpu_addr >> 12);
-
-	/* flush hdp cache */
-	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0));
-	radeon_ring_write(ring, (0xf << 16) | (HDP_MEM_COHERENCY_FLUSH_CNTL >> 2));
-	radeon_ring_write(ring, 1);
-
-	/* bits 0-7 are the VM contexts0-7 */
-	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0));
-	radeon_ring_write(ring, (0xf << 16) | (VM_INVALIDATE_REQUEST >> 2));
-	radeon_ring_write(ring, 1 << vm->id);
-}
-
diff --git a/drivers/gpu/drm/radeon/ni_dma.c b/drivers/gpu/drm/radeon/ni_dma.c
new file mode 100644
index 000000000000..0f3c0baea4a6
--- /dev/null
+++ b/drivers/gpu/drm/radeon/ni_dma.c
@@ -0,0 +1,338 @@
+/*
+ * Copyright 2010 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Alex Deucher
+ */
+#include <drm/drmP.h>
+#include "radeon.h"
+#include "radeon_asic.h"
+#include "nid.h"
+
+u32 cayman_gpu_check_soft_reset(struct radeon_device *rdev);
+
+/*
+ * DMA
+ * Starting with R600, the GPU has an asynchronous
+ * DMA engine.  The programming model is very similar
+ * to the 3D engine (ring buffer, IBs, etc.), but the
+ * DMA controller has it's own packet format that is
+ * different form the PM4 format used by the 3D engine.
+ * It supports copying data, writing embedded data,
+ * solid fills, and a number of other things.  It also
+ * has support for tiling/detiling of buffers.
+ * Cayman and newer support two asynchronous DMA engines.
+ */
+
+/**
+ * cayman_dma_ring_ib_execute - Schedule an IB on the DMA engine
+ *
+ * @rdev: radeon_device pointer
+ * @ib: IB object to schedule
+ *
+ * Schedule an IB in the DMA ring (cayman-SI).
+ */
+void cayman_dma_ring_ib_execute(struct radeon_device *rdev,
+				struct radeon_ib *ib)
+{
+	struct radeon_ring *ring = &rdev->ring[ib->ring];
+
+	if (rdev->wb.enabled) {
+		u32 next_rptr = ring->wptr + 4;
+		while ((next_rptr & 7) != 5)
+			next_rptr++;
+		next_rptr += 3;
+		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 1));
+		radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
+		radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xff);
+		radeon_ring_write(ring, next_rptr);
+	}
+
+	/* The indirect buffer packet must end on an 8 DW boundary in the DMA ring.
+	 * Pad as necessary with NOPs.
+	 */
+	while ((ring->wptr & 7) != 5)
+		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0));
+	radeon_ring_write(ring, DMA_IB_PACKET(DMA_PACKET_INDIRECT_BUFFER, ib->vm ? ib->vm->id : 0, 0));
+	radeon_ring_write(ring, (ib->gpu_addr & 0xFFFFFFE0));
+	radeon_ring_write(ring, (ib->length_dw << 12) | (upper_32_bits(ib->gpu_addr) & 0xFF));
+
+}
+
+/**
+ * cayman_dma_stop - stop the async dma engines
+ *
+ * @rdev: radeon_device pointer
+ *
+ * Stop the async dma engines (cayman-SI).
+ */
+void cayman_dma_stop(struct radeon_device *rdev)
+{
+	u32 rb_cntl;
+
+	radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
+
+	/* dma0 */
+	rb_cntl = RREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET);
+	rb_cntl &= ~DMA_RB_ENABLE;
+	WREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET, rb_cntl);
+
+	/* dma1 */
+	rb_cntl = RREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET);
+	rb_cntl &= ~DMA_RB_ENABLE;
+	WREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET, rb_cntl);
+
+	rdev->ring[R600_RING_TYPE_DMA_INDEX].ready = false;
+	rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX].ready = false;
+}
+
+/**
+ * cayman_dma_resume - setup and start the async dma engines
+ *
+ * @rdev: radeon_device pointer
+ *
+ * Set up the DMA ring buffers and enable them. (cayman-SI).
+ * Returns 0 for success, error for failure.
+ */
+int cayman_dma_resume(struct radeon_device *rdev)
+{
+	struct radeon_ring *ring;
+	u32 rb_cntl, dma_cntl, ib_cntl;
+	u32 rb_bufsz;
+	u32 reg_offset, wb_offset;
+	int i, r;
+
+	/* Reset dma */
+	WREG32(SRBM_SOFT_RESET, SOFT_RESET_DMA | SOFT_RESET_DMA1);
+	RREG32(SRBM_SOFT_RESET);
+	udelay(50);
+	WREG32(SRBM_SOFT_RESET, 0);
+
+	for (i = 0; i < 2; i++) {
+		if (i == 0) {
+			ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
+			reg_offset = DMA0_REGISTER_OFFSET;
+			wb_offset = R600_WB_DMA_RPTR_OFFSET;
+		} else {
+			ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
+			reg_offset = DMA1_REGISTER_OFFSET;
+			wb_offset = CAYMAN_WB_DMA1_RPTR_OFFSET;
+		}
+
+		WREG32(DMA_SEM_INCOMPLETE_TIMER_CNTL + reg_offset, 0);
+		WREG32(DMA_SEM_WAIT_FAIL_TIMER_CNTL + reg_offset, 0);
+
+		/* Set ring buffer size in dwords */
+		rb_bufsz = drm_order(ring->ring_size / 4);
+		rb_cntl = rb_bufsz << 1;
+#ifdef __BIG_ENDIAN
+		rb_cntl |= DMA_RB_SWAP_ENABLE | DMA_RPTR_WRITEBACK_SWAP_ENABLE;
+#endif
+		WREG32(DMA_RB_CNTL + reg_offset, rb_cntl);
+
+		/* Initialize the ring buffer's read and write pointers */
+		WREG32(DMA_RB_RPTR + reg_offset, 0);
+		WREG32(DMA_RB_WPTR + reg_offset, 0);
+
+		/* set the wb address whether it's enabled or not */
+		WREG32(DMA_RB_RPTR_ADDR_HI + reg_offset,
+		       upper_32_bits(rdev->wb.gpu_addr + wb_offset) & 0xFF);
+		WREG32(DMA_RB_RPTR_ADDR_LO + reg_offset,
+		       ((rdev->wb.gpu_addr + wb_offset) & 0xFFFFFFFC));
+
+		if (rdev->wb.enabled)
+			rb_cntl |= DMA_RPTR_WRITEBACK_ENABLE;
+
+		WREG32(DMA_RB_BASE + reg_offset, ring->gpu_addr >> 8);
+
+		/* enable DMA IBs */
+		ib_cntl = DMA_IB_ENABLE | CMD_VMID_FORCE;
+#ifdef __BIG_ENDIAN
+		ib_cntl |= DMA_IB_SWAP_ENABLE;
+#endif
+		WREG32(DMA_IB_CNTL + reg_offset, ib_cntl);
+
+		dma_cntl = RREG32(DMA_CNTL + reg_offset);
+		dma_cntl &= ~CTXEMPTY_INT_ENABLE;
+		WREG32(DMA_CNTL + reg_offset, dma_cntl);
+
+		ring->wptr = 0;
+		WREG32(DMA_RB_WPTR + reg_offset, ring->wptr << 2);
+
+		ring->rptr = RREG32(DMA_RB_RPTR + reg_offset) >> 2;
+
+		WREG32(DMA_RB_CNTL + reg_offset, rb_cntl | DMA_RB_ENABLE);
+
+		ring->ready = true;
+
+		r = radeon_ring_test(rdev, ring->idx, ring);
+		if (r) {
+			ring->ready = false;
+			return r;
+		}
+	}
+
+	radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
+
+	return 0;
+}
+
+/**
+ * cayman_dma_fini - tear down the async dma engines
+ *
+ * @rdev: radeon_device pointer
+ *
+ * Stop the async dma engines and free the rings (cayman-SI).
+ */
+void cayman_dma_fini(struct radeon_device *rdev)
+{
+	cayman_dma_stop(rdev);
+	radeon_ring_fini(rdev, &rdev->ring[R600_RING_TYPE_DMA_INDEX]);
+	radeon_ring_fini(rdev, &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX]);
+}
+
+/**
+ * cayman_dma_is_lockup - Check if the DMA engine is locked up
+ *
+ * @rdev: radeon_device pointer
+ * @ring: radeon_ring structure holding ring information
+ *
+ * Check if the async DMA engine is locked up.
+ * Returns true if the engine appears to be locked up, false if not.
+ */
+bool cayman_dma_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
+{
+	u32 reset_mask = cayman_gpu_check_soft_reset(rdev);
+	u32 mask;
+
+	if (ring->idx == R600_RING_TYPE_DMA_INDEX)
+		mask = RADEON_RESET_DMA;
+	else
+		mask = RADEON_RESET_DMA1;
+
+	if (!(reset_mask & mask)) {
+		radeon_ring_lockup_update(ring);
+		return false;
+	}
+	/* force ring activities */
+	radeon_ring_force_activity(rdev, ring);
+	return radeon_ring_test_lockup(rdev, ring);
+}
+
+/**
+ * cayman_dma_vm_set_page - update the page tables using the DMA
+ *
+ * @rdev: radeon_device pointer
+ * @ib: indirect buffer to fill with commands
+ * @pe: addr of the page entry
+ * @addr: dst addr to write into pe
+ * @count: number of page entries to update
+ * @incr: increase next addr by incr bytes
+ * @flags: access flags
+ * @r600_flags: hw access flags 
+ *
+ * Update the page tables using the DMA (cayman/TN).
+ */
+void cayman_dma_vm_set_page(struct radeon_device *rdev,
+			    struct radeon_ib *ib,
+			    uint64_t pe,
+			    uint64_t addr, unsigned count,
+			    uint32_t incr, uint32_t flags)
+{
+	uint32_t r600_flags = cayman_vm_page_flags(rdev, flags);
+	uint64_t value;
+	unsigned ndw;
+
+	if ((flags & RADEON_VM_PAGE_SYSTEM) || (count == 1)) {
+		while (count) {
+			ndw = count * 2;
+			if (ndw > 0xFFFFE)
+				ndw = 0xFFFFE;
+
+			/* for non-physically contiguous pages (system) */
+			ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_WRITE, 0, 0, ndw);
+			ib->ptr[ib->length_dw++] = pe;
+			ib->ptr[ib->length_dw++] = upper_32_bits(pe) & 0xff;
+			for (; ndw > 0; ndw -= 2, --count, pe += 8) {
+				if (flags & RADEON_VM_PAGE_SYSTEM) {
+					value = radeon_vm_map_gart(rdev, addr);
+					value &= 0xFFFFFFFFFFFFF000ULL;
+				} else if (flags & RADEON_VM_PAGE_VALID) {
+					value = addr;
+				} else {
+					value = 0;
+				}
+				addr += incr;
+				value |= r600_flags;
+				ib->ptr[ib->length_dw++] = value;
+				ib->ptr[ib->length_dw++] = upper_32_bits(value);
+			}
+		}
+	} else {
+		while (count) {
+			ndw = count * 2;
+			if (ndw > 0xFFFFE)
+				ndw = 0xFFFFE;
+
+			if (flags & RADEON_VM_PAGE_VALID)
+				value = addr;
+			else
+				value = 0;
+			/* for physically contiguous pages (vram) */
+			ib->ptr[ib->length_dw++] = DMA_PTE_PDE_PACKET(ndw);
+			ib->ptr[ib->length_dw++] = pe; /* dst addr */
+			ib->ptr[ib->length_dw++] = upper_32_bits(pe) & 0xff;
+			ib->ptr[ib->length_dw++] = r600_flags; /* mask */
+			ib->ptr[ib->length_dw++] = 0;
+			ib->ptr[ib->length_dw++] = value; /* value */
+			ib->ptr[ib->length_dw++] = upper_32_bits(value);
+			ib->ptr[ib->length_dw++] = incr; /* increment size */
+			ib->ptr[ib->length_dw++] = 0;
+			pe += ndw * 4;
+			addr += (ndw / 2) * incr;
+			count -= ndw / 2;
+		}
+	}
+	while (ib->length_dw & 0x7)
+		ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0);
+}
+
+void cayman_dma_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm)
+{
+	struct radeon_ring *ring = &rdev->ring[ridx];
+
+	if (vm == NULL)
+		return;
+
+	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0));
+	radeon_ring_write(ring, (0xf << 16) | ((VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2)) >> 2));
+	radeon_ring_write(ring, vm->pd_gpu_addr >> 12);
+
+	/* flush hdp cache */
+	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0));
+	radeon_ring_write(ring, (0xf << 16) | (HDP_MEM_COHERENCY_FLUSH_CNTL >> 2));
+	radeon_ring_write(ring, 1);
+
+	/* bits 0-7 are the VM contexts0-7 */
+	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0));
+	radeon_ring_write(ring, (0xf << 16) | (VM_INVALIDATE_REQUEST >> 2));
+	radeon_ring_write(ring, 1 << vm->id);
+}
+
diff --git a/drivers/gpu/drm/radeon/r600.c b/drivers/gpu/drm/radeon/r600.c
index 3a08ef92d33f..087cff444ba2 100644
--- a/drivers/gpu/drm/radeon/r600.c
+++ b/drivers/gpu/drm/radeon/r600.c
@@ -1374,7 +1374,7 @@ static bool r600_is_display_hung(struct radeon_device *rdev)
 	return true;
 }
 
-static u32 r600_gpu_check_soft_reset(struct radeon_device *rdev)
+u32 r600_gpu_check_soft_reset(struct radeon_device *rdev)
 {
 	u32 reset_mask = 0;
 	u32 tmp;
@@ -1622,28 +1622,6 @@ bool r600_gfx_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
 	return radeon_ring_test_lockup(rdev, ring);
 }
 
-/**
- * r600_dma_is_lockup - Check if the DMA engine is locked up
- *
- * @rdev: radeon_device pointer
- * @ring: radeon_ring structure holding ring information
- *
- * Check if the async DMA engine is locked up.
- * Returns true if the engine appears to be locked up, false if not.
- */
-bool r600_dma_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
-{
-	u32 reset_mask = r600_gpu_check_soft_reset(rdev);
-
-	if (!(reset_mask & RADEON_RESET_DMA)) {
-		radeon_ring_lockup_update(ring);
-		return false;
-	}
-	/* force ring activities */
-	radeon_ring_force_activity(rdev, ring);
-	return radeon_ring_test_lockup(rdev, ring);
-}
-
 u32 r6xx_remap_render_backend(struct radeon_device *rdev,
 			      u32 tiling_pipe_num,
 			      u32 max_rb_num,
@@ -2494,176 +2472,6 @@ void r600_cp_fini(struct radeon_device *rdev)
 }
 
 /*
- * DMA
- * Starting with R600, the GPU has an asynchronous
- * DMA engine.  The programming model is very similar
- * to the 3D engine (ring buffer, IBs, etc.), but the
- * DMA controller has it's own packet format that is
- * different form the PM4 format used by the 3D engine.
- * It supports copying data, writing embedded data,
- * solid fills, and a number of other things.  It also
- * has support for tiling/detiling of buffers.
- */
-
-/**
- * r600_dma_get_rptr - get the current read pointer
- *
- * @rdev: radeon_device pointer
- * @ring: radeon ring pointer
- *
- * Get the current rptr from the hardware (r6xx+).
- */
-uint32_t r600_dma_get_rptr(struct radeon_device *rdev,
-			   struct radeon_ring *ring)
-{
-	return (radeon_ring_generic_get_rptr(rdev, ring) & 0x3fffc) >> 2;
-}
-
-/**
- * r600_dma_get_wptr - get the current write pointer
- *
- * @rdev: radeon_device pointer
- * @ring: radeon ring pointer
- *
- * Get the current wptr from the hardware (r6xx+).
- */
-uint32_t r600_dma_get_wptr(struct radeon_device *rdev,
-			   struct radeon_ring *ring)
-{
-	return (RREG32(ring->wptr_reg) & 0x3fffc) >> 2;
-}
-
-/**
- * r600_dma_set_wptr - commit the write pointer
- *
- * @rdev: radeon_device pointer
- * @ring: radeon ring pointer
- *
- * Write the wptr back to the hardware (r6xx+).
- */
-void r600_dma_set_wptr(struct radeon_device *rdev,
-		       struct radeon_ring *ring)
-{
-	WREG32(ring->wptr_reg, (ring->wptr << 2) & 0x3fffc);
-}
-
-/**
- * r600_dma_stop - stop the async dma engine
- *
- * @rdev: radeon_device pointer
- *
- * Stop the async dma engine (r6xx-evergreen).
- */
-void r600_dma_stop(struct radeon_device *rdev)
-{
-	u32 rb_cntl = RREG32(DMA_RB_CNTL);
-
-	radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
-
-	rb_cntl &= ~DMA_RB_ENABLE;
-	WREG32(DMA_RB_CNTL, rb_cntl);
-
-	rdev->ring[R600_RING_TYPE_DMA_INDEX].ready = false;
-}
-
-/**
- * r600_dma_resume - setup and start the async dma engine
- *
- * @rdev: radeon_device pointer
- *
- * Set up the DMA ring buffer and enable it. (r6xx-evergreen).
- * Returns 0 for success, error for failure.
- */
-int r600_dma_resume(struct radeon_device *rdev)
-{
-	struct radeon_ring *ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
-	u32 rb_cntl, dma_cntl, ib_cntl;
-	u32 rb_bufsz;
-	int r;
-
-	/* Reset dma */
-	if (rdev->family >= CHIP_RV770)
-		WREG32(SRBM_SOFT_RESET, RV770_SOFT_RESET_DMA);
-	else
-		WREG32(SRBM_SOFT_RESET, SOFT_RESET_DMA);
-	RREG32(SRBM_SOFT_RESET);
-	udelay(50);
-	WREG32(SRBM_SOFT_RESET, 0);
-
-	WREG32(DMA_SEM_INCOMPLETE_TIMER_CNTL, 0);
-	WREG32(DMA_SEM_WAIT_FAIL_TIMER_CNTL, 0);
-
-	/* Set ring buffer size in dwords */
-	rb_bufsz = drm_order(ring->ring_size / 4);
-	rb_cntl = rb_bufsz << 1;
-#ifdef __BIG_ENDIAN
-	rb_cntl |= DMA_RB_SWAP_ENABLE | DMA_RPTR_WRITEBACK_SWAP_ENABLE;
-#endif
-	WREG32(DMA_RB_CNTL, rb_cntl);
-
-	/* Initialize the ring buffer's read and write pointers */
-	WREG32(DMA_RB_RPTR, 0);
-	WREG32(DMA_RB_WPTR, 0);
-
-	/* set the wb address whether it's enabled or not */
-	WREG32(DMA_RB_RPTR_ADDR_HI,
-	       upper_32_bits(rdev->wb.gpu_addr + R600_WB_DMA_RPTR_OFFSET) & 0xFF);
-	WREG32(DMA_RB_RPTR_ADDR_LO,
-	       ((rdev->wb.gpu_addr + R600_WB_DMA_RPTR_OFFSET) & 0xFFFFFFFC));
-
-	if (rdev->wb.enabled)
-		rb_cntl |= DMA_RPTR_WRITEBACK_ENABLE;
-
-	WREG32(DMA_RB_BASE, ring->gpu_addr >> 8);
-
-	/* enable DMA IBs */
-	ib_cntl = DMA_IB_ENABLE;
-#ifdef __BIG_ENDIAN
-	ib_cntl |= DMA_IB_SWAP_ENABLE;
-#endif
-	WREG32(DMA_IB_CNTL, ib_cntl);
-
-	dma_cntl = RREG32(DMA_CNTL);
-	dma_cntl &= ~CTXEMPTY_INT_ENABLE;
-	WREG32(DMA_CNTL, dma_cntl);
-
-	if (rdev->family >= CHIP_RV770)
-		WREG32(DMA_MODE, 1);
-
-	ring->wptr = 0;
-	WREG32(DMA_RB_WPTR, ring->wptr << 2);
-
-	ring->rptr = RREG32(DMA_RB_RPTR) >> 2;
-
-	WREG32(DMA_RB_CNTL, rb_cntl | DMA_RB_ENABLE);
-
-	ring->ready = true;
-
-	r = radeon_ring_test(rdev, R600_RING_TYPE_DMA_INDEX, ring);
-	if (r) {
-		ring->ready = false;
-		return r;
-	}
-
-	radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
-
-	return 0;
-}
-
-/**
- * r600_dma_fini - tear down the async dma engine
- *
- * @rdev: radeon_device pointer
- *
- * Stop the async dma engine and free the ring (r6xx-evergreen).
- */
-void r600_dma_fini(struct radeon_device *rdev)
-{
-	r600_dma_stop(rdev);
-	radeon_ring_fini(rdev, &rdev->ring[R600_RING_TYPE_DMA_INDEX]);
-}
-
-/*
  * GPU scratch registers helpers function.
  */
 void r600_scratch_init(struct radeon_device *rdev)
@@ -2718,60 +2526,6 @@ int r600_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
 	return r;
 }
 
-/**
- * r600_dma_ring_test - simple async dma engine test
- *
- * @rdev: radeon_device pointer
- * @ring: radeon_ring structure holding ring information
- *
- * Test the DMA engine by writing using it to write an
- * value to memory. (r6xx-SI).
- * Returns 0 for success, error for failure.
- */
-int r600_dma_ring_test(struct radeon_device *rdev,
-		       struct radeon_ring *ring)
-{
-	unsigned i;
-	int r;
-	void __iomem *ptr = (void *)rdev->vram_scratch.ptr;
-	u32 tmp;
-
-	if (!ptr) {
-		DRM_ERROR("invalid vram scratch pointer\n");
-		return -EINVAL;
-	}
-
-	tmp = 0xCAFEDEAD;
-	writel(tmp, ptr);
-
-	r = radeon_ring_lock(rdev, ring, 4);
-	if (r) {
-		DRM_ERROR("radeon: dma failed to lock ring %d (%d).\n", ring->idx, r);
-		return r;
-	}
-	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 1));
-	radeon_ring_write(ring, rdev->vram_scratch.gpu_addr & 0xfffffffc);
-	radeon_ring_write(ring, upper_32_bits(rdev->vram_scratch.gpu_addr) & 0xff);
-	radeon_ring_write(ring, 0xDEADBEEF);
-	radeon_ring_unlock_commit(rdev, ring);
-
-	for (i = 0; i < rdev->usec_timeout; i++) {
-		tmp = readl(ptr);
-		if (tmp == 0xDEADBEEF)
-			break;
-		DRM_UDELAY(1);
-	}
-
-	if (i < rdev->usec_timeout) {
-		DRM_INFO("ring test on %d succeeded in %d usecs\n", ring->idx, i);
-	} else {
-		DRM_ERROR("radeon: ring %d test failed (0x%08X)\n",
-			  ring->idx, tmp);
-		r = -EINVAL;
-	}
-	return r;
-}
-
 /*
  * CP fences/semaphores
  */
@@ -2839,59 +2593,6 @@ void r600_semaphore_ring_emit(struct radeon_device *rdev,
 	radeon_ring_write(ring, (upper_32_bits(addr) & 0xff) | sel);
 }
 
-/*
- * DMA fences/semaphores
- */
-
-/**
- * r600_dma_fence_ring_emit - emit a fence on the DMA ring
- *
- * @rdev: radeon_device pointer
- * @fence: radeon fence object
- *
- * Add a DMA fence packet to the ring to write
- * the fence seq number and DMA trap packet to generate
- * an interrupt if needed (r6xx-r7xx).
- */
-void r600_dma_fence_ring_emit(struct radeon_device *rdev,
-			      struct radeon_fence *fence)
-{
-	struct radeon_ring *ring = &rdev->ring[fence->ring];
-	u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
-
-	/* write the fence */
-	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_FENCE, 0, 0, 0));
-	radeon_ring_write(ring, addr & 0xfffffffc);
-	radeon_ring_write(ring, (upper_32_bits(addr) & 0xff));
-	radeon_ring_write(ring, lower_32_bits(fence->seq));
-	/* generate an interrupt */
-	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_TRAP, 0, 0, 0));
-}
-
-/**
- * r600_dma_semaphore_ring_emit - emit a semaphore on the dma ring
- *
- * @rdev: radeon_device pointer
- * @ring: radeon_ring structure holding ring information
- * @semaphore: radeon semaphore object
- * @emit_wait: wait or signal semaphore
- *
- * Add a DMA semaphore packet to the ring wait on or signal
- * other rings (r6xx-SI).
- */
-void r600_dma_semaphore_ring_emit(struct radeon_device *rdev,
-				  struct radeon_ring *ring,
-				  struct radeon_semaphore *semaphore,
-				  bool emit_wait)
-{
-	u64 addr = semaphore->gpu_addr;
-	u32 s = emit_wait ? 0 : 1;
-
-	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SEMAPHORE, 0, s, 0));
-	radeon_ring_write(ring, addr & 0xfffffffc);
-	radeon_ring_write(ring, upper_32_bits(addr) & 0xff);
-}
-
 /**
  * r600_copy_cpdma - copy pages using the CP DMA engine
  *
@@ -2976,80 +2677,6 @@ int r600_copy_cpdma(struct radeon_device *rdev,
 	return r;
 }
 
-/**
- * r600_copy_dma - copy pages using the DMA engine
- *
- * @rdev: radeon_device pointer
- * @src_offset: src GPU address
- * @dst_offset: dst GPU address
- * @num_gpu_pages: number of GPU pages to xfer
- * @fence: radeon fence object
- *
- * Copy GPU paging using the DMA engine (r6xx).
- * Used by the radeon ttm implementation to move pages if
- * registered as the asic copy callback.
- */
-int r600_copy_dma(struct radeon_device *rdev,
-		  uint64_t src_offset, uint64_t dst_offset,
-		  unsigned num_gpu_pages,
-		  struct radeon_fence **fence)
-{
-	struct radeon_semaphore *sem = NULL;
-	int ring_index = rdev->asic->copy.dma_ring_index;
-	struct radeon_ring *ring = &rdev->ring[ring_index];
-	u32 size_in_dw, cur_size_in_dw;
-	int i, num_loops;
-	int r = 0;
-
-	r = radeon_semaphore_create(rdev, &sem);
-	if (r) {
-		DRM_ERROR("radeon: moving bo (%d).\n", r);
-		return r;
-	}
-
-	size_in_dw = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT) / 4;
-	num_loops = DIV_ROUND_UP(size_in_dw, 0xFFFE);
-	r = radeon_ring_lock(rdev, ring, num_loops * 4 + 8);
-	if (r) {
-		DRM_ERROR("radeon: moving bo (%d).\n", r);
-		radeon_semaphore_free(rdev, &sem, NULL);
-		return r;
-	}
-
-	if (radeon_fence_need_sync(*fence, ring->idx)) {
-		radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring,
-					    ring->idx);
-		radeon_fence_note_sync(*fence, ring->idx);
-	} else {
-		radeon_semaphore_free(rdev, &sem, NULL);
-	}
-
-	for (i = 0; i < num_loops; i++) {
-		cur_size_in_dw = size_in_dw;
-		if (cur_size_in_dw > 0xFFFE)
-			cur_size_in_dw = 0xFFFE;
-		size_in_dw -= cur_size_in_dw;
-		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_COPY, 0, 0, cur_size_in_dw));
-		radeon_ring_write(ring, dst_offset & 0xfffffffc);
-		radeon_ring_write(ring, src_offset & 0xfffffffc);
-		radeon_ring_write(ring, (((upper_32_bits(dst_offset) & 0xff) << 16) |
-					 (upper_32_bits(src_offset) & 0xff)));
-		src_offset += cur_size_in_dw * 4;
-		dst_offset += cur_size_in_dw * 4;
-	}
-
-	r = radeon_fence_emit(rdev, fence, ring->idx);
-	if (r) {
-		radeon_ring_unlock_undo(rdev, ring);
-		return r;
-	}
-
-	radeon_ring_unlock_commit(rdev, ring);
-	radeon_semaphore_free(rdev, &sem, *fence);
-
-	return r;
-}
-
 int r600_set_surface_reg(struct radeon_device *rdev, int reg,
 			 uint32_t tiling_flags, uint32_t pitch,
 			 uint32_t offset, uint32_t obj_size)
@@ -3409,104 +3036,6 @@ free_scratch:
 	return r;
 }
 
-/**
- * r600_dma_ib_test - test an IB on the DMA engine
- *
- * @rdev: radeon_device pointer
- * @ring: radeon_ring structure holding ring information
- *
- * Test a simple IB in the DMA ring (r6xx-SI).
- * Returns 0 on success, error on failure.
- */
-int r600_dma_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
-{
-	struct radeon_ib ib;
-	unsigned i;
-	int r;
-	void __iomem *ptr = (void *)rdev->vram_scratch.ptr;
-	u32 tmp = 0;
-
-	if (!ptr) {
-		DRM_ERROR("invalid vram scratch pointer\n");
-		return -EINVAL;
-	}
-
-	tmp = 0xCAFEDEAD;
-	writel(tmp, ptr);
-
-	r = radeon_ib_get(rdev, ring->idx, &ib, NULL, 256);
-	if (r) {
-		DRM_ERROR("radeon: failed to get ib (%d).\n", r);
-		return r;
-	}
-
-	ib.ptr[0] = DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 1);
-	ib.ptr[1] = rdev->vram_scratch.gpu_addr & 0xfffffffc;
-	ib.ptr[2] = upper_32_bits(rdev->vram_scratch.gpu_addr) & 0xff;
-	ib.ptr[3] = 0xDEADBEEF;
-	ib.length_dw = 4;
-
-	r = radeon_ib_schedule(rdev, &ib, NULL);
-	if (r) {
-		radeon_ib_free(rdev, &ib);
-		DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
-		return r;
-	}
-	r = radeon_fence_wait(ib.fence, false);
-	if (r) {
-		DRM_ERROR("radeon: fence wait failed (%d).\n", r);
-		return r;
-	}
-	for (i = 0; i < rdev->usec_timeout; i++) {
-		tmp = readl(ptr);
-		if (tmp == 0xDEADBEEF)
-			break;
-		DRM_UDELAY(1);
-	}
-	if (i < rdev->usec_timeout) {
-		DRM_INFO("ib test on ring %d succeeded in %u usecs\n", ib.fence->ring, i);
-	} else {
-		DRM_ERROR("radeon: ib test failed (0x%08X)\n", tmp);
-		r = -EINVAL;
-	}
-	radeon_ib_free(rdev, &ib);
-	return r;
-}
-
-/**
- * r600_dma_ring_ib_execute - Schedule an IB on the DMA engine
- *
- * @rdev: radeon_device pointer
- * @ib: IB object to schedule
- *
- * Schedule an IB in the DMA ring (r6xx-r7xx).
- */
-void r600_dma_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
-{
-	struct radeon_ring *ring = &rdev->ring[ib->ring];
-
-	if (rdev->wb.enabled) {
-		u32 next_rptr = ring->wptr + 4;
-		while ((next_rptr & 7) != 5)
-			next_rptr++;
-		next_rptr += 3;
-		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 1));
-		radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
-		radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xff);
-		radeon_ring_write(ring, next_rptr);
-	}
-
-	/* The indirect buffer packet must end on an 8 DW boundary in the DMA ring.
-	 * Pad as necessary with NOPs.
-	 */
-	while ((ring->wptr & 7) != 5)
-		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0));
-	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_INDIRECT_BUFFER, 0, 0, 0));
-	radeon_ring_write(ring, (ib->gpu_addr & 0xFFFFFFE0));
-	radeon_ring_write(ring, (ib->length_dw << 16) | (upper_32_bits(ib->gpu_addr) & 0xFF));
-
-}
-
 /*
  * Interrupts
  *
diff --git a/drivers/gpu/drm/radeon/r600_dma.c b/drivers/gpu/drm/radeon/r600_dma.c
new file mode 100644
index 000000000000..bff05576266b
--- /dev/null
+++ b/drivers/gpu/drm/radeon/r600_dma.c
@@ -0,0 +1,497 @@
+/*
+ * Copyright 2013 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Alex Deucher
+ */
+#include <drm/drmP.h>
+#include "radeon.h"
+#include "radeon_asic.h"
+#include "r600d.h"
+
+u32 r600_gpu_check_soft_reset(struct radeon_device *rdev);
+
+/*
+ * DMA
+ * Starting with R600, the GPU has an asynchronous
+ * DMA engine.  The programming model is very similar
+ * to the 3D engine (ring buffer, IBs, etc.), but the
+ * DMA controller has it's own packet format that is
+ * different form the PM4 format used by the 3D engine.
+ * It supports copying data, writing embedded data,
+ * solid fills, and a number of other things.  It also
+ * has support for tiling/detiling of buffers.
+ */
+
+/**
+ * r600_dma_get_rptr - get the current read pointer
+ *
+ * @rdev: radeon_device pointer
+ * @ring: radeon ring pointer
+ *
+ * Get the current rptr from the hardware (r6xx+).
+ */
+uint32_t r600_dma_get_rptr(struct radeon_device *rdev,
+			   struct radeon_ring *ring)
+{
+	return (radeon_ring_generic_get_rptr(rdev, ring) & 0x3fffc) >> 2;
+}
+
+/**
+ * r600_dma_get_wptr - get the current write pointer
+ *
+ * @rdev: radeon_device pointer
+ * @ring: radeon ring pointer
+ *
+ * Get the current wptr from the hardware (r6xx+).
+ */
+uint32_t r600_dma_get_wptr(struct radeon_device *rdev,
+			   struct radeon_ring *ring)
+{
+	return (RREG32(ring->wptr_reg) & 0x3fffc) >> 2;
+}
+
+/**
+ * r600_dma_set_wptr - commit the write pointer
+ *
+ * @rdev: radeon_device pointer
+ * @ring: radeon ring pointer
+ *
+ * Write the wptr back to the hardware (r6xx+).
+ */
+void r600_dma_set_wptr(struct radeon_device *rdev,
+		       struct radeon_ring *ring)
+{
+	WREG32(ring->wptr_reg, (ring->wptr << 2) & 0x3fffc);
+}
+
+/**
+ * r600_dma_stop - stop the async dma engine
+ *
+ * @rdev: radeon_device pointer
+ *
+ * Stop the async dma engine (r6xx-evergreen).
+ */
+void r600_dma_stop(struct radeon_device *rdev)
+{
+	u32 rb_cntl = RREG32(DMA_RB_CNTL);
+
+	radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
+
+	rb_cntl &= ~DMA_RB_ENABLE;
+	WREG32(DMA_RB_CNTL, rb_cntl);
+
+	rdev->ring[R600_RING_TYPE_DMA_INDEX].ready = false;
+}
+
+/**
+ * r600_dma_resume - setup and start the async dma engine
+ *
+ * @rdev: radeon_device pointer
+ *
+ * Set up the DMA ring buffer and enable it. (r6xx-evergreen).
+ * Returns 0 for success, error for failure.
+ */
+int r600_dma_resume(struct radeon_device *rdev)
+{
+	struct radeon_ring *ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
+	u32 rb_cntl, dma_cntl, ib_cntl;
+	u32 rb_bufsz;
+	int r;
+
+	/* Reset dma */
+	if (rdev->family >= CHIP_RV770)
+		WREG32(SRBM_SOFT_RESET, RV770_SOFT_RESET_DMA);
+	else
+		WREG32(SRBM_SOFT_RESET, SOFT_RESET_DMA);
+	RREG32(SRBM_SOFT_RESET);
+	udelay(50);
+	WREG32(SRBM_SOFT_RESET, 0);
+
+	WREG32(DMA_SEM_INCOMPLETE_TIMER_CNTL, 0);
+	WREG32(DMA_SEM_WAIT_FAIL_TIMER_CNTL, 0);
+
+	/* Set ring buffer size in dwords */
+	rb_bufsz = drm_order(ring->ring_size / 4);
+	rb_cntl = rb_bufsz << 1;
+#ifdef __BIG_ENDIAN
+	rb_cntl |= DMA_RB_SWAP_ENABLE | DMA_RPTR_WRITEBACK_SWAP_ENABLE;
+#endif
+	WREG32(DMA_RB_CNTL, rb_cntl);
+
+	/* Initialize the ring buffer's read and write pointers */
+	WREG32(DMA_RB_RPTR, 0);
+	WREG32(DMA_RB_WPTR, 0);
+
+	/* set the wb address whether it's enabled or not */
+	WREG32(DMA_RB_RPTR_ADDR_HI,
+	       upper_32_bits(rdev->wb.gpu_addr + R600_WB_DMA_RPTR_OFFSET) & 0xFF);
+	WREG32(DMA_RB_RPTR_ADDR_LO,
+	       ((rdev->wb.gpu_addr + R600_WB_DMA_RPTR_OFFSET) & 0xFFFFFFFC));
+
+	if (rdev->wb.enabled)
+		rb_cntl |= DMA_RPTR_WRITEBACK_ENABLE;
+
+	WREG32(DMA_RB_BASE, ring->gpu_addr >> 8);
+
+	/* enable DMA IBs */
+	ib_cntl = DMA_IB_ENABLE;
+#ifdef __BIG_ENDIAN
+	ib_cntl |= DMA_IB_SWAP_ENABLE;
+#endif
+	WREG32(DMA_IB_CNTL, ib_cntl);
+
+	dma_cntl = RREG32(DMA_CNTL);
+	dma_cntl &= ~CTXEMPTY_INT_ENABLE;
+	WREG32(DMA_CNTL, dma_cntl);
+
+	if (rdev->family >= CHIP_RV770)
+		WREG32(DMA_MODE, 1);
+
+	ring->wptr = 0;
+	WREG32(DMA_RB_WPTR, ring->wptr << 2);
+
+	ring->rptr = RREG32(DMA_RB_RPTR) >> 2;
+
+	WREG32(DMA_RB_CNTL, rb_cntl | DMA_RB_ENABLE);
+
+	ring->ready = true;
+
+	r = radeon_ring_test(rdev, R600_RING_TYPE_DMA_INDEX, ring);
+	if (r) {
+		ring->ready = false;
+		return r;
+	}
+
+	radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
+
+	return 0;
+}
+
+/**
+ * r600_dma_fini - tear down the async dma engine
+ *
+ * @rdev: radeon_device pointer
+ *
+ * Stop the async dma engine and free the ring (r6xx-evergreen).
+ */
+void r600_dma_fini(struct radeon_device *rdev)
+{
+	r600_dma_stop(rdev);
+	radeon_ring_fini(rdev, &rdev->ring[R600_RING_TYPE_DMA_INDEX]);
+}
+
+/**
+ * r600_dma_is_lockup - Check if the DMA engine is locked up
+ *
+ * @rdev: radeon_device pointer
+ * @ring: radeon_ring structure holding ring information
+ *
+ * Check if the async DMA engine is locked up.
+ * Returns true if the engine appears to be locked up, false if not.
+ */
+bool r600_dma_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
+{
+	u32 reset_mask = r600_gpu_check_soft_reset(rdev);
+
+	if (!(reset_mask & RADEON_RESET_DMA)) {
+		radeon_ring_lockup_update(ring);
+		return false;
+	}
+	/* force ring activities */
+	radeon_ring_force_activity(rdev, ring);
+	return radeon_ring_test_lockup(rdev, ring);
+}
+
+
+/**
+ * r600_dma_ring_test - simple async dma engine test
+ *
+ * @rdev: radeon_device pointer
+ * @ring: radeon_ring structure holding ring information
+ *
+ * Test the DMA engine by writing using it to write an
+ * value to memory. (r6xx-SI).
+ * Returns 0 for success, error for failure.
+ */
+int r600_dma_ring_test(struct radeon_device *rdev,
+		       struct radeon_ring *ring)
+{
+	unsigned i;
+	int r;
+	void __iomem *ptr = (void *)rdev->vram_scratch.ptr;
+	u32 tmp;
+
+	if (!ptr) {
+		DRM_ERROR("invalid vram scratch pointer\n");
+		return -EINVAL;
+	}
+
+	tmp = 0xCAFEDEAD;
+	writel(tmp, ptr);
+
+	r = radeon_ring_lock(rdev, ring, 4);
+	if (r) {
+		DRM_ERROR("radeon: dma failed to lock ring %d (%d).\n", ring->idx, r);
+		return r;
+	}
+	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 1));
+	radeon_ring_write(ring, rdev->vram_scratch.gpu_addr & 0xfffffffc);
+	radeon_ring_write(ring, upper_32_bits(rdev->vram_scratch.gpu_addr) & 0xff);
+	radeon_ring_write(ring, 0xDEADBEEF);
+	radeon_ring_unlock_commit(rdev, ring);
+
+	for (i = 0; i < rdev->usec_timeout; i++) {
+		tmp = readl(ptr);
+		if (tmp == 0xDEADBEEF)
+			break;
+		DRM_UDELAY(1);
+	}
+
+	if (i < rdev->usec_timeout) {
+		DRM_INFO("ring test on %d succeeded in %d usecs\n", ring->idx, i);
+	} else {
+		DRM_ERROR("radeon: ring %d test failed (0x%08X)\n",
+			  ring->idx, tmp);
+		r = -EINVAL;
+	}
+	return r;
+}
+
+/**
+ * r600_dma_fence_ring_emit - emit a fence on the DMA ring
+ *
+ * @rdev: radeon_device pointer
+ * @fence: radeon fence object
+ *
+ * Add a DMA fence packet to the ring to write
+ * the fence seq number and DMA trap packet to generate
+ * an interrupt if needed (r6xx-r7xx).
+ */
+void r600_dma_fence_ring_emit(struct radeon_device *rdev,
+			      struct radeon_fence *fence)
+{
+	struct radeon_ring *ring = &rdev->ring[fence->ring];
+	u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
+
+	/* write the fence */
+	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_FENCE, 0, 0, 0));
+	radeon_ring_write(ring, addr & 0xfffffffc);
+	radeon_ring_write(ring, (upper_32_bits(addr) & 0xff));
+	radeon_ring_write(ring, lower_32_bits(fence->seq));
+	/* generate an interrupt */
+	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_TRAP, 0, 0, 0));
+}
+
+/**
+ * r600_dma_semaphore_ring_emit - emit a semaphore on the dma ring
+ *
+ * @rdev: radeon_device pointer
+ * @ring: radeon_ring structure holding ring information
+ * @semaphore: radeon semaphore object
+ * @emit_wait: wait or signal semaphore
+ *
+ * Add a DMA semaphore packet to the ring wait on or signal
+ * other rings (r6xx-SI).
+ */
+void r600_dma_semaphore_ring_emit(struct radeon_device *rdev,
+				  struct radeon_ring *ring,
+				  struct radeon_semaphore *semaphore,
+				  bool emit_wait)
+{
+	u64 addr = semaphore->gpu_addr;
+	u32 s = emit_wait ? 0 : 1;
+
+	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SEMAPHORE, 0, s, 0));
+	radeon_ring_write(ring, addr & 0xfffffffc);
+	radeon_ring_write(ring, upper_32_bits(addr) & 0xff);
+}
+
+/**
+ * r600_dma_ib_test - test an IB on the DMA engine
+ *
+ * @rdev: radeon_device pointer
+ * @ring: radeon_ring structure holding ring information
+ *
+ * Test a simple IB in the DMA ring (r6xx-SI).
+ * Returns 0 on success, error on failure.
+ */
+int r600_dma_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
+{
+	struct radeon_ib ib;
+	unsigned i;
+	int r;
+	void __iomem *ptr = (void *)rdev->vram_scratch.ptr;
+	u32 tmp = 0;
+
+	if (!ptr) {
+		DRM_ERROR("invalid vram scratch pointer\n");
+		return -EINVAL;
+	}
+
+	tmp = 0xCAFEDEAD;
+	writel(tmp, ptr);
+
+	r = radeon_ib_get(rdev, ring->idx, &ib, NULL, 256);
+	if (r) {
+		DRM_ERROR("radeon: failed to get ib (%d).\n", r);
+		return r;
+	}
+
+	ib.ptr[0] = DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 1);
+	ib.ptr[1] = rdev->vram_scratch.gpu_addr & 0xfffffffc;
+	ib.ptr[2] = upper_32_bits(rdev->vram_scratch.gpu_addr) & 0xff;
+	ib.ptr[3] = 0xDEADBEEF;
+	ib.length_dw = 4;
+
+	r = radeon_ib_schedule(rdev, &ib, NULL);
+	if (r) {
+		radeon_ib_free(rdev, &ib);
+		DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
+		return r;
+	}
+	r = radeon_fence_wait(ib.fence, false);
+	if (r) {
+		DRM_ERROR("radeon: fence wait failed (%d).\n", r);
+		return r;
+	}
+	for (i = 0; i < rdev->usec_timeout; i++) {
+		tmp = readl(ptr);
+		if (tmp == 0xDEADBEEF)
+			break;
+		DRM_UDELAY(1);
+	}
+	if (i < rdev->usec_timeout) {
+		DRM_INFO("ib test on ring %d succeeded in %u usecs\n", ib.fence->ring, i);
+	} else {
+		DRM_ERROR("radeon: ib test failed (0x%08X)\n", tmp);
+		r = -EINVAL;
+	}
+	radeon_ib_free(rdev, &ib);
+	return r;
+}
+
+/**
+ * r600_dma_ring_ib_execute - Schedule an IB on the DMA engine
+ *
+ * @rdev: radeon_device pointer
+ * @ib: IB object to schedule
+ *
+ * Schedule an IB in the DMA ring (r6xx-r7xx).
+ */
+void r600_dma_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
+{
+	struct radeon_ring *ring = &rdev->ring[ib->ring];
+
+	if (rdev->wb.enabled) {
+		u32 next_rptr = ring->wptr + 4;
+		while ((next_rptr & 7) != 5)
+			next_rptr++;
+		next_rptr += 3;
+		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 1));
+		radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
+		radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xff);
+		radeon_ring_write(ring, next_rptr);
+	}
+
+	/* The indirect buffer packet must end on an 8 DW boundary in the DMA ring.
+	 * Pad as necessary with NOPs.
+	 */
+	while ((ring->wptr & 7) != 5)
+		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0));
+	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_INDIRECT_BUFFER, 0, 0, 0));
+	radeon_ring_write(ring, (ib->gpu_addr & 0xFFFFFFE0));
+	radeon_ring_write(ring, (ib->length_dw << 16) | (upper_32_bits(ib->gpu_addr) & 0xFF));
+
+}
+
+/**
+ * r600_copy_dma - copy pages using the DMA engine
+ *
+ * @rdev: radeon_device pointer
+ * @src_offset: src GPU address
+ * @dst_offset: dst GPU address
+ * @num_gpu_pages: number of GPU pages to xfer
+ * @fence: radeon fence object
+ *
+ * Copy GPU paging using the DMA engine (r6xx).
+ * Used by the radeon ttm implementation to move pages if
+ * registered as the asic copy callback.
+ */
+int r600_copy_dma(struct radeon_device *rdev,
+		  uint64_t src_offset, uint64_t dst_offset,
+		  unsigned num_gpu_pages,
+		  struct radeon_fence **fence)
+{
+	struct radeon_semaphore *sem = NULL;
+	int ring_index = rdev->asic->copy.dma_ring_index;
+	struct radeon_ring *ring = &rdev->ring[ring_index];
+	u32 size_in_dw, cur_size_in_dw;
+	int i, num_loops;
+	int r = 0;
+
+	r = radeon_semaphore_create(rdev, &sem);
+	if (r) {
+		DRM_ERROR("radeon: moving bo (%d).\n", r);
+		return r;
+	}
+
+	size_in_dw = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT) / 4;
+	num_loops = DIV_ROUND_UP(size_in_dw, 0xFFFE);
+	r = radeon_ring_lock(rdev, ring, num_loops * 4 + 8);
+	if (r) {
+		DRM_ERROR("radeon: moving bo (%d).\n", r);
+		radeon_semaphore_free(rdev, &sem, NULL);
+		return r;
+	}
+
+	if (radeon_fence_need_sync(*fence, ring->idx)) {
+		radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring,
+					    ring->idx);
+		radeon_fence_note_sync(*fence, ring->idx);
+	} else {
+		radeon_semaphore_free(rdev, &sem, NULL);
+	}
+
+	for (i = 0; i < num_loops; i++) {
+		cur_size_in_dw = size_in_dw;
+		if (cur_size_in_dw > 0xFFFE)
+			cur_size_in_dw = 0xFFFE;
+		size_in_dw -= cur_size_in_dw;
+		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_COPY, 0, 0, cur_size_in_dw));
+		radeon_ring_write(ring, dst_offset & 0xfffffffc);
+		radeon_ring_write(ring, src_offset & 0xfffffffc);
+		radeon_ring_write(ring, (((upper_32_bits(dst_offset) & 0xff) << 16) |
+					 (upper_32_bits(src_offset) & 0xff)));
+		src_offset += cur_size_in_dw * 4;
+		dst_offset += cur_size_in_dw * 4;
+	}
+
+	r = radeon_fence_emit(rdev, fence, ring->idx);
+	if (r) {
+		radeon_ring_unlock_undo(rdev, ring);
+		return r;
+	}
+
+	radeon_ring_unlock_commit(rdev, ring);
+	radeon_semaphore_free(rdev, &sem, *fence);
+
+	return r;
+}
diff --git a/drivers/gpu/drm/radeon/rv770.c b/drivers/gpu/drm/radeon/rv770.c
index aaab7b1bba27..b811296462a3 100644
--- a/drivers/gpu/drm/radeon/rv770.c
+++ b/drivers/gpu/drm/radeon/rv770.c
@@ -1650,80 +1650,6 @@ static int rv770_mc_init(struct radeon_device *rdev)
 	return 0;
 }
 
-/**
- * rv770_copy_dma - copy pages using the DMA engine
- *
- * @rdev: radeon_device pointer
- * @src_offset: src GPU address
- * @dst_offset: dst GPU address
- * @num_gpu_pages: number of GPU pages to xfer
- * @fence: radeon fence object
- *
- * Copy GPU paging using the DMA engine (r7xx).
- * Used by the radeon ttm implementation to move pages if
- * registered as the asic copy callback.
- */
-int rv770_copy_dma(struct radeon_device *rdev,
-		  uint64_t src_offset, uint64_t dst_offset,
-		  unsigned num_gpu_pages,
-		  struct radeon_fence **fence)
-{
-	struct radeon_semaphore *sem = NULL;
-	int ring_index = rdev->asic->copy.dma_ring_index;
-	struct radeon_ring *ring = &rdev->ring[ring_index];
-	u32 size_in_dw, cur_size_in_dw;
-	int i, num_loops;
-	int r = 0;
-
-	r = radeon_semaphore_create(rdev, &sem);
-	if (r) {
-		DRM_ERROR("radeon: moving bo (%d).\n", r);
-		return r;
-	}
-
-	size_in_dw = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT) / 4;
-	num_loops = DIV_ROUND_UP(size_in_dw, 0xFFFF);
-	r = radeon_ring_lock(rdev, ring, num_loops * 5 + 8);
-	if (r) {
-		DRM_ERROR("radeon: moving bo (%d).\n", r);
-		radeon_semaphore_free(rdev, &sem, NULL);
-		return r;
-	}
-
-	if (radeon_fence_need_sync(*fence, ring->idx)) {
-		radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring,
-					    ring->idx);
-		radeon_fence_note_sync(*fence, ring->idx);
-	} else {
-		radeon_semaphore_free(rdev, &sem, NULL);
-	}
-
-	for (i = 0; i < num_loops; i++) {
-		cur_size_in_dw = size_in_dw;
-		if (cur_size_in_dw > 0xFFFF)
-			cur_size_in_dw = 0xFFFF;
-		size_in_dw -= cur_size_in_dw;
-		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_COPY, 0, 0, cur_size_in_dw));
-		radeon_ring_write(ring, dst_offset & 0xfffffffc);
-		radeon_ring_write(ring, src_offset & 0xfffffffc);
-		radeon_ring_write(ring, upper_32_bits(dst_offset) & 0xff);
-		radeon_ring_write(ring, upper_32_bits(src_offset) & 0xff);
-		src_offset += cur_size_in_dw * 4;
-		dst_offset += cur_size_in_dw * 4;
-	}
-
-	r = radeon_fence_emit(rdev, fence, ring->idx);
-	if (r) {
-		radeon_ring_unlock_undo(rdev, ring);
-		return r;
-	}
-
-	radeon_ring_unlock_commit(rdev, ring);
-	radeon_semaphore_free(rdev, &sem, *fence);
-
-	return r;
-}
-
 static int rv770_startup(struct radeon_device *rdev)
 {
 	struct radeon_ring *ring;
diff --git a/drivers/gpu/drm/radeon/rv770_dma.c b/drivers/gpu/drm/radeon/rv770_dma.c
new file mode 100644
index 000000000000..f9b02e3d6830
--- /dev/null
+++ b/drivers/gpu/drm/radeon/rv770_dma.c
@@ -0,0 +1,101 @@
+/*
+ * Copyright 2013 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Alex Deucher
+ */
+#include <drm/drmP.h>
+#include "radeon.h"
+#include "radeon_asic.h"
+#include "rv770d.h"
+
+/**
+ * rv770_copy_dma - copy pages using the DMA engine
+ *
+ * @rdev: radeon_device pointer
+ * @src_offset: src GPU address
+ * @dst_offset: dst GPU address
+ * @num_gpu_pages: number of GPU pages to xfer
+ * @fence: radeon fence object
+ *
+ * Copy GPU paging using the DMA engine (r7xx).
+ * Used by the radeon ttm implementation to move pages if
+ * registered as the asic copy callback.
+ */
+int rv770_copy_dma(struct radeon_device *rdev,
+		  uint64_t src_offset, uint64_t dst_offset,
+		  unsigned num_gpu_pages,
+		  struct radeon_fence **fence)
+{
+	struct radeon_semaphore *sem = NULL;
+	int ring_index = rdev->asic->copy.dma_ring_index;
+	struct radeon_ring *ring = &rdev->ring[ring_index];
+	u32 size_in_dw, cur_size_in_dw;
+	int i, num_loops;
+	int r = 0;
+
+	r = radeon_semaphore_create(rdev, &sem);
+	if (r) {
+		DRM_ERROR("radeon: moving bo (%d).\n", r);
+		return r;
+	}
+
+	size_in_dw = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT) / 4;
+	num_loops = DIV_ROUND_UP(size_in_dw, 0xFFFF);
+	r = radeon_ring_lock(rdev, ring, num_loops * 5 + 8);
+	if (r) {
+		DRM_ERROR("radeon: moving bo (%d).\n", r);
+		radeon_semaphore_free(rdev, &sem, NULL);
+		return r;
+	}
+
+	if (radeon_fence_need_sync(*fence, ring->idx)) {
+		radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring,
+					    ring->idx);
+		radeon_fence_note_sync(*fence, ring->idx);
+	} else {
+		radeon_semaphore_free(rdev, &sem, NULL);
+	}
+
+	for (i = 0; i < num_loops; i++) {
+		cur_size_in_dw = size_in_dw;
+		if (cur_size_in_dw > 0xFFFF)
+			cur_size_in_dw = 0xFFFF;
+		size_in_dw -= cur_size_in_dw;
+		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_COPY, 0, 0, cur_size_in_dw));
+		radeon_ring_write(ring, dst_offset & 0xfffffffc);
+		radeon_ring_write(ring, src_offset & 0xfffffffc);
+		radeon_ring_write(ring, upper_32_bits(dst_offset) & 0xff);
+		radeon_ring_write(ring, upper_32_bits(src_offset) & 0xff);
+		src_offset += cur_size_in_dw * 4;
+		dst_offset += cur_size_in_dw * 4;
+	}
+
+	r = radeon_fence_emit(rdev, fence, ring->idx);
+	if (r) {
+		radeon_ring_unlock_undo(rdev, ring);
+		return r;
+	}
+
+	radeon_ring_unlock_commit(rdev, ring);
+	radeon_semaphore_free(rdev, &sem, *fence);
+
+	return r;
+}
diff --git a/drivers/gpu/drm/radeon/si.c b/drivers/gpu/drm/radeon/si.c
index f3f79089405e..f5307e6bb92b 100644
--- a/drivers/gpu/drm/radeon/si.c
+++ b/drivers/gpu/drm/radeon/si.c
@@ -78,6 +78,11 @@ extern void evergreen_mc_resume(struct radeon_device *rdev, struct evergreen_mc_
 extern u32 evergreen_get_number_of_dram_channels(struct radeon_device *rdev);
 extern void evergreen_print_gpu_status_regs(struct radeon_device *rdev);
 extern bool evergreen_is_display_hung(struct radeon_device *rdev);
+extern void si_dma_vm_set_page(struct radeon_device *rdev,
+			       struct radeon_ib *ib,
+			       uint64_t pe,
+			       uint64_t addr, unsigned count,
+			       uint32_t incr, uint32_t flags);
 
 static const u32 verde_rlc_save_restore_register_list[] =
 {
@@ -3495,7 +3500,7 @@ static int si_cp_resume(struct radeon_device *rdev)
 	return 0;
 }
 
-static u32 si_gpu_check_soft_reset(struct radeon_device *rdev)
+u32 si_gpu_check_soft_reset(struct radeon_device *rdev)
 {
 	u32 reset_mask = 0;
 	u32 tmp;
@@ -3744,34 +3749,6 @@ bool si_gfx_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
 	return radeon_ring_test_lockup(rdev, ring);
 }
 
-/**
- * si_dma_is_lockup - Check if the DMA engine is locked up
- *
- * @rdev: radeon_device pointer
- * @ring: radeon_ring structure holding ring information
- *
- * Check if the async DMA engine is locked up.
- * Returns true if the engine appears to be locked up, false if not.
- */
-bool si_dma_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
-{
-	u32 reset_mask = si_gpu_check_soft_reset(rdev);
-	u32 mask;
-
-	if (ring->idx == R600_RING_TYPE_DMA_INDEX)
-		mask = RADEON_RESET_DMA;
-	else
-		mask = RADEON_RESET_DMA1;
-
-	if (!(reset_mask & mask)) {
-		radeon_ring_lockup_update(ring);
-		return false;
-	}
-	/* force ring activities */
-	radeon_ring_force_activity(rdev, ring);
-	return radeon_ring_test_lockup(rdev, ring);
-}
-
 /* MC */
 static void si_mc_program(struct radeon_device *rdev)
 {
@@ -4710,58 +4687,7 @@ void si_vm_set_page(struct radeon_device *rdev,
 		}
 	} else {
 		/* DMA */
-		if (flags & RADEON_VM_PAGE_SYSTEM) {
-			while (count) {
-				ndw = count * 2;
-				if (ndw > 0xFFFFE)
-					ndw = 0xFFFFE;
-
-				/* for non-physically contiguous pages (system) */
-				ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 0, ndw);
-				ib->ptr[ib->length_dw++] = pe;
-				ib->ptr[ib->length_dw++] = upper_32_bits(pe) & 0xff;
-				for (; ndw > 0; ndw -= 2, --count, pe += 8) {
-					if (flags & RADEON_VM_PAGE_SYSTEM) {
-						value = radeon_vm_map_gart(rdev, addr);
-						value &= 0xFFFFFFFFFFFFF000ULL;
-					} else if (flags & RADEON_VM_PAGE_VALID) {
-						value = addr;
-					} else {
-						value = 0;
-					}
-					addr += incr;
-					value |= r600_flags;
-					ib->ptr[ib->length_dw++] = value;
-					ib->ptr[ib->length_dw++] = upper_32_bits(value);
-				}
-			}
-		} else {
-			while (count) {
-				ndw = count * 2;
-				if (ndw > 0xFFFFE)
-					ndw = 0xFFFFE;
-
-				if (flags & RADEON_VM_PAGE_VALID)
-					value = addr;
-				else
-					value = 0;
-				/* for physically contiguous pages (vram) */
-				ib->ptr[ib->length_dw++] = DMA_PTE_PDE_PACKET(ndw);
-				ib->ptr[ib->length_dw++] = pe; /* dst addr */
-				ib->ptr[ib->length_dw++] = upper_32_bits(pe) & 0xff;
-				ib->ptr[ib->length_dw++] = r600_flags; /* mask */
-				ib->ptr[ib->length_dw++] = 0;
-				ib->ptr[ib->length_dw++] = value; /* value */
-				ib->ptr[ib->length_dw++] = upper_32_bits(value);
-				ib->ptr[ib->length_dw++] = incr; /* increment size */
-				ib->ptr[ib->length_dw++] = 0;
-				pe += ndw * 4;
-				addr += (ndw / 2) * incr;
-				count -= ndw / 2;
-			}
-		}
-		while (ib->length_dw & 0x7)
-			ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0, 0);
+		si_dma_vm_set_page(rdev, ib, pe, addr, count, incr, flags);
 	}
 }
 
@@ -4808,32 +4734,6 @@ void si_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm)
 	radeon_ring_write(ring, 0x0);
 }
 
-void si_dma_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm)
-{
-	struct radeon_ring *ring = &rdev->ring[ridx];
-
-	if (vm == NULL)
-		return;
-
-	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0));
-	if (vm->id < 8) {
-		radeon_ring_write(ring, (0xf << 16) | ((VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2)) >> 2));
-	} else {
-		radeon_ring_write(ring, (0xf << 16) | ((VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((vm->id - 8) << 2)) >> 2));
-	}
-	radeon_ring_write(ring, vm->pd_gpu_addr >> 12);
-
-	/* flush hdp cache */
-	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0));
-	radeon_ring_write(ring, (0xf << 16) | (HDP_MEM_COHERENCY_FLUSH_CNTL >> 2));
-	radeon_ring_write(ring, 1);
-
-	/* bits 0-7 are the VM contexts0-7 */
-	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0));
-	radeon_ring_write(ring, (0xf << 16) | (VM_INVALIDATE_REQUEST >> 2));
-	radeon_ring_write(ring, 1 << vm->id);
-}
-
 /*
  *  Power and clock gating
  */
@@ -6177,80 +6077,6 @@ restart_ih:
 	return IRQ_HANDLED;
 }
 
-/**
- * si_copy_dma - copy pages using the DMA engine
- *
- * @rdev: radeon_device pointer
- * @src_offset: src GPU address
- * @dst_offset: dst GPU address
- * @num_gpu_pages: number of GPU pages to xfer
- * @fence: radeon fence object
- *
- * Copy GPU paging using the DMA engine (SI).
- * Used by the radeon ttm implementation to move pages if
- * registered as the asic copy callback.
- */
-int si_copy_dma(struct radeon_device *rdev,
-		uint64_t src_offset, uint64_t dst_offset,
-		unsigned num_gpu_pages,
-		struct radeon_fence **fence)
-{
-	struct radeon_semaphore *sem = NULL;
-	int ring_index = rdev->asic->copy.dma_ring_index;
-	struct radeon_ring *ring = &rdev->ring[ring_index];
-	u32 size_in_bytes, cur_size_in_bytes;
-	int i, num_loops;
-	int r = 0;
-
-	r = radeon_semaphore_create(rdev, &sem);
-	if (r) {
-		DRM_ERROR("radeon: moving bo (%d).\n", r);
-		return r;
-	}
-
-	size_in_bytes = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT);
-	num_loops = DIV_ROUND_UP(size_in_bytes, 0xfffff);
-	r = radeon_ring_lock(rdev, ring, num_loops * 5 + 11);
-	if (r) {
-		DRM_ERROR("radeon: moving bo (%d).\n", r);
-		radeon_semaphore_free(rdev, &sem, NULL);
-		return r;
-	}
-
-	if (radeon_fence_need_sync(*fence, ring->idx)) {
-		radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring,
-					    ring->idx);
-		radeon_fence_note_sync(*fence, ring->idx);
-	} else {
-		radeon_semaphore_free(rdev, &sem, NULL);
-	}
-
-	for (i = 0; i < num_loops; i++) {
-		cur_size_in_bytes = size_in_bytes;
-		if (cur_size_in_bytes > 0xFFFFF)
-			cur_size_in_bytes = 0xFFFFF;
-		size_in_bytes -= cur_size_in_bytes;
-		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_COPY, 1, 0, 0, cur_size_in_bytes));
-		radeon_ring_write(ring, dst_offset & 0xffffffff);
-		radeon_ring_write(ring, src_offset & 0xffffffff);
-		radeon_ring_write(ring, upper_32_bits(dst_offset) & 0xff);
-		radeon_ring_write(ring, upper_32_bits(src_offset) & 0xff);
-		src_offset += cur_size_in_bytes;
-		dst_offset += cur_size_in_bytes;
-	}
-
-	r = radeon_fence_emit(rdev, fence, ring->idx);
-	if (r) {
-		radeon_ring_unlock_undo(rdev, ring);
-		return r;
-	}
-
-	radeon_ring_unlock_commit(rdev, ring);
-	radeon_semaphore_free(rdev, &sem, *fence);
-
-	return r;
-}
-
 /*
  * startup/shutdown callbacks
  */
diff --git a/drivers/gpu/drm/radeon/si_dma.c b/drivers/gpu/drm/radeon/si_dma.c
new file mode 100644
index 000000000000..49909d23dfce
--- /dev/null
+++ b/drivers/gpu/drm/radeon/si_dma.c
@@ -0,0 +1,235 @@
+/*
+ * Copyright 2013 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Alex Deucher
+ */
+#include <drm/drmP.h>
+#include "radeon.h"
+#include "radeon_asic.h"
+#include "sid.h"
+
+u32 si_gpu_check_soft_reset(struct radeon_device *rdev);
+
+/**
+ * si_dma_is_lockup - Check if the DMA engine is locked up
+ *
+ * @rdev: radeon_device pointer
+ * @ring: radeon_ring structure holding ring information
+ *
+ * Check if the async DMA engine is locked up.
+ * Returns true if the engine appears to be locked up, false if not.
+ */
+bool si_dma_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
+{
+	u32 reset_mask = si_gpu_check_soft_reset(rdev);
+	u32 mask;
+
+	if (ring->idx == R600_RING_TYPE_DMA_INDEX)
+		mask = RADEON_RESET_DMA;
+	else
+		mask = RADEON_RESET_DMA1;
+
+	if (!(reset_mask & mask)) {
+		radeon_ring_lockup_update(ring);
+		return false;
+	}
+	/* force ring activities */
+	radeon_ring_force_activity(rdev, ring);
+	return radeon_ring_test_lockup(rdev, ring);
+}
+
+/**
+ * si_dma_vm_set_page - update the page tables using the DMA
+ *
+ * @rdev: radeon_device pointer
+ * @ib: indirect buffer to fill with commands
+ * @pe: addr of the page entry
+ * @addr: dst addr to write into pe
+ * @count: number of page entries to update
+ * @incr: increase next addr by incr bytes
+ * @flags: access flags
+ *
+ * Update the page tables using the DMA (SI).
+ */
+void si_dma_vm_set_page(struct radeon_device *rdev,
+			struct radeon_ib *ib,
+			uint64_t pe,
+			uint64_t addr, unsigned count,
+			uint32_t incr, uint32_t flags)
+{
+	uint32_t r600_flags = cayman_vm_page_flags(rdev, flags);
+	uint64_t value;
+	unsigned ndw;
+
+	if (flags & RADEON_VM_PAGE_SYSTEM) {
+		while (count) {
+			ndw = count * 2;
+			if (ndw > 0xFFFFE)
+				ndw = 0xFFFFE;
+
+			/* for non-physically contiguous pages (system) */
+			ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 0, ndw);
+			ib->ptr[ib->length_dw++] = pe;
+			ib->ptr[ib->length_dw++] = upper_32_bits(pe) & 0xff;
+			for (; ndw > 0; ndw -= 2, --count, pe += 8) {
+				if (flags & RADEON_VM_PAGE_SYSTEM) {
+					value = radeon_vm_map_gart(rdev, addr);
+					value &= 0xFFFFFFFFFFFFF000ULL;
+				} else if (flags & RADEON_VM_PAGE_VALID) {
+					value = addr;
+				} else {
+					value = 0;
+				}
+				addr += incr;
+				value |= r600_flags;
+				ib->ptr[ib->length_dw++] = value;
+				ib->ptr[ib->length_dw++] = upper_32_bits(value);
+			}
+		}
+	} else {
+		while (count) {
+			ndw = count * 2;
+			if (ndw > 0xFFFFE)
+				ndw = 0xFFFFE;
+
+			if (flags & RADEON_VM_PAGE_VALID)
+				value = addr;
+			else
+				value = 0;
+			/* for physically contiguous pages (vram) */
+			ib->ptr[ib->length_dw++] = DMA_PTE_PDE_PACKET(ndw);
+			ib->ptr[ib->length_dw++] = pe; /* dst addr */
+			ib->ptr[ib->length_dw++] = upper_32_bits(pe) & 0xff;
+			ib->ptr[ib->length_dw++] = r600_flags; /* mask */
+			ib->ptr[ib->length_dw++] = 0;
+			ib->ptr[ib->length_dw++] = value; /* value */
+			ib->ptr[ib->length_dw++] = upper_32_bits(value);
+			ib->ptr[ib->length_dw++] = incr; /* increment size */
+			ib->ptr[ib->length_dw++] = 0;
+			pe += ndw * 4;
+			addr += (ndw / 2) * incr;
+			count -= ndw / 2;
+		}
+	}
+	while (ib->length_dw & 0x7)
+		ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0, 0);
+}
+
+void si_dma_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm)
+{
+	struct radeon_ring *ring = &rdev->ring[ridx];
+
+	if (vm == NULL)
+		return;
+
+	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0));
+	if (vm->id < 8) {
+		radeon_ring_write(ring, (0xf << 16) | ((VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2)) >> 2));
+	} else {
+		radeon_ring_write(ring, (0xf << 16) | ((VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((vm->id - 8) << 2)) >> 2));
+	}
+	radeon_ring_write(ring, vm->pd_gpu_addr >> 12);
+
+	/* flush hdp cache */
+	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0));
+	radeon_ring_write(ring, (0xf << 16) | (HDP_MEM_COHERENCY_FLUSH_CNTL >> 2));
+	radeon_ring_write(ring, 1);
+
+	/* bits 0-7 are the VM contexts0-7 */
+	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0));
+	radeon_ring_write(ring, (0xf << 16) | (VM_INVALIDATE_REQUEST >> 2));
+	radeon_ring_write(ring, 1 << vm->id);
+}
+
+/**
+ * si_copy_dma - copy pages using the DMA engine
+ *
+ * @rdev: radeon_device pointer
+ * @src_offset: src GPU address
+ * @dst_offset: dst GPU address
+ * @num_gpu_pages: number of GPU pages to xfer
+ * @fence: radeon fence object
+ *
+ * Copy GPU paging using the DMA engine (SI).
+ * Used by the radeon ttm implementation to move pages if
+ * registered as the asic copy callback.
+ */
+int si_copy_dma(struct radeon_device *rdev,
+		uint64_t src_offset, uint64_t dst_offset,
+		unsigned num_gpu_pages,
+		struct radeon_fence **fence)
+{
+	struct radeon_semaphore *sem = NULL;
+	int ring_index = rdev->asic->copy.dma_ring_index;
+	struct radeon_ring *ring = &rdev->ring[ring_index];
+	u32 size_in_bytes, cur_size_in_bytes;
+	int i, num_loops;
+	int r = 0;
+
+	r = radeon_semaphore_create(rdev, &sem);
+	if (r) {
+		DRM_ERROR("radeon: moving bo (%d).\n", r);
+		return r;
+	}
+
+	size_in_bytes = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT);
+	num_loops = DIV_ROUND_UP(size_in_bytes, 0xfffff);
+	r = radeon_ring_lock(rdev, ring, num_loops * 5 + 11);
+	if (r) {
+		DRM_ERROR("radeon: moving bo (%d).\n", r);
+		radeon_semaphore_free(rdev, &sem, NULL);
+		return r;
+	}
+
+	if (radeon_fence_need_sync(*fence, ring->idx)) {
+		radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring,
+					    ring->idx);
+		radeon_fence_note_sync(*fence, ring->idx);
+	} else {
+		radeon_semaphore_free(rdev, &sem, NULL);
+	}
+
+	for (i = 0; i < num_loops; i++) {
+		cur_size_in_bytes = size_in_bytes;
+		if (cur_size_in_bytes > 0xFFFFF)
+			cur_size_in_bytes = 0xFFFFF;
+		size_in_bytes -= cur_size_in_bytes;
+		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_COPY, 1, 0, 0, cur_size_in_bytes));
+		radeon_ring_write(ring, dst_offset & 0xffffffff);
+		radeon_ring_write(ring, src_offset & 0xffffffff);
+		radeon_ring_write(ring, upper_32_bits(dst_offset) & 0xff);
+		radeon_ring_write(ring, upper_32_bits(src_offset) & 0xff);
+		src_offset += cur_size_in_bytes;
+		dst_offset += cur_size_in_bytes;
+	}
+
+	r = radeon_fence_emit(rdev, fence, ring->idx);
+	if (r) {
+		radeon_ring_unlock_undo(rdev, ring);
+		return r;
+	}
+
+	radeon_ring_unlock_commit(rdev, ring);
+	radeon_semaphore_free(rdev, &sem, *fence);
+
+	return r;
+}
+