drm/radeon/kms: Add initial support for async DMA on SI

Pretty much the same as cayman.  Some changes to the copy
packets.

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

drivers/gpu/drm/radeon/radeon.h

@@ -797,6 +797,10 @@ void r600_dma_stop(struct radeon_device *rdev);
 int r600_dma_resume(struct radeon_device *rdev);
 void r600_dma_fini(struct radeon_device *rdev);
 
+void cayman_dma_stop(struct radeon_device *rdev);
+int cayman_dma_resume(struct radeon_device *rdev);
+void cayman_dma_fini(struct radeon_device *rdev);
+
 /*
  * CS.
  */

drivers/gpu/drm/radeon/radeon_asic.c

@@ -1731,6 +1731,26 @@ static struct radeon_asic si_asic = {
 			.ib_test = &r600_ib_test,
 			.is_lockup = &si_gpu_is_lockup,
 			.vm_flush = &si_vm_flush,
+		},
+		[R600_RING_TYPE_DMA_INDEX] = {
+			.ib_execute = &cayman_dma_ring_ib_execute,
+			.emit_fence = &evergreen_dma_fence_ring_emit,
+			.emit_semaphore = &r600_dma_semaphore_ring_emit,
+			.cs_parse = NULL,
+			.ring_test = &r600_dma_ring_test,
+			.ib_test = &r600_dma_ib_test,
+			.is_lockup = &cayman_dma_is_lockup,
+			.vm_flush = &si_dma_vm_flush,
+		},
+		[CAYMAN_RING_TYPE_DMA1_INDEX] = {
+			.ib_execute = &cayman_dma_ring_ib_execute,
+			.emit_fence = &evergreen_dma_fence_ring_emit,
+			.emit_semaphore = &r600_dma_semaphore_ring_emit,
+			.cs_parse = NULL,
+			.ring_test = &r600_dma_ring_test,
+			.ib_test = &r600_dma_ib_test,
+			.is_lockup = &cayman_dma_is_lockup,
+			.vm_flush = &si_dma_vm_flush,
 		}
 	},
 	.irq = {
@@ -1747,8 +1767,8 @@ static struct radeon_asic si_asic = {
 	.copy = {
 		.blit = NULL,
 		.blit_ring_index = RADEON_RING_TYPE_GFX_INDEX,
-		.dma = NULL,
+		.dma = &si_copy_dma,
-		.dma_ring_index = RADEON_RING_TYPE_GFX_INDEX,
+		.dma_ring_index = R600_RING_TYPE_DMA_INDEX,
 		.copy = NULL,
 		.copy_ring_index = RADEON_RING_TYPE_GFX_INDEX,
 	},

drivers/gpu/drm/radeon/radeon_asic.h

@@ -501,5 +501,10 @@ void si_vm_set_page(struct radeon_device *rdev, uint64_t pe,
 void si_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm);
 int si_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib);
 uint64_t si_get_gpu_clock(struct radeon_device *rdev);
+int si_copy_dma(struct radeon_device *rdev,
+		uint64_t src_offset, uint64_t dst_offset,
+		unsigned num_gpu_pages,
+		struct radeon_fence **fence);
+void si_dma_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm);
 
 #endif

drivers/gpu/drm/radeon/si.c

@@ -1660,6 +1660,8 @@ static void si_gpu_init(struct radeon_device *rdev)
 	WREG32(GB_ADDR_CONFIG, gb_addr_config);
 	WREG32(DMIF_ADDR_CONFIG, gb_addr_config);
 	WREG32(HDP_ADDR_CONFIG, gb_addr_config);
+	WREG32(DMA_TILING_CONFIG + DMA0_REGISTER_OFFSET, gb_addr_config);
+	WREG32(DMA_TILING_CONFIG + DMA1_REGISTER_OFFSET, gb_addr_config);
 
 	si_tiling_mode_table_init(rdev);
 
@@ -1836,6 +1838,9 @@ static void si_cp_enable(struct radeon_device *rdev, bool enable)
 		radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
 		WREG32(CP_ME_CNTL, (CP_ME_HALT | CP_PFP_HALT | CP_CE_HALT));
 		WREG32(SCRATCH_UMSK, 0);
+		rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
+		rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false;
+		rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false;
 	}
 	udelay(50);
 }
@@ -2891,6 +2896,32 @@ 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);
+}
+
 /*
  * RLC
  */
@@ -3059,6 +3090,10 @@ static void si_disable_interrupt_state(struct radeon_device *rdev)
 	WREG32(CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
 	WREG32(CP_INT_CNTL_RING1, 0);
 	WREG32(CP_INT_CNTL_RING2, 0);
+	tmp = RREG32(DMA_CNTL + DMA0_REGISTER_OFFSET) & ~TRAP_ENABLE;
+	WREG32(DMA_CNTL + DMA0_REGISTER_OFFSET, tmp);
+	tmp = RREG32(DMA_CNTL + DMA1_REGISTER_OFFSET) & ~TRAP_ENABLE;
+	WREG32(DMA_CNTL + DMA1_REGISTER_OFFSET, tmp);
 	WREG32(GRBM_INT_CNTL, 0);
 	WREG32(INT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
 	WREG32(INT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
@@ -3178,6 +3213,7 @@ int si_irq_set(struct radeon_device *rdev)
 	u32 hpd1, hpd2, hpd3, hpd4, hpd5, hpd6;
 	u32 grbm_int_cntl = 0;
 	u32 grph1 = 0, grph2 = 0, grph3 = 0, grph4 = 0, grph5 = 0, grph6 = 0;
+	u32 dma_cntl, dma_cntl1;
 
 	if (!rdev->irq.installed) {
 		WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
@@ -3198,6 +3234,9 @@ int si_irq_set(struct radeon_device *rdev)
 	hpd5 = RREG32(DC_HPD5_INT_CONTROL) & ~DC_HPDx_INT_EN;
 	hpd6 = RREG32(DC_HPD6_INT_CONTROL) & ~DC_HPDx_INT_EN;
 
+	dma_cntl = RREG32(DMA_CNTL + DMA0_REGISTER_OFFSET) & ~TRAP_ENABLE;
+	dma_cntl1 = RREG32(DMA_CNTL + DMA1_REGISTER_OFFSET) & ~TRAP_ENABLE;
+
 	/* enable CP interrupts on all rings */
 	if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
 		DRM_DEBUG("si_irq_set: sw int gfx\n");
@@ -3211,6 +3250,15 @@ int si_irq_set(struct radeon_device *rdev)
 		DRM_DEBUG("si_irq_set: sw int cp2\n");
 		cp_int_cntl2 |= TIME_STAMP_INT_ENABLE;
 	}
+	if (atomic_read(&rdev->irq.ring_int[R600_RING_TYPE_DMA_INDEX])) {
+		DRM_DEBUG("si_irq_set: sw int dma\n");
+		dma_cntl |= TRAP_ENABLE;
+	}
+
+	if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_DMA1_INDEX])) {
+		DRM_DEBUG("si_irq_set: sw int dma1\n");
+		dma_cntl1 |= TRAP_ENABLE;
+	}
 	if (rdev->irq.crtc_vblank_int[0] ||
 	    atomic_read(&rdev->irq.pflip[0])) {
 		DRM_DEBUG("si_irq_set: vblank 0\n");
@@ -3270,6 +3318,9 @@ int si_irq_set(struct radeon_device *rdev)
 	WREG32(CP_INT_CNTL_RING1, cp_int_cntl1);
 	WREG32(CP_INT_CNTL_RING2, cp_int_cntl2);
 
+	WREG32(DMA_CNTL + DMA0_REGISTER_OFFSET, dma_cntl);
+	WREG32(DMA_CNTL + DMA1_REGISTER_OFFSET, dma_cntl1);
+
 	WREG32(GRBM_INT_CNTL, grbm_int_cntl);
 
 	WREG32(INT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, crtc1);
@@ -3728,9 +3779,17 @@ int si_irq_process(struct radeon_device *rdev)
 				break;
 			}
 			break;
+		case 224: /* DMA trap event */
+			DRM_DEBUG("IH: DMA trap\n");
+			radeon_fence_process(rdev, R600_RING_TYPE_DMA_INDEX);
+			break;
 		case 233: /* GUI IDLE */
 			DRM_DEBUG("IH: GUI idle\n");
 			break;
+		case 244: /* DMA trap event */
+			DRM_DEBUG("IH: DMA1 trap\n");
+			radeon_fence_process(rdev, CAYMAN_RING_TYPE_DMA1_INDEX);
+			break;
 		default:
 			DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
 			break;
@@ -3754,6 +3813,80 @@ int si_irq_process(struct radeon_device *rdev)
 	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
  */
@@ -3825,6 +3958,18 @@ static int si_startup(struct radeon_device *rdev)
 		return r;
 	}
 
+	r = radeon_fence_driver_start_ring(rdev, R600_RING_TYPE_DMA_INDEX);
+	if (r) {
+		dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r);
+		return r;
+	}
+
+	r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_DMA1_INDEX);
+	if (r) {
+		dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r);
+		return r;
+	}
+
 	/* Enable IRQ */
 	r = si_irq_init(rdev);
 	if (r) {
@@ -3855,6 +4000,22 @@ static int si_startup(struct radeon_device *rdev)
 	if (r)
 		return r;
 
+	ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
+	r = radeon_ring_init(rdev, ring, ring->ring_size, R600_WB_DMA_RPTR_OFFSET,
+			     DMA_RB_RPTR + DMA0_REGISTER_OFFSET,
+			     DMA_RB_WPTR + DMA0_REGISTER_OFFSET,
+			     2, 0x3fffc, DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0, 0));
+	if (r)
+		return r;
+
+	ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
+	r = radeon_ring_init(rdev, ring, ring->ring_size, CAYMAN_WB_DMA1_RPTR_OFFSET,
+			     DMA_RB_RPTR + DMA1_REGISTER_OFFSET,
+			     DMA_RB_WPTR + DMA1_REGISTER_OFFSET,
+			     2, 0x3fffc, DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0, 0));
+	if (r)
+		return r;
+
 	r = si_cp_load_microcode(rdev);
 	if (r)
 		return r;
@@ -3862,6 +4023,10 @@ static int si_startup(struct radeon_device *rdev)
 	if (r)
 		return r;
 
+	r = cayman_dma_resume(rdev);
+	if (r)
+		return r;
+
 	r = radeon_ib_pool_init(rdev);
 	if (r) {
 		dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
@@ -3903,9 +4068,7 @@ int si_resume(struct radeon_device *rdev)
 int si_suspend(struct radeon_device *rdev)
 {
 	si_cp_enable(rdev, false);
-	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
+	cayman_dma_stop(rdev);
-	rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false;
-	rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false;
 	si_irq_suspend(rdev);
 	radeon_wb_disable(rdev);
 	si_pcie_gart_disable(rdev);
@@ -3983,6 +4146,14 @@ int si_init(struct radeon_device *rdev)
 	ring->ring_obj = NULL;
 	r600_ring_init(rdev, ring, 1024 * 1024);
 
+	ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
+	ring->ring_obj = NULL;
+	r600_ring_init(rdev, ring, 64 * 1024);
+
+	ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
+	ring->ring_obj = NULL;
+	r600_ring_init(rdev, ring, 64 * 1024);
+
 	rdev->ih.ring_obj = NULL;
 	r600_ih_ring_init(rdev, 64 * 1024);
 
@@ -3995,6 +4166,7 @@ int si_init(struct radeon_device *rdev)
 	if (r) {
 		dev_err(rdev->dev, "disabling GPU acceleration\n");
 		si_cp_fini(rdev);
+		cayman_dma_fini(rdev);
 		si_irq_fini(rdev);
 		si_rlc_fini(rdev);
 		radeon_wb_fini(rdev);
@@ -4023,6 +4195,7 @@ void si_fini(struct radeon_device *rdev)
 	r600_blit_fini(rdev);
 #endif
 	si_cp_fini(rdev);
+	cayman_dma_fini(rdev);
 	si_irq_fini(rdev);
 	si_rlc_fini(rdev);
 	radeon_wb_fini(rdev);

drivers/gpu/drm/radeon/sid.h

@@ -936,4 +936,51 @@
 #define	PACKET3_WAIT_ON_AVAIL_BUFFER			0x8A
 #define	PACKET3_SWITCH_BUFFER				0x8B
 
+/* ASYNC DMA - first instance at 0xd000, second at 0xd800 */
+#define DMA0_REGISTER_OFFSET                              0x0 /* not a register */
+#define DMA1_REGISTER_OFFSET                              0x800 /* not a register */
+
+#define DMA_RB_CNTL                                       0xd000
+#       define DMA_RB_ENABLE                              (1 << 0)
+#       define DMA_RB_SIZE(x)                             ((x) << 1) /* log2 */
+#       define DMA_RB_SWAP_ENABLE                         (1 << 9) /* 8IN32 */
+#       define DMA_RPTR_WRITEBACK_ENABLE                  (1 << 12)
+#       define DMA_RPTR_WRITEBACK_SWAP_ENABLE             (1 << 13)  /* 8IN32 */
+#       define DMA_RPTR_WRITEBACK_TIMER(x)                ((x) << 16) /* log2 */
+#define DMA_RB_BASE                                       0xd004
+#define DMA_RB_RPTR                                       0xd008
+#define DMA_RB_WPTR                                       0xd00c
+
+#define DMA_RB_RPTR_ADDR_HI                               0xd01c
+#define DMA_RB_RPTR_ADDR_LO                               0xd020
+
+#define DMA_IB_CNTL                                       0xd024
+#       define DMA_IB_ENABLE                              (1 << 0)
+#       define DMA_IB_SWAP_ENABLE                         (1 << 4)
+#define DMA_IB_RPTR                                       0xd028
+#define DMA_CNTL                                          0xd02c
+#       define TRAP_ENABLE                                (1 << 0)
+#       define SEM_INCOMPLETE_INT_ENABLE                  (1 << 1)
+#       define SEM_WAIT_INT_ENABLE                        (1 << 2)
+#       define DATA_SWAP_ENABLE                           (1 << 3)
+#       define FENCE_SWAP_ENABLE                          (1 << 4)
+#       define CTXEMPTY_INT_ENABLE                        (1 << 28)
+#define DMA_TILING_CONFIG  				  0xd0b8
+
+#define DMA_PACKET(cmd, b, t, s, n)	((((cmd) & 0xF) << 28) |	\
+					 (((b) & 0x1) << 26) |		\
+					 (((t) & 0x1) << 23) |		\
+					 (((s) & 0x1) << 22) |		\
+					 (((n) & 0xFFFFF) << 0))
+/* async DMA Packet types */
+#define	DMA_PACKET_WRITE				  0x2
+#define	DMA_PACKET_COPY					  0x3
+#define	DMA_PACKET_INDIRECT_BUFFER			  0x4
+#define	DMA_PACKET_SEMAPHORE				  0x5
+#define	DMA_PACKET_FENCE				  0x6
+#define	DMA_PACKET_TRAP					  0x7
+#define	DMA_PACKET_SRBM_WRITE				  0x9
+#define	DMA_PACKET_CONSTANT_FILL			  0xd
+#define	DMA_PACKET_NOP					  0xf
+
 #endif