1 files changed, 221 insertions, 37 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c b/drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c
index e074f7ed388c..def583916294 100644
--- a/drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c
@@ -44,10 +44,11 @@
 
 #define mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET			0x431
 #define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET		0x3b4
-#define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET 	0x3b5
+#define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET		0x3b5
 #define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET			0x25c
 
-#define VCN_INSTANCES_SIENNA_CICHLID	 				2
+#define VCN_INSTANCES_SIENNA_CICHLID				2
+#define DEC_SW_RING_ENABLED					FALSE
 
 static int amdgpu_ih_clientid_vcns[] = {
 	SOC15_IH_CLIENTID_VCN,
@@ -55,8 +56,8 @@ static int amdgpu_ih_clientid_vcns[] = {
 };
 
 static int amdgpu_ucode_id_vcns[] = {
-       AMDGPU_UCODE_ID_VCN,
-       AMDGPU_UCODE_ID_VCN1
+	AMDGPU_UCODE_ID_VCN,
+	AMDGPU_UCODE_ID_VCN1
 };
 
 static int vcn_v3_0_start_sriov(struct amdgpu_device *adev);
@@ -155,6 +156,13 @@ static int vcn_v3_0_sw_init(void *handle)
 	if (r)
 		return r;
 
+	/*
+	 * Note: doorbell assignment is fixed for SRIOV multiple VCN engines
+	 * Formula:
+	 *   vcn_db_base  = adev->doorbell_index.vcn.vcn_ring0_1 << 1;
+	 *   dec_ring_i   = vcn_db_base + i * (adev->vcn.num_enc_rings + 1)
+	 *   enc_ring_i,j = vcn_db_base + i * (adev->vcn.num_enc_rings + 1) + 1 + j
+	 */
 	if (amdgpu_sriov_vf(adev)) {
 		vcn_doorbell_index = adev->doorbell_index.vcn.vcn_ring0_1;
 		/* get DWORD offset */
@@ -162,6 +170,7 @@ static int vcn_v3_0_sw_init(void *handle)
 	}
 
 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
+		volatile struct amdgpu_fw_shared *fw_shared;
 		if (adev->vcn.harvest_config & (1 << i))
 			continue;
 
@@ -192,9 +201,7 @@ static int vcn_v3_0_sw_init(void *handle)
 		ring = &adev->vcn.inst[i].ring_dec;
 		ring->use_doorbell = true;
 		if (amdgpu_sriov_vf(adev)) {
-			ring->doorbell_index = vcn_doorbell_index;
-			/* NOTE: increment so next VCN engine use next DOORBELL DWORD */
-			vcn_doorbell_index++;
+			ring->doorbell_index = vcn_doorbell_index + i * (adev->vcn.num_enc_rings + 1);
 		} else {
 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i;
 		}
@@ -216,9 +223,7 @@ static int vcn_v3_0_sw_init(void *handle)
 			ring = &adev->vcn.inst[i].ring_enc[j];
 			ring->use_doorbell = true;
 			if (amdgpu_sriov_vf(adev)) {
-				ring->doorbell_index = vcn_doorbell_index;
-				/* NOTE: increment so next VCN engine use next DOORBELL DWORD */
-				vcn_doorbell_index++;
+				ring->doorbell_index = vcn_doorbell_index + i * (adev->vcn.num_enc_rings + 1) + 1 + j;
 			} else {
 				ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + j + 8 * i;
 			}
@@ -230,6 +235,11 @@ static int vcn_v3_0_sw_init(void *handle)
 			if (r)
 				return r;
 		}
+
+		fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
+		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_SW_RING_FLAG) |
+					     cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG);
+		fw_shared->sw_ring.is_enabled = cpu_to_le32(DEC_SW_RING_ENABLED);
 	}
 
 	if (amdgpu_sriov_vf(adev)) {
@@ -253,7 +263,17 @@ static int vcn_v3_0_sw_init(void *handle)
 static int vcn_v3_0_sw_fini(void *handle)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-	int r;
+	int i, r;
+
+	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
+		volatile struct amdgpu_fw_shared *fw_shared;
+
+		if (adev->vcn.harvest_config & (1 << i))
+			continue;
+		fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
+		fw_shared->present_flag_0 = 0;
+		fw_shared->sw_ring.is_enabled = false;
+	}
 
 	if (amdgpu_sriov_vf(adev))
 		amdgpu_virt_free_mm_table(adev);
@@ -291,17 +311,19 @@ static int vcn_v3_0_hw_init(void *handle)
 				continue;
 
 			ring = &adev->vcn.inst[i].ring_dec;
-			ring->wptr = 0;
-			ring->wptr_old = 0;
-			vcn_v3_0_dec_ring_set_wptr(ring);
-			ring->sched.ready = true;
+			if (ring->sched.ready) {
+				ring->wptr = 0;
+				ring->wptr_old = 0;
+				vcn_v3_0_dec_ring_set_wptr(ring);
+			}
 
 			for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
 				ring = &adev->vcn.inst[i].ring_enc[j];
-				ring->wptr = 0;
-				ring->wptr_old = 0;
-				vcn_v3_0_enc_ring_set_wptr(ring);
-				ring->sched.ready = true;
+				if (ring->sched.ready) {
+					ring->wptr = 0;
+					ring->wptr_old = 0;
+					vcn_v3_0_enc_ring_set_wptr(ring);
+				}
 			}
 		}
 	} else {
@@ -461,6 +483,15 @@ static void vcn_v3_0_mc_resume(struct amdgpu_device *adev, int inst)
 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET2, 0);
 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
+
+	/* non-cache window */
+	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
+		lower_32_bits(adev->vcn.inst[inst].fw_shared_gpu_addr));
+	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
+		upper_32_bits(adev->vcn.inst[inst].fw_shared_gpu_addr));
+	WREG32_SOC15(VCN, inst, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
+	WREG32_SOC15(VCN, inst, mmUVD_VCPU_NONCACHE_SIZE0,
+		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
 }
 
 static void vcn_v3_0_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
@@ -543,13 +574,16 @@ static void vcn_v3_0_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx
 
 	/* non-cache window */
 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
-			VCN, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW), 0, 0, indirect);
+			VCN, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
+			lower_32_bits(adev->vcn.inst[inst_idx].fw_shared_gpu_addr), 0, indirect);
 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
-			VCN, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH), 0, 0, indirect);
+			VCN, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
+			upper_32_bits(adev->vcn.inst[inst_idx].fw_shared_gpu_addr), 0, indirect);
 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 			VCN, inst_idx, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
-			VCN, inst_idx, mmUVD_VCPU_NONCACHE_SIZE0), 0, 0, indirect);
+			VCN, inst_idx, mmUVD_VCPU_NONCACHE_SIZE0),
+			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
 }
 
 static void vcn_v3_0_disable_static_power_gating(struct amdgpu_device *adev, int inst)
@@ -902,6 +936,7 @@ static void vcn_v3_0_enable_clock_gating(struct amdgpu_device *adev, int inst)
 
 static int vcn_v3_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
 {
+	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared_cpu_addr;
 	struct amdgpu_ring *ring;
 	uint32_t rb_bufsz, tmp;
 
@@ -915,7 +950,7 @@ static int vcn_v3_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, boo
 	WREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS, tmp);
 
 	if (indirect)
-		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t*)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
+		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
 
 	/* enable clock gating */
 	vcn_v3_0_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
@@ -1011,6 +1046,12 @@ static int vcn_v3_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, boo
 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_CNTL, tmp);
 
+	/* Stall DPG before WPTR/RPTR reset */
+	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
+		UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
+		~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
+	fw_shared->multi_queue.decode_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
+
 	/* set the write pointer delay */
 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
 
@@ -1033,11 +1074,17 @@ static int vcn_v3_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, boo
 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR,
 		lower_32_bits(ring->wptr));
 
+	fw_shared->multi_queue.decode_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
+	/* Unstall DPG */
+	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
+		0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
+
 	return 0;
 }
 
 static int vcn_v3_0_start(struct amdgpu_device *adev)
 {
+	volatile struct amdgpu_fw_shared *fw_shared;
 	struct amdgpu_ring *ring;
 	uint32_t rb_bufsz, tmp;
 	int i, j, k, r;
@@ -1180,6 +1227,9 @@ static int vcn_v3_0_start(struct amdgpu_device *adev)
 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_CNTL, tmp);
 
+		fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
+		fw_shared->multi_queue.decode_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
+
 		/* programm the RB_BASE for ring buffer */
 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
 			lower_32_bits(ring->gpu_addr));
@@ -1192,19 +1242,25 @@ static int vcn_v3_0_start(struct amdgpu_device *adev)
 		ring->wptr = RREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR);
 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_WPTR,
 			lower_32_bits(ring->wptr));
+		fw_shared->multi_queue.decode_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
+
+		fw_shared->multi_queue.encode_generalpurpose_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
 		ring = &adev->vcn.inst[i].ring_enc[0];
 		WREG32_SOC15(VCN, i, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
 		WREG32_SOC15(VCN, i, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO, ring->gpu_addr);
 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
 		WREG32_SOC15(VCN, i, mmUVD_RB_SIZE, ring->ring_size / 4);
+		fw_shared->multi_queue.encode_generalpurpose_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
 
+		fw_shared->multi_queue.encode_lowlatency_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
 		ring = &adev->vcn.inst[i].ring_enc[1];
 		WREG32_SOC15(VCN, i, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
 		WREG32_SOC15(VCN, i, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO2, ring->gpu_addr);
 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
 		WREG32_SOC15(VCN, i, mmUVD_RB_SIZE2, ring->ring_size / 4);
+		fw_shared->multi_queue.encode_lowlatency_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
 	}
 
 	return 0;
@@ -1227,12 +1283,12 @@ static int vcn_v3_0_start_sriov(struct amdgpu_device *adev)
 	uint32_t table_size;
 	uint32_t size, size_dw;
 
+	bool is_vcn_ready;
+
 	struct mmsch_v3_0_cmd_direct_write
 		direct_wt = { {0} };
 	struct mmsch_v3_0_cmd_direct_read_modify_write
 		direct_rd_mod_wt = { {0} };
-	struct mmsch_v3_0_cmd_direct_polling
-		direct_poll = { {0} };
 	struct mmsch_v3_0_cmd_end end = { {0} };
 	struct mmsch_v3_0_init_header header;
 
@@ -1240,8 +1296,6 @@ static int vcn_v3_0_start_sriov(struct amdgpu_device *adev)
 		MMSCH_COMMAND__DIRECT_REG_WRITE;
 	direct_rd_mod_wt.cmd_header.command_type =
 		MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
-	direct_poll.cmd_header.command_type =
-		MMSCH_COMMAND__DIRECT_REG_POLLING;
 	end.cmd_header.command_type =
 		MMSCH_COMMAND__END;
 
@@ -1364,14 +1418,14 @@ static int vcn_v3_0_start_sriov(struct amdgpu_device *adev)
 		MMSCH_V3_0_INSERT_END();
 
 		/* refine header */
-		header.inst[i].init_status = 1;
+		header.inst[i].init_status = 0;
 		header.inst[i].table_offset = header.total_size;
 		header.inst[i].table_size = table_size;
 		header.total_size += table_size;
 	}
 
 	/* Update init table header in memory */
-        size = sizeof(struct mmsch_v3_0_init_header);
+	size = sizeof(struct mmsch_v3_0_init_header);
 	table_loc = (uint32_t *)table->cpu_addr;
 	memcpy((void *)table_loc, &header, size);
 
@@ -1422,6 +1476,30 @@ static int vcn_v3_0_start_sriov(struct amdgpu_device *adev)
 		}
 	}
 
+	/* 6, check each VCN's init_status
+	 * if it remains as 0, then this VCN is not assigned to current VF
+	 * do not start ring for this VCN
+	 */
+	size = sizeof(struct mmsch_v3_0_init_header);
+	table_loc = (uint32_t *)table->cpu_addr;
+	memcpy(&header, (void *)table_loc, size);
+
+	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
+		if (adev->vcn.harvest_config & (1 << i))
+			continue;
+
+		is_vcn_ready = (header.inst[i].init_status == 1);
+		if (!is_vcn_ready)
+			DRM_INFO("VCN(%d) engine is disabled by hypervisor\n", i);
+
+		ring = &adev->vcn.inst[i].ring_dec;
+		ring->sched.ready = is_vcn_ready;
+		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
+			ring = &adev->vcn.inst[i].ring_enc[j];
+			ring->sched.ready = is_vcn_ready;
+		}
+	}
+
 	return 0;
 }
 
@@ -1531,6 +1609,7 @@ static int vcn_v3_0_stop(struct amdgpu_device *adev)
 static int vcn_v3_0_pause_dpg_mode(struct amdgpu_device *adev,
 		   int inst_idx, struct dpg_pause_state *new_state)
 {
+	volatile struct amdgpu_fw_shared *fw_shared;
 	struct amdgpu_ring *ring;
 	uint32_t reg_data = 0;
 	int ret_code;
@@ -1556,23 +1635,36 @@ static int vcn_v3_0_pause_dpg_mode(struct amdgpu_device *adev,
 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
 
+				/* Stall DPG before WPTR/RPTR reset */
+				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
+					UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
+					~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
+
 				/* Restore */
+				fw_shared = adev->vcn.inst[inst_idx].fw_shared_cpu_addr;
+				fw_shared->multi_queue.encode_generalpurpose_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
 				ring = &adev->vcn.inst[inst_idx].ring_enc[0];
+				ring->wptr = 0;
 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4);
 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
+				fw_shared->multi_queue.encode_generalpurpose_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
 
+				fw_shared->multi_queue.encode_lowlatency_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
 				ring = &adev->vcn.inst[inst_idx].ring_enc[1];
+				ring->wptr = 0;
 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4);
 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
+				fw_shared->multi_queue.encode_lowlatency_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
 
-				WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR,
-					RREG32_SOC15(VCN, inst_idx, mmUVD_SCRATCH2) & 0x7FFFFFFF);
+				/* Unstall DPG */
+				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
+					0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
 
 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS,
 					UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
@@ -1630,10 +1722,6 @@ static void vcn_v3_0_dec_ring_set_wptr(struct amdgpu_ring *ring)
 {
 	struct amdgpu_device *adev = ring->adev;
 
-	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
-		WREG32_SOC15(VCN, ring->me, mmUVD_SCRATCH2,
-			lower_32_bits(ring->wptr) | 0x80000000);
-
 	if (ring->use_doorbell) {
 		adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
@@ -1642,6 +1730,98 @@ static void vcn_v3_0_dec_ring_set_wptr(struct amdgpu_ring *ring)
 	}
 }
 
+static void vcn_v3_0_dec_sw_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
+				u64 seq, uint32_t flags)
+{
+	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
+
+	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_FENCE);
+	amdgpu_ring_write(ring, addr);
+	amdgpu_ring_write(ring, upper_32_bits(addr));
+	amdgpu_ring_write(ring, seq);
+	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_TRAP);
+}
+
+static void vcn_v3_0_dec_sw_ring_insert_end(struct amdgpu_ring *ring)
+{
+	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_END);
+}
+
+static void vcn_v3_0_dec_sw_ring_emit_ib(struct amdgpu_ring *ring,
+			       struct amdgpu_job *job,
+			       struct amdgpu_ib *ib,
+			       uint32_t flags)
+{
+	uint32_t vmid = AMDGPU_JOB_GET_VMID(job);
+
+	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_IB);
+	amdgpu_ring_write(ring, vmid);
+	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
+	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
+	amdgpu_ring_write(ring, ib->length_dw);
+}
+
+static void vcn_v3_0_dec_sw_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
+				uint32_t val, uint32_t mask)
+{
+	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_REG_WAIT);
+	amdgpu_ring_write(ring, reg << 2);
+	amdgpu_ring_write(ring, mask);
+	amdgpu_ring_write(ring, val);
+}
+
+static void vcn_v3_0_dec_sw_ring_emit_vm_flush(struct amdgpu_ring *ring,
+				uint32_t vmid, uint64_t pd_addr)
+{
+	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
+	uint32_t data0, data1, mask;
+
+	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
+
+	/* wait for register write */
+	data0 = hub->ctx0_ptb_addr_lo32 + vmid * hub->ctx_addr_distance;
+	data1 = lower_32_bits(pd_addr);
+	mask = 0xffffffff;
+	vcn_v3_0_dec_sw_ring_emit_reg_wait(ring, data0, data1, mask);
+}
+
+static void vcn_v3_0_dec_sw_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg, uint32_t val)
+{
+	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_REG_WRITE);
+	amdgpu_ring_write(ring,	reg << 2);
+	amdgpu_ring_write(ring, val);
+}
+
+static const struct amdgpu_ring_funcs vcn_v3_0_dec_sw_ring_vm_funcs = {
+	.type = AMDGPU_RING_TYPE_VCN_DEC,
+	.align_mask = 0x3f,
+	.nop = VCN_DEC_SW_CMD_NO_OP,
+	.vmhub = AMDGPU_MMHUB_0,
+	.get_rptr = vcn_v3_0_dec_ring_get_rptr,
+	.get_wptr = vcn_v3_0_dec_ring_get_wptr,
+	.set_wptr = vcn_v3_0_dec_ring_set_wptr,
+	.emit_frame_size =
+		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
+		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
+		4 + /* vcn_v3_0_dec_sw_ring_emit_vm_flush */
+		5 + 5 + /* vcn_v3_0_dec_sw_ring_emit_fdec_swe x2 vm fdec_swe */
+		1, /* vcn_v3_0_dec_sw_ring_insert_end */
+	.emit_ib_size = 5, /* vcn_v3_0_dec_sw_ring_emit_ib */
+	.emit_ib = vcn_v3_0_dec_sw_ring_emit_ib,
+	.emit_fence = vcn_v3_0_dec_sw_ring_emit_fence,
+	.emit_vm_flush = vcn_v3_0_dec_sw_ring_emit_vm_flush,
+	.test_ring = amdgpu_vcn_dec_sw_ring_test_ring,
+	.test_ib = NULL,//amdgpu_vcn_dec_sw_ring_test_ib,
+	.insert_nop = amdgpu_ring_insert_nop,
+	.insert_end = vcn_v3_0_dec_sw_ring_insert_end,
+	.pad_ib = amdgpu_ring_generic_pad_ib,
+	.begin_use = amdgpu_vcn_ring_begin_use,
+	.end_use = amdgpu_vcn_ring_end_use,
+	.emit_wreg = vcn_v3_0_dec_sw_ring_emit_wreg,
+	.emit_reg_wait = vcn_v3_0_dec_sw_ring_emit_reg_wait,
+	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
+};
+
 static const struct amdgpu_ring_funcs vcn_v3_0_dec_ring_vm_funcs = {
 	.type = AMDGPU_RING_TYPE_VCN_DEC,
 	.align_mask = 0xf,
@@ -1779,9 +1959,13 @@ static void vcn_v3_0_set_dec_ring_funcs(struct amdgpu_device *adev)
 		if (adev->vcn.harvest_config & (1 << i))
 			continue;
 
-		adev->vcn.inst[i].ring_dec.funcs = &vcn_v3_0_dec_ring_vm_funcs;
+		if (!DEC_SW_RING_ENABLED)
+			adev->vcn.inst[i].ring_dec.funcs = &vcn_v3_0_dec_ring_vm_funcs;
+		else
+			adev->vcn.inst[i].ring_dec.funcs = &vcn_v3_0_dec_sw_ring_vm_funcs;
 		adev->vcn.inst[i].ring_dec.me = i;
-		DRM_INFO("VCN(%d) decode is enabled in VM mode\n", i);
+		DRM_INFO("VCN(%d) decode%s is enabled in VM mode\n", i,
+			  DEC_SW_RING_ENABLED?"(Software Ring)":"");
 	}
 }