drm/amdgpu/vcn2.5: Add firmware w/r ptr reset sync

Add firmware write/read point reset sync through shared memory

Signed-off-by: James Zhu <James.Zhu@amd.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

1 files changed, 48 insertions, 5 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c b/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c
index dcb884ab0c71..8470e09eca0b 100644
--- a/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c
+++ b/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c
@@ -165,6 +165,8 @@ static int vcn_v2_5_sw_init(void *handle)
 		return r;
 
 	for (j = 0; j < adev->vcn.num_vcn_inst; j++) {
+		volatile struct amdgpu_fw_shared *fw_shared;
+
 		if (adev->vcn.harvest_config & (1 << j))
 			continue;
 		adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
@@ -207,6 +209,9 @@ static int vcn_v2_5_sw_init(void *handle)
 			if (r)
 				return r;
 		}
+
+		fw_shared = adev->vcn.inst[j].fw_shared_cpu_addr;
+		fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG);
 	}
 
 	if (amdgpu_sriov_vf(adev)) {
@@ -230,8 +235,16 @@ static int vcn_v2_5_sw_init(void *handle)
  */
 static int vcn_v2_5_sw_fini(void *handle)
 {
-	int r;
+	int i, r;
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	volatile struct amdgpu_fw_shared *fw_shared;
+
+	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
+		if (adev->vcn.harvest_config & (1 << i))
+			continue;
+		fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
+		fw_shared->present_flag_0 = 0;
+	}
 
 	if (amdgpu_sriov_vf(adev))
 		amdgpu_virt_free_mm_table(adev);
@@ -415,6 +428,15 @@ static void vcn_v2_5_mc_resume(struct amdgpu_device *adev)
 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
 		WREG32_SOC15(UVD, i, mmUVD_VCPU_CACHE_OFFSET2, 0);
 		WREG32_SOC15(UVD, i, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
+
+		/* non-cache window */
+		WREG32_SOC15(UVD, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
+			lower_32_bits(adev->vcn.inst[i].fw_shared_gpu_addr));
+		WREG32_SOC15(UVD, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
+			upper_32_bits(adev->vcn.inst[i].fw_shared_gpu_addr));
+		WREG32_SOC15(UVD, i, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
+		WREG32_SOC15(UVD, i, mmUVD_VCPU_NONCACHE_SIZE0,
+			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
 	}
 }
 
@@ -498,13 +520,16 @@ static void vcn_v2_5_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx
 
 	/* non-cache window */
 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
-		UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW), 0, 0, indirect);
+		UVD, 0, 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_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
-		UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH), 0, 0, indirect);
+		UVD, 0, 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_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
 		UVD, 0, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
-		UVD, 0, mmUVD_VCPU_NONCACHE_SIZE0), 0, 0, indirect);
+		UVD, 0, mmUVD_VCPU_NONCACHE_SIZE0),
+		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
 
 	/* VCN global tiling registers */
 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
@@ -741,6 +766,7 @@ static void vcn_v2_5_enable_clock_gating(struct amdgpu_device *adev)
 
 static int vcn_v2_5_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;
 
@@ -850,7 +876,7 @@ static int vcn_v2_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, boo
 	WREG32_P(SOC15_REG_OFFSET(UVD, 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 |= fw_queue_ring_reset;
+	fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
 
 	/* set the write pointer delay */
 	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
@@ -874,6 +900,7 @@ static int vcn_v2_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, boo
 	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_WPTR,
 		lower_32_bits(ring->wptr));
 
+	fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
 	/* Unstall DPG */
 	WREG32_P(SOC15_REG_OFFSET(UVD, inst_idx, mmUVD_POWER_STATUS),
 		0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
@@ -963,6 +990,7 @@ static int vcn_v2_5_start(struct amdgpu_device *adev)
 	vcn_v2_5_mc_resume(adev);
 
 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
+		volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
 		if (adev->vcn.harvest_config & (1 << i))
 			continue;
 		/* VCN global tiling registers */
@@ -1036,6 +1064,7 @@ static int vcn_v2_5_start(struct amdgpu_device *adev)
 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
 		WREG32_SOC15(UVD, i, mmUVD_RBC_RB_CNTL, tmp);
 
+		fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
 		/* programm the RB_BASE for ring buffer */
 		WREG32_SOC15(UVD, i, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
 			lower_32_bits(ring->gpu_addr));
@@ -1048,19 +1077,25 @@ static int vcn_v2_5_start(struct amdgpu_device *adev)
 		ring->wptr = RREG32_SOC15(UVD, i, mmUVD_RBC_RB_RPTR);
 		WREG32_SOC15(UVD, i, mmUVD_RBC_RB_WPTR,
 				lower_32_bits(ring->wptr));
+		fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
+
+		fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
 		ring = &adev->vcn.inst[i].ring_enc[0];
 		WREG32_SOC15(UVD, i, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
 		WREG32_SOC15(UVD, i, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
 		WREG32_SOC15(UVD, i, mmUVD_RB_BASE_LO, ring->gpu_addr);
 		WREG32_SOC15(UVD, i, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
 		WREG32_SOC15(UVD, i, mmUVD_RB_SIZE, ring->ring_size / 4);
+		fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
 
+		fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
 		ring = &adev->vcn.inst[i].ring_enc[1];
 		WREG32_SOC15(UVD, i, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
 		WREG32_SOC15(UVD, i, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
 		WREG32_SOC15(UVD, i, mmUVD_RB_BASE_LO2, ring->gpu_addr);
 		WREG32_SOC15(UVD, i, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
 		WREG32_SOC15(UVD, i, mmUVD_RB_SIZE2, ring->ring_size / 4);
+		fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
 	}
 
 	return 0;
@@ -1381,6 +1416,8 @@ static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
 
 			if (!ret_code) {
+				volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared_cpu_addr;
+
 				/* pause DPG */
 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
 				WREG32_SOC15(UVD, inst_idx, mmUVD_DPG_PAUSE, reg_data);
@@ -1396,6 +1433,7 @@ static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
 					   ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
 
 				/* Restore */
+				fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
 				ring = &adev->vcn.inst[inst_idx].ring_enc[0];
 				ring->wptr = 0;
 				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
@@ -1403,7 +1441,9 @@ static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
 				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4);
 				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
 				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
+				fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
 
+				fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
 				ring = &adev->vcn.inst[inst_idx].ring_enc[1];
 				ring->wptr = 0;
 				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
@@ -1411,9 +1451,12 @@ static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
 				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4);
 				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
 				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
+				fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
 
+				fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
 				WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_WPTR,
 					   RREG32_SOC15(UVD, inst_idx, mmUVD_SCRATCH2) & 0x7FFFFFFF);
+				fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
 
 				/* Unstall DPG */
 				WREG32_P(SOC15_REG_OFFSET(UVD, inst_idx, mmUVD_POWER_STATUS),