drm/amdgpu: implement gmc_v9_0_emit_flush_gpu_tlb

Unify tlb flushing for gmc v9.
Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h

@@ -69,7 +69,6 @@ struct amdgpu_gmc_funcs {
 	/* get the pde for a given mc addr */
 	void (*get_vm_pde)(struct amdgpu_device *adev, int level,
 			   u64 *dst, u64 *flags);
-	uint32_t (*get_invalidate_req)(unsigned int vmid);
 };
 
 struct amdgpu_gmc {

drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c

@@ -3688,31 +3688,16 @@ static void gfx_v9_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
 					uint64_t pd_addr)
 {
 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
-	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
-	uint32_t req = ring->adev->gmc.gmc_funcs->get_invalidate_req(vmid);
-	uint64_t flags = AMDGPU_PTE_VALID;
 	unsigned eng = ring->vm_inv_eng;
 
-	amdgpu_gmc_get_vm_pde(ring->adev, -1, &pd_addr, &flags);
-	pd_addr |= flags;
-
-	gfx_v9_0_write_data_to_reg(ring, usepfp, true,
-				   hub->ctx0_ptb_addr_lo32 + (2 * vmid),
-				   lower_32_bits(pd_addr));
-
-	gfx_v9_0_write_data_to_reg(ring, usepfp, true,
-				   hub->ctx0_ptb_addr_hi32 + (2 * vmid),
-				   upper_32_bits(pd_addr));
-
-	gfx_v9_0_write_data_to_reg(ring, usepfp, true,
-				   hub->vm_inv_eng0_req + eng, req);
+	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pasid, pd_addr);
 
 	/* wait for the invalidate to complete */
-	gfx_v9_0_wait_reg_mem(ring, 0, 0, 0, hub->vm_inv_eng0_ack +
-			      eng, 0, 1 << vmid, 1 << vmid, 0x20);
+	gfx_v9_0_wait_reg_mem(ring, 0, 0, 0, hub->vm_inv_eng0_ack + eng,
+			      0, 1 << vmid, 1 << vmid, 0x20);
 
 	/* compute doesn't have PFP */
-	if (usepfp) {
+	if (ring->funcs->type == AMDGPU_RING_TYPE_GFX) {
 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
 		amdgpu_ring_write(ring, 0x0);
@@ -4312,7 +4297,7 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_gfx = {
 	.emit_frame_size = /* totally 242 maximum if 16 IBs */
 		5 +  /* COND_EXEC */
 		7 +  /* PIPELINE_SYNC */
-		24 + /* VM_FLUSH */
+		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 + 9 + /* VM_FLUSH */
 		8 +  /* FENCE for VM_FLUSH */
 		20 + /* GDS switch */
 		4 + /* double SWITCH_BUFFER,
@@ -4361,7 +4346,7 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = {
 		7 + /* gfx_v9_0_ring_emit_hdp_flush */
 		5 + /* gfx_v9_0_ring_emit_hdp_invalidate */
 		7 + /* gfx_v9_0_ring_emit_pipeline_sync */
-		24 + /* gfx_v9_0_ring_emit_vm_flush */
+		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 + 9 + /* gfx_v9_0_ring_emit_vm_flush */
 		8 + 8 + 8, /* gfx_v9_0_ring_emit_fence x3 for user fence, vm fence */
 	.emit_ib_size =	4, /* gfx_v9_0_ring_emit_ib_compute */
 	.emit_ib = gfx_v9_0_ring_emit_ib_compute,
@@ -4393,7 +4378,7 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_kiq = {
 		7 + /* gfx_v9_0_ring_emit_hdp_flush */
 		5 + /* gfx_v9_0_ring_emit_hdp_invalidate */
 		7 + /* gfx_v9_0_ring_emit_pipeline_sync */
-		24 + /* gfx_v9_0_ring_emit_vm_flush */
+		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 + 9 + /* gfx_v9_0_ring_emit_vm_flush */
 		8 + 8 + 8, /* gfx_v9_0_ring_emit_fence_kiq x3 for user fence, vm fence */
 	.emit_ib_size =	4, /* gfx_v9_0_ring_emit_ib_compute */
 	.emit_ib = gfx_v9_0_ring_emit_ib_compute,

drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c

@@ -366,6 +366,29 @@ static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev,
 	spin_unlock(&adev->gmc.invalidate_lock);
 }
 
+static uint64_t gmc_v9_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
+					    unsigned vmid, unsigned pasid,
+					    uint64_t pd_addr)
+{
+	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
+	uint32_t req = gmc_v9_0_get_invalidate_req(vmid);
+	uint64_t flags = AMDGPU_PTE_VALID;
+	unsigned eng = ring->vm_inv_eng;
+
+	amdgpu_gmc_get_vm_pde(ring->adev, -1, &pd_addr, &flags);
+	pd_addr |= flags;
+
+	amdgpu_ring_emit_wreg(ring, hub->ctx0_ptb_addr_lo32 + (2 * vmid),
+			      lower_32_bits(pd_addr));
+
+	amdgpu_ring_emit_wreg(ring, hub->ctx0_ptb_addr_hi32 + (2 * vmid),
+			      upper_32_bits(pd_addr));
+
+	amdgpu_ring_emit_wreg(ring, hub->vm_inv_eng0_req + eng, req);
+
+	return pd_addr;
+}
+
 /**
  * gmc_v9_0_set_pte_pde - update the page tables using MMIO
  *
@@ -491,8 +514,8 @@ static void gmc_v9_0_get_vm_pde(struct amdgpu_device *adev, int level,
 
 static const struct amdgpu_gmc_funcs gmc_v9_0_gmc_funcs = {
 	.flush_gpu_tlb = gmc_v9_0_flush_gpu_tlb,
+	.emit_flush_gpu_tlb = gmc_v9_0_emit_flush_gpu_tlb,
 	.set_pte_pde = gmc_v9_0_set_pte_pde,
-	.get_invalidate_req = gmc_v9_0_get_invalidate_req,
 	.get_vm_pte_flags = gmc_v9_0_get_vm_pte_flags,
 	.get_vm_pde = gmc_v9_0_get_vm_pde
 };

drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c

@@ -1137,28 +1137,9 @@ static void sdma_v4_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
 					 uint64_t pd_addr)
 {
 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
-	uint32_t req = ring->adev->gmc.gmc_funcs->get_invalidate_req(vmid);
-	uint64_t flags = AMDGPU_PTE_VALID;
 	unsigned eng = ring->vm_inv_eng;
 
-	amdgpu_gmc_get_vm_pde(ring->adev, -1, &pd_addr, &flags);
-	pd_addr |= flags;
-
-	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SRBM_WRITE) |
-			  SDMA_PKT_SRBM_WRITE_HEADER_BYTE_EN(0xf));
-	amdgpu_ring_write(ring, hub->ctx0_ptb_addr_lo32 + vmid * 2);
-	amdgpu_ring_write(ring, lower_32_bits(pd_addr));
-
-	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SRBM_WRITE) |
-			  SDMA_PKT_SRBM_WRITE_HEADER_BYTE_EN(0xf));
-	amdgpu_ring_write(ring, hub->ctx0_ptb_addr_hi32 + vmid * 2);
-	amdgpu_ring_write(ring, upper_32_bits(pd_addr));
-
-	/* flush TLB */
-	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SRBM_WRITE) |
-			  SDMA_PKT_SRBM_WRITE_HEADER_BYTE_EN(0xf));
-	amdgpu_ring_write(ring, hub->vm_inv_eng0_req + eng);
-	amdgpu_ring_write(ring, req);
+	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pasid, pd_addr);
 
 	/* wait for flush */
 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_POLL_REGMEM) |
@@ -1604,7 +1585,7 @@ static const struct amdgpu_ring_funcs sdma_v4_0_ring_funcs = {
 		6 + /* sdma_v4_0_ring_emit_hdp_flush */
 		3 + /* sdma_v4_0_ring_emit_hdp_invalidate */
 		6 + /* sdma_v4_0_ring_emit_pipeline_sync */
-		18 + /* sdma_v4_0_ring_emit_vm_flush */
+		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 + 6 + /* sdma_v4_0_ring_emit_vm_flush */
 		10 + 10 + 10, /* sdma_v4_0_ring_emit_fence x3 for user fence, vm fence */
 	.emit_ib_size = 7 + 6, /* sdma_v4_0_ring_emit_ib */
 	.emit_ib = sdma_v4_0_ring_emit_ib,

drivers/gpu/drm/amd/amdgpu/soc15.h

@@ -27,6 +27,8 @@
 #include "nbio_v6_1.h"
 #include "nbio_v7_0.h"
 
+#define SOC15_FLUSH_GPU_TLB_NUM_WREG	3
+
 extern const struct amd_ip_funcs soc15_common_ip_funcs;
 
 struct soc15_reg_golden {

drivers/gpu/drm/amd/amdgpu/uvd_v7_0.c

@@ -25,6 +25,7 @@
 #include <drm/drmP.h>
 #include "amdgpu.h"
 #include "amdgpu_uvd.h"
+#include "soc15.h"
 #include "soc15d.h"
 #include "soc15_common.h"
 #include "mmsch_v1_0.h"
@@ -1295,32 +1296,17 @@ static void uvd_v7_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
 					uint64_t pd_addr)
 {
 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
-	uint32_t req = ring->adev->gmc.gmc_funcs->get_invalidate_req(vmid);
-	uint64_t flags = AMDGPU_PTE_VALID;
 	unsigned eng = ring->vm_inv_eng;
 	uint32_t data0, data1, mask;
 
-	amdgpu_gmc_get_vm_pde(ring->adev, -1, &pd_addr, &flags);
-	pd_addr |= flags;
-
-	data0 = (hub->ctx0_ptb_addr_hi32 + vmid * 2) << 2;
-	data1 = upper_32_bits(pd_addr);
-	uvd_v7_0_ring_emit_wreg(ring, data0, data1);
-
-	data0 = (hub->ctx0_ptb_addr_lo32 + vmid * 2) << 2;
-	data1 = lower_32_bits(pd_addr);
-	uvd_v7_0_ring_emit_wreg(ring, data0, data1);
+	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pasid, pd_addr);
 
+	/* wait for reg writes */
 	data0 = (hub->ctx0_ptb_addr_lo32 + vmid * 2) << 2;
 	data1 = lower_32_bits(pd_addr);
 	mask = 0xffffffff;
 	uvd_v7_0_vm_reg_wait(ring, data0, data1, mask);
 
-	/* flush TLB */
-	data0 = (hub->vm_inv_eng0_req + eng) << 2;
-	data1 = req;
-	uvd_v7_0_ring_emit_wreg(ring, data0, data1);
-
 	/* wait for flush */
 	data0 = (hub->vm_inv_eng0_ack + eng) << 2;
 	data1 = 1 << vmid;
@@ -1348,31 +1334,16 @@ static void uvd_v7_0_enc_ring_emit_vm_flush(struct amdgpu_ring *ring,
 					    uint64_t pd_addr)
 {
 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
-	uint32_t req = ring->adev->gmc.gmc_funcs->get_invalidate_req(vmid);
-	uint64_t flags = AMDGPU_PTE_VALID;
 	unsigned eng = ring->vm_inv_eng;
 
-	amdgpu_gmc_get_vm_pde(ring->adev, -1, &pd_addr, &flags);
-	pd_addr |= flags;
-
-	amdgpu_ring_write(ring, HEVC_ENC_CMD_REG_WRITE);
-	amdgpu_ring_write(ring,	(hub->ctx0_ptb_addr_hi32 + vmid * 2) << 2);
-	amdgpu_ring_write(ring, upper_32_bits(pd_addr));
-
-	amdgpu_ring_write(ring, HEVC_ENC_CMD_REG_WRITE);
-	amdgpu_ring_write(ring,	(hub->ctx0_ptb_addr_lo32 + vmid * 2) << 2);
-	amdgpu_ring_write(ring, lower_32_bits(pd_addr));
+	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pasid, pd_addr);
 
+	/* wait for reg writes */
 	amdgpu_ring_write(ring, HEVC_ENC_CMD_REG_WAIT);
 	amdgpu_ring_write(ring,	(hub->ctx0_ptb_addr_lo32 + vmid * 2) << 2);
 	amdgpu_ring_write(ring, 0xffffffff);
 	amdgpu_ring_write(ring, lower_32_bits(pd_addr));
 
-	/* flush TLB */
-	amdgpu_ring_write(ring, HEVC_ENC_CMD_REG_WRITE);
-	amdgpu_ring_write(ring,	(hub->vm_inv_eng0_req + eng) << 2);
-	amdgpu_ring_write(ring, req);
-
 	/* wait for flush */
 	amdgpu_ring_write(ring, HEVC_ENC_CMD_REG_WAIT);
 	amdgpu_ring_write(ring, (hub->vm_inv_eng0_ack + eng) << 2);
@@ -1724,7 +1695,7 @@ static const struct amdgpu_ring_funcs uvd_v7_0_ring_vm_funcs = {
 	.emit_frame_size =
 		2 + /* uvd_v7_0_ring_emit_hdp_flush */
 		2 + /* uvd_v7_0_ring_emit_hdp_invalidate */
-		34 + /* uvd_v7_0_ring_emit_vm_flush */
+		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 + 16 + /* uvd_v7_0_ring_emit_vm_flush */
 		14 + 14, /* uvd_v7_0_ring_emit_fence x2 vm fence */
 	.emit_ib_size = 8, /* uvd_v7_0_ring_emit_ib */
 	.emit_ib = uvd_v7_0_ring_emit_ib,
@@ -1751,7 +1722,7 @@ static const struct amdgpu_ring_funcs uvd_v7_0_enc_ring_vm_funcs = {
 	.get_wptr = uvd_v7_0_enc_ring_get_wptr,
 	.set_wptr = uvd_v7_0_enc_ring_set_wptr,
 	.emit_frame_size =
-		17 + /* uvd_v7_0_enc_ring_emit_vm_flush */
+		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 + 8 + /* uvd_v7_0_enc_ring_emit_vm_flush */
 		5 + 5 + /* uvd_v7_0_enc_ring_emit_fence x2 vm fence */
 		1, /* uvd_v7_0_enc_ring_insert_end */
 	.emit_ib_size = 5, /* uvd_v7_0_enc_ring_emit_ib */

drivers/gpu/drm/amd/amdgpu/vce_v4_0.c

@@ -28,6 +28,7 @@
 #include <drm/drmP.h>
 #include "amdgpu.h"
 #include "amdgpu_vce.h"
+#include "soc15.h"
 #include "soc15d.h"
 #include "soc15_common.h"
 #include "mmsch_v1_0.h"
@@ -969,31 +970,16 @@ static void vce_v4_0_emit_vm_flush(struct amdgpu_ring *ring,
 				   uint64_t pd_addr)
 {
 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
-	uint32_t req = ring->adev->gmc.gmc_funcs->get_invalidate_req(vmid);
-	uint64_t flags = AMDGPU_PTE_VALID;
 	unsigned eng = ring->vm_inv_eng;
 
-	amdgpu_gmc_get_vm_pde(ring->adev, -1, &pd_addr, &flags);
-	pd_addr |= flags;
-
-	amdgpu_ring_write(ring, VCE_CMD_REG_WRITE);
-	amdgpu_ring_write(ring,	(hub->ctx0_ptb_addr_hi32 + vmid * 2) << 2);
-	amdgpu_ring_write(ring, upper_32_bits(pd_addr));
-
-	amdgpu_ring_write(ring, VCE_CMD_REG_WRITE);
-	amdgpu_ring_write(ring,	(hub->ctx0_ptb_addr_lo32 + vmid * 2) << 2);
-	amdgpu_ring_write(ring, lower_32_bits(pd_addr));
+	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pasid, pd_addr);
 
+	/* wait for reg writes */
 	amdgpu_ring_write(ring, VCE_CMD_REG_WAIT);
 	amdgpu_ring_write(ring,	(hub->ctx0_ptb_addr_lo32 + vmid * 2) << 2);
 	amdgpu_ring_write(ring, 0xffffffff);
 	amdgpu_ring_write(ring, lower_32_bits(pd_addr));
 
-	/* flush TLB */
-	amdgpu_ring_write(ring, VCE_CMD_REG_WRITE);
-	amdgpu_ring_write(ring,	(hub->vm_inv_eng0_req + eng) << 2);
-	amdgpu_ring_write(ring, req);
-
 	/* wait for flush */
 	amdgpu_ring_write(ring, VCE_CMD_REG_WAIT);
 	amdgpu_ring_write(ring, (hub->vm_inv_eng0_ack + eng) << 2);
@@ -1078,7 +1064,7 @@ static const struct amdgpu_ring_funcs vce_v4_0_ring_vm_funcs = {
 	.set_wptr = vce_v4_0_ring_set_wptr,
 	.parse_cs = amdgpu_vce_ring_parse_cs_vm,
 	.emit_frame_size =
-		17 + /* vce_v4_0_emit_vm_flush */
+		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 + 8 + /* vce_v4_0_emit_vm_flush */
 		5 + 5 + /* amdgpu_vce_ring_emit_fence x2 vm fence */
 		1, /* vce_v4_0_ring_insert_end */
 	.emit_ib_size = 5, /* vce_v4_0_ring_emit_ib */

drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c

@@ -25,6 +25,7 @@
 #include <drm/drmP.h>
 #include "amdgpu.h"
 #include "amdgpu_vcn.h"
+#include "soc15.h"
 #include "soc15d.h"
 #include "soc15_common.h"
 
@@ -852,22 +853,6 @@ static void vcn_v1_0_dec_ring_emit_ib(struct amdgpu_ring *ring,
 	amdgpu_ring_write(ring, ib->length_dw);
 }
 
-static void vcn_v1_0_dec_vm_reg_write(struct amdgpu_ring *ring,
-				uint32_t data0, uint32_t data1)
-{
-	struct amdgpu_device *adev = ring->adev;
-
-	amdgpu_ring_write(ring,
-		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0));
-	amdgpu_ring_write(ring, data0);
-	amdgpu_ring_write(ring,
-		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1), 0));
-	amdgpu_ring_write(ring, data1);
-	amdgpu_ring_write(ring,
-		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0));
-	amdgpu_ring_write(ring, VCN_DEC_CMD_WRITE_REG << 1);
-}
-
 static void vcn_v1_0_dec_vm_reg_wait(struct amdgpu_ring *ring,
 				uint32_t data0, uint32_t data1, uint32_t mask)
 {
@@ -892,32 +877,17 @@ static void vcn_v1_0_dec_ring_emit_vm_flush(struct amdgpu_ring *ring,
 					    uint64_t pd_addr)
 {
 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
-	uint32_t req = ring->adev->gmc.gmc_funcs->get_invalidate_req(vmid);
-	uint64_t flags = AMDGPU_PTE_VALID;
 	unsigned eng = ring->vm_inv_eng;
 	uint32_t data0, data1, mask;
 
-	amdgpu_gmc_get_vm_pde(ring->adev, -1, &pd_addr, &flags);
-	pd_addr |= flags;
-
-	data0 = (hub->ctx0_ptb_addr_hi32 + vmid * 2) << 2;
-	data1 = upper_32_bits(pd_addr);
-	vcn_v1_0_dec_vm_reg_write(ring, data0, data1);
-
-	data0 = (hub->ctx0_ptb_addr_lo32 + vmid * 2) << 2;
-	data1 = lower_32_bits(pd_addr);
-	vcn_v1_0_dec_vm_reg_write(ring, data0, data1);
+	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pasid, pd_addr);
 
+	/* wait for register write */
 	data0 = (hub->ctx0_ptb_addr_lo32 + vmid * 2) << 2;
 	data1 = lower_32_bits(pd_addr);
 	mask = 0xffffffff;
 	vcn_v1_0_dec_vm_reg_wait(ring, data0, data1, mask);
 
-	/* flush TLB */
-	data0 = (hub->vm_inv_eng0_req + eng) << 2;
-	data1 = req;
-	vcn_v1_0_dec_vm_reg_write(ring, data0, data1);
-
 	/* wait for flush */
 	data0 = (hub->vm_inv_eng0_ack + eng) << 2;
 	data1 = 1 << vmid;
@@ -1026,34 +996,17 @@ static void vcn_v1_0_enc_ring_emit_vm_flush(struct amdgpu_ring *ring,
 					    uint64_t pd_addr)
 {
 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
-	uint32_t req = ring->adev->gmc.gmc_funcs->get_invalidate_req(vmid);
-	uint64_t flags = AMDGPU_PTE_VALID;
 	unsigned eng = ring->vm_inv_eng;
 
-	amdgpu_gmc_get_vm_pde(ring->adev, -1, &pd_addr, &flags);
-	pd_addr |= flags;
-
-	amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WRITE);
-	amdgpu_ring_write(ring,
-			  (hub->ctx0_ptb_addr_hi32 + vmid * 2) << 2);
-	amdgpu_ring_write(ring, upper_32_bits(pd_addr));
-
-	amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WRITE);
-	amdgpu_ring_write(ring,
-			  (hub->ctx0_ptb_addr_lo32 + vmid * 2) << 2);
-	amdgpu_ring_write(ring, lower_32_bits(pd_addr));
+	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pasid, pd_addr);
 
+	/* wait for reg writes */
 	amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WAIT);
 	amdgpu_ring_write(ring,
 			  (hub->ctx0_ptb_addr_lo32 + vmid * 2) << 2);
 	amdgpu_ring_write(ring, 0xffffffff);
 	amdgpu_ring_write(ring, lower_32_bits(pd_addr));
 
-	/* flush TLB */
-	amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WRITE);
-	amdgpu_ring_write(ring,	(hub->vm_inv_eng0_req + eng) << 2);
-	amdgpu_ring_write(ring, req);
-
 	/* wait for flush */
 	amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WAIT);
 	amdgpu_ring_write(ring, (hub->vm_inv_eng0_ack + eng) << 2);
@@ -1144,7 +1097,7 @@ static const struct amdgpu_ring_funcs vcn_v1_0_dec_ring_vm_funcs = {
 	.set_wptr = vcn_v1_0_dec_ring_set_wptr,
 	.emit_frame_size =
 		2 + /* vcn_v1_0_dec_ring_emit_hdp_invalidate */
-		34 + /* vcn_v1_0_dec_ring_emit_vm_flush */
+		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 + 16 + /* vcn_v1_0_dec_ring_emit_vm_flush */
 		14 + 14 + /* vcn_v1_0_dec_ring_emit_fence x2 vm fence */
 		6,
 	.emit_ib_size = 8, /* vcn_v1_0_dec_ring_emit_ib */
@@ -1173,7 +1126,7 @@ static const struct amdgpu_ring_funcs vcn_v1_0_enc_ring_vm_funcs = {
 	.get_wptr = vcn_v1_0_enc_ring_get_wptr,
 	.set_wptr = vcn_v1_0_enc_ring_set_wptr,
 	.emit_frame_size =
-		17 + /* vcn_v1_0_enc_ring_emit_vm_flush */
+		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 + 8 + /* vcn_v1_0_enc_ring_emit_vm_flush */
 		5 + 5 + /* vcn_v1_0_enc_ring_emit_fence x2 vm fence */
 		1, /* vcn_v1_0_enc_ring_insert_end */
 	.emit_ib_size = 5, /* vcn_v1_0_enc_ring_emit_ib */