drm/amdgpu: move waiting for VM flush into gmc_v9_0_emit_flush_gpu_tlb

Keep that at a common place instead of spread over all engines.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

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

@@ -3679,15 +3679,8 @@ static void gfx_v9_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
 					unsigned vmid, unsigned pasid,
 					uint64_t pd_addr)
 {
-	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
-	unsigned eng = ring->vm_inv_eng;
-
 	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);
-
 	/* compute doesn't have PFP */
 	if (ring->funcs->type == AMDGPU_RING_TYPE_GFX) {
 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
@@ -4295,7 +4288,9 @@ 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 */
-		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 + 9 + /* VM_FLUSH */
+		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
+		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
+		2 + /* VM_FLUSH */
 		8 +  /* FENCE for VM_FLUSH */
 		20 + /* GDS switch */
 		4 + /* double SWITCH_BUFFER,
@@ -4344,7 +4339,9 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = {
 		7 + /* gfx_v9_0_ring_emit_hdp_flush */
 		5 + /* hdp invalidate */
 		7 + /* gfx_v9_0_ring_emit_pipeline_sync */
-		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 + 9 + /* gfx_v9_0_ring_emit_vm_flush */
+		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
+		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
+		2 + /* 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,
@@ -4376,7 +4373,9 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_kiq = {
 		7 + /* gfx_v9_0_ring_emit_hdp_flush */
 		5 + /* hdp invalidate */
 		7 + /* gfx_v9_0_ring_emit_pipeline_sync */
-		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 + 9 + /* gfx_v9_0_ring_emit_vm_flush */
+		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
+		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
+		2 + /* 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

@@ -396,6 +396,10 @@ static uint64_t gmc_v9_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
 
 	amdgpu_ring_emit_wreg(ring, hub->vm_inv_eng0_req + eng, req);
 
+	/* wait for the invalidate to complete */
+	amdgpu_ring_emit_reg_wait(ring, hub->vm_inv_eng0_ack + eng,
+				  1 << vmid, 1 << vmid);
+
 	return pd_addr;
 }
 

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

@@ -1126,21 +1126,7 @@ static void sdma_v4_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
 					 unsigned vmid, unsigned pasid,
 					 uint64_t pd_addr)
 {
-	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
-	unsigned eng = ring->vm_inv_eng;
-
 	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) |
-			  SDMA_PKT_POLL_REGMEM_HEADER_HDP_FLUSH(0) |
-			  SDMA_PKT_POLL_REGMEM_HEADER_FUNC(3)); /* equal */
-	amdgpu_ring_write(ring, (hub->vm_inv_eng0_ack + eng) << 2);
-	amdgpu_ring_write(ring, 0);
-	amdgpu_ring_write(ring, 1 << vmid); /* reference */
-	amdgpu_ring_write(ring, 1 << vmid); /* mask */
-	amdgpu_ring_write(ring, SDMA_PKT_POLL_REGMEM_DW5_RETRY_COUNT(0xfff) |
-			  SDMA_PKT_POLL_REGMEM_DW5_INTERVAL(10));
 }
 
 static void sdma_v4_0_ring_emit_wreg(struct amdgpu_ring *ring,
@@ -1589,7 +1575,9 @@ static const struct amdgpu_ring_funcs sdma_v4_0_ring_funcs = {
 		6 + /* sdma_v4_0_ring_emit_hdp_flush */
 		3 + /* hdp invalidate */
 		6 + /* sdma_v4_0_ring_emit_pipeline_sync */
-		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 + 6 + /* sdma_v4_0_ring_emit_vm_flush */
+		/* sdma_v4_0_ring_emit_vm_flush */
+		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
+		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 6 +
 		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

@@ -28,6 +28,7 @@
 #include "nbio_v7_0.h"
 
 #define SOC15_FLUSH_GPU_TLB_NUM_WREG		4
+#define SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT	1
 
 extern const struct amd_ip_funcs soc15_common_ip_funcs;
 

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

@@ -1265,7 +1265,6 @@ 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];
-	unsigned eng = ring->vm_inv_eng;
 	uint32_t data0, data1, mask;
 
 	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pasid, pd_addr);
@@ -1275,12 +1274,6 @@ static void uvd_v7_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
 	data1 = lower_32_bits(pd_addr);
 	mask = 0xffffffff;
 	uvd_v7_0_ring_emit_reg_wait(ring, data0, data1, mask);
-
-	/* wait for flush */
-	data0 = hub->vm_inv_eng0_ack + eng;
-	data1 = 1 << vmid;
-	mask =  1 << vmid;
-	uvd_v7_0_ring_emit_reg_wait(ring, data0, data1, mask);
 }
 
 static void uvd_v7_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
@@ -1313,17 +1306,12 @@ 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];
-	unsigned eng = ring->vm_inv_eng;
 
 	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pasid, pd_addr);
 
 	/* wait for reg writes */
 	uvd_v7_0_enc_ring_emit_reg_wait(ring, hub->ctx0_ptb_addr_lo32 + vmid * 2,
 					lower_32_bits(pd_addr), 0xffffffff);
-
-	/* wait for flush */
-	uvd_v7_0_enc_ring_emit_reg_wait(ring, hub->vm_inv_eng0_ack + eng,
-					1 << vmid, 1 << vmid);
 }
 
 static void uvd_v7_0_enc_ring_emit_wreg(struct amdgpu_ring *ring,
@@ -1669,7 +1657,9 @@ static const struct amdgpu_ring_funcs uvd_v7_0_ring_vm_funcs = {
 	.set_wptr = uvd_v7_0_ring_set_wptr,
 	.emit_frame_size =
 		6 + 6 + /* hdp flush / invalidate */
-		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 + 16 + /* uvd_v7_0_ring_emit_vm_flush */
+		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
+		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
+		8 + /* 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,
@@ -1696,7 +1686,9 @@ static const struct amdgpu_ring_funcs uvd_v7_0_enc_ring_vm_funcs = {
 	.set_wptr = uvd_v7_0_enc_ring_set_wptr,
 	.emit_frame_size =
 		3 + 3 + /* hdp flush / invalidate */
-		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 + 8 + /* uvd_v7_0_enc_ring_emit_vm_flush */
+		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
+		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
+		4 + /* 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

@@ -979,17 +979,12 @@ 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];
-	unsigned eng = ring->vm_inv_eng;
 
 	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pasid, pd_addr);
 
 	/* wait for reg writes */
 	vce_v4_0_emit_reg_wait(ring, hub->ctx0_ptb_addr_lo32 + vmid * 2,
 			       lower_32_bits(pd_addr), 0xffffffff);
-
-	/* wait for flush */
-	vce_v4_0_emit_reg_wait(ring, hub->vm_inv_eng0_ack + eng,
-			       1 << vmid, 1 << vmid);
 }
 
 static void vce_v4_0_emit_wreg(struct amdgpu_ring *ring,
@@ -1069,7 +1064,9 @@ 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 =
-		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 + 8 + /* vce_v4_0_emit_vm_flush */
+		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
+		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
+		4 + /* 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

@@ -863,7 +863,6 @@ 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];
-	unsigned eng = ring->vm_inv_eng;
 	uint32_t data0, data1, mask;
 
 	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pasid, pd_addr);
@@ -873,12 +872,6 @@ static void vcn_v1_0_dec_ring_emit_vm_flush(struct amdgpu_ring *ring,
 	data1 = lower_32_bits(pd_addr);
 	mask = 0xffffffff;
 	vcn_v1_0_dec_ring_emit_reg_wait(ring, data0, data1, mask);
-
-	/* wait for flush */
-	data0 = hub->vm_inv_eng0_ack + eng;
-	data1 = 1 << vmid;
-	mask =  1 << vmid;
-	vcn_v1_0_dec_ring_emit_reg_wait(ring, data0, data1, mask);
 }
 
 static void vcn_v1_0_dec_ring_emit_wreg(struct amdgpu_ring *ring,
@@ -1008,17 +1001,12 @@ 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];
-	unsigned eng = ring->vm_inv_eng;
 
 	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pasid, pd_addr);
 
 	/* wait for reg writes */
 	vcn_v1_0_enc_ring_emit_reg_wait(ring, hub->ctx0_ptb_addr_lo32 + vmid * 2,
 					lower_32_bits(pd_addr), 0xffffffff);
-
-	/* wait for flush */
-	vcn_v1_0_enc_ring_emit_reg_wait(ring, hub->vm_inv_eng0_ack + eng,
-					1 << vmid, 1 << vmid);
 }
 
 static void vcn_v1_0_enc_ring_emit_wreg(struct amdgpu_ring *ring,
@@ -1104,7 +1092,9 @@ 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 =
 		6 + 6 + /* hdp invalidate / flush */
-		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 + 16 + /* vcn_v1_0_dec_ring_emit_vm_flush */
+		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
+		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
+		8 + /* 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 */
@@ -1133,7 +1123,9 @@ 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 =
-		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 + 8 + /* vcn_v1_0_enc_ring_emit_vm_flush */
+		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
+		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
+		4 + /* 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 */