drm/amdgpu: implement gmc_v7_0_emit_flush_gpu_tlb

Unify tlb flushing for gmc v7.
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/cik.h

@@ -24,6 +24,8 @@
 #ifndef __CIK_H__
 #define __CIK_H__
 
+#define CIK_FLUSH_GPU_TLB_NUM_WREG	2
+
 void cik_srbm_select(struct amdgpu_device *adev,
 		     u32 me, u32 pipe, u32 queue, u32 vmid);
 int cik_set_ip_blocks(struct amdgpu_device *adev);

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

@@ -886,18 +886,7 @@ static void cik_sdma_ring_emit_vm_flush(struct amdgpu_ring *ring,
 	u32 extra_bits = (SDMA_POLL_REG_MEM_EXTRA_OP(0) |
 			  SDMA_POLL_REG_MEM_EXTRA_FUNC(0)); /* always */
 
-	amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
-	if (vmid < 8) {
-		amdgpu_ring_write(ring, (mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + vmid));
-	} else {
-		amdgpu_ring_write(ring, (mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + vmid - 8));
-	}
-	amdgpu_ring_write(ring, pd_addr >> 12);
-
-	/* flush TLB */
-	amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
-	amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST);
-	amdgpu_ring_write(ring, 1 << vmid);
+	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pasid, pd_addr);
 
 	amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_POLL_REG_MEM, 0, extra_bits));
 	amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST << 2);
@@ -1290,7 +1279,7 @@ static const struct amdgpu_ring_funcs cik_sdma_ring_funcs = {
 		6 + /* cik_sdma_ring_emit_hdp_flush */
 		3 + /* cik_sdma_ring_emit_hdp_invalidate */
 		6 + /* cik_sdma_ring_emit_pipeline_sync */
-		12 + /* cik_sdma_ring_emit_vm_flush */
+		CIK_FLUSH_GPU_TLB_NUM_WREG * 3 + 6 + /* cik_sdma_ring_emit_vm_flush */
 		9 + 9 + 9, /* cik_sdma_ring_emit_fence x3 for user fence, vm fence */
 	.emit_ib_size = 7 + 4, /* cik_sdma_ring_emit_ib */
 	.emit_ib = cik_sdma_ring_emit_ib,

drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c

@@ -3244,26 +3244,7 @@ static void gfx_v7_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
 {
 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
 
-	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
-	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |
-				 WRITE_DATA_DST_SEL(0)));
-	if (vmid < 8) {
-		amdgpu_ring_write(ring,
-				  (mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + vmid));
-	} else {
-		amdgpu_ring_write(ring,
-				  (mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + vmid - 8));
-	}
-	amdgpu_ring_write(ring, 0);
-	amdgpu_ring_write(ring, pd_addr >> 12);
-
-	/* bits 0-15 are the VM contexts0-15 */
-	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
-	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
-				 WRITE_DATA_DST_SEL(0)));
-	amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST);
-	amdgpu_ring_write(ring, 0);
-	amdgpu_ring_write(ring, 1 << vmid);
+	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pasid, pd_addr);
 
 	/* wait for the invalidate to complete */
 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
@@ -5132,7 +5113,7 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = {
 		5 + /* gfx_v7_0_ring_emit_hdp_invalidate */
 		12 + 12 + 12 + /* gfx_v7_0_ring_emit_fence_gfx x3 for user fence, vm fence */
 		7 + 4 + /* gfx_v7_0_ring_emit_pipeline_sync */
-		17 + 6 + /* gfx_v7_0_ring_emit_vm_flush */
+		CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + 6 + /* gfx_v7_0_ring_emit_vm_flush */
 		3 + 4, /* gfx_v7_ring_emit_cntxcntl including vgt flush*/
 	.emit_ib_size = 4, /* gfx_v7_0_ring_emit_ib_gfx */
 	.emit_ib = gfx_v7_0_ring_emit_ib_gfx,
@@ -5163,7 +5144,7 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = {
 		7 + /* gfx_v7_0_ring_emit_hdp_flush */
 		5 + /* gfx_v7_0_ring_emit_hdp_invalidate */
 		7 + /* gfx_v7_0_ring_emit_pipeline_sync */
-		17 + /* gfx_v7_0_ring_emit_vm_flush */
+		CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v7_0_ring_emit_vm_flush */
 		7 + 7 + 7, /* gfx_v7_0_ring_emit_fence_compute x3 for user fence, vm fence */
 	.emit_ib_size =	4, /* gfx_v7_0_ring_emit_ib_compute */
 	.emit_ib = gfx_v7_0_ring_emit_ib_compute,

drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c

@@ -435,6 +435,24 @@ static void gmc_v7_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid)
 	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
 }
 
+static uint64_t gmc_v7_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
+					    unsigned vmid, unsigned pasid,
+					    uint64_t pd_addr)
+{
+	uint32_t reg;
+
+	if (vmid < 8)
+		reg = mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + vmid;
+	else
+		reg = mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + vmid - 8;
+	amdgpu_ring_emit_wreg(ring, reg, pd_addr >> 12);
+
+	/* bits 0-15 are the VM contexts0-15 */
+	amdgpu_ring_emit_wreg(ring, mmVM_INVALIDATE_REQUEST, 1 << vmid);
+
+	return pd_addr;
+}
+
 /**
  * gmc_v7_0_set_pte_pde - update the page tables using MMIO
  *
@@ -1305,6 +1323,7 @@ static const struct amd_ip_funcs gmc_v7_0_ip_funcs = {
 
 static const struct amdgpu_gmc_funcs gmc_v7_0_gmc_funcs = {
 	.flush_gpu_tlb = gmc_v7_0_flush_gpu_tlb,
+	.emit_flush_gpu_tlb = gmc_v7_0_emit_flush_gpu_tlb,
 	.set_pte_pde = gmc_v7_0_set_pte_pde,
 	.set_prt = gmc_v7_0_set_prt,
 	.get_vm_pte_flags = gmc_v7_0_get_vm_pte_flags,