drm/amdgpu: implement gmc_v6_0_emit_flush_gpu_tlb

Unify tlb flushing for gmc v6.
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/gfx_v6_0.c

@@ -38,6 +38,7 @@
 #include "dce/dce_6_0_sh_mask.h"
 #include "gca/gfx_7_2_enum.h"
 #include "si_enums.h"
+#include "si.h"
 
 static void gfx_v6_0_set_ring_funcs(struct amdgpu_device *adev);
 static void gfx_v6_0_set_irq_funcs(struct amdgpu_device *adev);
@@ -2359,25 +2360,7 @@ static void gfx_v6_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
 {
 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
 
-	/* write new base address */
-	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
-	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) |
-				 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(1) |
-				 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));
@@ -3528,7 +3511,7 @@ static const struct amdgpu_ring_funcs gfx_v6_0_ring_funcs_gfx = {
 		5 + /* gfx_v6_0_ring_emit_hdp_invalidate */
 		14 + 14 + 14 + /* gfx_v6_0_ring_emit_fence x3 for user fence, vm fence */
 		7 + 4 + /* gfx_v6_0_ring_emit_pipeline_sync */
-		17 + 6 + /* gfx_v6_0_ring_emit_vm_flush */
+		SI_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + 6 + /* gfx_v6_0_ring_emit_vm_flush */
 		3 + 2, /* gfx_v6_ring_emit_cntxcntl including vgt flush */
 	.emit_ib_size = 6, /* gfx_v6_0_ring_emit_ib */
 	.emit_ib = gfx_v6_0_ring_emit_ib,
@@ -3555,7 +3538,7 @@ static const struct amdgpu_ring_funcs gfx_v6_0_ring_funcs_compute = {
 		5 + /* gfx_v6_0_ring_emit_hdp_flush */
 		5 + /* gfx_v6_0_ring_emit_hdp_invalidate */
 		7 + /* gfx_v6_0_ring_emit_pipeline_sync */
-		17 + /* gfx_v6_0_ring_emit_vm_flush */
+		SI_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v6_0_ring_emit_vm_flush */
 		14 + 14 + 14, /* gfx_v6_0_ring_emit_fence x3 for user fence, vm fence */
 	.emit_ib_size = 6, /* gfx_v6_0_ring_emit_ib */
 	.emit_ib = gfx_v6_0_ring_emit_ib,

drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c

@@ -362,6 +362,25 @@ static void gmc_v6_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid)
 	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
 }
 
+static uint64_t gmc_v6_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
+					    unsigned vmid, unsigned pasid,
+					    uint64_t pd_addr)
+{
+	uint32_t reg;
+
+	/* write new base address */
+	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;
+}
+
 static int gmc_v6_0_set_pte_pde(struct amdgpu_device *adev, void *cpu_pt_addr,
 				uint32_t gpu_page_idx, uint64_t addr,
 				uint64_t flags)
@@ -1126,6 +1145,7 @@ static const struct amd_ip_funcs gmc_v6_0_ip_funcs = {
 
 static const struct amdgpu_gmc_funcs gmc_v6_0_gmc_funcs = {
 	.flush_gpu_tlb = gmc_v6_0_flush_gpu_tlb,
+	.emit_flush_gpu_tlb = gmc_v6_0_emit_flush_gpu_tlb,
 	.set_pte_pde = gmc_v6_0_set_pte_pde,
 	.set_prt = gmc_v6_0_set_prt,
 	.get_vm_pde = gmc_v6_0_get_vm_pde,

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

@@ -24,6 +24,8 @@
 #ifndef __SI_H__
 #define __SI_H__
 
+#define SI_FLUSH_GPU_TLB_NUM_WREG	2
+
 void si_srbm_select(struct amdgpu_device *adev,
 		     u32 me, u32 pipe, u32 queue, u32 vmid);
 int si_set_ip_blocks(struct amdgpu_device *adev);

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

@@ -24,6 +24,7 @@
 #include <drm/drmP.h>
 #include "amdgpu.h"
 #include "amdgpu_trace.h"
+#include "si.h"
 #include "sid.h"
 
 const u32 sdma_offsets[SDMA_MAX_INSTANCE] =
@@ -476,17 +477,7 @@ static void si_dma_ring_emit_vm_flush(struct amdgpu_ring *ring,
 				      unsigned vmid, unsigned pasid,
 				      uint64_t pd_addr)
 {
-	amdgpu_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0));
-	if (vmid < 8)
-		amdgpu_ring_write(ring, (0xf << 16) | (VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + vmid));
-	else
-		amdgpu_ring_write(ring, (0xf << 16) | (VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + (vmid - 8)));
-	amdgpu_ring_write(ring, pd_addr >> 12);
-
-	/* bits 0-7 are the VM contexts0-7 */
-	amdgpu_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0));
-	amdgpu_ring_write(ring, (0xf << 16) | (VM_INVALIDATE_REQUEST));
-	amdgpu_ring_write(ring, 1 << vmid);
+	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pasid, pd_addr);
 
 	/* wait for invalidate to complete */
 	amdgpu_ring_write(ring, DMA_PACKET(DMA_PACKET_POLL_REG_MEM, 0, 0, 0, 0));
@@ -784,7 +775,7 @@ static const struct amdgpu_ring_funcs si_dma_ring_funcs = {
 		3 + /* si_dma_ring_emit_hdp_flush */
 		3 + /* si_dma_ring_emit_hdp_invalidate */
 		6 + /* si_dma_ring_emit_pipeline_sync */
-		12 + /* si_dma_ring_emit_vm_flush */
+		SI_FLUSH_GPU_TLB_NUM_WREG * 3 + 6 + /* si_dma_ring_emit_vm_flush */
 		9 + 9 + 9, /* si_dma_ring_emit_fence x3 for user fence, vm fence */
 	.emit_ib_size = 7 + 3, /* si_dma_ring_emit_ib */
 	.emit_ib = si_dma_ring_emit_ib,