drm/amdgpu: sync ce and me with SWITCH_BUFFER(2)

we used to adopt wait_reg_mem to let CE wait before DE finish page
updating, but from Tonga+, CE doesn't support wait_reg_mem package so
this logic no longer works.

so here is another approach to do same thing:
Insert two of SWITCH_BUFFER at both front and end of vm_flush can
guarantee that CE not go further to process IB_const before vm_flush
done.

Insert two of SWITCH_BUFFER also works on CI, so remove legency method
to sync CE and ME

v2:
Insert double SWITCH_BUFFER at front of vm flush as well.
Signed-off-by: monk.liu <monk.liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

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

@@ -1202,8 +1202,6 @@ struct amdgpu_gfx {
 	struct amdgpu_irq_src		priv_inst_irq;
 	/* gfx status */
 	uint32_t gfx_current_status;
-	/* sync signal for const engine */
-	unsigned ce_sync_offs;
 	/* ce ram size*/
 	unsigned ce_ram_size;
 };

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

@@ -3610,41 +3610,6 @@ static int gfx_v7_0_cp_resume(struct amdgpu_device *adev)
 	return 0;
 }
 
-static void gfx_v7_0_ce_sync_me(struct amdgpu_ring *ring)
-{
-	struct amdgpu_device *adev = ring->adev;
-	u64 gpu_addr = adev->wb.gpu_addr + adev->gfx.ce_sync_offs * 4;
-
-	/* instruct DE to set a magic number */
-	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
-	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
-							 WRITE_DATA_DST_SEL(5)));
-	amdgpu_ring_write(ring, gpu_addr & 0xfffffffc);
-	amdgpu_ring_write(ring, upper_32_bits(gpu_addr) & 0xffffffff);
-	amdgpu_ring_write(ring, 1);
-
-	/* let CE wait till condition satisfied */
-	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
-	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */
-							 WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
-							 WAIT_REG_MEM_FUNCTION(3) |  /* == */
-							 WAIT_REG_MEM_ENGINE(2)));   /* ce */
-	amdgpu_ring_write(ring, gpu_addr & 0xfffffffc);
-	amdgpu_ring_write(ring, upper_32_bits(gpu_addr) & 0xffffffff);
-	amdgpu_ring_write(ring, 1);
-	amdgpu_ring_write(ring, 0xffffffff);
-	amdgpu_ring_write(ring, 4); /* poll interval */
-
-	/* instruct CE to reset wb of ce_sync to zero */
-	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
-	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(2) |
-							 WRITE_DATA_DST_SEL(5) |
-							 WR_CONFIRM));
-	amdgpu_ring_write(ring, gpu_addr & 0xfffffffc);
-	amdgpu_ring_write(ring, upper_32_bits(gpu_addr) & 0xffffffff);
-	amdgpu_ring_write(ring, 0);
-}
-
 /*
  * vm
  * VMID 0 is the physical GPU addresses as used by the kernel.
@@ -3663,6 +3628,13 @@ static void gfx_v7_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
 					unsigned vm_id, uint64_t pd_addr)
 {
 	int usepfp = (ring->type == AMDGPU_RING_TYPE_GFX);
+	if (usepfp) {
+		/* synce CE with ME to prevent CE fetch CEIB before context switch done */
+		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
+		amdgpu_ring_write(ring, 0);
+		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
+		amdgpu_ring_write(ring, 0);
+	}
 
 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |
@@ -3703,7 +3675,10 @@ static void gfx_v7_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
 		amdgpu_ring_write(ring, 0x0);
 
 		/* synce CE with ME to prevent CE fetch CEIB before context switch done */
-		gfx_v7_0_ce_sync_me(ring);
+		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
+		amdgpu_ring_write(ring, 0);
+		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
+		amdgpu_ring_write(ring, 0);
 	}
 }
 
@@ -4805,12 +4780,6 @@ static int gfx_v7_0_sw_init(void *handle)
 		return r;
 	}
 
-	r = amdgpu_wb_get(adev, &adev->gfx.ce_sync_offs);
-	if (r) {
-		DRM_ERROR("(%d) gfx.ce_sync_offs wb alloc failed\n", r);
-		return r;
-	}
-
 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
 		ring = &adev->gfx.gfx_ring[i];
 		ring->ring_obj = NULL;
@@ -4889,8 +4858,6 @@ static int gfx_v7_0_sw_fini(void *handle)
 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
 		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
 
-	amdgpu_wb_free(adev, adev->gfx.ce_sync_offs);
-
 	gfx_v7_0_cp_compute_fini(adev);
 	gfx_v7_0_rlc_fini(adev);
 	gfx_v7_0_mec_fini(adev);

drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c

@@ -940,12 +940,6 @@ static int gfx_v8_0_sw_init(void *handle)
 		return r;
 	}
 
-	r = amdgpu_wb_get(adev, &adev->gfx.ce_sync_offs);
-	if (r) {
-		DRM_ERROR("(%d) gfx.ce_sync_offs wb alloc failed\n", r);
-		return r;
-	}
-
 	/* set up the gfx ring */
 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
 		ring = &adev->gfx.gfx_ring[i];
@@ -1033,8 +1027,6 @@ static int gfx_v8_0_sw_fini(void *handle)
 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
 		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
 
-	amdgpu_wb_free(adev, adev->gfx.ce_sync_offs);
-
 	gfx_v8_0_mec_fini(adev);
 
 	return 0;
@@ -4006,41 +3998,6 @@ static bool gfx_v8_0_ring_emit_semaphore(struct amdgpu_ring *ring,
 	return true;
 }
 
-static void gfx_v8_0_ce_sync_me(struct amdgpu_ring *ring)
-{
-	struct amdgpu_device *adev = ring->adev;
-	u64 gpu_addr = adev->wb.gpu_addr + adev->gfx.ce_sync_offs * 4;
-
-	/* instruct DE to set a magic number */
-	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
-	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
-							 WRITE_DATA_DST_SEL(5)));
-	amdgpu_ring_write(ring, gpu_addr & 0xfffffffc);
-	amdgpu_ring_write(ring, upper_32_bits(gpu_addr) & 0xffffffff);
-	amdgpu_ring_write(ring, 1);
-
-	/* let CE wait till condition satisfied */
-	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
-	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */
-							 WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
-							 WAIT_REG_MEM_FUNCTION(3) |  /* == */
-							 WAIT_REG_MEM_ENGINE(2)));   /* ce */
-	amdgpu_ring_write(ring, gpu_addr & 0xfffffffc);
-	amdgpu_ring_write(ring, upper_32_bits(gpu_addr) & 0xffffffff);
-	amdgpu_ring_write(ring, 1);
-	amdgpu_ring_write(ring, 0xffffffff);
-	amdgpu_ring_write(ring, 4); /* poll interval */
-
-	/* instruct CE to reset wb of ce_sync to zero */
-	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
-	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(2) |
-							 WRITE_DATA_DST_SEL(5) |
-							 WR_CONFIRM));
-	amdgpu_ring_write(ring, gpu_addr & 0xfffffffc);
-	amdgpu_ring_write(ring, upper_32_bits(gpu_addr) & 0xffffffff);
-	amdgpu_ring_write(ring, 0);
-}
-
 static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
 					unsigned vm_id, uint64_t pd_addr)
 {
@@ -4057,6 +4014,14 @@ static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
 	amdgpu_ring_write(ring, 0xffffffff);
 	amdgpu_ring_write(ring, 4); /* poll interval */
 
+	if (usepfp) {
+		/* synce CE with ME to prevent CE fetch CEIB before context switch done */
+		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
+		amdgpu_ring_write(ring, 0);
+		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
+		amdgpu_ring_write(ring, 0);
+	}
+
 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |
 				 WRITE_DATA_DST_SEL(0)) |
@@ -4096,9 +4061,10 @@ static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
 		/* 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);
-
-		/* synce CE with ME to prevent CE fetch CEIB before context switch done */
-		gfx_v8_0_ce_sync_me(ring);
+		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
+		amdgpu_ring_write(ring, 0);
+		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
+		amdgpu_ring_write(ring, 0);
 	}
 }