drm/amdgpu: split pipeline sync and vm flush

This allows us to use the pipeline sync for other tasks as well.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

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

@@ -326,6 +326,7 @@ struct amdgpu_ring_funcs {
 			struct amdgpu_ib *ib);
 	void (*emit_fence)(struct amdgpu_ring *ring, uint64_t addr,
 			   uint64_t seq, unsigned flags);
+	void (*emit_pipeline_sync)(struct amdgpu_ring *ring);
 	void (*emit_vm_flush)(struct amdgpu_ring *ring, unsigned vm_id,
 			      uint64_t pd_addr);
 	void (*emit_hdp_flush)(struct amdgpu_ring *ring);
@@ -2234,6 +2235,7 @@ amdgpu_get_sdma_instance(struct amdgpu_ring *ring)
 #define amdgpu_ring_get_wptr(r) (r)->funcs->get_wptr((r))
 #define amdgpu_ring_set_wptr(r) (r)->funcs->set_wptr((r))
 #define amdgpu_ring_emit_ib(r, ib) (r)->funcs->emit_ib((r), (ib))
+#define amdgpu_ring_emit_pipeline_sync(r) (r)->funcs->emit_pipeline_sync((r))
 #define amdgpu_ring_emit_vm_flush(r, vmid, addr) (r)->funcs->emit_vm_flush((r), (vmid), (addr))
 #define amdgpu_ring_emit_fence(r, addr, seq, flags) (r)->funcs->emit_fence((r), (addr), (seq), (flags))
 #define amdgpu_ring_emit_gds_switch(r, v, db, ds, wb, ws, ab, as) (r)->funcs->emit_gds_switch((r), (v), (db), (ds), (wb), (ws), (ab), (as))

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

@@ -257,6 +257,8 @@ void amdgpu_vm_flush(struct amdgpu_ring *ring,
 
 	if (pd_addr != AMDGPU_VM_NO_FLUSH) {
 		trace_amdgpu_vm_flush(pd_addr, ring->idx, vm_id);
+		if (ring->funcs->emit_pipeline_sync)
+			amdgpu_ring_emit_pipeline_sync(ring);
 		amdgpu_ring_emit_vm_flush(ring, vm_id, pd_addr);
 	}
 

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

@@ -3041,6 +3041,26 @@ static int gfx_v7_0_cp_resume(struct amdgpu_device *adev)
 	return 0;
 }
 
+/**
+ * gfx_v7_0_ring_emit_vm_flush - cik vm flush using the CP
+ *
+ * @ring: the ring to emmit the commands to
+ *
+ * Sync the command pipeline with the PFP. E.g. wait for everything
+ * to be completed.
+ */
+static void gfx_v7_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
+{
+	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);
+	}
+}
+
 /*
  * vm
  * VMID 0 is the physical GPU addresses as used by the kernel.
@@ -3059,13 +3079,6 @@ 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) |
@@ -5147,6 +5160,7 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = {
 	.parse_cs = NULL,
 	.emit_ib = gfx_v7_0_ring_emit_ib_gfx,
 	.emit_fence = gfx_v7_0_ring_emit_fence_gfx,
+	.emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync,
 	.emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
 	.emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
 	.emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
@@ -5164,6 +5178,7 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = {
 	.parse_cs = NULL,
 	.emit_ib = gfx_v7_0_ring_emit_ib_compute,
 	.emit_fence = gfx_v7_0_ring_emit_fence_compute,
+	.emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync,
 	.emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
 	.emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
 	.emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,

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

@@ -4692,8 +4692,7 @@ static void gfx_v8_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr,
 
 }
 
-static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
-					unsigned vm_id, uint64_t pd_addr)
+static void gfx_v8_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
 {
 	int usepfp = (ring->type == AMDGPU_RING_TYPE_GFX);
 	uint32_t seq = ring->fence_drv.sync_seq;
@@ -4715,6 +4714,12 @@ static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
 		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)
+{
+	int usepfp = (ring->type == AMDGPU_RING_TYPE_GFX);
 
 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |
@@ -5037,6 +5042,7 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = {
 	.parse_cs = NULL,
 	.emit_ib = gfx_v8_0_ring_emit_ib_gfx,
 	.emit_fence = gfx_v8_0_ring_emit_fence_gfx,
+	.emit_pipeline_sync = gfx_v8_0_ring_emit_pipeline_sync,
 	.emit_vm_flush = gfx_v8_0_ring_emit_vm_flush,
 	.emit_gds_switch = gfx_v8_0_ring_emit_gds_switch,
 	.emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush,
@@ -5054,6 +5060,7 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = {
 	.parse_cs = NULL,
 	.emit_ib = gfx_v8_0_ring_emit_ib_compute,
 	.emit_fence = gfx_v8_0_ring_emit_fence_compute,
+	.emit_pipeline_sync = gfx_v8_0_ring_emit_pipeline_sync,
 	.emit_vm_flush = gfx_v8_0_ring_emit_vm_flush,
 	.emit_gds_switch = gfx_v8_0_ring_emit_gds_switch,
 	.emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush,