drm/amdgpu: execution barrier after fence v2

Insert wait for reg mem after EOP to fix potential issue with vm context switch

v2: move wait to vm_flush() use equal instead of greater than.
Signed-off-by: NAnatoli Antonovitch <anatoli.antonovitch@amd.com>
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>

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

@@ -3965,6 +3965,7 @@ static void gfx_v8_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr,
 			  DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
 	amdgpu_ring_write(ring, lower_32_bits(seq));
 	amdgpu_ring_write(ring, upper_32_bits(seq));
+
 }
 
 /**
@@ -4044,6 +4045,17 @@ 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);
+	uint32_t seq = ring->fence_drv.sync_seq[ring->idx];
+	uint64_t addr = ring->fence_drv.gpu_addr;
+
+	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
+	amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
+		 WAIT_REG_MEM_FUNCTION(3))); /* equal */
+	amdgpu_ring_write(ring, addr & 0xfffffffc);
+	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
+	amdgpu_ring_write(ring, seq);
+	amdgpu_ring_write(ring, 0xffffffff);
+	amdgpu_ring_write(ring, 4); /* poll interval */
 
 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |