drm/amdgpu: fix memory leak in amdgpu_cs_sync_rings

amdgpu_sync_get_fence deletes the returned fence from the
syncobj, so the refcount of fence needs to lowered to avoid
a memory leak.

Bug: https://gitlab.freedesktop.org/drm/amd/-/issues/2360Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Tested-by: Mikhail Gavrilov <mikhail.v.gavrilov@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Bert Karwatzki <spasswolf@web.de>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/3b590ba0f11d24b8c6c39c3d38250129c1116af4.camel@web.de

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

@@ -1220,10 +1220,13 @@ static int amdgpu_cs_sync_rings(struct amdgpu_cs_parser *p)
 		 * next job actually sees the results from the previous one
 		 * before we start executing on the same scheduler ring.
 		 */
-		if (!s_fence || s_fence->sched != sched)
+		if (!s_fence || s_fence->sched != sched) {
+			dma_fence_put(fence);
 			continue;
+		}
 
 		r = amdgpu_sync_fence(&p->gang_leader->explicit_sync, fence);
+		dma_fence_put(fence);
 		if (r)
 			return r;
 	}