Implement the vmid reservation.

v2: move sync waiting only when flush needs
v3: fix racy
v4: peek fence instead of get fence, and fix potential context starved.
Signed-off-by: NChunming Zhou <David1.Zhou@amd.com>
Reviewed-by: NJunwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>