IB/mlx4: Fix memory ordering of VLAN insertion control bits

We must fully update the control segment before marking it as valid,
so that hardware doesn't start executing it before we're ready.

Signed-off-by: Eli Cohen <eli@mellanox.co.il>

[ Move VLAN control bit setting to before wmb().  - Roland ]

Signed-off-by: Roland Dreier <rolandd@cisco.com>