qdisc_restart locking bug?

When qdisc_restart() is called from qdisc_run(), dev->queue_lock is 
obtained via spin_lock_bh().
When dev->hard_start_xmit() is called from qdisc_restart(), 
dev->queue_lock is dropped, and then re-acquired using only spin_lock().
So, doesn't qdisc_restart need s/spin_lock/spin_lock_bh/ for 
dev->queue_lock?
When looking at the dev_queue_xmit -> qdisc_run -> qdisc_restart call 
path, it seems so to me.
        Jeff