The break lock happens while the system panicked but it is not an useful
operation as, at that point, the other CPUs might be already stopped.
It is more correct to just skip locking entirely for the interfaces,
as long as the asserts won't whine in panicstr != NULL conditions (to
be added asap).
The 'acquire_lock' and 'release_lock' semantic is just left for the
normal RB_DUMP reboot(2) invokation case (reboot -d, for example).

Discussed with: emaste, rstone