More PMAP performance improvements: skip 256 MB segments entirely if they
are are not mapped during ranged operations and reduce the scope of the
tlbie lock only to the actual tlbie instruction instead of the entire
sequence. There are a few more optimization possibilities here as well.