mac80211: fix CCMP races

Since we can process multiple packets at the
same time for different ACs, but the PN is
allocated from a single counter, we need to
use an atomic value there. Use atomic64_t to
make this cheaper on 64-bit platforms, other
platforms will support this through software
emulation, see lib/atomic64.c.

We also need to use an on-stack scratch buf
so that multiple packets won't corrupt each
others scratch buffers.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>