Adrian's Todo list

Embedded

Fix if_arge.c RX ring handling so mbufs aren't being wasted each time the interface is up/down'ed (PR?)
Re-architect the whole phymask mess with if_arge
Work with juli@ to optimise if_arge throughput
hwpmc mips24k support for sampled/interrupt driven measurement

Figure out a good way to handle fragments, now that r190579 broke fragment handling (as IF_ENQUEUE and IF_DEQUEUE both clear the m_nextpkt chained fragments, leaking fragment mbufs.)
Figure out why scan/bgscan actually issues the first scan even though traffic is definitely ongoing
In the case of a scan being interrupted, how do we inform any consumers (eg via scan ioctl from userland) that the scan was interrupted, rather than having it simply hang indefinitely.

Tidy up the rate control API - it's very messy and could do with a bi of a sanity check
.. then, begin work to merge the rate control stuff back into the net80211 rate control API.
Survey/add missing locking surrounding how ieee80211_node entries are handled for nodes that aren't in a node table. (Commit? PR?)
Implement if_transmit and handle fragments correctly (see above.)
Finish BAR TX handling (which involves correctly debugging node pause/unpause)
Reduce the size of the hardware queue when doing non-aggregation - don't just fill it full of non-aggregate frames
Teach ath_start() to only allow ~ half of the TX buffers to be allocated before delaying TX - that way space is left over for management frames (eg ADDBA, BAR.)
Export the rate control statistics using a command line program or a sysctl which returns the text data - stop doing it via kernel printf()s
Correctly handle frames when doing