[ath9k-devel] [Make-wifi-fast] Diagram of the ath9k TX path

The other hack that I've seen ath9k do is they actually assign
sequence numbers and CCMP IVs at the point of enqueue-to-hardware,
rather than enqueue-to-driver. I tried this in FreeBSD and dropped it
for other reasons, mostly in favour of shallower (configurable!)
per-station queue depths.

That way you /could/ drop data frames in the driver/wifi stack
per-STA/TID queues as part of queue discipline, way before it gets
sent out on the air. If you're modeling airtime on the fly then you
could see the queue for a given station is way too deep now that the
rate to it has dropped, so don't bother keeping those frames around.

Once you have assigned the seqno/IV then you're "committed" so to speak.


-adrian