K 10
svn:author
V 6
adrian
K 8
svn:date
V 27
2013-04-21T19:33:55.612993Z
K 7
svn:log
V 1507
Bring over my initial power save queue work.

Since this is a little more complicated than I thought, I'll need some
subversion history for this.

So far:

* Do node pause/unpause when the stack enters/leaves powersave for
  the given station;
* Remove the use of ath_node lock outside of the rate control code -
  now the rest of the ath_node stat is protected by the tx lock;
* When a node reassociates, a bunch of frames are in the software
  queue and hardware queue.  We need to tidy that up - so for now,
  just go through the normal cleanup path.  But that may not work
  in the long run.
* .. and whilst doing this, reset some of the state (eg BAR TX
  state, filtered frames, etc.)
* .. this still doesn't entirely close the queue stalls; but they
  may be BAR TX related rather than this particular work.

* Add support for leaking frames out whilst the node is in sleep
  state.  That way I can do interesting things for PS-POLL.
* Tie into the PS-POLL hook from net80211; leak out a software queued
  frame if we have any; else just mark the node as ready to leak
  a frame and leave it up to the stack.

I'm still looking into whether there's any possible races with
ps-poll handling - notably, if this happens:

* STA sends PS-POLL
* Stack generates a null frame response;
* .. but in another thread, the stack gets handed a frame;
* .. and that thread wins the transmit race and it gets queued
  out first.

Now hopefully the TX path locking in net80211 stops this from occuring.
But still..

END