K 10
svn:author
V 6
adrian
K 8
svn:date
V 27
2011-08-28T03:05:16.267535Z
K 7
svn:log
V 1199
Migrate back to the locking structure the atheros reference driver and
linux ath9k uses - hide the per-TID state behind the associated hardware
TXQ.

This is less fine grained than what I was doing before, but it results in:

* The code being cleaner (ie, there's not a mess of locks everywhere);
* The code being closer to what the reference/linux drivers do, making it
  easier to port code over;
* A bunch of state changes become much more predictable - eg, the BAW
  window updates occur in one bunch rather than interleaved with other
  packet TX.

This isn't yet complete, as there are a few things to fix:

* recursive mutexes during ht node cleanup
* remove the current txq drain which uses ieee80211_iterate_nodes() which
  grabs the IEEE80211_NODE_LOCK() and replace it with a similar setup
  to what Linux/Reference driver does.
* Figure out how to properly halt the software TXQ (via ath_raw_xmit()
  and via ath_start()) during the reset process so frames aren't queued
  to the software queue whilst things are drained.
* Investigate whether a STA which is going away can have traffic queued
  to its ieee80211_node, which is in the process of being freed or has
  already been freed.

END