ath 11n update: locking, regressions, testing

[Adrian Chadd]

.. and here's another example LOR, this time between the net80211 node
lock and the ath lock, again during raw TX.

This occurs because of an ioctl call through setmlme_dropsta(), which
holds the node lock whilst it calls domlme(), which sends a management
frame via IEEE80211_SEND_MGMT ->ic_raw_xmit->ath_raw_xmit().

Again, I'm not entirely sure how to fix this.



Adrian

lock order reversal:
 1st 0xc08316cc ath0_node_lock (ath0_node_lock) @
/data/freebsd/mips/if_ath_tx/src/sys/net80211/ieee80211_ioctl.c:1326
 2nd 0x80a01440 ath0 (ath0) @
/data/freebsd/mips/if_ath_tx/src/sys/dev/ath/if_ath_tx.c:1716
KDB: stack backtrace:
db_trace_thread+30 (?,?,?,?) ra 80361b8c sp c7be9750 sz 24
db_trace_self+1c (?,?,?,?) ra 80074dbc sp c7be9768 sz 24
80074d88+34 (?,?,?,?) ra 801e103c sp c7be9780 sz 416
kdb_backtrace+44 (?,?,?,?) ra 801f8f38 sp c7be9920 sz 24
801f8f04+34 (?,?,?,?) ra 801f9bf4 sp c7be9938 sz 32
witness_checkorder+9cc (?,?,803b67cc,6b4) ra 80199a50 sp c7be9958 sz 80
_mtx_lock_flags+c4 (?,?,?,?) ra 8008b064 sp c7be99a8 sz 48
ath_raw_xmit+60 (?,?,?,?) ra 802b20a0 sp c7be99d8 sz 128
ieee80211_mgmt_output+2a4 (?,?,?,?) ra 802b4fac sp c7be9a58 sz 64
ieee80211_send_mgmt+9e4 (?,?,?,?) ra 802a29d4 sp c7be9a98 sz 96
802a296c+68 (?,?,?,?) ra 802a3824 sp c7be9af8 sz 32
802a3710+114 (?,?,?,?) ra 802a3fac sp c7be9b18 sz 64
802a3f14+98 (?,?,?,?) ra 802a4778 sp c7be9b58 sz 72
802a4224+554 (?,?,?,?) ra 802a6434 sp c7be9ba0 sz 128
ieee80211_ioctl+2c8 (?,?,?,?) ra 802cc71c sp c7be9c20 sz 48
in_control+1c8 (?,?,?,?) ra 80265488 sp c7be9c50 sz 88
ifioctl+13cc (?,?,80aa9de0,8072e300) ra 80201a90 sp c7be9ca8 sz 144
soo_ioctl+3b0 (?,?,?,?) ra 801fc354 sp c7be9d38 sz 40
kern_ioctl+248 (?,?,?,?) ra 801fc4fc sp c7be9d60 sz 64
ioctl+130 (?,?,?,?) ra 80355b8c sp c7be9da0 sz 56
trap+8a4 (?,?,?,?) ra 8034e01c sp c7be9dd8 sz 168
MipsUserGenException+10c (?,?,?,40658290) ra 0 sp c7be9e80 sz 0
pid 171