Using WPA over a Trend TEW-443PI works fine for low packets/second.
At 100 KBps or thereabouts, sylog becomes flooded with:
rix <nnn> bad ratekbps 0 mode 6
The Tx queue will lock for about 15 seconds, wherein no packets are sent.
At 1 MBps and above, the kernel will panic randomly after a minute or so of send -> wait 15 seconds for tx to resume -> send -> repeat
Unlike one user (seniorr) who reported a panic, the kernel here panics in the interrupt handler, locking all processes, preventing the oops from being recorded in syslog. Alt-sysrq-boot is the only option at that point.
System is an Athlon 64 (running in i686 32 bit mode) on Slackware 10.2 (glibc 2.3.5, gcc 3.3.6) with a hand-compiled kernel for obvious reasons.
Sample oops copied by hand from screen:
EIP: 0060:<f486c02b> (calc_usecs_unicast_packet+0x2b/0x280)
EFLAGS: 00010286
eax: 00000792, ebx: f88e6228, ecx: 00000640, edx: 00000286
esi: 00000000, edi: 0000001f, ebp: f88e43c0, esp: c2163cb8
ds: 007b, es: 007b, ss: 0068
stack: f486db60 0000000b 00000000 00000000 00000006 00000000
00000000 00000000 00000000 00000000 00000005 00000008
f33712dc 00000002 f486c07a = update_stats
update_stats+0x17a/0x420
ath_rate_tx_complete+0x70e/0x790
zz005b88fd+0x14/0x130
ath_uapsd_processtriggers+0xe6/0x550
ieee80211_input_monitor+0x32/0x510
ath_tx_capture+0xc1/0x1a0
ath_tx_processq+0x4ac/0x6c0
handle_IRQ_event+0x30/0x70
ath_tx_tasklet+0x11/0x120
tasklet_action+0x46/0x70
__do_softirq+0x7d/0x90
local_bh_enable+0x56/0x90
rt_run_flush+0x40/0x90
rt_run_flush+0x5b/0x90
fib_netdev_event+0x5a/0x90
notifier_call_chain+0x2d/0x50
netdev_state_change+0x38/0x60
linkwatch_run_queue+0xb7/0x120
linkwatch_event+0x2b/0x60
worker_thread+0x0/0x2a0
code: 55 57 bf 1f 00 00 00 56 31 f6 53 83 ec 28 8b 44 24 3c
8b 54 24 44 8b a8 fc 0e 00 00 8d 0f 52 85 ed 8d 5c 85
20 0f 84 24 02 00 00 <66> 83 7b 06 00 0f 84 f2 01 00 00
0f b6 43 05 83 f8 02 0f 84 c8
