How to profile this strange effect under high load?

[Bruno Randolf <br1@einfach.org>]

Hello!

I am trying to debug a strange effect, I am seeing on soekris net48xx boards 
with two ath5k interfaces:

 * wlan0 [phy0] configured in ad-hoc mode, A band, and it is getting a lot of 
traffic routed thru it
 * wlan1 [phy1] also configured in ad-hoc mode, G band, but not actively 
sending (no IP address).

I run iperf between two PCs attached by ethernet, UDP 25Mbps and I am sure all 
traffic is routed thru wlan0 (phy0). The load on the SENDING box gets very 
high:

Mem: 27304K used, 99040K free, 0K shrd, 616K buff, 14128K cached
CPU:   0% usr   0% sys   0% nic   0% idle   0% io   5% irq  93% sirq
Load average: 1.73 1.01 2.23 2/38 3128
  PID  PPID USER     STAT   VSZ %MEM %CPU COMMAND
 2772     2 root     RW       0   0%  71% [phy1]
 2757     2 root     SW       0   0%  11% [phy0]
 3128   436 root     R      988   1%   8% top 
    3     2 root     SW       0   0%   5% [ksoftirqd/0]
    5     2 root     SW       0   0%   3% [events/0]
   56     2 root     SW       0   0%   1% [bdi-default]
 1383     1 root     S     1824   1%   0% /usr/bin/oprofiled --session-dir=/r

...and the interesting thing is that [phy1] - the INACTIVE interface - 
consumes much more CPU time than [phy0] which is actually transmitting.

I have disabled ANI on wlan1, to avoid getting many MIB interrupts, but it 
does not change much. 

I have dumped frames on phy1, and all it sees are a few beacon frames.

I have tried wlan1 in managed mode - same effect. The effect is slightly less 
in AP mode, though...

I tried oprofile, but it shows mostly tx related things.

So my question is: How can I find out, what [phy1] is so busy doing? Any ideas 
how to profile this?

Thanks in advance,
bruno