In message <tkrbswqd663.fsf@parity-error.sm.sony.co.jp>,
enami tsugutomo writes:


>if_wi.c itself does busy loop.  Even though it's a busy loop without
>delay, but ...

I saw the loops that retry up to 65536 times. The "insanity" i see is
after replacing the pccb delay() calls with a tsleep. if I do that,
and when the USB speakers are doing output, ifconfi'ging  the wi device
up consistently causes a slew of wi_seek timeout messages. They
persist until I pop the  card out

When the usb speakers are not active, it works fine.

The other devices I tried -- a tlp0, and a Xircom ne2000 clone, both
worked.


I guess is right about locks, but i just got lucky: I hacked up a
script that does a thousand ps axlws and greps for the waitchannels i
added.  its mostly the cardbus0 kernel thread which incurs the tsleeps
at insertion, and the ifconfig process incurs the tsleeps for ifconfig
up/down.