rt73usb (rt2x00) crash bug on (big-endian) ARM

rt73usb (rt2x00) crash bug on (big-endian) ARM

[Gert Boddaert]

Hi,

I tested 3 compat-wireless versions:-

we intend to use compat-wireless-2008-04-08. 
For our embedded puposes this is relatively stable (although it contains the
mentioned bug).
Also the stable version compat-wireless-2.6.30 kernel was tested, and the latest
"bleeding edge" compat-wireless-2009-06-25.

The drivers were cross-compiled (after some integration work) with the windriver
toolchain for the big-endian IXP420 (x-scale). 
We are using the 2.6.27.18 kernel.

the rt73usb is used in AP mode through hostapd, after some patching in hostapd
(the TX not acked thing). 
This works remarkably well, accept for the TX rate bug (low throughput ~800Kb/s
from AP to client),
which can be solved in 2008-04-08 by fixing the rate manually (e.g. iwconfig
wlan0 rate 24M). 
I am happy to inform, that in the 2009-06-25 version this particular automatic
rate problem seems solved.

Anyway, the bug I want to post about is presumably somewhere in the RX path of
the rt73usb driver (or related stacked drivers).

When running iperf (iperf -c 192.168.x.y -u -t 10 -i 1 -b 20M -r ) and
increasing the load on the AP-RX-side IXP420 to >12 Mb/s, the system crashes.

We are investigating how to get a decent crash dump from the embedded device,
but that will take some work.
 
I also tested the same setup on a big-endian powerpc-mac G4 (same 2.6.27.18
kernel), but it doesn't crash there probably because the rt73usb is connected to
a USB 1.0/1.1 port and the througput does not rise above 8-9Mb/s.

To summarize...

Problem 1
The device crashes when the load on its RX side goes higher than ~12Mb/s on all
tested driver versions. 
Is this a known big-endian issue or does this also
happen on low-spec'ed i386 processors?

Problem 2, 
related to compat-wireless-2009-06-25: I can no longer connect from Windows XP
to our embedded AP when using this driver (alignment or endianess issue),
although this does work perfectly with the other driver versions.
Strange thing is, that when booting my test laptop into (ubuntu) linux, 
I CAN connect to the embedded AP with the compat-wireless-2009-06-25 driver.

Re: rt73usb (rt2x00) crash bug on (big-endian) ARM

[Gert Boddaert]

We found the culprit of the 'crash' for problem 1: 
our watchdog implementation.
The high load on the device, stops the watchdog refresh... hence crash and reset.

We disabled the watchdog to do further testing... next.

However, when the load on the RX side gets to high, after a while (a few seconds
of sustained RX load), 
the connection is not dropped but we are unable to get
any more packets through. This state does not resolve itself.