Resets and kernel freezes

[Milan Krcmar]

Thank you very much for all the answers.

> > - Do I still have to watch for card's functionality by generating
> > some traffic and "iwpriv wlanx reset 1" it when it fails to
> > communicate?
> 
> I'm not sure whether I understood this question correctly. Do you mean
> that the network connection does not work (i.e., cannot either send or
> receive packets) but otherwise devices seem to be operational? Do you
> see any error messages in 'dmesg' output?

Exactly. The card is not able to receive or transmit packets. I haven't
given you more information, because the purpose of my first posting was
to discover the current situation. Continue reading, please.

> 
> > - Is the above mentioned reset command fatal with some probability to
> >   the kernel (i.e. does the kernel freeze when I reset my card several
> >   times).
> 
> I haven't been able to easily (if at all) to reproduce this kind of
> issues with the latest driver versions. It looks likely that there is
> some kind of hardware dependency and I just don't have "suitable"
> combination in my testbed.
> 
> > Because of the resets being able to freeze the whole router, we have to
> > add a hardware watchdog to every router. (The resets are triggered both
> > by the driver itself and by the scripts watching for its functionality,
> > mostly at time of heavy traffic load.)
> 
> Do you have any way of getting any other debug information than "it's
> dead" when this happens? Do you see anything on the console when this
> happen if full debug is enabled ('dmesg -n 8')? Do you have any system
> that would support NMI watchdog that could help in debugging this?

Yes, I can and will give you more information. I've got a private reply
to this thread from a man using the same card without problems, he
described particular versions of kernel/driver/firware he is using.
I will start at this combination (which is, in general, the most current
one) and give you a precise report after several days. That report will
include debug information.

When the kernel freezes (in consequence to the frequent card resets), it
does not print anything to the console but the statement it is resetting
the card. Then it is not functional at all (from what I see at the
console). I will use kernel with APIC support to get the watchdog.

Please, now wait for several days for my report...

Milan