WLAN status update

Werner Almesberger werner at openmoko.org
Mon Jan 12 19:34:37 CET 2009


Michael Buesch wrote:
> I was talking about BMIrompatchInstall and friends.

Ah ! I hadn't poked around in the BMI yet. That looks quite interesting
indeed.

> I think it should be handed to the Supplicant. The Supplicant will try
> to re-establish the connection or drop it.

Sounds good to me. If we're allowed to do something radical, such
as just discarding the whole MAC state, that would be perfect for
the AR6k in terms of reliability of the rfkill mechanism and also
in terms of implementation complexity.

More specifically, SDIO gives us a single-bit on/off switch. But
that affects the whole module, not just the transmitter. And since
this is a "full MAC", there's quite a lot of state in the module.

This still leaves the Atheros-specific extensions, such as --power
maxperf, but I think we can worry about these later.

> So the real question is, does the firmware implement part of the
> Supplicant? I guess so.

Hmm, I'm not sure what you mean here. The firmware does scan for
networks on its own and connects if it finds matching ESSID and
encryption.

> Well, wireless extensions semantics are a little bit crappy.

Hmm, or at least "vague" :)

> But as far as I
> understand the semantics, SIWESSID is supposed to associate to the
> service (except in a few cornercases).
> So IMO it should fail, because without a transmitter, we can't handshake a
> connection.

My understanding is that SIWESSID is allowed to return successfully
without establishing an association. The MAC keeps on trying on its
own.

> Yeah, as I said. Current mainline drivers don't mess with state at all.

Yes, that's what I used as my baseline so far. I.e., that rfkill is
not allowed to do damage that goes beyond what happens anyway when
you have a complete loss of signal.

I also noticed that the - otherwise very detailed - documentation of
rfkill leaves this aspect completely open to interpretation.

Perhaps we should resume this on linux-wireless, since I guess that's
where all the rfkill folks hang out.

Thanks,
- Werner



More information about the devel mailing list