Bug#434527: Bug#408207: Madwifi problems associating using wext/NetworkManager

[Kel Modderman]

Hi Tim,

On Thu, 26 Jul 2007 10:54:13 am Tim Hull wrote:
> I have spent a significant amount of time dealing with issues between
> MadWifi and wext/wpasupplicant/NetworkManager on Debian (etch, lenny, AND
> sid).  As it stands, there are still significant issues - in factI cannot
> associate *at all* to my stock, unencrypted WRT54G using
> NetworkManager+madwifi on my Atheros 5424 card (bundled with MacBook Core
> Duo). This is about as standard of a configuration as you can get, so it's
> obviously a widespread bug.

http://madwifi.org/ticket/275

>
> I have managed to isolate the issue to a degree.  Though I'm somewhat
> C-impaired, the issue does seem to stem from issues w/wext compliance - and
> in particular AP scanning and the "preempt_scan" function.  This seems to
> be a long-standing issue, and has existed in every build of madwifi I've
> tried.  The upstream bug mentioned earlier in this thread is still open,
> and there appears to be no progress whatsoever on that front.

Add your progress to it.

>
> Anyway, there is currently two workarounds that I've tested and know to
> work.  The first, and the one used by Ubuntu, is to change NetworkManager's
> source such that the wpasupplicant's "madwifi" driver is used for cards
> using madwifi instead of the generic "wext" driver.  This seems to work
> well, though the Debian maintainer for NetworkManager didn't like the idea
> of special behavior for madwifi.  Secondly, there is the option of totally
> removing the preempt_scan() function and all calls to it from
> ieee80211_wireless.c.  This seems to work perfectly to me, but may come at
> the loss of some functionality (namely, wireless scan timeout).  This
> function was only added in the last 6-9 months, though, so the driver is
> known to work without it.

It seems that with or without preempt_scan, people still have various 
different problems associating to access points.

>
> I've attached my patch to remove the function in question (preempt_scan)
> from ieee80211.c. I've attached patches to both the release used in
> unstable and the trunk from madwifi.org (which, in my experience, blows
> away the release in wireless performance - in addition to adding support
> for the Atheros 802.11n chipsets).
>
> I hope a workaround such as the one(s) I've suggested can be uploaded to
> the archive - this can be quite a nasty bug for some Atheros users...

I'll only apply to the debian package what has been accepted by upstream.

In the ticket referenced above, I was a vocal advocate *for* prempt_scan, so 
it'd be best to convince some of the currently active madwifi developers that 
preempt_scan is crap, to remove it, and to "fix" madwifi scanning behaviour 
with something better.

Thanks, Kel.