[ath9k-devel] ath9k and kernel versions

[Pavel Roskin <proski@gnu.org>]

On Fri, 2008-07-25 at 10:52 -0400, Jack Howarth wrote:
> Does anyone know if the current ath9k drivers will work
> against either the tip linux kernel or the current 2.6.26-git12
> stable linux kernel? I have been able to generate an ath9k
> patch against the tip kernel git which applies cleanly to both
> and builds the ath9k driver. I've only tested the ath9k driver
> under the tip kernel so far using...

The current tip kernel is known to have some problems with wireless
devices, which is discussed in linux-wireless.  I don't think it's a
good idea to use it for wireless work at this time.

ath9k is meant to be used with wireless-testing only:
git://git.kernel.org/pub/scm/linux/kernel/git/linville/wireless-testing.git
(master branch)

There is a repository based on wireless-testing with ath9k added:
git://git.kernel.org/pub/scm/linux/kernel/git/mcgrof/my-wireless-testing.git
(ath9k branch)

> alias ath0 ath9k
> alias wifi0 ath9k
> 
> in /etc/modprobe.conf which gives the following ath9k kernel messages
> in dmesg...

Do you really need all that?  You must be using an ancient distro if the
modules for existing PCI devices are not loaded automatically.  Besides,
you can create a file under /etc/modprobe.d instead of modifying the
stock /etc/modprobe.conf

> ath0: authenticate with AP 00:13:f7:05:e8:92
> ath0: authenticated
> ath0: associate with AP 00:13:f7:05:e8:92
> ath0: associate with AP 00:13:f7:05:e8:92
> ath0: associate with AP 00:13:f7:05:e8:92
> ath0: association with AP 00:13:f7:05:e8:92 timed out

I believe that's the problem reported in the mainline git.  As for the
locking, there was a recent discussion in linux-wireless that may be
relevant.

> However the NetworkManager seems unable to get an address
> from the AP. Since I have been using the madwifi drivers before
> these kernels (built using the rawhide development srpm as
> a framework), I have to remove...
> 
> options ath9k autocreate=sta
> 
> from modprobe.conf.

Why was it there?  ath9k has never had that parameter in its short
lifetime.

>  I am wondering if my problems with the
> AP not getting an ip address could be due to the absence of
> some option for the ath9k driver which would set it in sta mode?

"modinfo ath9k" doesn't show any options at all.  Options are
discouraged in modern drivers.

> Also I am unclear if the ath5k/ath9k drivers would require more
> careful configuration in the NetworkManager (since they don't have
> the 'autocreate=sta' line in modprobe.conf)?

I don't know much about NetworkManager except that I tried it a while
ago.  I don't think anyone is trying to make your life hard and require
additional options, more careful configuration or anything like that.
Free software normally becomes easier to use over time, not harder.

>  Or is it likely that
> I am missing some critical wireless patch in tip and/or the stable
> kernel.

I'm afraid it's not only you who is missing that patch.  Somebody needs
to check what broke the mainline.

>  I was hoping to be able to test this driver under at least
> the linus-kernel git or the current stable git so that I can keep
> using the rawhide srpm and associated patches for Fedora 9. Thanks
> in advance for any advice as I am baffled as to why the driver
> isn't working.

You are lucky that it builds at all, because I tried it on RawHide a few
days ago and it didn't compile.

I believe something got mismerged in the mainline, perhaps some patches
were pulled from wireless testing in a wrong order.

Unfortunately, ath9k is not the easiest driver to use for debugging that
issue, as it requires some of those patches.  There may be no route
between the mainline and wireless-testing that would be bisectable with
ath9k.

-- 
Regards,
Pavel Roskin