On Monday 09 May 2011 17:20:49 John Nielsen wrote: > On May 6, 2011, at 4:14 PM, Mick wrote: > > On Friday 06 May 2011 18:04:31 John Nielsen wrote: > >> Doesn't look like this went through the first time; re-sending without > >> attachment. > >> > >> On May 5, 2011, at 6:52 PM, Paul Hartman wrote: > >>> On Thu, May 5, 2011 at 12:40 PM, John Nielsen wrote: > >>>> I am trying to set up a 5GHz wireless access point on an Alix 3d2 > >>>> board with an AR9220 (ath9k "Merlin") PCI card. I have done so > >>>> successfully using Fedora 14 on identical hardware but I would > >>>> greatly prefer to use Gentoo so I can use a more recent kernel and > >>>> customize things appropriately for the platform. > >>> > >>> It seems like everything is pretty much the same, other than the > >>> kernel (and presumably the ath9k driver). But I would look at the udev > >>> rules for CRDA to be sure they match and are being applied the same on > >>> both systems. You shouldn't ever need to "iw reg set" on a system with > >>> CRDA, it should do it for you. So I wonder if you're setting it, and > >>> then CRDA is immediately setting it back to 00... > >> > >> The udev rule for CRDA is the same on both systems, and matches what is > >> shown on http://wireless.kernel.org/en/developers/Regulatory/CRDA. The > >> Fedora box also has a magic rule to call /sbin/setregdomain when an > >> 80211 interface is added, which is a shell script that infers the > >> regdomain from the currently set timezone. The punch line of the script > >> is simply a call to "iw reg set $COUNTRY" > >> > >> However, I'm not sure crda is being called appropriately on the Gentoo > >> box. > > > > On the Fedora machine I see this in dmesg: > >>>> [ 17.248674] cfg80211: Calling CRDA for country: US > >>>> [ 18.848206] cfg80211: Regulatory domain changed to country: US > >> > >> and I don't ever see anything similar on the Gentoo machine, even when > >> running "iw reg set" by hand. Further, I don't see anything in the > >> output of "udevadm monitor --environment kernel". > >> > >> I just dropped the kernel from the Fedora machine on to the Gentoo box > >> and (somewhat surprisingly) it works just fine. The reg domain gets set > >> no problem, hostapd starts, life is good--except that now I feel like > >> I've sinned against nature and I'd like to get my own, smaller kernel > >> back. > >> > >> While it's possible the new kernel version is broken I rather suspect > >> that I have configured it badly. I set out to configure a minimal > >> kernel with just the features and drivers I want on this hardware and > >> no need for modules or an initramfs. Does cfg80211 need to be a module > >> to work properly? I wouldn't think so. > >> > >> I'll do some more experimenting but in the mean time here's my kernel > >> config in case anyone has ideas on what could be wrong. Thanks! > >> > >> Config file here: http://pastebin.com/S68ye6Pz > > > > I suggest that you run a diff --suppress-common-lines -y between the > > Fedora and your own kernel to find out what's different between the two > > as far as your driver is concerned. > > The outright diff was far too verbose to be useful, but looking at "lsmod" > output when running the Fedora kernel was instructive. I recognized > everything as being in by Gentoo kernel except for "rfkill", which was > listed as a dependency of cfg80211. Long story short, I got it working > like I want by: > > 1) Including RFKILL in my kernel config and > 2) Building cfg80211, mac80211, rfkill, and all parts of the ath9k driver > as modules > > Just adding RFKILL to a static (module-less) kernel didn't fix it, and just > building cfg80211 and friends as modules without RFKILL also didn't fix > it; I had to do both. > > Thanks all for your responses so far. I'll be happy to ditch Fedora. :) > > For future reference, does anyone know why either of the above would be > required? If I see the same behavior in the latest git kernel should I > file a bug? If so, where (since I doubt this is a Gentoo issue)? I wouldn't have thought that you *must* have rfkill built in as a module for it to work. However, I have my wireless drivers (inc. rfkill) built in as modules. I did this because I need to pass some options to the wireless driver at some point. The only time I had to build something as a module was alsa, because otherwise I couldn't configure my sound card on a particular box (alsaconf/alsactl expected it to be a module so that it can probe it). I ended up compiling it as a module since. Perhaps someone else can advise on this rfkill issue. -- Regards, Mick