HAL changes wireless interface (net.80211) to wired (net.80203) in info.category after suspend

[Constantine Evans]

On a Dell D600 running Edgy with an Intel 2200BG, HAL changes the information published on the network device after coming out of suspend. Instead of listing it as a wireless device, it changes info.category and info.capabilities to advertise it as net.80203.

Network Manager thus detects the card as a wired interface instead of a wireless interface after suspending (everything is fine before suspending). It then goes on to claim that it is connected to a wired connection, but it does not actually configure the interface correctly, since it is using ZeroConf.

[Constantine Evans (cevans)]

Confirming, as others have reported this as well.

[Constantine Evans (cevans)]

This actually appears to be a bug in hal - I'll move it as soon as I get more information on it. Hal seems to be changing the information it publishes after a suspend. Restarting hal fixes the problem.

I'm attaching the NM log. Search for "Going to sleep" to find point where suspend occurs.

[Constantine Evans (cevans)]

rejecting - this is a hal bug. oops.

[Constantine Evans (cevans)]

I'm attaching a log of hald's verbose output when this happens, but it doesn't appear to have any obviously useful information on the problem.

Search for "event is 'button/sleep" to get to what I believe is the relevant portion.

[Ben Roberts (ben-l-roberts)]

I seem to have exactly this problem with a Toshiba Portege m300/ ipw2200

[Crispin Flowerday (crispin-flowerday-deactivatedaccount)]

It seems there is a monster race condition, at line 220 of classdev.c (where it opens /proc/net/wireless), /proc/net/wireless is totally empty:

Inter-| sta-| Quality | Discarded packets | Missed | WE
 face | tus | link level noise | nwid crypt frag retry misc | beacon | 20

investigations continue ...

[Crispin Flowerday (crispin-flowerday-deactivatedaccount)]

Further investigations show that when hal is invoked, nothing in /proc/net is setup yet (seems especially to be a problem for the ipw2200 driver, my other e100 driver works fine).

However, the stuff in /sys/class/net/eth* is setup fine. If I put a tiny sleep (not a real solution I know) in the classdev.c file just before it reads /proc/net/wireless everything works fine.

There is some code in classdev.c that just looks for the /sys/class/net/wireless directory. I don't claim to know why it isn't used, but if I use that code instead (patch attached) everything works fine for me.

Perhaps this should be given to the kernel people to work out why there is this race condition ? I would have thought /proc should be as up to date as /sys when the notification occurs

[Soren Hansen (soren)]

Marked as duplicate of 59981. In the comments for 59981 there are links to packages for testing a patch which is basically identical to the one attached to this bug report. :-)

[Constantine Evans (cevans)]

Rejecting hal part as well.