Bugzilla Bug 184804

Madwifi-ng doesn't compile on kernel 2.6.22 (specially, for me,
gentoo-sources-2.6.22). This is a known bug (see URL for madwifi's ticket).
there is a patch posted (I haven't tested it), so maybe we should use that
untill the next version of madwifi is out?

Reproducible: Always

Steps to Reproduce:
1.Run 2.6.22
2.emerge madwifi-ng

Actual Results:  
The same error as in the attached ticket.

Expected Results:  
It should have compiled.

I was using AMD64. I think this qualifies as 'major' since my desktop is unable
to access the internet w/ the latest kernel, but feel free to change that.

Sorry, forgot to name the package that didn't compile. 

What is your madwifi version? It works perfectly fine for me.

http://madwifi.org/attachment/ticket/1315/141-Madwifi-fixes.patch

does that patch help for you?

(In reply to comment #3)
> http://madwifi.org/attachment/ticket/1315/141-Madwifi-fixes.patch
> 
> does that patch help for you?
> 

In regards to that patch, it needed some tweaking to apply in an ebuild.  The
line numbers aren't quite right, the headers needed altering, it had a section
meant for in-kernel use, part of it was already in the madwifi-2.6.22.patch in
portage, and it used some hacky #if's so it would build against 2.6.22_rc1. 
Also, it created an interesting problem.  I'm using two Trendnet NICs with
AR5212 chips.  My linux machine is the AP, with hostapd, using iptables to do
masquerading (and using ip forwarding, of course.  The Atheros NIC is bridged
with my onboard NIC (e1000).  So far, I haven't had much trouble with it, until
applying the patch and rebooting with a 2.6.22-gentoo kernel.  Hostapd still
started, the bridge was fine, the station associated (XP box), and packets came
merrily through the air, marched over the bridge, and out ppp0 into the great
big internet. Unfortunately, the reverse was not true.  Nothing would pass back
over the bridge and to the station.  From that end, everything just timed out.
Since it was needed, I had to go reboot with the 2.6.21 kernel, so I didn't get
a chance to test some stuff I wanted to. There's another station, a Wii, which
wasn't on at the time, and I plan to check that.  I also didn't get around to
sniffing the wireless NIC.  Currently, I've been checking the madwifi tree on
the net.  There's a couple changes that aren't in the patch that I'm going to
try tomorrow. Well, there's actually been a lot of changes, but I've isolated a
couple small ones that I think may at least alter the current behavior.

I'm sorry, I can't seem to get that patch to apply cleanly at all (as comment
#4 indicated, it's kinda trick). I'm using
madwifi-ng-0.9.3 and gentoo sources 2.6.22 and linux headers 2.6.22 as well.

drivers/net/wireless/madwifi/net80211/ieee80211_input.c: In function
'ieee80211_defrag':
drivers/net/wireless/madwifi/net80211/ieee80211_input.c:1051: error: invalid
operands to binary -
drivers/net/wireless/madwifi/net80211/ieee80211_input.c:1054: error: invalid
operands to binary -

.. and then a bunch of warnings before it errors out.

Created an attachment (id=124404) [edit]
New madwifi-2.6.22.patch

OK, tried out the changes. No dice.  Packets go out, they don't come back in.
Same story for the Wii.  I tcpdumped ath0, and realized arp was working. I had
forgotten that it would show up there.  So, I was seeing llc traffic, arp,
802.11 stuff, and outgoing IP traffic. Just nothing incoming on IP. Not a
surprise, since it wasn't showing up on br0. Then I realized: I forgot to
rebuild hostapd against the new kernel. Son of a..anyway, since I can't test
this in station mode, I'm posting it for someone else to try. It replaces the
current madwifi-2.6.22 patch in portage
(/usr/portage/net-wireless/madwifi-ng/files/madwifi-2.6.22.patch, for your
pasting convenience)

I'll rebuild hostapd and retest tomorrow (well, technically, later today).
Sleep now.

Thanks for the updated patch, Thomas. It compiled on my machine and seems to be
working pretty good :)

I'm not using the driver for anything like routing, though - just connecting to
an access point and using the internet. I won't close the bug since Thomas is
still having an issue.

Ok, rebuilt hostapd, which had it's own problems (see bug #184830, for those
who care).  Rebuilt bridge-utils, just in case.  Nope. Nothing can cross back
over the bridge to the stations.  Reboot with the 2.6.21 kernel, everything
works. I traced a ping.  The ICMP reply packets come into eth0, which is the
ppp0 link, and go...somewhere. At this point, I don't know if madwifi is to
blame, so maybe this bug should be closed. It could be hostapd, the kernel,
madwifi, or all three. I'd like to know if anyone is using a different wireless
driver w/a bridge and seeing the same behavior.

The new patch works for me too, just a basic wifi internet connection setup
with NetworkManger, none of the fancy routing stuff.  Seems like the other
issues are probably outside of madwifi-ng.  

Is it possible to get the latest version of the patch into portage before
fixing    all the other problems?  That way at least the basic functions will
work which is better than it not being able to compile and will do it for most
folks.

Thanks a lot! 

thank you, I added the new patch.