[ath9k-devel] 2.6.27-rc5 + w1.fi patches = crash (No ProbeResp + WARNING: at kernel/softirq.c:136)

[Mats Johannesson <spamcan1@abrasax.se>]

On Fri, 29 Aug 2008 10:49:43 +0300 Jouni Malinen wrote:
> On Aug 29, 2008, at 9:32 AM, Mats Johannesson wrote:
[...]
> > "Submitted for inclusion; no known open issues") cause a disconnect
> > after some minutes and throws a kernel warning. After that, the
> > link is dead.
> 
> One thing to keep in mind with those patches is that they are
> against wireless-testing tree, not Linus git tree. I haven't looked
> how different mac80211 and ath9k are between those trees, but there
> may be differences (more likely in mac80211) that can cause issues.

Yeah, I was a bit uneasy with the tree separation but felt it better
to shout an "early wolf" (in case the patches have some subtle issues)
than to git-bisect a massive tangle of 2.6.28-WIDE_MERGE_WINDOWS

> > Most of the patches applied without issues
> > (mac80211_iwassocreqie.patch FAILED because it already was in -rc5
> > and mac80211_fix_debugfs_netdev.patch was successfully detected as  
> > being in -rc5) and the two which applied with a slight fuzz I
> > checked manually afterwards.
> 
> wireless-testing has been a bit behind over the past weeks, but it  
> was updated yesterday (to rc4 or somewhere between rc4 and rc5) and
> I updated the patch series based on that.

Someone upstreams cherry-picked the two bad boys then... Not good...

[...] 
> Would you be able to easily test whether you'll get the same problem  
> with wireless-testing plus the patches from w1.fi? I'm currently  
> running that on my laptop and haven't been able to reproduce this  
> issue so far. Did you do anything special with the connection when  
> the issue showed up?

Didn't do more than the usual web-surfing. First disconnect came after
three minutes - rebooted - then disconnect after twelve minutes.

I've been out of the loop for a good while so will need a bit of time
to install git and pull the wireless-testing branch.

> > Aug 29 04:07:28 sleipner kernel: WARNING: at kernel/softirq.c:136 ()
> 
> That's WARN_ON_ONCE(in_irq() || irqs_disabled()) in  
> _local_bh_enable_ip(). The call trace did not show any clear ath9k  
> functions, so I'm not sure where exactly this is triggered.

I'll do my best to get a clean environment with more and better output.
Might as well do that with the -rc5+w1.fi also (though sometimes the
debug stuff hides/eradicates subtle bugs).

Will return,
Mats