[BUG] reproduceable PANIC in recent hostap driver

[JuanJo Ciarlante]

On Sun, Oct 19, 2003 at 06:01:53PM -0700, Jouni Malinen wrote:
> On Sun, Oct 19, 2003 at 09:23:40PM -0300, JuanJo Ciarlante wrote:
> 
> > System: P-166Mhz, linux-2.4.22, Debian 3.0r1, hostap-0.1.[01];
> >         Prism2.5 DWL-520, PRI v1.1.1, STA v1.7.4
> > 
> > Doing ifconfig wlan0 PANICs "immediatly" if ospfd (from zebra pkg) is running.
> 
> Do you mean 'ifconfig wlan0' or 'ifconfig wlan0 <something else,
> like up>'?

stupid me |-(  :   ifconfig wlan0 down

> > Oops always shows ospfd process paniced; I guess is something related to
> > netlink stuff changed around 0.0.4 (I'd been getting panics but didn't find
> > the possible cause, ... 'til today ;), ospfd holds a netlink socket to get
> > reported about interface/routing/etc news.
> 
> Just monitoring the netlink messages with 'ip monitor all' is not enough
> to trigger anything (at least on Linux 2.6.0-test8).

interesting... ip monitor all on my 2.4.22 shows the following errors about
2~3/second while wlan0 is up (although is doesn't panic as ospfd when
downing the iface):

  bash# ip monitor all
  BUG: nil ifname
  BUG: nil ifname
  BUG: nil ifname
  :
  
> Would you happen to
> know what ospfd does after noticing a change in network devices?

dunno, but I'll try to strace it ``to death''.
What I'm sure is that ospfd process hold a raw socket (from lsof) that I
assume to be netlink .

> I would assume it could be something like changing routing entries etc.
> 
> > Another interesting thing is that the panic is the infamous
> > "Aieee killing interrupt handler"; given that it would seem to happen under
> > process context --*always* ospfd--. 
> > Obviously after that big red button is the only escape.
> > 
> > Feeding ksymoops with hand written address (1st in the stack trace) gives
> >   c0113509 <schedule+4d/314>
> 
> Could you please try to resolve couple of more functions from the back
> trace? schedule() is not supposed to be called from interrupt handler,
> so it is certainly not good sign.. I would like to know whether there
> are any functions from Host AP in the back trace.

Sometimes PANICs show unreal stack traces (eg. adresses not within "0xc"
kernel region). I'll try to catch more info...

Regards
-- 
--Juanjo

#  Juan Jose Ciarlante (JuanJo PGP) jjo ;at; mendoza.gov.ar              #
#  Key fingerprint = 76 60 A5 76 FD D2 53 E3  50 C7 90 20 22 8C F1 2D    #