On 21st Oct 2005, Christos Zoulas wrote:
> > My 3.0_BETA i386 box with i82550 network (fxp0) occasionally panics. 
> > When it does so, the port light on my switch flashes like mad and the whole 
> > network grinds to a halt.
> >
> > trace says (from handwritten notes):
> >
> > kernel: page fault trap code=0
> > Stopped at netbsd:llcintr+0xcc: movzbl 0x2(%edi),%eax
> > llcintr(x0f98e00,c01022cd,c0f98e00,c0102bde,4a360010) at netbsd:llcintr+0xcc
> > ccittintr(4a360010,20030,27bc0010,10,c08f4000) at netbsd:ccittintr+0x35
> > DDB lost frame for netbsd:Xsoftnet+0x82 trying 0xc08f7e80
> > Xsoftnet() at netbsd:Xsoftnet+0x82
> > --- interrupt ---
> > 0x240:
> >
> > As soon as I type reboot, the network starts working again. Last time 
> > it died, I tried tcpdumping on another machine, but didn't see anything.
> >
> >This machine's been running 1.6 for years without any downtime.
> >
>
> Looks like a bug in netccitt.

I see the code is described as pretty stagnant (though I can't remember 
where I saw the reference). There seems to be some regression here, should 
it be enabled by default in GENERIC?

> Compile a kernel without it. Can you put the kernel and a dump somewhere?

The latter depends on not running the former. :-)

I've got a kernel and a dump, but I guess it would be better with debug 
symbols compiled into the kernel. I'll compile that up and wait for 
another crash (they seem to happen every 5 days or so).

-- 
Stephen