Panic on 6.1REL using linux compat

[Pieter de Goeje]

On Monday 28 August 2006 17:00, you wrote:
> On Sunday 27 August 2006 15:02, Pieter de Goeje wrote:
> > Hi John,
> >
> > On Monday 21 August 2006 18:49, you wrote:
> > > On Saturday 19 August 2006 17:59, Pieter de Goeje wrote:
> > > > The box in question is a dual core athlon64, running 6.1-RELEASE-p3
> > > > amd64
> > >
> > > with
> > >
> > > > SMP enabled. The box has been panicing every month or so, and i've
> > > > obtained
> > >
> > > a
> > >
> > > > crashdump the last time. The primary use for the box is hosting
> > > > Half-Life Dedicated Servers, using linux compatibility. One of them
> > > > was the current process seen in the panic below.
> > >
> > > [snip crashdump]
> > >
> > > This doesn't make any sense.. m0 is passed directly to m_length, it's
> > > value shouldn't change.  You either have a random memory corruption bug
> > > or you have faulty hardware.
> >
> > Well today I had exactly the same panic. In my experience faulty memory
> > leads to completely unrelated crashes, which makes me wonder what caused
> > this panic.
>
> On a whim, I'd do a make clean and rebuild your entire kernel from scratch
> and try again.  I had a recent frustration at work with amd64 boxes where
> the kernel seemed to be corrupting a call safe register during a simple
> leaf function call but would do some on multiple boxes, and always the same
> corruption (a pointer magically became NULL).  Recompiling a fresh kernel
> from scratch "fixed" it.

Ok, I'll give that a try.
>
> > Also I forgot to mention that I have a linux binary which uses sockets
> > and only runs correctly when used with debug.mpsafenet=0, which made me
> > believe there was some issue with SMP & Linux compat. Unfortunatly I
> > don't have it's source.
>
> Hmm, I've cc'd Robert Watson (though he's on travel in India currently).