Page fault in FreeBSD 5.3 on IBM e325, Dual AMD64 2.2GHz, 4GB RAM, ServeRAID 6M - problem goes away without TCP sack

Robert Watson rwatson at freebsd.org
Wed Nov 24 03:14:44 PST 2004


On Wed, 24 Nov 2004, Ganbold wrote:

> At 06:43 PM 11/22/2004, you wrote:
> 
> > >
> > > It seems to me the problem is related to network stack and threading.
> > > Am I right? How to solve this problem?
> >
> >I've seen reports of this problem with and without debug.mpsafenet=1,
> >which suggests it is a network stack bug but not specific to locking. I've
> >also seen reports that disabling TCP SACK will make the problem go away,
> >which would be good to confirm.  I spent the weekend building up some more
> >expertise in TCP and reading a lot of TCP code, and hope to look at this
> >problem in more detail today.  You may want to try turning off TCP sack
> >using net.inet.tcp.sack.enable=0 in sysctl.conf (or loader.conf).
> 
> I turned off TCP sack using net.inet.tcp.sack.enable=0 in sysctl.conf
> and it seems like the problem goes away.  It is working for more than 20
> hours without any crash.  Robert, did you find anything in network stack
> code? I'm just curious. 

I have not yet identified a (the?) bug, although it seems likely that it
has to do with the internal TCP state relating to the SACK blocks.  What
we need right now, I think, is a core dump, preferably on an i386 kernel
since our debugging tools work best with that.  My recollection, though,
is that you're using an amd64 kernel, but there have been reports from
others that likely are on i386.

Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert at fledge.watson.org      Principal Research Scientist, McAfee Research





More information about the freebsd-current mailing list