Re: NMI watchdog detected lockup

[Randy.Dunlap]

Andi Kleen wrote:
> On Mon, 18 Oct 2004 10:13:11 -0700
> "Randy.Dunlap" <rddunlap@xxxxxxxx> wrote:
>
>
> > Marc Bevand wrote:
> >
> > > On 2004-10-17, Randy.Dunlap <rddunlap@xxxxxxxx> wrote:
> > > | 
> > > |  I'm seeing this often during a kernel build on AIC79xx.
> > > |  I did one kernel build on SATA without seeing this.
> > > |  This is on a dual-Opteron IBM Workstation A with
> > > |  2 GB RAM, SATA, & SCSI.
> > > |  [...]
> > > |  NMI Watchdog detected LOCKUP on CPU0, registers:
> > > |  [...]
> > >
> > > You are not the first one to observe frequent watchdog timeout
> > > lockup on dual Opteron systems during intense I/O operations,
> > > see this thread:
> > >
> > >  http://thread.gmane.org/gmane.linux.ide/1933
> > >
> > > Note: this does *not* seem to be SATA-related.
> >
> > Hi,
> >
> > Zwane suspected NMI spikes and advised me to disable nmi_watchdog
> > (nmi_watchdog=0).  After doing that, a kernel build completes
> > successfully, although with many messages like these:
> >
> > Uhhuh. NMI received for unknown reason 21.
>
>
> Something on your system creates bogus NMI interrupts. What chipset
> are you using exactly?
>
> Sometimes chipsets can be programmed to raise NMIs when an PCI bus
> error occurs. 
>
> 21 is the normal state (PIT timer running, but no errors logged) 
>
> If you have an AMD 8131 it could be in theory erratum 54, but then
> normally one of the error bits in reason should be set.

Yes, it's an AMD-8111 / 8131 / 8151 / K8-northbridge machine.

--
~Randy
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/