Re: Linux 2.6.21-rc1

From: Daniel Walker
Date: Wed Feb 21 2007 - 13:25:49 EST


On Wed, 2007-02-21 at 19:18 +0100, Thomas Gleixner wrote:
> On Wed, 2007-02-21 at 09:38 -0800, Daniel Walker wrote:
> > > >
> > > > Could be the switch over then which confuses the NMI .
> > >
> > > Why? The switch just stops the PIT/HPET. It does not fiddle with IO_APIC
> > > and friends at all.
> >
> > I'm not an expert on the io-apic, but the check_timer() function seemed
> > to assume IRQ0 was happening regularly ..
>
> Again:
>
> check_timer() is called _BEFORE_ we even touch the local APIC timers. At
> this point PIT/HPET _IS_ firing IRQ0 with HZ frequency.

Right, but eventually there isn't a regular timer interrupt through the
io-apic. I don't think in the past IRQ0 stops without the system
crashing, so check_timer() could assume the timer (IRQ0) is _always_
regular.

do you know what the requirement are for routing the NMI through the
io-apic?

> > Well, I'm pretty sure it's HRT, cause in prior versions this only
> > happened when HRT is enabled. Then you guys went to the lapic all the
> > time, and now this is happening all the time ..
>
> The NMI is stuck:
>
> if (nmi_count(cpu) - prev_nmi_count[cpu] <= 5) {
> printk("CPU#%d: NMI appears to be stuck (%d->%d)!\n",
> cpu,
> prev_nmi_count[cpu],
> nmi_count(cpu));
>
> This has nothing to do with jiffies.

I think it has to do with IRQ0 . Did I mention this doesn't happen in
2.6.20 .

> There have been a bunch of changes in arch/i386/kernel/nmi.c as well.
>
> > You can't reproduce this?
>
> Nope.

Do you use nmi_watchdog=1 ?

Daniel

-
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/