2004-11-17 17:39:35

by Joerg Sommrey

[permalink] [raw]
Subject: local-/io-apic nmi watchdog failing on S2466

Hello,

I'm still having problems with nmi watchdog on my S2466 board. I tried lots
of different configurations with a large number of 2.6 kernels (vanilla,
-mm, -ac) all with the same result: no working nmi watchdog, neither with
local- nor with io-apic. I still wonder if anybody out there has ever
succeeded with a working nmi watchdog on Tyan Tiger MPX.

The symptoms are:

nmi_watchdog=1:
===============
dmesg:
testing NMI watchdog ... CPU#0: NMI appears to be stuck!
/proc/interrupts:
no NMI count

nmi_watchdog=2:
===============
dmesg:
testing NMI watchdog ... OK.
/proc/interupts:
NMI count increments
recovery from lockup:
none

nmi_watchdog=2 clock=pit:
=========================
dmesg:
testing NMI watchdog ... CPU#0: NMI appears to be stuck!
/proc/interrupts:
NMI count increments, but rate is ~ 1/20s
recovery from lockup:
none

The lockup-test is done with a little program that Ingo posted on this
list:

int
main(void) {
iopl(3);
while (1)
asm("cli");
return 0;
}

The only reaction I can see from this test: after some seconds (5+) the LEDs
on the keyboard start blinking when nmi_watchdog=2 and clock!=pit.
Always need to hit the reset button :-(

What else could I try? Are there any BIOS-settings relevant to a
working nmi-watchdog? What information is needed to track down this
problem?

Maybe it's the board's failure, but as there *are* counted NMIs I still
hope there is a software solution to this problem.

Thanks,
-jo

--
-rw-r--r-- 1 jo users 63 2004-11-17 17:45 /home/jo/.signature


2004-11-17 21:42:09

by Chris Wedgwood

[permalink] [raw]
Subject: Re: local-/io-apic nmi watchdog failing on S2466

On Wed, Nov 17, 2004 at 06:31:18PM +0100, Joerg Sommrey wrote:

> Maybe it's the board's failure, but as there *are* counted NMIs I
> still hope there is a software solution to this problem.

All 2466 boards seem to do be broken like this, maybe you can contact
Tyan about it and see if they have any comment?

2004-11-17 23:08:43

by Joerg Sommrey

[permalink] [raw]
Subject: Re: local-/io-apic nmi watchdog failing on S2466

On Wed, Nov 17, 2004 at 01:33:49PM -0800, Chris Wedgwood wrote:
> On Wed, Nov 17, 2004 at 06:31:18PM +0100, Joerg Sommrey wrote:
>
> > Maybe it's the board's failure, but as there *are* counted NMIs I
> > still hope there is a software solution to this problem.
>
> All 2466 boards seem to do be broken like this, maybe you can contact
> Tyan about it and see if they have any comment?
>
Just sent a problem report to Tyan. Let's see...

-jo

--
-rw-r--r-- 1 jo users 63 2004-11-17 23:09 /home/jo/.signature