Re: NMI error

[Thomas HERAULT]

On Wed, 25 Jul 2001 04:38:51 -0700 (PDT)
Kannan Soundarapandian <[e-mail]> wrote:

> Hello,
> 
> I am a grad student at the Oregon State univ. I have
> the same problem as you. I have a dual p3 1gig
> processor system with the Asus CUR-DLS motherboard
> (which uses the serverworks LE chipset), and 512 mb 
> Registered samsung RAm as one module. I am also using
> a quantum atlas10k2scsi hdd with an on motherboard
> symbios controller.
> 
> I get the following error continuously.. after which
> the system dies! I'm using redhat7.1.
> 
> Can u please help me fix the problem??
> 
> Uhhuh. NMI received. Dazed and confused, but trying to
> continue 
> Do you have a strange power saving mode enabled?
> 
> Please help. Thank you very much
> 
> Kannan
> 
> 
> =====
> --
> Kannan Soundarapandian,
> OSU, Corvallis.
> 
>

Hi.
I *almost* "solved" this problem :
Since linux-2.4.2-ac18, by default nmi_watchdog is disabled.
If you re-enable it with nmi_watchdog=1 as kernel boot parameter,
the symptom disapear.
But it is only a work around which works : the fact is that in
arch/i386/kernel/traps.c, there is a test,
  if (nmi_watchdog)
    {
        nmi_watchdog_tick(regs);
        return;
    }
Thus, our NMIs are catched an treated like watchdog NMIs,
but even if you disable nmi_watchdog (like it is by default),
these NMIs occure on my (our ?) CPUs (approximatively 300 interruption / second)
and produces these annoying error messages.

On other multi-processors motherboards I know, when you set
nmi_watchdog=0, the cat /proc/interrupts says that there is
0 NMI on every CPU ; on mine, I still have 300 NMI / sec.

For those of the lkm which are interested in this problem,
my motherboard is a 694D Pro (MS-6321) from MSI, with
a VIA VT82C694X chipset and an Apollo Pro133A north bridge,
featuring 2 Pentium III at 800 Mhz. Every kernel I tested
(2.2.x included) had the same problem.
The reasons of the interruptions are (2d then 3d)*

Hope that help and that someone could help us find what
produces these interruptions.