Re: apic errors and looping with 2.4, none with 2.2

[Mikael Pettersson <mikpe@csd.uu.se>]

Chris Stromsoe writes:
 > I have one machine that won't run 2.4.  As soon as a 2.4 kernel boots, it
 > starts throwing APIC errors.
 > 
 > The machine is a dual CPU pIII 933MHz system with 512Mb ram on a
 > SuperMicro motherboard, either a P3TDLR or a 370DLR, with the ServerWorks
 > LE chipset.  I'm booting using lilo with append="noapic".
 > 
 > As soon as I boot into a 2.4 kernel, I start getting APIC errors on both
 > CPUs.  Varying combinations of:
 > 
 > Mar 23 00:40:45 dahlia kernel: APIC error on CPU0: 02(08)
 > Mar 23 00:40:45 dahlia kernel: APIC error on CPU1: 01(08)
...
 > After a few hours of uptime, the box stops responding to keyboard input.
 > It begins printing the above messages to console over and over.  I have
 > several other identical machines that I received in the same batch that
 > run 2.4 without any problems (though they do seem to require "noapic").
 > 
 > It runs fine with 2.2 and is running 2.2.26 right now.

Like Maciej wrote, the machine's local APIC bus corrupts messages.
The fact that your other supposedly-identical machines don't have
this problem indicates strongly that this particular box has a HW
problem. It could be a weak power-supply, inadequate cooling, a
bad CPU, or a bad motherboard.

I just checked the 2.2.25 kernel and it doesn't seem to enable
APIC_LVTERR. You're problably getting APIC bus errors with 2.2
too, you just won't see them logged anywhere.

(Regarding your other boxes that need "noapic": you have tried
w/o ACPI, right? acpi=off or pci=noacpi.)

/Mikael