ACPI,SMP(?), and 5.2.1-RELEASE

[Nate Lawson]

On Thu, 8 Apr 2004, David E. Cross wrote:
> I recently upgraded my odl ASUS P2B-DS runnig BIOS version 1012 to
> 5.2.1-RELEASE (from 4.9.1) and noticed that the system is spending 50% of
> its time in interrupt while idle running any kernel (including GENERIC).
> When the system is booted without ACPI everything runs correctly and no
> interrupt load is observed.

Send the output of vmstat -i so I can see the interrupt load.
Try booting with:
    hint.apic.0.disabled="1"
This is likely an APIC issue if this fixes it.

> Additionally on shutting down the kernel messages after disabling ACPI are
> completely mangled.  Like 90% of the time on a "halt" the "System is
> halted\nSystem is safe to power down or press any key to reboot" is
> completely missing or mangled to the point of non recognition.  When it is
> visible I sometimes see a message like "Stray irq 2", and recently I got
> "stray irq 20\n  rq 20"  (just to give an example of how things are
> mangled after ACPI shutdown).  the "2" bit in the IRQ is consistent
> whenever I am able to observe it, however I don't have anything in dmesg
> on irq 2, 12, 20, 21, 22, or 23 (where the IRQs max out according to the
> IOAPIC line) and I cannot think of anything else that has "2" in the irq
> number.  So why am I not getting stray IRQ errors earlier?  Is there any
> way to tell on a running system which IRQ line is getting spammed and
> "why"?
>
> Any help on this would be greatly appreciated as I would like to run ACPI
> on this system so that I can do power offs for things like UPS-Battery
> failure, etc.

This also indicates an APIC problem.  jhb@ has a patch that hopefully will
be committed to allow disabling irq src overrides.  Just for kicks, also
reply with the output of "acpidump -t" on your system.

-Nate