Re: [acpi-jp 2879] 5.1R: page fault from AcpiNsMapHandleToNode

On Fri, 12 Dec 2003, Georg-W. Koltermann wrote:
> On Fr, 2003-12-12 at 00:29, Nate Lawson wrote:
> > Please try 5.2-current if you can.  However, I'm not sure this is fixed
> > in it.  More info is needed.
>
> Will try one of the RC isos soon anyway, but given that it takes some
> hours to trigger I am not sure how much I'll get from that.
>
> > Parent pointer is invalid, as you show.  The panic was triggered by an
> > attempt to turn the fan on.
>
> I guessed that.
>
> > You should be able to trigger it quickly by
> > turning off CPU idling (sysctl machdep.cpu_idle_hlt=0) and doing something
> > CPU intensive (dd if=/dev/random of=/dev/null bs=1m).
>
> Notice that the fan turns on and off just fine during normal work, I use
> this machine as a swdev system and it does get loaded during builds.
> Didn't try turning off cpu idling yet, though.  Hmm the fan briefly
> turned off (???) for 10 secs when I started bzipping the dump, then went
> back on, then switched to high.  This is on AC where it usually keeps
> running on low speed even if idle.  It turns off when I run on battery.

There are different ways of turning on the fan.  acpi_thermal switches
them on based on the _ACx objects, which is what caused the panic.  So the
fan being turned on by the BIOS or through some other mechanism won't
trigger the fault.  We can also lower the value at which acpi_thermal does
the switch or just manually turn it on.  Output from sysctl hw.acpi would
help.

> >  Let's get some more
> > debugging info.
> >
> > Please send acpidump -t -d > georg-CpqEvo800w.asl
>
> acpidump doesn't like these switches. I'll attach the raw acpi dump from
> acpidump -o gwk.aml below, uuencoded.

Oh, for 5.1 it's acpidump > georg-CpqEvo800w.asl

> > Please send me a link to the bzip2 kernel.debug and dump you have from
> > above.
>
> vmcore.bz2 is 336M, will put it on some server at work tomorrow.  I
> don't have that much space from my private ISP.  Yes, 1GB RAM does have
> a drawback!

Don't bother, that's too big.  I thought we have support for sparse kernel
dumps?  If not, that's a big todo.

> > Also, please recompile your acpi.ko with ACPI_DEBUG, then set these in
> > sysctl after your system has booted:
> >     debug.acpi.layer="ACPI_THERMAL ACPI_NAMESPACE"
> >     debug.acpi.level="ACPI_LV_VALUES ACPI_LV_OBJECTS"
> >     hw.acpi.verbose="1"
> >
> > Then trigger the panic as described above.  You should get some helpful
> > dmesg output.
>
> I'll try tomorrow (getting late here), but might be difficult to
> trigger.

Ok.  You can manually turn on the active cooling with:
    sysctl hw.acpi.thermal.active=0
(or 1,2,3, etc.)  This is another way to trigger the panic.

-Nate