[olug] Need guru help

[Matthew G. Marsh]

On 17 Jun 2002, Carl Lundstedt wrote:

>
> > What kernel version - and below by stock I assume you really mean "Vendor
> > Patched" kernel?
>
> Stock == 'Vendor patched' (In this case 2.4.18-6mdksmp)
> Recompiled == 2.4.18 (grabbed from kernel.org, no patches)

Ah - ok. Hmmmm.

> > Is your system BIOS fully MP 1.4 compliant? If you have a choice of MP
> > versions try setting the version to 1.4 (assuming a 'true' stock kernel)
> > and then seeing. If you still get the errors then try 1.1 - if the errors
> > go away with 1.1 bummer as that would imply you have a single APIC vector
> > table. 8-{
>
> I have no idea.  How can this be determined and set?

Most system BIOSes that I have seen that allow this setting usually bury
it in the "Advanced" settings although I have also seen it under the
PNP/PCI settings as well. Usually it is called the MP Compliance setting
although I have seen it called the MP Version and the PIC Table. If you
have it the only two settings are "1.1" which is the older original x86
SMP specification for x>=5 and 1.4 which is the latest (~2-3 years old)
version. If your board has such a toggle sometimes setting the 1.1 version
will help with APIC errors as 1.1 had a looser specification for the
interrupt routing (APIC). Otherwise if it is set to 1.1 then I would try
1.4 as the 2.4.xx series kernels work better with the full compliance.

> > > I didn't see these with with the stock SMP kernel.  I can change the
> > > frequency by monkeying with some kernel settings before recompile (these
> > > errors were coming in every 15 seconds), but I'm not sure (now) what
> > > settings helped curb them (I may have just shut down the APIC
> > > reporting.)

Hmmm - on rereading the above para you may want to check and see what
patches the MDK pople have put into the kernel. They may have a patch
against that particular problem only (one of many reasons I asked about
the definition of stock - most vendors have all sorts of tweaks they
apply)

Also check out the -ac series of kernels as Alan is wont to have those
type of patches in place (and MDK follows RH faily closely).

> > What settings - and 'stock' = ????
> stock==see above.
> Doing a diff on the .config and .config.old for the kernel compile
> (which cut back the number of APIC errors and the lost interrupts):
>
> < # CONFIG_MICROCODE is not set
> < # CONFIG_X86_MSR is not set
> < # CONFIG_X86_CPUID is not set
> ---
> > CONFIG_MICROCODE=y
> > CONFIG_X86_MSR=y
> > CONFIG_X86_CPUID=y
> 93,95c93,111
> < # CONFIG_PM is not set
> < # CONFIG_ACPI is not set
> < # CONFIG_APM is not set
> ---
> > CONFIG_PM=y

Mmmm - do you really need/want Power management on an SMP server? It is
not a "good thing" usually due to the multi interrupt structure. Oh wait -
is this the .config or the old.config?

> <snipped a bunch of power management options>
>
> 319c335
> < CONFIG_BLK_DEV_ATARAID_PDC=y
> ---
> > # CONFIG_BLK_DEV_ATARAID_PDC is not set
> 556a573
> > # CONFIG_SERIAL_ACPI is not set
>
> As a further note, this machine is does not receive (nor will it ever
> receive) heavy traffic.  The machine just needs to be UP and able to see
> hard drives on the dang Promise RAID controller (which was the whole
> reason that I had to tinker with the kernel in the first place).  If I
> need to shut off the SMP ability of the MB, that is an option since I
> think a single processor will handle the traffic fine (1 Ghz is probably
> overkill).

Given that the MDK SMP kernel did not give these errors then there is some
tweak that may be needed. I suspect that trying an -ac series may work
without mods. Also turn off the Power Management stuff.

> The second net interface on the board is un-used.  (Mostly this is a
> reply to Nick's response about ping floods.  I may go with the
> resolution of -noapic in lilo.)

That would be worst case only. They performance and possible deadlocks
that brings may not be worth it.

> Thanks, feeling stupid(er),

Nah - Think of Linux as the best exercise a mind can get... ;-}

> Carl Lundstedt
> UNL

--------------------------------------------------
Matthew G. Marsh,  President
Paktronix Systems LLC
1506 North 59th Street
Omaha  NE  68104
Phone: (402) 932-7250
Email: mgm at midwestlinux.com
WWW:  http://www.midwestlinux.com
--------------------------------------------------


-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_

For help contact olug-help at olug.org - run by ezmlm
to unsubscribe, send mail to olug-unsubscribe at olug.org
or `mail olug-unsubscribe at olug.org < /dev/null`
(c)1998-2002 OLUG http://www.olug.org

-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_