Problems with SMP on 6.1-STABLE-200608

[youshi10 at u.washington.edu]

On Wed, 28 Mar 2007, Lowell Gilbert wrote:

> "Don O'Neil" <lists at lizardhill.com> writes:
>
>> I've been having problems with my server freezing up, having the #2 CPU
>> 'shut down', kernel panics, and all sorts of nastyness....
>>
>> Originally I thought it was exim, or possibly bind, or bad hardware (mb, cpu
>> or memory)... I've swapped out the motherboard & CPU's & memory from an old
>> server that was running 4.11 ROCK SOLID for years...
>>
>> At first I thought the problem was solved, but now it's popping up again...
>> The 2nd CPU gets 'shut down', or kernel panics, esentially taking the system
>> offline.
>
> There are lots of things this could be, and I certainly wouldn't rule
> out hardware problems (power supply?).  Figuring out the problems
> directly would certainly involve looking at more details than you're
> listing here.
>
>> If I install a single CPU (non-smp) kernel, then the system works fine... (I
>> did this on the old motherboard before I swapped it out, and it worked fine
>> too).. So I'm wondering if there is an SMP bug or problem I'm running into.
>>
>> I'm running 6.1-STABLE-200608, an ISO image I downloaded from the archives
>> when I built the box (NOT 6.1-RELEASE).
>
> The whole point of making releases is that it's much easier to support
> a small number of known reference software configurations.
>
>> I'm runining an Intel Serverworks motherboard with 2 1.4 GHz PIII's... The
>> problem only seems to show up under high load.
>
> I don't think I've heard of anything similar.  I think there are a
> bunch of these boards out there.
>
>> I'm wondering what I should do here...
>>
>> I'm concerned about doing a binary upgrade to 6.2 won't fix the problem, and
>> I've tried using freebsd-update, but it complains about the version not
>> being compatible.
>>
>> If I do a binary upgrade from CD, will it also update the kernel sources so
>> I can build a new one? Will it complain about it not being compatible?
>
> It can give you the sources; that's a menu option during the install.
> That should work fine.
>
>> Is there a way to 'force' the ID of the system to be 6.1-RELEASE so that
>> freebsd-update will work?
>
> Well, yes, but there's a reason for the check, you know...
>
>> Will doing the 6.1-6.2 binary upgrade as posted by Colin also update the
>> kernel sources?
>
> I don't know what procedure he described, so I don't know.  But if you
> update to 6.2-RELEASE, then it will be easy to get the right sources
> afterwards.  Again, that is the advantage of having releases.
>
>> Would my best option really be to start over with a fresh install rather
>> than upgrade? (this would be painful)
>
> If it's that painful, you'd probably be well served to have a spare
> system to stage changes on.  In addition to being good risk
> management, it saves you time, which is worth something too.
>
>> I'm going to try to test out 6.2 on the old MB/CPU combo to see if I can
>> re-create it under 6.2 as well before I do anything. As well as try doing an
>> upgrade on the bench from CD from 6.1-STABLE-200608 to 6.2-RELEASE... Since
>> this is a production server (and for months it was burned in with no
>> apparent issues) I only have 1 shot at this to do it right.
>>
>> Any help/recomendation would be appreciated.
>
> Good luck.

Honestly I would probe around your motherboard a bit checking voltages (power supply) and/or heat dissipation, because those are the most likely cases if it _only_ fails under high load. Next thing to check would be RAM integrity.

-Garrett