FreeBSD 6.1-STABLE: Unexplained power off

[Android Andrew [:]]

Peter Jeremy wrote:
> On Tue, 2006-Aug-15 21:17:21 +0300, Apatewna wrote:
>> O/H Android Andrew [:] ??????:
>>> I've checked supply voltages by digital multimeter,
> 
> This won't show a noisy supply rail (eg due to high ESR capacitors).
> If you suspect power, check the rails under load (eg lots of disk
> seeks and I/O) using an oscilloscope.
I've checked voltages on MB contacts exactly under load (during port 
compilation). I have no oscilloscope, so I can't control voltage 
impulse, especially on multiple channels.
In this case I could only replace PSU for testing.

> 
>>> and temperature I've 
>>> checked "manually" by putting my hand on supply case
> 
> Assuming the PSU cooling fan is running, the temperature of the
> exhaust air would be more accurate.  (If the cooling fan isn't
> running, I'd check why).
It's ok with cooling fan, and temperature of the exhaust air is within 
the bounds of normal in comparison with other computers around me.

> 
>>> The same system has been working properly for the last 6 months under 
>>> amd64 version of FreeBSD 6.0-STABLE...
> 
> Are you in a position to revert and see if the system starts working
> stably?
I wouldn't like to do it because, first of all, it will take much 
additional time and some hardware (on-board network adapter, on-board 
sound card, video adapter) are not supported under amd64 platform. If I 
will not find a solution of this problem I'll have to revert.

> 
>> Check your BIOS to see if it has the option "CPU thermal shutdown".
>> I always disable it on client systems since it has the potential of 
>> driving you nuts with no apparent reason.
> 
> I've had my laptop thermal sensor glitch once but it logged an
> overtemperature event before it shutdown.  The downside of inhibiting
> the thermal shutdown is that if a real problem eventuates (and CPU
> fans do fail), you will destroy the CPU and maybe mobo.
I haven't found any overtemperature event but I'll try to disable "CPU 
thermal shutdown" for a testing time.

> 
> It might be worthwhile setting up a serial console and logging it
> on another box to see if anything is written to the console before
> it dies.

It's great idea, I'll try it. Do I need a null-model cable to do it?