My system keeps shutting down when running a standard KnoppMyth R5F27 install. I have a celeron (Pentium 3), and I tested the system by running some CPU burn-in / stress test software, which had no problem running indefinitely. I also previously had an earlier version of KnoppMyth R5E50, and it did not have the problem. I finally managed to capture some output that shows the problem occurring. Any ideas why KnoppMyth suddenly decided to care about the temperature of my CPU, and also why it gets it *wrong* ?

Message from syslogd@mythtv at Fri Sep 21 00:25:31 2007 ...
mythtv kernel: Critical temperature reached (-264 C), shutting down.

Broadcast message from root@mythtv (Fri Sep 21 00:25:32 2007):

The system is going down for system halt NOW!

There are any number of variables that could be causing overheating. If it is overheating. A stuck fan? -I've had stray power wires from the fan stop the blades turning fall into the blades and causing overheating. Have you moved it?
If you've moved it from where you tested it perhaps there is insufficient ventilation. Have you taken the heatsink off at any point? Could just be a case of reseating the heatsink?
What sorts of temps are being reported in the bios?
On your install did you opt for the monitoring software? Could be related to that?
I am guessing here.

_________________
Version:R8
Intel C2D 7400, Nvidia 5600 via HDMI to Samsung B37B650TW (PAL), Asus P5QL-E mobo, 4Gb PC6400 DDR2 ram, Samsung Spinpoint 500 Gb & 1Tb drive, Nova-HD-S2 (x2)

I don't think -264c is an overheating problem :p

Check the temp in your bios, see what it says, maybe it's reading it wrong and it's a bad sensor?

_________________
"The amount of time needed to solve a problem is inversely proportionate to the complexity of the solution" -- Me

KM: R5.5
CPU: Athlon 3800+
Vid: nvidia 7300GT
Snd: Chaintech av-710
Tuner: PVR150
HD: 1000gb sata + 750gb sata + 500gb usb

Computer temp reporting is notorious for being inaccurate- yes, even in systems with on-die sensors reading the temperature.

Here are some of the reasons:

1) Even on-die sensors read only a specific spot of the cpu core, and temperature does vary inside a cpu
2) BIOS code takes digital data and converts it to a temp, even in BIOS- a change in code will change the indicated temp
3) Software which uses the reported BIOS temp can also adjust the indicated temperature
4) Airflow in a case can be misleading- a surface mount sensor with air blowing directly on it may read much cooler than the surrounding components may actually BE
5) Software that monitors temps may try to read a sensors that is not actually in use- most systems have at least 3 POSSIBLE sensors, but it is common for only 2 to be in use

I actually suspect that the issue you are having is the last in that list- the sensor that is being read is NOT actually in use and a "ghost" temp (which is invalid) is being read.

That being said, a physical check of the cpu IS the first thing to do!
Make sure the heatsink is not clogged and that the fan spins. Make sure the heatsink is not loose- if it is not evenly, securely attached to the cpu, temp CAN get hot quickly. (Generally speaking a P3 heatsink should not tilt at all, but it should be possible to twist it slightly- be careful!)

If the heatsink feels secure, you can carefully touch the heatsink with a finger while running and see how warm it feels...
This is inaccurate, of course, but if it feels like it is on fire, you have a problem.
If not, and the heatsink feels secure you are PROBABLY ok. You'd have to remove and reinstall it to be sure, and if you aren't experienced at this, could make a new problem.

I'm still gonna bet on reading an invalid sensor, myself.

I do know that it is not actually overheating. Like I said, I tried running some CPU burn-in/stress tests, and I let them run overnight and the CPU did not overheat. I would like to know why specifically R5E50 did not have the problem and R5F27 does. Also, does anyone know if I can either calibrate the temp sensor, or tell KnoppMyth not to check it? The message seems to indicate it is a kernel thing. I found a post on another forum to try putting acpi=off apm=off on the kernel command line, so I am trying that.

I have not yet installed R5F27, so I am not sure what it is doing with sensors.

In general I know a little about linux and sensor use though-

Test something-
get a command line running and type sensors

That should get you a readout of any working sensors in the system.

Usually, if no config has been done, you will either get nothing or just one sensor reading. The single reading is supposed to indicate cpu temp as reported via apci, I think- and it sometimes works, sometimes does not.

If you get MORE readings, the install is trying to configure lm_sensors, but I am betting there will be just one.

You could try running sensors-detect which is the config program for lm_sensors and is more likely to generate realistic readings, as well as MORE of them. As long as your chipset is supported.

Keep in mind this is with acpi=off on the kernel command line, so it is probably not as helpful. However, I should point out that with that, my mythtv box has stayed up for about 5 days now, and it's recording programs just fine.

root@mythtv:/home/mythtv# sensors
wm8775-i2c-1-1b
Adapter: ivtv i2c driver #0

cx25840-i2c-1-44
Adapter: ivtv i2c driver #0

tveeprom-i2c-1-50
Adapter: ivtv i2c driver #0

TCL 2002N-i2c-1-61
Adapter: ivtv i2c driver #0

Ok, it looks like everything reported is related to ivtv- none of them are actual temp sensors.

Any temp reading you are getting is from the very limited support built into a kernel. (Note that I have one system whose sensor chip is NOT supported by lm_sensors and I do get a ballpark reading; I do think the reading is is acpi related, though I have not investigated much.)

I still have not tried the latest on a test machine yet, and I should.

I only have a few possible suggestions for you, and make no claim to them being the best options or even if they might not alter some config you need...

Use at your own risk!

Try detecting all the sensors on the system- sensors-detect will run the config utility if it is installed. If it is not installed, lm_sensors is what you need to install (alternative package spelling is sometimes lm-sensors.)
The lm_sensors readings may take precedence over acpi, I think.
I seem to recall reading somewhere of difficulty with lm_sensors and ivtv on some hardware combinations, so it is possible that this could cause problems.



Finding out just what is monitoring temps may allow you to adjust the setpoint or what is monitored and get rid of the issue.

I must admit that I have seen erroneous reading like yours before, but never had anything which by default had a shutdown temp configured and active.

I am not sure if having acpi on would change the report from sensors or not- never have tried.