CPU scaling issues

[bigdweeb]

I've been having problems with my CPU scaling for several months. I have a Core2 Duo T9600 in my Dell Latitude E6400. The problem I have is when I do something CPU intensive like compiling, the CPU will clock down to 800MHz and not allow me to manually scale it back up with cpufreq-set. This has at least been happening since 2.6.32-r7. I just upgraded to 2.6.34-r1 and it's still happening. I was originally using the ondemand CPU governor built into the kernel. I also tried compiling as modules a. I have gone back to builtin with the current kernel I just built, switching to the performance governor and removing the ondemand one. Does anyone have suggestions?

I don't think the CPU is overheating. I've been using the lm_sensors tools to check that. I don't know how to figure out what's triggering these events.

This is happening right now, and this is what cpufreq-info looks like.

[snilsson]

I notice that the ondemand govoner is not listed among the avaiable govoners.
In my setup I have ondemand compiled into the kernel at it lists that both in cpufreq-info and like this:

[alexandref75]

I have a similar machine, Studio XPS 16 with a T9550 and had a similar problem that was temperature related. When a certain threshold was achieved the BIOS would throttle the machine to 800Mhz and would only return after a reboot. It got much better with the new BIOS, A14, and a laptop cooler.
To test, check the temperature and speed with a intensive job in the background.

AFAIK, that is a common problem for this machines and DELL released new BIOS to fix (or mitigate) the problem.

[bigdweeb]

In my original post I had changed to userspace in order to try forcing the CPU speed back up to 2.80GHz.

I have checked the BIOS on my laptop and I'm running the most current version.

I rebooted this morning and added cpufreq.debug=7 to the kernel. I emerged something large and watched the log. It quickly throttled down to 800MHz again. I again tried switching to userspace and forcing the speed up. Here's what I got in my logs.

[bigdweeb]

Also, is there a way to figure out what threshold is being met that triggers this event? The fans are quite often running on this machine. The laptop does sit on a docking station almost all the time which props it off the table and allows air to flow underneath, so I would think it would do better with heat dissipation.

[snilsson]

This seems more tricky then I first thought.
I don't know much about the details but if it is the BIOS that makes the restriction it should be visible using

[bigdweeb]

If I look at the bios_limit, it is indeed set down to 800MHz.

[alexandref75]

I am using a script that sets the thermal cooling at /sys/class/thermal/cooling_device at certain thresholds. Setting a higher number at this cooling_devices (cur_state) reduces the CPU speed a little. This is better than the hard threshold that the BIOS set (800MhZ).

When the BIOS threshold is tripped and the machine is stuck at 800MHz, I had some luck of returning the normal speed by removing the ac adapter for 10-15 seconds. it is easier to do by checking the output of the script below:

#!/bin/bash

while true
do
sleep 1
echo `date` `cat /sys/class/thermal/thermal_zone0/temp` `cat /sys/class/thermal/thermal_zone1/temp` `cat /sys/class/thermal/thermal_zone2/temp` `cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq`
done


This script can also help to log the temperatures and speed to determine when the threshold of the BIOS is tripped.