Using acpi: thermal/sysfs-api To Extend Battery Life

[dd_wizard]

This link, acpi: thermal/sysfs-api, explains how the new thermal management sysfs class is built, but doesn't give much information about using it. Using watch, I can see that the cur_state of cooling_device2 changes from 0 to 5 when I check "Dim display when idle" in Power Management Preferences. But I haven't found an applet that changes cooling_device0 or cooling_device1.

Echoing different integers to the cur_state files limits the maximum cpu frequency for cpu0 and cpu1, respectively. This behaviour is expected from what I've read, and mimics the options in Windows power manager for extending battery life by throttling the CPUs. I've had no luck with google and local man pages, so has anybody has seen an applet for controlling /sys/class/thermal/cooling_device[0|1]/cur_state?

On a side note, a value of 1 does slow the CPU down, but it will still hit 100C (normal for an Intel mobile duo core). However, values of 2 and larger throttle enough to lower the maximum CPU temp. Since the CPU temp is a good indicator of power consumption, it's pretty obvious that these two cur_state files are intended to extend battery life.

dd_wizard

---------- Post added at 02:02 PM CST ---------- Previous post was at 01:59 PM CST ----------

Bah! Can someone change the thead title to his one for me, please?? TITLE: Using acpi: thermal/sysfs-api To Extend Battery Life

[GallComp]

Originally Posted by dd_wizard

Bah! Can someone change the thead title to his one for me, please?? TITLE: Using acpi: thermal/sysfs-api To Extend Battery Life

Done...

[dd_wizard]

Since nobody replied, I decided to figure out how things work. The following bash script has been tested in Fedora 12, and most likely won't work in other versions. First, the script:

#!/bin/sh
# Script to throttle cpu speed.
# Works for any number of cpus, but only tested on Fedora 12.
# I have no idea how long this will be around,
# see http://patchwork.kernel.org/patch/43071 for more info.

if [ -n "$1" ]
then
  let "state = $1 + 0"
  if [ "$state" -ne "$1" ] || [ "$state" -lt "0" ]
  then
    echo "Usage: `basename $0` [ non-negative integer ]"
    exit 1
  fi
  for device in `ls -d /sys/devices/virtual/thermal/cooling_device*`
  do
    if [ "`cat $device/type`" = "Processor" ]
    then
      let "max_state = `cat $device/max_state`"
      if [ "$state" -gt "$max_state" ]
      then
        echo $max_state  > "$device/cur_state"
      else
        echo $state  > "$device/cur_state"
      fi
    fi
  done
fi
echo "Current CPU thermal cooling states:"
for device in `ls -d /sys/devices/virtual/thermal/cooling_device*`
do
  if [ "`cat $device/type`" = "Processor" ]
  then
    echo "$device = `cat $device/cur_state` "
  fi
done
exit 0

For testing, I ran two threads of Prime95 to push both CPUs to 100%. Setting cur_state to 1 capped the CPU frequency at roughly 75%, 1.33GHz on a 1.73GHz duo core processor. It's a laptop, so both CPUs bumped against 99C even at 1.33GHz. Setting cur_state to 2 dropped the frequency to 800MHz, and dropped the temperatures to 86C. With just one thread running, which simulates one runaway process, the maximum temperature was 94C for a cur_state of 1. I'm not sure if it's my BIOS, or Fedora, but the thread switches between CPUs to balance the temperatures. See the attached screenshot.

This mimics the behaviour of the Windows power manager. The default maximum battery life scheme for Toshiba laptops limits the CPU frequency to 50%.

dd_wizard

P.S. You can run the script as an unprivileged user to display the current states, but you need to run it with su -c "cputhrottle 2" to write to the files.