Linux Kernel – Setting CPU Governor to On Demand or Conservative

cpu-frequencylinux-kernelpower-management

I'm struggling with cpupower on ArchLinux. I want to set governor to ondemand or even to conservative.

First if I do $ sudo cpupower frequency-info --governors, I only get performance powersave.

So I look for available modules like this

ls -1 /lib/modules/`uname -r`/kernel/drivers/cpufreq/

…and I get

acpi-cpufreq.ko.gz
amd_freq_sensitivity.ko.gz
cpufreq_conservative.ko.gz
cpufreq_powersave.ko.gz
cpufreq_stats.ko.gz
cpufreq_userspace.ko.gz
p4-clockmod.ko.gz
pcc-cpufreq.ko.gz
powernow-k8.ko.gz
speedstep-lib.ko.gz

So, first of all no modules for "ondemand" seems to be available. What do I miss?

Then I try to enable at least conservative:

$ sudo modprobe cpufreq_conservative

then I check the module is actually loaded

$ lsmod | grep cpufreq

and check if it is now avaliable

$ sudo cpupower frequency-info --governors

but unfortunately I still get the same: performance powersave only,
and if I try to enable conservative

$ sudo cpupower frequency-set -g conservative

It says that the module is not avaliable.

So basically I have two questions:

What do I need to install in order to have ondemand module
How can I enable it?

Best Answer

Assuming your governor is the intel_pstate (default for Intel Sandy Bridge and Ivy Bridge CPUs as of kernel 3.9). This issue is not specific to Arch, but all distros using the new Intel pstate driver for managing CPU frequency/power management. Arch linux CPU frequency scaling.

Theodore Ts'o wrote his explanation on Google+:

intel_pstate can be disabled at boot-time with kernel arg intel_pstate=disable
The problem with the ondemand governor is that it doesn't know the specific capabilities of the CPU
Executing some tasks with higher frequency will consume less power than would a lower frequency taking more time e.g. arithmetic stuff, but not true for all tasks e.g. loading something from memory
The intel_pstate driver knows the details of the how the CPU works and it does a better job than the generic ACPI solution
intel_pstate offers only two governors, powersave and performance. Intel claims that the intel_pstate "powersave" is faster than the generic acpi governor with "performance"

To change back to the ACPI driver, reboot and set the kernel arg intel_pstate=disable
Then execute modprobe acpi-cpufreq and you should have the ondemand governor available.

You can make the changes permanent by editing /etc/default/grub and adding

GRUB_CMDLINE_LINUX_DEFAULT="intel_pstate=disable"

And then updating grub.cfg ala grub-mkconfig -o /boot/grub/grub.cfg

Follow the instructions for Arch kernel module loading and add the acpi-cpufreq module.

For 3.x kernels

The interface to CPUFreq has changed in the newer kernels. This would include CentOS 6. You can read about the entire interface here in the Red Hat Enterprise Linux (RHEL) documentation titled: Chapter 3. Core Infrastructure and Mechanics.

Specifically the section on CPUFreq Setup. Here are the steps required to set it up.

CPUFreq drivers

$ ls -1 /lib/modules/`uname -r`/kernel/arch/x86/kernel/cpu/cpufreq/
acpi-cpufreq.ko
mperf.ko
p4-clockmod.ko
pcc-cpufreq.ko
powernow-k8.ko
speedstep-lib.ko

load appropriate driver

$ modprobe acpi-cpufreq

install cpupower tool

$ yum install cpupowerutils

view governors

$ cpupower frequency-info --governors
analyzing CPU 0:
ondemand userspace performance

So we currently only have these 3 governors loaded: ondemand, userspace, and performance.

loading governors that are missing

You can get a list of all governors that are available like so.

$ ls -1 /lib/modules/`uname -r`/kernel/drivers/cpufreq/
cpufreq_conservative.ko
cpufreq_ondemand.ko
cpufreq_powersave.ko
cpufreq_stats.ko
freq_table.ko

$ modprobe cpufreq_powersave

confirm modules thus far:

$ lsmod |grep cpuf
cpufreq_powersave       1196  0 
cpufreq_ondemand       10544  8 
acpi_cpufreq            7763  0 
freq_table              4936  2 cpufreq_ondemand,acpi_cpufreq
mperf                   1557  1 acpi_cpufreq

confirm which governors are loaded

$ cpupower frequency-info --governors
analyzing CPU 0:
powersave ondemand userspace performance

viewing current policy

$ cpupower frequency-info
analyzing CPU 0:
  driver: acpi-cpufreq
  CPUs which run at the same hardware frequency: 0 1 2 3 4 5 6 7
  CPUs which need to have their frequency coordinated by software: 0
  maximum transition latency: 10.0 us.
  hardware limits: 1.60 GHz - 3.20 GHz
  available frequency steps: 3.20 GHz, 3.20 GHz, 3.07 GHz, 2.93 GHz, 2.80 GHz, 2.67 GHz, 2.53 GHz, 2.40 GHz, 2.27 GHz, 2.13 GHz, 2.00 GHz, 1.87 GHz, 1.73 GHz, 1.60 GHz
  available cpufreq governors: powersave, ondemand, userspace, performance
  current policy: frequency should be within 1.60 GHz and 3.20 GHz.
                  The governor "ondemand" may decide which speed to use
                  within this range.
  current CPU frequency is 1.60 GHz (asserted by call to hardware).
  boost state support:
    Supported: yes
    Active: yes
    2500 MHz max turbo 4 active cores
    2500 MHz max turbo 3 active cores
    2500 MHz max turbo 2 active cores
    2600 MHz max turbo 1 active cores

In the above output you can see my current policy is ondemand. To tune the policy and speed you use this command to do so:

$ cpupower frequency-set --governor performance
Setting cpu: 0
Setting cpu: 1
Setting cpu: 2
Setting cpu: 3
Setting cpu: 4
Setting cpu: 5
Setting cpu: 6
Setting cpu: 7

confirm new governor

$ cpupower frequency-info
analyzing CPU 0:
  driver: acpi-cpufreq
  CPUs which run at the same hardware frequency: 0 1 2 3 4 5 6 7
  CPUs which need to have their frequency coordinated by software: 0
  maximum transition latency: 10.0 us.
  hardware limits: 1.60 GHz - 3.20 GHz
  available frequency steps: 3.20 GHz, 3.20 GHz, 3.07 GHz, 2.93 GHz, 2.80 GHz, 2.67 GHz, 2.53 GHz, 2.40 GHz, 2.27 GHz, 2.13 GHz, 2.00 GHz, 1.87 GHz, 1.73 GHz, 1.60 GHz
  available cpufreq governors: powersave, ondemand, userspace, performance
  current policy: frequency should be within 1.60 GHz and 3.20 GHz.
                  The governor "performance" may decide which speed to use
                  within this range.
  current CPU frequency is 3.20 GHz (asserted by call to hardware).
  boost state support:
    Supported: yes
    Active: yes
    2500 MHz max turbo 4 active cores
    2500 MHz max turbo 3 active cores
    2500 MHz max turbo 2 active cores
    2600 MHz max turbo 1 active cores

You can also tune the min/max CPU frequencies within a policy using the cpupower frequency-set --min <freq> --max <freq>. See this page for more details on what you can do with cpupower frequency-set.

doing the above without cpupowerutils

So finally, if you don't have the cpupowerutils package installed, you can interact with it similar to how you did in the previous 2.6 kernels. Mainly you echo values into the sysfs filesystem.

for example

$ echo 360000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq

For 2.6 kernels

You can read about the various cpufreq features over on this site.

excerpt from CPU frequency scaling in Linux with cpufreq

ignore_nice_load - This parameter takes a value of '0' or '1'. When set to '0' (its default), all processes are counted towards the 'cpu utilization' value. When set to '1', the processes that are run with a 'nice' value will not count (and thus be ignored) in the overall usage calculation. This is useful if you are running a CPU intensive calculation on your laptop that you do not care how long it takes to complete as you can 'nice' it and prevent it from taking part in the deciding process of whether to increase your CPU frequency. To turn this on do the following.

sudo sh -c "echo 1 > /sys/devices/system/cpu/cpu0/cpufreq/ondemand/ignore_nice_load"

I'd put a 0 in this file since this should be the default. If you have any long running niced process, which I highly doubt, you can set it to 1.

Linux – How to add CPU frequency governors to the Linux kernel

You'll have to find the code for that specific governor and add it to your kernel before recompiling it or you can also write the proper Makefile and compile the governor as a module. The code for the governor should be in drivers/cpufreq/. For example, for the lulzactive governor: drivers/cpufreq/cpufreq_lulzactive.c