Power management/Processor

From Gentoo Wiki
Jump to: navigation, search
This page contains changes which are not marked for translation.

This article describes the setup of power management for processors.

CPU frequency scaling

External resources

CPU frequency scaling is a technique whereby the frequency of a processor can be automatically adjusted "on the fly" to conserve power and to reduce the amount of heat generated by the chip.

Installation

BIOS

Some functions can be enabled or disabled in the BIOS. Check that them are enabled, e.g.:

  • Processor C1E support
  • Enhanced Speedstep (EIST)

Kernel

You need to activate the following kernel options:

Kernel configuration

Power management and ACPI options  --->
     [*] ACPI (Advanced Configuration and Power Interface) Support  --->
         <*>   Processor
     CPU Frequency scaling  --->
         [*] CPU Frequency scaling
             Default CPUFreq governor (ondemand)  --->
                 Select a default governor; see below table
                 Default is:
                 ondemand
             x86 CPU frequency scaling drivers  --->
                 Select a driver, see below table

You need a governor, who changes the frequency based on his profile:

Default CPUFreq governor
Option Module Supported Processors Note
'performance' governor cpufreq_performance Sets the frequency statically to the highest available CPU frequency.
'powersave' governor cpufreq_powersave Sets the frequency statically to the lowest available CPU frequency. Can't be set as default governor
'userspace' governor for userspace frequency scaling cpufreq_userspace To set the CPU frequency manually or when a userspace program shall be able to set the CPU dynamically.
'ondemand' cpufreq policy governor cpufreq_ondemand Does a periodic polling and changes frequency based on the CPU utilization. recommend
'conservative' cpufreq governor cpufreq_conservative Similar to ondemand. The frequency is gracefully increased and decreased rather than jumping to 100% when speed is required.
x86 CPU frequency scaling drivers
Option Module Supported Processors Note
Processor Clocking Control interface driver pcc-cpufreq
ACPI Processor P-States driver acpi-cpufreq Intel Core, Intel Core 2, Intel Atom, Intel Pentium M
AMD Opteron/Athlon64 PowerNow! powernow-k8 AMD Opteron, AMD Athlon 64, AMD Turion 64
Intel Enhanced SpeedStep (deprecated) speedstep_centrino Intel Pentium M (Centrino), Intel Xeon Deprecate, use ACPI Processor P-States driver instead.
Intel Pentium 4 clock modulation p4-clockmod Intel Pentium 4, Intel XEON Causes severe slowdowns and noticeable latencies

Ticks

The processor saves the most energy when the processor stays longer in power savings modes, so you have to reduce the amount of ticks that wakes the processor up.

Installation

BIOS

Some functions can be enabled or disabled in the BIOS. Check that them are enabled, e.g.:

  • HPET
  • Multimedia timer

Kernel

You need to activate the following kernel options:

Kernel configuration

Processor type and features  --->
   [*] Tickless System (Dynamic Ticks)
   [*] High Resolution Timer Support
   [*] HPET Timer Support
Note
These options now appear to be under "General setup > Timers subsystem".

PowerTOP

PowerTOP is a utility designed to measure, explain and minimise a computer's electrical power consumption.

Installation

Kernel

Kernel configuration

Power management and ACPI options  --->
   [*] Power Management Debug Support
   [*]   Extra PM attributes in sysfs for low-level debugging/testing
Kernel hacking  --->
   [*] Kernel debugging
   [*] Collect kernel timers statistics
External resources
External resources

Software

Install sys-power/powertop:

→ Information about USE flags
USE flag Default Recommended Description
unicode Yes Adds support for Unicode
root # emerge --ask sys-power/powertop
Important
For kernel 3.3 onwards, one will need to enable the following kernel options (maybe incomplete list) to get proper support. Some of the following config are already enabled in the previous graphical kernel configuration. One can use the following to check quickly if the not so required options are set. One can get almost everything working or rather tunable power management without the last two (tested on Intel GM45).
File/usr/src/linux/.config

CONFIG_NO_HZ
CONFIG_HIGH_RES_TIMERS
CONFIG_HPET_TIMER
CONFIG_CPU_FREQ_GOV_ONDEMAND
CONFIG_USB_SUSPEND
CONFIG_SND_AC97_POWER_SAVE
CONFIG_TIMER_STATS
CONFIG_PERF_EVENTS
CONFIG_PERF_COUNTERS
CONFIG_TRACEPOINTS
CONFIG_TRACING
CONFIG_EVENT_POWER_TRACING_DEPRECATED
CONFIG_ACPI_PROCFS_POWER

I used a few time sys-power/powertop-1 before and the software was pretty good at giving good advise on what power management options to set up depending on the system usage. Now, version 2 become very good in that I nearly get 10-15° less CPU temperature enabling every power management option which is almost what I can get on cold weather (~40°C) while my laptop runs almost at 50-55°C in summer (quite hot... compiling/chunking around C++ objetcs make it jump to almost 100°C! (qt-libraries, libreoffice, chromium) or simply with Wifi card on). With everything enabled, the CPU runs almost all the time (when no power hungry application is running obviously) at idle or the lowest C-state (800MHz). I could not expect better. The spot that is missing is a deamon-ish application or at least a configuration/profile file that PowerTOP will pick up at start up. Lets submit the idea to upstream.

Usage

powertop - Sorts the running processes in order how often they cause the processor to wake up.

Cpufrequtils

Install the package to get started. And do not forget to add the init service in the boot level afterwards.

root # emerge --ask sys-power/cpufrequtils
root # rc-update add cpufrequtils boot

And do not forget to set up the default governor that fits the machine in the configuration file.

File/etc/conf.d/cpufrequtils

START_OPTS="--governor ondemand"

I used sys-power/cpufrequtils and other equivalent CPU frequency utilities in the past only to remove them after a few days. The thing is... one can do just that with seting up or building the necessary modules or rather CPU governors and set up a default governor in the kernel. That should suffice for most users especially with ondemand governor.

Hprofile

When one needs to change the CPU governor, one could just use a simple command such as the following which nullify the usage of cpufrequtils or other CPU governor and frequency utilities.

root # for c in $(ls -d /sys/devices/system/cpu/cpu[0-9]*); do echo ondemand >$c/cpufreq/scaling_governor; done

However, that kind of command incanatation needs a little automation... think if you're using a laptop and want it to be a little dynamic in power saving mode, you will surely want a power saving profile if the laptop is not wired to AC power. This is where Hprofile comes into play, follow reading in that article for more information.