Difference between revisions of "How to reduce power consumption"

From ThinkWiki
Jump to: navigation, search
(Ethernet Controler: clarify)
(LCD Backlight/Brightness)
 
(120 intermediate revisions by 39 users not shown)
Line 4: Line 4:
 
to a fine tuned system. The general rules are :
 
to a fine tuned system. The general rules are :
 
* Unload drivers for unused devices (ie. USB 1.1, Yenta/PCMCIA, Wireless, IRDA, Bluetooth, ...)
 
* Unload drivers for unused devices (ie. USB 1.1, Yenta/PCMCIA, Wireless, IRDA, Bluetooth, ...)
* Reduce polling on devices (drives, USB subsystem, nvram, ...)
+
* Reduce polling on devices (drives, USB subsystem, nvram, use SATA AN, ...)
 
* Reduce hard drive activity
 
* Reduce hard drive activity
 
* Reduce LCD brightness to the minimum you can stand
 
* Reduce LCD brightness to the minimum you can stand
 
* Reduce CPU wakeups, so it can stay longer in deep power saving c-states
 
* Reduce CPU wakeups, so it can stay longer in deep power saving c-states
 +
* Make use of every hardware devices availables power saving features (AHCI ALPM, USB autosuspend, Alsa and Wireless powersaving modes, HPET timers, ...)
  
 
==Tools==
 
==Tools==
Arjan van de Ven's [http://www.linuxpowertop.org/ PowerTOP] utility
+
Arjan van de Ven's [[PowerTOP]] utility
is a gold mine to improve energy efficiency. This tool helps to easily detect
+
is a gold mine to improve energy efficiency, but is almost only CPU-oriented. This tool helps to easily detect
the top power offenders, both userland and kernel modules, and sometime suggest  
+
the top power offenders, both userland and kernel modules, which prevent the use of CPU power saving mechanisms and sometime suggest  
 
fixes accordingly.
 
fixes accordingly.
PowerTOP users collected some [http://www.linuxpowertop.org/known.php tips & tricks]
+
PowerTOP users collected some [http://www.lesswatts.org/projects/powertop/known.php tips & tricks]
and an informative [http://www.linuxpowertop.org/faq.php faq].
+
and an informative [http://www.lesswatts.org/projects/powertop/faq.php faq].
When you see an application causing a lot of wakeups in PowerTOP, you should
 
strace it to understand what is happening.
 
  
To detect what make your disk spinning (a costly operation),  
+
Alternatively (or complementary) to PowerTOP, running <code>strace -p $(pidof yourapp)</code>
 +
for all your favorite or background running applications while they are expected to be
 +
idle, will show the misbehaviors.
 +
 
 +
Beside CPU wakeups, disks spins are also power hungry. To detect what make your disk spinning,
  
<code>
 
 
  sysctl vm.block_dump=1
 
  sysctl vm.block_dump=1
</code>
 
  
 
will list all applications causing disks wakeups on the kernel's dmesg.
 
will list all applications causing disks wakeups on the kernel's dmesg.
Other tools for this purpose are blktrace and iostat.
+
Other useful tools for this purpose are blktrace, iostat and lm-profiler
 +
(from laptop-mode-tools suite).
 +
 
 +
See also [[#Misbehaving Userland]].
  
 
==BIOS settings==
 
==BIOS settings==
 +
===Enabling Power Management===
 
Some Thinkpad BIOS (like 2.08 BIOS on {{X40}}) offer two very lame options,
 
Some Thinkpad BIOS (like 2.08 BIOS on {{X40}}) offer two very lame options,
 
with a very misleading online help (saying "Usually not needed"). That's
 
with a very misleading online help (saying "Usually not needed"). That's
  
<code>
 
 
  CPU power management: (default disabled)
 
  CPU power management: (default disabled)
 
  PCI bus power management: (default disabled)
 
  PCI bus power management: (default disabled)
</code>
 
  
You should indeed enable them, else the deepest C3 and C4 ACPI C-states
+
You should indeed ''enable'' them, else the deepest C3 and C4 ACPI C-states
 
are disabled.
 
are disabled.
 +
 +
===Disabling I/O Ports===
 +
The BIOS (at least version 3.11 on {{X200}}) can also be used to disable I/O ports, like PCMCIA/CardBus.  Although this requires a reboot to change settings, using the BIOS rather than a configuration file will survive distribution changes and may make it easier to remember how to re-enable a port.  Disabling these devices can reduce power consumption by several watts.
  
 
==CPU==
 
==CPU==
Line 48: Line 54:
 
a trivial light job, reduce the time the CPU stays on a deep power
 
a trivial light job, reduce the time the CPU stays on a deep power
 
saving C-state (like C3 or C4). Therefore you should ensure your applications
 
saving C-state (like C3 or C4). Therefore you should ensure your applications
stay really idle when they meant to be idle (track to short select timeouts
+
stay really idle when they meant to be idle (track shorts select timeouts
in loop, etc.).
+
in loop, etc. with [[PowerTOP]]).
  
 
Also note that manually locking the CPU in the lowest P-state (frequency)  
 
Also note that manually locking the CPU in the lowest P-state (frequency)  
 
available is '''not''' an efficient way to improve battery lifetime. This will
 
available is '''not''' an efficient way to improve battery lifetime. This will
cause the CPU to stay longer in C0 (power angry C-state) doing hard work when  
+
cause the CPU to stay longer in C0 (power hungry C-state) doing hard work when  
 
there is something to do, while it could have done this work faster by augmenting
 
there is something to do, while it could have done this work faster by augmenting
the CPU freq, and returned back faster to deeper, economic, C-state.
+
the CPU freq, and returned back faster to a deeper, economic, C-state and to a
The best is to let the kernel choose the appropriate CPU frequencies by itself.
+
lower frequency (P-state).
 +
The best is to let the kernel select the appropriates CPU frequencies by itself
 +
with the help of in kernel CPU governors.
 
Have a look at [http://www.bughost.org/pipermail/power/2007-May/000166.html this explanation]
 
Have a look at [http://www.bughost.org/pipermail/power/2007-May/000166.html this explanation]
 
from Intel's kernel developer Arjan van de Ven.
 
from Intel's kernel developer Arjan van de Ven.
Line 63: Line 71:
  
 
===General settings===
 
===General settings===
The 2.6.21 kernel brought some very effective changes (like dynticks). If
+
The 2.6.21 kernel brought some very effective changes (like [[dynticks]]).  
it's not already on your distribution and you value power efficiency, you
+
Later, 2.6.24-rc2 brought a lot of other power efficiency improvements.
may think about compiling it (or a more recent one) yourself.
+
If it's not already on your distribution and you value power efficiency,  
 +
you may think about compiling a recent kernel yourself.  
  
 
Here are a few options (beside the ACPI and APM related one) that matter to  
 
Here are a few options (beside the ACPI and APM related one) that matter to  
 
reduce power consumption or to help diagnosing consumers:
 
reduce power consumption or to help diagnosing consumers:
  
<code>
 
 
  # From PowerTOP's FAQ:
 
  # From PowerTOP's FAQ:
 
  CONFIG_NO_HZ
 
  CONFIG_NO_HZ
 
  CONFIG_HIGH_RES_TIMERS
 
  CONFIG_HIGH_RES_TIMERS
 
  CONFIG_HPET
 
  CONFIG_HPET
 +
CONFIG_HPET_TIMER
 
  CONFIG_CPU_FREQ_GOV_ONDEMAND
 
  CONFIG_CPU_FREQ_GOV_ONDEMAND
 
  CONFIG_USB_SUSPEND
 
  CONFIG_USB_SUSPEND
 
  CONFIG_SND_AC97_POWER_SAVE
 
  CONFIG_SND_AC97_POWER_SAVE
 +
CONFIG_SND_HDA_POWER_SAVE
 +
CONFIG_SND_HDA_POWER_SAVE_DEFAULT=3
 
  CONFIG_TIMER_STATS
 
  CONFIG_TIMER_STATS
 
  CONFIG_ACPI_BATTERY
 
  CONFIG_ACPI_BATTERY
 
  CONFIG_CPU_FREQ_STAT
 
  CONFIG_CPU_FREQ_STAT
  # not from the PowerTOP FAQ:
+
CONFIG_INOTIFY
 +
 
 +
  # Not from the PowerTOP FAQ:
 
  CONFIG_BLK_DEV_IO_TRACE
 
  CONFIG_BLK_DEV_IO_TRACE
 
  CONFIG_X86_ACPI_CPUFREQ
 
  CONFIG_X86_ACPI_CPUFREQ
  CONFIG_X86_SPEEDSTEP_CENTRINO
+
  CONFIG_X86_SPEEDSTEP_CENTRINO depreciated as of kernel 2.6.24, use CONFIG_X86_ACPI_CPUFREQ
 
  CONFIG_X86_SPEEDSTEP_ICH
 
  CONFIG_X86_SPEEDSTEP_ICH
 
  CONFIG_X86_SPEEDSTEP_SMI
 
  CONFIG_X86_SPEEDSTEP_SMI
</code>
+
CONFIG_CPU_IDLE
 +
CONFIG_CPU_IDLE_GOV_LADDER
 +
CONFIG_CPU_IDLE_GOV_MENU
  
Those options are already in Fedora Core 7 and Ubuntu Gutsy (not Feisty!) default kernels.
+
Those options are already in Fedora Core 7 and Ubuntu Gutsy (not Feisty) default i686 kernels.
 
PowerTOP FAQ also suggest to '''disable'''
 
PowerTOP FAQ also suggest to '''disable'''
CONFIG_IRQBALANCE et CONFIG_ACPI_DEBUG.
+
CONFIG_IRQBALANCE and CONFIG_ACPI_DEBUG.
  
 
Also, you need to properly set APM and ACPI. Look at:
 
Also, you need to properly set APM and ACPI. Look at:
Line 97: Line 112:
 
* [[How to make use of Power Management features]]
 
* [[How to make use of Power Management features]]
  
===Useful Patchs===
+
=== Kernel boot and module loading options ===
 +
If you have an Intel chipset ICH5 or later (cf. lspci output), as in most modern Thinkpads, you should
 +
be using the integrated HPET timer (saves about 30 CPU wake ups per second). To see if
 +
hpet is enabled on your laptop:
 +
 
 +
grep hpet /proc/timer_list
 +
 
 +
If this does not display "Clock Event Device: hpet", then add
 +
 
 +
hpet=force
 +
 
 +
{{WARN|The ICH4 does have an HPET, but it is disabled for a good reason: Intel didn't test/validade it!  Use of the ICH4 HPET is '''not''' recommended}}
 +
 
 +
to your kernel boot options (usualy in /boot/grub/menu.lst or lilo.conf).
 +
Note that "hpet=force" is only available by default in 2.6.24-rc2 and above
 +
(or as a separated patch for 2.6.22 and 2.6.23, see below).
 +
 
 +
On modern ThinkPads the HPET timer is automatically detected and enabled. On certain older machines hpet=force is required such as on the following machines:
 +
* {{T30}}, {{T40}}, {{T40p}}, {{T41}}, {{T41p}}, {{T42}}
 +
* {{X22}}, {{X23}}, {{X24}}, {{X30}}, {{X31}}, {{X40}}
 +
* {{A31}}
 +
* {{SL500}}
 +
 
 +
{{HELP|please add your ThinkPad to the above list, if <nowiki>hpet=force</nowiki> was required}}
 +
 
 +
===Useful Patches===
  
 
Thomas Gleixner High Resolution Timers (hrt) patchset brings many improvements,
 
Thomas Gleixner High Resolution Timers (hrt) patchset brings many improvements,
 
like the cpuidle work and Udo A. Steinberg and Venki Pallipadi "force
 
like the cpuidle work and Udo A. Steinberg and Venki Pallipadi "force
enable HPET" patches (non HPET timers causes about 20-40 CPU wakeups/second).
+
enable HPET" patches (non HPET timers causes about 20-40 CPU wakeups/second, but
 +
HPET is often hidden by the BIOS due to Windows XP deficiencies). Those are
 +
fully merged in 2.6.24-rc1 vanilla kernel.
 
See http://www.tglx.de/projects/hrtimers/
 
See http://www.tglx.de/projects/hrtimers/
  
 
Kristen Carlson Accardi from Intel has a patchset to turn on "Aggressive
 
Kristen Carlson Accardi from Intel has a patchset to turn on "Aggressive
Link Power Management" (ALPM) for the ahci driver. See:
+
Link Power Management" (ALPM) for the AHCI driver (for SATA bus). Also from
http://www.kernel.org/pub/linux/kernel/people/kristen/patches/SATA/alpm/
+
Accardi, SATA Asynchronous Notification (SATA AN), alows SATA link to notify
 +
media insertions (thus avoid hal polling the cdrom). Those patches were merged
 +
in 2.6.24-rc2 kernel (AN needs also support in hal to be used).
 +
See: http://www.kernel.org/pub/linux/kernel/people/kristen/patches/SATA/alpm/
 +
 
 +
As of now (2.6.24-rc8), the linux kernel doesn't support PCI Express power
 +
management (aka PCIe ASPM, aka PCIe LPM). Shaohua Li from Intel submited a
 +
patch on LKML (http://lkml.org/lkml/2008/1/17/544 ) though, and reported it
 +
to reduce power consumption by 1.3 watts for a system with three PCIe links.
 +
 
 +
The [[HDAPS]] disk protection systems can reduce battery life.
 +
Matthew Garrett provides [http://www.linuxpowertop.org/patches/hdaps.patch a patch]
 +
that prevents hdaps kernel module to generate interrupts when
 +
this feature isn't used.
  
 
===Useful sysctls===
 
===Useful sysctls===
The meaning of those settings are explained on the relevant sections of
+
The meaning of those settings is explained case by case on the relevant  
this document. But for convenience sake, we regroupe them here too.
+
sections of this document. But for convenience sake, we group them here too.
  
Note that the "ondemand" scaling governor is recommended by Intel developpers
+
Note that the "ondemand" scaling governor is recommended by Intel developers
 
for energy efficiency: it's expected to be more efficient than the "powersave"
 
for energy efficiency: it's expected to be more efficient than the "powersave"
 
governor, or than userspace daemons (like cpufreq-utils, cpufreqd, powernowd...).
 
governor, or than userspace daemons (like cpufreq-utils, cpufreqd, powernowd...).
 
Look [http://www.bughost.org/pipermail/power/2007-May/000071.html here],
 
Look [http://www.bughost.org/pipermail/power/2007-May/000071.html here],
[http://www.bughost.org/pipermail/power/2007-May/000073.html there] or
+
[http://www.bughost.org/pipermail/power/2007-May/000073.html here], or
 
[http://www.bughost.org/pipermail/power/2007-May/000166.html here] for a
 
[http://www.bughost.org/pipermail/power/2007-May/000166.html here] for a
 
kernel developer explanation about "ondemand" being better on modern Intel CPUs.
 
kernel developer explanation about "ondemand" being better on modern Intel CPUs.
  
 
The "link_power_management_policy" tunable won't be available unless you
 
The "link_power_management_policy" tunable won't be available unless you
run Kirsten patchset, have an Intel AHCI compatible chipset, and use SATA drives.
+
run a 2.6.24-rc2 or more kernel, or applied Kirsten patchset, have an Intel  
 +
AHCI compatible chipset, and use SATA drives.
  
<code>
 
 
  echo 5 > /proc/sys/vm/laptop_mode
 
  echo 5 > /proc/sys/vm/laptop_mode
 
  echo 0 > /proc/sys/kernel/nmi_watchdog
 
  echo 0 > /proc/sys/kernel/nmi_watchdog
Line 129: Line 184:
 
  echo 1 > /sys/devices/system/cpu/sched_mc_power_savings
 
  echo 1 > /sys/devices/system/cpu/sched_mc_power_savings
 
  echo ondemand > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
 
  echo ondemand > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
 +
echo 1500 > /proc/sys/vm/dirty_writeback_centisecs
 +
for i in /sys/bus/usb/devices/*/power/autosuspend; do echo 1 > $i; done
 +
# those sysctl's are only available if you have an AHCI compatible SATA
 +
# controler and use kernel > 2.6.24-rc2 (or use Kristen ALPM patchset) :
 
  echo min_power > /sys/class/scsi_host/host0/link_power_management_policy
 
  echo min_power > /sys/class/scsi_host/host0/link_power_management_policy
 
  echo min_power > /sys/class/scsi_host/host1/link_power_management_policy
 
  echo min_power > /sys/class/scsi_host/host1/link_power_management_policy
  echo 1500 > /proc/sys/vm/dirty_writeback_centisecs
+
  echo min_power > /sys/class/scsi_host/host2/link_power_management_policy
</code>
+
 
 +
If you're running a kernel older than 2.6.22 do this. Not needed for kernels 2.6.22 onward:
 +
 
 +
cd /sys/devices/system/cpu/cpu0/cpufreq
 +
cat ondemand/sampling_rate_max > ondemand/sampling_rate
  
 
==ATA drives==
 
==ATA drives==
Line 140: Line 203:
  
 
===Hard Drives===
 
===Hard Drives===
 +
 
The files access time update, while mandated by POSIX, is causing lots of
 
The files access time update, while mandated by POSIX, is causing lots of
disks access, even for file accessed while on cache. If you don't use this
+
disk write access; even accessing files on disk cache may wake the ATA or USB
feature, disable it via:
+
bus. If you don't use this feature, disable it by adding the <tt>relatime</tt>
 +
option to all relevant lines in the /etc/fstab, for example:
 +
/dev/sda1 / ext3 relatime,errors=remount-ro 0 1
 +
 
 +
(On older kernels you may need to use <tt>noatime</tt> instead of <tt>relatime</tt>.)
  
<code>
+
Also consider merely using a larger value for the <tt>commit</tt> option. This defines how often changed data is written to the disk (it is cached until then). The default value is 5 seconds.
mount -o remount,noatime /  # and so on for all mounted fs
 
</code>
 
  
The laptop_mode reduce disk usage by regrouping writes. You should enable
+
See man mount(8) for details on how the <tt>rel/noatime</tt> and <tt>commit</tt> options work.
 +
 
 +
Use laptop_mode to reduce disk usage by delaying and grouping writes. You should enable
 
it, at least while on battery. See [[Laptop-mode]] for more details:
 
it, at least while on battery. See [[Laptop-mode]] for more details:
  
<code>
 
 
  echo 5 > /proc/sys/vm/laptop_mode
 
  echo 5 > /proc/sys/vm/laptop_mode
</code>
 
  
 
The default kernel dirty page writeback frequency is very conservative. On
 
The default kernel dirty page writeback frequency is very conservative. On
 
a laptop running on battery, one might find more appropriate to reduce it:
 
a laptop running on battery, one might find more appropriate to reduce it:
  
<code>
 
 
  echo 1500 > /proc/sys/vm/dirty_writeback_centisecs
 
  echo 1500 > /proc/sys/vm/dirty_writeback_centisecs
</code>
 
  
Some power saving hard drives features can be activated with hdparm.
+
Some power saving hard drives features can be activated with hdparm (beware
 +
that "-B 1" may reduce your drive lifetime, if you have lot of intermittent
 +
disk activity causing lots of heads load/unloads: so reduce I/O activity first,
 +
as explained above, in order to get longer disks idling periods).
 
For more details look at [[How to make use of Power Management features]] :
 
For more details look at [[How to make use of Power Management features]] :
  
<code>
 
 
  hdparm -B 1 -S 12 /dev/sda # and/or any other disk device
 
  hdparm -B 1 -S 12 /dev/sda # and/or any other disk device
</code>
+
 
 +
====SATA Link Power Management====
 +
On kernels 2.6.24 and new this enables SATA Link Power Management:
 +
echo min_power > /sys/class/scsi_host/host0/link_power_management_policy
 +
echo min_power > /sys/class/scsi_host/host1/link_power_management_policy
 +
echo min_power > /sys/class/scsi_host/host2/link_power_management_policy
 +
 
 +
Disable it by replacing <code>min_power</code> with <code>max_performance</code>.
 +
 
 +
Which host-number to use depends on your laptop.  It is typically 0 for the main drive, but for example on the T61 the 0 and 1 are for the pata interface used for the optical drive, and the main hard-disk is on host2.
 +
 
 +
On Ubuntu Hardy Heron with a 2.6.24-16 kernel, a suspend/resume cycle is much quicker if you disable SATA Link Power Management before initiating the suspend. As of Intrepid Ibex and kernel 2.6.27, this should be fixed. ([https://bugs.launchpad.net/linux/+bug/234047 Launchpad bug 234047], [http://bugzilla.kernel.org/show_bug.cgi?id=10817 Kernel bug 10817])
 +
 
 +
====Laptop Mode Tools====
 +
 
 +
The [http://samwel.tk/laptop_mode/ Laptop Mode Tools] utility implements many of the above power-saving measures from disks, and some others.
  
 
===Optical drive===
 
===Optical drive===
The optical drive consumes about 1W even when not accessed. See  
+
The optical drive is reported to consume power even when not accessed. See  
  
 
* [[How to hotswap UltraBay devices|Eject the UltraBay optical drive]], or just turn off its power supply (i.e., run the appropriate eject script but leave the drive inserted).
 
* [[How to hotswap UltraBay devices|Eject the UltraBay optical drive]], or just turn off its power supply (i.e., run the appropriate eject script but leave the drive inserted).
Line 176: Line 257:
  
 
The hald daemon polling tends to maintain the ATA buses out of power saving
 
The hald daemon polling tends to maintain the ATA buses out of power saving
modes, and to wakeup CDROM drive. If you have a recent hald version, you
+
modes, and to wakeup CDROM drive (except if you have a kernel >= 2.6.24, hal >= 0.5.10,
 +
and SATA AN compatible devices). If you have a recent hald version, you
 
can stop this polling when on battery:
 
can stop this polling when on battery:
  
<code>
+
  hal-disable-polling --device /dev/scd0 # or whatever your CD drive is
  hal-disable-polling /dev/scd0 # or whatever your CD drive is
 
</code>
 
  
If your hald is not recent enough, consider stopping it on battery.
+
start polling again when on ac:
 +
 
 +
hal-disable-polling --enable-polling --device /dev/scd0 # or whatever your CD drive is
 +
 
 +
 
 +
If your hald is not recent enough, consider suspending it when running on battery. Some moderns SATA buses and drivers supports a notification mechanism (SATA AN - Asynchronous Events Notifications) that obsolete the need for polling on modern hardware; support for this feature had been merged in Linux 2.6.24-rc1 and HAL 0.5.10.
  
 
==LCD Backlight/Brightness==
 
==LCD Backlight/Brightness==
The LCD backlight is one of the major power drain (usually consumes 
+
The LCD backlight is one of the very major power drain.
5W or more at full brightness). So reducing brightness to the lowest readable
+
Reducing brightness to the lowest readable
level will save you a lot of battery lifetime. Also, don't forget to configure
+
level will save a lot of battery lifetime. Also, don't forget to configure
your screen saver to shutdown the screen backlight (and not to display some
+
your screen saver to shutdown the screen backlight (rather than displaying some
 
eye candy), when no activity for a few minutes.
 
eye candy), when no activity for a few minutes.
 +
 +
You can also let the system [[automatically reduce brightness]] after a
 +
period of inactivity.
  
 
If you're choosing your Thinkpad laptop model, keep in mind that the screen
 
If you're choosing your Thinkpad laptop model, keep in mind that the screen
size affect the battery time greatly (more power needed for larger screens).
+
size affect the battery time greatly: more power needed for larger screens.
 +
 
 +
The very recent, but xorg standard way to control backlight from CLI is
 +
using xbacklight. ie. to set backlight at half the brightness:
 +
 
 +
xbacklight -set 50
 +
 
 +
You should configure the DPMS to shutdown the screen when idle (rather than
 +
displaying a fancy but power consuming screensaver). ie. to turn off the
 +
display after 5 minutes of idling:
 +
 
 +
xset +dpms
 +
xset dpms 0 0 300
 +
 
 +
===Screensaver method===
 +
Alternatively, you can add a screensaver that does nothing but blank the screen and turn off the display. In this example, I'm using Linux Mint MATE 17.2, please adjust the keywords below accordingly. Also, you will need to enable DPMS as above, or as in [[How to make ACPI work#Screen blanking (Standby)]].
 +
 
 +
We need to create 2 files, the first one is <code>/usr/share/applications/screensavers/screenoff.desktop</code>:
 +
 
 +
[Desktop Entry]
 +
Name=Screen off
 +
Comment=Turns the screen off
 +
Exec=/usr/lib/mate-screensaver/screenoff
 +
TryExec=/bin/true
 +
StartupNotify=false
 +
Terminal=false
 +
Type=Application
 +
Categories=Screensaver;
 +
Keywords=MATE;screensaver;dpms;
 +
OnlyShowIn=MATE;
 +
 
 +
The second one is <code>/usr/lib/mate-screensaver/screenoff</code>, which needs to be executable:
 +
 
 +
#!/bin/sh
 +
pidof mate-screensaver-preferences | grep -w $PPID && exit 0
 +
xset q | grep -q "Monitor is On" && xset dpms force off
 +
 
 +
The <code>Keywords</code> section and the <code>mate-screensaver-preferences</code> should be customized according to your desktop (like Cinnamon, or GNOME, etc).
 +
 
 +
After creating those files, open the Screensaver preferences dialog and choose the screensaver with the name "Screen off". Now, when the screensaver is activated, the screen will fade to black and backlight will be turned off by DPMS.
  
 
==Graphic controllers==
 
==Graphic controllers==
Line 199: Line 326:
 
the same bug causing very frequent CPU wakeups when DRI is activated, even
 
the same bug causing very frequent CPU wakeups when DRI is activated, even
 
when you don't use any 3D application.
 
when you don't use any 3D application.
This problem is (partly) fixed on xorg git tree (but not released as of xorg
+
This problem is partly fixed on xorg git tree but not released as of xorg
7.2). If you value more battery than 3D, you should disable DRI: put this on
+
7.2. If you value more battery than 3D, you should disable DRI: put this on
 
the /etc/X11/xorg.conf "Device" of you graphic controller:
 
the /etc/X11/xorg.conf "Device" of you graphic controller:
  
<code>
 
 
  Option          "NoDRI"
 
  Option          "NoDRI"
</code>
 
  
Also be sure that DPMS is working (if not, had Option "DPMS" to your config).
+
Also be sure that DPMS is working: <code>grep DPMS /var/log/Xorg.0.log</code>
 +
should output "DPMS enabled". If not, put <code>Option "DPMS"</code> in your config.
 +
See the section above about how to enable dpms driven display power saving.
  
On recent xrandr/xorg versions, you can disable the TV output when you're
+
On recent xrandr/xorg versions, you can disable the TV output (or any other detected
not using it, it saves about 1W:
+
as connected but not used outputs) when you're not using it: it's known to consume power.
  
<code>
+
xrandr # see all displays listed here, but that you don't actually use and disable them.
  xrandr --output TV -off
+
  xrandr --output TV --off # for instance (if "xrandr" above listed a connected output named "TV" that you don't use)
</code>
 
  
See also:
+
When you don't have an external monitor plugged, disable CRT and DVI output
* [[How to make use of Graphics Chips Power Management features]]
+
(for some, this can make a difference in power usage) :
 +
 
 +
echo crt_disable > /proc/acpi/ibm/video
 +
echo dvi_disable > /proc/acpi/ibm/video
 +
 
 +
Some drivers have specials power saving mode, and/or allows underclocking the GPU. See also:
 +
* [[How to make use of Graphics Chips Power Management features]], or with [[Rovclock]] on ATI.
 
* [[Problem with high power drain in ACPI sleep]]
 
* [[Problem with high power drain in ACPI sleep]]
* [[Rovclock]]
 
* Reduce LCD brightness (~1-4W difference)
 
** Do this [[Automatically_reduce_brightness|automatically]] after a period of inactivity
 
* Disable CRT output (~0.5W difference)
 
  
 
==USB Subsystem==
 
==USB Subsystem==
Line 228: Line 356:
 
CONFIG_USB_SUSPEND. This may not trigger in when you have an USB device
 
CONFIG_USB_SUSPEND. This may not trigger in when you have an USB device
 
plugged (and beside, USB devices tends to suck power on their own), so avoid
 
plugged (and beside, USB devices tends to suck power on their own), so avoid
using such devices when on battery.
+
using such devices when on battery. To enable it by default, you must add the line
 +
 
 +
options usbcore autosuspend=1
 +
 
 +
to your <tt>/etc/modprobe.conf</tt> or add it to (and create if necessary) the file <tt>/etc/modprobe.d/usbcore</tt> depending on how your distribution organises modprobe configuration.
 +
 
 +
If on the other hand, you have <tt>usbcore</tt> built into your kernel, you can add this in the kernel boot options (ie. in grub's menu.lst):
 +
 
 +
usbcore.autosuspend=1
 +
 
 +
or at runtime, per device, with:
 +
 
 +
for i in /sys/bus/usb/devices/*/power/autosuspend; do echo 1 > $i; done
 +
for i in /sys/bus/usb/devices/*/power/level; do echo auto > $i; done
  
 
USB 1.1 is worst. It needs polling the bus frequently, hence can't really go
 
USB 1.1 is worst. It needs polling the bus frequently, hence can't really go
Line 234: Line 375:
 
plugged. You'd better remove it when you don't use a 1.1 device:
 
plugged. You'd better remove it when you don't use a 1.1 device:
  
<code>
 
 
  rmmod uhci_hcd
 
  rmmod uhci_hcd
</code>
 
  
If you don't intend to use any USB 1.1 only device (those tends to be rare
+
If you don't intend to use any device needing USB 1.1 (unfortunately, the built-in bluetooth and fingerprint-reader are USB 1.1 devices), the USB 1.1 support can also be totally avoided. On Debian and derivatives, just do:
anyway), the USB 1.1 support can also be totaly avoided. On Debian and
 
derivatives, just do:
 
  
<code>
+
  echo "blacklist uhci_hcd" >> /etc/modprobe.d/blacklist
  echo "blacklist uhci_hcd" > /etc/modprobe.d/blacklist.
+
 
</code>
+
==PCMCIA/CardBus==
 +
Same for PCMCIA/CardBus. Some users experiences interrupts clouds (sometime up to
 +
several thousands interrupts/second) causing CPU wakeups, thus totally preventing
 +
the CPU to reach lower C-states.  
 +
If you don't use PCMCIA, you may disable it the same way (unloading seems insufficient
 +
to restore the system properly, you have to boot without it):
 +
 
 +
echo "blacklist pcmcia" >> /etc/modprobe.d/blacklist
 +
echo "blacklist yenta_socket" >> /etc/modprobe.d/blacklist
  
 
==Sound==
 
==Sound==
 +
 
ALSA has a power saving feature that should be enabled on your kernel
 
ALSA has a power saving feature that should be enabled on your kernel
 
(CONFIG_SND_AC97_POWER_SAVE). Note that this low power mode won't trigger in
 
(CONFIG_SND_AC97_POWER_SAVE). Note that this low power mode won't trigger in
Line 252: Line 398:
 
to be activated with:
 
to be activated with:
  
<code>
+
amixer set Line mute nocap
 +
amixer set Mic mute nocap
 
  echo Y > /sys/module/snd_ac97_codec/parameters/power_save
 
  echo Y > /sys/module/snd_ac97_codec/parameters/power_save
</code>
 
  
More radical: you can unload all sound related modules when you are on
 
battery.
 
  
Seel also [[How to enable AC97 power saving]].
+
===Intel HD Audio===
 +
 
 +
If you have Intel HD audio as your onboard sound controller, substitute the following for the last line in the above sequence:
 +
 
 +
echo Y > /sys/module/snd_hda_intel/parameters/power_save_controller
 +
 
 +
You also may wish to decrease the sound poweroff timeout to something shorter, like 1 second after last playback:
 +
 
 +
echo 1 > /sys/module/snd_hda_intel/parameters/power_save
 +
 
 +
===Additional Tweaks===
 +
 
 +
You can unload all sound related modules when you are on
 +
battery, or mute the sound system (echo mute > /proc/acpi/ibm/volume).
 +
 
 +
See also [[How to enable audio codec power saving]].
  
 
==Wireless Interface==
 
==Wireless Interface==
Wireless network Consumes up to ~4W.
+
===intel wireless===
To save power on ipw adapters, you can kill the Wi-Fi radio when it's not in use:
+
Wireless network consume a lot of power.
 +
To save power, you can kill the Wi-Fi radio when it's not in use:
  
<code>
+
  echo 1 > /sys/bus/pci/devices/*/rf_kill
  echo 1 > /sys/bus/pci/drivers/ipw2200/*/rf_kill
 
</code>
 
  
 
If you need Wi-Fi, you can also reduce power consumption (at the price of
 
If you need Wi-Fi, you can also reduce power consumption (at the price of
 
performances) by activating the power saving modes:
 
performances) by activating the power saving modes:
  
<code>
+
  iwpriv eth1 set_power 5
  iwpriv eth0 set_power 7
+
 
</code>
+
For drivers using the new Wi-Fi kernel framework (mac80211/cfg80211),
 +
the canonical way to do this is now:
  
Reducing Beacon intervals on your Access Point to 1 per second will also
+
for i in /sys/bus/pci/devices/*/power_level ; do echo 5 > $i ; done
reduce network card interrupts, and helps power savings.
+
 
 +
Most drivers, like ipw2200, that don't use the new mac80211 framework place the
 +
interfaces in aggressive scanning mode when they are not associated with any
 +
Access Point, even when the interface is down (more info about this on Intel's
 +
[http://www.lesswatts.org/tips/wireless.php LessWatts] website).
 +
This behavior consumes a lot of power, even more than when the interface
 +
is plain active and in use. But this can disabled at module's load time :
 +
 
 +
rmmod ipw2200
 +
modprobe ipw2200 associate=0
 +
 
 +
You can fix this setting by placing the following in /etc/modprobe.d/options
 +
(Debian/Ubuntu) or in /etc/modprobe.conf (Red Hat/Fedora):
 +
 
 +
options ipw2200 associate=0
 +
 
 +
Reducing beacon intervals on your Access Point to 1 per second will also
 +
reduce network card interrupts, therefore power savings. This shouldn't have
 +
negatives side effects.
 +
 
 +
In recent kernels, the powersaving on the intel ipw3945 has been disabled, as
 +
for some it is unstable. For others it worked just fine.
 +
 
 +
See [http://git.kernel.org/?p=linux/kernel/git/stable/linux-2.6.33.y.git;a=blobdiff;f=drivers/net/wireless/iwlwifi/iwl-3945.c;h=e413bd35bc411a1113177f1576538eb0ac26f00c;hp=4609323d8436dc9e22c74be2f1f3cf5e97785cb2;hb=bc45a67079c916a9bd0a95b0b879cc0f259bac6e;hpb=b7bb1756cb6a610cdbac8cfdad9e79bb5670b63b this patch]. You may wish to try changing '.broken_powersave=true' to false, in /usr/src/linux/drivers/net/wireless/iwlwifi/iwl-3945.c if you wish to enable powersave.
  
 
See also, to activate power saving on the wireless network card:
 
See also, to activate power saving on the wireless network card:
Line 287: Line 469:
 
because it sucks a lot of power:
 
because it sucks a lot of power:
  
<code>
 
 
  ethtool -s eth0 wol d
 
  ethtool -s eth0 wol d
</code>
 
  
 
If you can, try to reduce useless network activity on your ethernet
 
If you can, try to reduce useless network activity on your ethernet
 
segment, coming to your NIC (ie. uneeded broadcasts), those cause  
 
segment, coming to your NIC (ie. uneeded broadcasts), those cause  
 
interrupts and CPU wakeups.
 
interrupts and CPU wakeups.
 +
 +
Forcing 100Mbps full-duplex speed on a gigabit ethernet NIC can also save a lot of power (~1W) on most network workloads. This also reduces components temperature (e.g., [[Thermal Sensors|thermal sensor]] 0xC0 on the {{T43}} cools down by 5 degree between 1000Mbps and 100Mbps, and another 1 degree for 10Mbps).
 +
 +
  ethtool -s eth0 autoneg off speed 100
 +
 +
Note, however, that if the network device on the other side has auto-negotiation enabled (which is very common) and you turn auto-negotiation off, the other side will assume half-duplex mode and you will experience a significant loss of performance.
  
 
==Bluetooth==
 
==Bluetooth==
When you don't need bluetooth, disable it. Because of it's radio,  
+
When you don't need bluetooth, disable it. Because of its radio,  
 
bluetooth is not power friendly.
 
bluetooth is not power friendly.
  
<code>
 
 
  hciconfig hci0 down ; rmmod hci_usb
 
  hciconfig hci0 down ; rmmod hci_usb
</code>
+
echo disable > /proc/acpi/ibm/bluetooth
 +
 
 +
==Modem==
 +
When was the last time you used your analog modem? If you can't remember, you probably just don't need it. If it is on a separate module in your laptop, simply remove it. Store it in a ESD safe place (like the bag in which your last addon card or hard drive was packed), in case you should need it again. This won't save you a lot of power and weight, but why carry something around you never use.
  
 
==System Fans==
 
==System Fans==
Consumes about 0.5W when running.
+
Fans consumes power when running, so you may look at the [[ACPI fan control script]].
* [[ACPI fan control script]]
 
  
 
==Misbehaving Userland==
 
==Misbehaving Userland==
Line 312: Line 499:
 
Look at the PowerTOP's [http://www.linuxpowertop.org/known.php known problems]
 
Look at the PowerTOP's [http://www.linuxpowertop.org/known.php known problems]
 
list.
 
list.
 +
 +
Deactivate desktop animations (blinking cursor on the terms, animated wallpapers, ...): they cause regular X (therefore kernel and CPU) wakeups.
 +
 +
In short, while on battery, you should stop all applications that don't really stay idle when you're not using them. This means applications that:
 +
* Wakes up the CPU too often (by polling something, because of too short select() timeouts, ...)
 +
* Access the disks at regular intervals
 +
* Access an hardware bus (USB, ATA, ...) at regular intervals
 +
To find those offenders run:
 +
* <code>strace -p $(pidof yourapp)</code> # for all your running applications
 +
* <code>powertop</code>
 +
* <code>dstat -t -c --power --top-cpu --top-io --top-bio --top-latency --top-cputime</code>
 +
* <code>sysctl vm.block_dump=1</code> # and look at dmesg
 +
* <code>ps aux | awk '{print$10,$11}' | sort -n</code> # will list all running softs sorted by used cpu time
 +
Please, don't forget to fill a bug when you find such a misbehaving software.
 +
{{NOTE|Not all software is evil, buggy or badly written. Some produce regular activity because they have to, in order to provide their intented functionality.  Think twice before filling bugs about this.}}
  
 
==See Also==
 
==See Also==
 
* [[How to measure power consumption]]
 
* [[How to measure power consumption]]
* [[Script_for_monitoring_power_consumption]]
+
* [[Script for monitoring power consumption]]
 
* Battery [[maintenance]]
 
* Battery [[maintenance]]
  
Line 323: Line 525:
 
* [http://www.gentoo.org/doc/en/power-management-guide.xml Power Management Guide] from the Gentoo Linux documentation
 
* [http://www.gentoo.org/doc/en/power-management-guide.xml Power Management Guide] from the Gentoo Linux documentation
 
* [http://mailman.linux-thinkpad.org/pipermail/linux-thinkpad/2005-November/030478.html When/where/what for low power consumption?] (thread on Linux-Thinkpad)
 
* [http://mailman.linux-thinkpad.org/pipermail/linux-thinkpad/2005-November/030478.html When/where/what for low power consumption?] (thread on Linux-Thinkpad)
* [http://www.linuxpowertop.org When/where/what for low power consumption? linuxpowertop] Intel utility for monitoring energy intensive processes (also available as Debian package)
+
* Intel's [http://www.lesswatts.org/ LessWatts] "''Saving power on Linux''" website
 
* ''8 hours of battery life on your lap(top)'' ([http://atrey.karlin.mff.cuni.cz/~pavel/swsusp/8hours.odp ODP]/[http://atrey.karlin.mff.cuni.cz/~pavel/swsusp/8hours.pdf PDF]), a presentation by Pavel Machek
 
* ''8 hours of battery life on your lap(top)'' ([http://atrey.karlin.mff.cuni.cz/~pavel/swsusp/8hours.odp ODP]/[http://atrey.karlin.mff.cuni.cz/~pavel/swsusp/8hours.pdf PDF]), a presentation by Pavel Machek
  
  
[[Category:600X]] [[Category:A20m]] [[Category:A20p]] [[Category:A21e]] [[Category:A21m]] [[Category:A21p]] [[Category:A22e]] [[Category:A22m]] [[Category:A22p]] [[Category:A30]] [[Category:A30p]] [[Category:A31]] [[Category:A31p]] [[Category:i1200]] [[Category:i1300]] [[Category:i1620]] [[Category:G40]] [[Category:G41]] [[Category:R30]] [[Category:R31]] [[Category:R32]] [[Category:R40]] [[Category:R40e]] [[Category:R50]] [[Category:R50e]] [[Category:R50p]] [[Category:R51]] [[Category:R52]] [[Category:R60]] [[Category:R60e]] [[Category:T20]] [[Category:T21]] [[Category:T22]] [[Category:T23]] [[Category:T30]] [[Category:T40]] [[Category:T40p]] [[Category:T41]] [[Category:T41p]] [[Category:T42]] [[Category:T42p]] [[Category:T43]] [[Category:T43p]] [[Category:T60]] [[Category:T60p]] [[Category:T61]] [[Category:X20]] [[Category:X21]] [[Category:X22]] [[Category:X23]] [[Category:X24]] [[Category:X30]] [[Category:X31]] [[Category:X32]] [[Category:X40]] [[Category:X41]] [[Category:X41 Tablet]] [[Category:X60]] [[Category:X60s]] [[Category:Z60m]] [[Category:Z60t]] [[Category:Z61t]] [[Category:Z61e]] [[Category:TransNote]]
+
[[Category:600X]] [[Category:A20m]] [[Category:A20p]] [[Category:A21e]] [[Category:A21m]] [[Category:A21p]] [[Category:A22e]] [[Category:A22m]] [[Category:A22p]] [[Category:A30]] [[Category:A30p]] [[Category:A31]] [[Category:A31p]] [[Category:i1200]] [[Category:i1300]] [[Category:i1620]] [[Category:G40]] [[Category:G41]] [[Category:R30]] [[Category:R31]] [[Category:R32]] [[Category:R40]] [[Category:R40e]] [[Category:R50]] [[Category:R50e]] [[Category:R50p]] [[Category:R51]] [[Category:R52]] [[Category:R60]] [[Category:R60e]] [[Category:T20]] [[Category:T21]] [[Category:T22]] [[Category:T23]] [[Category:T30]] [[Category:T40]] [[Category:T40p]] [[Category:T41]] [[Category:T41p]] [[Category:T42]] [[Category:T42p]] [[Category:T43]] [[Category:T43p]] [[Category:T60]] [[Category:T60p]] [[Category:T61]] [[Category:X20]] [[Category:X21]] [[Category:X22]] [[Category:X23]] [[Category:X24]] [[Category:X30]] [[Category:X31]] [[Category:X32]] [[Category:X40]] [[Category:X41]] [[Category:X41 Tablet]] [[Category:X60]] [[Category:X60s]] [[Category:X61]] [[Category:X61s]]  [[Category:Z60m]] [[Category:Z60t]] [[Category:Z61t]] [[Category:Z61e]] [[Category:TransNote]]

Latest revision as of 02:23, 28 October 2015

Reducing system power consumption will extend battery life, reduce system temperature and (on some models) reduce system fan noise. Power consumption can be greatly improved from a stock distribution configuration to a fine tuned system. The general rules are :

  • Unload drivers for unused devices (ie. USB 1.1, Yenta/PCMCIA, Wireless, IRDA, Bluetooth, ...)
  • Reduce polling on devices (drives, USB subsystem, nvram, use SATA AN, ...)
  • Reduce hard drive activity
  • Reduce LCD brightness to the minimum you can stand
  • Reduce CPU wakeups, so it can stay longer in deep power saving c-states
  • Make use of every hardware devices availables power saving features (AHCI ALPM, USB autosuspend, Alsa and Wireless powersaving modes, HPET timers, ...)

Tools

Arjan van de Ven's PowerTOP utility is a gold mine to improve energy efficiency, but is almost only CPU-oriented. This tool helps to easily detect the top power offenders, both userland and kernel modules, which prevent the use of CPU power saving mechanisms and sometime suggest fixes accordingly. PowerTOP users collected some tips & tricks and an informative faq.

Alternatively (or complementary) to PowerTOP, running strace -p $(pidof yourapp) for all your favorite or background running applications while they are expected to be idle, will show the misbehaviors.

Beside CPU wakeups, disks spins are also power hungry. To detect what make your disk spinning,

sysctl vm.block_dump=1

will list all applications causing disks wakeups on the kernel's dmesg. Other useful tools for this purpose are blktrace, iostat and lm-profiler (from laptop-mode-tools suite).

See also #Misbehaving Userland.

BIOS settings

Enabling Power Management

Some Thinkpad BIOS (like 2.08 BIOS on X40) offer two very lame options, with a very misleading online help (saying "Usually not needed"). That's

CPU power management: (default disabled)
PCI bus power management: (default disabled)

You should indeed enable them, else the deepest C3 and C4 ACPI C-states are disabled.

Disabling I/O Ports

The BIOS (at least version 3.11 on X200) can also be used to disable I/O ports, like PCMCIA/CardBus. Although this requires a reboot to change settings, using the BIOS rather than a configuration file will survive distribution changes and may make it easier to remember how to re-enable a port. Disabling these devices can reduce power consumption by several watts.

CPU

Look at:

A good thing to keep in mind is that every CPU wakeup, even if it's for a trivial light job, reduce the time the CPU stays on a deep power saving C-state (like C3 or C4). Therefore you should ensure your applications stay really idle when they meant to be idle (track shorts select timeouts in loop, etc. with PowerTOP).

Also note that manually locking the CPU in the lowest P-state (frequency) available is not an efficient way to improve battery lifetime. This will cause the CPU to stay longer in C0 (power hungry C-state) doing hard work when there is something to do, while it could have done this work faster by augmenting the CPU freq, and returned back faster to a deeper, economic, C-state and to a lower frequency (P-state). The best is to let the kernel select the appropriates CPU frequencies by itself with the help of in kernel CPU governors. Have a look at this explanation from Intel's kernel developer Arjan van de Ven.

Kernel settings and patches

General settings

The 2.6.21 kernel brought some very effective changes (like dynticks). Later, 2.6.24-rc2 brought a lot of other power efficiency improvements. If it's not already on your distribution and you value power efficiency, you may think about compiling a recent kernel yourself.

Here are a few options (beside the ACPI and APM related one) that matter to reduce power consumption or to help diagnosing consumers:

# From PowerTOP's FAQ:
CONFIG_NO_HZ
CONFIG_HIGH_RES_TIMERS
CONFIG_HPET
CONFIG_HPET_TIMER
CONFIG_CPU_FREQ_GOV_ONDEMAND
CONFIG_USB_SUSPEND
CONFIG_SND_AC97_POWER_SAVE
CONFIG_SND_HDA_POWER_SAVE
CONFIG_SND_HDA_POWER_SAVE_DEFAULT=3
CONFIG_TIMER_STATS
CONFIG_ACPI_BATTERY
CONFIG_CPU_FREQ_STAT
CONFIG_INOTIFY
# Not from the PowerTOP FAQ:
CONFIG_BLK_DEV_IO_TRACE
CONFIG_X86_ACPI_CPUFREQ
CONFIG_X86_SPEEDSTEP_CENTRINO depreciated as of kernel 2.6.24, use CONFIG_X86_ACPI_CPUFREQ
CONFIG_X86_SPEEDSTEP_ICH
CONFIG_X86_SPEEDSTEP_SMI
CONFIG_CPU_IDLE
CONFIG_CPU_IDLE_GOV_LADDER
CONFIG_CPU_IDLE_GOV_MENU

Those options are already in Fedora Core 7 and Ubuntu Gutsy (not Feisty) default i686 kernels. PowerTOP FAQ also suggest to disable CONFIG_IRQBALANCE and CONFIG_ACPI_DEBUG.

Also, you need to properly set APM and ACPI. Look at:

Kernel boot and module loading options

If you have an Intel chipset ICH5 or later (cf. lspci output), as in most modern Thinkpads, you should be using the integrated HPET timer (saves about 30 CPU wake ups per second). To see if hpet is enabled on your laptop:

grep hpet /proc/timer_list

If this does not display "Clock Event Device: hpet", then add

hpet=force
ATTENTION!
The ICH4 does have an HPET, but it is disabled for a good reason: Intel didn't test/validade it! Use of the ICH4 HPET is not recommended

to your kernel boot options (usualy in /boot/grub/menu.lst or lilo.conf). Note that "hpet=force" is only available by default in 2.6.24-rc2 and above (or as a separated patch for 2.6.22 and 2.6.23, see below).

On modern ThinkPads the HPET timer is automatically detected and enabled. On certain older machines hpet=force is required such as on the following machines:

Help needed
please add your ThinkPad to the above list, if hpet=force was required

Useful Patches

Thomas Gleixner High Resolution Timers (hrt) patchset brings many improvements, like the cpuidle work and Udo A. Steinberg and Venki Pallipadi "force enable HPET" patches (non HPET timers causes about 20-40 CPU wakeups/second, but HPET is often hidden by the BIOS due to Windows XP deficiencies). Those are fully merged in 2.6.24-rc1 vanilla kernel. See http://www.tglx.de/projects/hrtimers/

Kristen Carlson Accardi from Intel has a patchset to turn on "Aggressive Link Power Management" (ALPM) for the AHCI driver (for SATA bus). Also from Accardi, SATA Asynchronous Notification (SATA AN), alows SATA link to notify media insertions (thus avoid hal polling the cdrom). Those patches were merged in 2.6.24-rc2 kernel (AN needs also support in hal to be used). See: http://www.kernel.org/pub/linux/kernel/people/kristen/patches/SATA/alpm/

As of now (2.6.24-rc8), the linux kernel doesn't support PCI Express power management (aka PCIe ASPM, aka PCIe LPM). Shaohua Li from Intel submited a patch on LKML (http://lkml.org/lkml/2008/1/17/544 ) though, and reported it to reduce power consumption by 1.3 watts for a system with three PCIe links.

The HDAPS disk protection systems can reduce battery life. Matthew Garrett provides a patch that prevents hdaps kernel module to generate interrupts when this feature isn't used.

Useful sysctls

The meaning of those settings is explained case by case on the relevant sections of this document. But for convenience sake, we group them here too.

Note that the "ondemand" scaling governor is recommended by Intel developers for energy efficiency: it's expected to be more efficient than the "powersave" governor, or than userspace daemons (like cpufreq-utils, cpufreqd, powernowd...). Look here, here, or here for a kernel developer explanation about "ondemand" being better on modern Intel CPUs.

The "link_power_management_policy" tunable won't be available unless you run a 2.6.24-rc2 or more kernel, or applied Kirsten patchset, have an Intel AHCI compatible chipset, and use SATA drives.

echo 5 > /proc/sys/vm/laptop_mode
echo 0 > /proc/sys/kernel/nmi_watchdog
echo Y > /sys/module/snd_ac97_codec/parameters/power_save
echo 1 > /sys/devices/system/cpu/sched_mc_power_savings
echo ondemand > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
echo 1500 > /proc/sys/vm/dirty_writeback_centisecs
for i in /sys/bus/usb/devices/*/power/autosuspend; do echo 1 > $i; done
# those sysctl's are only available if you have an AHCI compatible SATA 
# controler and use kernel > 2.6.24-rc2 (or use Kristen ALPM patchset) : 
echo min_power > /sys/class/scsi_host/host0/link_power_management_policy
echo min_power > /sys/class/scsi_host/host1/link_power_management_policy
echo min_power > /sys/class/scsi_host/host2/link_power_management_policy

If you're running a kernel older than 2.6.22 do this. Not needed for kernels 2.6.22 onward:

cd /sys/devices/system/cpu/cpu0/cpufreq
cat ondemand/sampling_rate_max > ondemand/sampling_rate

ATA drives

Hard drives and CDRom drives spinning is very costly. To improve battery lifetime, you should reduce disks access (or devices polling) the more you can.

Hard Drives

The files access time update, while mandated by POSIX, is causing lots of disk write access; even accessing files on disk cache may wake the ATA or USB bus. If you don't use this feature, disable it by adding the relatime option to all relevant lines in the /etc/fstab, for example:

/dev/sda1 / ext3 relatime,errors=remount-ro 0 1

(On older kernels you may need to use noatime instead of relatime.)

Also consider merely using a larger value for the commit option. This defines how often changed data is written to the disk (it is cached until then). The default value is 5 seconds.

See man mount(8) for details on how the rel/noatime and commit options work.

Use laptop_mode to reduce disk usage by delaying and grouping writes. You should enable it, at least while on battery. See Laptop-mode for more details:

echo 5 > /proc/sys/vm/laptop_mode

The default kernel dirty page writeback frequency is very conservative. On a laptop running on battery, one might find more appropriate to reduce it:

echo 1500 > /proc/sys/vm/dirty_writeback_centisecs

Some power saving hard drives features can be activated with hdparm (beware that "-B 1" may reduce your drive lifetime, if you have lot of intermittent disk activity causing lots of heads load/unloads: so reduce I/O activity first, as explained above, in order to get longer disks idling periods). For more details look at How to make use of Power Management features :

hdparm -B 1 -S 12 /dev/sda # and/or any other disk device

SATA Link Power Management

On kernels 2.6.24 and new this enables SATA Link Power Management:

echo min_power > /sys/class/scsi_host/host0/link_power_management_policy
echo min_power > /sys/class/scsi_host/host1/link_power_management_policy
echo min_power > /sys/class/scsi_host/host2/link_power_management_policy

Disable it by replacing min_power with max_performance.

Which host-number to use depends on your laptop. It is typically 0 for the main drive, but for example on the T61 the 0 and 1 are for the pata interface used for the optical drive, and the main hard-disk is on host2.

On Ubuntu Hardy Heron with a 2.6.24-16 kernel, a suspend/resume cycle is much quicker if you disable SATA Link Power Management before initiating the suspend. As of Intrepid Ibex and kernel 2.6.27, this should be fixed. (Launchpad bug 234047, Kernel bug 10817)

Laptop Mode Tools

The Laptop Mode Tools utility implements many of the above power-saving measures from disks, and some others.

Optical drive

The optical drive is reported to consume power even when not accessed. See

The hald daemon polling tends to maintain the ATA buses out of power saving modes, and to wakeup CDROM drive (except if you have a kernel >= 2.6.24, hal >= 0.5.10, and SATA AN compatible devices). If you have a recent hald version, you can stop this polling when on battery:

hal-disable-polling --device /dev/scd0 # or whatever your CD drive is

start polling again when on ac:

hal-disable-polling --enable-polling --device /dev/scd0 # or whatever your CD drive is


If your hald is not recent enough, consider suspending it when running on battery. Some moderns SATA buses and drivers supports a notification mechanism (SATA AN - Asynchronous Events Notifications) that obsolete the need for polling on modern hardware; support for this feature had been merged in Linux 2.6.24-rc1 and HAL 0.5.10.

LCD Backlight/Brightness

The LCD backlight is one of the very major power drain. Reducing brightness to the lowest readable level will save a lot of battery lifetime. Also, don't forget to configure your screen saver to shutdown the screen backlight (rather than displaying some eye candy), when no activity for a few minutes.

You can also let the system automatically reduce brightness after a period of inactivity.

If you're choosing your Thinkpad laptop model, keep in mind that the screen size affect the battery time greatly: more power needed for larger screens.

The very recent, but xorg standard way to control backlight from CLI is using xbacklight. ie. to set backlight at half the brightness:

xbacklight -set 50

You should configure the DPMS to shutdown the screen when idle (rather than displaying a fancy but power consuming screensaver). ie. to turn off the display after 5 minutes of idling:

xset +dpms
xset dpms 0 0 300

Screensaver method

Alternatively, you can add a screensaver that does nothing but blank the screen and turn off the display. In this example, I'm using Linux Mint MATE 17.2, please adjust the keywords below accordingly. Also, you will need to enable DPMS as above, or as in How to make ACPI work#Screen blanking (Standby).

We need to create 2 files, the first one is /usr/share/applications/screensavers/screenoff.desktop:

[Desktop Entry]
Name=Screen off
Comment=Turns the screen off
Exec=/usr/lib/mate-screensaver/screenoff
TryExec=/bin/true
StartupNotify=false
Terminal=false
Type=Application
Categories=Screensaver;
Keywords=MATE;screensaver;dpms;
OnlyShowIn=MATE;

The second one is /usr/lib/mate-screensaver/screenoff, which needs to be executable:

#!/bin/sh
pidof mate-screensaver-preferences | grep -w $PPID && exit 0
xset q | grep -q "Monitor is On" && xset dpms force off

The Keywords section and the mate-screensaver-preferences should be customized according to your desktop (like Cinnamon, or GNOME, etc).

After creating those files, open the Screensaver preferences dialog and choose the screensaver with the name "Screen off". Now, when the screensaver is activated, the screen will fade to black and backlight will be turned off by DPMS.

Graphic controllers

All xorg Thinkpad graphics chipsets drivers (ati, radeon, fglrx, i810) have the same bug causing very frequent CPU wakeups when DRI is activated, even when you don't use any 3D application. This problem is partly fixed on xorg git tree but not released as of xorg 7.2. If you value more battery than 3D, you should disable DRI: put this on the /etc/X11/xorg.conf "Device" of you graphic controller:

Option          "NoDRI"

Also be sure that DPMS is working: grep DPMS /var/log/Xorg.0.log should output "DPMS enabled". If not, put Option "DPMS" in your config. See the section above about how to enable dpms driven display power saving.

On recent xrandr/xorg versions, you can disable the TV output (or any other detected as connected but not used outputs) when you're not using it: it's known to consume power.

xrandr # see all displays listed here, but that you don't actually use and disable them. 
xrandr --output TV --off # for instance (if "xrandr" above listed a connected output named "TV" that you don't use)

When you don't have an external monitor plugged, disable CRT and DVI output (for some, this can make a difference in power usage) :

echo crt_disable > /proc/acpi/ibm/video
echo dvi_disable > /proc/acpi/ibm/video

Some drivers have specials power saving mode, and/or allows underclocking the GPU. See also:

USB Subsystem

The kernel support an efficient USB 2.0 power saving feature if you enabled CONFIG_USB_SUSPEND. This may not trigger in when you have an USB device plugged (and beside, USB devices tends to suck power on their own), so avoid using such devices when on battery. To enable it by default, you must add the line

options usbcore autosuspend=1

to your /etc/modprobe.conf or add it to (and create if necessary) the file /etc/modprobe.d/usbcore depending on how your distribution organises modprobe configuration.

If on the other hand, you have usbcore built into your kernel, you can add this in the kernel boot options (ie. in grub's menu.lst):

usbcore.autosuspend=1

or at runtime, per device, with:

for i in /sys/bus/usb/devices/*/power/autosuspend; do echo 1 > $i; done 
for i in /sys/bus/usb/devices/*/power/level; do echo auto > $i; done

USB 1.1 is worst. It needs polling the bus frequently, hence can't really go in a low power mode when you enabled it, even if you don't have any device plugged. You'd better remove it when you don't use a 1.1 device:

rmmod uhci_hcd

If you don't intend to use any device needing USB 1.1 (unfortunately, the built-in bluetooth and fingerprint-reader are USB 1.1 devices), the USB 1.1 support can also be totally avoided. On Debian and derivatives, just do:

echo "blacklist uhci_hcd" >> /etc/modprobe.d/blacklist

PCMCIA/CardBus

Same for PCMCIA/CardBus. Some users experiences interrupts clouds (sometime up to several thousands interrupts/second) causing CPU wakeups, thus totally preventing the CPU to reach lower C-states. If you don't use PCMCIA, you may disable it the same way (unloading seems insufficient to restore the system properly, you have to boot without it):

echo "blacklist pcmcia" >> /etc/modprobe.d/blacklist
echo "blacklist yenta_socket" >> /etc/modprobe.d/blacklist

Sound

ALSA has a power saving feature that should be enabled on your kernel (CONFIG_SND_AC97_POWER_SAVE). Note that this low power mode won't trigger in unless you muted all sound inputs (micro, line in etc.). This feature has to be activated with:

amixer set Line mute nocap
amixer set Mic mute nocap
echo Y > /sys/module/snd_ac97_codec/parameters/power_save


Intel HD Audio

If you have Intel HD audio as your onboard sound controller, substitute the following for the last line in the above sequence:

echo Y > /sys/module/snd_hda_intel/parameters/power_save_controller

You also may wish to decrease the sound poweroff timeout to something shorter, like 1 second after last playback:

echo 1 > /sys/module/snd_hda_intel/parameters/power_save

Additional Tweaks

You can unload all sound related modules when you are on battery, or mute the sound system (echo mute > /proc/acpi/ibm/volume).

See also How to enable audio codec power saving.

Wireless Interface

intel wireless

Wireless network consume a lot of power. To save power, you can kill the Wi-Fi radio when it's not in use:

echo 1 > /sys/bus/pci/devices/*/rf_kill

If you need Wi-Fi, you can also reduce power consumption (at the price of performances) by activating the power saving modes:

iwpriv eth1 set_power 5

For drivers using the new Wi-Fi kernel framework (mac80211/cfg80211), the canonical way to do this is now:

for i in /sys/bus/pci/devices/*/power_level ; do echo 5 > $i ; done

Most drivers, like ipw2200, that don't use the new mac80211 framework place the interfaces in aggressive scanning mode when they are not associated with any Access Point, even when the interface is down (more info about this on Intel's LessWatts website). This behavior consumes a lot of power, even more than when the interface is plain active and in use. But this can disabled at module's load time :

rmmod ipw2200
modprobe ipw2200 associate=0

You can fix this setting by placing the following in /etc/modprobe.d/options (Debian/Ubuntu) or in /etc/modprobe.conf (Red Hat/Fedora):

options ipw2200 associate=0

Reducing beacon intervals on your Access Point to 1 per second will also reduce network card interrupts, therefore power savings. This shouldn't have negatives side effects.

In recent kernels, the powersaving on the intel ipw3945 has been disabled, as for some it is unstable. For others it worked just fine.

See this patch. You may wish to try changing '.broken_powersave=true' to false, in /usr/src/linux/drivers/net/wireless/iwlwifi/iwl-3945.c if you wish to enable powersave.

See also, to activate power saving on the wireless network card:

Ethernet Controler

If you don't use Wake-on-LAN, you should disable it for your network card, because it sucks a lot of power:

ethtool -s eth0 wol d

If you can, try to reduce useless network activity on your ethernet segment, coming to your NIC (ie. uneeded broadcasts), those cause interrupts and CPU wakeups.

Forcing 100Mbps full-duplex speed on a gigabit ethernet NIC can also save a lot of power (~1W) on most network workloads. This also reduces components temperature (e.g., thermal sensor 0xC0 on the T43 cools down by 5 degree between 1000Mbps and 100Mbps, and another 1 degree for 10Mbps).

 ethtool -s eth0 autoneg off speed 100

Note, however, that if the network device on the other side has auto-negotiation enabled (which is very common) and you turn auto-negotiation off, the other side will assume half-duplex mode and you will experience a significant loss of performance.

Bluetooth

When you don't need bluetooth, disable it. Because of its radio, bluetooth is not power friendly.

hciconfig hci0 down ; rmmod hci_usb
echo disable > /proc/acpi/ibm/bluetooth

Modem

When was the last time you used your analog modem? If you can't remember, you probably just don't need it. If it is on a separate module in your laptop, simply remove it. Store it in a ESD safe place (like the bag in which your last addon card or hard drive was packed), in case you should need it again. This won't save you a lot of power and weight, but why carry something around you never use.

System Fans

Fans consumes power when running, so you may look at the ACPI fan control script.

Misbehaving Userland

You should avoid using Beagle, Compiz, Beryl, XMMS, gnome-power-manager and Evolution while on battery. Look at the PowerTOP's known problems list.

Deactivate desktop animations (blinking cursor on the terms, animated wallpapers, ...): they cause regular X (therefore kernel and CPU) wakeups.

In short, while on battery, you should stop all applications that don't really stay idle when you're not using them. This means applications that:

  • Wakes up the CPU too often (by polling something, because of too short select() timeouts, ...)
  • Access the disks at regular intervals
  • Access an hardware bus (USB, ATA, ...) at regular intervals

To find those offenders run:

  • strace -p $(pidof yourapp) # for all your running applications
  • powertop
  • dstat -t -c --power --top-cpu --top-io --top-bio --top-latency --top-cputime
  • sysctl vm.block_dump=1 # and look at dmesg
  • ps aux | awk '{print$10,$11}' | sort -n # will list all running softs sorted by used cpu time

Please, don't forget to fill a bug when you find such a misbehaving software.

NOTE!
Not all software is evil, buggy or badly written. Some produce regular activity because they have to, in order to provide their intented functionality. Think twice before filling bugs about this.

See Also

External resources