The Linux kernel uses several power management strategies:

  • Dynamic voltage frequency scaling (DVFS) adjusts the CPU frequency and the different voltages depending on the system’s load. The combination of a CPU frequency and a set of voltages for that frequency is called an operating point.

  • Suspend to memory allows for the system to sleep waiting for an event. On suspend, all system devices, including CPU and memory, enter low power mode. On resume, the system will continue from the same state it was in before it suspended.

  • Power off, which brings the system to a halt until an event wakes the system. On power off the system power remains enabled and the system is placed on its lowest consumption mode. On wake up, the bootloader starts up again and the system is initialized.

Dynamic Voltage Frequency Scaling

Dynamic Voltage Frequency Scaling is enabled by default, so the CPU frequency and voltages of a running system will adapt to the system’s load. The DVFS subsystem is controlled through the /sys/devices/system/cpu/cpuN/cpufreq/ sysfs path, where N corresponds to the core number:

Sysfs attributes of CPU frequency scaling
# ls /sys/devices/system/cpu/cpu0/cpufreq/
affected_cpus                  related_cpus                   scaling_governor
cpuinfo_cur_freq               scaling_available_frequencies  scaling_max_freq
cpuinfo_max_freq               scaling_available_governors    scaling_min_freq
cpuinfo_min_freq               scaling_cur_freq               scaling_setspeed
cpuinfo_transition_latency     scaling_driver                 stats

The scaling_cur_freq file shows the current frequency of the core:

# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq

The available frequencies are shown in scaling_available_frequencies.

The DVFS subsystem can be configured with different governors that control its behavior. The available governors are listed in scaling_available_governors.

# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_governors
conservative ondemand userspace powersave interactive

The current governor is shown in scaling_governor, and by default is set to the interactive governor.

  • ondemand: This governor does a periodic polling to adjust the operating mode to the system load.

  • conservative: Similar to the ondemand governor, the conservative governor performs graceful frequency and voltage changes that are better suited to battery powered devices.

  • interactive: This governor also adjusts the operating mode to system load periodically, but in a way that is more responsive to interactive workloads.

  • powersave: This governor uses the lowest available frequency without dynamic adaptation.

  • userspace: This governor allows for user space to select the operating frequencies.

Manually select a working CPU frequency
# echo userspace > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies
396000 792000 996000 1200000
# echo 792000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_setspeed
# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq

Suspend to memory

The Linux kernel can also perform a suspend to memory or suspend to RAM operation. When entering this low-power mode, the system state is kept in self-refreshing RAM while the system enters a low-power-consumption mode. The system resumes when a previously selected interrupt is received, restores the previous state, and continues running from where it left off. There is often a trade-off between the depth of the low-power mode and the speed at which the system can resume.

Entering suspend mode

To enter suspend mode, either:

  • Run /bin/standby from the command line, or

  • Briefly press (less than two seconds) the power key

Resume events

The system can resume from any interrupt generating event, including:

  • GPIOs

  • RTC alarm

  • Power key event

  • Wake on LAN (if supported by the driver)

By default, the power key event will always be registered to resume a system from suspend. See Configuring wake up sources for additional details.

Power off

The Linux kernel can perform a power off operation that places the Power Management IC (PMIC) in power off mode, disabling all power sources that are not needed for wake up.

Entering power off

To enter power off mode you can do one of the following:

  • Run the command poweroff from the command line to perform a controlled software power off sequence

  • Press the power key for longer than two seconds, but less than nine seconds, to perform a controlled software power off sequence

  • Press the power key for nine seconds or longer to perform an unconditional hardware power off (not controlled by software)

Wake up events

You can wake up the target from power off mode with an interrupt event to the PMIC, but not to the CPU as it will not be powered. These include:

  • Power key event

  • PMIC’s GPIOs

  • PMIC’s RTC alarm

The power key will always wake up the system.

Configuring wake up sources

GPIO resume

To program a GPIO N to generate a wake up interrupt when it changes status:

  • Export the GPIO:

    echo -n N > /sys/class/gpio/export
  • Configure it as input:

    echo in > /sys/class/gpio/gpioN/direction
  • Configure it as interrupt (either rising or falling edge):

    echo rising > /sys/class/gpio/gpioN/edge
  • Configure it as wake up source:

    echo enabled > /sys/class/gpio/gpioN/device/power/wakeup

RTC alarm resume

To enable the RTC wake alarm to trigger in 60 seconds:

  • Enable the RTC device as a wake up source:

    echo enabled > /sys/class/rtc/rtc0/device/power/wakeup
  • Program the time when the alarm should trigger an interrupt (format is seconds since the epoch or, if there’s a leading +, seconds in the future or, if there’s a leading +=, seconds ahead of the current alarm):

    echo +60 > /sys/class/rtc/rtc0/wakealarm