Linux manages power in two broad ways: it can put the whole system into a sleep state, or it can reduce power use by individual devices and processors while the system remains active. These mechanisms are related but not interchangeable. Which states and behaviors are available depends on the kernel configuration, hardware, drivers, and platform firmware.
What is kernel power management?
Kernel power management coordinates hardware and software choices that reduce energy use or change performance. It covers both system-wide transitions, where userspace stops running, and working-state controls that act on devices or CPUs while the computer is otherwise on. The kernel documentation describes these as distinct areas; their effects depend on the system rather than guaranteeing a particular battery-life or performance result.
Four useful concepts are system sleep, device runtime power management, CPU idle, and CPU performance scaling. The first is a global state change. The other three manage components or processor behavior during normal operation.
How do Linux system sleep states differ?
System sleep affects the entire computer: userspace cannot execute, and the kernel and platform place hardware into lower-power states. The kernel may support up to four states, but not every machine or kernel configuration supports every option. The Linux kernel’s System Sleep States documentation, authored by Rafael J. Wysocki, describes the following distinctions:
#1 Best Overall
| State | What happens | Typical trade-off |
|---|---|---|
| Suspend-to-idle | Userspace is frozen, timekeeping is suspended, and I/O devices enter low-power states; CPUs can use deep idle states. | Uses less power than an active system, while resume behavior and available wake sources depend on the platform. |
| Standby | Non-boot CPUs are taken offline, with the platform entering a deeper state than suspend-to-idle. | Typically saves more power than suspend-to-idle, with longer resume latency. |
| Suspend-to-RAM | Memory remains in self-refresh while the rest of the system is placed in low-power states. | Requires platform support and continued memory power; wake behavior depends on supported wake sources. |
| Hibernation | The kernel writes a memory image to persistent storage and can then power down nearly all hardware. | Requires suitable persistent storage and a more involved save-and-restore transition. |
These descriptions are broad behaviors, not guarantees about a particular computer. The firmware, kernel configuration, and devices that can signal a wake event determine which sleep states are usable and how the machine resumes.
What is the difference between runtime power management and suspend?
Runtime power management can put an individual device into a low-power state while the system remains running and userspace continues to execute. As the kernel documentation puts it, “Many devices are able to dynamically power down while the system is still running.” This is coordinated among the device driver, its bus or subsystem, and the kernel’s PM core. Device dependencies and bus rules can constrain when a device may suspend. During system sleep or hibernation, runtime-suspended devices may also require special handling.
Rank #2
System sleep is different: it is a whole-system transition. A device’s runtime policy does not decide whether the machine enters suspend or hibernation.
Control runtime policy with power/control
For devices that expose the interface, the sysfs file power/control sets the runtime power-management policy:
Rank #3
autoallows runtime power management.onprevents runtime power management and brings the device back to full power if needed.
Changing this setting does not remove the device from system-wide suspend or hibernation. It controls runtime behavior, not the device’s participation in those global transitions. The interface is documented in the kernel’s Device Power Management Basics.
Wakeup capability and wakeup policy
A device may have the hardware ability to wake the system, but that does not mean wakeup is enabled. Where supported, the device’s power/wakeup sysfs file exposes the policy choice. Enabling wakeup can consume power, even though a wake-capable device may make a deeper system sleep practical. The kernel documentation treats device suspend and resume as coordinated operations involving driver and subsystem callbacks.
How do CPU idle and performance scaling fit in?
CPU idle management selects an idle state when a CPU has no work to run. CPU performance scaling changes processor performance behavior. Both are working-state power-management mechanisms, but they address different situations and are documented as separate kernel subsystems in the CPU idle documentation and CPU performance scaling documentation.
There is no universal CPU policy or driver that produces a fixed energy saving or performance result. Comparisons need to identify the processor, kernel version, active driver, and workload; behavior from one system should not be assumed to apply to another.
Best Value
Which mechanism applies to a particular situation?
- The whole computer should pause: consider system sleep. Check which states the platform and kernel support, what devices can wake it, and the acceptable resume behavior.
- A device is idle while Linux remains in use: runtime power management may let that device enter a lower-power state. Its driver, bus, and parent-child relationships affect what is possible.
- A processor has no work to run: CPU idle selects an idle state.
- Processor performance behavior should change: CPU performance scaling is the relevant subsystem; the result depends on the hardware, active driver, kernel, and workload.
These mechanisms can coexist. For example, a device may use runtime PM while the computer is active, and the system may later enter a global sleep state. Treating their controls as separate makes it easier to understand why changing one setting does not necessarily change the other.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
