CPU Frequency

Records changes in the CPU power management scheme through the Linux kernel ftrace infrastructure.

• Enabling the power/cpu_frequency ftrace event. (For details, see TraceConfig below.) This will record an event every time the in-kernel CPU frequency scaling driver changes the frequency. Note that this is not supported on all platforms. According to our experience, it works reliably on Arm-based SoCs but produces no data on most Intel-based modern platforms. This is because recent Intel CPUs use an internal DVFS which is directly controlled by the CPU and does not expose frequency change events to the kernel. Also note that even on Arm-based platforms, the event is triggered only when the CPU frequency changes. In many cases, the CPU frequency will not change for several seconds, which will show up as an empty block at the start of the trace. You are advised to combine this with polling to obtain a reliable snapshot of the initial frequency.
• Polling sysfs by enabling the linux.sys_stats data source and setting cpufreq_period_ms to a value greater than 0. This will periodically poll /sys/devices/system/cpu/cpu*/cpufreq/cpuinfo_cur_freq and record the current value in the trace buffer. This works on both Intel- and Arm-based platforms.

On most HarmonyOS devices, frequency scaling is performed based on the cluster (group of big/little cores). Therefore, it is common to see that four CPUs change the frequency at the same time.

This is not supported on modern Intel platforms for the same aforementioned reasons of power/cpu_frequency.

Idle transitions are relatively fast. A CPU can enter and leave idle states hundreds of times in a second. Idleness must not be confused with full device suspension, since the latter is a stronger and more invasive power saving state. CPUs can be idle even when the screen is on and the device looks operational.