Linux 5.9 Adding New Knob To Control Default Boost Value For Real-Time Workloads

Primarily driven currently by Arm big.LITTLE use-cases like high-end smartphones where you may be running on battery and not want to boost the CPU performance too high for real-time (RT) tasks, Linux 5.9 is adding new capabilities around setting the default boost value.

Contributed by Arm, Linux 5.9's sched/uclamp code is setting the capability to control the RT default boost value. This can be used for lowering the default boost value to in turn conserve energy consumption of real-time tasks. It's assumed that vendors will tune this value for the best performance/power for RT tasks depending upon the device and potentially differ depending upon the AC/battery power source.

This new sysctl tunable is configurable as sysctl_sched_uclamp_util_min_rt_default should you not want your RT tasks running on the highest capacity CPU and at peak frequency. While for now this is about Arm big.LITTLE, Intel may find use of the code next year with Alder Lake and the hybrid core configuration expected there with a mix of Atom and Core processors.

The addition is currently queued in sched/core ahead of the Linux 5.9 merge window. This patch adds more documentation around the sched util clamping for those interested in all the fine details.