Intel P-State Driver Preparing To Migrate From "Powersave" To Passive Schedutil Default

It looks like in the next one or two kernel releases we could see Intel transitioning their CPU frequency scaling governor default from the long-standing powersave to the modern schedutil governor. It's now believed schedutil should be at least as good as powersave.

Schedutil is the modern Linux CPU frequency scaling governor that makes use of scheduler utilization data (hence its name) in making decisions about ramping up or down the CPU clock frequencies. Schedutil has been around for a while now and continues to receive improvements. It's been looked at as the future of CPUfreq governors.

We're now hitting a state of maturity where Schedutil should be more competitive across the board. Intel's Rafael J. Wysocki has prepared patches in changing the default behavior of the P-State CPU frequency scaling driver that is used by all current Intel CPUs going back to the Sandy Bridge days.

In particular, since the recent frequency invariance patches, Rafael believes the Schedutil support is in good standing for making it the default.

He has a branch where intel_pstate now defaults to the passive mode for systems without HWP (Hardware P-States) and also uses the schedutil governor by default.

Another branch is carrying the patch about changing the behavior for the systems with HWP. This change is considered more experimental and thus staged separately from the modifying of the non-HWP system default.

As of writing these patches aren't queued in his power management "-next" branch, so they might not be vetted in time for the upcoming Linux 5.7 kernel merge window but otherwise we'd likely be looking at this default change to go into Linux 5.8. Once the change does happen, we'll be around with some power and performance benchmarks of the P-State governor impact.