Linux 5.18 Power Management Brings Improvements For Both Intel & AMD
In addition to the Intel Hardware Feedback Interface (HFI) support sent in through the thermal subsystem, the Linux 5.18 power management updates have some noteworthy changes too.
Intel's Idle Linux driver (intel_idle) now has proper/native support for upcoming Xeon Sapphire Rapids CPUs. There is also the ability to have better control over the the C-states on these forthcoming server CPUs. Sapphire Rapids also has a core C6 optimization with the intel_idle changes.
Meanwhile Intel's P-State driver will now use the default default Energy Performance Preference (EPP) exposed by the firmware. To now Intel P-State has used a hard-coded EPP default for Alder Lake to try to ensure a maximum 1 core turbo frequency is attainable out of the box, but moving forward the Linux kernel will respect the EPP default of the firmware. The hard-coded value will be used should the firmware not expose a default EPP value.
Over on the AMD CPU side, the CPUPower utility that lives within the kernel source tree now supports running in conjunction with the AMD P-State driver. Linux 5.17 introduced the AMD P-State driver as an alternative to ACPI CPUFreq for Zen 2 systems and newer while the CPUPower changes are now in place for working in the presence of amd_pstate. There is also new tracing tool for the AMD P-State driver.
Virtualized Linux guests will also now take the ACPI S4 hardware signature into account by default. See the power management pull for the full list of changes this cycle.
PM/ACPI maintainer Rafael Wysocki of Intel also sent in the ACPI updates at the same time last week. There is a new Arm Generic Diagnostic Dump and Reset device driver, various hardware quirks, and other routine ACPI churn.