ARM Now Defaulting To Schedutil Governor, Other Power Management Work For Linux 5.9
Sent out earlier this week was the main power management pull request. Included with that were some interesting bits:
- The Energy Model (EM) framework has been extended to support devices besides just CPUs.
- Ice Lake Xeon server idle states were added to the intel_idle driver along with other changes. This includes Ice Lake Xeons ramping up their frequencies slower and thus a new workaround introduced.
- Support for Intel Sapphire Rapids with the RAPL power-capping driver. There is also PL4 / Power Limit 4 control support.
- A new sysfs toggle for controlling CPU energy-efficency optimizations for the processor.
Various CPUfreq fixes and more are also included. This pull request has the initial batch of power management updates.
Intel's Rafael Wysocki who oversees the power management code then sent out a second PR on Friday. Notable with this second batch of work includes:
- A new NVIDIA Tegra CPUFreq driver. This new driver is the Tegra194 SoC CPUFreq driver for Xavier hardware.
- A fix to the Intel P-State driver to use the correct maximum frequency value when the MSR turbo ratio limit bit is set to zero.
- Making Schedutil the default governor on ARM and ARM64 (AArch64). This follows Intel P-State working to transition to Schedutil by default as well. ARM already has been defaulting to Schedutil for big.LITTLE systems as this governor is required for energy aware scheduling (EAS). Schedutil on ARM is also important for honoring utilization clamp performance requests and other functionality.
Ultimately the Linux kernel developers are still aiming at Schedutil (the scheduler utilization) CPU frequency scaling governor to become the default for all drivers and architectures but isn't there quite yet.
Rafael also sent in ACPI updates for Linux 5.9 this week. Notable there is eliminating "significant" AML processing overhead, updating the ACPICA code, and various fixes.
1 Comment
