More AMD P-State Driver Improvements Queued For Linux 6.11

[phoronix]

Phoronix: More AMD P-State Driver Improvements Queued For Linux 6.11

Earlier this month AMD Fast CPPC support was queued into the Linux power management subsystem's "-next" codebase ahead of the Linux 6.11 cycle. Additional AMD P-State driver enhancements are now deemed ready and submitted for staging ahead of next month's Linux 6.11 cycle kicking off...

https://www.phoronix.com/news/AMD-P-State-Linux-6.11-Round-2

[ptr1337]

Great! Then only cpb boost is left and the mainline kernel has a pretty nice implementation.

2 Days ago I have reported a bug to them, for the default enabling of amd-pstate. Got pretty fast resolved by Mario.
It appeared, that Zen 2 CPUs on Mainboards with an older BIOS version were not able to load amd-pstate, due no reported frequencies, this resulted into a stuck frequency on acpi-cpufreq. Gladly it got solved and looking forward to have it merged.

[mrg666]

I hope core hotplug capability also comes soon to completely turn off unused cores to save power especially in laptops.

[agd5f]

I hope core hotplug capability also comes soon to completely turn off unused cores to save power especially in laptops.

This happens automatically when CPU cores are idle.

[Mitch]

I hope core hotplug capability also comes soon to completely turn off unused cores to save power especially in laptops.

Is this what other platforms and OSes do? I've never heard of it but it seems great. For some reason I've only ever heard of core hotplugging for virtualization (e.g. dynamically give and take guest OS cores without rebooting the guest OS)

[mrg666]

Is this what other platforms and OSes do? I've never heard of it but it seems great. For some reason I've only ever heard of core hotplugging for virtualization (e.g. dynamically give and take guest OS cores without rebooting the guest OS)

i have worked on the hotplug drivers of ARM SCOs for Android. MPdecision is one of them, there are many others. Here is a link.

https://xdaforums.com/t/cpu-governors-hotplug-drivers-and-gpu-governors-explained-part-3-updated-30-12-16.3532210/

[mrg666]

This happens automatically when CPU cores are idle.

Not on x86. They just go to low energy, idle (C3 or lower) state. But they are not shut down like in ARM. I am curious to see if it is at all possible/feasible without a new C state added to X86.

[agd5f]

Not on x86. They just go to low energy, idle (C3 or lower) state. But they are not shut down like in ARM. I am curious to see if it is at all possible/feasible without a new C state added to X86.

Internally the platform firmware can shut them off when conditions allow.

[mrg666]

Internally the platform firmware can shut them off when conditions allow.

True. But after OS boots, OS needs to do that. Firmware cannot interfere with that. Right?

[agd5f]

True. But after OS boots, OS needs to do that. Firmware cannot interfere with that. Right?

Firmware handles it transparently, even at runtime.