CPUFreq & P-State Go Through A Redesign With Linux 4.6

[phoronix]

Phoronix: CPUFreq & P-State Go Through A Redesign With Linux 4.6

The power management updates for Linux 4.6 are quite hearty...

http://www.phoronix.com/scan.php?page=news_item&px=Linux-4.6-Power-Management

[Adarion]

I wonder if these changes will yield any power consumption improvements or if this is just to make the code more sustainable / manageable / robust...
Though even saving a few computing cycles by simplifications might sum up in some power savings.

It's mainly a slightly different construction site but I really wish the Intel folks would finally kick ACPI's ass so much that we don't end up with all these stupid problems and hiccups caused by it. Be it kernel freezes / box not booting without acpi=off or failing to S3 (or wake up from there) or just stuff that just won't work or causes other misfortunes.

[c117152]

I wonder if these changes will yield any power consumption improvements or if this is just to make the code more sustainable / manageable / robust...
Though even saving a few computing cycles by simplifications might sum up in some power savings.

Better performance. Marginally better power consumption. pstate is active while cstate is idle. So, presumably based on their pull request description, they reduced the overhead when the govonor throttles frequencies in various work loads (not idle). Like, when 3 threads are pushing 3 cores and are interchangeably locking on and off so the frequencies jump and burst but aren't getting into cstates. It's more games and databases loads, and less browsing loads.

Though they did mess around with acpi sleep*.... Who knows, maybe my i7 might finally go to sleep instead of hibernating

*

drivers/acpi/sleep.c | 35 +

[VikingGe]

Nice, this should make it possible to implement a governor that ramps up CPU clock speeds before any code is run in order to improve responsiveness and performance in mixed-load scenarios.