"CPU-IDLE route" chooser

[cyring]

Based on my R&D, Intel and AMD architectures don't have the same appropriated idle instruction:

1. Intel sleeps best in MONITOR/MWAIT
2. whereas AMD, especially Zen, sleeps in I/O WAIT, first, or in HALT, second.

Code is available for testings in develop branch.

Prerequisites are:

• CPU-IDLE driver has to be registered.
  This can be achieve from the UI
• However as mentioned in README.md, you will have to make some room !
• See FAQ how to register CoreFreq as the Kernel idle handler
  Depending of the architecture, blacklist straight from boot some Linux cpu-idle drivers.

[cyring]

This feature is marked as Experimental.

[cyring]

Latest commit f4a546a appears to be stable.

[cyring]

The I/O-MWAIT handler issue is debugged: the BAR value was missing per SMT. 😕

Especially with Intel/Westmere, Core traps the in instruction on a byte only; whereas with AMD/Ryzen/Matisse, the trigger works with both inb or inw.

Also with Intel, I/O MWAIT Redirection must be enabled ...
... in addition to CPU-IDLE route selected to I/O.
The prerequisite is to register CoreFreq as a the kernel cpu-idle driver. See FAQ

As a test I have reprogrammed the base address in Intel/AMD MSR registers (respectively 0xe4, 0xc0010073) just before launching the driver ... and I/O-MWAIT still works. 😎

About adding a new driver parameter IOMWAIT_BAR to let the User reprogram an address ?
This could help with buggy ACPI in BIOS.
But; User has to make sure of the BAR address chosen ... pretty risky.