x86_energy_perf_policy fix

[mikethebos]

https://github.com/erpalma/throttled/issues/107
I fixed my HWP mode issue by using x86_energy_perf_policy to set hints.

[erpalma]

This is not the right way to "fix" your issue. That subprocess call has higher overhead than regular code plus using shell=True is not a great idea in general. HWP_Mode is loaded from the AC profile only thus we should replicate it to the BATTERY profile at least.

[mikethebos]

Debian 9 "stretch" does not implement the Intel p_state driver in the cpufreq package, so this was the most convenient method I could find. Plus, I added to improve efficiency:
exit_event_wait()

[erpalma]

If you don't have the p_state driver, then you can't control the hardware p-states, unless I'm missing something...

[mikethebos]

Incorrect, according to: http://manpages.ubuntu.com/manpages/bionic/man8/x86_energy_perf_policy.8.html

x86_energy_perf_policy displays and updates energy-performance policy settings specific to
       Intel  Architecture  Processors.   Settings are accessed via Model Specific Register (MSR)
       updates, no matter if the Linux cpufreq sub-system is enabled or not.

[erpalma]

Ok I missed that x86_energy_perf_policy was using MSR directly. Still, to be consistent I prefer to replace the subprocess call with a piece of code writing the correct MSR.

[mikethebos]

C code

[mikethebos]

Let me know if you need help with the MSR... I can test if needed...
I recommend that you make 2 new config parameters like the following:

HWP_FALLBACK_AC: balance-performance
HWP_FALLBACK_BATT: balance-power

[erpalma]

Why do you need a separate entry for fallback? Still, HWP is quite tricky because is also set by tlp. The detection code that I use is quite ugly and it is not that much fail safe I guess, but it works and I don't see a better way of doing it.