Re: 50 Watt idle power regression bisected to Linux-3.10

[H. Peter Anvin]

This is a multi-part message in MIME format.On 12/11/2013 03:14 PM, Borislav Petkov wrote:
> On Wed, Dec 11, 2013 at 03:08:35PM -0800, H. Peter Anvin wrote:
>> So I would like to propose that we switch to using a percpu variable
>> which is a single cache line of nothing at all. It would only ever
>> be touched by MONITOR and for explicit wakeup. Hopefully that will
>> resolve this problem without the need for the CLFLUSH.
> 
> Yep, makes a lot of sense to me to have an exclusive (overloaded meaning
> here :-)) cacheline only for that. And, if it works, we'll save us the
> penalty from the CLFLUSH too, cool.
> 

Here is a POC patch... anyone willing to test it out?

Two obvious things to watch out for:

1. I couldn't actually spot any obvious cases of a deliberate monitor
   trigger.

2. Should we do cpu_relax() for all users, not just powerclamp?

	-hpa