3 Feb
2015
3 Feb
'15
3:04 p.m.
On Tuesday, February 03, 2015 09:40:58 AM Viresh Kumar wrote:
On 3 February 2015 at 06:11, Rafael J. Wysocki rjw@rjwysocki.net wrote:
This becomes quite inefficient on a system with many CPUs having different policies. My approach would be to somehow attach the fallback policy information to the CPU device object.
It will be same as the per-cpu approach which we are fed up of.
No, it won't be the same. The per-CPU memory is special.
The general idea of the existing code is sane in my view. It connects the fallback policy information with the given CPU directly, which generally is a good idea. Storing that information in the per-CPU memory isn't a good idea, however.
--
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.