Re: NetBSD Dom0 and (kind of) power management

[Jukka Ruohonen <jruohonen%iki.fi@localhost>]

On Fri, Mar 18, 2011 at 07:54:47AM +0100, Christoph Egger wrote:
> The key point is that the Dom0 has virtual cpus. And you don't know
> which vcpu is scheduled on which physical cpu.
> 
> Dom0 has to invoke hypercalls for all kind of things that must happen
> on the physical cpus.
> C- and P-States belong to those "things".

That is of course the basic scheme.

But when reading the hypervisor code and looking at some Intel's presentations
(see e.g. [1]), the Xen 4.0 hypervisor implements (or rather, duplicates) its
own full-blown, Linux-dependent, CPU PM facility.

From the x86 and ACPI standpoint this seems to imply that not only are the
basic things like hypercalls necessary, but we would have to implement normal
x86 drivers and interfaces to match the ever-changing Linux/Xen APIs. I don't
think that is desirable, so in my opinion the old "dom0-based" approach is
more reasonable for NetBSD.

- Jukka.

[1] http://www.xen.org/files/xensummit_oracle09/XenSummit09pm.pdf