Re: [PATCH 7/7] cpu: make preferred CPU policy writers mutually exclusive

From: Yury Norov

Date: Fri Nov 20 2026 - 11:26:22 EST


On Fri, Nov 20, 2026 at 05:09:57PM +0100, Peter Zijlstra wrote:
> On Fri, Nov 20, 2026 at 10:59:43AM -0500, Yury Norov wrote:
> > On Fri, Nov 20, 2026 at 04:38:51PM +0100, Peter Zijlstra wrote:
> > > On Fri, Nov 13, 2026 at 01:42:56PM -0500, Yury Norov wrote:
> > > > Serializing preferred mask updates still lets competing policy drivers
> > > > repeatedly override each other's selections.
> > >
> > > Competing controllers is not going to happen.
> >
> > That's the purpose of this patch. You can't stop people writing
> > alternative controllers,
>
> Why not? As I said last time; the whole scheme is co-operative, it
> relies on all guests doing the same thing.

The whole scheme is co-operative so far. I'm experimenting with
cgroups right now to teach the host to catch and punish cheaters.

Throttling may be driven by a different type of control signal,
like [1] does for overheating, or a mix of signals.

Even for the current steal time scheme, one may want to implement
their own controller, for example built into their driver.