From: Landry Breuil <landry@openbsd.org>
Subject: Re: scmi: hook up to cpu_* to get apm working
To: Tobias Heider <tobias.heider@stusta.de>
Cc: Mark Kettenis <mark.kettenis@xs4all.nl>, tech@openbsd.org, patrick@openbsd.org
Date: Sat, 14 Jun 2025 08:09:32 +0200

Le Sat, May 24, 2025 at 05:41:25AM +0200, Tobias Heider a écrit :
> On Fri, May 23, 2025 at 02:58:45PM +0200, Mark Kettenis wrote:

<snip>

> > Well, pl_ifreq is optional.  Firmware can set it to 0.  So you have to
> > use pl_perf.
> > 
> > I'd argue that pl_perf is better anyway, since out hw.setperf also is
> > a "abstract performance level".  It just has a different scaling.  To
> > determine the scaling you simply need to figure out the min and max
> > pl_perf for the domain.  See the arm64 acpicpu(4) diff I mailed out
> > recently.
> > 
> > (The ACPI mechanism was clearly designed to expose this SCMI
> > performance control mechanism to an ACPI aware OS).
> > 
> 
> Right, makes sense.
> 
> I can just collect the min and max while I am building the list of
> per-domain perf levels. How do you like this one? 

fwiw, im still running this diff (or a precedent iteration) to have perf
levels on the omnibook, and it works fine. ok for me too.

Landry