Re: [PATCH v2 08/11] dt-bindings: firmware: arm,scmi: Extend bindings for protocol@13

[Ulf Hansson]

On Wed, 26 Jul 2023 at 13:12, Ulf Hansson <ulf.hansson@xxxxxxxxxx> wrote:
>
> On Fri, 21 Jul 2023 at 20:38, Sudeep Holla <sudeep.holla@xxxxxxx> wrote:
> >
> > On Fri, Jul 21, 2023 at 08:33:04AM -0600, Rob Herring wrote:
> > > On Fri, Jul 21, 2023 at 12:55:35PM +0100, Sudeep Holla wrote:
> > > > On Fri, Jul 21, 2023 at 01:42:43PM +0200, Ulf Hansson wrote:
> > > > > On Wed, 19 Jul 2023 at 17:17, Sudeep Holla <sudeep.holla@xxxxxxx> wrote:
> > > > > >
> > > > > > On Thu, Jul 13, 2023 at 04:17:35PM +0200, Ulf Hansson wrote:
> > > > > > > The protocol@13 node is describing the performance scaling option for the
> > > > > > > ARM SCMI interface, as a clock provider. This is unnecessary limiting, as
> > > > > > > performance scaling is in many cases not limited to switching a clock's
> > > > > > > frequency.
> > > > > > >
> > > > > > > Therefore, let's extend the binding so the interface can be modelled as a
> > > > > > > generic performance domaintoo. The common way to describe this, is to use
> > > > > > > the "power-domain" DT bindings, so let's use that.
> > > > > > >
> > > > > >
> > > > > > One thing I forgot to ask earlier is how we can manage different domain IDs
> > > > > > for perf and power domains which is the case with current SCMI platforms as
> > > > > > the spec never mandated or can ever mandate the perf and power domains IDs
> > > > > > to match. They need not be same anyways.
> > > > >
> > > > > Based upon what you describe above, I have modelled the perf-domain
> > > > > and the power-domain as two separate power-domain providers.
> > > > >
> > > > > A consumer device being hooked up to both domains, would specify the
> > > > > domain IDs in the second power-domain-cell, along the lines of the
> > > > > below. Then we would use power-domain-names to specify what each
> > > > > power-domain represents.
> > > > >
> > > > > power-domains = <&scmi_pd 2>, <&scmi_dvfs 4>;
> > > > > power-domain-names = "power", "perf";
> > > > >
> > > > > I hope this makes it clearer!?
> > > >
> > > > Yes it make is clear definitely, but it does change the definition of the
> > > > generic binding of the "power-domains" property now. I am interesting in
> > > > the feedback from the binding maintainers with respect to that. Or is it
> > > > already present ? IIUC, the ones supported already are generally both
> > > > power and performance providers. May be it doesn't matter much, just
> > > > wanted to explicit ask and confirm those details.
> > >
> > > I commented on v1.
> > >
> > > Looks like abuse of "power-domains" to me, but nothing new really.
> > > Please define when to use a power domain vs. a perf domain and don't
> > > leave it up to the whims of the platform. Maybe perf domains was a
> > > mistake and they should be deprecated?
> > >
> >
> > Just a thought here, instead of deprecating it I was thinking if possible
> > to keep the power-domains and performance-domains separate and just extend
> > the genpd to handle the latter. There by we are not mixing up and creating
> > confusions that need more specific definitions in the binding(which is not
> > a big deal) but platforms getting it wrong inspite of that is a big problem.
> > Keep it separate makes it more aligned to the hardware and doesn't dilute
> > the definitions and probably avoids any possible mistakes due to that.
> >
> > Sorry Ulf I am just not yet convinced to mix them up yet 😉 and wish you
> > don't convince me to. Let me know why the above suggestion won't work.
>
> The main point I think we need to consider too, is that on some
> platforms, the power-domain and the performance-domain are managed
> together by the FW. It is not really two separate things and hence it
> would not quite be correct to describe it as two different types of
> providers in DT.
>
> If we should follow your suggestion above, to use the
> performance-domain bindings, then I think we need an additional new
> binding to cover the above mentioned case too. This would lead us into
> having one binding for the power-domain, another for the
> performance-domain and a third for the power+performance-domain.
>
> In my opinion this sounds quite like a mess. I would rather keep using
> the power-domain bindings for all these cases. Of course, it's a bit
> of a stretch too, but I think it should be less confusing in the end,
> assuming we extend/clarify the description of the power-domain
> bindings, of course.
>
> Did that convince you? :-)

Sudeep, Rob,

Can we try to conclude on the way forward?

Is it acceptable to keep using the power-domain bindings (with some
clarifications) for performance domains or should we start moving to
the performance-domain bindings?

If moving to the performance-domain binding, should we start migrating
existing users of the power-domain binding too - or what is your take
on this?

Kind regards
Uffe