Re: [PATCH 1/6] dt-bindings: clock: qcom,dispcc-sm8x50: add mmcx power domain

[Ulf Hansson]

On Thu, 1 Jul 2021 at 21:26, Bjorn Andersson <bjorn.andersson@xxxxxxxxxx> wrote:
>
> On Thu 01 Jul 11:58 CDT 2021, Ulf Hansson wrote:
>
> > On Thu, 1 Jul 2021 at 18:39, Dmitry Baryshkov
> > <dmitry.baryshkov@xxxxxxxxxx> wrote:
> > >
> > > On Thu, 1 Jul 2021 at 19:17, Ulf Hansson <ulf.hansson@xxxxxxxxxx> wrote:
> > > >
> > > > On Wed, 30 Jun 2021 at 15:31, Dmitry Baryshkov
> > > > <dmitry.baryshkov@xxxxxxxxxx> wrote:
> > > > >
> > > > > On sm8250 dispcc requires MMCX power domain to be powered up before
> > > > > clock controller's registers become available. For now sm8250 was using
> > > > > external regulator driven by the power domain to describe this
> > > > > relationship. Switch into specifying power-domain and required opp-state
> > > > > directly.
> > > > >
> > > > > Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@xxxxxxxxxx>
> > > > > ---
> > > > >  .../bindings/clock/qcom,dispcc-sm8x50.yaml    | 19 +++++++++++++++++++
> > > > >  1 file changed, 19 insertions(+)
> > > > >
> > > > > diff --git a/Documentation/devicetree/bindings/clock/qcom,dispcc-sm8x50.yaml b/Documentation/devicetree/bindings/clock/qcom,dispcc-sm8x50.yaml
> > > > > index 0cdf53f41f84..48d86fb34fa7 100644
> > > > > --- a/Documentation/devicetree/bindings/clock/qcom,dispcc-sm8x50.yaml
> > > > > +++ b/Documentation/devicetree/bindings/clock/qcom,dispcc-sm8x50.yaml
> > > > > @@ -55,6 +55,16 @@ properties:
> > > > >    reg:
> > > > >      maxItems: 1
> > > > >
> > > > > +  power-domains:
> > > > > +    description:
> > > > > +      A phandle and PM domain specifier for the MMCX power domain.
> > > > > +    maxItems: 1
> > > > > +
> > > >
> > > > Should you perhaps state that this is a parent domain? Or it isn't?
> > > >
> > > > Related to this and because this is a power domain provider, you
> > > > should probably reference the common power-domain bindings somewhere
> > > > here. Along the lines of this:
> > > >
> > > > - $ref: power-domain.yaml#
> > > >
> > > > As an example, you could have a look at
> > > > Documentation/devicetree/bindings/power/pd-samsung.yaml.
> > >
> > > I'll take a look.
> > >
> > > >
> > > > > +  required-opps:
> > > > > +    description:
> > > > > +      Performance state to use for MMCX to enable register access.
> > > > > +    maxItems: 1
> > > >
> > > > According to the previous discussions, I was under the assumption that
> > > > this property belongs to a consumer node rather than in the provider
> > > > node, no?
> > >
> > > It is both a consumer and a provider. It consumes SM8250_MMCX from
> > > rpmhpd and provides MMSC_GDSC.
> >
> > That sounds a bit weird to me.
> >
>
> dispcc is a hardware block powered by MMCX, so it is a consumer of it
> and needs to control MMCX.

Right, that sounds reasonable.

>
> > In my view and per the common power domain bindings (as pointed to
> > above): If a power domain provider is a consumer of another power
> > domain, that per definition means that there is a parent domain
> > specified.
> >
>
> And in addition to needing MMCX to access the dispcc, the exposed
> power-domain "MDSS_GDSC" is powered by the same MMCX and as such
> MDSS_GDSC should be a subdomain of MMCX.

What do you mean by "exposed"? It sounds like you are saying that
"MDSS_GDSC" is an artificial power domain, no?

If that's the case, more exactly, why is it like this?

My apologies if I bother you with details, but as a maintainer of
genpd, it is very useful to me to have the complete picture.

>
>
> But what I was trying to say yesterday is that the power-domain property
> should be sufficient and that we shouldn't need to drive MMCX to a
> particular performance_state in order to access the registers.
>
> Then as clients make votes on clock rates that requires higher
> performance_state, they would describe this in their opp-tables etc.
>
>
> But without any performance_state requests, pd->corner will in
> rpmhpd_power_on() be 0 and as such powering on the power-domain won't
> actually do anything. Similarly dev_pm_genpd_set_performance_state(dev,
> 0) on an active power-domain from rpmhpd will turn it off.

Yes, I noticed the patches you posted. Thanks for helping out here!

>
>
> So the reason why Dmitry is adding the required-opps to the binding is
> to get rpmhpd to actually tell the hardware to turn on the power domain.
> And I don't think this is in accordance with the framework's
> expectations.

I agree!

>
> Regards,
> Bjorn

Kind regards
Uffe