On Tue, 1 Jun 2021 at 13:44, Viresh Kumar [off-list ref] wrote:

On 01-06-21, 13:12, Stephan Gerhold wrote:

quoted

quoted

+    child4: consumer@12341000 {
+        compatible = "foo,consumer";
+        reg = <0x12341000 0x1000>;
+        power-domains = <&parent4>, <&parent5>;
+        assigned-performance-states = <0>, <256>;
+    };

Bjorn already asked this in v1 [1]:

quoted

May I ask how this is different from saying something like:

    required-opps = <&??>, <&rpmpd_opp_svs>;

and maybe this was already discussed further elsewhere. But I think at
the very least we need some clarification in the commit message + the
binding documentation how your new property relates to the existing
"required-opps" binding.

Because even if it might not be implemented at the moment,
Documentation/devicetree/bindings/power/power_domain.txt actually also
specifies "required-opps" for device nodes e.g. with the following example:

      leaky-device0@12350000 {
              compatible = "foo,i-leak-current";
              reg = <0x12350000 0x1000>;
              power-domains = <&power 0>;
              required-opps = <&domain0_opp_0>;
      };

It looks like Viresh added that in commit e856f078bcf1
("OPP: Introduce "required-opp" property").

And in general I think it's a bit inconsistent that we usually refer to
performance states with phandles into the OPP table, but the
assigned-performance-states suddenly use "raw numbers".

I must have missed that discussion, sorry about that.

The required-opps property, when present in device's node directly, is about the
(default) OPPs to choose for that device's normal functioning as they may not do
DVFS.

Alright, so it looks like we already have the DT binding that we need for this.

That leaves us with the question, at what place should we parse it
(call of_get_required_opp_performance_state()) and set the performance
state for the device?

Does it still make sense to do it while attaching the device to the
genpd, you think? Or, is there another better common path?

For fixed regulators, the binding is parsed during the probe and then
the performance state is set/dropped at regulator_enable/disable().

Good point Stephan.

Kind regards
Uffe