On Thu, Aug 10, 2023 at 09:51:05AM +0200, Jerome Brunet wrote: > > On Thu 10 Aug 2023 at 09:46, Krzysztof Kozlowski <krzysztof.kozlowski@xxxxxxxxxx> wrote: > > > On 10/08/2023 09:32, Jerome Brunet wrote: > >>>>> Then why do you have this huge, apparently unnecessary, oneOf? If it's > >>>>> the same, then drop the oneOf and make number of clocks fixed. > >>>> > >>>> But as far as I understand the number of clocks is not fixed. As Jerome pointed > >>>> out in the other post, it can have any combination of clocks and range from 1 > >>>> up to 11, where 'pclk' is always 1st clock. > >>>> I currently have no idea how to constraint that, despite limiting the number > >>>> of clock-names. > >>> > >>> The same as in all other clock controllers (was also present on my list > >>> of useful patterns - Variable length arrays (per variant)): > >>> https://elixir.bootlin.com/linux/v5.19-rc6/source/Documentation/devicetree/bindings/clock/samsung,exynos7-clock.yaml#L57 > >> > >> In the example provided, the number and list of clocks required by each > >> controller variant is fixed, if I'm reading it correctly > >> > >> Here the controller (regardless of the variant) accepts a maximum 29 > >> clock inputs. Only pclk is required. It is valid to have any of 28 > >> optional clocks at index 2, 3, etc ... > > > > I actually doubt that it is optional. These are valid clock inputs. I > > could imagine they are optional depending on the use-case, like some > > block being turned off or on... but then still the clock is there, just > > not actively used. > > > > Aren't you now describing existing Linux driver? > > They are valid inputs but not required. It is valid (and expected) to > have a fair share of them not connected. The slave clocks just don't exist > most of the time, and the IP can totally accept unwired master clock > inputs on SoC variant. Nothing is going to break down if some are > missing. > > >From the controller perspective, the description given here is correct > and the inputs are optional. > > The more generic question is how do we deal with multiple, > independent and optional ressources ? Because then, the order in which > they appear cannot be predicted. phandle values of 0 or -1 are treated as "NULL" entries. Rob
