* Krzysztof Kozlowski <krzysztof.kozlowski@xxxxxxxxxx> [231128 08:41]: > On 28/11/2023 09:32, Andreas Kemnade wrote: > > On Tue, 28 Nov 2023 09:00:16 +0100 > > Krzysztof Kozlowski <krzysztof.kozlowski@xxxxxxxxxx> wrote: > > > >>> +required: > >>> + - compatible > >>> + - clocks > >>> + - '#clock-cells' > >> > >> reg is required. Device cannot take "reg" from parent, DTS does not work > >> like this. > > > > Well, apparently they do... and I am just dealing with status quo and not > > how it should be. > > Look at commit 31fc1c63c2ae4a542e3c7ac572a10a59ece45c24 > > Who designed clock-controller binding with a device node per each clock? > This is ridiculous (although of course not your fault here)! Looking at > omap3xxx-clocks.dtsi - all its children should be just defined by the > driver, not by DTSI. Earlier all the clocks were separate nodes, the ti,clksel binding made things a bit better by grouping the seprate clock nodes so we don't have multiple nodes with the same reg.. But yeah clksel instance clocks should be clock@6 with reg = <6> if the clock bits are at bit 6. That would be fairly easy to do if that helps, but in general I doubt anybody's going to spend much effort to fix the omap3 legacy clocks atthis point. For omap4 and later, things are a bit better as they use the clkctrl clocks: Documentation/devicetree/bindings/clock/ti-clkctrl.txt I don't think omap3 has any clkctrl clocks but if it does then that could be used. Regards, Tony
