On Sun, Jul 20, 2025 at 07:35:24PM +0200, Otto Pflüger wrote: > On Sun, Jul 20, 2025 at 05:38:02PM +0200, Krzysztof Kozlowski wrote: > > > > > > The clocks should be the same on sharkl3 (sc9863a) and ums9230, but > > > the existing bindings don't really make sense here or are incomplete. > > > AFAIK there is no SoC in which this display controller is directly > > > connected to the PLL as shown in the example. The DSI controller is > > > > This is not the PLL. Gate either. You are looking from wrong side - how > > clock is generated. > > > > You describe here CLOCK INPUT. > > > > > connected to a clock gate. The DPU actually does have two clocks, both > > > of which are clock muxes that allow selecting different frequencies and > > > one of which is behind a clock gate. I can add the second clock for the > > > DPU if needed. > > > > > > Since nothing seems to be using these bindings at the moment, would it > > > be okay to drop the old clock names that refer to specific frequencies? > > > > It is still completely irrelevant whether these are muxes. Dropping > > existing properties is ABI change, but anyway first figure out what is > > here really. > > I was trying to point out that the existing clock names are incorrect > because they refer to a specific source that is not necessarily used > for these clocks, instead of giving a name for the clock input.
OK, if the old name refers to the same clock input as in your new device, you can deprecate old case in the binding. > > For the DPU, would "core" and "dpi" be more appropriate as clock names? > DPI refers to the interface used internally between the DPU and the DSI > controller. Sounds fine. > > For the DSI controller, it seems that the clock is actually an APB bus > clock needed for accessing the control registers. Again, it is not > required to be connected to a 96MHz clock source as the name used in the > binding suggests. Would something like "apb_clk" or "pclk" be more > descriptive? Yeah, both are correct. I think pclk is preferred. Best regards, Krzysztof