Re: [PATCH 07/15] dt-bindings: clock: Add Ambarella clock bindings

[Krzysztof Kozlowski]

On 06/02/2023 15:57, Li Chen wrote:
> Hi Krzysztof,
>  ---- On Mon, 06 Feb 2023 21:41:44 +0800  Krzysztof Kozlowski  wrote --- 
>  > On 06/02/2023 12:28, Li Chen wrote:
>  > > Hi Krzysztof ,
>  > > 
>  > >  ---- On Fri, 27 Jan 2023 23:08:09 +0800  Krzysztof Kozlowski  wrote --- 
>  > >  > On 27/01/2023 15:48, Li Chen wrote:
>  > >  > >  > 
>  > >  > >  > but what you are saying is that there is no separate clock controller
>  > >  > >  > device with its own IO address but these clocks are part of rct_syscon.
>  > >  > >  > Then model it that way in DTS. The rct_syscon is then your clock
>  > >  > >  > controller and all these fake gclk-core and gclk-ddr nodes should be gone.
>  > >  > > 
>  > >  > > Ok, I will remove these fake nodes, and model the hardware as:
>  > >  > > 
>  > >  > > rct_syscon node
>  > >  > > | clock node(pll, div, mux, composite  clocks live in the same driver)
>  > >  > > | other periphal nodes
>  > >  > 
>  > >  > You need clock node if it takes any resources. If it doesn't, you do not
>  > >  > need it.
>  > > 
>  > > If the only hardware resource the clock node can take is its parent clock(clocks = &osc;),
>  > > then can I have this clock node?
>  > 
>  > I am not sure if I understand. osc does not look like parent device, so
>  > this part of comment confuses me.
> 
> Sorry for the confusion. I mean osc is the root of clock tree:
> 
> osc
>   | pll A
>   | pll B
>   | ...
> 
> So if I have a clock node under rct_syscon node, I think it should take osc as the parent(node) clock:
> rct_syscon {
>     ......
>     clock_controller {
>           clocks = <&osc>;
>           ......
> 
> You have said "You need clock node if it takes any resources. ", do you think osc here can be counted as a used resource?

Yes, in that matter osc should be the input to this clock controller.

Best regards,
Krzysztof