Re: [PATCH v3 3/6] dt-bindings: clock: imx8mp: Add reset-controller sub-node

[Conor Dooley <conor@kernel.org>]

[-- Attachment #1: Type: text/plain, Size: 2419 bytes --]

On Fri, May 17, 2024 at 01:10:05PM -0400, Frank Li wrote:
> On Fri, May 17, 2024 at 05:21:32PM +0100, Conor Dooley wrote:
> > On Thu, May 16, 2024 at 11:56:27PM -0400, Frank Li wrote:
> > 
> > > Look like it is easy to register auxdev "reset" devices. But I have a
> > > problem. How to use it by DT phandle?  "reset" devices is service provider.
> > > Some client will use it.
> > > 
> > > Generally, reset node will used by other devices nodes. like
> > > 
> > > ABC: reset {
> > > 	compatible="simple-reset";
> > > 	...
> > > }
> > > 
> > > other node will use "reset = <&ABC 0>".  If use auxdev, how to get &ABC
> > > in dts file.
> > 
> > Whether or not you use auxdev or any other method etc, does not matter
> > in a DT system, the consumer will always have a phandle to the provider
> > node:
> > 
> > ABC: whatever {
> > 	compatible = "whatever";
> > 	#clock-cells = <...>;
> > 	#reset-cells = <...>;
> > }
> > 
> > something-else {
> > 	clocks = <&ABC ...>;
> > 	resets = <&ABC ...>;
> > }
> 
> 
> It goes back to old problem, "reset-cells" will be in "clock-controller".
> 
> clock-controller@30e20000 {
>         compatible = "fsl,imx8mp-audio-blk-ctrl", "syscon", "simple-mfd";
>         reg = <0x30e20000 0x10000>;
> 	...
> 	
> 	#reset-cells = <...>;
> 	^^^
>     };
> 
> If create new "whatever" auxdev bus driver which included two aux devices, 
> (clock and reset). 
> 
> it will be similar with mfd. Still need change
> clock-controller@30e20000 drivers.
> 
> "Which is I suspect is gonna require a change to your clock driver,
> because the range in the existing clock nodes:
> 	audio_blk_ctrl: clock-controller@30e20000 {
> 		compatible = "fsl,imx8mp-audio-blk-ctrl";
> 		reg = <0x30e20000 0x10000>;
> 	};
> would then have to move to the mfd parent node, and your clock child
> would have a reg property that overlaps the reset region. You'd need to
> then define a new binding that splits the range in two - obviously
> doable, but significantly more work and more disruptive than using an
> auxdev."
> 
> So I don't know why auxdev will be better than mfd.

I think Stephen and I have spent enough time trying to explain why using
auxdev is beneficial here. I, at least, won't be wasting any more of my
(metaphorical) breath.

> A possible benefit may be that Auxdev needn't binding doc for clock and
> reset node devices.


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]