Re: [PATCH v3 04/17] dt-bindings: soc: mobileye: add EyeQ5 OLB system controller

[Théo Lebrun <theo.lebrun@xxxxxxxxxxx>]

Hello,

On Wed Jan 24, 2024 at 8:22 PM CET, Rob Herring wrote:
> On Wed, Jan 24, 2024 at 11:40 AM Théo Lebrun <theo.lebrun@xxxxxxxxxxx> wrote:
> > On Wed Jan 24, 2024 at 6:28 PM CET, Théo Lebrun wrote:
> > > On Wed Jan 24, 2024 at 4:14 PM CET, Rob Herring wrote:
> > > > On Tue, Jan 23, 2024 at 07:46:49PM +0100, Théo Lebrun wrote:

[...]

> > > > > +      };
> > > > > +
> > > > > +      pinctrl-b {
> > > > > +        compatible = "mobileye,eyeq5-b-pinctrl";
> > > > > +        #pinctrl-cells = <1>;
> > > > > +      };
> > > > > +    };
> > > >
> > > > This can all be simplified to:
> > > >
> > > > system-controller@e00000 {
> > > >     compatible = "mobileye,eyeq5-olb", "syscon";
> > > >     reg = <0xe00000 0x400>;
> > > >     #reset-cells = <2>;
> > > >     #clock-cells = <1>;
> > > >     clocks = <&xtal>;
> > > >     clock-names = "ref";
> > > >
> > > >     pins { ... };
> > > > };
> > > >
> > > > There is no need for sub nodes unless you have reusable blocks or each
> > > > block has its own resources in DT.
> > >
> > > That is right, and it does simplify the devicetree as you have shown.
> > > However, the split nodes gives the following advantages:
> > >
> > >  - Devicetree-wise, it allows for one alias per function.
> > >    `clocks = <&clocks EQ5C_PLL_CPU>` is surely more intuitive
> > >    than `clocks = <&olb EQ5C_PLL_CPU>;`. Same for reset.
>
> clocks: resets: pinctrl: system-controller@e00000 {
>
> > >
> > >  - It means an MFD driver must be implemented, adding between 100 to 200
> > >    lines of boilerplate code to the kernel.
>
> From a binding perspective, not my problem... That's Linux details
> defining the binding. What about u-boot, BSD, future versions of Linux
> with different structure?
>
> I don't think an MFD is required here. A driver should be able to be
> both clock and reset provider. That's pretty common. pinctrl less so.

@Rob & @Krzysztof: following Krzysztof's question about the memory map
and adding ressources to the system-controller, I was wondering if the
following approach would be more suitable:

	olb: system-controller@e00000 {
		compatible = "mobileye,eyeq5-olb", "syscon", "simple-mfd";
		reg = <0 0xe00000 0x0 0x400>;
		#address-cells = <1>;
		#size-cells = <1>;

		clocks: clock-controller {
			compatible = "mobileye,eyeq5-clk";
			reg = <0x02c 0x7C>;
			#clock-cells = <1>;
			clocks = <&xtal>;
			clock-names = "ref";
		};

		reset: reset-controller {
			compatible = "mobileye,eyeq5-reset";
			reg = <0x004 0x08>, <0x120 0x04>, <0x200 0x34>;
			reg-names = "d0", "d2", "d1";
			#reset-cells = <2>;
		};

		pinctrl0: pinctrl-a {
			compatible = "mobileye,eyeq5-a-pinctrl";
			reg = <0x0B0 0x30>;
		};

		pinctrl1: pinctrl-b {
			compatible = "mobileye,eyeq5-b-pinctrl";
			reg = <0x0B0 0x30>;
		};
	};

It highlights that they are in fact separate controllers and not one
device. The common thing between them is that they were
custom-implemented by Mobileye and therefore all registers were put in
a single block.

Else we'll go with the driver that implements both the clock & reset
providers. It'd live in drivers/clk/ I believe, as this is where other
drivers of the sort live.

> > >  - It means one pinctrl device for the two banks. That addresses your
> > >    comment on [PATCH v3 10/17]. This is often done and would be doable
> > >    on this platform. However it means added logic to each individual
> > >    function of pinctrl-eyeq5.
>
> If it makes things easier, 2 'pins' sub-nodes is fine. That's just
> container nodes.
>
> > >    Overall it makes for less readable code, for code that already looks
> > >    more complex than it really is.
> > >
> > >    My initial non-public version of pinctrl-eyeq5 was using this method
> > >    (a device handling both banks) and I've leaned away from it.
> >
> > I had forgotten one other reason:
> >
> >  - Reusability does count for something. Other Mobileye platforms exist,
> >    and the system controller stuff is more complex on those. Multiple
> >    different OLB blocks, etc. But my understanding is that
> >    per-peripheral logic is reused across versions.
>
> IME, this stuff never stays exactly the same from chip to chip.

If it helps, I have access to the downstream vendor kernel to see how
things work there. It supports the next generation of Mobileye
hardware.

Regards,

--
Théo Lebrun, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com