[v3] dt/bindings: qoriq-clock: Add binding for FMan clock mux

[Scott Wood]

On Tue, 2015-05-05 at 16:02 -0500, Liberman Igal-B31950 wrote:
> 
> 
> Regards,
> Igal Liberman.
> 
> > -----Original Message-----
> > From: Wood Scott-B07421
> > Sent: Saturday, May 02, 2015 2:43 AM
> > To: Liberman Igal-B31950
> > Cc: devicetree at vger.kernel.org; linuxppc-dev at lists.ozlabs.org; Tang
> > Yuantian-B29983
> > Subject: Re: [v3] dt/bindings: qoriq-clock: Add binding for FMan clock mux
> > 
> > On Thu, 2015-04-30 at 09:28 -0500, Liberman Igal-B31950 wrote:
> > > > -----Original Message-----
> > > > From: Wood Scott-B07421
> > > > Sent: Thursday, April 30, 2015 3:31 AM
> > > > To: Liberman Igal-B31950
> > > > Cc: devicetree at vger.kernel.org; linuxppc-dev at lists.ozlabs.org; Tang
> > > > Yuantian-B29983
> > > > Subject: Re: [v3] dt/bindings: qoriq-clock: Add binding for FMan
> > > > clock mux
> > > >
> > > > On Wed, 2015-04-22 at 05:47 -0500, Liberman Igal-B31950 wrote:
> > > > > We have 2 cases:
> > > > > 	- Devices (T2/T4/B4) with CLKCG1HWACSR register.
> > > > > 	- Devices (Pxxxx, T1) without CLKCG1HWACSR register (Pxxxx
> > > > devices
> > > > > have many options, T1 has only one option)
> > > > >
> > > > > For the first group, we can have " qoriq-hwacsr" property in the clock
> > node.
> > > >
> > > > No, we're not going to describe every register with its own property.
> > > > Describe the chip and let the driver be the place with the knowledge
> > > > of what each chip is like.
> > > >
> > >
> > > I think that FMan clock mux (as we defined it) is similar to other muxes
> > available in our SoC.
> > 
> > I realize that.  I'm saying the way we described existing muxes seems to be a
> > mistake.  We're putting too much complexity in the device tree.
> > Better to put the complexity in a place that isn't stable ABI.
> > 
> > > If we take T4 as example, it has 3 muxes defined in the device tree (mux0,
> > mux1 and mux2).
> > > Each mux has its own reg property (and clock providers).
> > > 	mux2: mux2 at 40 {
> > > 		#clock-cells = <0>;
> > > 		reg = <0x40 0x4>;
> > > 		compatible = "fsl,qoriq-core-mux-2.0";
> > > 		clocks = <&pll3 0>, <&pll3 1>, <&pll3 2>,
> > > 			<&pll4 0>, <&pll4 1>, <&pll4 2>;
> > > 		clock-names = "pll3", "pll3-div2", "pll3-div4",
> > > 			"pll4", "pll4-div2", "pll4-div4";
> > > 		clock-output-names = "cmux2";
> > > 	};
> > >
> > > I agree that "fm1-clk-mux" need to be moved from the "guts" node to
> > "clockgen" node.
> > 
> > That's not what I was saying.  I'm saying get rid of the node entirely, in favor
> > of having the clockgen node itself be a clock source with multiple post-mux
> > outputs.
> > 
> 
> Scott,
> Currently the clockgen node has a number of sub nodes (platform clock, sysclock, PLLs and muxes, the number of muxes/PLLs is SoC dependent).
> The clockgen node has reg property, however, all other nodes have reg property too (The clockgen points to the clocking block address and the sub nodes point to the specific register in the clock block).
> Do you want to change this structure completely?

Yes, I want to change it completely (while ensuring the kernel still
works, albeit without new functionality, with older device trees).

> I'm not sure I understand the exact way you want see those nodes (in the final state).

Something like:

clockgen: global-utilities at e1000 {
	compatible = "fsl,<whatever>-clockgen";
	clock-frequency = <...>;
	reg = <0xe1000 0x1000>;
	#clock-cells = <2>;
};

cpu0: PowerPC,e6500 at 0 {
	device_type = "cpu";
	reg = <0 1>;
	clocks = <&clockgen 0 0>;
	next-level-cache = <&L2_1>;
	fsl,portid-mapping = <0x80000000>;
};


cpu4: PowerPC,e6500 at 0 {
	device_type = "cpu";
	reg = <4 5>;
	clocks = <&clockgen 0 1>;
	next-level-cache = <&L2_1>;
	fsl,portid-mapping = <0x80000000>;
};

fman at ... {
	...
	clocks = <&clockgen 1 0>;
	...
};

...with the clockgen binding explaining the format of the clock
specifier, and the clockgen driver containing the chip-specific muxing
details.

-Scott