Re: [PATCH v5 2/3] arm64: dts: ti: k3-am62l: add initial infrastructure

[Bryan Brattlof <bb@ti.com>]

On May 16, 2025 thus sayeth Sascha Hauer:
> Hi All,
> 
> On Wed, May 07, 2025 at 10:09:20PM -0500, Bryan Brattlof wrote:
> > +	rti0: watchdog@e000000 {
> > +		compatible = "ti,j7-rti-wdt";
> > +		reg = <0x00 0x0e000000 0x00 0x100>;
> > +		clocks = <&scmi_clk 273>;
> > +		power-domains = <&scmi_pds 60>;
> > +		assigned-clocks = <&scmi_clk 273>;
> > +		assigned-clock-parents = <&scmi_clk 1>;
> > +	};
> > +
> > +	rti1: watchdog@e010000 {
> > +		compatible = "ti,j7-rti-wdt";
> > +		reg = <0x00 0x0e010000 0x00 0x100>;
> > +		clocks = <&scmi_clk 279>;
> > +		power-domains = <&scmi_pds 61>;
> > +		assigned-clocks = <&scmi_clk 279>;
> > +		assigned-clock-parents = <&scmi_clk 1>;
> > +	};
> 
> In the TI downstream TF-A we have this in the readme:
> 
> | AM62L Clock SCMI ID List:
> | =========================
> | 
> | **Note:** For using the clock parents, the scmi clock ID will not be the actual number itself,
> | rather will be starting with 0 for respective parents similar to how it's being done in TI SCI
> | Documentation: https://software-dl.ti.com/tisci/esd/09_02_07/5_soc_doc/am62ax/clocks.html
> | 
> | For eg. for AM62LX_DEV_MCASP0_AUX_CLK device, clock ID will be 192, however if we require the parent
> | to be AM62LX_DEV_MCASP0_AUX_CLK_PARENT_HSDIV4_16FFT_WKUP_0_HSDIVOUT1_CLK then we would represent in
> | the device tree as:
> | 
> | ```
> | assigned-clocks = <&scmi_clk 192>;
> | assigned-clock-parents = <&scmi_clk 1>;
> 
> This doesn't work. It's not conforming to the SCMI specification and it
> also doesn't work with the Linux SCMI clk driver which will hang on a
> cat /sys/kernel/debug/clk/clk_summary. This is because all clock
> parents will be the SCMI clocks with the lowest ids including the clocks
> with the lowest ids which will expose themselves as parents.
> 
> It also prevents the userspace from being able to make a tree structure
> from the clock tree.
> 
> The corresponding TF-A code is requested for upstreaming here [1] where
> I also commented on with the same issue.
> 
> Please stop spreading this further. The parents must be the actual
> clk_id, not numbers from 0..n.
> 
> Sascha
> 
> [1] https://review.trustedfirmware.org/c/TF-A/trusted-firmware-a/+/34834
>

Ah my bad. I'll drop all nodes that need clock parents called out until 
this is resolved.

Thanks Sascha
~Bryan