Re: [PATCH v9 05/12] clk: zte: Add Clock registration infrastructure

Stefan Dösinger <[email protected]> Mon, 03 Aug 2026 20:48:18 +0300

Newsgroups	gmane.linux.kernel.clk,gmane.linux.drivers.devicetree,gmane.linux.kernel,gmane.linux.ports.arm.kernel
Message-ID	<[email protected]>

Hi Brian,

Thanks again for the review!

Am Montag, 3. August 2026, 19:03:36 Ostafrikanische Zeit schrieb Brian Masney:
> > +	/*
> > +	 * Sanity check: Make sure all parents are there and write a clear
> > message rather than +	 * leave potential orphans.
> > +	 */
> 
> It's not common for clk drivers to have a validate function like this.
> That said, I know you are reverse engineering what's in the downstream
> BSP without documentation, and there is value in having this. I think
> it'd be useful to clarify that in the comments because code usually gets
> copied and pasted between drivers over time.
>
> That said, once it's verified that the clock tree is correct, can these
> checks be eventually dropped so that this validation logic isn't ran on
> every boot?

Yes, I plan to remove it eventually when support for the SoC is more mature. 
It has saved me some head-scratching as I have been working and re-working the 
clock tree. I'll clarify its driver specific role in the next submission.

Another option I contemplated is wrapping it in #ifdef DEBUG. What do you 
think about that?

> > +struct zx_clk_export {
> > +	unsigned int priv;
> > +	unsigned int binding;
> > +};
> 
> Sashiko flagged this as unused.

Indeed, that is a leftover from earlier - before I realized a simple unsigned 
int[] does the job fine.

---

I have a long-standing question about tristate/module support for drivers like 
this: I don't think the driver can realistically be unloaded. I have been 
testing driver unloading by removing the UART clocks from the DT (otherwise 
the clock driver is busy) and marking all clocks critical (otherwise 
unloading/unbinding will shut down the UART (and more) and lock me out of the 
system).

I have made it tristate because from early research into clock driver state of 
the art I gathered it was desired, even for drivers necessary for fundamental 
operation [0]. Did I understand this correctly? It also uncovered some linking 
errors that weren't obvious when compiling the driver into the kernel.

That said, I'd prefer to mark my init tables discardable with __initdata and 
think the module load support is going to go untested and bitrot over time.

Cheers,
Stefan

0: Chen-Yu Tsai's tak here mostly: https://www.youtube.com/watch?v=d1VIAnVb3hI

signature.asc (application/pgp-signature, 870 B)