On Mon, Oct 05, 2026 at 08:53:05PM +0700, Michael Zavertkin wrote:
> On Mon, Oct 05, 2026 at 11:38:51AM +0100, Conor Dooley wrote:
> > On Sun, Oct 04, 2026 at 10:32:35PM +0700, Michael Zavertkin wrote:
> > > For RTL9607C driver used to configure IO_MODE register. This register is
> > > global and is a scope of pinctrl driver (either pinctrl-single or WIP
> > > driver for SoC).
> > > 
> > > Drop realtek,scl requirement for that SoC and require pinctrl instead.
> > 
> > This patch itself is probably fine, but the driver changes will cause
> > regressions because all existing devicetrees are going to have
> > realtek,scl and not the pinctrl properties. You need to retain the
> > driver code and make the register setting optional depending on if the
> > scl is there (or similar).
> > pw-bot: changes-requested
> > 
> > Thanks,
> > Conor.
> 
> I understand backward compatibilyti requirement. However all known users
> of this driver are here (me and Rustam Adilov). Does compatibility
> strictly required for such case?

How do you know there's only the two of you?

> If so, what would be better - print warning about realtek,scl (only for
> RTL9607C) and work as earlier, potentially interfering with pinctrl
> driver - or abort probe with error if there is no pinctrl property?
> Explicit error or message is better then silent regression, if I
> understand you correctly.

If there's genuinely only two users, then I suppose you can break it.
But two known users doesn't meant two users. Otherwise, use pintrl if
available and use the fallback to the existing property if not. Mark the
property deprecated and probably emit a warning.