Re: [PATCH v10 3/5] dt-bindings: clock: meson: add A1 PLL and Peripherals clkcs bindings

[Dmitry Rokosov]

On Tue, Mar 14, 2023 at 05:37:04PM +0100, Krzysztof Kozlowski wrote:
> On 14/03/2023 16:56, Dmitry Rokosov wrote:
> > On Tue, Mar 14, 2023 at 04:40:19PM +0100, neil.armstrong@xxxxxxxxxx wrote:
> >> On 14/03/2023 16:37, Krzysztof Kozlowski wrote:
> >>> On 14/03/2023 16:33, neil.armstrong@xxxxxxxxxx wrote:
> >>>>> There are many ways - depend on your driver. For example like this:
> >>>>> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/clk/samsung/clk-exynos5420.c#n975
> >>>>>
> >>>>> The first argument is the clock ID (or ignore).
> >>>>>
> >>>>> BTW, quite likely the problem is generic to all Meson clock drivers.
> >>>>
> >>>> This issue about "public" non-continuous defined was already discussed at https://lore.kernel.org/all/c088e01c-0714-82be-8347-6140daf56640@xxxxxxxxxx/
> >>>>
> >>>> I don't see what's different with this one.
> >>>
> >>> So you are aware that all undocumented clock IDs are still allowed to
> >>> use in DTS and they are ABI? Changing them will be an ABI break.
> >>
> >> Yes of course.
> >>
> >> Neil
> >>
> >>>
> >>> Best regards,
> >>> Krzysztof
> >>>
> >>
> > 
> > Sorry, guys, I'm little bit confused.
> > In the discussion pointed by Neil not-by-one-increment ID with public and
> > private parts are acked by Krzysztof due to explicit explanation in the
> > gxbb header. Have I to comment out my situation and stay it as is?
> 
> I did not NAK your solution here. I just pointed my usual remarks that
> it has certain outcome and minuses (undocumented ABI). But it is OK.
> 

Got it, thank you.

> > 
> > BTW, I think changing IDs value would not affect logic, because
> > it's not connected to driver logic 'by values', but 'by constants
> 
> You cannot change the IDs, neither their values nor the names (with
> exceptions). IDs - so the numbers - are ABI.
> 
> "Constant names" - I assume you mean the names of defines - do not exist
> after preprocessing, so also not really relevant here...
> 

Ah, you mean the situation when dtb blob is old and module or kernel
image is new, so ABI is broken. Yep, agree with you.

[...]

-- 
Thank you,
Dmitry