Re: [PATCH 1/3] dt-bindings: pwm: kona: Add new compatible for new version pwm-kona

[Scott Branden]

Hi Rob,

On 2019-01-21 3:11 p.m., Rob Herring wrote:
> On Tue, Jan 15, 2019 at 04:14:21PM -0800, Scott Branden wrote:
> > Hi Uwe,
> >
> > On 2019-01-12 7:05 a.m., Uwe Kleine-KÃnig wrote:
> > > Hello Scott,
> > >
> > > On Fri, Jan 11, 2019 at 01:28:45PM -0800, Scott Branden wrote:
> > > > On 2019-01-11 12:48 p.m., Uwe Kleine-KÃnig wrote:
> > > > > On Fri, Jan 11, 2019 at 10:51:14AM +0530, Sheetal Tigadoli wrote:
> > > > > > From: Praveen Kumar B <praveen.b@xxxxxxxxxxxx>
> > > > > >
> > > > > > Add new compatible string for new version of pwm-kona
> > > > > >
> > > > > > Signed-off-by: Praveen Kumar B <praveen.b@xxxxxxxxxxxx>
> > > > > > Reviewed-by: Ray Jui <ray.jui@xxxxxxxxxxxx>
> > > > > > Reviewed-by: Scott Branden <scott.branden@xxxxxxxxxxxx>
> > > > > > Signed-off-by: Sheetal Tigadoli <sheetal.tigadoli@xxxxxxxxxxxx>
> > > > > > ---
> > > > > >     Documentation/devicetree/bindings/pwm/brcm,kona-pwm.txt | 2 +-
> > > > > >     1 file changed, 1 insertion(+), 1 deletion(-)
> > > > > >
> > > > > > diff --git a/Documentation/devicetree/bindings/pwm/brcm,kona-pwm.txt b/Documentation/devicetree/bindings/pwm/brcm,kona-pwm.txt
> > > > > > index 8eae9fe..d37f697 100644
> > > > > > --- a/Documentation/devicetree/bindings/pwm/brcm,kona-pwm.txt
> > > > > > +++ b/Documentation/devicetree/bindings/pwm/brcm,kona-pwm.txt
> > > > > > @@ -3,7 +3,7 @@ Broadcom Kona PWM controller device tree bindings
> > > > > >     This controller has 6 channels.
> > > > > >     Required Properties :
> > > > > > -- compatible: should contain "brcm,kona-pwm"
> > > > > > +- compatible: should contain "brcm,kona-pwm" or "brcm,kona-pwm-v2"
> > > > > Is v2 used on a newer generation of kona SoCs? On i.MX these variants
> > > > > are usually named after the first SoC that came with the new variant. Is
> > > > > this sensible here, too?
> > > > It doesn't make as much sense here as different revs of the IP block are
> > > > picked up based on various decisions.
> > > >
> > > > A new SoC could decide to use an old version.
> > > IMHO this is no reason to not use the name of the oldest SoC with this
> > > variant. I don't know how the SoC names are in the broadcom family, but
> > > if they were (in order of age, oldest first):
> > >
> > > 	ant
> > > 	bear
> > > 	crocodile
> > >
> > > and ant and crocodile use the same IP block we would have
> > >
> > > a) with v1, v2:
> > >
> > > 	ant:
> > > 		compatible = "brcm,kona-pwm-v1";
> > > 	bear:
> > > 		compatible = "brcm,kona-pwm-v2";
> > > 	crocodile:
> > > 		compatible = "brcm,kona-pwm-v1";
> Version numbers can be fine, but generally only as fallbacks as even the
> same IP version can be integrated into an SoC differently.
>
> The other issue with versions is they should be meanful such as
> corresponding to version tags in IP repos. Often, I'd guess anything
> with a 'v1' is just what some s/w person made up. Of course, we only
> can really know that for opensource IP or programmable logic IP.
>
> If you do use versions, document what the versioning scheme is.
>
> > > ; and
> > >
> > > b) with the SoC naming:
> > >
> > > 	ant:
> > > 		compatible = "brcm,kona-ant-pwm";
> > > 	bear:
> > > 		compatible = "brcm,kona-bear-pwm";
> > > 	crocodile:
> > > 		compatible = "brcm,kona-crocodile-pwm", "brcm,kona-ant-pwm";
> This is the recommended practice.
>
> > > (If you want, drop "brcm,kona-crocodile-pwm", but keeping it is more
> > > defensive.)
> Generally, you should have "brcm,kona-crocodile-pwm" in case there's
> some difference found later. Then you can support the bug or feature
> without a DT change.

No DT change would be necessary in any case.

A check against the SOC type in the driver without additional DT 
compatibility strings could be done.

> > > I like b) (with "...-crocodile-...") better than a). crocodile using
> > > "...-ant-..." is not more ugly than crocodile using "...-v1". This is
> > > also a tad more robust because if broadcom releases kona-dolphin and
> > > someone finds a minor difference between the IPs used on ant and
> > > crocodile it depends on the order of these events who gets v3, while
> > > with the SoC naming the result is clear.
> > >
> > > (OK, and given that "brcm,kona-pwm" is already fixed, both approaches
> > > need slight adaption, but I guess you still get what I meant.)
> > Thanks for your thoughts and explanation.
> >
> > It is unfortunate devicetree has no proper guidelines or documentation on
> >
> > binding naming. In the interest of getting this upstream we can name it
> Surely we've captured that somewhere...

Please point me at such documentation.

There is no consistency in kernel drivers from what I have seen.

> > "brcm, omega-pwm". We can drop kona from the binding name as that
> > architecture
> >
> > is really no more - only IP derived from it is - hence the name kona-v2
> > previously.
> >
> > > Best regards
> > > Uwe
> > Cheers,
> > Scott