Re: [RFC v2 3/3] ARM: dts: omap3: bulk convert compatible to be explicitly ti,omap3430 or ti,omap36xx

[H. Nikolaus Schaller]

Hi Tony,

> Am 06.09.2019 um 17:47 schrieb Tony Lindgren <tony@xxxxxxxxxxx>:
> 
> * H. Nikolaus Schaller <hns@xxxxxxxxxxxxx> [190906 07:53]:
>>> Am 05.09.2019 um 16:27 schrieb Tony Lindgren <tony@xxxxxxxxxxx>:
>>> compatible = "ti,omap3-ldp", "ti,omap3430", "ti,omap34xx", "ti,omap3";
>> 
>> After thinking a little about the whole topic the main rule of this change must be:
>> 
>> * do not break any existing in-tree DTS
>> 	=> only *add* to compatible what we need to distinguish between omap34 and omap36
>> 
>> * additions shall only follow new scheme
>> 	=> we only add "ti,omap34xx" or "ti,omap36xx"
>>           but neither "ti,omap3630" nor "ti,omap3430"
> 
> Sorry I don't follow you on this one.. We should always add "ti,omap3630"
> where "ti,omap36xx" is currently used so we can eventually get rid of
> "ti,omap36xx". And the same for 34xx.

Ah, ok now I see.

You want to make the "ti,omap3630" the official one and "ti,omap36xx" legacy.
It is probably an arbitrary choice if we want to get rid of the xx or the 30.

I had thought to do it the other way round because I had done this statistics:

for i in 3430 34xx 3630 36xx; do echo $i $(fgrep '"'ti,omap$i'"' arch/arm/boot/dts/*.dts* | wc -l); done

3430 12
34xx 28
3630 3
36xx 23

which would indicate that 34xx and 36xx are more common.

>> * cover some out-of-tree DTS
>> 	=> make the ti-cpufreq driver still match "ti,omap3430" or "ti,omap3630"
>> 	   even if this duplicates compatibility
>> 
>> This would mean that the logicpd-som-lv-37xx-devkit.dts gets the additional "ti,omap36xx"
>> while the omap3-ldp.dts would only get an "ti,omap34xx" but no "ti,omap3430" (since we
>> do not use it anywhere).
>> 
>> Could you agree on this approach?
> 
> Yeah sounds like logicpd-som-lv-37xx-devkit.dts currently still needs
> "ti,omap36xx" for now.
> 
> If modifying omap3-ldp.dts, also add "ti,omap3430" in additon to
> "ti,omap34xx" that it already has.
> 
> So basically let's assume the following:
> 
> "ti,omap3430" == "ti,omap34xx"
> "ti,omap3630" == "ti,omap36xx"
> 
> This means code needs to parse both.

Yes, it already does everywhere.

BTW there is also some code that does special SoC detection based on
soc_device_match(), mainly in omapdrm/dss.

If we were to use this mechanism in the ti-cpufreq driver we could
match it to ti,omap3 and could avoid all these changes.

But make it less maintainable and code more complex.

> 
> And eventually we just drop the "xx" variants.

> 
> So while patching compatibles, let's also update for this to
> avoid multiple patches churning the same compatibles over and
> over.

Ok. I'll rework the patch so that we never add "ti,34xx" or "ti,36xx"
but add "ti,3430" or "ti,3630" if missing.

I'll also take a look at omap.txt bindings since that likely needs
an update as well to better describe what the official ones are
and which are legacy.

BR and thanks,
Nikolaus