Re: [PATCH 2/3] dt-bindings: arm: qcom: Add MSM8926 and Samsung Galaxy Tab 4 10.1 LTE

[Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>]

On 28/01/2023 22:43, Luca Weiss wrote:
> On Montag, 23. Jänner 2023 19:08:03 CET Krzysztof Kozlowski wrote:
>> On 23/01/2023 18:41, Stefan Hansson wrote:
>>>>> 2. You base on other SoC but you do not include its compatibles. Why? Is
>>>>> it intended? None of the properties applicable to other SoC will match
>>>>> here, thus I actually wonder if you run dtbs_check...
>>>
>>> Sorry, I forgot about running dtbs_check. However, I'm not sure I
>>> understand the question. What do you mean by that I don't include its
>>> compatibles?
>>
>> I understood you include the msm8226.dtsi which is a different SoC. If
>> you include it, you get all of its content. We do it only for compatible
>> devices, but your device does not indicate compatibility with msm8226.
> 
> Hi Krzysztof,
> 
> the way the earlier Qualcomm SoCs work, especially regarding naming scheme is 
> the following.
> 
> There's for example the msm8x74 family which includes msm8974, msm8674, 
> msm8274, and the a bit differently named apq8074 where the significant 
> different are the RF capabilities, I think with those only 8974 had LTE, 8674 
> and 8274 only 3G but different band support, and the apq8074 has no mobile 
> radio.
> 
> The same exists for sure also for 8x16 and 8x26, probably a bunch of other 
> SoCs as well.
> 
> So from software side (apart from modem firmware of course) it can be treated 
> in practise as the same SoC so that's why we included the dtsi in this case in 
> msm8226 but also msm8926 and apq8026.

First, there is distinction between SoC having and not having modem. I
guess small enough that we just include the DTSI.

Second, there is distinction between different families, even if they
share a lot. All of the SoCs here share something, because Qualcomm has
versionable IP blocks which they re-use.

> 
> But the compatible on board-level is in practise (to my knowledge) not really 
> used for anything important other than having a nice string in the dts file. I 
> know some software uses compatible from user space but there for 
> differentiating between different devices and ignoring the SoC compatibles.

It's not only about board, but about all devices in the SoC.

> 
> But while they are software-compatible for the most part, they *are* distinct 
> SoCs with different capabilities and I just don't see the point in trying to 
> establish some kinds of relationships between different SoCs that are somewhat 
> or very similar (msm8226 and msm8974 also share many components but are 
> obviously different SoCs).

You don't have to create such relationships. You don't have to include
other DTSI, either. What yo have to is - quoting Linux docs:
"DO make 'compatible' properties specific. DON'T use wildcards in
compatible strings. DO use fallback compatibles when devices are the
same as or a subset of prior implementations. DO add new compatibles in
case there are new features or bugs."

> 
> And also e.g. (nearly) all apq* dts files we already have in mainline only 
> have apq compatible and not the corresponding msm* compatible. And I think 
> that's totally legitimate.

We do not talk here about apq, actually, at all. We talk about one
msm8xxx including other msm8xxx...


Best regards,
Krzysztof