Re: [PATCH 3/5] arm64: dts: imx8mp: Enable HS400-ES

[Krzysztof Kozlowski <krzk@kernel.org>]

On 28/03/2022 15:07, Ahmad Fatoum wrote:
> Hello Krzysztof,
> 
> On 28.03.22 14:56, Krzysztof Kozlowski wrote:
>> On 28/03/2022 14:45, Adam Ford wrote:
>>> On Mon, Mar 28, 2022 at 6:49 AM Krzysztof Kozlowski <krzk@kernel.org> wrote:
>>>>
>>>> On 28/03/2022 13:09, Ahmad Fatoum wrote:
>>>>> Hello Adam,
>>>>>
>>>>> On 28.03.22 12:47, Adam Ford wrote:
>>>>>> On Mon, Mar 28, 2022 at 2:20 AM Ahmad Fatoum <a.fatoum@pengutronix.de> wrote:
>>>>>>>
>>>>>>> Hello Adam,
>>>>>>>
>>>>>>> On 27.03.22 14:38, Adam Ford wrote:
>>>>>>>> The SDHC controller in the imx8mp has the same controller
>>>>>>>> as the imx8mm which supports HS400-ES. Change the compatible
>>>>>>>> fallback to imx8mm to enable it.
>>>>>>>
>>>>>>> I believe that's a shortcoming of the Linux driver, which should explicitly list
>>>>>>> fsl,imx8mp-usdhc in its compatibles and enable HS400-ES for it.
>>>>>>>
>>>>>>> I find dropping compatibles problematic, because like Linux matching
>>>>>>> fsl,imx8mm-usdhc, but not fsl,imx8mp-usdhc, other software may match
>>>>>>> fsl,imx7d-usdhc, but not fsl,imx8[mp]-usdhc.
>>>>>>>
>>>>>>> I'd prefer that either the kernel driver gains extra compatibles or that
>>>>>>> the DTS lists extra compatibles and we refrain from dropping existing
>>>>>>> (correct) ones.
>>>>>>>
>>>>>>
>>>>>> I would argue that imx7d is not correct since the IP blocks between
>>>>>> imx7d and imx8mm have different flags/quirks.  One of which includes
>>>>>> HS400-ES, but there are other differences as well.
>>>>>
>>>>> The DTS currently says that an fsl,imx7d-usdhc is a subset of an
>>>>> fsl,imx8mm-usdhc. So a driver could treat both HW the exact same
>>>>> by focusing on the i.MX7D parts. Linux apparently did exactly
>>>>> that so far. Is this not accurate?
>>>>>
>>>>>
>>>>>>> What do you think?
>>>>>>
>>>>>> From my understanding of the fallback compatibility strings is to
>>>>>> avoid having to add more and more compatible strings to the drivers
>>>>>> when they do not serve a functional purpose. Based On a conversation
>>>>>> with Krzysztof [1], he suggested we update the YAML file based on the
>>>>>> fallback, but he wanted NXP to give their feedback as to what the
>>>>>> right fallback strings should be.  Haibo from NXP sent me a hierarchy
>>>>>> [1] which is what I used to update the YAML file.  Based on the YAML
>>>>>> file, the fallback in each DTSI file was updated to ensure the use of
>>>>>> the proper IP block.
>>>>>
>>>>> Myself I am in favor of moving to three compatibles instead of dropping one.
>>>>> For some theoretical fsl,imx8mf-usdhc that's supposed to be exactly the same
>>>>> as a fsl,imx8mm-usdhc, I don't mind omitting the fsl,imx7d-usdhc compatible,
>>>>> but for existing device trees, this may introduce needless potential breakage
>>>>> for other software that also uses Linux device trees.
>>>>>
>>>>
>>>> Affecting existing users is indeed a concern with this approach, because
>>>> in-kernel DTS might be used in other projects as well.
>>>>
>>>> I still cannot find here the answer whether fsl,imx8mm-usdhc is actually
>>>> compatible with fsl,imx7d-usdhc. It's not about driver, but about
>>>> hardware and programming model. imx8mm can support additional features
>>>> and still be compatible with imx7d. However if any flags of imx7d are
>>>> actually not valid for imx8mm, then it's different case.
>>>
>>> The imx7d flags are:
>>>  ESDHC_FLAG_USDHC
>>> ESDHC_FLAG_STD_TUNING
>>>  ESDHC_FLAG_HAVE_CAP1
>>> ESDHC_FLAG_HS200
>>>  ESDHC_FLAG_HS400
>>>  ESDHC_FLAG_STATE_LOST_IN_LPMODE
>>>  ESDHC_FLAG_BROKEN_AUTO_CMD23,
>>>
>>> The imx8mm flags are:
>>>  ESDHC_FLAG_USDHC
>>>  ESDHC_FLAG_STD_TUNING
>>>  ESDHC_FLAG_HAVE_CAP1
>>> ESDHC_FLAG_HS200
>>>  ESDHC_FLAG_HS400
>>>  ESDHC_FLAG_HS400_ES
>>>  ESDHC_FLAG_STATE_LOST_IN_LPMODE
>>>
>>> It does not have the ESDHC_FLAG_BROKEN_AUTO_CMD23 that is present in the imx7d.
>>
>> AFAIU, it looks imx8mm is compatible with imx7d, because the broken
>> acmd23 only limits the features. If imx8mm binds according to imx7d, it
>> will not support acmd23 and HS400-ES.
>>
>> Having three compatibles is therefore also OK.
> 
> My thoughts, exactly.
> 
>> You could also add two cases:
>> 1. three compatibles, deprecated: True,
>> 2. two compatibles, without imx7d.
>>
>> Existing DTS stays with three compatibles for two years and later gets
>> converted to two compatibles. New DTS should use two compatibles.
>>
>> It's quite a lot of churn, but would make in the long term bindings
>> correct and also not break other users/projects.
> 
> I don't see why we need to deprecate the old binding. New SoCs
> can be imx8mm-usdhc compatible from the beginning and need not
> care about the old binding. Existing SoCs can just remain imx7d-usdhc
> compatible as they are now.
> 
> I don't see what the deprecation accomplishes.

It avoids to have too many entries of imx8mm (imx8mm alone,
imx8mm+imx7d, imx8xx+imx8mm+imx7d).

Best regards,
Krzysztof