Message-ID: <c687756a-1dcc-4103-b2ac-7c117ad792e2@linaro.org>
Date: Wed, 7 Feb 2024 10:57:16 +0100
From: Krzysztof Kozlowski <krzysztof.kozlowski@...aro.org>
To: Marco Felsch <m.felsch@...gutronix.de>
Cc: gregkh@...uxfoundation.org, robh+dt@...nel.org,
 krzysztof.kozlowski+dt@...aro.org, conor+dt@...nel.org, linux@...ck-us.net,
 heikki.krogerus@...ux.intel.com, linux-usb@...r.kernel.org,
 devicetree@...r.kernel.org, linux-kernel@...r.kernel.org,
 kernel@...gutronix.de
Subject: Re: [PATCH 1/4] dt-bindings: usb: typec-tcpci: add tcpci compatible
 binding

On 07/02/2024 10:05, Marco Felsch wrote:
> On 24-02-06, Krzysztof Kozlowski wrote:
>> On 06/02/2024 15:52, Marco Felsch wrote:
>>> On 24-02-06, Krzysztof Kozlowski wrote:
>>>> On 05/02/2024 17:43, Marco Felsch wrote:
>>>>> This binding descripes the generic TCPCI specification [1]. So add the
>>>>
>>>> Typo: describes.
>>>
>>> Argh.
>>>
>>>> No, this binding describes PTN5110, not generic TCPCI. This is not
>>>> accurate commit description.
>>>
>>> This binding is currently missued if another TCPCI conform chip is used
>>
>> Why would people misuse binding instead of doing things properly? :)
> 
> You know people... ;)
> 
> ...
> 
>>>>>  properties:
>>>>>    compatible:
>>>>> -    const: nxp,ptn5110
>>>>> +    enum:
>>>>> +      - nxp,ptn5110
>>>>> +      - tcpci
>>>>
>>>> I don't think this is correct. First, this is binding for NXP chip, so
>>>> why generic implementation should be here? I would expect it in its own
>>>> dedicated binding.
>>>
>>> The nxp,ptn5110 device was the first driver which implements an TCPCI
>>> conform driver. The driver already support the tcpci binding for i2c-id
>>> devices as I mentioned above. IMHO this whole binding (file) should be
>>> converted and the nxp,ptn5110 compatible should be marked as deprecated.
>>
>> You speak about driver, but I was speaking about binding.
> 
> I know and I was afraid of mention the driver within this conversation
> since this is all about bindings and devices :)
> 
> Nevertheless this particular NXP device does support the generic "tcpci"
> compatible already. The support is pulled indirectly via the
> i2c_device_id.name which is in the end used for of/acpi/legacy devices.
> 
>>>> Second, we rarely want generic compatibles. Care to share more details?
>>>
>>> As said above this particular NXP chip is an TCPCI conform chip. There
>>> is nothing special about it. There are other vendors like OnSemi (in my
>>> case) which implement also an TCPCI conform chip. The (Linux) driver
>>> already binds to the generic tcpci compatible if the i2c-core falls back
>>> to the i2c-device id. It's even more confusing that the i2c-id supports
>>> only the generic binding the of-compatible support only the specifc one.
>>
>> I don't know much about TCPCI, so maybe questions are obvious: you are
>> claiming that there will be no differentiating hardware element, like
>> reset-gpios or power supply for none of TCPCI-conforming chips? All of
>> them will never need any different hardware configuration?
> 
> Of course TCPCI doesn't mention reset gpios or power supplies but if you
> use this argumentation the already supported NXP device shouldn't be
> available too since the binding is missing the VDD supply ;) Since we

The existing binding is incomplete and maybe, as you suggested, misused,
but this is not a reason to make it worse.

> never break compatibility, the vdd-supply have to be optional and the
> same can be done for reset-gpios.

So the answer to my questions is: They will not be 100% identical and
they will need customization?

> 
>> Is this what you claim?
> 
> Please see above.
> 
>> Just to remind: there was such claim for USB and PCI till we figured out
>> it was simply wrong and we are living now with on-board hubs and PCI
>> power-sequencing stuff.
> 
> Don't get me wrong, I get your point. In the end I don't care and can
> copy'n'paste the whole file and change the compatible to the OnSemi
> device or I can add the dedicated OnSemi compatible to this file. But I
> don't wanted to add an 2nd specific compatible while the device already
> supports the generic one but via i2c_device_id.name. Therefore I aligned
> the i2c_device_id with the of_device_id.

You can add generic compatible used as fallback. That's pretty common
practice.

> 
>>>> Are all details expected to follow spec, without need of quirks?
>>>
>>> Please see above, I hope this helps.
>>
>> Sorry, doesn't. You still speak about driver and how it can bind to
>> something. I did not ask about this at all.
>>
>> To be clear:
>> WE ARE NOT TALKING ABOUT LINUX DRIVER.
> 
> I KNOW
> 
>> We talk about hardware and how it is represented in Devicetree,
>> including its supplies, pins, GPIOs and any ideas hardware engineers
>> like to bring.

Then terms "driver" and "binding" (or matching) do not fit here as
arguments whether specific compatible should be there or not. There is
guideline for that: writing bindings, which exactly, 100% covers this
thing here.

Best regards,
Krzysztof

Powered by blists - more mailing lists