Date:   Tue, 18 Jul 2023 16:30:34 +0200
From:   Michal Simek <michal.simek@....com>
To:     Krzysztof Kozlowski <krzysztof.kozlowski@...aro.org>,
        Conor Dooley <conor@...nel.org>,
        Piyush Mehta <piyush.mehta@....com>
Cc:     robh+dt@...nel.org, krzysztof.kozlowski+dt@...aro.org,
        conor+dt@...nel.org, p.zabel@...gutronix.de,
        devicetree@...r.kernel.org, linux-kernel@...r.kernel.org,
        git@....com
Subject: Re: [PATCH 1/2] dt-bindings: reset: Updated binding for Versal-NET
 reset driver



On 7/18/23 16:04, Krzysztof Kozlowski wrote:
> On 18/07/2023 16:01, Michal Simek wrote:
>>
>>
>> On 7/18/23 15:20, Krzysztof Kozlowski wrote:
>>> On 18/07/2023 15:11, Michal Simek wrote:
>>>>>>
>>>>>> That numbers in DT are virtual no matter if you use ID from 0 to max or random
>>>>>> values it is up to code to handle them. Checking nr_pins against ID is done in
>>>>>> core but it is up to drivers.
>>>>>
>>>>> No, you confuse "virtual" and "ID". IDs are not virtual. IDs are real
>>>>> and have representation in Linux driver. You do not need to define
>>>>> anything virtual in the bindings.
>>>>
>>>> Not sure how you define ID itself. But HW doesn't know ID. HW knows only
>>>> register which you can use to perform the reset. It is not really 128bit
>>>> register where every bit targets to different IP.
>>>>
>>>> And this is SW-firmware interface like SCMI reset driver.
>>>>
>>>> Firmware is saying that ID 0 is QSPI, ID 1 is MMC.
>>>> Their Linux driver is asking for nr_reset via firmware call which can be
>>>> different for different SOC and that's fine and I have no problem with it.
>>>> But only SCMI server is dictating that ID 0 is QSPI and ID 1 is MMC. Different
>>>> SCMI server implementation can map it differently.
>>>
>>> Sure, and all this points to: no need for bindings.
>>>
>>>>
>>>>
>>>>>> In our case that IDs are coming from firmware and driver itself is just matching
>>>>>> them.
>>>>>
>>>>> So they are the same as if coming from hardware - no need for IDs.
>>>>
>>>> It is hard to say what hardware here exactly is. From my perspective and I am
>>>> not advocating not using IDs from 0 to max, it is just a number.
>>>>
>>>> If my firmware knows that QSPI reset is 0xc10402dU then I will just pass it to
>>>> reach my goal which is reset QSPI IP.
>>>>
>>>> If you think that we should use IDs from 0 to max NR I am happy to pass this
>>>> message to PM team and we should extend any SW to do translation between.
>>>
>>> When we talk about IDs and bindings, we mean IDs meaningful to Linux.
>>> Whatever is ignored by Linux and passed to anyone else - hardware or
>>> firmware - is not a ID anymore from bindings point of view. It's just
>>> some value.
>>
>> Please correct me if I am wrong. Description about ID should be removed from
>> commit message because it is not necessary. And
>> include/dt-bindings/reset/xlnx-versal-net-resets.h
>> should be added when we merge also DT for versal-net SOC.
> 
> No, the binding header is needed only if driver is using it. Adding DTS
> will not change that - still no kernel driver users. No benefits of such
> binding. If there are no users and no benefits - don't make it a binding.

But header in board folder should be fine like these.
arch/arm64/boot/dts/mediatek/mt2712-pinfunc.h
arch/arm64/boot/dts/mediatek/mt8516-pinfunc.h
arch/arm64/boot/dts/mediatek/mt8167-pinfunc.h
arch/arm64/boot/dts/mediatek/mt8173-pinfunc.h
arch/arm64/boot/dts/exynos/exynos-pinctrl.h
arch/arm64/boot/dts/freescale/imx8mq-pinfunc.h
arch/arm64/boot/dts/freescale/imx8mp-pinfunc.h
arch/arm64/boot/dts/freescale/imx8mn-pinfunc.h
arch/arm64/boot/dts/freescale/imx93-pinfunc.h
arch/arm64/boot/dts/freescale/imx8ulp-pinfunc.h
arch/arm64/boot/dts/freescale/imx8mm-pinfunc.h
arch/arm64/boot/dts/tesla/fsd-pinctrl.h

Thanks,
Michal

Powered by blists - more mailing lists