Date:   Sun, 12 Feb 2023 07:27:28 -0800
From:   Guenter Roeck <linux@...ck-us.net>
To:     Sergio Paracuellos <sergio.paracuellos@...il.com>,
        Krzysztof Kozlowski <krzysztof.kozlowski@...aro.org>
Cc:     Arınç ÜNAL <arinc.unal@...nc9.com>,
        linux-watchdog@...r.kernel.org, wim@...ux-watchdog.org,
        robh+dt@...nel.org, krzysztof.kozlowski+dt@...aro.org,
        matthias.bgg@...il.com, tsbogend@...ha.franken.de,
        p.zabel@...gutronix.de, linux-kernel@...r.kernel.org,
        devicetree@...r.kernel.org, linux-mips@...r.kernel.org
Subject: Re: [PATCH v4 1/5] dt-bindings: watchdog: mt7621-wdt: add phandle to
 access system controller registers

On 2/12/23 00:13, Sergio Paracuellos wrote:
> On Sat, Feb 11, 2023 at 12:42 PM Krzysztof Kozlowski
> <krzysztof.kozlowski@...aro.org> wrote:
>>
>> On 11/02/2023 12:01, Sergio Paracuellos wrote:
>>> On Sat, Feb 11, 2023 at 11:47 AM Arınç ÜNAL <arinc.unal@...nc9.com> wrote:
>>>>
>>>> On 11.02.2023 13:41, Sergio Paracuellos wrote:
>>>>> On Sat, Feb 11, 2023 at 10:10 AM Arınç ÜNAL <arinc.unal@...nc9.com> wrote:
>>>>>>
>>>>>> Is this mediatek,sysctl property required after your changes on the
>>>>>> watchdog code?
>>>>>
>>>>> I don't really understand the question :-) Yes, it is. Since we have
>>>>> introduced a new phandle in the watchdog node to be able to access the
>>>>> reset status register through the 'sysc' syscon node.
>>>>> We need the bindings to be aligned with the mt7621.dtsi file and we
>>>>> are getting the syscon regmap handler via
>>>>> 'syscon_regmap_lookup_by_phandle()'. See PATCH 5 of the series, Arınç.
>>>>
>>>> I believe you need to put mediatek,sysctl under "required:".
>>>
>>> Ah, I understood your question now :-). You meant 'required' property.
>>> I need more coffee, I guess :-). I am not sure if you can add
>>> properties as required after bindings are already mainlined for
>>> compatibility issues. The problem with this SoC is that drivers become
>>> mainlined before the device tree was so if things are properly fixed
>>> now this kind of issues appear.  Let's see Krzysztof and Rob comments
>>> for this.
>>
>> If your driver fails to probe without mediatek,sysctl, you already made
>> it required (thus broke the ABI) regardless what dt-binding is saying.
>> In such case you should update dt-binding to reflect reality.
>>
>> Now ABI break is different case. Usually you should not break it without
>> valid reasons (e.g. it was never working before). Your commit msg
>> suggests that you only improve the code, thus ABI break is not really
>> justified. In such case - binding is correct, driver should be reworked
>> to accept DTS without the new property.
> 
> Thanks for clarification, Krzysztof. Ok, so if this is the case I need
> to add this property required (as Arinc was properly pointing out in
> previous mail) since without it the driver is going to fail on probe
> (PATCH 5 of the series). I understand the "it was never working
> before" argument reason for ABI breaks. What happens if the old driver
> code was not ideal and totally dependent on architecture specific
> operations when this could be totally avoided and properly make arch
> independent agnostic drivers? This driver was added in 2016 [0]. There
> was not a device tree file in the kernel for this SoC mainlined until
> 2022 [1]. I also personally migrated this watchdog binding in 2022
> from text to YAML and maintained it without changes [2]. When this was
> mainlined not all drivers were properly reviewed and the current code
> was just maintained as it is. Most users of this SoC are in the
> openWRT community where the dtsi of the mainline is not used yet and
> they maintain their own mt7621.dtsi files. Also, when a new version of
> the openWRT selected kernel is added they also modify and align with
> its mt7621.dtsi file without maintaining previous dtb's. If "make the
> driver arch independent to be able to be compile tested" and this kind
> of arguments are not valid at all I need to know because I have
> started to review driver code for this SoC and other drivers also have
> the same arch dependency that ideally should be avoided in the same
> way. This at the end means to break the ABI again in the future for
> those drivers / bindings. So I can just let them be as it is and not
> provide any change at all and continue without being compile tested
> and other beneficial features to detect future driver breakage.
> 

Problem is that there are (presumably) shipped systems out there with
the old devicetree file. The watchdog driver would no longer instantiate
on those systems.

Guenter

Powered by blists - more mailing lists