Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <49834ea3-7fc2-d2cf-36a7-b509d36f260f@linaro.org>
Date:   Tue, 13 Dec 2022 08:50:16 +0100
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.5.1
Subject: Re: [PATCH v2 2/5] dt-bindings: net: add schema for NXP S32CC dwmac
 glue driver
To:     Chester Lin <clin@suse.com>
Cc:     =?UTF-8?Q?Andreas_F=c3=a4rber?= <afaerber@suse.de>,
        "David S. Miller" <davem@davemloft.net>,
        Eric Dumazet <edumazet@google.com>,
        Jakub Kicinski <kuba@kernel.org>,
        Paolo Abeni <pabeni@redhat.com>,
        Rob Herring <robh+dt@kernel.org>,
        Krzysztof Kozlowski <krzysztof.kozlowski+dt@linaro.org>,
        Jan Petrous <jan.petrous@nxp.com>,
        Andrew Lunn <andrew@lunn.ch>,
        Alexandre Torgue <alexandre.torgue@foss.st.com>,
        Giuseppe Cavallaro <peppe.cavallaro@st.com>,
        Jose Abreu <joabreu@synopsys.com>, netdev@vger.kernel.org,
        s32@nxp.com, devicetree@vger.kernel.org,
        linux-kernel@vger.kernel.org, linux-arm-kernel@lists.infradead.org,
        Matthias Brugger <mbrugger@suse.com>,
        ghennadi.procopciuc@oss.nxp.com
References: <20221128054920.2113-1-clin@suse.com>
 <20221128054920.2113-3-clin@suse.com>
 <4a7a9bf7-f831-e1c1-0a31-8afcf92ae84c@linaro.org>
 <560c38a5-318a-7a72-dc5f-8b79afb664ca@suse.de>
 <9778695f-f8a9-e361-e28f-f99525c96689@linaro.org>
 <Y42jqDiiq+rOurV+@linux-8mug>
 <3908e923-a063-0377-1854-ccbb3ecc704d@linaro.org>
 <Y5fnjhugp8cQSzwM@linux-8mug>
Content-Language: en-US
From:   Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
In-Reply-To: <Y5fnjhugp8cQSzwM@linux-8mug>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 13/12/2022 03:46, Chester Lin wrote:
> Hi Krzysztof,
> 
> Sorry for the late reply.
> 
> On Mon, Dec 05, 2022 at 09:55:40AM +0100, Krzysztof Kozlowski wrote:
>> On 05/12/2022 08:54, Chester Lin wrote:
>>>>>>> +examples:
>>>>>>> +  - |
>>>>>>> +    #include <dt-bindings/interrupt-controller/arm-gic.h>
>>>>>>> +    #include <dt-bindings/interrupt-controller/irq.h>
>>>>>>> +
>>>>>>> +    #define S32GEN1_SCMI_CLK_GMAC0_AXI
>>>>>>> +    #define S32GEN1_SCMI_CLK_GMAC0_TX_PCS
>>>>>>> +    #define S32GEN1_SCMI_CLK_GMAC0_TX_RGMII
>>>>>>> +    #define S32GEN1_SCMI_CLK_GMAC0_TX_RMII
>>>>>>> +    #define S32GEN1_SCMI_CLK_GMAC0_TX_MII
>>>>>>> +    #define S32GEN1_SCMI_CLK_GMAC0_RX_PCS
>>>>>>> +    #define S32GEN1_SCMI_CLK_GMAC0_RX_RGMII
>>>>>>> +    #define S32GEN1_SCMI_CLK_GMAC0_RX_RMII
>>>>>>> +    #define S32GEN1_SCMI_CLK_GMAC0_RX_MII
>>>>>>> +    #define S32GEN1_SCMI_CLK_GMAC0_TS
>>>>>>
>>>>>> Why defines? Your clock controller is not ready? If so, just use raw
>>>>>> numbers.
>>>>>
>>>>> Please compare v1: There is no Linux-driven clock controller here but 
>>>>> rather a fluid SCMI firmware interface. Work towards getting clocks into 
>>>>> a kernel-hosted .dtsi was halted in favor of (downstream) TF-A, which 
>>>>> also explains the ugly examples here and for pinctrl.
>>>>
>>>> This does not explain to me why you added defines in the example. Are
>>>> you saying these can change any moment?
>>>>
>>>
>>> Actually these GMAC-related SCMI clock IDs changed once in NXP's downstream TF-A,
>>> some redundant TS clock IDs were removed and the rest of clock IDs were all moved
>>> forward. 
>>
>> This is not accepted. Your downstream TF-A cannot change bindings. As an
>> upstream contributor you should push this back and definitely not try to
>> upstream such approach.
>>
>>> Apart from GMAC-related IDs, some other clock IDs were also appended
>>> in both base-clock IDs and platform-specific clock IDs [The first plat ID =
>>> The last base ID + 1]. Due to the current design of the clk-scmi driver and the
>>> SCMI clock protocol, IIUC, it's better to keep all clock IDs in sequence without
>>> a blank in order to avoid query miss, which could affect the probe speed.
>>
>> You miss here broken ABI! Any change in IDs causes all DTBs to be
>> broken. Downstream, upstream, other projects, everywhere.
>>
>> Therefore thank you for clarifying that we need to be more careful about
>> stuff coming from (or for) NXP. Here you need to drop all defines and
>> all your patches must assume the ID is fixed. Once there, it's
>> unchangeable without breaking the ABI.
>>
> 
> Please accept my apologies for submitting these bad patches. We were just
> confused since we thought that this approach might be acceptable because
> there were some similar examples got merged in the kernel tree:

How are these related to the talk of ABI break?

> 
> https://elixir.bootlin.com/linux/latest/source/Documentation/devicetree/bindings/net/intel,dwmac-plat.yaml#L57
> https://elixir.bootlin.com/linux/latest/source/Documentation/devicetree/bindings/usb/intel,keembay-dwc3.yaml#L55
> https://elixir.bootlin.com/linux/latest/source/Documentation/devicetree/bindings/sound/intel,keembay-i2s.yaml#L73
> https://elixir.bootlin.com/linux/latest/source/Documentation/devicetree/bindings/pwm/intel,keembay-pwm.yaml#L39
> https://elixir.bootlin.com/linux/latest/source/Documentation/devicetree/bindings/watchdog/intel,keembay-wdt.yaml#L46
> https://elixir.bootlin.com/linux/latest/source/Documentation/devicetree/bindings/pci/intel,keembay-pcie.yaml#L75
> https://elixir.bootlin.com/linux/latest/source/Documentation/devicetree/bindings/mmc/arasan,sdhci.yaml#L282

How are these even relevant here? The support for this platform is
incomplete, right? These are just few scattered pieces - only bindings -
without drivers and DTS. It proves nothing. You cannot take some
incomplete platform and build on top of it theory that you can keep
changing ABI!

And anyway it is entirely independent problem. You just said you want to
change defines which is not allowed. ABI break. No one changes the
Keembay defines, whatever they are (you even do not know what they are...).

> 
> The defines in these yaml files are not actually referred by kernel DTs or
> drivers so I assume that they should be provided by firmware as pure integer
> numbers and the clk-scmi driver should just take them to ask firmware for doing
> clk stuff.


Best regards,
Krzysztof