Re: [PATCH 4/4] dts: iot2050: Support IOT2050-SM variant

[Jan Kiszka <jan.kiszka@siemens.com>]

On 19.12.23 10:54, Jan Kiszka wrote:
> On 19.12.23 10:50, Krzysztof Kozlowski wrote:
>> On 19/12/2023 10:03, Jan Kiszka wrote:
>>> On 19.12.23 09:48, Krzysztof Kozlowski wrote:
>>>> On 19/12/2023 09:22, Jan Kiszka wrote:
>>>>>>
>>>>>>> +			gpios = <&wkup_gpio0 53 GPIO_ACTIVE_HIGH>;
>>>>>>
>>>>>> Ditto
>>>>>>
>>>>>
>>>>> This is adjusting the existing LED nodes in k3-am65-iot2050-common.dtsi,
>>>>> not introducing new ones. We can add the color properties in a separate
>>>>
>>>>
>>>> Then why aren't you overriding by phandle/label?
>>>>
>>>
>>> We could do that as well if we added labels first (they don't exist so 
>>> far). Not seeing any difference, though.
>>
>> Confusion? Your code suggests new node, thus you got review like you got.
>>
>>>
>>>>> patch, but the node names are now part of the kernel ABI. Changing them
>>>>> would break existing userland.
>>>>
>>>> You mean label. Why node names became the ABI? Which interface exposes them?
>>>
>>> root@iot2050-debian:~# ls -l /sys/class/leds/
>>> total 0
>>> lrwxrwxrwx 1 root root 0 Dec 19 08:55 mmc0:: -> ../../devices/platform/bus@100000/4fa0000.mmc/leds/mmc0::
>>> lrwxrwxrwx 1 root root 0 Dec 19 08:55 mmc1:: -> ../../devices/platform/bus@100000/4f80000.mmc/leds/mmc1::
>>> lrwxrwxrwx 1 root root 0 Dec 14 21:12 status-led-green -> ../../devices/platform/leds/leds/status-led-green
>>> lrwxrwxrwx 1 root root 0 Dec 19 08:55 status-led-red -> ../../devices/platform/leds/leds/status-led-red
>>> lrwxrwxrwx 1 root root 0 Dec 19 08:55 user-led1-green -> ../../devices/platform/leds/leds/user-led1-green
>>> lrwxrwxrwx 1 root root 0 Dec 19 08:55 user-led1-red -> ../../devices/platform/leds/leds/user-led1-red
>>> lrwxrwxrwx 1 root root 0 Dec 19 08:55 user-led2-green -> ../../devices/platform/leds/leds/user-led2-green
>>> lrwxrwxrwx 1 root root 0 Dec 19 08:55 user-led2-red -> ../../devices/platform/leds/leds/user-led2-red
>>
>> I replied too fast previous and did not include answer here:
>>
>> You have label for that... Somehow all these nodes are half-baked,
>> without all the expected properties and now you call node name as ABI.
>> The node name is not the ABI.
> 
> Well, existing userspace uses those names, and adding the properties
> would break that interface. Now, does Linux do that?
> 

Obviously, we could deviate from the existing naming scheme only for the
new variant, keeping it for the other 5, but that will be "fun" to maintain.

Jan

-- 
Siemens AG, Technology
Linux Expert Center