[PATCH 1/2] arm64: dts: meson-axg: add missing reset-names property

[Jerome Brunet]

On Mon 17 Jan 2022 at 10:49, Neil Armstrong <narmstrong at baylibre.com> wrote:

> Hi,
>
> On 16/01/2022 18:30, Jerome Brunet wrote:
>> 
>> On Sun 16 Jan 2022 at 10:49, Alexander Stein <alexander.stein at mailbox.org> wrote:
>> 
>>> Am Samstag, 15. Januar 2022, 16:04:10 CET schrieb Jerome Brunet:
>>>>
>>>> On Sat 15 Jan 2022 at 10:35, Alexander Stein <alexander.stein at mailbox.org> 
>>> wrote:
>>>>
>>>>> Bindings amlogic,axg-fifo.txt mandates that reset-names is a required
>>>>> property. Add it.
>>>>
>>>> Binginds *mandates* ?? the bindings you are adding mandates that, not the
>>>> previous doc, nor the driver.
>>>
>>> Well, under required properties 'reset-names' is listed as well as 'arb' is 
>>> required, only 'rst' is optional.
>> 
>> I think there is a misunderstanding then.
>> The arb reset is required, the "reset-names" is not - as long as there
>> is single reset.
>
> To be fair, it's not explicit in the .txt bindings at all:
>
- resets: list of reset phandle, one for each entry reset-names.
> -- reset-names: should contain the following:
> -  * "arb" : memory ARB line (required)
> -  * "rst" : dedicated device reset line (optional)

That was fairly usual way to describe clocks and reset with txt files
but I agree it could have been interpreted the other way around

>
> Anyway, this should be solved, it's pretty common to have reset-names mandatory even
> for a single reset if a second one is optional.

Binding should not decribe what's common but how the binding is supposed to be
used. Fact is the usage was defined by the first and only user which is linux
driver.

This driver does not care if the arb name is present or not. Mandating
something which is unused makes no sense.

If we want to be precise, then it just cares it is the first reset is
the arb one (and yes, this constraint is not described either).

The reason for that is simple, there was no 'rst' line on
first version of the IP, and it was 'fairly usual' to not have
'reset-names' when there is a single reset.

If you think the 'arb' name should be made mandatory, that's fine by
me but one should be able to rely on the name so the driver should be
updated to use it.

>
>> 
>>> So when creating the .yaml accordingly this leads to warnings this patch is 
>>> about to fix.
>>>
>>>> Modifying drivers and DT to accomodate made-up bindings requirement is
>>>> disturbing.
>>>>
>>>> The bindings should not require that because the driver does not, as it
>>>> stands. The driver requires the arb reset to be provided, not the name.
>>>> Please fix the bindings.
>>>
>>> Nothing is made up. When creating the .yaml file I took the .txt documentation 
>>> for granted. How should I know the bindings documentation is apparently wrong?
>>>
>>> When using your older bindings conversion [1] I'm fine with dropping this one.
>>>
>>> Best regards,
>>> Alexander
>>>
>>> [1] https://patchwork.kernel.org/project/linux-amlogic/list/?
>>> series=246453&state=%2A&archive=both
>>