[RFC][PATCH 2/3] drm/modes: Make width-mm/height-mm mandatory in of_get_drm_panel_display_mode()

[Marek Vasut]

On 4/2/22 19:08, Noralf Trønnes wrote:
> 
> 
> Den 02.04.2022 18.39, skrev Marek Vasut:
>> On 4/2/22 09:45, Noralf Trønnes wrote:
>>>
>>>
>>> Den 02.04.2022 06.28, skrev Marek Vasut:
>>>> On 4/2/22 05:19, Laurent Pinchart wrote:
>>>>> On Fri, Apr 01, 2022 at 10:36:24PM +0200, Marek Vasut wrote:
>>>>>> On 4/1/22 20:46, Laurent Pinchart wrote:
>>>>>>> On Fri, Apr 01, 2022 at 06:37:54PM +0200, Marek Vasut wrote:
>>>>>>>> Make the width-mm/height-mm panel properties mandatory in
>>>>>>>> of_get_drm_panel_display_mode(), print error message and
>>>>>>>> return -ve in case these DT properties are not present.
>>>>>>>> This is needed to correctly report panel dimensions.
>>>>>>>
>>>>>>> Can we guarantee this won't cause a regression ?
>>>>>>
>>>>>> For the upstream DTs, I think we can.
>>>>>> For downstream DTs, we cannot know.
>>>>>
>>>>> Are there users of this function whose DT bindings don't require the
>>>>> width-mm and height-mm properties ?
>>>>
>>>> There is literally one user of this function upstream:
>>>> drivers/gpu/drm/tiny/panel-mipi-dbi.c
>>>
>>> Yes, the function was added for that driver since it was so generic in
>>> nature. What about adding an argument to of_get_drm_panel_display_mode()
>>> that tells if the properties are mandatory or not?
>>
>> Sure, we can do that, but maybe the question here is even bigger than
>> this series.
>>
>> Should every panel set mandatory width_mm/height_mm so e.g. the user
>> space can infer DPI from it and set up scaling accordingly, or should
>> width_mm/height_mm be optional ?
>>
>> I think width_mm/height_mm should be mandatory for all panels.
>>
>> Thoughts ?
> 
> If this had come up during the review of the driver I would have no
> problem making it mandatory. It makes sense for DPI. Maybe it's possible
> to get around the ABI break by getting in a change through -fixes before
> 5.18 is released? I'm fine with that.

Well that's awesome, the dbi-spi.yaml didn't land in any kernel release 
yet, so we still have a chance to fix it ? Rob ?