Re: [PATCH] dt-bindings: soc: imx8mp-media-blk-ctrl: Add LDB into DT example

[Marek Vasut <marex@denx.de>]

On 7/1/22 02:09, Rob Herring wrote:
> On Thu, Jun 30, 2022 at 09:50:31PM +0200, Marek Vasut wrote:
>> On 6/30/22 21:31, Krzysztof Kozlowski wrote:
>>
>> Hi,
>>
>> [...]
>>
>>>>>> diff --git a/Documentation/devicetree/bindings/soc/imx/fsl,imx8mp-media-blk-ctrl.yaml b/Documentation/devicetree/bindings/soc/imx/fsl,imx8mp-media-blk-ctrl.yaml
>>>>>> index b246d8386ba4a..05a19d3229830 100644
>>>>>> --- a/Documentation/devicetree/bindings/soc/imx/fsl,imx8mp-media-blk-ctrl.yaml
>>>>>> +++ b/Documentation/devicetree/bindings/soc/imx/fsl,imx8mp-media-blk-ctrl.yaml
>>>>>> @@ -18,11 +18,18 @@ properties:
>>>>>>       compatible:
>>>>>>         items:
>>>>>>           - const: fsl,imx8mp-media-blk-ctrl
>>>>>> +      - const: simple-mfd
>>>>>
>>>>> Not really... simple-mfd means devices is really simple and you just use
>>>>> it to instantiate children. However this is not simple - it's a power
>>>>> domain controller with several clocks and power domains as input.
>>>>>
>>>>> It's not a simple MFD, but a regular device.
>>>>
>>>> I don't understand this comment. The LDB bridge is literally two
>>>> registers with a few bits in this media block controller register area.
>>>> Can you expand on why the simple-mfd is unsuitable and what should it be
>>>> instead ?
>>>
>>> Looking at the bindings you have there 10 power domains, 10 input clocks
>>> and a domain provider. The driver is also not that simple which is
>>> another argument that this is not simple-mfd. Simply, it is not simple.
>>>
>>> What I meant, is that probably you should populate children from the
>>> driver instead of adding simple-mfd compatible. Once you add simple-mfd,
>>> you cannot remove it and children cannot use anything from the parent.
>>
>> No, I don't think so.
>>
>> The block controller provides those 10 power domains, those are separate
>> things controlled by separate registers within the block control register
>> space.
>>
>> This LDB bridge are two more completely unrelated registers which have
>> nothing to do with those power domains . They are just in the same register
>> block because they had to put those registers somewhere. And they are mixed
>> literally in the middle of the register block, because there was space it
>> seems. Hence the simple-mfd is I think the right thing here.
> 
> Was 'simple-mfd' missing or LDB bridge wasn't a child node? I though
> this was a no functional change patch. Seems more than just update the
> example.

In the local patch I have to imx8mp.dtsi , the simple-mfd had to be 
added for the ldb bridge child node to bind.

> 'simple-mfd' is saying the child has 0 dependence on the parent. IMO,
> 'syscon' contradicts that, but that's an all to common pattern. You are
> saying all the clocks (or any other resources) in the parent can be off
> and the LDB bridge is still functional.

The bridge itself should work even if the block control is clocked off. 
The block controller apb (register) clock have to be enabled only when 
accessing these two LDB registers, which should be handled by runtime 
PM. Or what am I missing here ?