Re: [PATCH 1/2] dt-bindings: regulator: mediatek: Add MT8196 vmm controller

[Krzysztof Kozlowski <krzk@kernel.org>]

On 28/05/2025 06:01, Nancy Lin (林欣螢) wrote:
>>
>> Virtual regulators do not sound real, so feels like you want some
>> sort
>> of power domains?
>>
> This regulator supplies power to the power domain. Before the power
> domain can be powered up, the MediaTek PM driver first enables the
> regulator that the power domain relies on, and then powers up the
> domain. While "virtual" might not be the best term, it accurately
> describes a regulator used to control the power switch.
> 
> Ultimately, the VCP (uP) handles the on/off control of the regulator.
> Therefore, we encapsulate it as a standard regulator in the kernel.
> 
> The internal control of the VMM regulator is as follows:
>    kernel                 HWCCF                          VCP (uP)
> |--------------|      |--------------|   irq    |---------------------|
> | VMM reglator | ---> | hardware     |  ----->  | get buck on/off irq |
> |              |      | voter        |          | and then turn on/off|
> |              |      |              |          | buck                |
> |--------------|      |--------------|          |---------------------|
> 
> When the regulator needs to be turned on or off, it uses the voter
> provided by hwccf to cast a vote for on/off. The VCP then receives the
> corresponding IRQ for the regulator and performs the buck on/off
> accordingly.

OK, thanks for explanation.

> 
...

> 
>>> +
>>> +  mediatek,hw-ccf:
>>> +    $ref: /schemas/types.yaml#/definitions/phandle
>>> +    description: Phandle to the hardware common clock framework
>>> syscon controller.
>>
>> No, you cannot express clocks with syscon.
>>
> HWCCF is MediaTek's internal naming; the functionality of the hardware
> IP is voter. We use the IP to vote for buck on/off, and the VCP will
> get the voting result to perform buck on/off. I will refine the
> naming/description to "voter."


OK, that was a bit confusing, but now I have more questions:

Why isn't this a child of the HWCCF? You do not have here any resources,
so this looks just to instantiate device for Linux.

Looking at your diagram earlier (maybe include it in binding description
if this stays as is at the end) I would imagine phandle only to hwccf.
But you have two phandles - HWCCF and VCP.

So maybe this is part of VCP device, not HWCCF? I don't know. I am
afraid the moment I will suggest you something someone from Collabora
will be pointing out that my advice us wrong and I don't know this
hardware. Yeah, I don't know it. You should know it.

> 
>>> +
>>> +  mediatek,vcp:
>>> +    $ref: /schemas/types.yaml#/definitions/phandle
>>> +    description: Phandle to the Video Co-Processor (VCP) node.
>>
>> For what purpose?
>>
> This is used to ensure that the VCP uP is ready to receive the hwccf
> IRQ and handle the corresponding buck on/off operations.
> 
>>> +
>>> +patternProperties:
>>> +  "^(vdisp|vdec-vcore)$":
>>
>> Redundant nodes, useless. Drop these completely.
>>
> OK.
Best regards,
Krzysztof