Re: [PATCH v8 1/2] dt-bindings: phy: qcom: Add CSI2 C-PHY/DPHY schema

[Vladimir Zapolskiy <vladimir.zapolskiy@linaro.org>]

On 6/4/26 12:06, Bryan O'Donoghue wrote:
> On 04/06/2026 09:46, Vladimir Zapolskiy wrote:
>> On 6/4/26 03:30, Bryan O'Donoghue wrote:
>>> On 04/06/2026 01:07, Vladimir Zapolskiy wrote:
>>>> On 6/4/26 00:18, Bryan O'Donoghue wrote:
>>>>> On 03/06/2026 21:51, Vladimir Zapolskiy wrote:
>>>>>>> Actually, one more thing, Why isn't TITAN TOP GDSC here?>>>> +
>>>>>> If CSIPHYs are true subdevices under the umbrella CAMSS device and
>>>>>> well
>>>>>> described as subnodes, then likely none of power domains are needed
>>>>>> to be
>>>>>> repeatedly described in the children device nodes, since this
>>>>>> information
>>>>>> can be obtained from the parent device by the driver.
>>>>>>
>>>>>> Technically 'power-domains' property can be safely removed, I believe.
>>>>>
>>>>> The policy is to describe the power-domain dependency fully since DT
>>>>> describes hardware not software architecture.
>>>>
>>>> It brings no contardiction to the statement I've given above, the needed
>>>> power domans will be properly described in the parent device, and there
>>>> is no
>>>> sense to repeat the properties it again and again in every child
>>>> subdevice.
>>>>
>>>>> Also for a very practical reason a sub-devices can probe/run
>>>>> asynchronously of the parent device being active so in fact we do need
>>>>> to describe the PDs fully.
>>>>
>>>> In opposite to the above this one is precisely a software centric
>>>> argument,
>>>> which should be excluded from the consideration, as well it's not a big
>>>> deal to make a proper async initialization, removing excessive dt
>>>> properties
>>>> is worth it.
>>>>
>>>
>>> Right look forget about that.
>>>
>>> - DT requires you to describe your hardware. You're not entitled to have
>>>      some other device vote for a clock or a PD you rely on.
>>>
>>
>> Above are two uncorrelated between each other sentences.
>>
>> A device ("consumer") can ask another device ("provider") to behave in
>> one or another way, this is the only possible and thus natually selected
>> system design, and nothing behind it was asked. There is no justification
>> for the proposed flood of multiply repeated data, it's avoidable.
> 
> CAMSS or rather the components of CAMSS modelled in the current node,
> is/are not the provider of the GDSCs or the power-domains, it/they are
> consumers themselves from CAMCC.

Well, this is the argument about software, and software can be changed.

> The producer/consumer model is CAMCC to components within the Camera
> block. Some components depend on say MXA, MXC, some do not. Nothing in
> CAMSS itself is a power-domain provider.
>>>      That's exactly the type of downstream short cut we are trying to zap.
>>>
>>> - In our case we also need to vote on PDs individually when the PHY
>>>      is active.
>>>
>>> In extremis say we are only running the TPG then we have no reason to
>>> vote for CSIPHY specific rails or operating points in the parent device.
>>
>> So, TPG shall communicate with CAMSS, there is no CSIPHY in the equation.
> 
> Right but it would be inappropriate to enable all of the PDs for all of
> the components in the CAMSS block when we can do so more granularly.

Whenever it is actually necessary, it should be possible to split PDs into
generic/parent and subdevice specific groups, it's a part of software
implementation. In some cases there might be no need to define any child
side PDs, likely CSIPHY falls into this category.

> If you drive the CSID with a TPG you shouldn't be voting for MXA or MXC
> since these are PDs related to the CSIPHY only and TPG and CSIPHY do the
> same thing from the CSID perspective - input data.
> 
>>
>>> We could make the parent power-domain argument for CAMSS and CCI but we
>>> have TITAN_TOP_GDSC in CCI specifically because we have to model the
>>> hardware - including the PDs for that device.
>>
>> CCI is not described as a child of CAMSS, here the situation is different.
> 
> CCI probably _should_ be a child of CAMSS given the design we are going
> for here.
> 
> Leaving that aside it doesn't matter if a node appears as a child node
> or a peer node - the DT should describe the hardware setup.

If a node is child or a sibling is a hardware description, different
hardware descriptions bring different dt properties, this is normal.

> I can't imagine a patch that would remove a power-domain from a device
> being accepted simply because the node being moved is expressed as a
> child of another node.

I strongly believe every dt binding describing actual hardware will find
its way to be implemented in software in a proper way, it should be doable.

>>> If tomorrow we put CCI as a sub-device of top-level CAMSS, that won't
>>> negate the need to include that GDSC.
>>
>> Of course in this case a phandle to Titan GDSC will be marked as obsolete
>> or unused for CCI, no problem here.
>>
-- 
Best wishes,
Vladimir