[PATCH v5 02/10] dt-bindings: phy: qcom,qmp-usb: Add Glymur USB UNI PHY compatible

[Wesley Cheng]

On 10/18/2025 8:39 AM, Krzysztof Kozlowski wrote:
> On 18/10/2025 02:20, Wesley Cheng wrote:
>>
>>
>> On 10/16/2025 9:41 PM, Krzysztof Kozlowski wrote:
>>> On 17/10/2025 02:15, Wesley Cheng wrote:
>>>>>> Technically its all handling the same clock branch (CXO), we have the
>>>>>> TCSR clkref register that allows us to gate the CXO to the USB PHY, as
>>>>>
>>>>>
>>>>> Ah, exactly. Then clkref is not a clock. You need rather proper clock
>>>>> hierarchy.
>>>>>
>>>>>> CXO is shared across several HW blocks, so it allows us to properly
>>>>>> powerdown the PHY even though other clients are voting for CXO on.  Then
>>>>>> we obviously have to remove our vote to the overall CXO, so that it can
>>>>>> potentially be shutdown.
>>>>>>
>>>>>> Maybe we can rename it to "clkref" for the CXO handle and
>>>>>> "clkref_switch" for the TCSRCC handle?
>>>>>
>>>>> Naming is better, but it is still not correct. This is not independent
>>>>> clock signal. It is the same clock.
>>>>>
>>>>
>>>> Hmmm... I guess that's why I kept the same clkref tag, to denote that
>>>> its the same clock, but one is a switch/gate for it.  Would you happen
>>>> to have any suggestions you might have that makes it clearer for
>>>> everyone to understand?
>>> To me it looks like:
>>>
>>> |-----|            |-----------|           |------------------|
>>> |clock|------------|TCSRCC gate|-----------|clkref to this dev|
>>> |-----|            |-----------|           |------------------|
>>>
>>> So you need proper clock controller for TCSR (TCSR Clock Controller, in
>>> short TCSRCC, what a surprise!) which will take input, add gate and
>>> produce clock for this device.
>>>
>>> Nothing non-standard, all Qualcomm SoCs have it, every other platform
>>> has it in some way.
>>>
>>
>> Hi Krzystof,
>>
>> Yes, the design is exactly how you outlined it above.  How about clkref
> 
> Hm? There is no connection between the clock and the device. Do you see
> any line going there?
> 
>> for the clock and tcsrcc_switch for the clkref switch?  That removes any
>> notation that the gate/switch is an actual clock...
> 
> You really did not get the point of this entire discussion.
> 

I won't argue how I interpreted this conversation vs your understanding. 
  We can remove our vote to the clock itself, since tcsrcc registers the 
CXO as its parent, so it'll handle the reference counting for us 
whenever we vote on the tcsr clkref.  Thank you for taking the time to 
explain what you were looking for.

Thanks
Wesley Cheng