[PATCH 1/3] phy: stm32-usbphyc: add counter of PLL consumer

[Patrick DELAUNAY]

Hi Sean,

On 5/11/22 18:48, Sean Anderson wrote:
> On 5/10/22 5:51 AM, Amelie Delaunay wrote:
>> Hi Patrick,
>> Hi Sean,
>>
>> On 5/9/22 16:37, Patrick DELAUNAY wrote:
>>> Hi Sean,
>>>
>>> On 5/8/22 20:21, Sean Anderson wrote:
>>>> On 4/26/22 8:37 AM, Patrick Delaunay wrote:
>>>>> Add the counter of the PLL user n_pll_cons managed by the 2 functions
>>>>> stm32_usbphyc_pll_enable / stm32_usbphyc_pll_disable.
>>>>>
>>>>> This counter allow to remove the function stm32_usbphyc_is_init
>>>>> and it is a preliminary step for ck_usbo_48m introduction.
>>>>
>>>> Is it necessary to disable this clock before booting to Linux? If 
>>>> it isn't,
>>>> then perhaps it is simpler to just not disable the clock.
>>>>
>>>> --Sean
>>>
>>>
>>> No, it is not necessary, we only need to enable the clock for the 
>>> first user.
>>>
>>> I copy the clock behavior from kernel,
>>>
>>> but I agree that can be simpler.
>>>
>>>
>>> Amelie any notice about this point ?
>>>
>>> Do you prefer that I kept the behavior - same as kernel driver - or 
>>> I simplify the U-Boot driver ?
>>
>> In case the PLL has not been disabled before Kernel boot, usbphyc 
>> Kernel driver will wait for the PLL pwerdown.
>> USB could also not being used in Kernel, so PLL would remain enabled, 
>> and would waste power.
>> I am rather in favor of disabling the PLL.
>
> It should be disabled if clk_ignore_unused is not in the kernel 
> parameters,
> as long as Linux is also aware of the clock.
>
> Generally, I would like to avoid refcounting if possible. Many U-Boot
> drivers do not disable their clocks (because they don't do any cleanup),
> so you can end up with the clock staying on anyway.
>
> --Sean
>
>> Regards,
>> Amelie
>>
>>>
>>>
>>> Patrick
>>>
>>>
>>>>

In general, I'm also in favor of not disabling the clock in U-Boot.

But this PLL with vdda1v1 and vdda1v8 dependency is requested for

- USBPHYC himself or

- source clock of RCC,


And we have have see many issue for this PLL initialization sequence / 
dependencies.


So for clock support, I only reused the existing/working function called 
by the PHY ops init & exit =

stm32_usbphyc_phy_init & stm32_usbphyc_phy_exit

=> the PLL and the associated VDD  are already deactivated on USBPHYC exit.


And to be coherent I for the same for the clock to avoid conflict 
between these 2 users

(USBPHYC or RCC) as it is done also in Linux kernel.


Avoid to deactivate PLL on clock disable is a major objection or just a 
advice?


Regards

Patrick

.