[PATCH v6 05/12] power-domain: Add refcounting

[Miquel Raynal]

On 16/04/2025 at 16:20:06 +03, Wadim Egorov <w.egorov at phytec.de> wrote:

> Am 16.04.25 um 12:50 schrieb Neha Malcom Francis:
>> Hi Miquel
>> On 16/04/25 14:36, Miquel Raynal wrote:
>>> Hello,
>>>
>>>>>> -int power_domain_on(struct power_domain *power_domain)
>>>>>> +int power_domain_on_lowlevel(struct power_domain *power_domain)
>>>>>>   {
>>>>>> +	struct power_domain_priv *priv = dev_get_uclass_priv(power_domain->dev);
>>>>>>   	struct power_domain_ops *ops = power_domain_dev_ops(power_domain->dev);
>>>>>> +	int ret;
>>>>>>     	debug("%s(power_domain=%p)\n", __func__, power_domain);
>>>>>>   -	return ops->on ? ops->on(power_domain) : 0;
>>>>>> +	if (priv->on_count++ > 0)
>>>>>> +		return -EALREADY;
>>>>>
>>>>> This change is broken for power domain providers with #power-domain-cells = <1>,
>>>>> which can have multiple domains per provider device. There would need to be a
>>>>> separate reference count per domain, and currently the uclass doesn't know the
>>>>> range of valid domain IDs.
>>>>
>>>> I didn't see this reply earlier, would've saved some time debugging to
>>>> come to the same conclusion :) but yes this is the reason for
>>>> breaking.
>>>
>>> That's indeed the reason, thanks a lot for figuring this out. I am
>>> looking for a solution. I can reproduce on imx8mp by enabling the two
>>> LCD interfaces, as they have a similar pattern as on k3 platform: a
>>> single power domain node and one cell for figuring out which PD to
>>> enable.
>>>
>>> The uclass does not save any data, so I don't have an immediate fix to
>>> propose. Let me dig a bit more into that and find a solution.
>>>
>> Thanks!
>> Meanwhile, could we revert this patch to keep the platforms from
>> breaking?
>
> Would be nice, just took me the same route to pin this commit and find
> out about this discussion. I think more people will run into the same
> issue soon

Go ahead, unfortunately, it's not straightforward to fix properly.

Thanks,
Miquèl