Re: [PATCH 3/4] hwmon: Add support for Amphenol ChipCap 2

[Krzysztof Kozlowski]

On 09/11/2023 10:52, Javier Carrasco wrote:
> On 09.11.23 10:35, Krzysztof Kozlowski wrote:
>> On 09/11/2023 09:59, Javier Carrasco wrote:
>>>
>>>
>>> On 09.11.23 09:40, Krzysztof Kozlowski wrote:
>>>> On 08/11/2023 17:35, Javier Carrasco wrote:
>>>>>>> +
>>>>>>> +	data->regulator = devm_regulator_get_optional(dev, "vdd");
>>>>>>> +	if (!IS_ERR(data->regulator)) {
>>>>>>> +		ret = cc2_retrive_alarm_config(data);
>>>>>>> +		if (ret)
>>>>>>> +			goto cleanup;
>>>>>>> +	} else {
>>>>>>> +		/* No access to EEPROM without regulator: no alarm control */
>>>>>>
>>>>>> Test your code with deferred probe. Are you sure you handle it
>>>>>> correctly? To me, it looks like you handle deferred probe the same as
>>>>>> any error.
>>>>>>
>>>>> The -EPROBE_DEFER is propagated to the probe function and it is the
>>>>> returned value. I clarified the error path in v2 so no error messages
>>>>
>>>> Really?
>>>>
>>>> I see:
>>>> if (!IS_ERR(data->regulator)) {
>>>> 	// so you do not go here
>>>> } else {
>>>> 	goto dev_register;
>>>> }
>>>> dev_register is not error path. So how do you return EPROBE_DEFER?
>>>>
>>>> Which line of code does it?
>>>>
>>> EPROBE_DEFER is returned if the command window was missed, which is
>>
>> How "command window was missed" is related to the place I commented?
>>
> it is right below the comment you added and hence the misunderstanding.
> But focusing on the line where your comment is, there is no probe
> deferring in that case. This is why I asked if you were talking about
> devm_regulator_get_optional() failing, which is not covered by the
> deferring mechanism in the current form.
> 
> I have never experienced the case where the regulator was still not
> available, but I suppose there is no reason why that should never happen.

Defer on regulators, just like several other resources, is quite likely,
so all code must be ready for this.

> The regulator is not mandatory and there is no reason to retry if it is
> not defined. But in case it is defined and not available, the deferring
> would make sense. I could consider that case as well.

Your code should consider it always.


Best regards,
Krzysztof