Re: [PATCH v2] hwmon: (it87) Add support for IT8625E

[Guenter Roeck <linux@roeck-us.net>]

On 10/23/24 05:41, Ahmad Khalifa wrote:
> On 22/10/2024 14:40, Guenter Roeck wrote:
>> On 10/22/24 03:13, Frank Crawford wrote:
>>> On Tue, 2024-10-22 at 17:13 +0800, Ai Chao wrote:
>>>> Add support for IT8625E on Centerm P410.
> ...
>>> Can I just add that it isn't a good idea to use the same type for
>>> different chips.  There are some specific differences between the
>>> chips, which mean that it should have its own entry in
>>>
>>> static const struct it87_devices it87_devices[]
>>>
>>> even if currently they are very similar.
>>
>> According to the information I have, the ADC voltage is different,
>> and 8628 supports PECI but 8625 doesn't. Most importantly, 8625
>> has multiple register banks. There are also some differences in
>> fan control; 8628 can explicitly turn fans off using register bits.
>>
>> Just mapping the chip to it8628 may be convenient, but it is not
>> acceptable.
> 
> Side question here. The standard for an acceptable chip driver is pretty
> high (and rightfully so). But a common use case centres around readonly
> display of information: temp/fan/in readings. Even just 2-3 readings are
> better than nothing.
> 
> Example, I still have to use Frank's out of tree it87 for my IT8688.
> It works perfectly fine for me, but still not possible to merge that
> device into hwmon's it87.
> This IT8625 is another example. The NCT6701D-R will be one more shortly.
> 

For the most part you can use the kernel driver with force_id parameter.
That is no different than your suggested patch.

Nuvoton is usually very supportive, so I don't see a problem adding
support for NCT6701D-R if someone is willing to spend the time to write
a driver (or adding support to an existing driver if the chip is similar).

> A readonly driver that is configurable from userspace would help in that
> use case. It can be configured for known devices without datasheets or
> for testing new devices. Wouldn't even need to access superio config
> space.
> 
> Filesystem has fuse, i2c has i2c-dev, input has evdev, ...
> Would something similar be acceptable for hwmon?
> 

I would have to details of a proposal, but It seems unlikely.
After all, chip details have to be sufficiently known to provide
any driver, even a userspace one. Also, the hwmon ABI (i.s., its set
of sysfs entries) isn't that special that it would warrant such a
kernel infrastructure. If at all, it might make more sense to add
support for this, for example, to libsensors.

Guenter