Re: [PATCH v5 RESEND 1/6] platform: x86-android-tablets: other: Add swnode for Xiaomi pad2 indicator LED

[Hans de Goede <hdegoede@redhat.com>]

Hi,

On 3/27/24 12:05 PM, Andy Shevchenko wrote:
> On Wed, Mar 27, 2024 at 9:58 AM Kate Hsuan <hpa@redhat.com> wrote:
>> On Mon, Mar 25, 2024 at 3:30 AM Andy Shevchenko
>> <andy.shevchenko@gmail.com> wrote:
>>> On Sun, Mar 24, 2024 at 5:02 PM Kate Hsuan <hpa@redhat.com> wrote:
> 
> ...
> 
>>>> +/* main fwnode for ktd2026 */
>>>> +static const struct software_node ktd2026_node = {
>>>> +       .name = "ktd2026"
>>>
>>> Leave a comma, this is not a terminator.
>>>
>>>> +};
>>>
>>> When I asked about the name I relied on the fact that you have an idea
>>> how it works. So, assuming my understanding is correct, this platform
>>> may not have more than a single LED of this type. Dunno if we need a
>>> comment about this.
>>
>> I'll make a comment to describe the configuration.
>> This LED controller can be configured to an RGB LED like this. Also,
>> it can be configured as three single-color (RGB) LEDs to show red,
>> green, and blue only.
>> I think the name can be "ktd2026-multi-color". Is it good for you?
> 
> My point here is that the name is static and if you have more than one
> LED in the system, the second one won't be registered due to sysfs
> name collisions. Question here is how many of these types of LEDs are
> possible on the platform? If more than one, the name has to be
> dropped. Writing this I think a comment would be good to have in any
> case.
> 
> ...
> 
>>>> +static int __init xiaomi_mipad2_init(void)
>>>> +{
>>>> +       return software_node_register_node_group(ktd2026_node_group);
>>>> +}
>>>> +
>>>> +static void xiaomi_mipad2_exit(void)
>>>
>>> __exit ?
>> No need.
>> x86-andriod-tablet is based on platform_driver and platform_device so
>> it doesn't need __exit.
>>
>> I put __exit and the compiler complained about the warning.
>> ===
>> WARNING: modpost:
>> drivers/platform/x86/x86-android-tablets/x86-android-tablets: section
>> mismatch in reference: xiaomi_mipad2_info+0x50 (section: .init.rodata)
>> -> xiaomi_mipad2_exit (section: .exit.text)
>> ===
> 
> This is interesting. Why then do we call them symmetrically?
> 
> Hans, do we need to have anything here been amended?

No this is all as expected.

The platform driver's probe() function can be __init because
the platform device is registered with the special:
platform_create_bundle() function which takes a probe() function
and the actual "struct platform_driver" does not have .probe
set at all.

Since we need to do manual cleanup on remove() however we need
a remove() callback and that does sit in the struct platform_driver
and since remove() can normally also be called on manual unbind
of the driver through sysfs it cannot be in the __exit section.

I say normally because IIRC platform_create_bundle() disables
manual unbinding but the section checking code does not know this,
all it sees is that the "struct platform_driver" is not __exit
and that it references the remove() callback and therefor the
remove() callback itself cannot be __exit.

Regards,

Hans