Re: [PATCH 2/7] mfd / platform: cros_ec: move lightbar attributes to its own driver.

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

On 11/23/18 3:52 AM, Enric Balletbo i Serra wrote:
> Hi Guenter,
> 
> On 22/11/18 18:41, Guenter Roeck wrote:
>> Hi Enric,
>>
>> On Thu, Nov 22, 2018 at 12:33:51PM +0100, Enric Balletbo i Serra wrote:
>>> The entire way how cros sysfs attibutes are created is broken.
>>> cros_ec_lightbar should be its own driver and its attributes should be
>>> associated with a lightbar driver not the mfd driver. In order to retain
>>> the path, the lightbar attributes are attached to the cros_class.
>>>
>>> The patch also adds the sysfs documentation.
>>>
>>> Signed-off-by: Enric Balletbo i Serra <enric.balletbo@collabora.com>
>>> ---
>>>
>> ...
>>>   
>>> +int cros_ec_attach_attribute_group(struct cros_ec_dev *ec,
>>> +				   struct attribute_group *attrs)
>>> +{
>>> +	return sysfs_create_group(&ec->class_dev.kobj, attrs);
>>> +}
>>> +EXPORT_SYMBOL(cros_ec_attach_attribute_group);
>>> +
>>> +void cros_ec_detach_attribute_group(struct cros_ec_dev *ec,
>>> +				    struct attribute_group *attrs)
>>> +{
>>> +	sysfs_remove_group(&ec->class_dev.kobj, attrs);
>>> +}
>>> +EXPORT_SYMBOL(cros_ec_detach_attribute_group);
>>> +
>>
>> Are those two functions necessary ? Why not just call sysfs_create_group
>> and sysfs_remove_group directly from the calling code ?
>>
> 
> Actually we have cros_ec_dev which registers the cros_ec class, and sysfs/vbc
> and lightbar using this cros_ec class. I had problems unloading the different
> modules. For example, when I removed cros_ec_dev modules before
> cros_ec_sysfs/cros_ec_vbc/cros_ec_lightbar I got a hang.
> 
> To solve the hang I did the easy solution that is make these drivers depend on
> cros_ec_dev so you're not able to unload cros_ec_dev if first you don't unload
> the sysfs/vbc/lightbar.
> 

That seems like a side effect of the callbacks, which may increase the use count
of cros_ec_dev. If the lack of these callbacks causes problems, we should identify
the root cause and fix it, and not depend on side effects of a callback.

Thanks,
Guenter

> Thinking again about it, I don't really understand now why failed in the first
> place, cros_ec_dev is the parent, so, on remove should call mfd_remove_devices
> for the subdevices.
> 
> So, let me check again this and I'll back to you.
> 
> Thanks,
>   Enric
> 
> 
>> Thanks,
>> Guenter
>>
>