Re: [PATCH v2] hwmon: corsair-cpro: Change to HID driver

[Marius Zachmann <mail@mariuszachmann.de>]

On 15.07.20 at 22:11:39 CEST, Guenter Roeck wrote
> On 7/15/20 12:55 PM, Marius Zachmann wrote:
> > On 15.07.20 at 21:49:59 CEST, Guenter Roeck wrote
> >> On 7/15/20 12:12 PM, Marius Zachmann wrote:
> >>> On 15.07.20 at 17:35:08 CEST, Guenter Roeck wrote
> >>>> On 7/15/20 8:14 AM, Marius Zachmann wrote:
> >>>>> This changes corsair-cpro to a hid driver using hid reports.
> >>>>>
> >>>>> Signed-off-by: Marius Zachmann <mail@mariuszachmann.de>
> >>>>>
> >>>>> ---
> >>>>
> >>> ...
> >>>>> -module_usb_driver(ccp_driver);
> >>>>> +static int __init ccp_init(void)
> >>>>> +{
> >>>>> +	return hid_register_driver(&ccp_driver);
> >>>>> +}
> >>>>> +
> >>>>> +static void __exit ccp_exit(void)
> >>>>> +{
> >>>>> +	hid_unregister_driver(&ccp_driver);
> >>>>> +}
> >>>>> +
> >>>>> +/* make sure it is loaded after hid */
> >>>>> +late_initcall(ccp_init);
> >>>>> +module_exit(ccp_exit);
> >>>>
> >>>> Sorry for not noticing before, but can you use module_hid_driver() ?
> >>>> That seems to work for other hid drivers.
> >>>>
> >>>> Thanks,
> >>>> Guenter
> >>>>
> >>>
> >>> When using module_hid_driver() and compiling the driver as built-in
> >>> the driver init function will get called before hid and fail registering
> >>> the driver. late_initcall forces it to be loaded after hid.
> >>>
> >>
> >> That seems odd. Why does no other hid driver have that problem ?
> >> Your argument is pretty much that module_hid_driver() would not work,
> >> which seems unlikely.
> >>
> >> Thanks,
> >> Guenter
> >>
> > 
> > The order of initcalls is determined by the build order. Hwmon subsystem
> > gets build before hid subsystem and therefore the initcalls are earlier
> > than hid. But hid_register_driver needs the hid driver to be there.
> > 
> Hmm. I am not happy, but I guess we'll have to live with it.
> 
> Can you add the detailed explanation to the comment above ?
> "make sure it is loaded after hid" doesn't tell the whole story,
> and I want to avoid having to deal with "cleanup" patches later.
> 
> Alternatively, I wonder if it would make sense to force
> this driver to be built as module. Have you thought about that ?
> 
> Thanks,
> Guenter
> 

I am also not happy about this and thought about this, but I
think late_initcall is the least ugly solution to this.
I will add a more detailed explanation.

Greetings
Marius