On 11/09/2020 14:33:46+0200, Bartosz Golaszewski wrote:

I'm seeing this pattern elsewhere in the kernel too and I just
recently fixed this for MDIO. I think it's just a matter of people
copy-pasting a bad implementation.

quoted

was an actual reason this was done this way and it was the ordering of
the rtc_nvmem_register/rtc_nvmem_unregister with rtc_device_unregister.
I'm not sure this is still necessary though.

To me - each of these should have their own 'unregister' function and
appropriate devres helpers *OR* RTC-related nvmem structures could be
set up and assigned to struct rtc_device after
devm_rtc_allocate_device() and picked up by the registration function
(and also undone by rtc_unregister_device()).

I'll try to allocate some time to look into this but it's not like
it's urgent or anything - it's just a potential improvement.

Well, this could simply be done by adding a devres_add in
__rtc_register_device. I'm planning to remove rtc_nvmem_unregister after
the next LTS which will make that even easier.

-- 
Alexandre Belloni, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com