On Mon, Aug 14, 2017 at 4:15 PM, Michal Simek [off-list ref] wrote:

quoted

Can't you just move it all the way to device_initcall and
simply use the standard module init macros?
builtin_platform_driver(), module_platform_driver()?

When I grep the kernel I see this

[linux](master)$ git grep "^core_initcall" drivers/gpio/ | wc -l
1
[linux](master)$ git grep "^postcore_initcall" drivers/gpio/ | wc -l
12
[linux](master)$ git grep "^arch_initcall" drivers/gpio/ | wc -l
2
[linux](master)$ git grep "^subsys_initcall" drivers/gpio/ | wc -l
33
[linux](master)$ git grep "^device_initcall" drivers/gpio/ | wc -l
4


[linux](master)$ git grep "^core_initcall" drivers/pinctrl/ | wc -l
6
[linux](master)$ git grep "^postcore_initcall" drivers/pinctrl/ | wc -l
7
[linux](master)$ git grep "^arch_initcall" drivers/pinctrl/ | wc -l
62
[linux](master)$ git grep "^subsys_initcall" drivers/pinctrl/ | wc -l
12
[linux](master)$ git grep "^device_initcall" drivers/pinctrl/ | wc -l
0

Majority of gpio drivers are in subsys_initcall and pinctrl in
arch_initcall.

The majority is likely wrong, we don't vote about what is the
best code quality luckily :D

You do not see a lot of device_initicall because in the majority
of cases these come implicitly from module_platform_driver(),
builtin_platform_driver_probe() or builtin_platform_driver()
see include/linux/platform_device.h

It doesn't mean that I have strong opinion about doing
this change. I have also read internal tracking system and it is not
fully clear if this is fixing any issue rather than removing on
deferring probe message.

I think you can make it into module_platform_driver() please
try that approach.

Yours,
Linus Walleij