Re: [PATCH V4 5/5] clk: imx8qxp: Support building i.MX8QXP clock driver as module

[Dong Aisheng]

On Thu, Jul 2, 2020 at 11:55 AM Anson Huang <anson.huang@xxxxxxx> wrote:
[...]
> > > +{
> > > +       return platform_driver_register(&imx8qxp_lpcg_clk_driver);
> > > +}
> > > +device_initcall(imx8qxp_lpcg_clk_init);
> >
> > Any reason to change to device_initcall which looks a bit strange?
> > Is it because the following line?
> > +obj-$(CONFIG_MXC_CLK_SCU) += clk-imx-scu.o clk-imx-lpcg-scu.o
> > But it looks to me they're still two modules. Aren't they?
>
> It is suggested by Arnd to NOT use builtin_platform_driver(), in order to support module
> unload, although the clock driver normally does NOT support remove, but it is better to
> follow the right way.
>

By expanding builtin_platform_driver() marcro, you will find your
patch is exactly doing the same thing
as buildin_platform_driver() which obivously is unneccesary.

#define builtin_platform_driver(__platform_driver) \
        builtin_driver(__platform_driver, platform_driver_register)
#define builtin_driver(__driver, __register, ...) \

static int __init __driver##_init(void) \
{ \
        return __register(&(__driver) , ##__VA_ARGS__); \
} \
device_initcall(__driver##_init);

If we want to support unload, we need a .remove() callback as current
clocks are not allocated by devm_().
If don't support,  we probably can use builtin_platform_driver() first
and switch to module_platform_driver()
in the future once the driver supports release resource properly.

Regards
Aisheng

> The change in Makefile is just to link scu/lpcg library to i.MX8QXP clk driver, so that we can
> get rid of exports and below 2 .ko are needed for all i.MX SoCs with SCU inside as per your
> saying of i.MX8QXP clock driver can be extended for future SoCs with SCU.
>
> clk-imx-lpcg-scu.ko
> clk-imx-scu.ko
>
> Thanks,
> Anson