Re: [PATCH] clocksource/drivers/exynos_mct: Fix section mismatch from the module conversion

["Arnd Bergmann" <arnd@arndb.de>]

On Tue, Jul 15, 2025, at 14:18, Daniel Lezcano wrote:
> diff --git a/drivers/clocksource/exynos_mct.c b/drivers/clocksource/exynos_mct.c
> index 5075ebe052a7..80d263ee046d 100644
> --- a/drivers/clocksource/exynos_mct.c
> +++ b/drivers/clocksource/exynos_mct.c
> @@ -657,7 +657,7 @@ static int exynos4_timer_interrupts(struct device_node *np,
>  	return err;
>  }
> 
> -static int mct_init_dt(struct device_node *np, unsigned int int_type)
> +static __init_or_module int mct_init_dt(struct device_node *np, 
> unsigned int int_type)
>  {
>  	bool frc_shared = of_property_read_bool(np, "samsung,frc-shared");
>  	u32 local_idx[MCT_NR_LOCAL] = {0};

Something still feels off here, though I'm not sure what the best
solution would be. Some observations:

 - The exynos4_mct_probe() function is listed as permanent, but
   it indirectly calls an __init_or_module function, which is not
   technically allowed but does not produce a warning here
   because of the indirection.

 - if the driver is built-in and the platform_driver picks
   up the device, it can be manually unbound and rebound, but the
   mct_init_dt() will be discarded at this point.

 - In a loadable module, you cannot call register_current_timer_delay(),
   since this causes a link failure.

 - unloading the module fails to undo the registration, so the
   next call to delay(), sched_clock(), ktime_get() or an interrupt
   ends up in invalid memory.

I think the driver should use an explicit init function calling
platform_driver_probe(), so the probe is only done at initcall
time and both unbinding and unloading are prevented.

       Arnd