Re: [RFC PATCH 1/7] drivers: base: Add resource managed version of delayed work init

[Hans de Goede <hdegoede@redhat.com>]

Hi,

On 2/13/21 4:27 PM, Guenter Roeck wrote:
> On 2/13/21 7:03 AM, Hans de Goede wrote:
> [ ... ]
>>
>> I think something like this should work:
>>
>> static int devm_delayed_work_autocancel(struct device *dev, struct delayed_work *w,
>> 					void (*worker)(struct work_struct *work)) {
>> 	INIT_DELAYED_WORK(w, worker);
>> 	return devm_add_action(dev, (void (*action)(void *))cancel_delayed_work_sync, w);
>> }
>>
>> I'm not sure about the cast, that may need something like this instead:
>>
>> typedef void (*devm_action_func)(void *);
>>
>> static int devm_delayed_work_autocancel(struct device *dev, struct delayed_work *w,
>> 					void (*worker)(struct work_struct *work)) {
>> 	INIT_DELAYED_WORK(w, worker);
>> 	return devm_add_action(dev, (devm_action_func)cancel_delayed_work_sync, w);
> 
> Unfortunately, you can not type cast function pointers in C. It is against the C ABI.
> I am sure it is done in a few places in the kernel anyway, but those are wrong.

I see, bummer.

> This is the reason why many calls to devm_add_action() point to functions such as
> 
> static void visconti_clk_disable_unprepare(void *data)
> {
>         clk_disable_unprepare(data);
> }
> 
> which could otherwise be handled using typecasts.

Hmm, wouldn't something like this be a candidate for adding a:

devm_clk_prepare_enable() helper?

This seems better then having the driver(s) make + error check separate
clk_prepare_enable() + devm_add_action_or_reset() calls ?

I must admit I'm guilty myself of just using devm_add_action() sometimes
when a specific devm helper is missing, but this whole discussion makes
me think that it would be good to have some extra devm helpers for
common cases / driver cleanup patterns.

If we add a devm_clk_prepare_enable() helper that should probably be added
to drivers/clk/clk-devres.c and not to drivers/base/devres.c .

I also still wonder if we cannot find a better place for this new
devm_delayed_work_autocancel() helper but nothing comes to mind.

Regards,

Hans