Re: [PATCH] genirq/devres: Simplify API devm_free_irq() implementation

[quic_zijuhu <quic_zijuhu@quicinc.com>]

On 10/18/2024 4:57 PM, Thomas Gleixner wrote:
> On Fri, Oct 18 2024 at 06:36, Zijun Hu wrote:
>> On 2024/10/18 04:33, Thomas Gleixner wrote:
>>>> It is simpler if devm_free_irq() is undoing what any devm_request_irq()
>>>> variant did, otherwise, it can avoid wrong and undesired free_irq().
>>>
>>> This is confusing at best. What's the wrong an undesired free_irq()?
>>>
>> for current devm_free_irq(..., irq_A, ...):
>> it is undesired if irq_A is requested by request_irq().
>> it is wrong and dangerous if irq_A was never requested.
> 
> There is nothing dangerous about it if it was never requested.
> free_irq() won't find a irq action which matches devid and do nothing
> than emitting a warning.
> 

you are right. it is not dangerous but a unnecessary call of free_irq().

> But that's not relevant either because there is no matching devres entry
> when the interrupt was not requested via devres_request_irq(), so
> free_irq() will not be reached because devres_destroy() will return
> -ENOENT.
> 
no, WARN_ON(devres_destroy()) doesn't return, the next free_irq() will
be still reached.

> So all this change does is changing the logic from:
> 
> devres_free_irq()
>     if (devres_destroy())
>         return;
>     free_irq();
>     
> to
>     devres_release()
> 

if irq to free was ever requested by devm_request_irq()
then both logic is exactly same.

otherwise, actually change devres_free_irq()'s logic from

if (irq is not requested by devm_request_irq() {
	warn;
}
free_irq()

To

if (irq is not requested by devm_request_irq() {
	warn;
	return;
}


> where devres_release() does the same thing as the above, i.e. it looks
> up the devres for a match and if found, it removes and frees the devres
> pointer and invokes the release function which in turn invokes
> free_irq().
> 
> So in terms of code logic this is exactly the same and does neither
> avoid or prevent anything.
> 
> All it does is sparing a single line of code.
> 
> Thanks,
> 
>         tglx
> 
> 
> 
>