Re: [PATCH v1 2/5] extcon: Return -EPROBE_DEFER when extcon device is not found

From: Andy Shevchenko
Date: Wed Nov 14 2018 - 06:17:40 EST


On Wed, Nov 14, 2018 at 1:05 PM Chanwoo Choi <cw00.choi@xxxxxxxxxxx> wrote:
> On 2018ë 11ì 14ì 19:20, Andy Shevchenko wrote:
> > On Wed, Nov 14, 2018 at 11:48 AM Chanwoo Choi <cw00.choi@xxxxxxxxxxx> wrote:
> >> On 2018ë 11ì 14ì 18:36, Andy Shevchenko wrote:
> >>> On Wed, Nov 14, 2018 at 06:13:37PM +0900, Chanwoo Choi wrote:
> >>>> On 2018ë 11ì 14ì 17:35, Andy Shevchenko wrote:
> >>>>> On Wed, Nov 14, 2018 at 1:53 AM Chanwoo Choi <cw00.choi@xxxxxxxxxxx> wrote:
> >>>>>
> >>>>>> I was thinking about again to change from NULL to EPROBE_DEFER.
> >>>>>>
> >>>>>> extcon_get_extcon_dev() function was almost called in the probe function.
> >>>>>> But, this function might be called on other position instead of probe.
> >>>>>
> >>>>> *Might be* sounds like a theoretical thing, care to share what is in you mind?
> >>>>> Current users and more important the new coming one are *all* doing the same.
> >>>>>
> >>>>>> ENODEV is more correct error instead of EPROBE_DEFER.
> >>>>>
> >>>>> So, you are proposing to continue duplicating conversion from ENODEV
> >>>>> to EPROBE_DEFER in *each* caller?
> >>>>
> >>>> The extcon core don't know the caller situation is in either probe() or other position
> >>>> in the caller driver. The caller driver should decide the kind of error value
> >>>> by using the return value of extcon_get_extcon_dev().
> >>>>
> >>>> extcon_get_extcon_dev() function cannot be modified for only one case.
> >>>> If some device driver call extcon_get_extcon_dev() out of probe() fuction,
> >>>> EPROBE_DEFER is not always correct.
> >>>
> >>> I agree with this, but look at the current state of affairs. All users do the same.
> >>> If we need to have another case we may consider this later.
> >>
> >> Because we know the potential wrong case of this change, I can't agree this change.
> >> If extcon_get_extcon_dev() returns ENODEV instead of EPROBE_DEFER,
> >> it is clear and then there are no problem on both current and future.
> >
> > Changing NULL to -ENODEV is a trading bad to worse.
> > I would not go that way, so, it's your call.
>
> If you think that this change is not necessary, just keep the current code
> without the modification.

Not only this, the useless churn for no benefit to anyone, except some
*theoretical* cases no one saw.

> Please drop this patch on next version.

I will.

> >>> API inside the kernel are not carved in the stone.
> >
> > Only can repeat myself (see above). While I see *theoretical*
> > rationale on your side, mine has *practical* proofs.
> > So, I'm giving up on this and will duplicate same what it's done in 4
> > current callers.
> >
>
> I think that it is more important for removing the potential wrong case
> instead of removing the duplicate code. The many device drivers already
> decided the proper error value by using the return value of function of
> kernel framework.

The API has been introduced back in 2012.

commit 74c5d09bd562edc220d6e076b8f1e118819c178f
Author: Donggeun Kim <dg77.kim@xxxxxxxxxxx>
Date: Fri Apr 20 14:16:24 2012 +0900

So, you are insisting that 6.5 years of use in a way people are using
it is wrong?

I don't know what might change your mind, but for me it's a clear
win-win to switch to deferred probe error code based on the
*practical* evidence.
But as I said, I gave up now.

P.S. I still disagree with your arguments in relation to de facto use of an API.

--
With Best Regards,
Andy Shevchenko