Re: [PATCH 00/62] mtd: rawnand: Get rid of nand_release()

[Masahiro Yamada <masahiroy@xxxxxxxxxx>]

Hi Miquel,


On Sun, May 10, 2020 at 9:14 PM Miquel Raynal <miquel.raynal@xxxxxxxxxxx> wrote:
>
> Hello,
>
> The main goal of this big cleanup series is to get rid of
> nand_release(). I am always confused when it comes to use either
> nand_cleanup() or nand_release() and I need to check the code in the
> core, which is a sign of the API being badly designed.
>
> Historically, there was a "nand_release()" helper only, doing the MTD
> device unregistration and cleaning/freeing all objects belonging to
> the NAND core. Later (v4.9), nand_cleanup() was been added to take
> care of all the NAND cleanup, and was called directly by
> nand_release(). This new helper was very useful to cleanup all the
> NAND core objects in one go on _probe() error (after a successful
> nand_scan()).
>
> Unfortunately, because of the lack of symmetry in the API, a lot of
> people also used nand_release wrongly in the _probe() error path.
>
> So now, instead of:
>
>         nand_release(chip);
>
> I would like people to use:
>
>         ret = mtd_device_unregister(mtd);
>         WARN_ON(ret);
>         nand_cleanup(chip);
>
> Note: as it is currently not possible to handle properly an error on
> mtd_device_unregister(), WARN_ON() will at least inform the user.
>
> All of you receiving this cover letter will at least receive one patch
> applying this change to the driver you are maintaining.
>
> Thanks,
> Miquèl



At first, I just wondered
if we could add nand_register()
that wraps the combo of
nand_scan() and mtd_device_register()
so that nand_register() vs nand_release()
looks symmetrical.


But, not all drivers call
nand_scan() and mtd_device_register()
in a row.

So, I agree we should simply drop
nand_release().


Thanks.

-- 
Best Regards
Masahiro Yamada

______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/