On Thu, 7 May 2020 17:13:11 +0200
Miquel Raynal <miquel.raynal@xxxxxxxxxxx> wrote:
> Hi Boris,
>
> Boris Brezillon <boris.brezillon@xxxxxxxxxxxxx> wrote on Thu, 7 May
> 2020 14:11:03 +0200:
>
> > On Thu, 7 May 2020 13:00:33 +0200
> > Miquel Raynal <miquel.raynal@xxxxxxxxxxx> wrote:
> >
> >
> > > +
> > > +static void anfc_chips_cleanup(struct arasan_nfc *nfc)
> > > +{
> > > + struct anand *anand, *tmp;
> > > +
> > > + list_for_each_entry_safe(anand, tmp, &nfc->chips, node) {
> > > + nand_release(&anand->chip);
> >
> > ret = mtd_device_unregister(nand_to_mtd(&anand->chip));
> > WARN_ON(ret);
> > nand_cleanup(&anand->chip);
> >
> > Or maybe add this WARN_ON() to nand_release() so we don't have to ask
> > people to use mtd_device_unregister() + nand_cleanup().
>
> I don't get your point here? I'm not against adding a warn_on between
> both functions but it's not related to this driver?
We've asked people to not call nand_release() but instead call
mtd_device_unregister()+nand_cleanup(), which is not done here. My
point is, if even us can't get it right, maybe it's a sign we should
instead patch nand_release() to do the right thing.
>
> > We really
> > should fix that at some point (allocate nand_chip and mtd_info
> > separately and leave a dummy mtd_info object with all hooks returning
> > ENODEV when the unregister fails).
>
> Yes, we should fix that.
>
> >
> > > + list_del(&anand->node);
> > > + }
> > > +}
>
> Thanks,
> Miquèl
