Re: [PATCH net-next 2/3] net: ag71xx: use devm for of_mdiobus_register

[Rosen Penev]

On Mon, Aug 12, 2024 at 7:41 PM Andrew Lunn <andrew@xxxxxxx> wrote:
>
> On Mon, Aug 12, 2024 at 02:35:45PM -0700, Rosen Penev wrote:
> > On Mon, Aug 12, 2024 at 2:28 PM Andrew Lunn <andrew@xxxxxxx> wrote:
> > >
> > > On Mon, Aug 12, 2024 at 12:06:52PM -0700, Rosen Penev wrote:
> > > > Allows removing ag71xx_mdio_remove.
> > > >
> > > > Removed local mii_bus variable and assign struct members directly.
> > > > Easier to reason about.
> > >
> > > This mixes up two different things, making the patch harder to
> > > review. Ideally you want lots of little patches, each doing one thing,
> > > and being obviously correct.
> > >
> > > Is ag->mii_bus actually used anywhere, outside of ag71xx_mdio_probe()?
> > > Often swapping to devm_ means the driver does not need to keep hold of
> > > the resources. So i actually think you can remove ag->mii_bus. This
> > > might of been more obvious if you had first swapped to
> > > devm_of_mdiobus_register() without the other changes mixed in.
> > not sure I follow. mdiobus_unregister would need to be called in
> > remove without devm. That would need a private mii_bus of some kind.
> > So with devm this is unneeded?
>
> If you use devm_of_mdiobus_register(), the device core will call
> devm_mdiobus_unregister() on remove. Your patch removed
> mdiobus_unregister() in remove....
>
> Is there any user of ag->mii_bus left after converting to
> devm_of_mdiobus_register()?
There is not. I've applied the change removing mii_bus from ag locally.

>From testing, nothing has blown up. Although there's still a problem
with switched ports (except for lan1) dying after a while. I think
that bug's in qca8k though.

calling restart results in no surprises, unlike with some of my other
questionable patches.
>
>         Andrew