Re: [PATCH net-next v6 3/3] net: dsa: yt921x: Add support for Motorcomm YT921x

[Vladimir Oltean <olteanv@gmail.com>]

On Mon, Aug 25, 2025 at 10:14:58PM +0800, Yangfl wrote:
> On Mon, Aug 25, 2025 at 2:14 AM Russell King (Oracle)
> <linux@armlinux.org.uk> wrote:
> >
> > On Mon, Aug 25, 2025 at 12:38:20AM +0800, Yangfl wrote:
> > > They are used in phylink_get_caps(), since I don't want to declare a
> > > port which we know it does not exist on some chips. But the info_* set
> > > might be inlined and removed since it is not used elsewhere.
> >
> > The problem is... if you have a port in 0..N that DSA thinks should be
> > used, but is neither internal or external, DSA's initialisation of it
> > will fail, because without any caps declared for it, phylink_create()
> > will return an error, causing dsa_port_phylink_create() to fail,
> > dsa_shared_port_phylink_register() or dsa_user_phy_setup(),
> > dsa_shared_port_link_register_of() or dsa_user_create()... etc. It
> > eventually gets propagated up causing the entire switch probe to fail.
> >
> > Again... read the code!
> 
> What would you expect when you specify Port 0 in DT when only Port 1,
> 3, 8 are available on the chip (YT9213NB)? Probe error.

It depends. Unless the driver has logic which behaves otherwise, the
setup of user ports can fail and DSA will just skip them and bring up
the rest. See commit 86f8b1c01a0a ("net: dsa: Do not make user port
errors fatal"). The shared ports are more important and their setup
failures do lead to a switch setup abort though.