Re: [PATCH RFC net-next 3/6] net: hibmcge: create a software node for phy_led

[Jijie Shao]

on 2025/12/16 15:17, Andrew Lunn wrote:
> On Mon, Dec 15, 2025 at 08:57:02PM +0800, Jijie Shao wrote:
>
> > +int hbg_mdio_init(struct hbg_priv *priv)
> > +{
> > +	struct device *dev = &priv->pdev->dev;
> > +	struct hbg_mac *mac = &priv->mac;
> >   	struct mii_bus *mdio_bus;
> >   	int ret;
> >   
> > @@ -281,7 +389,7 @@ int hbg_mdio_init(struct hbg_priv *priv)
> >   
> >   	mdio_bus->parent = dev;
> >   	mdio_bus->priv = priv;
> > -	mdio_bus->phy_mask = ~(1 << mac->phy_addr);
> > +	mdio_bus->phy_mask = 0xFFFFFFFF; /* not scan phy device */
> >   	mdio_bus->name = "hibmcge mii bus";
> >   	mac->mdio_bus = mdio_bus;
> I think you are taking the wrong approach.
>
> Please consider trying to use of_mdiobus_register(). Either load a DT
> overlay, or see if you can construct a tree of properties as you have
> been doing, and pass it to of_mdiobus_register(). You then don't need
> to destroy and create PHY devices.

This is not easy.
`of_mdiobus_register()` requires a device_node, but I currently don't have one.
It is precisely because there is no DT or ACPI node that I am using a software node as a substitute.

In fact, this patch does not destroy and recreate the phydev.
It only delays the timing of phy registration:
first, register the MDIO (without scanning and registering the PHY devices),
then obtain the PHY device, register the software node, and finally register the PHY device.

Thanks,
Jijie Shao