Re: [PATCH net] net: emac: release DT PHY resources on teardown

[Rosen Penev]

On Tue Oct 20, 2026 at 6:05 PM PDT, Andrew Lunn wrote:
> On Tue, Oct 20, 2026 at 03:30:02PM -0700, Rosen Penev wrote:
>> Disconnect PHYLIB PHYs and deregister fixed PHYs during both probe
>> unwind and driver removal. Without this cleanup, unbinding a DT-based
>> EMAC can leave its PHY attached to the freed netdev, while fixed-link
>> instances retain their registered fixed PHY.
>
> This driver is quite convoluted with its PHY handling.
I agree. I got this LLM to convert the whole thing to phylink. It tested
a bunch of stuff over serial and ethernet, which works. That's net-next
material though. It'll probably get rejected.
>
> Please split this into two, phy_disconnect() and
> of_phy_deregister_fixed_link() and add better commit messages.
>
> 	if ((dev->phy_address == 0xffffffff && dev->phy_map == 0xffffffff) ||
> 	    of_phy_is_fixed_link(np)) {
>
> 		/* PHY-less configuration. */
> 		dev->phy.address = -1;
> ...
> 		if (of_phy_is_fixed_link(np)) {
> 			int res = emac_dt_mdio_probe(dev);
>
> 			if (res)
> 				return res;
>
> 			res = of_phy_register_fixed_link(np);
>
> The logic here is more complex than
>
>> +	if (of_phy_is_fixed_link(np))
>> +		of_phy_deregister_fixed_link(np);
>
> So you need to explain in the commit message why it is correct.
>
>> +static void emac_cleanup_phy(struct emac_instance *dev)
>> +{
>> +	struct device_node *np = dev->ofdev->dev.of_node;
>> +
>> +	if (dev->ndev->phydev && dev->phy.address >= 0)
>> +		phy_disconnect(dev->ndev->phydev);
>
> It is also not clear this is correct, since a fixed-link should be
> connected to, yet a fixed link has dev->phy.address == -1?
>
> 	Andrew