Hello Russell,

I'm re-replying here even though a more recent version was sent, as I 
realise I forgot to fully address that.

On 27/07/2025 11:56, Russell King (Oracle) wrote:

On Tue, Jul 22, 2025 at 02:16:12PM +0200, Maxime Chevallier wrote:

quoted

+static int phy_sfp_module_insert(void *upstream, const struct sfp_eeprom_id *id)
+{
+	struct phy_device *phydev = upstream;
+	struct phy_port *port = phy_get_sfp_port(phydev);
+
+	__ETHTOOL_DECLARE_LINK_MODE_MASK(sfp_support);
+	DECLARE_PHY_INTERFACE_MASK(interfaces);
+	phy_interface_t iface;
+
+	linkmode_zero(sfp_support);
+
+	if (!port)
+		return -EINVAL;
+
+	sfp_parse_support(phydev->sfp_bus, id, sfp_support, interfaces);
+
+	if (phydev->n_ports == 1)
+		phydev->port = sfp_parse_port(phydev->sfp_bus, id, sfp_support);
+
+	linkmode_and(sfp_support, port->supported, sfp_support);
+
+	if (linkmode_empty(sfp_support)) {
+		dev_err(&phydev->mdio.dev, "incompatible SFP module inserted\n");
+		return -EINVAL;
+	}
+
+	iface = sfp_select_interface(phydev->sfp_bus, sfp_support);

I've been moving phylink away from using sfp_select_interface() because
it requires two stages of translation - one from the module capabilties
to linkmodes, and then linkmodes to interfaces.

sfp_parse_support() now provides the interfaces that the optical module
supports, and the possible interfaces that a copper module _might_
support (but we don't know for certain about that until we discover a
PHY.)

The only place in phylink where this function continues to be used is
when there's an optical module which supports multiple different
speeds, and we need to select it based on the advertising mask provided
by userspace. Everywhere else shouldn't use this function, but should
instead use the interfaces returned from sfp_parse_support().

In any case, we'll eventually have to select one of the interfaces if 
there are multiple matches from the sfp_parse_support. phylink maintains 
a sorted list of interfaces used as a preference, I think we should use 
the same list for phy-driver SFP. I'm thinking about moving 
phylink_choose_sfp_interface() in the sfp code, would you be OK with that ?

Maxime