Re: [PATCH] net: mdio: force deassert MDIO reset signal

From: Lars-Peter Clausen
Date: Tue Jan 17 2023 - 08:49:26 EST


On 1/17/23 05:40, Andrew Lunn wrote:
So compatible "ethernet-phy-ieee802.3-c45" results in is_c45 being set
true. The if (is_c45 || is then true, so it does not need to call
fwnode_get_phy_id(child, &phy_id) so ignores whatever ID is in DT and
asks the PHY.

Try this, totally untested:

diff --git a/drivers/net/mdio/fwnode_mdio.c b/drivers/net/mdio/fwnode_mdio.c
index b782c35c4ac1..13be23f8ac97 100644
--- a/drivers/net/mdio/fwnode_mdio.c
+++ b/drivers/net/mdio/fwnode_mdio.c
@@ -134,10 +134,10 @@ int fwnode_mdiobus_register_phy(struct mii_bus *bus,
if (rc >= 0)
is_c45 = true;
- if (is_c45 || fwnode_get_phy_id(child, &phy_id))
+ if (fwnode_get_phy_id (child, &phy_id))
phy = get_phy_device(bus, addr, is_c45);
else
- phy = phy_device_create(bus, addr, phy_id, 0, NULL);
+ phy = phy_device_create(bus, addr, phy_id, is_c45, NULL);
if (IS_ERR(phy)) {
rc = PTR_ERR(phy);
goto clean_mii_ts;

I think part of the problem is that for C45 there are a few other fields
that get populated by the ID detection, such as devices_in_package and
mmds_present. Is this something we can do after running the PHY drivers
probe function? Or is it too late at that point?
As i hinted, it needs somebody to actually debug this and figure out
why it does not work.

I think what i did above is part of the solution. You need to actually
read the ID from the DT, which if you never call fwnode_get_phy_id()
you never do.

You then need to look at phy_bus_match() and probably remove the

return 0;
} else {

so that C22 style ID matching is performed if matching via
c45_ids.device_ids fails.

Sorry, I've should have been more clear. I did try your proposed change a while ago. The problem why it does not work is that there are other fields in the phy data structure that get initialized when reading the IDs for C45. Such as which MMD addresses are valid.