mdio-uclass.c: support fixed-link subnodes
authorRasmus Villemoes <rasmus.villemoes@prevas.dk>
Mon, 5 Oct 2020 13:15:16 +0000 (15:15 +0200)
committerTom Rini <trini@konsulko.com>
Sun, 21 Feb 2021 23:46:41 +0000 (18:46 -0500)
commitbdf319273b4a752664c089fbffee5bb2024c8586
tree50436e04043b99e921e95f6738bd962afaaef0bc
parent397ecc30b8fa23d0b4fda52cb1aeee1b8e9d07b8
mdio-uclass.c: support fixed-link subnodes

When trying to port our mpc8309-based board to DM_ETH, on top of
Heiko's patches, I found that nothing in mdio-uclass.c seems to
support the use of a fixed-link subnode of the ethernet DT node. That
is, the ethernet node looks like

enet0: ethernet@2000 {
device_type = "network";
compatible = "ucc_geth";
...
fixed-link {
reg = <0xffffffff>;
speed = <100>;
full-duplex;
};

but the current code expects there to be phy-handle property. Adding
that, i.e.

phy-handle = <&enet0phy>;
enet0phy: fixed-link {

just makes the code break a few lines later since a fixed-link node
doesn't have a reg property. Ignoring the dtc complaint and adding a
dummy reg property, we of course hit "can't find MDIO bus for node
ethernet@2000" since indeed, the parent node of the phy node does not
represent an MDIO bus. So that's obviously the wrong path.

Now, in linux, it seems that the fixed link case is treated specially;
in the of_phy_get_and_connect() which roughly corresponds to
dm_eth_connect_phy_handle() we have

    if (of_phy_is_fixed_link(np)) {
        ret = of_phy_register_fixed_link(np);
        ...
    } else {
        phy_np = of_parse_phandle(np, "phy-handle", 0);
...
    }

    phy = of_phy_connect(dev, phy_np, hndlr, 0, iface);

And U-Boot's phy_connect() does have support for fixed-link
subnodes. Calling phy_connect() directly with NULL bus and a dummy
address does seem to make the ethernet work.

Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Heiko Schocher <hs@denx.de>
net/mdio-uclass.c