On Thu, Jan 7, 2021 at 1:16 PM Fabio Estevam [off-list ref] wrote:

The driver drivers/net/phy/mdio_bus.c and the yaml description says
that it only supports one bus level PHY GPIO reset via reset-gpios
property.

I suppose it's this binding:
Documentation/devicetree/bindings/net/ethernet-phy.yaml

On the imx6ul-evk, there are two KSZ8081 PHYs in the same mdio bus, so
this is how I described it in dts:

https://pastebin.com/raw/xLjYUHdN

but the 'reset-gpios' properties are not found in this case. I think
we need to let drivers/net/phy/mdio_bus.c to search for 'reset-gpios'
also inside the mdio children.

This driver gives me headache.

The bindings say that it should populate devices from the compatible
of the subnodes with names like "ethernet-phy-id0141.0e90"
and stuff like that.

Indeed, but I don't understand why the phy in this example does not
have a compatible string?

There is some hackery going on to probe the driver from the bus
level since
commit 46abc02175b3c246dd5141d878f565a8725060c9
"phylib: give mdio buses a device tree presence"

But as far as I read the code this driver should be probing
devices inidividually for ther reset-gpios, I think the problem is
maybe that no proper devices (mdiodev) are added for these
devices, does mdiobus_register_device() even get called
for them? Does of_mdiobus_register_phy() get called?
If this is the problem I think you need to add
compatible for your phy devices and make sure there is
some code to probe them as well?

of_mdiobus_register() must be called by the MDIO bus master/host
driver. This is what traverses the children with
for_each_available_child_of_node().
Does your host driver properly call this function?

Sorry if I sound a bit confused.

Yours,
Linus Walleij