[PATCH 2/4] dt-bindings: net: dsa: document internal MDIO bus

Arınç ÜNAL arinc.unal at arinc9.com
Tue Sep 5 04:00:21 PDT 2023


On 5.09.2023 05:42, Luiz Angelo Daros de Luca wrote:
>>> [1] ...this. The SMI-controlled and MDIO-controlled Realtek switches are
>>> otherwise the same, right? So why would they have different dt-bindings?
>>
>> Honestly, I'm wondering the answer to this as well. For some reason, when
>> probing the SMI controlled Realtek switches, instead of just letting
>> dsa_switch_setup() populate ds->slave_mii_bus, on realtek_smi_setup_mdio()
>> on realtek-smi.c:
>>
>> - priv->slave_mii_bus is allocated.
>> - mdio_np = of_get_compatible_child(priv->dev->of_node, "realtek,smi-mdio");
>> - priv->slave_mii_bus->dev.of_node = mdio_np;
>> - ds->slave_mii_bus = priv->slave_mii_bus;
> 
> I might be able to help here. The Realtek SMI version created a custom
> slave_mii driver because it was the only way to associate it with an
> MDIO DT node. And that DT node was required to specify the interrupts
> for each phy0.
> It would work without that mdio node, letting DSA setup handle the
> slave bus, but it would rely only on polling for port status.
> 
> As we only have a single internal MDIO, the compatible string
> "realtek,smi-mdio" would not be necessary if the driver checks for a
> "mdio"-named child node. Maybe the code was just inspired by another
> DSA driver that uses more MDIO buses or external ones. The "mdio" name
> is suggested by docs since it was committed
> (https://www.kernel.org/doc/Documentation/devicetree/bindings/net/dsa/realtek-smi.txt).
> That name was also kept in the YAML translation
> (https://www.kernel.org/doc/Documentation/devicetree/bindings/net/dsa/realtek.yaml).
> 
> The Realtek MDIO driver was merged at the same release that included
> the change that allows dsa_switch_setup() to reference the "mdio"
> OF-node if present. That way, it could avoid creating a custom
> slave_mii_bus driver.
> 
> I submitted a small series of patches to unify that behavior between
> those two drivers:
> 
> https://lore.kernel.org/netdev/CAJq09z44SNGFkCi_BCpQ+3DuXhKfGVsMubRYE7AezJsGGOboVA@mail.gmail.com/
> (This is my answer to the series opening message to include the first
> paragraph ate by the editor)
> 
> There was some discussion but not NAC, ACK or RFC. It would have
> dropped some lines of code. I can revive it if there is interest.

I'd like this to happen, thanks Luiz!

Arınç



More information about the Linux-mediatek mailing list