[Intel-wired-lan] [PATCH net-next v4 0/2] Add mii_bus to ixgbe driver for dsa devs

[Florian Fainelli]

On 12/3/18 3:42 PM, Steve Douthit wrote:
>> Not directly related to this patch series, are you using the legacy or
>> "new" way of passing platform data in order to register the DSA switch?
>> Since you mentioned 6390, I would assume this must be the "new" way of
>> registering DSA devices with mdio_boardinfo etc. In that case, have you
>> found yourself limited by the current dsa_chip_platform_data?
> 
> With the new method I had an off-by-one error where the 'enabled_ports'
> bitmask and 'port_names' array didn't match up in my first attempt.
> That's definitely my fault, but the loop that searches for the "cpu"
> string didn't like it and segfaulted.
> 
> I've got something of a chicken-and-the-egg problem between where I'm
> deferring while waiting for netdevs to show up, and registering the
> mdio_board_info that needs those same pointers.  For testing I exported
> mdiobus_setup_mdiodev_from_board_info() and mdiobus_create_device() so I
> could call the setup function again when the data was actually ready.

Yes the current solution whereby we need to get a hold on the network
device's struct device reference is not quite ideal, AFAIR, Andrew has
had the same problem.

> 
> It'd be nice to have more than one "cpu" port on a switch and have some
> way to associate downstream and "cpu" ports.  Not sure exactly what that
> would look like, and it's not something I'm going to try and tackle at
> the moment, but it's one for the wish list.

Yes, we have been discussing that topic with Andrew and have a few ideas
on how that could be achieved, but not code to use that at the moment.
One of the idea was to finally allow enslaving the DSA master network
device, that way you could assign specific ports to a specific CPU/DSA
master network port within the switch, though that requires setting up a
bridge. Would that work for your use case?
--
Florian