Re: [PATCH v1] net: phy: tja11xx: add TJA1102 support

[Heiner Kallweit <hkallweit1@gmail.com>]

On 03.03.2020 09:56, Marc Kleine-Budde wrote:
> On 3/3/20 9:46 AM, Heiner Kallweit wrote:
>> On 03.03.2020 08:37, Oleksij Rempel wrote:
>>> TJA1102 is an dual T1 PHY chip. Both PHYs are separately addressable.
>>> PHY 0 can be identified by PHY ID. PHY 1 has no PHY ID and can be
>>> configured in device tree by setting compatible =
>>> "ethernet-phy-id0180.dc81".
>     ^^^^^^^^^^^^^^^^^^^^^^^^
> 
>> I'm not sure I understand what you're doing here. The two ports of the chip
>> are separate PHY's on individual MDIO bus addresses?
> 
> Yes, Port 0 and Port 1 have seperate MFIO bus addresses, but only Port 0
> has a proper phy_id (== PHY_ID_TJA1102), while Port 1 just has 0.
> 
> Currently we register Port 1 via the device tree compatible
> "ethernet-phy-id0180.dc81".
> 
Thanks for the explanation. Missed that part, then reading the PHY ID
of the chip is skipped. For checking whether we are port 0 or 1 it wouldn't
even be needed to read both PHY ID registers.
One drawback of the solution might be that it works on DT-configured
systems only (not sure how relevant this is). A little bit more detailed
comment may be helpful, to avoid misunderstandings like mine.

>> Reading the PHY ID registers here seems to repeat what phylib did already
>> to populate phydev->phy_id. If port 1 has PHD ID 0 then the driver wouldn't
>> bind and tja11xx_probe() would never be called (see phy_bus_match)
> 
> But we "force" it via the DT compatible. Another option would be to make
> up a phyid for Port 1 and hope that it doesn't collide with a real phy
> id in other or upcoming chips. But that sounds not like a clean solution
> either.
> 
> Marc
> 
Heiner