Re: [PATCH 2/6] net: bcmgenet: refactor phy mode configuration

From: Florian Fainelli
Date: Mon Feb 03 2020 - 13:55:13 EST


On 2/3/20 10:46 AM, Jeremy Linton wrote:
> Hi,
>
> On 2/2/20 9:24 PM, Florian Fainelli wrote:
>>
>>
>> On 2/2/2020 5:17 PM, Andrew Lunn wrote:
>>> On Sat, Feb 01, 2020 at 08:24:14AM -0800, Florian Fainelli wrote:
>>>>
>>>>
>>>> On 1/31/2020 11:46 PM, Jeremy Linton wrote:
>>>>> The DT phy mode is similar to what we want for ACPI
>>>>> lets factor it out of the of path, and change the
>>>>> of_ call to device_. Further if the phy-mode property
>>>>> cannot be found instead of failing the driver load lets
>>>>> just default it to RGMII.
>>>>
>>>> Humm no please do not provide a fallback, if we cannot find a valid
>>>> 'phy-mode' property we error out. This controller can be used with a
>>>> variety of configurations (internal EPHY/GPHY, MoCA, external
>>>> MII/Reverse MII/RGMII) and from a support perspective it is much easier
>>>> for us if the driver errors out if one of those essential properties
>>>> are
>>>> omitted.
>>>
>>> Hi Florian
>>>
>>> Does any of the silicon variants have two or more MACs sharing one
>>> MDIO bus? I'm thinking about the next patch in the series.
>>
>> Have not come across a customer doing that, but the hardware
>> definitively permits it, and so does the top-level chip pinmuxing.
>>
>
> Does the genet driver?
>
> I might be missing something in the driver, but it looks like the whole
> thing is 1:1:1:1 with platform dev:mdio:phy:netdev at the moment. Given
> the way bcmgenet_mii_register is throwing a bcmgenet MII bus for every
> device _probe().

Andrew's question was hypothetical, and I answered it the best way I
could. I did not say this was a *currently* supported combination from a
software perspective or that it would work today, but the hardware would
allow it. This question probably belongs to patch #3 which is about
using (or not phy_find_first).
--
Florian