[PATCH net-next v3 01/10] net: stmmac: move XPCS lifetime management to platform drivers

[Maxime Chevallier]

On 9/3/26 10:51, Coia Prant wrote:
> Maxime Chevallier <maxime.chevallier at bootlin.com> 于2026年9月3日周四 16:32写道：
>>
>> Hi again,
>>
>> Sorry I had a second look at that after a good night's sleep, and actually here
>> we may regress. The original logic checked for "pcs-handle" presence, then
>> for the pcs mask, but if no PCS is found we didn't error out, we returned 0.
>>
>> This is making it mandatory to have a PCS. Please handle gracefully the "there's no
>> PCS" case :(
>>
>>
>> Maxime
> 
> Hi,
> 
> This change covers all callbacks that use `select_pcs`. Drivers
> without `select_pcs` callbacks will not have their pcs used by
> Phylink.
> 
> Currently, only the Intel mgbe driver uses xpcs, and Intel has also
> set a mask for pcs-handle not set.
> 
> This means that on Intel platforms, the parts that implement
> `select_pcs` always require xpcs to exist.
> 
> I think there shouldn't be any problems here, but I agree with you,
> regression issues can indeed easily occur here.

Alight, let's keep it this way. The assumption becomes "if you have
a pcs_init() callback, it means you expect a PCS to be there, ENODEV
is an actual error".

My reviewed-by still stands then.

Maxime