[PATCH net v4 2/2] net: phy: mdio-i2c: defer RollBall bridge probe to PHY discovery

[Maxime Chevallier]

Hi Petr

On 6/26/26 18:35, Petr Wozniak wrote:
> Maxime Chevallier wrote:
>> I finally got time to test this with a RollBall module, and I
>> confirm what Aleksander says, the RollBall module's PHY doesn't
>> get detected even with this patch. It does work on v7.0 though,
>> so before the bridge probing was introduced.
> 
> Thanks a lot for taking the time to test, Maxime - and Aleksander
> for the original report.
> 
> That settles it: the deferred-probe approach in patch 2/2 doesn't
> actually restore genuine RollBall PHY detection, and as you both
> confirm it worked before 8fe125892f40 introduced the bridge probing.
> Sashiko's static review flagged the same thing (the probe bypasses the
> PHY discovery retry loop for slow-initializing modules), so the static
> analysis and the two hardware reports all point at the same flaw.
> 
> I only have RTL8261BE-based copper modules here, not a genuine RollBall
> one, so I can't develop and verify a proper slow-init timing fix
> (module_t_wait / a retry that waits for the bridge) without the
> hardware to test against.
> 
> Given that, my suggestion:
> 
> - Please drop patch 2/2 from the series.
> 
> - Since 8fe125892f40 regressed genuine RollBall detection and the
>   deferred probe doesn't restore it, I think the cleanest fix is to
>   revert 8fe125892f40. I'm happy to send that revert if you'd prefer.
>   The 5-minute RTL8261BE probe loop it was addressing is handled in our
>   downstream tree, so reverting it upstream is fine on our side.
> 
> - Patch 1/2 (the mii_bus leak fix) is independent of all this and
>   already has Reviewed-by from Maxime and Larysa - it would be good to
>   take that one regardless. I can resend it standalone if that's easier.
> 
> A proper fix covering slow-firmware modules really needs a genuine
> RollBall module to validate, so it's better owned by someone who has
> that hardware - happy to help review.
> 

I only have a single rollball module here, and none with the missing
bridge, so I can't test the full thing either.

In that case, I agree with reverting 8fe125892f40 as this is breaking
all rollball modules right now.

Can you send the revert ?

Maxime