Re: [PATCH net-next v13 09/10] net: ethtool: Introduce ethtool command to list ports

[Maxime Chevallier <maxime.chevallier@bootlin.com>]

On 7/7/26 15:46, Paolo Abeni wrote:
> On 7/7/26 12:47 PM, Maxime Chevallier wrote:
>> On 7/7/26 12:12, Paolo Abeni wrote:
>>>
>>> Both sashikos instances points to multiple races with dumps. I suspect
>>> some explicit locking is required.
>>>
>>> https://sashiko.dev/#/patchset/20260701110427.143945-1-maxime.chevallier%40bootlin.com
>>> https://netdev-ai.bots.linux.dev/sashiko/#/patchset/20260701110427.143945-1-maxime.chevallier%40bootlin.com
>>>
>>> I *think* this is hopefully the last pending issue.
>>
>> True indeed, the other issues are sashiko saying "what if we detach a PHY and attach it
>> to another netdev", which can't happen :) I agree with some of the locking points that
>> were found, I'll address that likely as a separate cleanup.
> 
> I'm sorry, but I think it's better to address the above problems with
> the main series to avoid a window of opportunity that could cause a lot
> of quite randomic/drop by submissions to try to addresses them with a
> lot of noise on the ML.

I was thinking the opposite actually, send a cleanup series that gets the
locking part correct before continuing with phy_port, as it seems for now
the phy_port/topology code is avoiding races just because all the callpaths
happen to be under rtnl(), but this isn't formalised properly with asserts
and dedicated locks in the problematic parts of the code.

I otherwise agree, let's not send buggy code to be fixed later and I definitely
see what you mean about the random submissions to fix these.

Maxime