Re: [PATCH net-next v2 2/3] selftests: drv-net: pause: Validate the pause autonegotiation with a partner

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

On 10/13/26 14:37, Andrew Lunn wrote:
>>> Firmware-managed NICs such as ice also keep the link on close by default.
>>> In drivers/net/ethernet/intel/ice/ice_main.c, the PHY is only taken down
>>> on close when ICE_FLAG_LINK_DOWN_ON_CLOSE_ENA is set:
>>>
>>> 	if (test_bit(ICE_FLAG_LINK_DOWN_ON_CLOSE_ENA, vsi->back->flags)) {
>>> 		int link_err = ice_phy_cfg(vsi, false);
>>>
>>> On this kind of hardware, a directly connected peer would be cached as
>>> cfg.lp_controllable = False. pause_aneg_resolution(), and the four later
>>> tests in the series, would then skip with "is not directly connected".
>>> That message points at cabling, when the real cause is that the PHY
>>> stayed powered.
>>>
>>> The new remote lp_advertising checks in pause_advertising_test() would
>>> also be skipped without any notice.
>>
>> Not much we can do if the peer doesn't behave as expected here. Not sure this
>> requires attention.
> 
> I sounds like anything with a BMC is going to hit this. So i doubt you
> will see it with your stmmac boards. Maybe sometime in the future we
> can query the NIC to see if it supports NCSI and change the test
> behaviour to fit?

Maybe the strategy is wrong then, the goal is to check that we're directly
connected to our test peer through a cable, or indirectly through a switch.

Link flapping seemed to me like the most generic solution, but maybe we can
do something entirely different ? Like for example trigger a re-negotiation
with ethtool -r and check for link flap, or maybe more reliably, ask the user
to pass a flag to the selftest command saying if yes or no we're directly
connected ?

Maxime