Re: Stmmac in the netdev CI ?

Hi

+netdev@

On 9/14/26 22:30, Andrew Lunn wrote:

On Mon, Sep 14, 2026 at 10:45:57AM +0200, Maxime Chevallier wrote:

quoted

Hi,

sorry this is off-list, I don't know what's the correct channel for
this kind of discussion, maybe directly in the nipa GH ?

[Bringing this to netdev@, for the context, this is me considering the addition
of stmmac to the netdev CI infra]

quoted

As part of the stmmac maintenance work I've undertaken, I've setup
a local CI for stmmac, allowing to make sure we detect breakages
early on, and on a good variety of HW platforms covering most variants
of the IP.

As an experiment I've plugged that to netdev@, with nightly tests of submitted
series, and ajudsted my infra so that I can fully build + run the whole set
of patches under a few hours.

The next logical step is to plug that to the netdev CI, but I have a few questions
for you :

- I currently have 8 boards with stmmac on them, and 7 more on their way to me, so
  in total I'd be adding around 15 more devices to NIPA, is it OK to add 15 or
  should I somehow try to condense these results ?

Do you have enough interfaces on these boards you can have 4 pairs of
boards with directly connected interfaces?

Most of the boards with a few rare exceptions are single-port, and for now they
are either facing non-stmmac embedded boards, or cheap USB<->Ethernet adapters
plugged to my testing infra. I still have a couple of devices that are reaching
the CI infra through a switch, but not for long.

I can run tests that involve configuring the peer :)

quoted

- The current kselftest don't mean much for stmmac, I don't have any board that has
  RSS, queueing stuff also doesn't really apply, the meaningful ksft are mostly
  csum and timestamping.

Does kselftest at least do the correct thing and declare a skip? I
guess most NICs don't actually have RSS and queue stuff.

It's a mix of both, some gracefully skip, some fail.

quoted

I'd be happy to have stmmac plugged to NIPA though, I think it's quite a good
candidate for an embedded driver to run these, hopefully this could set the example
for other vendors...

I also hope it is something bootlin can turn into a training course,
so more vendors setup their own runners.

It can at least be a good topic for netdevconf / LPC / ELCE :)

Maxime

quoted

quoted

- The current kselftest don't mean much for stmmac, I don't have any board that has
  RSS, queueing stuff also doesn't really apply, the meaningful ksft are mostly
  csum and timestamping.

Does kselftest at least do the correct thing and declare a skip? I
guess most NICs don't actually have RSS and queue stuff.

It's a mix of both, some gracefully skip, some fail.

It would be nice to report those that fail. Maybe the author of the
test will then fix it.

     Andrew