On Wed, Oct 08, 2025 at 06:02:56AM -0700, Breno Leitao wrote:
Note that with the create_netdevsim() helper from tools/testing/selftests/net/lib.sh you could create the netdevsim device directly in the target namespace and avoid some duplicate code.
Awesome. I am more than happy to create_netdevsim() in this selftest, and move the others to use it as well.
It would be probably safer to create both rx and tx devices in child namespaces.
Sure, that is doable, but, I need to change a few common helpers, to start netconsole from inside the "tx namespace" instead of the default namespace.
Given all the other netconsole selftest uses TX from the default net namespace, I would like to move them at all the same time.
Do you think it is Ok to have this test using TX interfaces from the main net namespace (as is now), and then I submit a follow patch to migrate all the netcons tests (including this one) to use a TX namespace? Then I can change the helpers at the same time, simplifying the code review.
In fact, I was able to isolate the functions for the isolation in the self test, and now I have a test that relies on create_netdevsim() and have two namespaces, one for TX and one for RX.
The problem is that it hits a bug on netdevsim that doesn't allow it to run properly. Basically create_netdevsim() put the interface up, and I need to get the interface down in order to enslaved it, losing the carrier.
I've propsoed a fix in here:
https://lore.kernel.org/all/20251013-netdevsim_fix-v1-1-357b265dd9d0@debian....
So, I will send a new version for this patchset soon, but, it will probably not pass the CI, given it will not have the fix aboveyet. But I will send it anyway, so, people can further review the patchset.
Thanks --breno