On Mon, 2026-03-23 at 18:17 +0100, Sabrina Dubroca wrote: > 2026-03-23, 09:32:43 -0700, Jakub Kicinski wrote: > > On Mon, 23 Mar 2026 16:02:59 +0100 Sabrina Dubroca wrote: > > > 2026-03-23, 14:42:00 +0000, Cosmin Ratiu wrote: > > > > On Mon, 2026-03-23 at 15:28 +0100, Sabrina Dubroca wrote: > > > > > Why not? Being able to test without accessing real HW is > > > > > still > > > > > useful. > > > > > > > > > The tests now send macsec traffic over VLANs and nsim, it's > > > > just that > > > > nsim doesn't deal with VLAN filters at all and there are no > > > > stubbed > > > > vlan filters in debugfs, since real hw doesn't have that > > > > interface. > > > > > > Since netdevsim doesn't deal with VLAN filters at all, the "tests > > > should be written so that they can run both against ``netdevsim`` > > > and > > > a real device" bit of the docs doesn't fully apply here? > > > > > > Anyway, I think the original tests had value, even if they're > > > more > > > limited in some ways than traffic tests. HW/driver behavior could > > > be > > > hiding problems in the stack with VLAN propagation, those simpler > > > tests don't have that risk. > > > > To be clear running the HW test without NETIF= should provide > > similar functionality to what the old tests could do. It's entirely > > okay to add netdevsim-specific subtests/test cases or asserts. > > > > Is there anything specific that you'd like to be tested? > > In v2/v3, nsim was exposing a debugfs file that contained the list of > VLAN filters on that interface, and the selftest was grepping through > that file to check if the correct entry was added/removed after each > operation. I see that as testing the actual propagation of filters, > while the traffic tests check the visible behavior of > stack+driver+HW, > which may not be correlated to actual propagation. > > > Let's not make this about HW tests vs nsim-only tests. > > That was not my intention. But since nsim doesn't currently implement > VLAN filters, it seems running the HW test on nsim doesn't test > anything at all. >
The problem with using the nsim-only VLAN filter debugfs is that it's a test-only interface for figuring out a property of the stack. Real HW doesn't have that interface and thus the tests actually have to generate traffic to ensure VLANs are propagated. The new VLAN tests don't actually ensure anything on nsim, given that it doesn't support VLANs. But on real HW they do - without the last patch, the new tests fail on any hw supporting VLANs and MACsec. Adding back the nsim debugfs file and test assertions guarded by "if cfg._ns" would ensure filter propagation correctness, but would feel non-Pythonic and a little hacky given that the same assertion can't work on real HW. So what would you like to happen? Bring back nsim patches + nsim- specific test assertions? Cosmin.
