On Mon, 10 Aug 2026 14:30:19 +0000 Cosmin Ratiu wrote: > I'll move the hw gro tests into a dedicated psp_gro.py test for the > next version and increase the assertion strength similar to other HW > GRO tests. > > I found a way to spawn multiple psp_responders to verify cross-stream > assembly errors and other weird edge cases.
No, please follow how the GRO tests work. Can we not pass 2 Rx keys encoded in hex when you spawn the sender, and use AF_PACKET to check the geometry of the received data? > In order to avoid duplicating a lot of psp connection management code, > I hope it's fine to move shared stuff in a psp_lib.py file (similar to > hw/devmem_lib.py). I know you're against frameworkifying stuff, but in > this case I think it's a good choice. > > All of this will push the series to 15 patches. Try to think what the point of the rule is. It's to keep series small _in order to_ make incremental progress, in reliable (form process perspective), comprehensible chunks. You could split the first 9 patches out and merge them separately. Internal metadata reshuffles and decap may be required for your implementations but aren't inherently part of HW-GRO, IMO... No point keeping the first 9 patches hostage to HW-GRO and ksft work.
