On Tue, Oct 01, 2024 at 11:12:54AM +0200, David Marchand wrote: > Following two upstream Linux kernel changes (see links), the mac address > of a iavf port, serviced by a i40e PF driver, is lost when the DPDK iavf > driver probes the port again (which may be triggered at any point of a > DPDK application life, like when a reset event is triggered by the PF). > > A first change results in the mac address of the VF port being reset to 0 > during the VIRTCHNL_OP_GET_VF_RESOURCES query. > The i40e PF driver change is pretty obscure but the iavf Linux driver does > set VIRTCHNL_VF_OFFLOAD_USO. > Announcing such a capability in the DPDK driver does not seem to be an > issue, so do the same in DPDK to keep the legacy behavior of a fixed mac. > > Then a second change in the kernel results in the VF mac address being > cleared when the VF driver remove its default mac address. > Removing (unicast or multicast) mac addresses is not done by the kernel VF > driver in general. > The reason why the DPDK driver behaves like this is undocumented > (and lost because the authors are not active anymore). > Aligning DPDK behavior to the upstream kernel driver is safer in any > case. > > Cc: sta...@dpdk.org > > Link: > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=fed0d9f13266 > Link: > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=ceb29474bbbc > Signed-off-by: David Marchand <david.march...@redhat.com> > ---
Seems reasonable to me. Acked-by: Bruce Richardson <bruce.richard...@intel.com>
