On Tue, 17 Jun 2025 18:42:52 +0300 Maayan Kashani <mkash...@nvidia.com> wrote:
> when using --no-numa option, > gro should be initiated with SOCKET_ID_ANY > to allow using cross NUMA. > > API changes notice: > Changed socket_id field in struct rte_gro_param from uint16 to > int32 to support negative value -1 (== SOCKET_ID_ANY). > Changed gro_tbl_create typedef to receive negative socket input. > > Signed-off-by: Maayan Kashani <mkash...@nvidia.com> GRO has experimental in the comments, but not the exported symbol that was a mistake. It either is experimental or not. Given that the tools check the exported symbol table, I would argue it is not experimental now. Most of the API's in ethdev use "unsigned int socket_id", I see no reason for GRO to be different. Why not use function versioning to do it now?