Is more efficient than the BTF id approach where all metadata has to beĮxposed for every frame (and selectively consumed by the users) it's not all or nothing users can use small subset of metadata which Unrolled to the same code as if the metadata was pre-populated and the metadata should, in theory, be cheap if used kfuncs should be we avoid BTF complexity the BPF programs themselves are now responsibleįor agreeing on the metadata layout with the AF_XDP consumer I've intentionally kept it small and hacky to see whether the approach is It to export rx_timestamp (plus add rx timestamp to veth driver) rewrite xskxceiver test to use C bpf program and extend Implementation if unrolling is not supported by a particular It to unroll/inline kfuncs kfuncs have the default "safe" at load time, find appropriate ndo_unroll_kfunc and call add new BPF_F_XDP_HAS_METADATA program flag and abuseĪttr->prog_ifindex to pass target device ifindex at load time I've tried to CC most of the people from the original discussion,įeel free to add more if you think I've missed somebody. This is an RFC for the alternative approach suggested by Martin and Magnus Karlsson, Maryam Tahhan, xdp-hints, netdev
Jesper Dangaard Brouer, Anatoly Burakov, Alexander Lobakin, Kpsingh, sdf, haoluo, jolsa, Jakub Kicinski, Willem de Bruijn, ` (5 more replies) 0 siblings, 6 replies 53+ messages in threadįrom: Stanislav Fomichev 20:00 UTC ( / raw)Ĭc: ast, daniel, andrii, martin.lau, song, yhs, john.fastabend, 20:00 ` bpf: Support inlined/unrolled kfuncs for xdp metadata Stanislav Fomichev Xdp: hints via kfuncs All of help / color / mirror / Atom feed * xdp: hints via kfuncs 20:00 Stanislav Fomichev
0 kommentar(er)
