On Fri, Jul 1, 2022 at 5:58 AM Mike Pattrick <[email protected]> wrote: > > From: Flavio Leitner <[email protected]> > > Document the implementation of netdev hardware offloading > in userspace datapath.
netdevs other than dpdk could support RX/TX IP checksum offload etc... And some of those netdevs are virtual. This is a netdev level feature, so I would rename this new file as netdev-offloads.rst. Idem, the term "driver" could be replaced with "netdev". Maybe others have a better idea. There is one typo, see below. [snip] > +Rules > +----- > +1) OVS should strive to forward all packets regardless of checksum. > + > +2) OVS must not correct a bad packet/checksum. > + > +3) Packet with flag DP_PACKET_OL_RX_IP_CSUM_GOOD means that the > + IP checksum is present in the packet and it is good. > + > +4) Packet with flag DP_PACKET_OL_RX_IP_CSUM_BAD means that the > + IP checksum is present in the packet and it is BAD. Extra care > + should be taken to not fix the packet during data path processing. > + > +5) The ingress packet parser can only set DP_PACKET_OL_TX_IP_CSUM > + if the packet has DP_PACKET_OL_RX_L4_CKSUM_GOOD to not violate DP_PACKET_OL_RX_IP_CSUM_GOOD* > + rule #2. > + > +6) Packet with flag DP_PACKET_OL_TX_IPV4 is a IPv4 packet. > + > +7) Packet with flag DP_PACKET_OL_TX_IPV6 is a IPv6 packet. > + > +8) Packet with flag DP_PACKET_OL_TX_IP_CSUM tells the data path > + to skip updating the IP checksum if the packet is modified. The > + IP checksum will be calculated by the egress port if that > + supports IP checksum offload, otherwise the IP checksum will > + be done in software before handing over the packet to the port. > + > +9) When there are modifications to the packet that requires checksum > + update, the data path needs to remove DP_PACKET_OL_RX_IP_CSUM_GOOD > + flag, otherwise the checksum is assumed to be good in the packet. -- David Marchand _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
