Hi, I have enabled the TCP/UDP Checksum Offload feature using the enable-tcp-udp-checksum in the startup.conf file. The feature gets enabled and the received locally destined IPV6 UDP/TCP packets at dpdk-input node have the bits corresponding to PKT_RX_IP_CKSUM_GOOD and PKT_RX_L4_CKSUM_GOOD set as part of the MBUF Packet Offload Flags.
What is observed is that, for the locally destined IPv6 TCP/UDP packets being RX'ed, when the packet eventually lands up in the ip6-local node, it tries verifying the checksum status in this node using the vlib_buffer_t flags field(on the basis of the VNET_BUFFER_F_L4_CHECKSUM_CORRECT, VNET_BUFFER_F_OFFLOAD_TCP_CKSUM, VNET_BUFFER_F_OFFLOAD_UDP_CKSUM), and as these particular bits are not set(from the code, the transition of the RX offload checksum status from the DPDK MBUF rx checksum offload bits to VLIB buffer bits is also not evident(currently using VPP 21.01, also checked this on the master branch)), it invokes ip6_tcp_udp_icmp_validate_checksum for the checksum verification which re-computes the L4 checksum. My query is that, when the checksum offload is enabled in the Rx direction, for the locally destined IPv6 packets, is this understanding correct that the L4 checksum would get recomputed in the software again, and if not, is there any other configuration setting available which I may be missing on my end, to avoid this re-computation. Thanks, Pankaj Malhotra
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#20392): https://lists.fd.io/g/vpp-dev/message/20392 Mute This Topic: https://lists.fd.io/mt/86623150/21656 Group Owner: vpp-dev+ow...@lists.fd.io Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com] -=-=-=-=-=-=-=-=-=-=-=-
