<div dir="ltr"><a href="https://github.com/rodgarrison/reinvent">https://github.com/rodgarrison/reinvent</a> demonstrates ether/UDP chec-kum offload running on AWS which uses virtio. I don't think TCP offload<div>is substantially different. Look here: <a href="https://github.com/rodgarrison/reinvent/blob/main/integration_tests/reinvent_dpdk_udp/reinvent_dpdk_udp_integration_test.cpp#L345">https://github.com/rodgarrison/reinvent/blob/main/integration_tests/reinvent_dpdk_udp/reinvent_dpdk_udp_integration_test.cpp#L345</a> for where the mbuf needs to be setup. But I kind of agree with Stephan, although I haven't tested it, avoiding more sets into mbuf which might involve cache misses it might be faster to do by hand since the data is already in cache.</div></div> <div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Sat, Mar 26, 2022 at 2:59 PM Stephen Hemminger <<a href="mailto:stephen@networkplumber.org">stephen@networkplumber.org</a>> wrote: </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">On Sat, 26 Mar 2022 08:13:21 +0000 "jiangheng (H)" <<a href="mailto:jiangheng12@huawei.com" target="_blank">jiangheng12@huawei.com</a>> wrote: > Hi all, > > I tried using the checksum offloads feature in DPDK and it did not see working under virtual machine. > > Port only support TCP checksum and do not support IP checksum: > rx_offload_capa = DEV_RX_OFFLOAD_TCP_CKSUM > tx_offload_capa = DEV_TX_OFFLOAD_TCP_CKSUM > > so I config rxmode.offload txmode.offloads as below: > rxmode.offloads = DEV_RX_OFFLOAD_TCP_CKSUM > txmode.offloads = DEV_TX_OFFLOAD_TCP_CKSUM > > For TX, I set the following parameters, it works good. > mbuf->l2_len = sizeof(*ethhdr) > mbuf->l3_len = ip header len > mbuf-ol_flags = RTE_MBUF_F_TX_IPV4 | RTE_MBUF_F_TX_TCP_CKSUM Virtio does not support IP checksum offload. Because Virtio passes packets to Linux kernel, and Linux kernel does not do IP checksum offload. The IP checksum is so trivial it is faster for most things to just do it in software; the header is only 20 bytes and it will be in cache. You should always check device capability before enabling an offload. > For RX, It will execute the following code: > In drivers/net/virtio/virtio_rxtx.c virtio_rx_offload function : > if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { > hdrlen = hdr_lens.l2_len + hdr_lens.l3_len + hdr_lens.l4_len; > if (hdr->csum_start <= hdrlen && l4_supported) { > m->ol_flags |= RTE_MBUF_F_RX_L4_CKSUM_NONE; > } else { > > m->ol_flags set to RTE_MBUF_F_RX_L4_CKSUM_NONE, causing the TCP RX checksum failed. > How do I avoid the above code going into this branch? > If you want TCP checksum offload you have to set RTE_ETH_RX_OFFLOAD_TCP_CKSUM in the rxmode when port is configured. This will tell virtio to ask the host to do rx offload. Again, virtio does not do IP checksum offload and you should always query device capability first. </blockquote></div>