On Wed, May 24, 2023 at 8:56 AM Ilya Maximets <[email protected]> wrote: > > On 5/17/23 05:11, Mike Pattrick wrote: > > From: Flavio Leitner <[email protected]> > > > > The netdev receiving packets is supposed to provide the flags > > indicating if the L4 checksum was verified and it is OK or BAD, > > otherwise the stack will check when appropriate by software. > > > > If the packet comes with good checksum, then postpone the > > checksum calculation to the egress device if needed. > > > > When encapsulate a packet with that flag, set the checksum > > of the inner L4 header since that is not yet supported. > > > > Calculate the L4 checksum when the packet is going to be sent > > over a device that doesn't support the feature. > > > > Linux tap devices allows enabling L3 and L4 offload, so this > > patch enables the feature. However, Linux socket interface > > remains disabled because the API doesn't allow enabling > > those two features without enabling TSO too. > > > > Signed-off-by: Flavio Leitner <[email protected]> > > Co-authored-by: Mike Pattrick <[email protected]> > > Signed-off-by: Mike Pattrick <[email protected]> > > > > --- > > Since v9: > > - Extended miniflow_extract changes into avx512 code > > - Formatting changes > > - Note that we cannot currently enable checksum offloading in > > CONFIGURE_VETH_OFFLOADS for check-system-userspace as > > netdev-linux.c currently only parses the vnet header if TSO > > is enabled. > > Since v10: > > - No change > > Since v11: > > - Added AVX512 IPv6 checksum offload support. > > - Improved error messages and logging. > > Since v12: > > - Added missing mutex annotations > > > > Signed-off-by: Mike Pattrick <[email protected]> > > --- > > lib/conntrack.c | 15 +- > > lib/dp-packet.c | 25 ++++ > > lib/dp-packet.h | 78 +++++++++- > > lib/dpif-netdev-extract-avx512.c | 62 +++++++- > > lib/flow.c | 23 +++ > > lib/netdev-dpdk.c | 176 +++++++++++++++------- > > lib/netdev-linux.c | 243 +++++++++++++++++++++---------- > > lib/netdev-native-tnl.c | 32 +--- > > lib/netdev.c | 46 ++---- > > lib/odp-execute-avx512.c | 88 ++++++----- > > lib/packets.c | 175 +++++++++++++++++----- > > lib/packets.h | 3 + > > 12 files changed, 688 insertions(+), 278 deletions(-) > > > > <snip> > > > @@ -1403,7 +1409,6 @@ static int > > netdev_linux_batch_rxq_recv_tap(struct netdev_rxq_linux *rx, int mtu, > > struct dp_packet_batch *batch) > > { > > - int virtio_net_hdr_size; > > ssize_t retval; > > size_t std_len; > > int iovlen; > > @@ -1413,16 +1418,14 @@ netdev_linux_batch_rxq_recv_tap(struct > > netdev_rxq_linux *rx, int mtu, > > /* Use the buffer from the allocated packet below to receive MTU > > * sized packets and an aux_buf for extra TSO data. */ > > iovlen = IOV_TSO_SIZE; > > - virtio_net_hdr_size = sizeof(struct virtio_net_hdr); > > } else { > > /* Use only the buffer from the allocated packet. */ > > iovlen = IOV_STD_SIZE; > > - virtio_net_hdr_size = 0; > > } > > > > /* The length here needs to be accounted in the same way when the > > * aux_buf is allocated so that it can be prepended to TSO buffer. */ > > - std_len = virtio_net_hdr_size + VLAN_ETH_HEADER_LEN + mtu; > > + std_len = sizeof(struct virtio_net_hdr) + VLAN_ETH_HEADER_LEN + mtu; > > for (i = 0; i < NETDEV_MAX_BURST; i++) { > > struct dp_packet *buffer; > > struct dp_packet *pkt; > > @@ -1462,7 +1465,7 @@ netdev_linux_batch_rxq_recv_tap(struct > > netdev_rxq_linux *rx, int mtu, > > pkt = buffer; > > } > > > > - if (virtio_net_hdr_size && netdev_linux_parse_vnet_hdr(pkt)) { > > + if (netdev_linux_parse_vnet_hdr(pkt)) { > > If TUNSETOFFLOAD failed, we will not have a vnet header, right?
The vnet header is linked to TUNSETIFF not TUNSETOFFLOAD. Support for this was added in 2.6.34 (Feb 2010): https://github.com/torvalds/linux/commit/b9fb9ee07e67fce0b7bfd517a48710465706c30a#diff-75b86939050dfaa1707f33042b068bd819e6cd34efae37535046e648a4ecd413R529 That said, I notice the releases FAQ says: "Open vSwitch userspace is not sensitive to the Linux kernel version. It should build against almost any kernel, certainly against 2.6.32 and later." This statement is probably still true, it might be able to build but wouldn't run properly if a userspace tap port was used. I'll update that line in the documentation to reflect this. Thanks, M N.B. For posterity, and because I tracked it down anyways, the support for TUNSETOFFLOAD was added in 3.11 in 2013 https://github.com/torvalds/linux/commit/2be5c76794b0e570aa87b012df5ac864ce668a74 > > > struct netdev *netdev_ = netdev_rxq_get_netdev(&rx->up); > > struct netdev_linux *netdev = netdev_linux_cast(netdev_); > > > > @@ -1611,7 +1614,7 @@ netdev_linux_sock_batch_send(int sock, int ifindex, > > bool tso, int mtu, > > * on other interface types because we attach a socket filter to the rx > > * socket. */ > > static int > > -netdev_linux_tap_batch_send(struct netdev *netdev_, bool tso, int mtu, > > +netdev_linux_tap_batch_send(struct netdev *netdev_, int mtu, > > struct dp_packet_batch *batch) > > { > > struct netdev_linux *netdev = netdev_linux_cast(netdev_); > > @@ -1632,9 +1635,7 @@ netdev_linux_tap_batch_send(struct netdev *netdev_, > > bool tso, int mtu, > > ssize_t retval; > > int error; > > > > - if (tso) { > > - netdev_linux_prepend_vnet_hdr(packet, mtu); > > - } > > + netdev_linux_prepend_vnet_hdr(packet, mtu); > > Is it allowed to add vnet header if TUNSETOFFLOAD failed? > > > > > size = dp_packet_size(packet); > > do { > > @@ -1765,7 +1766,7 @@ netdev_linux_send(struct netdev *netdev_, int qid > > OVS_UNUSED, > > > > error = netdev_linux_sock_batch_send(sock, ifindex, tso, mtu, > > batch); > > } else { > > - error = netdev_linux_tap_batch_send(netdev_, tso, mtu, batch); > > + error = netdev_linux_tap_batch_send(netdev_, mtu, batch); > > } > > if (error) { > > if (error == ENOBUFS) { > _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
