[ovs-dev] [PATCH dpdk-latest v2 4/4] netdev-dpdk: Add TCP Segmentation Offload support

Stokes, Ian ian.stokes at intel.com
Mon Jan 6 20:24:48 UTC 2020



On 12/31/2019 8:14 PM, Flavio Leitner wrote:
> Abbreviated as TSO, TCP Segmentation Offload is a feature which enables
> the network stack to delegate the TCP segmentation to the NIC reducing
> the per packet CPU overhead.
> 
> A guest using vhostuser interface with TSO enabled can send TCP packets
> much bigger than the MTU, which saves CPU cycles normally used to break
> the packets down to MTU size and to calculate checksums.
> 
> It also saves CPU cycles used to parse multiple packets/headers during
> the packet processing inside virtual switch.
> 
> If the destination of the packet is another guest in the same host, then
> the same big packet can be sent through a vhostuser interface skipping
> the segmentation completely. However, if the destination is not local,
> the NIC hardware is instructed to do the TCP segmentation and checksum
> calculation.
> 
> It is recommended to check if NIC hardware supports TSO before enabling
> the feature, which is off by default.

Might be useful to add a link to the DPDK Networking Device feature 
support as it lists whether TSO is supported for a given device currently

https://doc.dpdk.org/guides/nics/overview.html

> 
> Signed-off-by: Flavio Leitner <fbl at sysclose.org>
> ---
>   Documentation/automake.mk           |   1 +
>   Documentation/topics/dpdk/index.rst |   1 +
>   Documentation/topics/dpdk/tso.rst   |  89 ++++++++
>   NEWS                                |   1 +
>   lib/automake.mk                     |   2 +
>   lib/conntrack.c                     |  29 ++-
>   lib/dp-packet.h                     | 152 +++++++++++++-
>   lib/ipf.c                           |  32 +--
>   lib/netdev-dpdk.c                   | 311 ++++++++++++++++++++++++----
>   lib/netdev-linux-private.h          |   4 +
>   lib/netdev-linux.c                  | 295 +++++++++++++++++++++++---
>   lib/netdev-provider.h               |  10 +
>   lib/netdev.c                        |  52 ++++-
>   lib/tso.c                           |  54 +++++
>   lib/tso.h                           |  23 ++
>   vswitchd/bridge.c                   |   2 +
>   vswitchd/vswitch.xml                |  12 ++
>   17 files changed, 982 insertions(+), 88 deletions(-)
>   create mode 100644 Documentation/topics/dpdk/tso.rst
>   create mode 100644 lib/tso.c
>   create mode 100644 lib/tso.h
> 
> diff --git a/Documentation/automake.mk b/Documentation/automake.mk
> index 5f7c3e07b..abe5aaed1 100644
> --- a/Documentation/automake.mk
> +++ b/Documentation/automake.mk
> @@ -35,6 +35,7 @@ DOC_SOURCE = \
>   	Documentation/topics/dpdk/index.rst \
>   	Documentation/topics/dpdk/bridge.rst \
>   	Documentation/topics/dpdk/jumbo-frames.rst \
> +	Documentation/topics/dpdk/tso.rst \
>   	Documentation/topics/dpdk/memory.rst \
>   	Documentation/topics/dpdk/pdump.rst \
>   	Documentation/topics/dpdk/phy.rst \
> diff --git a/Documentation/topics/dpdk/index.rst b/Documentation/topics/dpdk/index.rst
> index f2862ea70..400d56051 100644
> --- a/Documentation/topics/dpdk/index.rst
> +++ b/Documentation/topics/dpdk/index.rst
> @@ -40,4 +40,5 @@ DPDK Support
>      /topics/dpdk/qos
>      /topics/dpdk/pdump
>      /topics/dpdk/jumbo-frames
> +   /topics/dpdk/tso
>      /topics/dpdk/memory
> diff --git a/Documentation/topics/dpdk/tso.rst b/Documentation/topics/dpdk/tso.rst
> new file mode 100644
> index 000000000..0724513bd
> --- /dev/null
> +++ b/Documentation/topics/dpdk/tso.rst
> @@ -0,0 +1,89 @@
> +..
> +      Copyright 2019, Red Hat, Inc.
Minor but probably 2020 above now for the next revision, goes for the 
other new files added also.

> +
> +      Licensed under the Apache License, Version 2.0 (the "License"); you may
> +      not use this file except in compliance with the License. You may obtain
> +      a copy of the License at
> +
> +          http://www.apache.org/licenses/LICENSE-2.0
> +
> +      Unless required by applicable law or agreed to in writing, software
> +      distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
> +      WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
> +      License for the specific language governing permissions and limitations
> +      under the License.
> +
> +      Convention for heading levels in Open vSwitch documentation:
> +
> +      =======  Heading 0 (reserved for the title in a document)
> +      -------  Heading 1
> +      ~~~~~~~  Heading 2
> +      +++++++  Heading 3
> +      '''''''  Heading 4
> +
> +      Avoid deeper levels because they do not render well.
> +
> +========================
> +Userspace Datapath - TSO
> +========================
> +
> +**Note:** This feature is considered experimental.
> +
> +TCP Segmentation Offload (TSO) enables a network stack to delegate segmentation
> +of an oversized TCP segment to the underlying physical NIC. Offload of frame
> +segmentation achieves computational savings in the core, freeing up CPU cycles
> +for more useful work.
> +
> +A common use case for TSO is when using virtualization, where traffic that's
> +coming in from a VM can offload the TCP segmentation, thus avoiding the
> +fragmentation in software. Additionally, if the traffic is headed to a VM
> +within the same host further optimization can be expected. As the traffic never
> +leaves the machine, no MTU needs to be accounted for, and thus no segmentation
> +and checksum calculations are required, which saves yet more cycles. Only when
> +the traffic actually leaves the host the segmentation needs to happen, in which
> +case it will be performed by the egress NIC. Consult your controller's
> +datasheet for compatibility. Secondly, the NIC must have an associated DPDK
> +Poll Mode Driver (PMD) which supports `TSO`.
> +

Possibly a better place for the link I posted earlier would be here.

> +Enabling TSO
> +~~~~~~~~~~~~
> +
> +The TSO support may be enabled via a global config value ``tso-support``.
> +Setting this to ``true`` enables TSO support for all ports.
> +
> +    $ ovs-vsctl set Open_vSwitch . other_config:tso-support=true
> +
> +The default value is ``false``.
> +
> +When using :doc:`vHost User ports <vhost-user>`, TSO may be enabled as follows.
> +
> +`TSO` is enabled in OvS by the DPDK vHost User backend; when a new guest
> +connection is established, `TSO` is thus advertised to the guest as an
> +available feature:
> +
> +QEMU Command Line Parameter::
> +
> +    $ sudo $QEMU_DIR/x86_64-softmmu/qemu-system-x86_64 \
> +    ...
> +    -device virtio-net-pci,mac=00:00:00:00:00:01,netdev=mynet1,\
> +    csum=on,guest_csum=on,guest_tso4=on,guest_tso6=on\
> +    ...
> +
> +2. Ethtool. Assuming that the guest's OS also supports `TSO`, ethtool can be
> +used to enable same::
> +
> +    $ ethtool -K eth0 sg on     # scatter-gather is a prerequisite for TSO
> +    $ ethtool -K eth0 tso on
> +    $ ethtool -k eth0
> +
> +~~~~~~~~~~~
> +Limitations
> +~~~~~~~~~~~
Minor: white space needed here to separate heading from text?
> +The current OvS userspace `TSO` implementation supports flat and VLAN networks
> +only (i.e. no support for `TSO` over tunneled connection [VxLAN, GRE, IPinIP,
> +etc.]).
> +
> +There is no software implementation of TSO, so all ports attached to the
> +datapath must support TSO or packets using that feature will be dropped.
> +That also means guests using vhost-user in client mode will receive TSO
> +packet regardless of TSO being enabled or disabled within the guest.
> diff --git a/NEWS b/NEWS
> index 0d65d5a7f..df32930bf 100644
> --- a/NEWS
> +++ b/NEWS
> @@ -14,6 +14,7 @@ Post-v2.12.0
>        * DPDK pdump packet capture support disabled by default. New configure
>          option '--enable-dpdk-pdump' to enable it.
>        * DPDK pdump support is deprecated and will be removed in next releases.
> +     * Add experimental support for TSO in vhost-user ports.
Is it just for vhost-user ports though? If supported by the NIC, TSO 
will be handled there as well right and this work enabled that case also?

>   
>   v2.12.0 - 03 Sep 2019
>   ---------------------
> diff --git a/lib/automake.mk b/lib/automake.mk
> index 17b36b43d..01c54d7f3 100644
> --- a/lib/automake.mk
> +++ b/lib/automake.mk
> @@ -302,6 +302,8 @@ lib_libopenvswitch_la_SOURCES = \
>   	lib/tnl-neigh-cache.h \
>   	lib/tnl-ports.c \
>   	lib/tnl-ports.h \
> +	lib/tso.c \
> +	lib/tso.h \
>   	lib/netdev-native-tnl.c \
>   	lib/netdev-native-tnl.h \
>   	lib/token-bucket.c \
> diff --git a/lib/conntrack.c b/lib/conntrack.c
> index df7b9fa7a..188f58dd8 100644
> --- a/lib/conntrack.c
> +++ b/lib/conntrack.c
> @@ -1885,7 +1885,8 @@ conn_key_extract(struct conntrack *ct, struct dp_packet *pkt, ovs_be16 dl_type,
>           if (hwol_bad_l3_csum) {
>               ok = false;
>           } else {
> -            bool hwol_good_l3_csum = dp_packet_ip_checksum_valid(pkt);
> +            bool hwol_good_l3_csum = dp_packet_ip_checksum_valid(pkt)
> +                                     || dp_packet_hwol_tx_ip_checksum(pkt);
>               /* Validate the checksum only when hwol is not supported. */
>               ok = extract_l3_ipv4(&ctx->key, l3, dp_packet_l3_size(pkt), NULL,
>                                    !hwol_good_l3_csum);
> @@ -1899,7 +1900,8 @@ conn_key_extract(struct conntrack *ct, struct dp_packet *pkt, ovs_be16 dl_type,
>       if (ok) {
>           bool hwol_bad_l4_csum = dp_packet_l4_checksum_bad(pkt);
>           if (!hwol_bad_l4_csum) {
> -            bool  hwol_good_l4_csum = dp_packet_l4_checksum_valid(pkt);
> +            bool  hwol_good_l4_csum = dp_packet_l4_checksum_valid(pkt)
> +                                      || dp_packet_hwol_tx_l4_checksum(pkt);
>               /* Validate the checksum only when hwol is not supported. */
>               if (extract_l4(&ctx->key, l4, dp_packet_l4_size(pkt),
>                              &ctx->icmp_related, l3, !hwol_good_l4_csum,
> @@ -3100,8 +3102,11 @@ handle_ftp_ctl(struct conntrack *ct, const struct conn_lookup_ctx *ctx,
>                   }
>                   if (seq_skew) {
>                       ip_len = ntohs(l3_hdr->ip_tot_len) + seq_skew;
> -                    l3_hdr->ip_csum = recalc_csum16(l3_hdr->ip_csum,
> -                                          l3_hdr->ip_tot_len, htons(ip_len));
> +                    if (!dp_packet_hwol_tx_ip_checksum(pkt)) {
> +                        l3_hdr->ip_csum = recalc_csum16(l3_hdr->ip_csum,
> +                                                        l3_hdr->ip_tot_len,
> +                                                        htons(ip_len));
> +                    }
>                       l3_hdr->ip_tot_len = htons(ip_len);
>                   }
>               }
> @@ -3119,13 +3124,15 @@ handle_ftp_ctl(struct conntrack *ct, const struct conn_lookup_ctx *ctx,
>       }
>   
>       th->tcp_csum = 0;
> -    if (ctx->key.dl_type == htons(ETH_TYPE_IPV6)) {
> -        th->tcp_csum = packet_csum_upperlayer6(nh6, th, ctx->key.nw_proto,
> -                           dp_packet_l4_size(pkt));
> -    } else {
> -        uint32_t tcp_csum = packet_csum_pseudoheader(l3_hdr);
> -        th->tcp_csum = csum_finish(
> -             csum_continue(tcp_csum, th, dp_packet_l4_size(pkt)));
> +    if (!dp_packet_hwol_tx_l4_checksum(pkt)) {
> +        if (ctx->key.dl_type == htons(ETH_TYPE_IPV6)) {
> +            th->tcp_csum = packet_csum_upperlayer6(nh6, th, ctx->key.nw_proto,
> +                               dp_packet_l4_size(pkt));
> +        } else {
> +            uint32_t tcp_csum = packet_csum_pseudoheader(l3_hdr);
> +            th->tcp_csum = csum_finish(
> +                 csum_continue(tcp_csum, th, dp_packet_l4_size(pkt)));
> +        }
>       }
>   
>       if (seq_skew) {
> diff --git a/lib/dp-packet.h b/lib/dp-packet.h
> index 325924eaa..5cd6fcc68 100644
> --- a/lib/dp-packet.h
> +++ b/lib/dp-packet.h
> @@ -109,6 +109,8 @@ static inline void dp_packet_set_size(struct dp_packet *, uint32_t);
>   static inline uint16_t dp_packet_get_allocated(const struct dp_packet *);
>   static inline void dp_packet_set_allocated(struct dp_packet *, uint16_t);
>   
> +void dp_packet_prepend_vnet_hdr(struct dp_packet *, int mtu);
> +
>   void *dp_packet_resize_l2(struct dp_packet *, int increment);
>   void *dp_packet_resize_l2_5(struct dp_packet *, int increment);
>   static inline void *dp_packet_eth(const struct dp_packet *);
> @@ -451,7 +453,7 @@ dp_packet_init_specific(struct dp_packet *p)
>   {
>       /* This initialization is needed for packets that do not come from DPDK
>        * interfaces, when vswitchd is built with --with-dpdk. */
> -    p->mbuf.tx_offload = p->mbuf.packet_type = 0;
> +    p->mbuf.ol_flags = p->mbuf.tx_offload = p->mbuf.packet_type = 0;
>       p->mbuf.nb_segs = 1;
>       p->mbuf.next = NULL;
>   }
> @@ -514,6 +516,80 @@ dp_packet_set_allocated(struct dp_packet *b, uint16_t s)
>       b->mbuf.buf_len = s;
>   }
>   
> +static inline bool
> +dp_packet_hwol_is_tso(const struct dp_packet *b)
> +{
> +    return (b->mbuf.ol_flags & (PKT_TX_TCP_SEG | PKT_TX_L4_MASK))
> +           ? true
> +           : false;
> +}
> +
> +static inline bool
> +dp_packet_hwol_is_ipv4(const struct dp_packet *b)
> +{
> +    return b->mbuf.ol_flags & PKT_TX_IPV4 ? true : false;
> +}
> +
> +static inline uint64_t
> +dp_packet_hwol_l4_mask(const struct dp_packet *b)
> +{
> +    return b->mbuf.ol_flags & PKT_TX_L4_MASK;
> +}
> +
> +static inline bool
> +dp_packet_hwol_l4_is_tcp(const struct dp_packet *b)
> +{
> +    return (b->mbuf.ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM
> +           ? true
> +           : false;
> +}
> +
> +static inline bool
> +dp_packet_hwol_l4_is_udp(struct dp_packet *b)
> +{
> +    return (b->mbuf.ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM
> +           ? true
> +           : false;
> +}
> +
> +static inline bool
> +dp_packet_hwol_l4_is_sctp(struct dp_packet *b)
> +{
> +    return (b->mbuf.ol_flags & PKT_TX_L4_MASK) == PKT_TX_SCTP_CKSUM
> +           ? true
> +           : false;
> +}
> +
> +static inline void
> +dp_packet_hwol_set_tx_ipv4(struct dp_packet *b) {
> +    b->mbuf.ol_flags |= PKT_TX_IPV4;
> +}
> +
> +static inline void
> +dp_packet_hwol_set_tx_ipv6(struct dp_packet *b) {
> +    b->mbuf.ol_flags |= PKT_TX_IPV6;
> +}
> +
> +static inline void
> +dp_packet_hwol_set_csum_tcp(struct dp_packet *b) {
> +    b->mbuf.ol_flags |= PKT_TX_TCP_CKSUM;
> +}
> +
> +static inline void
> +dp_packet_hwol_set_csum_udp(struct dp_packet *b) {
> +    b->mbuf.ol_flags |= PKT_TX_UDP_CKSUM;
> +}
> +
> +static inline void
> +dp_packet_hwol_set_csum_sctp(struct dp_packet *b) {
> +    b->mbuf.ol_flags |= PKT_TX_SCTP_CKSUM;
> +}
> +
> +static inline void
> +dp_packet_hwol_set_tcp_seg(struct dp_packet *b) {
> +    b->mbuf.ol_flags |= PKT_TX_TCP_SEG;
> +}
> +
>   /* Returns the RSS hash of the packet 'p'.  Note that the returned value is
>    * correct only if 'dp_packet_rss_valid(p)' returns true */
>   static inline uint32_t
> @@ -643,6 +719,66 @@ dp_packet_set_allocated(struct dp_packet *b, uint16_t s)
>       b->allocated_ = s;
>   }
>   
> +static inline bool
> +dp_packet_hwol_is_tso(const struct dp_packet *b OVS_UNUSED)
> +{
> +    return false;
> +}
> +
> +static inline bool
> +dp_packet_hwol_is_ipv4(const struct dp_packet *b OVS_UNUSED)
> +{
> +    return false;
> +}
> +
> +static inline uint64_t
> +dp_packet_hwol_l4_mask(const struct dp_packet *b OVS_UNUSED)
> +{
> +    return 0;
> +}
> +
> +static inline bool
> +dp_packet_hwol_l4_is_tcp(const struct dp_packet *b OVS_UNUSED)
> +{
> +    return false;
> +}
> +
> +static inline bool
> +dp_packet_hwol_l4_is_udp(const struct dp_packet *b OVS_UNUSED)
> +{
> +    return false;
> +}
> +
> +static inline bool
> +dp_packet_hwol_l4_is_sctp(const struct dp_packet *b OVS_UNUSED)
> +{
> +    return false;
> +}
> +
> +static inline void
> +dp_packet_hwol_set_tx_ipv4(struct dp_packet *b OVS_UNUSED) {
> +}
> +
> +static inline void
> +dp_packet_hwol_set_tx_ipv6(struct dp_packet *b OVS_UNUSED) {
> +}
> +
> +static inline void
> +dp_packet_hwol_set_csum_tcp(struct dp_packet *b OVS_UNUSED) {
> +}
> +
> +static inline void
> +dp_packet_hwol_set_csum_udp(struct dp_packet *b OVS_UNUSED) {
> +}
> +
> +static inline void
> +dp_packet_hwol_set_csum_sctp(struct dp_packet *b OVS_UNUSED) {
> +}
> +
> +static inline void
> +dp_packet_hwol_set_tcp_seg(struct dp_packet *b OVS_UNUSED) {
> +}
> +
>   /* Returns the RSS hash of the packet 'p'.  Note that the returned value is
>    * correct only if 'dp_packet_rss_valid(p)' returns true */
>   static inline uint32_t
> @@ -934,6 +1070,20 @@ dp_packet_batch_reset_cutlen(struct dp_packet_batch *batch)
>       }
>   }
>   
> +static inline bool
> +dp_packet_hwol_tx_ip_checksum(const struct dp_packet *p)
> +{
> +
> +    return dp_packet_hwol_l4_mask(p) ? true : false;
> +}
> +
> +static inline bool
> +dp_packet_hwol_tx_l4_checksum(const struct dp_packet *p)
> +{
> +
> +    return dp_packet_hwol_l4_mask(p) ? true : false;
> +}
> +
>   #ifdef  __cplusplus
>   }
>   #endif
> diff --git a/lib/ipf.c b/lib/ipf.c
> index 4cc0f2df6..052867d90 100644
> --- a/lib/ipf.c
> +++ b/lib/ipf.c
> @@ -433,9 +433,11 @@ ipf_reassemble_v4_frags(struct ipf_list *ipf_list)
>       len += rest_len;
>       l3 = dp_packet_l3(pkt);
>       ovs_be16 new_ip_frag_off = l3->ip_frag_off & ~htons(IP_MORE_FRAGMENTS);
> -    l3->ip_csum = recalc_csum16(l3->ip_csum, l3->ip_frag_off,
> -                                new_ip_frag_off);
> -    l3->ip_csum = recalc_csum16(l3->ip_csum, l3->ip_tot_len, htons(len));
> +    if (!dp_packet_hwol_tx_ip_checksum(pkt)) {
> +        l3->ip_csum = recalc_csum16(l3->ip_csum, l3->ip_frag_off,
> +                                    new_ip_frag_off);
> +        l3->ip_csum = recalc_csum16(l3->ip_csum, l3->ip_tot_len, htons(len));
> +    }
>       l3->ip_tot_len = htons(len);
>       l3->ip_frag_off = new_ip_frag_off;
>       dp_packet_set_l2_pad_size(pkt, 0);
> @@ -606,6 +608,7 @@ ipf_is_valid_v4_frag(struct ipf *ipf, struct dp_packet *pkt)
>       }
>   
>       if (OVS_UNLIKELY(!dp_packet_ip_checksum_valid(pkt)
> +                     && !dp_packet_hwol_tx_ip_checksum(pkt)
>                        && csum(l3, ip_hdr_len) != 0)) {
>           goto invalid_pkt;
>       }
> @@ -1180,16 +1183,21 @@ ipf_post_execute_reass_pkts(struct ipf *ipf,
>                   } else {
>                       struct ip_header *l3_frag = dp_packet_l3(frag_0->pkt);
>                       struct ip_header *l3_reass = dp_packet_l3(pkt);
> -                    ovs_be32 reass_ip = get_16aligned_be32(&l3_reass->ip_src);
> -                    ovs_be32 frag_ip = get_16aligned_be32(&l3_frag->ip_src);
> -                    l3_frag->ip_csum = recalc_csum32(l3_frag->ip_csum,
> -                                                     frag_ip, reass_ip);
> -                    l3_frag->ip_src = l3_reass->ip_src;
> +                    if (!dp_packet_hwol_tx_ip_checksum(frag_0->pkt)) {
> +                        ovs_be32 reass_ip =
> +                            get_16aligned_be32(&l3_reass->ip_src);
> +                        ovs_be32 frag_ip =
> +                            get_16aligned_be32(&l3_frag->ip_src);
> +
> +                        l3_frag->ip_csum = recalc_csum32(l3_frag->ip_csum,
> +                                                         frag_ip, reass_ip);
> +                        reass_ip = get_16aligned_be32(&l3_reass->ip_dst);
> +                        frag_ip = get_16aligned_be32(&l3_frag->ip_dst);
> +                        l3_frag->ip_csum = recalc_csum32(l3_frag->ip_csum,
> +                                                         frag_ip, reass_ip);
> +                    }
>   
> -                    reass_ip = get_16aligned_be32(&l3_reass->ip_dst);
> -                    frag_ip = get_16aligned_be32(&l3_frag->ip_dst);
> -                    l3_frag->ip_csum = recalc_csum32(l3_frag->ip_csum,
> -                                                     frag_ip, reass_ip);
> +                    l3_frag->ip_src = l3_reass->ip_src;
>                       l3_frag->ip_dst = l3_reass->ip_dst;
>                   }
>   
> diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c
> index 57bff5e58..9f4360ae9 100644
> --- a/lib/netdev-dpdk.c
> +++ b/lib/netdev-dpdk.c
> @@ -63,6 +63,7 @@
>   #include "smap.h"
>   #include "sset.h"
>   #include "timeval.h"
> +#include "tso.h"
>   #include "unaligned.h"
>   #include "unixctl.h"
>   #include "util.h"
> @@ -355,7 +356,8 @@ struct ingress_policer {
>   enum dpdk_hw_ol_features {
>       NETDEV_RX_CHECKSUM_OFFLOAD = 1 << 0,
>       NETDEV_RX_HW_CRC_STRIP = 1 << 1,
> -    NETDEV_RX_HW_SCATTER = 1 << 2
> +    NETDEV_RX_HW_SCATTER = 1 << 2,
> +    NETDEV_TX_TSO_OFFLOAD = 1 << 3,
>   };
>   
>   /*
> @@ -935,6 +937,12 @@ dpdk_eth_dev_port_config(struct netdev_dpdk *dev, int n_rxq, int n_txq)
>           conf.rxmode.offloads |= DEV_RX_OFFLOAD_KEEP_CRC;
>       }
>   
> +    if (dev->hw_ol_features & NETDEV_TX_TSO_OFFLOAD) {
> +        conf.txmode.offloads |= DEV_TX_OFFLOAD_TCP_TSO;
> +        conf.txmode.offloads |= DEV_TX_OFFLOAD_TCP_CKSUM;
> +        conf.txmode.offloads |= DEV_TX_OFFLOAD_IPV4_CKSUM;
> +    }
> +
>       /* Limit configured rss hash functions to only those supported
>        * by the eth device. */
>       conf.rx_adv_conf.rss_conf.rss_hf &= info.flow_type_rss_offloads;
> @@ -1036,6 +1044,9 @@ dpdk_eth_dev_init(struct netdev_dpdk *dev)
>       uint32_t rx_chksm_offload_capa = DEV_RX_OFFLOAD_UDP_CKSUM |
>                                        DEV_RX_OFFLOAD_TCP_CKSUM |
>                                        DEV_RX_OFFLOAD_IPV4_CKSUM;
> +    uint32_t tx_tso_offload_capa = DEV_TX_OFFLOAD_TCP_TSO |
> +                                   DEV_TX_OFFLOAD_TCP_CKSUM |
> +                                   DEV_TX_OFFLOAD_IPV4_CKSUM;
>   
>       rte_eth_dev_info_get(dev->port_id, &info);
>   
> @@ -1062,6 +1073,14 @@ dpdk_eth_dev_init(struct netdev_dpdk *dev)
>           dev->hw_ol_features &= ~NETDEV_RX_HW_SCATTER;
>       }
>   
> +    if (info.tx_offload_capa & tx_tso_offload_capa) {
> +        dev->hw_ol_features |= NETDEV_TX_TSO_OFFLOAD;
> +    } else {
> +        dev->hw_ol_features &= ~NETDEV_TX_TSO_OFFLOAD;
> +        VLOG_WARN("Tx TSO offload is not supported on port "
> +                  DPDK_PORT_ID_FMT, dev->port_id);

To make this warning more verbose we could include the interface name as 
well as the port ID?

> +    }
> +
>       n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq);
>       n_txq = MIN(info.max_tx_queues, dev->up.n_txq);
>   
> @@ -1310,14 +1329,16 @@ netdev_dpdk_vhost_construct(struct netdev *netdev)
>           goto out;
>       }
>   
> -    err = rte_vhost_driver_disable_features(dev->vhost_id,
> -                                1ULL << VIRTIO_NET_F_HOST_TSO4
> -                                | 1ULL << VIRTIO_NET_F_HOST_TSO6
> -                                | 1ULL << VIRTIO_NET_F_CSUM);
> -    if (err) {
> -        VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user "
> -                 "port: %s\n", name);
> -        goto out;
> +    if (!tso_enabled()) {
> +        err = rte_vhost_driver_disable_features(dev->vhost_id,
> +                                    1ULL << VIRTIO_NET_F_HOST_TSO4
> +                                    | 1ULL << VIRTIO_NET_F_HOST_TSO6
> +                                    | 1ULL << VIRTIO_NET_F_CSUM);
> +        if (err) {
> +            VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user "
> +                     "port: %s\n", name);
> +            goto out;
> +        }
>       }
>   
>       err = rte_vhost_driver_start(dev->vhost_id);
> @@ -1652,6 +1673,11 @@ netdev_dpdk_get_config(const struct netdev *netdev, struct smap *args)
>           } else {
>               smap_add(args, "rx_csum_offload", "false");
>           }
> +        if (dev->hw_ol_features & NETDEV_TX_TSO_OFFLOAD) {
> +            smap_add(args, "tx_tso_offload", "true");
> +        } else {
> +            smap_add(args, "tx_tso_offload", "false");
> +        }
>           smap_add(args, "lsc_interrupt_mode",
>                    dev->lsc_interrupt_mode ? "true" : "false");
>       }
> @@ -2051,6 +2077,64 @@ netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq)
>       rte_free(rx);
>   }
>   
> +/* Prepare the packet for HWOL.
> + * Return True if the packet is OK to continue. */
> +static bool
> +netdev_dpdk_prep_hwol_packet(struct netdev_dpdk *dev, struct rte_mbuf *mbuf)
> +{
> +    struct dp_packet *pkt = CONTAINER_OF(mbuf, struct dp_packet, mbuf);
> +
> +    if (mbuf->ol_flags & PKT_TX_L4_MASK) {
> +        mbuf->l2_len = (char *)dp_packet_l3(pkt) - (char *)dp_packet_eth(pkt);
> +        mbuf->l3_len = (char *)dp_packet_l4(pkt) - (char *)dp_packet_l3(pkt);
> +        mbuf->outer_l2_len = 0;
> +        mbuf->outer_l3_len = 0;
> +    }
> +
> +    if (mbuf->ol_flags & PKT_TX_TCP_SEG) {
> +        struct tcp_header *th = dp_packet_l4(pkt);
> +
> +        if (!th) {
> +            VLOG_WARN_RL(&rl, "%s: TCP Segmentation without L4 header"
> +                         " pkt len: %"PRIu32"", dev->up.name, mbuf->pkt_len);
> +            return false;
> +        }
> +
> +        mbuf->l4_len = TCP_OFFSET(th->tcp_ctl) * 4;
Can we replace the use of magic number 4 above?

> +        mbuf->ol_flags |= PKT_TX_TCP_CKSUM | PKT_TX_IP_CKSUM;
> +        mbuf->tso_segsz = dev->mtu - mbuf->l3_len - mbuf->l4_len;
> +    }
> +
> +    return true;
> +}
> +
> +/* Prepare a batch for HWOL.
> + * Return the number of good packets in the batch. */
> +static int
> +netdev_dpdk_prep_hwol_batch(struct netdev_dpdk *dev, struct rte_mbuf **pkts,
> +                            int pkt_cnt)
> +{
> +    int i = 0;
> +    int cnt = 0;
> +    struct rte_mbuf *pkt;
> +
> +    /* Prepare and filter bad HWOL packets */

Minor, missing period for comment above.

> +    for (i = 0; i < pkt_cnt; i++) {
> +        pkt = pkts[i];
> +        if (!netdev_dpdk_prep_hwol_packet(dev, pkt)) {
> +            rte_pktmbuf_free(pkt);
> +            continue;
> +        }
> +
> +        if (OVS_UNLIKELY(i != cnt)) {
> +            pkts[cnt] = pkt;
> +        }
> +        cnt++;
> +    }
> +
> +    return cnt;
> +}
> +
>   /* Tries to transmit 'pkts' to txq 'qid' of device 'dev'.  Takes ownership of
>    * 'pkts', even in case of failure.
>    *
> @@ -2060,11 +2144,22 @@ netdev_dpdk_eth_tx_burst(struct netdev_dpdk *dev, int qid,
>                            struct rte_mbuf **pkts, int cnt)
>   {
>       uint32_t nb_tx = 0;
> +    uint16_t nb_tx_prep = cnt;
> +
> +    if (tso_enabled()) {
> +        nb_tx_prep = rte_eth_tx_prepare(dev->port_id, qid, pkts, cnt);
> +        if (nb_tx_prep != cnt) {
> +            VLOG_WARN_RL(&rl, "%s: Output batch contains invalid packets. "
> +                         "Only %u/%u are valid: %s", dev->up.name, nb_tx_prep,
> +                         cnt, rte_strerror(rte_errno));
> +        }
> +    }
>   
> -    while (nb_tx != cnt) {
> +    while (nb_tx != nb_tx_prep) {
>           uint32_t ret;
>   
> -        ret = rte_eth_tx_burst(dev->port_id, qid, pkts + nb_tx, cnt - nb_tx);
> +        ret = rte_eth_tx_burst(dev->port_id, qid, pkts + nb_tx,
> +                               nb_tx_prep - nb_tx);
>           if (!ret) {
>               break;
>           }
> @@ -2348,11 +2443,14 @@ netdev_dpdk_filter_packet_len(struct netdev_dpdk *dev, struct rte_mbuf **pkts,
>       int cnt = 0;
>       struct rte_mbuf *pkt;
>   
> +    /* Filter oversized packets, unless are marked for TSO. */
>       for (i = 0; i < pkt_cnt; i++) {
>           pkt = pkts[i];
> -        if (OVS_UNLIKELY(pkt->pkt_len > dev->max_packet_len)) {
> -            VLOG_WARN_RL(&rl, "%s: Too big size %" PRIu32 " max_packet_len %d",
> -                         dev->up.name, pkt->pkt_len, dev->max_packet_len);
> +        if (OVS_UNLIKELY((pkt->pkt_len > dev->max_packet_len)
> +            && !(pkt->ol_flags & PKT_TX_TCP_SEG))) {
> +            VLOG_WARN_RL(&rl, "%s: Too big size %" PRIu32 " "
> +                         "max_packet_len %d", dev->up.name, pkt->pkt_len,
> +                         dev->max_packet_len);
>               rte_pktmbuf_free(pkt);
>               continue;
>           }
> @@ -2401,7 +2499,7 @@ __netdev_dpdk_vhost_send(struct netdev *netdev, int qid,
>       struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts;
>       struct netdev_dpdk_sw_stats sw_stats_add;
>       unsigned int n_packets_to_free = cnt;
> -    unsigned int total_packets = cnt;
> +    unsigned int total_packets;
>       int i, retries = 0;
>       int max_retries = VHOST_ENQ_RETRY_MIN;
>       int vid = netdev_dpdk_get_vid(dev);
> @@ -2421,7 +2519,8 @@ __netdev_dpdk_vhost_send(struct netdev *netdev, int qid,
>           rte_spinlock_lock(&dev->tx_q[qid].tx_lock);
>       }
>   
> -    cnt = netdev_dpdk_filter_packet_len(dev, cur_pkts, cnt);
> +    total_packets = netdev_dpdk_prep_hwol_batch(dev, cur_pkts, cnt);
> +    cnt = netdev_dpdk_filter_packet_len(dev, cur_pkts, total_packets);
>       sw_stats_add.tx_mtu_exceeded_drops = total_packets - cnt;
>   
>       /* Check has QoS has been configured for the netdev */
> @@ -2470,6 +2569,123 @@ out:
>       }
>   }
>   
> +static void
> +netdev_dpdk_extbuf_free(void *addr OVS_UNUSED, void *opaque)
> +{
> +    rte_free(opaque);
> +}
> +
> +static struct rte_mbuf *
> +dpdk_pktmbuf_attach_extbuf(struct rte_mbuf *pkt, uint32_t data_len)
> +{
> +    uint32_t total_len = RTE_PKTMBUF_HEADROOM + data_len;
> +    struct rte_mbuf_ext_shared_info *shinfo = NULL;
> +    uint16_t buf_len;
> +    void *buf;
> +
> +    if (rte_pktmbuf_tailroom(pkt) >= sizeof(*shinfo)) {
> +        shinfo = rte_pktmbuf_mtod(pkt, struct rte_mbuf_ext_shared_info *);
> +    } else {
> +        total_len += sizeof(*shinfo) + sizeof(uintptr_t);
> +        total_len = RTE_ALIGN_CEIL(total_len, sizeof(uintptr_t));
> +    }
> +
> +    if (unlikely(total_len > UINT16_MAX)) {
> +        VLOG_ERR("Can't copy packet: too big %u", total_len);
> +        return NULL;
> +    }
> +
> +    buf_len = total_len;
> +    buf = rte_malloc(NULL, buf_len, RTE_CACHE_LINE_SIZE);
> +    if (unlikely(buf == NULL)) {
> +        VLOG_ERR("Failed to allocate mem using rte_malloc: %u", buf_len);
> +        return NULL;
> +    }
> +
> +    /* Initialize shinfo */
> +    if (shinfo) {
> +        shinfo->free_cb = netdev_dpdk_extbuf_free;
> +        shinfo->fcb_opaque = buf;
> +        rte_mbuf_ext_refcnt_set(shinfo, 1);
> +    } else {
> +        shinfo = rte_pktmbuf_ext_shinfo_init_helper(buf, &buf_len,
> +                                                    netdev_dpdk_extbuf_free,
> +                                                    buf);
> +        if (unlikely(shinfo == NULL)) {
> +            rte_free(buf);
> +            VLOG_ERR("Failed to initialize shinfo");

I'm not sure this error message will be clear to an end user, 
particualrly the meaning and context of 'shinfo'. Could we make it more 
verbose i.e. Failed to initialize shared info for mbuf while atteptming 
to attach external mbuf.

> +            return NULL;
> +        }
> +    }
> +
> +    rte_pktmbuf_attach_extbuf(pkt, buf, rte_malloc_virt2iova(buf), buf_len,
> +                              shinfo);
> +    rte_pktmbuf_reset_headroom(pkt);
> +
> +    return pkt;
> +}
> +
> +static struct rte_mbuf *
> +dpdk_pktmbuf_alloc(struct rte_mempool *mp, uint32_t data_len)
> +{
> +    struct rte_mbuf *pkt = rte_pktmbuf_alloc(mp);
> +
> +    if (OVS_UNLIKELY(!pkt)) {
> +        return NULL;
> +    }
> +
> +    dp_packet_init_specific((struct dp_packet *)pkt);
> +    if (rte_pktmbuf_tailroom(pkt) >= data_len) {
> +        return pkt;
> +    }
> +
> +    if (dpdk_pktmbuf_attach_extbuf(pkt, data_len)) {
> +        return pkt;
> +    }
> +
> +    rte_pktmbuf_free(pkt);
> +
> +    return NULL;
> +}
> +

For the function below, is this expected to be used when a packet 
originates outside of DPDK interface I assume?

> +static struct dp_packet *
> +dpdk_copy_dp_packet_to_mbuf(struct rte_mempool *mp, struct dp_packet *pkt_orig)
> +{
> +    struct rte_mbuf *mbuf_dest;
> +    struct dp_packet *pkt_dest;
> +    uint32_t size;
> +    uint32_t headroom;
> +
> +    size = dp_packet_size(pkt_orig);
> +    mbuf_dest = dpdk_pktmbuf_alloc(mp, size);
> +    if (OVS_UNLIKELY(mbuf_dest == NULL)) {
> +            return NULL;
> +    }
> +
> +    pkt_dest = CONTAINER_OF(mbuf_dest, struct dp_packet, mbuf);
> +    headroom = dp_packet_headroom(pkt_orig);
> +    dp_packet_set_data(pkt_dest, (char *)dp_packet_data(pkt_dest) + headroom);
> +    memcpy(dp_packet_data(pkt_dest), dp_packet_data(pkt_orig), size);
> +    dp_packet_set_size(pkt_dest, size);
> +
> +    mbuf_dest->tx_offload = pkt_orig->mbuf.tx_offload;
> +    mbuf_dest->packet_type = pkt_orig->mbuf.packet_type;
> +    mbuf_dest->ol_flags |= (pkt_orig->mbuf.ol_flags &
> +                            ~(EXT_ATTACHED_MBUF | IND_ATTACHED_MBUF));
> +
> +    memcpy(&pkt_dest->l2_pad_size, &pkt_orig->l2_pad_size,
> +           sizeof(struct dp_packet) - offsetof(struct dp_packet, l2_pad_size));
> +
> +    if (mbuf_dest->ol_flags & PKT_TX_L4_MASK) {
> +        mbuf_dest->l2_len = (char *)dp_packet_l3(pkt_dest)
> +                                - (char *)dp_packet_eth(pkt_dest);
> +        mbuf_dest->l3_len = (char *)dp_packet_l4(pkt_dest)
> +                                - (char *) dp_packet_l3(pkt_dest);
> +    }
> +
> +    return pkt_dest;
> +}
> +
>   /* Tx function. Transmit packets indefinitely */
>   static void
>   dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet_batch *batch)
> @@ -2483,7 +2699,7 @@ dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet_batch *batch)
>       enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST };
>   #endif
>       struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
> -    struct rte_mbuf *pkts[PKT_ARRAY_SIZE];
> +    struct dp_packet *pkts[PKT_ARRAY_SIZE];
>       struct netdev_dpdk_sw_stats *sw_stats = dev->sw_stats;
>       uint32_t cnt = batch_cnt;
>       uint32_t dropped = 0;
> @@ -2504,34 +2720,30 @@ dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet_batch *batch)
>           struct dp_packet *packet = batch->packets[i];
>           uint32_t size = dp_packet_size(packet);
>   
> -        if (OVS_UNLIKELY(size > dev->max_packet_len)) {
> -            VLOG_WARN_RL(&rl, "Too big size %u max_packet_len %d",
> -                         size, dev->max_packet_len);
> -
> +        if (size > dev->max_packet_len
> +            && !(packet->mbuf.ol_flags & PKT_TX_TCP_SEG)) {
> +            VLOG_WARN_RL(&rl, "Too big size %u max_packet_len %d", size,
> +                         dev->max_packet_len);
>               mtu_drops++;
>               continue;
>           }
>   
> -        pkts[txcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp);
> +        pkts[txcnt] = dpdk_copy_dp_packet_to_mbuf(dev->dpdk_mp->mp, packet);

Ah, ignore the previous questions, it's clearer now.

>           if (OVS_UNLIKELY(!pkts[txcnt])) {
>               dropped = cnt - i;
>               break;
>           }
>   
> -        /* We have to do a copy for now */
> -        memcpy(rte_pktmbuf_mtod(pkts[txcnt], void *),
> -               dp_packet_data(packet), size);
> -        dp_packet_set_size((struct dp_packet *)pkts[txcnt], size);
> -
>           txcnt++;
>       }
>   
>       if (OVS_LIKELY(txcnt)) {
>           if (dev->type == DPDK_DEV_VHOST) {
> -            __netdev_dpdk_vhost_send(netdev, qid, (struct dp_packet **) pkts,
> -                                     txcnt);
> +            __netdev_dpdk_vhost_send(netdev, qid, pkts, txcnt);
>           } else {
> -            tx_failure = netdev_dpdk_eth_tx_burst(dev, qid, pkts, txcnt);
> +            tx_failure += netdev_dpdk_eth_tx_burst(dev, qid,
> +                                                   (struct rte_mbuf **)pkts,
> +                                                   txcnt);
>           }
>       }
>   
> @@ -2589,6 +2801,7 @@ netdev_dpdk_send__(struct netdev_dpdk *dev, int qid,
>           int batch_cnt = dp_packet_batch_size(batch);
>           struct rte_mbuf **pkts = (struct rte_mbuf **) batch->packets;
>   
> +        batch_cnt = netdev_dpdk_prep_hwol_batch(dev, pkts, batch_cnt);
>           tx_cnt = netdev_dpdk_filter_packet_len(dev, pkts, batch_cnt);
>           mtu_drops = batch_cnt - tx_cnt;
>           qos_drops = tx_cnt;
> @@ -4277,6 +4490,12 @@ netdev_dpdk_reconfigure(struct netdev *netdev)
>   
>       rte_free(dev->tx_q);
>       err = dpdk_eth_dev_init(dev);
> +    if (dev->hw_ol_features & NETDEV_TX_TSO_OFFLOAD) {
> +        netdev->ol_flags |= NETDEV_TX_OFFLOAD_TCP_TSO;
> +        netdev->ol_flags |= NETDEV_TX_OFFLOAD_TCP_CKSUM;
> +        netdev->ol_flags |= NETDEV_TX_OFFLOAD_IPV4_CKSUM;
> +    }
> +
>       dev->tx_q = netdev_dpdk_alloc_txq(netdev->n_txq);
>       if (!dev->tx_q) {
>           err = ENOMEM;
> @@ -4306,6 +4525,11 @@ dpdk_vhost_reconfigure_helper(struct netdev_dpdk *dev)
>           dev->tx_q[0].map = 0;
>       }
>   
> +    if (tso_enabled()) {
> +        dev->hw_ol_features |= NETDEV_TX_TSO_OFFLOAD;
> +        VLOG_DBG("%s: TSO enabled on vhost port", dev->up.name);

Can we return the device name via netdev_get_name(dev->up), I know there 
is an incinsistency with how the name is returned in the logs but I 
think the preference is to use this function when possible.

> +    }
> +
>       netdev_dpdk_remap_txqs(dev);
>   
>       err = netdev_dpdk_mempool_configure(dev);
> @@ -4378,6 +4602,11 @@ netdev_dpdk_vhost_client_reconfigure(struct netdev *netdev)
>               vhost_flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
>           }
>   
> +        /* Enable External Buffers if TCP Segmentation Offload is enabled */
Minor, missing period.

> +        if (tso_enabled()) {
> +            vhost_flags |= RTE_VHOST_USER_EXTBUF_SUPPORT;
> +        }
> +
>           err = rte_vhost_driver_register(dev->vhost_id, vhost_flags);
>           if (err) {
>               VLOG_ERR("vhost-user device setup failure for device %s\n",
> @@ -4402,14 +4631,20 @@ netdev_dpdk_vhost_client_reconfigure(struct netdev *netdev)
>               goto unlock;
>           }
>   
> -        err = rte_vhost_driver_disable_features(dev->vhost_id,
> -                                    1ULL << VIRTIO_NET_F_HOST_TSO4
> -                                    | 1ULL << VIRTIO_NET_F_HOST_TSO6
> -                                    | 1ULL << VIRTIO_NET_F_CSUM);
> -        if (err) {
> -            VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user "
> -                     "client port: %s\n", dev->up.name);
> -            goto unlock;
> +        if (tso_enabled()) {
> +            netdev->ol_flags |= NETDEV_TX_OFFLOAD_TCP_TSO;
> +            netdev->ol_flags |= NETDEV_TX_OFFLOAD_TCP_CKSUM;
> +            netdev->ol_flags |= NETDEV_TX_OFFLOAD_IPV4_CKSUM;
> +        } else {
> +            err = rte_vhost_driver_disable_features(dev->vhost_id,
> +                                        1ULL << VIRTIO_NET_F_HOST_TSO4
> +                                        | 1ULL << VIRTIO_NET_F_HOST_TSO6
> +                                        | 1ULL << VIRTIO_NET_F_CSUM);
> +            if (err) {
> +                VLOG_ERR("rte_vhost_driver_disable_features failed for "
> +                         "vhost user client port: %s\n", dev->up.name);
> +                goto unlock;
> +            }
>           }
>   
>           err = rte_vhost_driver_start(dev->vhost_id);
> diff --git a/lib/netdev-linux-private.h b/lib/netdev-linux-private.h
> index c14f2fb81..2eb8badd9 100644
> --- a/lib/netdev-linux-private.h
> +++ b/lib/netdev-linux-private.h
> @@ -37,10 +37,14 @@
>   
>   struct netdev;
>   
> +#define LINUX_RXQ_TSO_MAX_LEN 65536
> +
>   struct netdev_rxq_linux {
>       struct netdev_rxq up;
>       bool is_tap;
>       int fd;
> +    char *bufaux;          /* Extra buffer to recv TSO pkt */
> +    int bufaux_len;        /* Extra buffer length */
>   };
>   
>   int netdev_linux_construct(struct netdev *);
> diff --git a/lib/netdev-linux.c b/lib/netdev-linux.c
> index 0a32cf9bc..187206fc5 100644
> --- a/lib/netdev-linux.c
> +++ b/lib/netdev-linux.c
> @@ -29,16 +29,18 @@
>   #include <linux/filter.h>
>   #include <linux/gen_stats.h>
>   #include <linux/if_ether.h>
> +#include <linux/if_packet.h>
>   #include <linux/if_tun.h>
>   #include <linux/types.h>
>   #include <linux/ethtool.h>
>   #include <linux/mii.h>
>   #include <linux/rtnetlink.h>
>   #include <linux/sockios.h>
> +#include <linux/virtio_net.h>
>   #include <sys/ioctl.h>
>   #include <sys/socket.h>
> +#include <sys/uio.h>
>   #include <sys/utsname.h>
> -#include <netpacket/packet.h>
>   #include <net/if.h>
>   #include <net/if_arp.h>
>   #include <net/route.h>
> @@ -72,6 +74,7 @@
>   #include "socket-util.h"
>   #include "sset.h"
>   #include "tc.h"
> +#include "tso.h"
>   #include "timer.h"
>   #include "unaligned.h"
>   #include "openvswitch/vlog.h"
> @@ -501,6 +504,8 @@ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
>    * changes in the device miimon status, so we can use atomic_count. */
>   static atomic_count miimon_cnt = ATOMIC_COUNT_INIT(0);
>   
> +static int netdev_linux_parse_vnet_hdr(struct dp_packet *b);
> +static void netdev_linux_prepend_vnet_hdr(struct dp_packet *b, int mtu);
>   static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *,
>                                      int cmd, const char *cmd_name);
>   static int get_flags(const struct netdev *, unsigned int *flags);
> @@ -902,6 +907,13 @@ netdev_linux_common_construct(struct netdev *netdev_)
>       /* The device could be in the same network namespace or in another one. */
>       netnsid_unset(&netdev->netnsid);
>       ovs_mutex_init(&netdev->mutex);
> +
> +    if (tso_enabled()) {
> +        netdev_->ol_flags |= NETDEV_TX_OFFLOAD_TCP_TSO;
> +        netdev_->ol_flags |= NETDEV_TX_OFFLOAD_TCP_CKSUM;
> +        netdev_->ol_flags |= NETDEV_TX_OFFLOAD_IPV4_CKSUM;
> +    }
> +
>       return 0;
>   }
>   
> @@ -961,6 +973,10 @@ netdev_linux_construct_tap(struct netdev *netdev_)
>       /* Create tap device. */
>       get_flags(&netdev->up, &netdev->ifi_flags);
>       ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
> +    if (tso_enabled()) {
> +        ifr.ifr_flags |= IFF_VNET_HDR;
> +    }
> +
>       ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
>       if (ioctl(netdev->tap_fd, TUNSETIFF, &ifr) == -1) {
>           VLOG_WARN("%s: creating tap device failed: %s", name,
> @@ -1024,6 +1040,13 @@ static struct netdev_rxq *
>   netdev_linux_rxq_alloc(void)
>   {
>       struct netdev_rxq_linux *rx = xzalloc(sizeof *rx);
> +    if (tso_enabled()) {
> +        rx->bufaux = xmalloc(LINUX_RXQ_TSO_MAX_LEN);
> +        if (rx->bufaux) {
> +            rx->bufaux_len = LINUX_RXQ_TSO_MAX_LEN;
> +        }
> +    }
> +
>       return &rx->up;
>   }
>   
> @@ -1069,6 +1092,17 @@ netdev_linux_rxq_construct(struct netdev_rxq *rxq_)
>               goto error;
>           }
>   
> +        if (tso_enabled()) {
> +            error = setsockopt(rx->fd, SOL_PACKET, PACKET_VNET_HDR, &val,
> +                               sizeof val);
> +            if (error) {
> +                error = errno;
> +                VLOG_ERR("%s: failed to enable vnet hdr in txq raw socket: %s",
> +                         netdev_get_name(netdev_), ovs_strerror(errno));
> +                goto error;
> +            }
> +        }
> +
>           /* Set non-blocking mode. */
>           error = set_nonblocking(rx->fd);
>           if (error) {
> @@ -1123,6 +1157,8 @@ netdev_linux_rxq_destruct(struct netdev_rxq *rxq_)
>       if (!rx->is_tap) {
>           close(rx->fd);
>       }
> +
> +    free(rx->bufaux);
>   }
>   
>   static void
> @@ -1151,12 +1187,15 @@ auxdata_has_vlan_tci(const struct tpacket_auxdata *aux)
>       return aux->tp_vlan_tci || aux->tp_status & TP_STATUS_VLAN_VALID;
>   }
>   
> +
Is the extra white space needed here?
>   static int
> -netdev_linux_rxq_recv_sock(int fd, struct dp_packet *buffer)
> +netdev_linux_rxq_recv_sock(int fd, char *bufaux, int bufaux_len,
> +                           struct dp_packet *buffer)
>   {
> -    size_t size;
> +    size_t std_len;
> +    size_t total_len;
>       ssize_t retval;
> -    struct iovec iov;
> +    struct iovec iov[2];
>       struct cmsghdr *cmsg;
>       union {
>           struct cmsghdr cmsg;
> @@ -1166,14 +1205,17 @@ netdev_linux_rxq_recv_sock(int fd, struct dp_packet *buffer)
>   
>       /* Reserve headroom for a single VLAN tag */
>       dp_packet_reserve(buffer, VLAN_HEADER_LEN);
> -    size = dp_packet_tailroom(buffer);
> +    std_len = dp_packet_tailroom(buffer);
> +    total_len = std_len + bufaux_len;
>   
> -    iov.iov_base = dp_packet_data(buffer);
> -    iov.iov_len = size;
> +    iov[0].iov_base = dp_packet_data(buffer);
> +    iov[0].iov_len = std_len;
> +    iov[1].iov_base = bufaux;
> +    iov[1].iov_len = bufaux_len;
>       msgh.msg_name = NULL;
>       msgh.msg_namelen = 0;
> -    msgh.msg_iov = &iov;
> -    msgh.msg_iovlen = 1;
> +    msgh.msg_iov = iov;
> +    msgh.msg_iovlen = 2;
>       msgh.msg_control = &cmsg_buffer;
>       msgh.msg_controllen = sizeof cmsg_buffer;
>       msgh.msg_flags = 0;
> @@ -1184,11 +1226,26 @@ netdev_linux_rxq_recv_sock(int fd, struct dp_packet *buffer)
>   
>       if (retval < 0) {
>           return errno;
> -    } else if (retval > size) {
> +    } else if (retval > total_len) {
>           return EMSGSIZE;
>       }
>   
> -    dp_packet_set_size(buffer, dp_packet_size(buffer) + retval);
> +    if (retval > std_len) {
> +        /* Build a single linear TSO packet */
Minor, missing period.
> +        size_t extra_len = retval - std_len;
> +
> +        dp_packet_set_size(buffer, dp_packet_size(buffer) + std_len);
> +        dp_packet_prealloc_tailroom(buffer, extra_len);
> +        memcpy(dp_packet_tail(buffer), bufaux, extra_len);
> +        dp_packet_set_size(buffer, dp_packet_size(buffer) + extra_len);
> +    } else {
> +        dp_packet_set_size(buffer, dp_packet_size(buffer) + retval);
> +    }
> +
> +    if (tso_enabled() && netdev_linux_parse_vnet_hdr(buffer)) {
> +        VLOG_WARN_RL(&rl, "Invalid virtio net header");
> +        return EINVAL;
> +    }
>   
>       for (cmsg = CMSG_FIRSTHDR(&msgh); cmsg; cmsg = CMSG_NXTHDR(&msgh, cmsg)) {
>           const struct tpacket_auxdata *aux;
> @@ -1221,20 +1278,44 @@ netdev_linux_rxq_recv_sock(int fd, struct dp_packet *buffer)
>   }
>   
>   static int
> -netdev_linux_rxq_recv_tap(int fd, struct dp_packet *buffer)
> +netdev_linux_rxq_recv_tap(int fd, char *bufaux, int bufaux_len,
> +                          struct dp_packet *buffer)
>   {
>       ssize_t retval;
> -    size_t size = dp_packet_tailroom(buffer);
> +    size_t std_len;
> +    struct iovec iov[2];
> +
> +    std_len = dp_packet_tailroom(buffer);
> +    iov[0].iov_base = dp_packet_data(buffer);
> +    iov[0].iov_len = std_len;
> +    iov[1].iov_base = bufaux;
> +    iov[1].iov_len = bufaux_len;
>   
>       do {
> -        retval = read(fd, dp_packet_data(buffer), size);
> +        retval = readv(fd, iov, 2);
>       } while (retval < 0 && errno == EINTR);
>   
>       if (retval < 0) {
>           return errno;
>       }
>   
> -    dp_packet_set_size(buffer, dp_packet_size(buffer) + retval);
> +    if (retval > std_len) {
> +        /* Build a single linear TSO packet */
Minor, missing period.

> +        size_t extra_len = retval - std_len;
> +
> +        dp_packet_set_size(buffer, dp_packet_size(buffer) + std_len);
> +        dp_packet_prealloc_tailroom(buffer, extra_len);
> +        memcpy(dp_packet_tail(buffer), bufaux, extra_len);
> +        dp_packet_set_size(buffer, dp_packet_size(buffer) + extra_len);
> +    } else {
> +        dp_packet_set_size(buffer, dp_packet_size(buffer) + retval);
> +    }
> +
> +    if (tso_enabled() && netdev_linux_parse_vnet_hdr(buffer)) {
> +        VLOG_WARN_RL(&rl, "Invalid virtio net header");
> +        return EINVAL;
> +    }
> +
>       return 0;
>   }
>   
> @@ -1245,6 +1326,7 @@ netdev_linux_rxq_recv(struct netdev_rxq *rxq_, struct dp_packet_batch *batch,
>       struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_);
>       struct netdev *netdev = rx->up.netdev;
>       struct dp_packet *buffer;
> +    size_t buffer_len;
>       ssize_t retval;
>       int mtu;
>   
> @@ -1252,12 +1334,18 @@ netdev_linux_rxq_recv(struct netdev_rxq *rxq_, struct dp_packet_batch *batch,
>           mtu = ETH_PAYLOAD_MAX;
>       }
>   
> +    buffer_len = VLAN_ETH_HEADER_LEN + mtu;
> +    if (tso_enabled()) {
> +            buffer_len += sizeof(struct virtio_net_hdr);
> +    }
> +
>       /* Assume Ethernet port. No need to set packet_type. */
> -    buffer = dp_packet_new_with_headroom(VLAN_ETH_HEADER_LEN + mtu,
> -                                           DP_NETDEV_HEADROOM);
> +    buffer = dp_packet_new_with_headroom(buffer_len, DP_NETDEV_HEADROOM);
>       retval = (rx->is_tap
> -              ? netdev_linux_rxq_recv_tap(rx->fd, buffer)
> -              : netdev_linux_rxq_recv_sock(rx->fd, buffer));
> +              ? netdev_linux_rxq_recv_tap(rx->fd, rx->bufaux, rx->bufaux_len,
> +                                          buffer)
> +              : netdev_linux_rxq_recv_sock(rx->fd, rx->bufaux, rx->bufaux_len,
> +                                           buffer));
>   
>       if (retval) {
>           if (retval != EAGAIN && retval != EMSGSIZE) {
> @@ -1302,7 +1390,7 @@ netdev_linux_rxq_drain(struct netdev_rxq *rxq_)
>   }
>   
>   static int
> -netdev_linux_sock_batch_send(int sock, int ifindex,
> +netdev_linux_sock_batch_send(int sock, int ifindex, bool tso, int mtu,
>                                struct dp_packet_batch *batch)
>   {
>       const size_t size = dp_packet_batch_size(batch);
> @@ -1316,6 +1404,10 @@ netdev_linux_sock_batch_send(int sock, int ifindex,
>   
>       struct dp_packet *packet;
>       DP_PACKET_BATCH_FOR_EACH (i, packet, batch) {
> +        if (tso) {
> +            netdev_linux_prepend_vnet_hdr(packet, mtu);
> +        }
> +
>           iov[i].iov_base = dp_packet_data(packet);
>           iov[i].iov_len = dp_packet_size(packet);
>           mmsg[i].msg_hdr = (struct msghdr) { .msg_name = &sll,
> @@ -1348,7 +1440,7 @@ netdev_linux_sock_batch_send(int sock, int ifindex,
>    * on other interface types because we attach a socket filter to the rx
>    * socket. */
>   static int
> -netdev_linux_tap_batch_send(struct netdev *netdev_,
> +netdev_linux_tap_batch_send(struct netdev *netdev_, bool tso, int mtu,
>                               struct dp_packet_batch *batch)
>   {
>       struct netdev_linux *netdev = netdev_linux_cast(netdev_);
> @@ -1365,10 +1457,15 @@ netdev_linux_tap_batch_send(struct netdev *netdev_,
>       }
>   
>       DP_PACKET_BATCH_FOR_EACH (i, packet, batch) {
> -        size_t size = dp_packet_size(packet);
> +        size_t size;
>           ssize_t retval;
>           int error;
>   
> +        if (tso) {
> +            netdev_linux_prepend_vnet_hdr(packet, mtu);
> +        }
> +
> +        size = dp_packet_size(packet);
>           do {
>               retval = write(netdev->tap_fd, dp_packet_data(packet), size);
>               error = retval < 0 ? errno : 0;
> @@ -1403,9 +1500,15 @@ netdev_linux_send(struct netdev *netdev_, int qid OVS_UNUSED,
>                     struct dp_packet_batch *batch,
>                     bool concurrent_txq OVS_UNUSED)
>   {
> +    bool tso = tso_enabled();
> +    int mtu = ETH_PAYLOAD_MAX;
>       int error = 0;
>       int sock = 0;
>   
> +    if (tso) {
> +        netdev_linux_get_mtu__(netdev_linux_cast(netdev_), &mtu);
> +    }
> +
>       if (!is_tap_netdev(netdev_)) {
>           if (netdev_linux_netnsid_is_remote(netdev_linux_cast(netdev_))) {
>               error = EOPNOTSUPP;
> @@ -1424,9 +1527,9 @@ netdev_linux_send(struct netdev *netdev_, int qid OVS_UNUSED,
>               goto free_batch;
>           }
>   
> -        error = netdev_linux_sock_batch_send(sock, ifindex, batch);
> +        error = netdev_linux_sock_batch_send(sock, ifindex, tso, mtu, batch);
>       } else {
> -        error = netdev_linux_tap_batch_send(netdev_, batch);
> +        error = netdev_linux_tap_batch_send(netdev_, tso, mtu, batch);
>       }
>       if (error) {
>           if (error == ENOBUFS) {
> @@ -6170,6 +6273,19 @@ af_packet_sock(void)
>                   close(sock);
>                   sock = -error;
>               }
> +
> +            if (tso_enabled()) {
> +                int val = 1;
> +                error = setsockopt(sock, SOL_PACKET, PACKET_VNET_HDR, &val,
> +                                   sizeof val);
> +                if (error) {
> +                    error = errno;
> +                    VLOG_ERR("failed to enable vnet hdr in raw socket: %s",
> +                             ovs_strerror(errno));
> +                    close(sock);
> +                    sock = -error;
> +                }
> +            }
>           } else {
>               sock = -errno;
>               VLOG_ERR("failed to create packet socket: %s",
> @@ -6180,3 +6296,134 @@ af_packet_sock(void)
>   
>       return sock;
>   }
> +
> +static int
> +netdev_linux_parse_l2(struct dp_packet *b, uint16_t *l4proto)
> +{
> +    struct eth_header *eth_hdr;
> +    ovs_be16 eth_type;
> +    int l2_len;
> +
> +    eth_hdr = dp_packet_at(b, 0, ETH_HEADER_LEN);
> +    if (!eth_hdr) {
> +        return -EINVAL;
> +    }
> +
> +    l2_len = ETH_HEADER_LEN;
> +    eth_type = eth_hdr->eth_type;
> +    if (eth_type_vlan(eth_type)) {
> +        struct vlan_header *vlan = dp_packet_at(b, l2_len, VLAN_HEADER_LEN);
> +
> +        if (!vlan) {
> +            return -EINVAL;
> +        }
> +
> +        eth_type = vlan->vlan_next_type;
> +        l2_len += VLAN_HEADER_LEN;
> +    }
> +
> +    if (eth_type == htons(ETH_TYPE_IP)) {
> +        struct ip_header *ip_hdr = dp_packet_at(b, l2_len, IP_HEADER_LEN);
> +
> +        if (!ip_hdr) {
> +            return -EINVAL;
> +        }
> +
> +        *l4proto = ip_hdr->ip_proto;
> +        dp_packet_hwol_set_tx_ipv4(b);
> +    } else if (eth_type == htons(ETH_TYPE_IPV6)) {
> +        struct ovs_16aligned_ip6_hdr *nh6;
> +
> +        nh6 = dp_packet_at(b, l2_len, IPV6_HEADER_LEN);
> +        if (!nh6) {
> +            return -EINVAL;
> +        }
> +
> +        *l4proto = nh6->ip6_ctlun.ip6_un1.ip6_un1_nxt;
> +        dp_packet_hwol_set_tx_ipv6(b);
> +    }
> +
> +    return 0;
> +}
> +
> +static int
> +netdev_linux_parse_vnet_hdr(struct dp_packet *b)
> +{
> +    struct virtio_net_hdr *vnet = dp_packet_pull(b, sizeof *vnet);
> +    uint16_t l4proto = 0;
> +
> +    if (OVS_UNLIKELY(!vnet)) {
> +        return -EINVAL;
> +    }
> +
> +    if (vnet->flags == 0 && vnet->gso_type == VIRTIO_NET_HDR_GSO_NONE) {
> +        return 0;
> +    }
> +
> +    if (netdev_linux_parse_l2(b, &l4proto)) {
> +        return -EINVAL;
> +    }
> +
> +    if (vnet->flags == VIRTIO_NET_HDR_F_NEEDS_CSUM) {
> +        if (l4proto == IPPROTO_TCP) {
> +            dp_packet_hwol_set_csum_tcp(b);
> +        } else if (l4proto == IPPROTO_UDP) {
> +            dp_packet_hwol_set_csum_udp(b);
> +        } else if (l4proto == IPPROTO_SCTP) {
> +            dp_packet_hwol_set_csum_sctp(b);
> +        }
> +    }
> +
> +    if (l4proto && vnet->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
> +        uint8_t allowed_mask = VIRTIO_NET_HDR_GSO_TCPV4
> +                                | VIRTIO_NET_HDR_GSO_TCPV6
> +                                | VIRTIO_NET_HDR_GSO_UDP;
> +        uint8_t type = vnet->gso_type & allowed_mask;
> +
> +        if (type == VIRTIO_NET_HDR_GSO_TCPV4
> +            || type == VIRTIO_NET_HDR_GSO_TCPV6) {
> +            dp_packet_hwol_set_tcp_seg(b);
> +        }
> +    }
> +
> +    return 0;
> +}
> +
> +static void
> +netdev_linux_prepend_vnet_hdr(struct dp_packet *b, int mtu)
> +{
> +    struct virtio_net_hdr *vnet = dp_packet_push_zeros(b, sizeof *vnet);
> +
> +    if ((dp_packet_size(b) > mtu) && dp_packet_hwol_is_tso(b)) {
> +
> +        vnet->hdr_len = (char *)dp_packet_l4(b) - (char *)dp_packet_eth(b);
> +        vnet->hdr_len += TCP_HEADER_LEN;
> +        vnet->gso_size = mtu - vnet->hdr_len;
> +        if (dp_packet_hwol_is_ipv4(b)) {
> +            vnet->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
> +        } else {
> +            vnet->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
> +        }
> +
> +    } else {
> +        vnet->flags = VIRTIO_NET_HDR_GSO_NONE;
> +    }
> +
> +    if (dp_packet_hwol_l4_mask(b)) {
> +        vnet->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
> +        vnet->csum_start = (char *)dp_packet_l4(b) - (char *)dp_packet_eth(b);
> +
> +        if (dp_packet_hwol_l4_is_tcp(b)) {
> +            vnet->csum_offset = __builtin_offsetof(struct tcp_header,
> +                                                   tcp_csum);
> +        } else if (dp_packet_hwol_l4_is_udp(b)) {
> +            vnet->csum_offset = __builtin_offsetof(struct udp_header,
> +                                                   udp_csum);
> +        } else if (dp_packet_hwol_l4_is_sctp(b)) {
> +            vnet->csum_offset = __builtin_offsetof(struct sctp_header,
> +                                                   sctp_csum);
> +        } else {
> +            VLOG_WARN_RL(&rl, "Unsupported L4 protocol");
> +        }
> +    }
> +}
> diff --git a/lib/netdev-provider.h b/lib/netdev-provider.h
> index 1e5a40c89..be1db62f8 100644
> --- a/lib/netdev-provider.h
> +++ b/lib/netdev-provider.h
> @@ -37,6 +37,12 @@ extern "C" {
>   struct netdev_tnl_build_header_params;
>   #define NETDEV_NUMA_UNSPEC OVS_NUMA_UNSPEC
>   
> +enum netdev_ol_flags {
> +    NETDEV_TX_OFFLOAD_IPV4_CKSUM = 1 << 0,
> +    NETDEV_TX_OFFLOAD_TCP_CKSUM = 1 << 1,
> +    NETDEV_TX_OFFLOAD_TCP_TSO = 1 << 2,
> +};
> +
>   /* A network device (e.g. an Ethernet device).
>    *
>    * Network device implementations may read these members but should not modify
> @@ -51,6 +57,10 @@ struct netdev {
>        * opening this device, and therefore got assigned to the "system" class */
>       bool auto_classified;
>   
> +    /* This bitmask of the offloading features enabled/supported by the
> +     * supported by the netdev. */
> +    uint64_t ol_flags;
> +
>       /* If this is 'true', the user explicitly specified an MTU for this
>        * netdev.  Otherwise, Open vSwitch is allowed to override it. */
>       bool mtu_user_config;
> diff --git a/lib/netdev.c b/lib/netdev.c
> index af8f8560d..c33378803 100644
> --- a/lib/netdev.c
> +++ b/lib/netdev.c
> @@ -782,6 +782,52 @@ netdev_get_pt_mode(const struct netdev *netdev)
>               : NETDEV_PT_LEGACY_L2);
>   }
>   
> +/* Check if a 'packet' is compatible with 'netdev_flags'.
> + * If a packet is incompatible, return 'false' with the 'errormsg'
> + * pointing to a reason. */
> +static bool
> +netdev_send_prepare_packet(const uint64_t netdev_flags,
> +                           struct dp_packet *packet, char **errormsg)
> +{
> +    if (dp_packet_hwol_is_tso(packet)
> +        && !(netdev_flags & NETDEV_TX_OFFLOAD_TCP_TSO)) {
> +            /* fall back to GSO in software */
Minor formatting, capilize start of comment, add missing period.

> +            *errormsg = "No TSO support";
> +            return false;
> +    }
> +
> +    if (dp_packet_hwol_l4_mask(packet)
> +        && !(netdev_flags & NETDEV_TX_OFFLOAD_TCP_CKSUM)) {
> +            /* fall back to L4 csum in software */
> +            *errormsg = "No L4 checksum support";
> +            return false;
> +    }
> +
> +    return true;
> +}
> +
> +/* Check if each packet in 'batch' is compatible with 'netdev' features,
> + * otherwise either fall back to software implementation or drop it. */
> +static void
> +netdev_send_prepare_batch(const struct netdev *netdev,
> +                          struct dp_packet_batch *batch)
> +{
> +    struct dp_packet *packet;
> +    size_t i, size = dp_packet_batch_size(batch);
> +
> +    DP_PACKET_BATCH_REFILL_FOR_EACH (i, size, packet, batch) {
> +        char *errormsg = NULL;
> +
> +        if (netdev_send_prepare_packet(netdev->ol_flags, packet, &errormsg)) {
> +            dp_packet_batch_refill(batch, packet, i);
> +        } else {
> +            VLOG_WARN_RL(&rl, "%s: Packet dropped: %s",
> +                         errormsg ? errormsg : "Unsupported feature",
> +                         netdev_get_name(netdev));
> +        }
> +    }
> +}
> +
>   /* Sends 'batch' on 'netdev'.  Returns 0 if successful (for every packet),
>    * otherwise a positive errno value.  Returns EAGAIN without blocking if
>    * at least one the packets cannot be queued immediately.  Returns EMSGSIZE
> @@ -811,8 +857,10 @@ int
>   netdev_send(struct netdev *netdev, int qid, struct dp_packet_batch *batch,
>               bool concurrent_txq)
>   {
> -    int error = netdev->netdev_class->send(netdev, qid, batch,
> -                                           concurrent_txq);
> +    int error;
> +
> +    netdev_send_prepare_batch(netdev, batch);
> +    error = netdev->netdev_class->send(netdev, qid, batch, concurrent_txq);
>       if (!error) {
>           COVERAGE_INC(netdev_sent);
>       }
> diff --git a/lib/tso.c b/lib/tso.c
> new file mode 100644
> index 000000000..2b062d14a
> --- /dev/null
> +++ b/lib/tso.c
> @@ -0,0 +1,54 @@
> +/*
> + * Copyright (c) 2019 Red Hat, Inc.
> + *
> + * Licensed under the Apache License, Version 2.0 (the "License");
> + * you may not use this file except in compliance with the License.
> + * You may obtain a copy of the License at:
> + *
> + *     http://www.apache.org/licenses/LICENSE-2.0
> + *
> + * Unless required by applicable law or agreed to in writing, software
> + * distributed under the License is distributed on an "AS IS" BASIS,
> + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
> + * See the License for the specific language governing permissions and
> + * limitations under the License.
> + */
> +
> +#include <config.h>
> +
> +#include "smap.h"
> +#include "ovs-thread.h"
> +#include "openvswitch/vlog.h"
> +#include "dpdk.h"
> +#include "tso.h"
> +#include "vswitch-idl.h"
> +
> +VLOG_DEFINE_THIS_MODULE(tso);
> +
> +static bool tso_support_enabled = false;
> +
> +void
> +tso_init(const struct smap *ovs_other_config)
> +{
> +    if (smap_get_bool(ovs_other_config, "tso-support", false)) {
> +        static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
> +
> +        if (ovsthread_once_start(&once)) {
> +            if (dpdk_available()) {
> +                VLOG_INFO("TCP Segmentation Offloading (TSO) support enabled");
> +                tso_support_enabled = true;
> +            } else {
> +                VLOG_ERR("TCP Segmentation Offloading (TSO) is unsupported "
> +                         "without enabling DPDK");
> +                tso_support_enabled = false;
> +            }
> +            ovsthread_once_done(&once);
> +        }
> +    }
> +}
> +
> +bool
> +tso_enabled(void)
> +{
> +    return tso_support_enabled;
> +}
> diff --git a/lib/tso.h b/lib/tso.h
> new file mode 100644
> index 000000000..5cc6993f3
> --- /dev/null
> +++ b/lib/tso.h
> @@ -0,0 +1,23 @@
> +/*
> + * Copyright (c) 2019 Red Hat Inc.
> + *
> + * Licensed under the Apache License, Version 2.0 (the "License");
> + * you may not use this file except in compliance with the License.
> + * You may obtain a copy of the License at:
> + *
> + *     http://www.apache.org/licenses/LICENSE-2.0
> + *
> + * Unless required by applicable law or agreed to in writing, software
> + * distributed under the License is distributed on an "AS IS" BASIS,
> + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
> + * See the License for the specific language governing permissions and
> + * limitations under the License.
> + */
> +
> +#ifndef TSO_H
> +#define TSO_H 1
> +
> +void tso_init(const struct smap *ovs_other_config);
> +bool tso_enabled(void);
> +
> +#endif /* tso.h */
> diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c
> index 9095ebf5d..a1cd1c541 100644
> --- a/vswitchd/bridge.c
> +++ b/vswitchd/bridge.c
> @@ -65,6 +65,7 @@
>   #include "system-stats.h"
>   #include "timeval.h"
>   #include "tnl-ports.h"
> +#include "tso.h"
>   #include "util.h"
>   #include "unixctl.h"
>   #include "lib/vswitch-idl.h"
> @@ -3234,6 +3235,7 @@ bridge_run(void)
>       if (cfg) {
>           netdev_set_flow_api_enabled(&cfg->other_config);
>           dpdk_init(&cfg->other_config);
> +        tso_init(&cfg->other_config);
>       }
>   
>       /* Initialize the ofproto library.  This only needs to run once, but
> diff --git a/vswitchd/vswitch.xml b/vswitchd/vswitch.xml
> index efdfb83bb..ae7f4f265 100644
> --- a/vswitchd/vswitch.xml
> +++ b/vswitchd/vswitch.xml
> @@ -690,6 +690,18 @@
>            once in few hours or a day or a week.
>           </p>
>         </column>
> +      <column name="other_config" key="tso-support"
> +              type='{"type": "boolean"}'>
> +        <p>
> +          Set this value to <code>true</code> to enable support for TSO (TCP
> +          Segmentation Offloading). When TSO is enabled, vhost-user client
> +          interfaces can transmit packets up to 64KB.
> +        </p>
> +        <p>
> +          The default value is <code>false</code>. Changing this value requires
> +          restarting the daemon
Minor missing period at end of sentence.

Also woulld it be worthing flagging this as part of the TSO doc you've 
introduced?

> +        </p>
> +      </column>
>       </group>
>       <group title="Status">
>         <column name="next_cfg">
> 


More information about the dev mailing list