[ovs-dev] [PATCH 4/4] netdev-dpdk: Add TCP Segmentation Offload support
Flavio Leitner
fbl at sysclose.org
Mon Dec 2 13:44:26 UTC 2019
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.
Signed-off-by: Flavio Leitner <fbl at sysclose.org>
---
Documentation/automake.mk | 1 +
Documentation/topics/dpdk/index.rst | 1 +
Documentation/topics/dpdk/tso.rst | 83 +++++++++++++++++++++
NEWS | 1 +
lib/automake.mk | 2 +
lib/dp-packet.c | 46 ++++++++++++
lib/dp-packet.h | 30 ++++++--
lib/netdev-bsd.c | 7 ++
lib/netdev-dpdk.c | 109 +++++++++++++++++++++++-----
lib/netdev-dummy.c | 7 ++
lib/netdev-linux.c | 71 ++++++++++++++++--
lib/tso.c | 54 ++++++++++++++
lib/tso.h | 23 ++++++
vswitchd/bridge.c | 2 +
vswitchd/vswitch.xml | 14 ++++
15 files changed, 418 insertions(+), 33 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..92393a399
--- /dev/null
+++ b/Documentation/topics/dpdk/tso.rst
@@ -0,0 +1,83 @@
+..
+ Copyright 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.
+
+ 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.
+
+===
+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`.
+
+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
+~~~~~~~~~~~
+The current OvS `TSO` implementation supports flat and VLAN networks only (i.e.
+no support for `TSO` over tunneled connection [VxLAN, GRE, IPinIP, etc.]).
+
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.
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/dp-packet.c b/lib/dp-packet.c
index e02891449..095f543f4 100644
--- a/lib/dp-packet.c
+++ b/lib/dp-packet.c
@@ -17,6 +17,7 @@
#include <config.h>
#include <stdlib.h>
#include <string.h>
+#include <linux/virtio_net.h>
#include "dp-packet.h"
#include "netdev-afxdp.h"
@@ -509,3 +510,48 @@ dp_packet_resize_l2(struct dp_packet *b, int increment)
dp_packet_adjust_layer_offset(&b->l2_5_ofs, increment);
return dp_packet_data(b);
}
+
+#ifdef DPDK_NETDEV
+void
+dp_packet_prepend_vnet_hdr(struct dp_packet *b, int mtu)
+{
+ struct virtio_net_hdr *vnet;
+ uint64_t ol_flags = b->mbuf.ol_flags;
+
+ vnet = dp_packet_push_zeros(b, sizeof *vnet);
+ if ((ol_flags & PKT_TX_TCP_SEG) && (dp_packet_size(b) > vnet->gso_size)) {
+
+ if (ol_flags & PKT_TX_IPV4) {
+ vnet->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
+ vnet->hdr_len = ETH_HLEN + IP_HEADER_LEN + TCP_HEADER_LEN;
+ } else {
+ vnet->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
+ vnet->hdr_len = ETH_HLEN + IPV6_HEADER_LEN + TCP_HEADER_LEN;
+ }
+
+ vnet->gso_size = mtu - vnet->hdr_len;
+ } else {
+ vnet->flags = VIRTIO_NET_HDR_GSO_NONE;
+ }
+
+ uint64_t tx_l4_mask = ol_flags & PKT_TX_L4_MASK;
+ if (tx_l4_mask) {
+ vnet->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
+ vnet->csum_start = (char *) dp_packet_l4(b) - (char *) dp_packet_eth(b);
+
+ if (tx_l4_mask == PKT_TX_TCP_CKSUM) {
+ vnet->csum_offset = __builtin_offsetof(struct tcp_header, tcp_csum);
+ } else if (tx_l4_mask == PKT_TX_UDP_CKSUM) {
+ vnet->csum_offset = __builtin_offsetof(struct udp_header, udp_csum);
+ } else if (tx_l4_mask == PKT_TX_SCTP_CKSUM) {
+ vnet->csum_offset = __builtin_offsetof(struct sctp_header,
+ sctp_csum);
+ }
+ }
+}
+#else
+void
+dp_packet_prepend_vnet_hdr(struct dp_packet *b OVS_UNUSED, int mtu OVS_UNUSED)
+{
+}
+#endif
diff --git a/lib/dp-packet.h b/lib/dp-packet.h
index 325924eaa..d962daa4b 100644
--- a/lib/dp-packet.h
+++ b/lib/dp-packet.h
@@ -109,6 +109,9 @@ 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);
+static inline bool dp_packet_is_tso(struct dp_packet *b);
+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 +454,7 @@ 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 +517,14 @@ dp_packet_set_allocated(struct dp_packet *b, uint16_t s)
b->mbuf.buf_len = s;
}
+static inline bool
+dp_packet_is_tso(struct dp_packet *b)
+{
+ return (b->mbuf.ol_flags & (PKT_TX_TCP_SEG | PKT_TX_L4_MASK))
+ ? true
+ : false;
+}
+
/* 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
@@ -544,8 +555,10 @@ dp_packet_reset_offload(struct dp_packet *p)
static inline bool
dp_packet_ip_checksum_valid(const struct dp_packet *p)
{
- return (p->mbuf.ol_flags & PKT_RX_IP_CKSUM_MASK) ==
- PKT_RX_IP_CKSUM_GOOD;
+
+ return (p->mbuf.ol_flags & PKT_TX_L4_MASK)
+ || ((p->mbuf.ol_flags & PKT_RX_IP_CKSUM_MASK) ==
+ PKT_RX_IP_CKSUM_GOOD);
}
static inline bool
@@ -558,8 +571,9 @@ dp_packet_ip_checksum_bad(const struct dp_packet *p)
static inline bool
dp_packet_l4_checksum_valid(const struct dp_packet *p)
{
- return (p->mbuf.ol_flags & PKT_RX_L4_CKSUM_MASK) ==
- PKT_RX_L4_CKSUM_GOOD;
+ return (p->mbuf.ol_flags & PKT_TX_L4_MASK)
+ || ((p->mbuf.ol_flags & PKT_RX_L4_CKSUM_MASK) ==
+ PKT_RX_L4_CKSUM_GOOD);
}
static inline bool
@@ -643,6 +657,12 @@ dp_packet_set_allocated(struct dp_packet *b, uint16_t s)
b->allocated_ = s;
}
+static inline bool
+dp_packet_is_tso(struct dp_packet *b OVS_UNUSED)
+{
+ return false;
+}
+
/* 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
diff --git a/lib/netdev-bsd.c b/lib/netdev-bsd.c
index 7875636cc..ad6669f9f 100644
--- a/lib/netdev-bsd.c
+++ b/lib/netdev-bsd.c
@@ -703,6 +703,13 @@ netdev_bsd_send(struct netdev *netdev_, int qid OVS_UNUSED,
const void *data = dp_packet_data(packet);
size_t size = dp_packet_size(packet);
+ /* No TSO support in BSD netdev */
+ if (dp_packet_is_tso(packet)) {
+ VLOG_WARN("%s: Dropping unsupported TSO packet of size %"PRIuSIZE,
+ name, size);
+ continue;
+ }
+
while (!error) {
ssize_t retval;
if (dev->tap_fd >= 0) {
diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c
index cd035f76e..0551a7921 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);
+ }
+
n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq);
n_txq = MIN(info.max_tx_queues, dev->up.n_txq);
@@ -1307,14 +1326,16 @@ 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);
@@ -1649,6 +1670,11 @@ 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");
}
@@ -2048,6 +2074,30 @@ netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq)
rte_free(rx);
}
+/* Prepare the packet for transmit TSO.
+ * It should be called only if PKT_TX_TCP_SEG is set in ol_flags.
+ * The flags PKT_TX_TCP_CKSUM and PKT_TX_IP_CKSUM are also set. */
+static void
+netdev_dpdk_prep_tso_packet(struct rte_mbuf *mbuf, int mtu)
+{
+ struct dp_packet *pkt = CONTAINER_OF(mbuf, struct dp_packet, mbuf);
+ struct tcp_header *th = dp_packet_l4(pkt);
+
+ 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);
+ /* There's no layer 4 in the packet. */
+ if (!th) {
+ VLOG_WARN_RL(&rl, "TSO packet without L4 header");
+ return;
+ }
+
+ mbuf->l4_len = TCP_OFFSET(th->tcp_ctl) * 4;
+ mbuf->ol_flags |= PKT_TX_TCP_CKSUM | PKT_TX_IP_CKSUM;
+ mbuf->tso_segsz = mtu - mbuf->l3_len - mbuf->l4_len;
+ mbuf->outer_l2_len = 0;
+ mbuf->outer_l3_len = 0;
+}
+
/* Tries to transmit 'pkts' to txq 'qid' of device 'dev'. Takes ownership of
* 'pkts', even in case of failure.
*
@@ -2345,11 +2395,18 @@ 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 ((pkt->ol_flags & PKT_TX_TCP_SEG)) {
+ /* vhost user client doesn't require 'pkt' to be prepared */
+ if (dev->type != DPDK_DEV_VHOST) {
+ netdev_dpdk_prep_tso_packet(pkt, dev->mtu);
+ }
+ } else 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);
rte_pktmbuf_free(pkt);
continue;
}
@@ -4303,6 +4360,11 @@ 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);
+ }
+
netdev_dpdk_remap_txqs(dev);
err = netdev_dpdk_mempool_configure(dev);
@@ -4375,6 +4437,11 @@ 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 */
+ 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",
@@ -4399,14 +4466,16 @@ 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()) {
+ 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-dummy.c b/lib/netdev-dummy.c
index 71df29184..6e86149d7 100644
--- a/lib/netdev-dummy.c
+++ b/lib/netdev-dummy.c
@@ -1108,6 +1108,13 @@ netdev_dummy_send(struct netdev *netdev, int qid OVS_UNUSED,
const void *buffer = dp_packet_data(packet);
size_t size = dp_packet_size(packet);
+ /* TSO not supported in Dummy netdev */
+ if (dp_packet_is_tso(packet)) {
+ VLOG_WARN("%s: Dropping unsupported TSO packet of size %"PRIuSIZE,
+ netdev_get_name(netdev), size);
+ continue;
+ }
+
if (!dp_packet_is_eth(packet)) {
error = EPFNOSUPPORT;
break;
diff --git a/lib/netdev-linux.c b/lib/netdev-linux.c
index 0a32cf9bc..154bd7ac1 100644
--- a/lib/netdev-linux.c
+++ b/lib/netdev-linux.c
@@ -29,16 +29,17 @@
#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/utsname.h>
-#include <netpacket/packet.h>
#include <net/if.h>
#include <net/if_arp.h>
#include <net/route.h>
@@ -72,6 +73,7 @@
#include "socket-util.h"
#include "sset.h"
#include "tc.h"
+#include "tso.h"
#include "timer.h"
#include "unaligned.h"
#include "openvswitch/vlog.h"
@@ -961,6 +963,10 @@ 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,
@@ -1069,6 +1075,17 @@ 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) {
@@ -1151,6 +1168,7 @@ auxdata_has_vlan_tci(const struct tpacket_auxdata *aux)
return aux->tp_vlan_tci || aux->tp_status & TP_STATUS_VLAN_VALID;
}
+
static int
netdev_linux_rxq_recv_sock(int fd, struct dp_packet *buffer)
{
@@ -1245,6 +1263,7 @@ 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;
@@ -1252,9 +1271,13 @@ 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));
@@ -1266,6 +1289,10 @@ netdev_linux_rxq_recv(struct netdev_rxq *rxq_, struct dp_packet_batch *batch,
}
dp_packet_delete(buffer);
} else {
+ if (tso_enabled()) {
+ dp_packet_pull(buffer, sizeof(struct virtio_net_hdr));
+ }
+
dp_packet_batch_init_packet(batch, buffer);
}
@@ -1302,7 +1329,7 @@ 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 +1343,10 @@ netdev_linux_sock_batch_send(int sock, int ifindex,
struct dp_packet *packet;
DP_PACKET_BATCH_FOR_EACH (i, packet, batch) {
+ if (tso) {
+ dp_packet_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 +1379,7 @@ 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 +1396,15 @@ 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) {
+ dp_packet_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 +1439,15 @@ 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 +1466,9 @@ 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 +6212,19 @@ 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",
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..69aec3154 100644
--- a/vswitchd/vswitch.xml
+++ b/vswitchd/vswitch.xml
@@ -690,6 +690,20 @@
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 which can be delivered
+ directly to other vhost-user client ports or to network cards with
+ TSO support.
+ </p>
+ <p>
+ The default value is <code>false</code>. Changing this value requires
+ restarting the daemon
+ </p>
+ </column>
</group>
<group title="Status">
<column name="next_cfg">
--
2.23.0
More information about the dev
mailing list