2 * Copyright (c) 2014 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
31 #include "dpif-netdev.h"
33 #include "netdev-dpdk.h"
34 #include "netdev-provider.h"
35 #include "netdev-vport.h"
37 #include "ofp-print.h"
39 #include "ovs-thread.h"
44 #include "unaligned.h"
49 VLOG_DEFINE_THIS_MODULE(dpdk);
50 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
52 #define DPDK_PORT_WATCHDOG_INTERVAL 5
54 #define OVS_CACHE_LINE_SIZE CACHE_LINE_SIZE
55 #define OVS_VPORT_DPDK "ovs_dpdk"
58 * need to reserve tons of extra space in the mbufs so we can align the
59 * DMA addresses to 4KB.
62 #define MTU_TO_MAX_LEN(mtu) ((mtu) + ETHER_HDR_LEN + ETHER_CRC_LEN)
63 #define MBUF_SIZE(mtu) (MTU_TO_MAX_LEN(mtu) + (512) + \
64 sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
66 /* TODO: mempool size should be based on system resources. */
67 #define NB_MBUF (4096 * 64)
68 #define MP_CACHE_SZ (256 * 2)
71 #define NON_PMD_THREAD_TX_QUEUE 0
73 /* TODO: Needs per NIC value for these constants. */
74 #define RX_PTHRESH 32 /* Default values of RX prefetch threshold reg. */
75 #define RX_HTHRESH 32 /* Default values of RX host threshold reg. */
76 #define RX_WTHRESH 16 /* Default values of RX write-back threshold reg. */
78 #define TX_PTHRESH 36 /* Default values of TX prefetch threshold reg. */
79 #define TX_HTHRESH 0 /* Default values of TX host threshold reg. */
80 #define TX_WTHRESH 0 /* Default values of TX write-back threshold reg. */
82 static const struct rte_eth_conf port_conf = {
84 .mq_mode = ETH_MQ_RX_RSS,
86 .header_split = 0, /* Header Split disabled */
87 .hw_ip_checksum = 0, /* IP checksum offload disabled */
88 .hw_vlan_filter = 0, /* VLAN filtering disabled */
89 .jumbo_frame = 0, /* Jumbo Frame Support disabled */
95 .rss_hf = ETH_RSS_IPV4_TCP | ETH_RSS_IPV4 | ETH_RSS_IPV6,
99 .mq_mode = ETH_MQ_TX_NONE,
103 static const struct rte_eth_rxconf rx_conf = {
105 .pthresh = RX_PTHRESH,
106 .hthresh = RX_HTHRESH,
107 .wthresh = RX_WTHRESH,
111 static const struct rte_eth_txconf tx_conf = {
113 .pthresh = TX_PTHRESH,
114 .hthresh = TX_HTHRESH,
115 .wthresh = TX_WTHRESH,
121 enum { MAX_RX_QUEUE_LEN = 64 };
122 enum { MAX_TX_QUEUE_LEN = 64 };
123 enum { DRAIN_TSC = 200000ULL };
125 static int rte_eal_init_ret = ENODEV;
127 static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER;
129 /* Contains all 'struct dpdk_dev's. */
130 static struct list dpdk_list OVS_GUARDED_BY(dpdk_mutex)
131 = LIST_INITIALIZER(&dpdk_list);
133 static struct list dpdk_mp_list OVS_GUARDED_BY(dpdk_mutex)
134 = LIST_INITIALIZER(&dpdk_mp_list);
136 static pthread_t watchdog_thread;
139 struct rte_mempool *mp;
143 struct list list_node OVS_GUARDED_BY(dpdk_mutex);
146 struct dpdk_tx_queue {
147 rte_spinlock_t tx_lock;
150 struct rte_mbuf *burst_pkts[MAX_TX_QUEUE_LEN];
158 struct dpdk_tx_queue tx_q[NR_QUEUE];
160 struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex);
162 struct dpdk_mp *dpdk_mp;
166 struct netdev_stats stats_offset;
167 struct netdev_stats stats;
169 uint8_t hwaddr[ETH_ADDR_LEN];
170 enum netdev_flags flags;
172 struct rte_eth_link link;
176 struct list list_node OVS_GUARDED_BY(dpdk_mutex);
179 struct netdev_rxq_dpdk {
180 struct netdev_rxq up;
184 static int netdev_dpdk_construct(struct netdev *);
187 is_dpdk_class(const struct netdev_class *class)
189 return class->construct == netdev_dpdk_construct;
192 /* TODO: use dpdk malloc for entire OVS. infact huge page shld be used
193 * for all other sengments data, bss and text. */
196 dpdk_rte_mzalloc(size_t sz)
200 ptr = rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE);
208 free_dpdk_buf(struct ofpbuf *b)
210 struct rte_mbuf *pkt = (struct rte_mbuf *) b;
212 rte_mempool_put(pkt->pool, pkt);
216 __rte_pktmbuf_init(struct rte_mempool *mp,
217 void *opaque_arg OVS_UNUSED,
219 unsigned i OVS_UNUSED)
221 struct rte_mbuf *m = _m;
222 uint32_t buf_len = mp->elt_size - sizeof(struct ofpbuf);
224 RTE_MBUF_ASSERT(mp->elt_size >= sizeof(struct ofpbuf));
226 memset(m, 0, mp->elt_size);
228 /* start of buffer is just after mbuf structure */
229 m->buf_addr = (char *)m + sizeof(struct ofpbuf);
230 m->buf_physaddr = rte_mempool_virt2phy(mp, m) +
231 sizeof(struct ofpbuf);
232 m->buf_len = (uint16_t)buf_len;
234 /* keep some headroom between start of buffer and data */
235 m->pkt.data = (char*) m->buf_addr + RTE_MIN(RTE_PKTMBUF_HEADROOM, m->buf_len);
237 /* init some constant fields */
238 m->type = RTE_MBUF_PKT;
241 m->pkt.in_port = 0xff;
245 ovs_rte_pktmbuf_init(struct rte_mempool *mp,
246 void *opaque_arg OVS_UNUSED,
248 unsigned i OVS_UNUSED)
250 struct rte_mbuf *m = _m;
252 __rte_pktmbuf_init(mp, opaque_arg, _m, i);
254 ofpbuf_init_dpdk((struct ofpbuf *) m, m->buf_len);
257 static struct dpdk_mp *
258 dpdk_mp_get(int socket_id, int mtu) OVS_REQUIRES(dpdk_mutex)
260 struct dpdk_mp *dmp = NULL;
261 char mp_name[RTE_MEMPOOL_NAMESIZE];
263 LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) {
264 if (dmp->socket_id == socket_id && dmp->mtu == mtu) {
270 dmp = dpdk_rte_mzalloc(sizeof *dmp);
271 dmp->socket_id = socket_id;
275 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, "ovs_mp_%d", dmp->mtu);
276 dmp->mp = rte_mempool_create(mp_name, NB_MBUF, MBUF_SIZE(mtu),
278 sizeof(struct rte_pktmbuf_pool_private),
279 rte_pktmbuf_pool_init, NULL,
280 ovs_rte_pktmbuf_init, NULL,
283 if (dmp->mp == NULL) {
287 list_push_back(&dpdk_mp_list, &dmp->list_node);
292 dpdk_mp_put(struct dpdk_mp *dmp)
300 ovs_assert(dmp->refcount >= 0);
303 /* I could not find any API to destroy mp. */
304 if (dmp->refcount == 0) {
305 list_delete(dmp->list_node);
306 /* destroy mp-pool. */
312 check_link_status(struct netdev_dpdk *dev)
314 struct rte_eth_link link;
316 rte_eth_link_get_nowait(dev->port_id, &link);
318 if (dev->link.link_status != link.link_status) {
319 netdev_change_seq_changed(&dev->up);
321 dev->link_reset_cnt++;
323 if (dev->link.link_status) {
324 VLOG_DBG_RL(&rl, "Port %d Link Up - speed %u Mbps - %s",
325 dev->port_id, (unsigned)dev->link.link_speed,
326 (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
327 ("full-duplex") : ("half-duplex"));
329 VLOG_DBG_RL(&rl, "Port %d Link Down", dev->port_id);
335 dpdk_watchdog(void *dummy OVS_UNUSED)
337 struct netdev_dpdk *dev;
339 pthread_detach(pthread_self());
342 ovs_mutex_lock(&dpdk_mutex);
343 LIST_FOR_EACH (dev, list_node, &dpdk_list) {
344 ovs_mutex_lock(&dev->mutex);
345 check_link_status(dev);
346 ovs_mutex_unlock(&dev->mutex);
348 ovs_mutex_unlock(&dpdk_mutex);
349 xsleep(DPDK_PORT_WATCHDOG_INTERVAL);
356 dpdk_eth_dev_init(struct netdev_dpdk *dev) OVS_REQUIRES(dpdk_mutex)
358 struct rte_pktmbuf_pool_private *mbp_priv;
359 struct ether_addr eth_addr;
363 if (dev->port_id < 0 || dev->port_id >= rte_eth_dev_count()) {
367 diag = rte_eth_dev_configure(dev->port_id, NR_QUEUE, NR_QUEUE, &port_conf);
369 VLOG_ERR("eth dev config error %d",diag);
373 for (i = 0; i < NR_QUEUE; i++) {
374 diag = rte_eth_tx_queue_setup(dev->port_id, i, 64, 0, &tx_conf);
376 VLOG_ERR("eth dev tx queue setup error %d",diag);
381 for (i = 0; i < NR_QUEUE; i++) {
382 diag = rte_eth_rx_queue_setup(dev->port_id, i, 64, 0, &rx_conf,
385 VLOG_ERR("eth dev rx queue setup error %d",diag);
390 diag = rte_eth_dev_start(dev->port_id);
392 VLOG_ERR("eth dev start error %d",diag);
396 rte_eth_promiscuous_enable(dev->port_id);
397 rte_eth_allmulticast_enable(dev->port_id);
399 memset(ð_addr, 0x0, sizeof(eth_addr));
400 rte_eth_macaddr_get(dev->port_id, ð_addr);
401 VLOG_INFO_RL(&rl, "Port %d: "ETH_ADDR_FMT"",
402 dev->port_id, ETH_ADDR_ARGS(eth_addr.addr_bytes));
404 memcpy(dev->hwaddr, eth_addr.addr_bytes, ETH_ADDR_LEN);
405 rte_eth_link_get_nowait(dev->port_id, &dev->link);
407 mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp);
408 dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM;
410 dev->flags = NETDEV_UP | NETDEV_PROMISC;
414 static struct netdev_dpdk *
415 netdev_dpdk_cast(const struct netdev *netdev)
417 return CONTAINER_OF(netdev, struct netdev_dpdk, up);
420 static struct netdev *
421 netdev_dpdk_alloc(void)
423 struct netdev_dpdk *netdev = dpdk_rte_mzalloc(sizeof *netdev);
428 netdev_dpdk_construct(struct netdev *netdev_)
430 struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_);
431 unsigned int port_no;
436 if (rte_eal_init_ret) {
437 return rte_eal_init_ret;
440 ovs_mutex_lock(&dpdk_mutex);
441 cport = netdev_->name + 4; /* Names always start with "dpdk" */
443 if (strncmp(netdev_->name, "dpdk", 4)) {
448 port_no = strtol(cport, 0, 0); /* string must be null terminated */
450 for (i = 0; i < NR_QUEUE; i++) {
451 rte_spinlock_init(&netdev->tx_q[i].tx_lock);
454 ovs_mutex_init(&netdev->mutex);
456 ovs_mutex_lock(&netdev->mutex);
459 netdev->mtu = ETHER_MTU;
460 netdev->max_packet_len = MTU_TO_MAX_LEN(netdev->mtu);
462 /* TODO: need to discover device node at run time. */
463 netdev->socket_id = SOCKET0;
464 netdev->port_id = port_no;
466 netdev->dpdk_mp = dpdk_mp_get(netdev->socket_id, netdev->mtu);
467 if (!netdev->dpdk_mp) {
472 err = dpdk_eth_dev_init(netdev);
476 netdev_->n_rxq = NR_QUEUE;
478 list_push_back(&dpdk_list, &netdev->list_node);
481 ovs_mutex_unlock(&netdev->mutex);
483 ovs_mutex_unlock(&dpdk_mutex);
488 netdev_dpdk_destruct(struct netdev *netdev_)
490 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_);
492 ovs_mutex_lock(&dev->mutex);
493 rte_eth_dev_stop(dev->port_id);
494 ovs_mutex_unlock(&dev->mutex);
496 ovs_mutex_lock(&dpdk_mutex);
497 list_remove(&dev->list_node);
498 dpdk_mp_put(dev->dpdk_mp);
499 ovs_mutex_unlock(&dpdk_mutex);
501 ovs_mutex_destroy(&dev->mutex);
505 netdev_dpdk_dealloc(struct netdev *netdev_)
507 struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_);
513 netdev_dpdk_get_config(const struct netdev *netdev_, struct smap *args)
515 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_);
517 ovs_mutex_lock(&dev->mutex);
519 /* TODO: Allow to configure number of queues. */
520 smap_add_format(args, "configured_rx_queues", "%u", netdev_->n_rxq);
521 smap_add_format(args, "configured_tx_queues", "%u", netdev_->n_rxq);
522 ovs_mutex_unlock(&dev->mutex);
527 static struct netdev_rxq *
528 netdev_dpdk_rxq_alloc(void)
530 struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx);
535 static struct netdev_rxq_dpdk *
536 netdev_rxq_dpdk_cast(const struct netdev_rxq *rx)
538 return CONTAINER_OF(rx, struct netdev_rxq_dpdk, up);
542 netdev_dpdk_rxq_construct(struct netdev_rxq *rxq_)
544 struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_);
545 struct netdev_dpdk *netdev = netdev_dpdk_cast(rx->up.netdev);
547 ovs_mutex_lock(&netdev->mutex);
548 rx->port_id = netdev->port_id;
549 ovs_mutex_unlock(&netdev->mutex);
555 netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq_ OVS_UNUSED)
560 netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq_)
562 struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_);
568 dpdk_queue_flush(struct netdev_dpdk *dev, int qid)
570 struct dpdk_tx_queue *txq = &dev->tx_q[qid];
573 if (txq->count == 0) {
576 rte_spinlock_lock(&txq->tx_lock);
577 nb_tx = rte_eth_tx_burst(dev->port_id, qid, txq->burst_pkts, txq->count);
578 if (nb_tx != txq->count) {
579 /* free buffers if we couldn't transmit packets */
580 rte_mempool_put_bulk(dev->dpdk_mp->mp,
581 (void **) &txq->burst_pkts[nb_tx],
582 (txq->count - nb_tx));
585 rte_spinlock_unlock(&txq->tx_lock);
589 netdev_dpdk_rxq_recv(struct netdev_rxq *rxq_, struct ofpbuf **packets, int *c)
591 struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_);
592 struct netdev *netdev = rx->up.netdev;
593 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
596 dpdk_queue_flush(dev, rxq_->queue_id);
598 nb_rx = rte_eth_rx_burst(rx->port_id, rxq_->queue_id,
599 (struct rte_mbuf **) packets, MAX_RX_QUEUE_LEN);
610 dpdk_queue_pkt(struct netdev_dpdk *dev, int qid,
611 struct rte_mbuf *pkt)
613 struct dpdk_tx_queue *txq = &dev->tx_q[qid];
618 rte_spinlock_lock(&txq->tx_lock);
619 txq->burst_pkts[txq->count++] = pkt;
620 if (txq->count == MAX_TX_QUEUE_LEN) {
623 cur_tsc = rte_get_timer_cycles();
624 if (txq->count == 1) {
627 diff_tsc = cur_tsc - txq->tsc;
628 if (diff_tsc >= DRAIN_TSC) {
631 rte_spinlock_unlock(&txq->tx_lock);
635 nb_tx = rte_eth_tx_burst(dev->port_id, qid, txq->burst_pkts, txq->count);
636 if (nb_tx != txq->count) {
637 /* free buffers if we couldn't transmit packets */
638 rte_mempool_put_bulk(dev->dpdk_mp->mp,
639 (void **) &txq->burst_pkts[nb_tx],
640 (txq->count - nb_tx));
643 rte_spinlock_unlock(&txq->tx_lock);
646 /* Tx function. Transmit packets indefinitely */
648 dpdk_do_tx_copy(struct netdev *netdev, char *buf, int size)
650 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
651 struct rte_mbuf *pkt;
653 pkt = rte_pktmbuf_alloc(dev->dpdk_mp->mp);
655 ovs_mutex_lock(&dev->mutex);
656 dev->stats.tx_dropped++;
657 ovs_mutex_unlock(&dev->mutex);
661 /* We have to do a copy for now */
662 memcpy(pkt->pkt.data, buf, size);
664 rte_pktmbuf_data_len(pkt) = size;
665 rte_pktmbuf_pkt_len(pkt) = size;
667 dpdk_queue_pkt(dev, NON_PMD_THREAD_TX_QUEUE, pkt);
668 dpdk_queue_flush(dev, NON_PMD_THREAD_TX_QUEUE);
672 netdev_dpdk_send(struct netdev *netdev,
673 struct ofpbuf *ofpbuf, bool may_steal)
675 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
678 if (ofpbuf_size(ofpbuf) > dev->max_packet_len) {
679 VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d",
680 (int)ofpbuf_size(ofpbuf) , dev->max_packet_len);
682 ovs_mutex_lock(&dev->mutex);
683 dev->stats.tx_dropped++;
684 ovs_mutex_unlock(&dev->mutex);
690 if (!may_steal || ofpbuf->source != OFPBUF_DPDK) {
691 dpdk_do_tx_copy(netdev, (char *) ofpbuf_data(ofpbuf), ofpbuf_size(ofpbuf));
694 ofpbuf_delete(ofpbuf);
699 qid = rte_lcore_id() % NR_QUEUE;
701 dpdk_queue_pkt(dev, qid, (struct rte_mbuf *)ofpbuf);
711 netdev_dpdk_set_etheraddr(struct netdev *netdev,
712 const uint8_t mac[ETH_ADDR_LEN])
714 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
716 ovs_mutex_lock(&dev->mutex);
717 if (!eth_addr_equals(dev->hwaddr, mac)) {
718 memcpy(dev->hwaddr, mac, ETH_ADDR_LEN);
719 netdev_change_seq_changed(netdev);
721 ovs_mutex_unlock(&dev->mutex);
727 netdev_dpdk_get_etheraddr(const struct netdev *netdev,
728 uint8_t mac[ETH_ADDR_LEN])
730 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
732 ovs_mutex_lock(&dev->mutex);
733 memcpy(mac, dev->hwaddr, ETH_ADDR_LEN);
734 ovs_mutex_unlock(&dev->mutex);
740 netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup)
742 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
744 ovs_mutex_lock(&dev->mutex);
746 ovs_mutex_unlock(&dev->mutex);
752 netdev_dpdk_set_mtu(const struct netdev *netdev, int mtu)
754 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
756 struct dpdk_mp *old_mp;
759 ovs_mutex_lock(&dpdk_mutex);
760 ovs_mutex_lock(&dev->mutex);
761 if (dev->mtu == mtu) {
766 mp = dpdk_mp_get(dev->socket_id, dev->mtu);
772 rte_eth_dev_stop(dev->port_id);
775 old_mp = dev->dpdk_mp;
778 dev->max_packet_len = MTU_TO_MAX_LEN(dev->mtu);
780 err = dpdk_eth_dev_init(dev);
785 dev->dpdk_mp = old_mp;
786 dev->max_packet_len = MTU_TO_MAX_LEN(dev->mtu);
787 dpdk_eth_dev_init(dev);
792 netdev_change_seq_changed(netdev);
794 ovs_mutex_unlock(&dev->mutex);
795 ovs_mutex_unlock(&dpdk_mutex);
800 netdev_dpdk_get_carrier(const struct netdev *netdev_, bool *carrier);
803 netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
805 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
806 struct rte_eth_stats rte_stats;
809 netdev_dpdk_get_carrier(netdev, &gg);
810 ovs_mutex_lock(&dev->mutex);
811 rte_eth_stats_get(dev->port_id, &rte_stats);
813 *stats = dev->stats_offset;
815 stats->rx_packets += rte_stats.ipackets;
816 stats->tx_packets += rte_stats.opackets;
817 stats->rx_bytes += rte_stats.ibytes;
818 stats->tx_bytes += rte_stats.obytes;
819 stats->rx_errors += rte_stats.ierrors;
820 stats->tx_errors += rte_stats.oerrors;
821 stats->multicast += rte_stats.imcasts;
823 stats->tx_dropped += dev->stats.tx_dropped;
824 ovs_mutex_unlock(&dev->mutex);
830 netdev_dpdk_set_stats(struct netdev *netdev, const struct netdev_stats *stats)
832 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
834 ovs_mutex_lock(&dev->mutex);
835 dev->stats_offset = *stats;
836 ovs_mutex_unlock(&dev->mutex);
842 netdev_dpdk_get_features(const struct netdev *netdev_,
843 enum netdev_features *current,
844 enum netdev_features *advertised OVS_UNUSED,
845 enum netdev_features *supported OVS_UNUSED,
846 enum netdev_features *peer OVS_UNUSED)
848 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_);
849 struct rte_eth_link link;
851 ovs_mutex_lock(&dev->mutex);
853 ovs_mutex_unlock(&dev->mutex);
855 if (link.link_duplex == ETH_LINK_AUTONEG_DUPLEX) {
856 if (link.link_speed == ETH_LINK_SPEED_AUTONEG) {
857 *current = NETDEV_F_AUTONEG;
859 } else if (link.link_duplex == ETH_LINK_HALF_DUPLEX) {
860 if (link.link_speed == ETH_LINK_SPEED_10) {
861 *current = NETDEV_F_10MB_HD;
863 if (link.link_speed == ETH_LINK_SPEED_100) {
864 *current = NETDEV_F_100MB_HD;
866 if (link.link_speed == ETH_LINK_SPEED_1000) {
867 *current = NETDEV_F_1GB_HD;
869 } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) {
870 if (link.link_speed == ETH_LINK_SPEED_10) {
871 *current = NETDEV_F_10MB_FD;
873 if (link.link_speed == ETH_LINK_SPEED_100) {
874 *current = NETDEV_F_100MB_FD;
876 if (link.link_speed == ETH_LINK_SPEED_1000) {
877 *current = NETDEV_F_1GB_FD;
879 if (link.link_speed == ETH_LINK_SPEED_10000) {
880 *current = NETDEV_F_10GB_FD;
888 netdev_dpdk_get_ifindex(const struct netdev *netdev)
890 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
893 ovs_mutex_lock(&dev->mutex);
894 ifindex = dev->port_id;
895 ovs_mutex_unlock(&dev->mutex);
901 netdev_dpdk_get_carrier(const struct netdev *netdev_, bool *carrier)
903 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_);
905 ovs_mutex_lock(&dev->mutex);
906 check_link_status(dev);
907 *carrier = dev->link.link_status;
908 ovs_mutex_unlock(&dev->mutex);
914 netdev_dpdk_get_carrier_resets(const struct netdev *netdev_)
916 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_);
917 long long int carrier_resets;
919 ovs_mutex_lock(&dev->mutex);
920 carrier_resets = dev->link_reset_cnt;
921 ovs_mutex_unlock(&dev->mutex);
923 return carrier_resets;
927 netdev_dpdk_set_miimon(struct netdev *netdev_ OVS_UNUSED,
928 long long int interval OVS_UNUSED)
934 netdev_dpdk_update_flags__(struct netdev_dpdk *dev,
935 enum netdev_flags off, enum netdev_flags on,
936 enum netdev_flags *old_flagsp)
937 OVS_REQUIRES(dev->mutex)
941 if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) {
945 *old_flagsp = dev->flags;
949 if (dev->flags == *old_flagsp) {
953 if (dev->flags & NETDEV_UP) {
954 err = rte_eth_dev_start(dev->port_id);
959 if (dev->flags & NETDEV_PROMISC) {
960 rte_eth_promiscuous_enable(dev->port_id);
963 if (!(dev->flags & NETDEV_UP)) {
964 rte_eth_dev_stop(dev->port_id);
971 netdev_dpdk_update_flags(struct netdev *netdev_,
972 enum netdev_flags off, enum netdev_flags on,
973 enum netdev_flags *old_flagsp)
975 struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_);
978 ovs_mutex_lock(&netdev->mutex);
979 error = netdev_dpdk_update_flags__(netdev, off, on, old_flagsp);
980 ovs_mutex_unlock(&netdev->mutex);
986 netdev_dpdk_get_status(const struct netdev *netdev_, struct smap *args)
988 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_);
989 struct rte_eth_dev_info dev_info;
991 if (dev->port_id <= 0)
994 ovs_mutex_lock(&dev->mutex);
995 rte_eth_dev_info_get(dev->port_id, &dev_info);
996 ovs_mutex_unlock(&dev->mutex);
998 smap_add_format(args, "driver_name", "%s", dev_info.driver_name);
1000 smap_add_format(args, "numa_id", "%d", rte_eth_dev_socket_id(dev->port_id));
1001 smap_add_format(args, "driver_name", "%s", dev_info.driver_name);
1002 smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize);
1003 smap_add_format(args, "max_rx_pktlen", "%u", dev_info.max_rx_pktlen);
1004 smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues);
1005 smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues);
1006 smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs);
1007 smap_add_format(args, "max_hash_mac_addrs", "%u", dev_info.max_hash_mac_addrs);
1008 smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs);
1009 smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools);
1011 smap_add_format(args, "pci-vendor_id", "0x%u", dev_info.pci_dev->id.vendor_id);
1012 smap_add_format(args, "pci-device_id", "0x%x", dev_info.pci_dev->id.device_id);
1018 netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state)
1019 OVS_REQUIRES(dev->mutex)
1021 enum netdev_flags old_flags;
1024 netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags);
1026 netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags);
1031 netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc,
1032 const char *argv[], void *aux OVS_UNUSED)
1036 if (!strcasecmp(argv[argc - 1], "up")) {
1038 } else if ( !strcasecmp(argv[argc - 1], "down")) {
1041 unixctl_command_reply_error(conn, "Invalid Admin State");
1046 struct netdev *netdev = netdev_from_name(argv[1]);
1047 if (netdev && is_dpdk_class(netdev->netdev_class)) {
1048 struct netdev_dpdk *dpdk_dev = netdev_dpdk_cast(netdev);
1050 ovs_mutex_lock(&dpdk_dev->mutex);
1051 netdev_dpdk_set_admin_state__(dpdk_dev, up);
1052 ovs_mutex_unlock(&dpdk_dev->mutex);
1054 netdev_close(netdev);
1056 unixctl_command_reply_error(conn, "Not a DPDK Interface");
1057 netdev_close(netdev);
1061 struct netdev_dpdk *netdev;
1063 ovs_mutex_lock(&dpdk_mutex);
1064 LIST_FOR_EACH (netdev, list_node, &dpdk_list) {
1065 ovs_mutex_lock(&netdev->mutex);
1066 netdev_dpdk_set_admin_state__(netdev, up);
1067 ovs_mutex_unlock(&netdev->mutex);
1069 ovs_mutex_unlock(&dpdk_mutex);
1071 unixctl_command_reply(conn, "OK");
1075 dpdk_class_init(void)
1079 if (rte_eal_init_ret) {
1083 result = rte_pmd_init_all();
1085 VLOG_ERR("Cannot init PMD");
1089 result = rte_eal_pci_probe();
1091 VLOG_ERR("Cannot probe PCI");
1095 if (rte_eth_dev_count() < 1) {
1096 VLOG_ERR("No Ethernet devices found. Try assigning ports to UIO.");
1099 VLOG_INFO("Ethernet Device Count: %d", (int)rte_eth_dev_count());
1101 list_init(&dpdk_list);
1102 list_init(&dpdk_mp_list);
1104 unixctl_command_register("netdev-dpdk/set-admin-state",
1105 "[netdev] up|down", 1, 2,
1106 netdev_dpdk_set_admin_state, NULL);
1108 xpthread_create(&watchdog_thread, NULL, dpdk_watchdog, NULL);
1112 static struct netdev_class netdev_dpdk_class = {
1114 dpdk_class_init, /* init */
1115 NULL, /* netdev_dpdk_run */
1116 NULL, /* netdev_dpdk_wait */
1119 netdev_dpdk_construct,
1120 netdev_dpdk_destruct,
1121 netdev_dpdk_dealloc,
1122 netdev_dpdk_get_config,
1123 NULL, /* netdev_dpdk_set_config */
1124 NULL, /* get_tunnel_config */
1126 netdev_dpdk_send, /* send */
1127 NULL, /* send_wait */
1129 netdev_dpdk_set_etheraddr,
1130 netdev_dpdk_get_etheraddr,
1131 netdev_dpdk_get_mtu,
1132 netdev_dpdk_set_mtu,
1133 netdev_dpdk_get_ifindex,
1134 netdev_dpdk_get_carrier,
1135 netdev_dpdk_get_carrier_resets,
1136 netdev_dpdk_set_miimon,
1137 netdev_dpdk_get_stats,
1138 netdev_dpdk_set_stats,
1139 netdev_dpdk_get_features,
1140 NULL, /* set_advertisements */
1142 NULL, /* set_policing */
1143 NULL, /* get_qos_types */
1144 NULL, /* get_qos_capabilities */
1147 NULL, /* get_queue */
1148 NULL, /* set_queue */
1149 NULL, /* delete_queue */
1150 NULL, /* get_queue_stats */
1151 NULL, /* queue_dump_start */
1152 NULL, /* queue_dump_next */
1153 NULL, /* queue_dump_done */
1154 NULL, /* dump_queue_stats */
1159 NULL, /* add_router */
1160 NULL, /* get_next_hop */
1161 netdev_dpdk_get_status,
1162 NULL, /* arp_lookup */
1164 netdev_dpdk_update_flags,
1166 netdev_dpdk_rxq_alloc,
1167 netdev_dpdk_rxq_construct,
1168 netdev_dpdk_rxq_destruct,
1169 netdev_dpdk_rxq_dealloc,
1170 netdev_dpdk_rxq_recv,
1171 NULL, /* rxq_wait */
1172 NULL, /* rxq_drain */
1176 dpdk_init(int argc, char **argv)
1180 if (strcmp(argv[1], "--dpdk"))
1186 /* Make sure things are initialized ... */
1187 result = rte_eal_init(argc, argv);
1189 ovs_abort(result, "Cannot init EAL\n");
1192 rte_eal_init_ret = 0;
1198 netdev_dpdk_register(void)
1200 netdev_register_provider(&netdev_dpdk_class);
1204 pmd_thread_setaffinity_cpu(int cpu)
1210 CPU_SET(cpu, &cpuset);
1211 err = pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), &cpuset);
1213 VLOG_ERR("Thread affinity error %d",err);
1216 RTE_PER_LCORE(_lcore_id) = cpu;