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 "connectivity.h"
32 #include "dpif-netdev.h"
34 #include "netdev-dpdk.h"
35 #include "netdev-provider.h"
36 #include "netdev-vport.h"
38 #include "ofp-print.h"
40 #include "ovs-thread.h"
46 #include "unaligned.h"
51 VLOG_DEFINE_THIS_MODULE(dpdk);
52 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
54 #define DPDK_PORT_WATCHDOG_INTERVAL 5
56 #define OVS_CACHE_LINE_SIZE CACHE_LINE_SIZE
57 #define OVS_VPORT_DPDK "ovs_dpdk"
60 * need to reserve tons of extra space in the mbufs so we can align the
61 * DMA addresses to 4KB.
64 #define MTU_TO_MAX_LEN(mtu) ((mtu) + ETHER_HDR_LEN + ETHER_CRC_LEN)
65 #define MBUF_SIZE(mtu) (MTU_TO_MAX_LEN(mtu) + (512) + \
66 sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
68 /* TODO: mempool size should be based on system resources. */
69 #define NB_MBUF (4096 * 64)
70 #define MP_CACHE_SZ (256 * 2)
73 #define NON_PMD_THREAD_TX_QUEUE 0
75 /* TODO: Needs per NIC value for these constants. */
76 #define RX_PTHRESH 32 /* Default values of RX prefetch threshold reg. */
77 #define RX_HTHRESH 32 /* Default values of RX host threshold reg. */
78 #define RX_WTHRESH 16 /* Default values of RX write-back threshold reg. */
80 #define TX_PTHRESH 36 /* Default values of TX prefetch threshold reg. */
81 #define TX_HTHRESH 0 /* Default values of TX host threshold reg. */
82 #define TX_WTHRESH 0 /* Default values of TX write-back threshold reg. */
84 static const struct rte_eth_conf port_conf = {
86 .mq_mode = ETH_MQ_RX_RSS,
88 .header_split = 0, /* Header Split disabled */
89 .hw_ip_checksum = 0, /* IP checksum offload disabled */
90 .hw_vlan_filter = 0, /* VLAN filtering disabled */
91 .jumbo_frame = 0, /* Jumbo Frame Support disabled */
97 .rss_hf = ETH_RSS_IPV4_TCP | ETH_RSS_IPV4 | ETH_RSS_IPV6,
101 .mq_mode = ETH_MQ_TX_NONE,
105 static const struct rte_eth_rxconf rx_conf = {
107 .pthresh = RX_PTHRESH,
108 .hthresh = RX_HTHRESH,
109 .wthresh = RX_WTHRESH,
113 static const struct rte_eth_txconf tx_conf = {
115 .pthresh = TX_PTHRESH,
116 .hthresh = TX_HTHRESH,
117 .wthresh = TX_WTHRESH,
123 enum { MAX_RX_QUEUE_LEN = 64 };
124 enum { MAX_TX_QUEUE_LEN = 64 };
125 enum { DRAIN_TSC = 200000ULL };
127 static int rte_eal_init_ret = ENODEV;
129 static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER;
131 /* Contains all 'struct dpdk_dev's. */
132 static struct list dpdk_list OVS_GUARDED_BY(dpdk_mutex)
133 = LIST_INITIALIZER(&dpdk_list);
135 static struct list dpdk_mp_list OVS_GUARDED_BY(dpdk_mutex)
136 = LIST_INITIALIZER(&dpdk_mp_list);
138 static pthread_t watchdog_thread;
141 struct rte_mempool *mp;
145 struct list list_node OVS_GUARDED_BY(dpdk_mutex);
148 struct dpdk_tx_queue {
149 rte_spinlock_t tx_lock;
152 struct rte_mbuf *burst_pkts[MAX_TX_QUEUE_LEN];
160 struct dpdk_tx_queue tx_q[NR_QUEUE];
162 struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex);
164 struct dpdk_mp *dpdk_mp;
168 struct netdev_stats stats_offset;
169 struct netdev_stats stats;
171 uint8_t hwaddr[ETH_ADDR_LEN];
172 enum netdev_flags flags;
174 struct rte_eth_link link;
178 struct list list_node OVS_GUARDED_BY(dpdk_mutex);
181 struct netdev_rxq_dpdk {
182 struct netdev_rxq up;
186 static int netdev_dpdk_construct(struct netdev *);
189 is_dpdk_class(const struct netdev_class *class)
191 return class->construct == netdev_dpdk_construct;
194 /* TODO: use dpdk malloc for entire OVS. infact huge page shld be used
195 * for all other sengments data, bss and text. */
198 dpdk_rte_mzalloc(size_t sz)
202 ptr = rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE);
210 free_dpdk_buf(struct ofpbuf *b)
212 struct rte_mbuf *pkt = (struct rte_mbuf *) b;
214 rte_mempool_put(pkt->pool, pkt);
218 __rte_pktmbuf_init(struct rte_mempool *mp,
219 void *opaque_arg OVS_UNUSED,
221 unsigned i OVS_UNUSED)
223 struct rte_mbuf *m = _m;
224 uint32_t buf_len = mp->elt_size - sizeof(struct ofpbuf);
226 RTE_MBUF_ASSERT(mp->elt_size >= sizeof(struct ofpbuf));
228 memset(m, 0, mp->elt_size);
230 /* start of buffer is just after mbuf structure */
231 m->buf_addr = (char *)m + sizeof(struct ofpbuf);
232 m->buf_physaddr = rte_mempool_virt2phy(mp, m) +
233 sizeof(struct ofpbuf);
234 m->buf_len = (uint16_t)buf_len;
236 /* keep some headroom between start of buffer and data */
237 m->pkt.data = (char*) m->buf_addr + RTE_MIN(RTE_PKTMBUF_HEADROOM, m->buf_len);
239 /* init some constant fields */
240 m->type = RTE_MBUF_PKT;
243 m->pkt.in_port = 0xff;
247 ovs_rte_pktmbuf_init(struct rte_mempool *mp,
248 void *opaque_arg OVS_UNUSED,
250 unsigned i OVS_UNUSED)
252 struct rte_mbuf *m = _m;
254 __rte_pktmbuf_init(mp, opaque_arg, _m, i);
256 ofpbuf_init_dpdk((struct ofpbuf *) m, m->buf_len);
259 static struct dpdk_mp *
260 dpdk_mp_get(int socket_id, int mtu) OVS_REQUIRES(dpdk_mutex)
262 struct dpdk_mp *dmp = NULL;
263 char mp_name[RTE_MEMPOOL_NAMESIZE];
265 LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) {
266 if (dmp->socket_id == socket_id && dmp->mtu == mtu) {
272 dmp = dpdk_rte_mzalloc(sizeof *dmp);
273 dmp->socket_id = socket_id;
277 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, "ovs_mp_%d", dmp->mtu);
278 dmp->mp = rte_mempool_create(mp_name, NB_MBUF, MBUF_SIZE(mtu),
280 sizeof(struct rte_pktmbuf_pool_private),
281 rte_pktmbuf_pool_init, NULL,
282 ovs_rte_pktmbuf_init, NULL,
285 if (dmp->mp == NULL) {
289 list_push_back(&dpdk_mp_list, &dmp->list_node);
294 dpdk_mp_put(struct dpdk_mp *dmp)
302 ovs_assert(dmp->refcount >= 0);
305 /* I could not find any API to destroy mp. */
306 if (dmp->refcount == 0) {
307 list_delete(dmp->list_node);
308 /* destroy mp-pool. */
314 check_link_status(struct netdev_dpdk *dev)
316 struct rte_eth_link link;
318 rte_eth_link_get_nowait(dev->port_id, &link);
320 if (dev->link.link_status != link.link_status) {
321 seq_change(connectivity_seq_get());
323 dev->link_reset_cnt++;
325 if (dev->link.link_status) {
326 VLOG_DBG_RL(&rl, "Port %d Link Up - speed %u Mbps - %s",
327 dev->port_id, (unsigned)dev->link.link_speed,
328 (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
329 ("full-duplex") : ("half-duplex"));
331 VLOG_DBG_RL(&rl, "Port %d Link Down", dev->port_id);
337 dpdk_watchdog(void *dummy OVS_UNUSED)
339 struct netdev_dpdk *dev;
341 pthread_detach(pthread_self());
344 ovs_mutex_lock(&dpdk_mutex);
345 LIST_FOR_EACH (dev, list_node, &dpdk_list) {
346 ovs_mutex_lock(&dev->mutex);
347 check_link_status(dev);
348 ovs_mutex_unlock(&dev->mutex);
350 ovs_mutex_unlock(&dpdk_mutex);
351 xsleep(DPDK_PORT_WATCHDOG_INTERVAL);
358 dpdk_eth_dev_init(struct netdev_dpdk *dev) OVS_REQUIRES(dpdk_mutex)
360 struct rte_pktmbuf_pool_private *mbp_priv;
361 struct ether_addr eth_addr;
365 if (dev->port_id < 0 || dev->port_id >= rte_eth_dev_count()) {
369 diag = rte_eth_dev_configure(dev->port_id, NR_QUEUE, NR_QUEUE, &port_conf);
371 VLOG_ERR("eth dev config error %d",diag);
375 for (i = 0; i < NR_QUEUE; i++) {
376 diag = rte_eth_tx_queue_setup(dev->port_id, i, 64, 0, &tx_conf);
378 VLOG_ERR("eth dev tx queue setup error %d",diag);
383 for (i = 0; i < NR_QUEUE; i++) {
384 diag = rte_eth_rx_queue_setup(dev->port_id, i, 64, 0, &rx_conf,
387 VLOG_ERR("eth dev rx queue setup error %d",diag);
392 diag = rte_eth_dev_start(dev->port_id);
394 VLOG_ERR("eth dev start error %d",diag);
398 rte_eth_promiscuous_enable(dev->port_id);
399 rte_eth_allmulticast_enable(dev->port_id);
401 memset(ð_addr, 0x0, sizeof(eth_addr));
402 rte_eth_macaddr_get(dev->port_id, ð_addr);
403 VLOG_INFO_RL(&rl, "Port %d: "ETH_ADDR_FMT"",
404 dev->port_id, ETH_ADDR_ARGS(eth_addr.addr_bytes));
406 memcpy(dev->hwaddr, eth_addr.addr_bytes, ETH_ADDR_LEN);
407 rte_eth_link_get_nowait(dev->port_id, &dev->link);
409 mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp);
410 dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM;
412 dev->flags = NETDEV_UP | NETDEV_PROMISC;
416 static struct netdev_dpdk *
417 netdev_dpdk_cast(const struct netdev *netdev)
419 return CONTAINER_OF(netdev, struct netdev_dpdk, up);
422 static struct netdev *
423 netdev_dpdk_alloc(void)
425 struct netdev_dpdk *netdev = dpdk_rte_mzalloc(sizeof *netdev);
430 netdev_dpdk_construct(struct netdev *netdev_)
432 struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_);
433 unsigned int port_no;
438 if (rte_eal_init_ret) {
439 return rte_eal_init_ret;
442 ovs_mutex_lock(&dpdk_mutex);
443 cport = netdev_->name + 4; /* Names always start with "dpdk" */
445 if (strncmp(netdev_->name, "dpdk", 4)) {
450 port_no = strtol(cport, 0, 0); /* string must be null terminated */
452 for (i = 0; i < NR_QUEUE; i++) {
453 rte_spinlock_init(&netdev->tx_q[i].tx_lock);
456 ovs_mutex_init(&netdev->mutex);
458 ovs_mutex_lock(&netdev->mutex);
461 netdev->mtu = ETHER_MTU;
462 netdev->max_packet_len = MTU_TO_MAX_LEN(netdev->mtu);
464 /* TODO: need to discover device node at run time. */
465 netdev->socket_id = SOCKET0;
466 netdev->port_id = port_no;
468 netdev->dpdk_mp = dpdk_mp_get(netdev->socket_id, netdev->mtu);
469 if (!netdev->dpdk_mp) {
474 err = dpdk_eth_dev_init(netdev);
478 netdev_->n_rxq = NR_QUEUE;
480 list_push_back(&dpdk_list, &netdev->list_node);
483 ovs_mutex_unlock(&netdev->mutex);
485 ovs_mutex_unlock(&dpdk_mutex);
490 netdev_dpdk_destruct(struct netdev *netdev_)
492 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_);
494 ovs_mutex_lock(&dev->mutex);
495 rte_eth_dev_stop(dev->port_id);
496 ovs_mutex_unlock(&dev->mutex);
498 ovs_mutex_lock(&dpdk_mutex);
499 list_remove(&dev->list_node);
500 dpdk_mp_put(dev->dpdk_mp);
501 ovs_mutex_unlock(&dpdk_mutex);
503 ovs_mutex_destroy(&dev->mutex);
507 netdev_dpdk_dealloc(struct netdev *netdev_)
509 struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_);
515 netdev_dpdk_get_config(const struct netdev *netdev_, struct smap *args)
517 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_);
519 ovs_mutex_lock(&dev->mutex);
521 /* TODO: Allow to configure number of queues. */
522 smap_add_format(args, "configured_rx_queues", "%u", netdev_->n_rxq);
523 smap_add_format(args, "configured_tx_queues", "%u", netdev_->n_rxq);
524 ovs_mutex_unlock(&dev->mutex);
529 static struct netdev_rxq *
530 netdev_dpdk_rxq_alloc(void)
532 struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx);
537 static struct netdev_rxq_dpdk *
538 netdev_rxq_dpdk_cast(const struct netdev_rxq *rx)
540 return CONTAINER_OF(rx, struct netdev_rxq_dpdk, up);
544 netdev_dpdk_rxq_construct(struct netdev_rxq *rxq_)
546 struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_);
547 struct netdev_dpdk *netdev = netdev_dpdk_cast(rx->up.netdev);
549 ovs_mutex_lock(&netdev->mutex);
550 rx->port_id = netdev->port_id;
551 ovs_mutex_unlock(&netdev->mutex);
557 netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq_ OVS_UNUSED)
562 netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq_)
564 struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_);
570 dpdk_queue_flush(struct netdev_dpdk *dev, int qid)
572 struct dpdk_tx_queue *txq = &dev->tx_q[qid];
575 if (txq->count == 0) {
578 rte_spinlock_lock(&txq->tx_lock);
579 nb_tx = rte_eth_tx_burst(dev->port_id, qid, txq->burst_pkts, txq->count);
580 if (nb_tx != txq->count) {
581 /* free buffers if we couldn't transmit packets */
582 rte_mempool_put_bulk(dev->dpdk_mp->mp,
583 (void **) &txq->burst_pkts[nb_tx],
584 (txq->count - nb_tx));
587 rte_spinlock_unlock(&txq->tx_lock);
591 netdev_dpdk_rxq_recv(struct netdev_rxq *rxq_, struct ofpbuf **packets, int *c)
593 struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_);
594 struct netdev *netdev = rx->up.netdev;
595 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
598 dpdk_queue_flush(dev, rxq_->queue_id);
600 nb_rx = rte_eth_rx_burst(rx->port_id, rxq_->queue_id,
601 (struct rte_mbuf **) packets, MAX_RX_QUEUE_LEN);
612 dpdk_queue_pkt(struct netdev_dpdk *dev, int qid,
613 struct rte_mbuf *pkt)
615 struct dpdk_tx_queue *txq = &dev->tx_q[qid];
620 rte_spinlock_lock(&txq->tx_lock);
621 txq->burst_pkts[txq->count++] = pkt;
622 if (txq->count == MAX_TX_QUEUE_LEN) {
625 cur_tsc = rte_get_timer_cycles();
626 if (txq->count == 1) {
629 diff_tsc = cur_tsc - txq->tsc;
630 if (diff_tsc >= DRAIN_TSC) {
633 rte_spinlock_unlock(&txq->tx_lock);
637 nb_tx = rte_eth_tx_burst(dev->port_id, qid, txq->burst_pkts, txq->count);
638 if (nb_tx != txq->count) {
639 /* free buffers if we couldn't transmit packets */
640 rte_mempool_put_bulk(dev->dpdk_mp->mp,
641 (void **) &txq->burst_pkts[nb_tx],
642 (txq->count - nb_tx));
645 rte_spinlock_unlock(&txq->tx_lock);
648 /* Tx function. Transmit packets indefinitely */
650 dpdk_do_tx_copy(struct netdev *netdev, char *buf, int size)
652 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
653 struct rte_mbuf *pkt;
655 pkt = rte_pktmbuf_alloc(dev->dpdk_mp->mp);
657 ovs_mutex_lock(&dev->mutex);
658 dev->stats.tx_dropped++;
659 ovs_mutex_unlock(&dev->mutex);
663 /* We have to do a copy for now */
664 memcpy(pkt->pkt.data, buf, size);
666 rte_pktmbuf_data_len(pkt) = size;
667 rte_pktmbuf_pkt_len(pkt) = size;
669 dpdk_queue_pkt(dev, NON_PMD_THREAD_TX_QUEUE, pkt);
670 dpdk_queue_flush(dev, NON_PMD_THREAD_TX_QUEUE);
674 netdev_dpdk_send(struct netdev *netdev,
675 struct ofpbuf *ofpbuf, bool may_steal)
677 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
680 if (ofpbuf_size(ofpbuf) > dev->max_packet_len) {
681 VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d",
682 (int)ofpbuf_size(ofpbuf) , dev->max_packet_len);
684 ovs_mutex_lock(&dev->mutex);
685 dev->stats.tx_dropped++;
686 ovs_mutex_unlock(&dev->mutex);
692 if (!may_steal || ofpbuf->source != OFPBUF_DPDK) {
693 dpdk_do_tx_copy(netdev, (char *) ofpbuf_data(ofpbuf), ofpbuf_size(ofpbuf));
696 ofpbuf_delete(ofpbuf);
701 qid = rte_lcore_id() % NR_QUEUE;
703 dpdk_queue_pkt(dev, qid, (struct rte_mbuf *)ofpbuf);
713 netdev_dpdk_set_etheraddr(struct netdev *netdev,
714 const uint8_t mac[ETH_ADDR_LEN])
716 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
718 ovs_mutex_lock(&dev->mutex);
719 if (!eth_addr_equals(dev->hwaddr, mac)) {
720 memcpy(dev->hwaddr, mac, ETH_ADDR_LEN);
722 ovs_mutex_unlock(&dev->mutex);
728 netdev_dpdk_get_etheraddr(const struct netdev *netdev,
729 uint8_t mac[ETH_ADDR_LEN])
731 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
733 ovs_mutex_lock(&dev->mutex);
734 memcpy(mac, dev->hwaddr, ETH_ADDR_LEN);
735 ovs_mutex_unlock(&dev->mutex);
741 netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup)
743 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
745 ovs_mutex_lock(&dev->mutex);
747 ovs_mutex_unlock(&dev->mutex);
753 netdev_dpdk_set_mtu(const struct netdev *netdev, int mtu)
755 struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
757 struct dpdk_mp *old_mp;
760 ovs_mutex_lock(&dpdk_mutex);
761 ovs_mutex_lock(&dev->mutex);
762 if (dev->mtu == mtu) {
767 mp = dpdk_mp_get(dev->socket_id, dev->mtu);
773 rte_eth_dev_stop(dev->port_id);
776 old_mp = dev->dpdk_mp;
779 dev->max_packet_len = MTU_TO_MAX_LEN(dev->mtu);
781 err = dpdk_eth_dev_init(dev);
786 dev->dpdk_mp = old_mp;
787 dev->max_packet_len = MTU_TO_MAX_LEN(dev->mtu);
788 dpdk_eth_dev_init(dev);
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;