4 * Copyright (c) 2016 IGEL Co., Ltd.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of IGEL Co.,Ltd. nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
36 #ifdef RTE_LIBRTE_VHOST_NUMA
41 #include <rte_ethdev.h>
42 #include <rte_malloc.h>
43 #include <rte_memcpy.h>
45 #include <rte_kvargs.h>
46 #include <rte_virtio_net.h>
47 #include <rte_spinlock.h>
49 #include "rte_eth_vhost.h"
51 #define ETH_VHOST_IFACE_ARG "iface"
52 #define ETH_VHOST_QUEUES_ARG "queues"
53 #define ETH_VHOST_CLIENT_ARG "client"
54 #define ETH_VHOST_DEQUEUE_ZERO_COPY "dequeue-zero-copy"
56 static const char *drivername = "VHOST PMD";
58 static const char *valid_arguments[] = {
62 ETH_VHOST_DEQUEUE_ZERO_COPY,
66 static struct ether_addr base_eth_addr = {
77 enum vhost_xstats_pkts {
78 VHOST_UNDERSIZE_PKT = 0,
83 VHOST_512_TO_1023_PKT,
84 VHOST_1024_TO_1522_PKT,
85 VHOST_1523_TO_MAX_PKT,
90 VHOST_ERRORS_FRAGMENTED,
92 VHOST_UNKNOWN_PROTOCOL,
100 uint64_t xstats[VHOST_XSTATS_MAX];
105 rte_atomic32_t allow_queuing;
106 rte_atomic32_t while_queuing;
107 struct pmd_internal *internal;
108 struct rte_mempool *mb_pool;
110 uint16_t virtqueue_id;
111 struct vhost_stats stats;
114 struct pmd_internal {
120 volatile uint16_t once;
123 struct internal_list {
124 TAILQ_ENTRY(internal_list) next;
125 struct rte_eth_dev *eth_dev;
128 TAILQ_HEAD(internal_list_head, internal_list);
129 static struct internal_list_head internal_list =
130 TAILQ_HEAD_INITIALIZER(internal_list);
132 static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
134 static rte_atomic16_t nb_started_ports;
135 static pthread_t session_th;
137 static struct rte_eth_link pmd_link = {
139 .link_duplex = ETH_LINK_FULL_DUPLEX,
140 .link_status = ETH_LINK_DOWN
143 struct rte_vhost_vring_state {
146 bool cur[RTE_MAX_QUEUES_PER_PORT * 2];
147 bool seen[RTE_MAX_QUEUES_PER_PORT * 2];
149 unsigned int max_vring;
152 static struct rte_vhost_vring_state *vring_states[RTE_MAX_ETHPORTS];
154 #define VHOST_XSTATS_NAME_SIZE 64
156 struct vhost_xstats_name_off {
157 char name[VHOST_XSTATS_NAME_SIZE];
161 /* [rx]_is prepended to the name string here */
162 static const struct vhost_xstats_name_off vhost_rxport_stat_strings[] = {
164 offsetof(struct vhost_queue, stats.pkts)},
166 offsetof(struct vhost_queue, stats.bytes)},
168 offsetof(struct vhost_queue, stats.missed_pkts)},
169 {"broadcast_packets",
170 offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
171 {"multicast_packets",
172 offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
174 offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
175 {"undersize_packets",
176 offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
178 offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
179 {"size_65_to_127_packets",
180 offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
181 {"size_128_to_255_packets",
182 offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
183 {"size_256_to_511_packets",
184 offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
185 {"size_512_to_1023_packets",
186 offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
187 {"size_1024_to_1522_packets",
188 offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
189 {"size_1523_to_max_packets",
190 offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
191 {"errors_with_bad_CRC",
192 offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
193 {"fragmented_errors",
194 offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_FRAGMENTED])},
196 offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_JABBER])},
197 {"unknown_protos_packets",
198 offsetof(struct vhost_queue, stats.xstats[VHOST_UNKNOWN_PROTOCOL])},
201 /* [tx]_ is prepended to the name string here */
202 static const struct vhost_xstats_name_off vhost_txport_stat_strings[] = {
204 offsetof(struct vhost_queue, stats.pkts)},
206 offsetof(struct vhost_queue, stats.bytes)},
208 offsetof(struct vhost_queue, stats.missed_pkts)},
209 {"broadcast_packets",
210 offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
211 {"multicast_packets",
212 offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
214 offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
215 {"undersize_packets",
216 offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
218 offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
219 {"size_65_to_127_packets",
220 offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
221 {"size_128_to_255_packets",
222 offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
223 {"size_256_to_511_packets",
224 offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
225 {"size_512_to_1023_packets",
226 offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
227 {"size_1024_to_1522_packets",
228 offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
229 {"size_1523_to_max_packets",
230 offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
231 {"errors_with_bad_CRC",
232 offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
235 #define VHOST_NB_XSTATS_RXPORT (sizeof(vhost_rxport_stat_strings) / \
236 sizeof(vhost_rxport_stat_strings[0]))
238 #define VHOST_NB_XSTATS_TXPORT (sizeof(vhost_txport_stat_strings) / \
239 sizeof(vhost_txport_stat_strings[0]))
242 vhost_dev_xstats_reset(struct rte_eth_dev *dev)
244 struct vhost_queue *vq = NULL;
247 for (i = 0; i < dev->data->nb_rx_queues; i++) {
248 vq = dev->data->rx_queues[i];
251 memset(&vq->stats, 0, sizeof(vq->stats));
253 for (i = 0; i < dev->data->nb_tx_queues; i++) {
254 vq = dev->data->tx_queues[i];
257 memset(&vq->stats, 0, sizeof(vq->stats));
262 vhost_dev_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
263 struct rte_eth_xstat_name *xstats_names,
264 unsigned int limit __rte_unused)
268 int nstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
272 for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
273 snprintf(xstats_names[count].name,
274 sizeof(xstats_names[count].name),
275 "rx_%s", vhost_rxport_stat_strings[t].name);
278 for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
279 snprintf(xstats_names[count].name,
280 sizeof(xstats_names[count].name),
281 "tx_%s", vhost_txport_stat_strings[t].name);
288 vhost_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
293 unsigned int count = 0;
294 struct vhost_queue *vq = NULL;
295 unsigned int nxstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
300 for (i = 0; i < dev->data->nb_rx_queues; i++) {
301 vq = dev->data->rx_queues[i];
304 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
305 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
306 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
308 for (i = 0; i < dev->data->nb_tx_queues; i++) {
309 vq = dev->data->tx_queues[i];
312 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
313 + vq->stats.missed_pkts
314 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
315 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
317 for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
318 xstats[count].value = 0;
319 for (i = 0; i < dev->data->nb_rx_queues; i++) {
320 vq = dev->data->rx_queues[i];
323 xstats[count].value +=
324 *(uint64_t *)(((char *)vq)
325 + vhost_rxport_stat_strings[t].offset);
327 xstats[count].id = count;
330 for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
331 xstats[count].value = 0;
332 for (i = 0; i < dev->data->nb_tx_queues; i++) {
333 vq = dev->data->tx_queues[i];
336 xstats[count].value +=
337 *(uint64_t *)(((char *)vq)
338 + vhost_txport_stat_strings[t].offset);
340 xstats[count].id = count;
347 vhost_count_multicast_broadcast(struct vhost_queue *vq,
348 struct rte_mbuf *mbuf)
350 struct ether_addr *ea = NULL;
351 struct vhost_stats *pstats = &vq->stats;
353 ea = rte_pktmbuf_mtod(mbuf, struct ether_addr *);
354 if (is_multicast_ether_addr(ea)) {
355 if (is_broadcast_ether_addr(ea))
356 pstats->xstats[VHOST_BROADCAST_PKT]++;
358 pstats->xstats[VHOST_MULTICAST_PKT]++;
363 vhost_update_packet_xstats(struct vhost_queue *vq,
364 struct rte_mbuf **bufs,
367 uint32_t pkt_len = 0;
370 struct vhost_stats *pstats = &vq->stats;
372 for (i = 0; i < count ; i++) {
373 pkt_len = bufs[i]->pkt_len;
375 pstats->xstats[VHOST_64_PKT]++;
376 } else if (pkt_len > 64 && pkt_len < 1024) {
377 index = (sizeof(pkt_len) * 8)
378 - __builtin_clz(pkt_len) - 5;
379 pstats->xstats[index]++;
382 pstats->xstats[VHOST_UNDERSIZE_PKT]++;
383 else if (pkt_len <= 1522)
384 pstats->xstats[VHOST_1024_TO_1522_PKT]++;
385 else if (pkt_len > 1522)
386 pstats->xstats[VHOST_1523_TO_MAX_PKT]++;
388 vhost_count_multicast_broadcast(vq, bufs[i]);
393 eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
395 struct vhost_queue *r = q;
396 uint16_t i, nb_rx = 0;
398 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
401 rte_atomic32_set(&r->while_queuing, 1);
403 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
406 /* Dequeue packets from guest TX queue */
407 nb_rx = rte_vhost_dequeue_burst(r->vid,
408 r->virtqueue_id, r->mb_pool, bufs, nb_bufs);
410 r->stats.pkts += nb_rx;
412 for (i = 0; likely(i < nb_rx); i++) {
413 bufs[i]->port = r->port;
414 r->stats.bytes += bufs[i]->pkt_len;
417 vhost_update_packet_xstats(r, bufs, nb_rx);
420 rte_atomic32_set(&r->while_queuing, 0);
426 eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
428 struct vhost_queue *r = q;
429 uint16_t i, nb_tx = 0;
431 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
434 rte_atomic32_set(&r->while_queuing, 1);
436 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
439 /* Enqueue packets to guest RX queue */
440 nb_tx = rte_vhost_enqueue_burst(r->vid,
441 r->virtqueue_id, bufs, nb_bufs);
443 r->stats.pkts += nb_tx;
444 r->stats.missed_pkts += nb_bufs - nb_tx;
446 for (i = 0; likely(i < nb_tx); i++)
447 r->stats.bytes += bufs[i]->pkt_len;
449 vhost_update_packet_xstats(r, bufs, nb_tx);
451 /* According to RFC2863 page42 section ifHCOutMulticastPkts and
452 * ifHCOutBroadcastPkts, the counters "multicast" and "broadcast"
453 * are increased when packets are not transmitted successfully.
455 for (i = nb_tx; i < nb_bufs; i++)
456 vhost_count_multicast_broadcast(r, bufs[i]);
458 for (i = 0; likely(i < nb_tx); i++)
459 rte_pktmbuf_free(bufs[i]);
461 rte_atomic32_set(&r->while_queuing, 0);
467 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
472 static inline struct internal_list *
473 find_internal_resource(char *ifname)
476 struct internal_list *list;
477 struct pmd_internal *internal;
482 pthread_mutex_lock(&internal_list_lock);
484 TAILQ_FOREACH(list, &internal_list, next) {
485 internal = list->eth_dev->data->dev_private;
486 if (!strcmp(internal->iface_name, ifname)) {
492 pthread_mutex_unlock(&internal_list_lock);
503 struct rte_eth_dev *eth_dev;
504 struct internal_list *list;
505 struct pmd_internal *internal;
506 struct vhost_queue *vq;
508 char ifname[PATH_MAX];
509 #ifdef RTE_LIBRTE_VHOST_NUMA
513 rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
514 list = find_internal_resource(ifname);
516 RTE_LOG(INFO, PMD, "Invalid device name: %s\n", ifname);
520 eth_dev = list->eth_dev;
521 internal = eth_dev->data->dev_private;
523 #ifdef RTE_LIBRTE_VHOST_NUMA
524 newnode = rte_vhost_get_numa_node(vid);
526 eth_dev->data->numa_node = newnode;
529 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
530 vq = eth_dev->data->rx_queues[i];
534 vq->internal = internal;
535 vq->port = eth_dev->data->port_id;
537 for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
538 vq = eth_dev->data->tx_queues[i];
542 vq->internal = internal;
543 vq->port = eth_dev->data->port_id;
546 for (i = 0; i < rte_vhost_get_queue_num(vid) * VIRTIO_QNUM; i++)
547 rte_vhost_enable_guest_notification(vid, i, 0);
549 eth_dev->data->dev_link.link_status = ETH_LINK_UP;
551 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
552 vq = eth_dev->data->rx_queues[i];
555 rte_atomic32_set(&vq->allow_queuing, 1);
557 for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
558 vq = eth_dev->data->tx_queues[i];
561 rte_atomic32_set(&vq->allow_queuing, 1);
564 RTE_LOG(INFO, PMD, "New connection established\n");
566 _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
572 destroy_device(int vid)
574 struct rte_eth_dev *eth_dev;
575 struct vhost_queue *vq;
576 struct internal_list *list;
577 char ifname[PATH_MAX];
579 struct rte_vhost_vring_state *state;
581 rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
582 list = find_internal_resource(ifname);
584 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
587 eth_dev = list->eth_dev;
589 /* Wait until rx/tx_pkt_burst stops accessing vhost device */
590 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
591 vq = eth_dev->data->rx_queues[i];
594 rte_atomic32_set(&vq->allow_queuing, 0);
595 while (rte_atomic32_read(&vq->while_queuing))
598 for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
599 vq = eth_dev->data->tx_queues[i];
602 rte_atomic32_set(&vq->allow_queuing, 0);
603 while (rte_atomic32_read(&vq->while_queuing))
607 eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
609 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
610 vq = eth_dev->data->rx_queues[i];
615 for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
616 vq = eth_dev->data->tx_queues[i];
622 state = vring_states[eth_dev->data->port_id];
623 rte_spinlock_lock(&state->lock);
624 for (i = 0; i <= state->max_vring; i++) {
625 state->cur[i] = false;
626 state->seen[i] = false;
628 state->max_vring = 0;
629 rte_spinlock_unlock(&state->lock);
631 RTE_LOG(INFO, PMD, "Connection closed\n");
633 _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
637 vring_state_changed(int vid, uint16_t vring, int enable)
639 struct rte_vhost_vring_state *state;
640 struct rte_eth_dev *eth_dev;
641 struct internal_list *list;
642 char ifname[PATH_MAX];
644 rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
645 list = find_internal_resource(ifname);
647 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
651 eth_dev = list->eth_dev;
653 state = vring_states[eth_dev->data->port_id];
654 rte_spinlock_lock(&state->lock);
655 state->cur[vring] = enable;
656 state->max_vring = RTE_MAX(vring, state->max_vring);
657 rte_spinlock_unlock(&state->lock);
659 RTE_LOG(INFO, PMD, "vring%u is %s\n",
660 vring, enable ? "enabled" : "disabled");
662 _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE, NULL);
668 rte_eth_vhost_get_queue_event(uint8_t port_id,
669 struct rte_eth_vhost_queue_event *event)
671 struct rte_vhost_vring_state *state;
675 if (port_id >= RTE_MAX_ETHPORTS) {
676 RTE_LOG(ERR, PMD, "Invalid port id\n");
680 state = vring_states[port_id];
682 RTE_LOG(ERR, PMD, "Unused port\n");
686 rte_spinlock_lock(&state->lock);
687 for (i = 0; i <= state->max_vring; i++) {
688 idx = state->index++ % (state->max_vring + 1);
690 if (state->cur[idx] != state->seen[idx]) {
691 state->seen[idx] = state->cur[idx];
692 event->queue_id = idx / 2;
694 event->enable = state->cur[idx];
695 rte_spinlock_unlock(&state->lock);
699 rte_spinlock_unlock(&state->lock);
705 rte_eth_vhost_get_vid_from_port_id(uint8_t port_id)
707 struct internal_list *list;
708 struct rte_eth_dev *eth_dev;
709 struct vhost_queue *vq;
712 if (!rte_eth_dev_is_valid_port(port_id))
715 pthread_mutex_lock(&internal_list_lock);
717 TAILQ_FOREACH(list, &internal_list, next) {
718 eth_dev = list->eth_dev;
719 if (eth_dev->data->port_id == port_id) {
720 vq = eth_dev->data->rx_queues[0];
728 pthread_mutex_unlock(&internal_list_lock);
734 vhost_driver_session(void *param __rte_unused)
736 static struct virtio_net_device_ops vhost_ops;
738 /* set vhost arguments */
739 vhost_ops.new_device = new_device;
740 vhost_ops.destroy_device = destroy_device;
741 vhost_ops.vring_state_changed = vring_state_changed;
742 if (rte_vhost_driver_callback_register(&vhost_ops) < 0)
743 RTE_LOG(ERR, PMD, "Can't register callbacks\n");
745 /* start event handling */
746 rte_vhost_driver_session_start();
752 vhost_driver_session_start(void)
756 ret = pthread_create(&session_th,
757 NULL, vhost_driver_session, NULL);
759 RTE_LOG(ERR, PMD, "Can't create a thread\n");
765 vhost_driver_session_stop(void)
769 ret = pthread_cancel(session_th);
771 RTE_LOG(ERR, PMD, "Can't cancel the thread\n");
773 ret = pthread_join(session_th, NULL);
775 RTE_LOG(ERR, PMD, "Can't join the thread\n");
779 eth_dev_start(struct rte_eth_dev *dev)
781 struct pmd_internal *internal = dev->data->dev_private;
784 if (rte_atomic16_cmpset(&internal->once, 0, 1)) {
785 ret = rte_vhost_driver_register(internal->iface_name,
791 /* We need only one message handling thread */
792 if (rte_atomic16_add_return(&nb_started_ports, 1) == 1)
793 ret = vhost_driver_session_start();
799 eth_dev_stop(struct rte_eth_dev *dev)
801 struct pmd_internal *internal = dev->data->dev_private;
803 if (rte_atomic16_cmpset(&internal->once, 1, 0))
804 rte_vhost_driver_unregister(internal->iface_name);
806 if (rte_atomic16_sub_return(&nb_started_ports, 1) == 0)
807 vhost_driver_session_stop();
811 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
812 uint16_t nb_rx_desc __rte_unused,
813 unsigned int socket_id,
814 const struct rte_eth_rxconf *rx_conf __rte_unused,
815 struct rte_mempool *mb_pool)
817 struct vhost_queue *vq;
819 vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
820 RTE_CACHE_LINE_SIZE, socket_id);
822 RTE_LOG(ERR, PMD, "Failed to allocate memory for rx queue\n");
826 vq->mb_pool = mb_pool;
827 vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
828 dev->data->rx_queues[rx_queue_id] = vq;
834 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
835 uint16_t nb_tx_desc __rte_unused,
836 unsigned int socket_id,
837 const struct rte_eth_txconf *tx_conf __rte_unused)
839 struct vhost_queue *vq;
841 vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
842 RTE_CACHE_LINE_SIZE, socket_id);
844 RTE_LOG(ERR, PMD, "Failed to allocate memory for tx queue\n");
848 vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
849 dev->data->tx_queues[tx_queue_id] = vq;
855 eth_dev_info(struct rte_eth_dev *dev,
856 struct rte_eth_dev_info *dev_info)
858 struct pmd_internal *internal;
860 internal = dev->data->dev_private;
861 if (internal == NULL) {
862 RTE_LOG(ERR, PMD, "Invalid device specified\n");
866 dev_info->driver_name = drivername;
867 dev_info->max_mac_addrs = 1;
868 dev_info->max_rx_pktlen = (uint32_t)-1;
869 dev_info->max_rx_queues = internal->max_queues;
870 dev_info->max_tx_queues = internal->max_queues;
871 dev_info->min_rx_bufsize = 0;
875 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
878 unsigned long rx_total = 0, tx_total = 0, tx_missed_total = 0;
879 unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
880 struct vhost_queue *vq;
882 for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
883 i < dev->data->nb_rx_queues; i++) {
884 if (dev->data->rx_queues[i] == NULL)
886 vq = dev->data->rx_queues[i];
887 stats->q_ipackets[i] = vq->stats.pkts;
888 rx_total += stats->q_ipackets[i];
890 stats->q_ibytes[i] = vq->stats.bytes;
891 rx_total_bytes += stats->q_ibytes[i];
894 for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
895 i < dev->data->nb_tx_queues; i++) {
896 if (dev->data->tx_queues[i] == NULL)
898 vq = dev->data->tx_queues[i];
899 stats->q_opackets[i] = vq->stats.pkts;
900 tx_missed_total += vq->stats.missed_pkts;
901 tx_total += stats->q_opackets[i];
903 stats->q_obytes[i] = vq->stats.bytes;
904 tx_total_bytes += stats->q_obytes[i];
907 stats->ipackets = rx_total;
908 stats->opackets = tx_total;
909 stats->oerrors = tx_missed_total;
910 stats->ibytes = rx_total_bytes;
911 stats->obytes = tx_total_bytes;
915 eth_stats_reset(struct rte_eth_dev *dev)
917 struct vhost_queue *vq;
920 for (i = 0; i < dev->data->nb_rx_queues; i++) {
921 if (dev->data->rx_queues[i] == NULL)
923 vq = dev->data->rx_queues[i];
927 for (i = 0; i < dev->data->nb_tx_queues; i++) {
928 if (dev->data->tx_queues[i] == NULL)
930 vq = dev->data->tx_queues[i];
933 vq->stats.missed_pkts = 0;
938 eth_queue_release(void *q)
944 eth_link_update(struct rte_eth_dev *dev __rte_unused,
945 int wait_to_complete __rte_unused)
951 * Disable features in feature_mask. Returns 0 on success.
954 rte_eth_vhost_feature_disable(uint64_t feature_mask)
956 return rte_vhost_feature_disable(feature_mask);
960 * Enable features in feature_mask. Returns 0 on success.
963 rte_eth_vhost_feature_enable(uint64_t feature_mask)
965 return rte_vhost_feature_enable(feature_mask);
968 /* Returns currently supported vhost features */
970 rte_eth_vhost_feature_get(void)
972 return rte_vhost_feature_get();
975 static const struct eth_dev_ops ops = {
976 .dev_start = eth_dev_start,
977 .dev_stop = eth_dev_stop,
978 .dev_configure = eth_dev_configure,
979 .dev_infos_get = eth_dev_info,
980 .rx_queue_setup = eth_rx_queue_setup,
981 .tx_queue_setup = eth_tx_queue_setup,
982 .rx_queue_release = eth_queue_release,
983 .tx_queue_release = eth_queue_release,
984 .link_update = eth_link_update,
985 .stats_get = eth_stats_get,
986 .stats_reset = eth_stats_reset,
987 .xstats_reset = vhost_dev_xstats_reset,
988 .xstats_get = vhost_dev_xstats_get,
989 .xstats_get_names = vhost_dev_xstats_get_names,
993 eth_dev_vhost_create(const char *name, char *iface_name, int16_t queues,
994 const unsigned numa_node, uint64_t flags)
996 struct rte_eth_dev_data *data = NULL;
997 struct pmd_internal *internal = NULL;
998 struct rte_eth_dev *eth_dev = NULL;
999 struct ether_addr *eth_addr = NULL;
1000 struct rte_vhost_vring_state *vring_state = NULL;
1001 struct internal_list *list = NULL;
1003 RTE_LOG(INFO, PMD, "Creating VHOST-USER backend on numa socket %u\n",
1006 /* now do all data allocation - for eth_dev structure, dummy pci driver
1007 * and internal (private) data
1009 data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
1013 internal = rte_zmalloc_socket(name, sizeof(*internal), 0, numa_node);
1014 if (internal == NULL)
1017 list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
1021 /* reserve an ethdev entry */
1022 eth_dev = rte_eth_dev_allocate(name);
1023 if (eth_dev == NULL)
1026 eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
1027 if (eth_addr == NULL)
1029 *eth_addr = base_eth_addr;
1030 eth_addr->addr_bytes[5] = eth_dev->data->port_id;
1032 vring_state = rte_zmalloc_socket(name,
1033 sizeof(*vring_state), 0, numa_node);
1034 if (vring_state == NULL)
1037 /* now put it all together
1038 * - store queue data in internal,
1039 * - store numa_node info in ethdev data
1040 * - point eth_dev_data to internals
1041 * - and point eth_dev structure to new eth_dev_data structure
1043 internal->dev_name = strdup(name);
1044 if (internal->dev_name == NULL)
1046 internal->iface_name = strdup(iface_name);
1047 if (internal->iface_name == NULL)
1049 internal->flags = flags;
1051 list->eth_dev = eth_dev;
1052 pthread_mutex_lock(&internal_list_lock);
1053 TAILQ_INSERT_TAIL(&internal_list, list, next);
1054 pthread_mutex_unlock(&internal_list_lock);
1056 rte_spinlock_init(&vring_state->lock);
1057 vring_states[eth_dev->data->port_id] = vring_state;
1059 data->dev_private = internal;
1060 data->port_id = eth_dev->data->port_id;
1061 memmove(data->name, eth_dev->data->name, sizeof(data->name));
1062 data->nb_rx_queues = queues;
1063 data->nb_tx_queues = queues;
1064 internal->max_queues = queues;
1065 data->dev_link = pmd_link;
1066 data->mac_addrs = eth_addr;
1068 /* We'll replace the 'data' originally allocated by eth_dev. So the
1069 * vhost PMD resources won't be shared between multi processes.
1071 eth_dev->data = data;
1072 eth_dev->dev_ops = &ops;
1073 eth_dev->driver = NULL;
1075 RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
1076 data->kdrv = RTE_KDRV_NONE;
1077 data->drv_name = internal->dev_name;
1078 data->numa_node = numa_node;
1080 /* finally assign rx and tx ops */
1081 eth_dev->rx_pkt_burst = eth_vhost_rx;
1082 eth_dev->tx_pkt_burst = eth_vhost_tx;
1084 return data->port_id;
1088 free(internal->dev_name);
1089 rte_free(vring_state);
1092 rte_eth_dev_release_port(eth_dev);
1101 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
1103 const char **iface_name = extra_args;
1108 *iface_name = value;
1114 open_int(const char *key __rte_unused, const char *value, void *extra_args)
1116 uint16_t *n = extra_args;
1118 if (value == NULL || extra_args == NULL)
1121 *n = (uint16_t)strtoul(value, NULL, 0);
1122 if (*n == USHRT_MAX && errno == ERANGE)
1129 rte_pmd_vhost_probe(const char *name, const char *params)
1131 struct rte_kvargs *kvlist = NULL;
1136 int client_mode = 0;
1137 int dequeue_zero_copy = 0;
1139 RTE_LOG(INFO, PMD, "Initializing pmd_vhost for %s\n", name);
1141 kvlist = rte_kvargs_parse(params, valid_arguments);
1145 if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
1146 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
1147 &open_iface, &iface_name);
1155 if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
1156 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
1157 &open_int, &queues);
1158 if (ret < 0 || queues > RTE_MAX_QUEUES_PER_PORT)
1164 if (rte_kvargs_count(kvlist, ETH_VHOST_CLIENT_ARG) == 1) {
1165 ret = rte_kvargs_process(kvlist, ETH_VHOST_CLIENT_ARG,
1166 &open_int, &client_mode);
1171 flags |= RTE_VHOST_USER_CLIENT;
1174 if (rte_kvargs_count(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY) == 1) {
1175 ret = rte_kvargs_process(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY,
1176 &open_int, &dequeue_zero_copy);
1180 if (dequeue_zero_copy)
1181 flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
1184 eth_dev_vhost_create(name, iface_name, queues, rte_socket_id(), flags);
1187 rte_kvargs_free(kvlist);
1192 rte_pmd_vhost_remove(const char *name)
1194 struct rte_eth_dev *eth_dev = NULL;
1195 struct pmd_internal *internal;
1196 struct internal_list *list;
1199 RTE_LOG(INFO, PMD, "Un-Initializing pmd_vhost for %s\n", name);
1201 /* find an ethdev entry */
1202 eth_dev = rte_eth_dev_allocated(name);
1203 if (eth_dev == NULL)
1206 internal = eth_dev->data->dev_private;
1207 if (internal == NULL)
1210 list = find_internal_resource(internal->iface_name);
1214 pthread_mutex_lock(&internal_list_lock);
1215 TAILQ_REMOVE(&internal_list, list, next);
1216 pthread_mutex_unlock(&internal_list_lock);
1219 eth_dev_stop(eth_dev);
1221 rte_free(vring_states[eth_dev->data->port_id]);
1222 vring_states[eth_dev->data->port_id] = NULL;
1224 free(internal->dev_name);
1225 free(internal->iface_name);
1227 for (i = 0; i < eth_dev->data->nb_rx_queues; i++)
1228 rte_free(eth_dev->data->rx_queues[i]);
1229 for (i = 0; i < eth_dev->data->nb_tx_queues; i++)
1230 rte_free(eth_dev->data->tx_queues[i]);
1232 rte_free(eth_dev->data->mac_addrs);
1233 rte_free(eth_dev->data);
1236 rte_eth_dev_release_port(eth_dev);
1241 static struct rte_vdev_driver pmd_vhost_drv = {
1242 .probe = rte_pmd_vhost_probe,
1243 .remove = rte_pmd_vhost_remove,
1246 RTE_PMD_REGISTER_VDEV(net_vhost, pmd_vhost_drv);
1247 RTE_PMD_REGISTER_ALIAS(net_vhost, eth_vhost);
1248 RTE_PMD_REGISTER_PARAM_STRING(net_vhost,