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 {
115 rte_atomic32_t dev_attached;
119 rte_atomic32_t started;
122 struct internal_list {
123 TAILQ_ENTRY(internal_list) next;
124 struct rte_eth_dev *eth_dev;
127 TAILQ_HEAD(internal_list_head, internal_list);
128 static struct internal_list_head internal_list =
129 TAILQ_HEAD_INITIALIZER(internal_list);
131 static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
133 static rte_atomic16_t nb_started_ports;
134 static pthread_t session_th;
136 static struct rte_eth_link pmd_link = {
138 .link_duplex = ETH_LINK_FULL_DUPLEX,
139 .link_status = ETH_LINK_DOWN
142 struct rte_vhost_vring_state {
145 bool cur[RTE_MAX_QUEUES_PER_PORT * 2];
146 bool seen[RTE_MAX_QUEUES_PER_PORT * 2];
148 unsigned int max_vring;
151 static struct rte_vhost_vring_state *vring_states[RTE_MAX_ETHPORTS];
153 #define VHOST_XSTATS_NAME_SIZE 64
155 struct vhost_xstats_name_off {
156 char name[VHOST_XSTATS_NAME_SIZE];
160 /* [rx]_is prepended to the name string here */
161 static const struct vhost_xstats_name_off vhost_rxport_stat_strings[] = {
163 offsetof(struct vhost_queue, stats.pkts)},
165 offsetof(struct vhost_queue, stats.bytes)},
167 offsetof(struct vhost_queue, stats.missed_pkts)},
168 {"broadcast_packets",
169 offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
170 {"multicast_packets",
171 offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
173 offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
174 {"undersize_packets",
175 offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
177 offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
178 {"size_65_to_127_packets",
179 offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
180 {"size_128_to_255_packets",
181 offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
182 {"size_256_to_511_packets",
183 offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
184 {"size_512_to_1023_packets",
185 offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
186 {"size_1024_to_1522_packets",
187 offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
188 {"size_1523_to_max_packets",
189 offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
190 {"errors_with_bad_CRC",
191 offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
192 {"fragmented_errors",
193 offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_FRAGMENTED])},
195 offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_JABBER])},
196 {"unknown_protos_packets",
197 offsetof(struct vhost_queue, stats.xstats[VHOST_UNKNOWN_PROTOCOL])},
200 /* [tx]_ is prepended to the name string here */
201 static const struct vhost_xstats_name_off vhost_txport_stat_strings[] = {
203 offsetof(struct vhost_queue, stats.pkts)},
205 offsetof(struct vhost_queue, stats.bytes)},
207 offsetof(struct vhost_queue, stats.missed_pkts)},
208 {"broadcast_packets",
209 offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
210 {"multicast_packets",
211 offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
213 offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
214 {"undersize_packets",
215 offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
217 offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
218 {"size_65_to_127_packets",
219 offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
220 {"size_128_to_255_packets",
221 offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
222 {"size_256_to_511_packets",
223 offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
224 {"size_512_to_1023_packets",
225 offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
226 {"size_1024_to_1522_packets",
227 offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
228 {"size_1523_to_max_packets",
229 offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
230 {"errors_with_bad_CRC",
231 offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
234 #define VHOST_NB_XSTATS_RXPORT (sizeof(vhost_rxport_stat_strings) / \
235 sizeof(vhost_rxport_stat_strings[0]))
237 #define VHOST_NB_XSTATS_TXPORT (sizeof(vhost_txport_stat_strings) / \
238 sizeof(vhost_txport_stat_strings[0]))
241 vhost_dev_xstats_reset(struct rte_eth_dev *dev)
243 struct vhost_queue *vq = NULL;
246 for (i = 0; i < dev->data->nb_rx_queues; i++) {
247 vq = dev->data->rx_queues[i];
250 memset(&vq->stats, 0, sizeof(vq->stats));
252 for (i = 0; i < dev->data->nb_tx_queues; i++) {
253 vq = dev->data->tx_queues[i];
256 memset(&vq->stats, 0, sizeof(vq->stats));
261 vhost_dev_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
262 struct rte_eth_xstat_name *xstats_names,
263 unsigned int limit __rte_unused)
267 int nstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
271 for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
272 snprintf(xstats_names[count].name,
273 sizeof(xstats_names[count].name),
274 "rx_%s", vhost_rxport_stat_strings[t].name);
277 for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
278 snprintf(xstats_names[count].name,
279 sizeof(xstats_names[count].name),
280 "tx_%s", vhost_txport_stat_strings[t].name);
287 vhost_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
292 unsigned int count = 0;
293 struct vhost_queue *vq = NULL;
294 unsigned int nxstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
299 for (i = 0; i < dev->data->nb_rx_queues; i++) {
300 vq = dev->data->rx_queues[i];
303 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
304 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
305 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
307 for (i = 0; i < dev->data->nb_tx_queues; i++) {
308 vq = dev->data->tx_queues[i];
311 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
312 + vq->stats.missed_pkts
313 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
314 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
316 for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
317 xstats[count].value = 0;
318 for (i = 0; i < dev->data->nb_rx_queues; i++) {
319 vq = dev->data->rx_queues[i];
322 xstats[count].value +=
323 *(uint64_t *)(((char *)vq)
324 + vhost_rxport_stat_strings[t].offset);
326 xstats[count].id = count;
329 for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
330 xstats[count].value = 0;
331 for (i = 0; i < dev->data->nb_tx_queues; i++) {
332 vq = dev->data->tx_queues[i];
335 xstats[count].value +=
336 *(uint64_t *)(((char *)vq)
337 + vhost_txport_stat_strings[t].offset);
339 xstats[count].id = count;
346 vhost_count_multicast_broadcast(struct vhost_queue *vq,
347 struct rte_mbuf *mbuf)
349 struct ether_addr *ea = NULL;
350 struct vhost_stats *pstats = &vq->stats;
352 ea = rte_pktmbuf_mtod(mbuf, struct ether_addr *);
353 if (is_multicast_ether_addr(ea)) {
354 if (is_broadcast_ether_addr(ea))
355 pstats->xstats[VHOST_BROADCAST_PKT]++;
357 pstats->xstats[VHOST_MULTICAST_PKT]++;
362 vhost_update_packet_xstats(struct vhost_queue *vq,
363 struct rte_mbuf **bufs,
366 uint32_t pkt_len = 0;
369 struct vhost_stats *pstats = &vq->stats;
371 for (i = 0; i < count ; i++) {
372 pkt_len = bufs[i]->pkt_len;
374 pstats->xstats[VHOST_64_PKT]++;
375 } else if (pkt_len > 64 && pkt_len < 1024) {
376 index = (sizeof(pkt_len) * 8)
377 - __builtin_clz(pkt_len) - 5;
378 pstats->xstats[index]++;
381 pstats->xstats[VHOST_UNDERSIZE_PKT]++;
382 else if (pkt_len <= 1522)
383 pstats->xstats[VHOST_1024_TO_1522_PKT]++;
384 else if (pkt_len > 1522)
385 pstats->xstats[VHOST_1523_TO_MAX_PKT]++;
387 vhost_count_multicast_broadcast(vq, bufs[i]);
392 eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
394 struct vhost_queue *r = q;
395 uint16_t i, nb_rx = 0;
397 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
400 rte_atomic32_set(&r->while_queuing, 1);
402 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
405 /* Dequeue packets from guest TX queue */
406 nb_rx = rte_vhost_dequeue_burst(r->vid,
407 r->virtqueue_id, r->mb_pool, bufs, nb_bufs);
409 r->stats.pkts += nb_rx;
411 for (i = 0; likely(i < nb_rx); i++) {
412 bufs[i]->port = r->port;
413 r->stats.bytes += bufs[i]->pkt_len;
416 vhost_update_packet_xstats(r, bufs, nb_rx);
419 rte_atomic32_set(&r->while_queuing, 0);
425 eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
427 struct vhost_queue *r = q;
428 uint16_t i, nb_tx = 0;
430 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
433 rte_atomic32_set(&r->while_queuing, 1);
435 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
438 /* Enqueue packets to guest RX queue */
439 nb_tx = rte_vhost_enqueue_burst(r->vid,
440 r->virtqueue_id, bufs, nb_bufs);
442 r->stats.pkts += nb_tx;
443 r->stats.missed_pkts += nb_bufs - nb_tx;
445 for (i = 0; likely(i < nb_tx); i++)
446 r->stats.bytes += bufs[i]->pkt_len;
448 vhost_update_packet_xstats(r, bufs, nb_tx);
450 /* According to RFC2863 page42 section ifHCOutMulticastPkts and
451 * ifHCOutBroadcastPkts, the counters "multicast" and "broadcast"
452 * are increased when packets are not transmitted successfully.
454 for (i = nb_tx; i < nb_bufs; i++)
455 vhost_count_multicast_broadcast(r, bufs[i]);
457 for (i = 0; likely(i < nb_tx); i++)
458 rte_pktmbuf_free(bufs[i]);
460 rte_atomic32_set(&r->while_queuing, 0);
466 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
471 static inline struct internal_list *
472 find_internal_resource(char *ifname)
475 struct internal_list *list;
476 struct pmd_internal *internal;
481 pthread_mutex_lock(&internal_list_lock);
483 TAILQ_FOREACH(list, &internal_list, next) {
484 internal = list->eth_dev->data->dev_private;
485 if (!strcmp(internal->iface_name, ifname)) {
491 pthread_mutex_unlock(&internal_list_lock);
500 update_queuing_status(struct rte_eth_dev *dev)
502 struct pmd_internal *internal = dev->data->dev_private;
503 struct vhost_queue *vq;
505 int allow_queuing = 1;
507 if (rte_atomic32_read(&internal->started) == 0 ||
508 rte_atomic32_read(&internal->dev_attached) == 0)
511 /* Wait until rx/tx_pkt_burst stops accessing vhost device */
512 for (i = 0; i < dev->data->nb_rx_queues; i++) {
513 vq = dev->data->rx_queues[i];
516 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
517 while (rte_atomic32_read(&vq->while_queuing))
521 for (i = 0; i < dev->data->nb_tx_queues; i++) {
522 vq = dev->data->tx_queues[i];
525 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
526 while (rte_atomic32_read(&vq->while_queuing))
534 struct rte_eth_dev *eth_dev;
535 struct internal_list *list;
536 struct pmd_internal *internal;
537 struct vhost_queue *vq;
539 char ifname[PATH_MAX];
540 #ifdef RTE_LIBRTE_VHOST_NUMA
544 rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
545 list = find_internal_resource(ifname);
547 RTE_LOG(INFO, PMD, "Invalid device name: %s\n", ifname);
551 eth_dev = list->eth_dev;
552 internal = eth_dev->data->dev_private;
554 #ifdef RTE_LIBRTE_VHOST_NUMA
555 newnode = rte_vhost_get_numa_node(vid);
557 eth_dev->data->numa_node = newnode;
560 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
561 vq = eth_dev->data->rx_queues[i];
565 vq->internal = internal;
566 vq->port = eth_dev->data->port_id;
568 for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
569 vq = eth_dev->data->tx_queues[i];
573 vq->internal = internal;
574 vq->port = eth_dev->data->port_id;
577 for (i = 0; i < rte_vhost_get_queue_num(vid) * VIRTIO_QNUM; i++)
578 rte_vhost_enable_guest_notification(vid, i, 0);
580 eth_dev->data->dev_link.link_status = ETH_LINK_UP;
582 rte_atomic32_set(&internal->dev_attached, 1);
583 update_queuing_status(eth_dev);
585 RTE_LOG(INFO, PMD, "New connection established\n");
587 _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
593 destroy_device(int vid)
595 struct rte_eth_dev *eth_dev;
596 struct pmd_internal *internal;
597 struct vhost_queue *vq;
598 struct internal_list *list;
599 char ifname[PATH_MAX];
601 struct rte_vhost_vring_state *state;
603 rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
604 list = find_internal_resource(ifname);
606 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
609 eth_dev = list->eth_dev;
610 internal = eth_dev->data->dev_private;
612 rte_atomic32_set(&internal->dev_attached, 0);
613 update_queuing_status(eth_dev);
615 eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
617 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
618 vq = eth_dev->data->rx_queues[i];
623 for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
624 vq = eth_dev->data->tx_queues[i];
630 state = vring_states[eth_dev->data->port_id];
631 rte_spinlock_lock(&state->lock);
632 for (i = 0; i <= state->max_vring; i++) {
633 state->cur[i] = false;
634 state->seen[i] = false;
636 state->max_vring = 0;
637 rte_spinlock_unlock(&state->lock);
639 RTE_LOG(INFO, PMD, "Connection closed\n");
641 _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
645 vring_state_changed(int vid, uint16_t vring, int enable)
647 struct rte_vhost_vring_state *state;
648 struct rte_eth_dev *eth_dev;
649 struct internal_list *list;
650 char ifname[PATH_MAX];
652 rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
653 list = find_internal_resource(ifname);
655 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
659 eth_dev = list->eth_dev;
661 state = vring_states[eth_dev->data->port_id];
662 rte_spinlock_lock(&state->lock);
663 state->cur[vring] = enable;
664 state->max_vring = RTE_MAX(vring, state->max_vring);
665 rte_spinlock_unlock(&state->lock);
667 RTE_LOG(INFO, PMD, "vring%u is %s\n",
668 vring, enable ? "enabled" : "disabled");
670 _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE, NULL);
676 rte_eth_vhost_get_queue_event(uint8_t port_id,
677 struct rte_eth_vhost_queue_event *event)
679 struct rte_vhost_vring_state *state;
683 if (port_id >= RTE_MAX_ETHPORTS) {
684 RTE_LOG(ERR, PMD, "Invalid port id\n");
688 state = vring_states[port_id];
690 RTE_LOG(ERR, PMD, "Unused port\n");
694 rte_spinlock_lock(&state->lock);
695 for (i = 0; i <= state->max_vring; i++) {
696 idx = state->index++ % (state->max_vring + 1);
698 if (state->cur[idx] != state->seen[idx]) {
699 state->seen[idx] = state->cur[idx];
700 event->queue_id = idx / 2;
702 event->enable = state->cur[idx];
703 rte_spinlock_unlock(&state->lock);
707 rte_spinlock_unlock(&state->lock);
713 rte_eth_vhost_get_vid_from_port_id(uint8_t port_id)
715 struct internal_list *list;
716 struct rte_eth_dev *eth_dev;
717 struct vhost_queue *vq;
720 if (!rte_eth_dev_is_valid_port(port_id))
723 pthread_mutex_lock(&internal_list_lock);
725 TAILQ_FOREACH(list, &internal_list, next) {
726 eth_dev = list->eth_dev;
727 if (eth_dev->data->port_id == port_id) {
728 vq = eth_dev->data->rx_queues[0];
736 pthread_mutex_unlock(&internal_list_lock);
742 vhost_driver_session(void *param __rte_unused)
744 static struct virtio_net_device_ops vhost_ops;
746 /* set vhost arguments */
747 vhost_ops.new_device = new_device;
748 vhost_ops.destroy_device = destroy_device;
749 vhost_ops.vring_state_changed = vring_state_changed;
750 if (rte_vhost_driver_callback_register(&vhost_ops) < 0)
751 RTE_LOG(ERR, PMD, "Can't register callbacks\n");
753 /* start event handling */
754 rte_vhost_driver_session_start();
760 vhost_driver_session_start(void)
764 ret = pthread_create(&session_th,
765 NULL, vhost_driver_session, NULL);
767 RTE_LOG(ERR, PMD, "Can't create a thread\n");
773 vhost_driver_session_stop(void)
777 ret = pthread_cancel(session_th);
779 RTE_LOG(ERR, PMD, "Can't cancel the thread\n");
781 ret = pthread_join(session_th, NULL);
783 RTE_LOG(ERR, PMD, "Can't join the thread\n");
787 eth_dev_start(struct rte_eth_dev *dev)
789 struct pmd_internal *internal = dev->data->dev_private;
791 rte_atomic32_set(&internal->started, 1);
792 update_queuing_status(dev);
798 eth_dev_stop(struct rte_eth_dev *dev)
800 struct pmd_internal *internal = dev->data->dev_private;
802 rte_atomic32_set(&internal->started, 0);
803 update_queuing_status(dev);
807 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
808 uint16_t nb_rx_desc __rte_unused,
809 unsigned int socket_id,
810 const struct rte_eth_rxconf *rx_conf __rte_unused,
811 struct rte_mempool *mb_pool)
813 struct vhost_queue *vq;
815 vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
816 RTE_CACHE_LINE_SIZE, socket_id);
818 RTE_LOG(ERR, PMD, "Failed to allocate memory for rx queue\n");
822 vq->mb_pool = mb_pool;
823 vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
824 dev->data->rx_queues[rx_queue_id] = vq;
830 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
831 uint16_t nb_tx_desc __rte_unused,
832 unsigned int socket_id,
833 const struct rte_eth_txconf *tx_conf __rte_unused)
835 struct vhost_queue *vq;
837 vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
838 RTE_CACHE_LINE_SIZE, socket_id);
840 RTE_LOG(ERR, PMD, "Failed to allocate memory for tx queue\n");
844 vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
845 dev->data->tx_queues[tx_queue_id] = vq;
851 eth_dev_info(struct rte_eth_dev *dev,
852 struct rte_eth_dev_info *dev_info)
854 struct pmd_internal *internal;
856 internal = dev->data->dev_private;
857 if (internal == NULL) {
858 RTE_LOG(ERR, PMD, "Invalid device specified\n");
862 dev_info->driver_name = drivername;
863 dev_info->max_mac_addrs = 1;
864 dev_info->max_rx_pktlen = (uint32_t)-1;
865 dev_info->max_rx_queues = internal->max_queues;
866 dev_info->max_tx_queues = internal->max_queues;
867 dev_info->min_rx_bufsize = 0;
871 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
874 unsigned long rx_total = 0, tx_total = 0, tx_missed_total = 0;
875 unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
876 struct vhost_queue *vq;
878 for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
879 i < dev->data->nb_rx_queues; i++) {
880 if (dev->data->rx_queues[i] == NULL)
882 vq = dev->data->rx_queues[i];
883 stats->q_ipackets[i] = vq->stats.pkts;
884 rx_total += stats->q_ipackets[i];
886 stats->q_ibytes[i] = vq->stats.bytes;
887 rx_total_bytes += stats->q_ibytes[i];
890 for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
891 i < dev->data->nb_tx_queues; i++) {
892 if (dev->data->tx_queues[i] == NULL)
894 vq = dev->data->tx_queues[i];
895 stats->q_opackets[i] = vq->stats.pkts;
896 tx_missed_total += vq->stats.missed_pkts;
897 tx_total += stats->q_opackets[i];
899 stats->q_obytes[i] = vq->stats.bytes;
900 tx_total_bytes += stats->q_obytes[i];
903 stats->ipackets = rx_total;
904 stats->opackets = tx_total;
905 stats->oerrors = tx_missed_total;
906 stats->ibytes = rx_total_bytes;
907 stats->obytes = tx_total_bytes;
911 eth_stats_reset(struct rte_eth_dev *dev)
913 struct vhost_queue *vq;
916 for (i = 0; i < dev->data->nb_rx_queues; i++) {
917 if (dev->data->rx_queues[i] == NULL)
919 vq = dev->data->rx_queues[i];
923 for (i = 0; i < dev->data->nb_tx_queues; i++) {
924 if (dev->data->tx_queues[i] == NULL)
926 vq = dev->data->tx_queues[i];
929 vq->stats.missed_pkts = 0;
934 eth_queue_release(void *q)
940 eth_link_update(struct rte_eth_dev *dev __rte_unused,
941 int wait_to_complete __rte_unused)
947 * Disable features in feature_mask. Returns 0 on success.
950 rte_eth_vhost_feature_disable(uint64_t feature_mask)
952 return rte_vhost_feature_disable(feature_mask);
956 * Enable features in feature_mask. Returns 0 on success.
959 rte_eth_vhost_feature_enable(uint64_t feature_mask)
961 return rte_vhost_feature_enable(feature_mask);
964 /* Returns currently supported vhost features */
966 rte_eth_vhost_feature_get(void)
968 return rte_vhost_feature_get();
971 static const struct eth_dev_ops ops = {
972 .dev_start = eth_dev_start,
973 .dev_stop = eth_dev_stop,
974 .dev_configure = eth_dev_configure,
975 .dev_infos_get = eth_dev_info,
976 .rx_queue_setup = eth_rx_queue_setup,
977 .tx_queue_setup = eth_tx_queue_setup,
978 .rx_queue_release = eth_queue_release,
979 .tx_queue_release = eth_queue_release,
980 .link_update = eth_link_update,
981 .stats_get = eth_stats_get,
982 .stats_reset = eth_stats_reset,
983 .xstats_reset = vhost_dev_xstats_reset,
984 .xstats_get = vhost_dev_xstats_get,
985 .xstats_get_names = vhost_dev_xstats_get_names,
989 eth_dev_vhost_create(const char *name, char *iface_name, int16_t queues,
990 const unsigned numa_node, uint64_t flags)
992 struct rte_eth_dev_data *data = NULL;
993 struct pmd_internal *internal = NULL;
994 struct rte_eth_dev *eth_dev = NULL;
995 struct ether_addr *eth_addr = NULL;
996 struct rte_vhost_vring_state *vring_state = NULL;
997 struct internal_list *list = NULL;
999 RTE_LOG(INFO, PMD, "Creating VHOST-USER backend on numa socket %u\n",
1002 /* now do all data allocation - for eth_dev structure, dummy pci driver
1003 * and internal (private) data
1005 data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
1009 internal = rte_zmalloc_socket(name, sizeof(*internal), 0, numa_node);
1010 if (internal == NULL)
1013 list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
1017 /* reserve an ethdev entry */
1018 eth_dev = rte_eth_dev_allocate(name);
1019 if (eth_dev == NULL)
1022 eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
1023 if (eth_addr == NULL)
1025 *eth_addr = base_eth_addr;
1026 eth_addr->addr_bytes[5] = eth_dev->data->port_id;
1028 vring_state = rte_zmalloc_socket(name,
1029 sizeof(*vring_state), 0, numa_node);
1030 if (vring_state == NULL)
1033 /* now put it all together
1034 * - store queue data in internal,
1035 * - store numa_node info in ethdev data
1036 * - point eth_dev_data to internals
1037 * - and point eth_dev structure to new eth_dev_data structure
1039 internal->dev_name = strdup(name);
1040 if (internal->dev_name == NULL)
1042 internal->iface_name = strdup(iface_name);
1043 if (internal->iface_name == NULL)
1046 list->eth_dev = eth_dev;
1047 pthread_mutex_lock(&internal_list_lock);
1048 TAILQ_INSERT_TAIL(&internal_list, list, next);
1049 pthread_mutex_unlock(&internal_list_lock);
1051 rte_spinlock_init(&vring_state->lock);
1052 vring_states[eth_dev->data->port_id] = vring_state;
1054 data->dev_private = internal;
1055 data->port_id = eth_dev->data->port_id;
1056 memmove(data->name, eth_dev->data->name, sizeof(data->name));
1057 data->nb_rx_queues = queues;
1058 data->nb_tx_queues = queues;
1059 internal->max_queues = queues;
1060 data->dev_link = pmd_link;
1061 data->mac_addrs = eth_addr;
1063 /* We'll replace the 'data' originally allocated by eth_dev. So the
1064 * vhost PMD resources won't be shared between multi processes.
1066 eth_dev->data = data;
1067 eth_dev->dev_ops = &ops;
1068 eth_dev->driver = NULL;
1070 RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
1071 data->kdrv = RTE_KDRV_NONE;
1072 data->drv_name = internal->dev_name;
1073 data->numa_node = numa_node;
1075 /* finally assign rx and tx ops */
1076 eth_dev->rx_pkt_burst = eth_vhost_rx;
1077 eth_dev->tx_pkt_burst = eth_vhost_tx;
1079 if (rte_vhost_driver_register(iface_name, flags))
1082 /* We need only one message handling thread */
1083 if (rte_atomic16_add_return(&nb_started_ports, 1) == 1) {
1084 if (vhost_driver_session_start())
1088 return data->port_id;
1092 free(internal->dev_name);
1093 rte_free(vring_state);
1096 rte_eth_dev_release_port(eth_dev);
1105 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
1107 const char **iface_name = extra_args;
1112 *iface_name = value;
1118 open_int(const char *key __rte_unused, const char *value, void *extra_args)
1120 uint16_t *n = extra_args;
1122 if (value == NULL || extra_args == NULL)
1125 *n = (uint16_t)strtoul(value, NULL, 0);
1126 if (*n == USHRT_MAX && errno == ERANGE)
1133 rte_pmd_vhost_probe(const char *name, const char *params)
1135 struct rte_kvargs *kvlist = NULL;
1140 int client_mode = 0;
1141 int dequeue_zero_copy = 0;
1143 RTE_LOG(INFO, PMD, "Initializing pmd_vhost for %s\n", name);
1145 kvlist = rte_kvargs_parse(params, valid_arguments);
1149 if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
1150 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
1151 &open_iface, &iface_name);
1159 if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
1160 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
1161 &open_int, &queues);
1162 if (ret < 0 || queues > RTE_MAX_QUEUES_PER_PORT)
1168 if (rte_kvargs_count(kvlist, ETH_VHOST_CLIENT_ARG) == 1) {
1169 ret = rte_kvargs_process(kvlist, ETH_VHOST_CLIENT_ARG,
1170 &open_int, &client_mode);
1175 flags |= RTE_VHOST_USER_CLIENT;
1178 if (rte_kvargs_count(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY) == 1) {
1179 ret = rte_kvargs_process(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY,
1180 &open_int, &dequeue_zero_copy);
1184 if (dequeue_zero_copy)
1185 flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
1188 eth_dev_vhost_create(name, iface_name, queues, rte_socket_id(), flags);
1191 rte_kvargs_free(kvlist);
1196 rte_pmd_vhost_remove(const char *name)
1198 struct rte_eth_dev *eth_dev = NULL;
1199 struct pmd_internal *internal;
1200 struct internal_list *list;
1203 RTE_LOG(INFO, PMD, "Un-Initializing pmd_vhost for %s\n", name);
1205 /* find an ethdev entry */
1206 eth_dev = rte_eth_dev_allocated(name);
1207 if (eth_dev == NULL)
1210 internal = eth_dev->data->dev_private;
1211 if (internal == NULL)
1214 list = find_internal_resource(internal->iface_name);
1218 pthread_mutex_lock(&internal_list_lock);
1219 TAILQ_REMOVE(&internal_list, list, next);
1220 pthread_mutex_unlock(&internal_list_lock);
1223 eth_dev_stop(eth_dev);
1225 rte_vhost_driver_unregister(internal->iface_name);
1227 if (rte_atomic16_sub_return(&nb_started_ports, 1) == 0)
1228 vhost_driver_session_stop();
1230 rte_free(vring_states[eth_dev->data->port_id]);
1231 vring_states[eth_dev->data->port_id] = NULL;
1233 free(internal->dev_name);
1234 free(internal->iface_name);
1236 for (i = 0; i < eth_dev->data->nb_rx_queues; i++)
1237 rte_free(eth_dev->data->rx_queues[i]);
1238 for (i = 0; i < eth_dev->data->nb_tx_queues; i++)
1239 rte_free(eth_dev->data->tx_queues[i]);
1241 rte_free(eth_dev->data->mac_addrs);
1242 rte_free(eth_dev->data);
1245 rte_eth_dev_release_port(eth_dev);
1250 static struct rte_vdev_driver pmd_vhost_drv = {
1251 .probe = rte_pmd_vhost_probe,
1252 .remove = rte_pmd_vhost_remove,
1255 RTE_PMD_REGISTER_VDEV(net_vhost, pmd_vhost_drv);
1256 RTE_PMD_REGISTER_ALIAS(net_vhost, eth_vhost);
1257 RTE_PMD_REGISTER_PARAM_STRING(net_vhost,