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"
54 static const char *drivername = "VHOST PMD";
56 static const char *valid_arguments[] = {
62 static struct ether_addr base_eth_addr = {
74 rte_atomic32_t allow_queuing;
75 rte_atomic32_t while_queuing;
76 struct virtio_net *device;
77 struct pmd_internal *internal;
78 struct rte_mempool *mb_pool;
80 uint16_t virtqueue_id;
93 volatile uint16_t once;
96 struct internal_list {
97 TAILQ_ENTRY(internal_list) next;
98 struct rte_eth_dev *eth_dev;
101 TAILQ_HEAD(internal_list_head, internal_list);
102 static struct internal_list_head internal_list =
103 TAILQ_HEAD_INITIALIZER(internal_list);
105 static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
107 static rte_atomic16_t nb_started_ports;
108 static pthread_t session_th;
110 static struct rte_eth_link pmd_link = {
112 .link_duplex = ETH_LINK_FULL_DUPLEX,
113 .link_status = ETH_LINK_DOWN
116 struct rte_vhost_vring_state {
119 bool cur[RTE_MAX_QUEUES_PER_PORT * 2];
120 bool seen[RTE_MAX_QUEUES_PER_PORT * 2];
122 unsigned int max_vring;
125 static struct rte_vhost_vring_state *vring_states[RTE_MAX_ETHPORTS];
128 eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
130 struct vhost_queue *r = q;
131 uint16_t i, nb_rx = 0;
133 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
136 rte_atomic32_set(&r->while_queuing, 1);
138 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
141 /* Dequeue packets from guest TX queue */
142 nb_rx = rte_vhost_dequeue_burst(r->device,
143 r->virtqueue_id, r->mb_pool, bufs, nb_bufs);
147 for (i = 0; likely(i < nb_rx); i++) {
148 bufs[i]->port = r->port;
149 r->rx_bytes += bufs[i]->pkt_len;
153 rte_atomic32_set(&r->while_queuing, 0);
159 eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
161 struct vhost_queue *r = q;
162 uint16_t i, nb_tx = 0;
164 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
167 rte_atomic32_set(&r->while_queuing, 1);
169 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
172 /* Enqueue packets to guest RX queue */
173 nb_tx = rte_vhost_enqueue_burst(r->device,
174 r->virtqueue_id, bufs, nb_bufs);
177 r->missed_pkts += nb_bufs - nb_tx;
179 for (i = 0; likely(i < nb_tx); i++)
180 r->tx_bytes += bufs[i]->pkt_len;
182 for (i = 0; likely(i < nb_tx); i++)
183 rte_pktmbuf_free(bufs[i]);
185 rte_atomic32_set(&r->while_queuing, 0);
191 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
196 static inline struct internal_list *
197 find_internal_resource(char *ifname)
200 struct internal_list *list;
201 struct pmd_internal *internal;
206 pthread_mutex_lock(&internal_list_lock);
208 TAILQ_FOREACH(list, &internal_list, next) {
209 internal = list->eth_dev->data->dev_private;
210 if (!strcmp(internal->iface_name, ifname)) {
216 pthread_mutex_unlock(&internal_list_lock);
225 new_device(struct virtio_net *dev)
227 struct rte_eth_dev *eth_dev;
228 struct internal_list *list;
229 struct pmd_internal *internal;
230 struct vhost_queue *vq;
232 #ifdef RTE_LIBRTE_VHOST_NUMA
237 RTE_LOG(INFO, PMD, "Invalid argument\n");
241 list = find_internal_resource(dev->ifname);
243 RTE_LOG(INFO, PMD, "Invalid device name\n");
247 eth_dev = list->eth_dev;
248 internal = eth_dev->data->dev_private;
250 #ifdef RTE_LIBRTE_VHOST_NUMA
251 newnode = rte_vhost_get_numa_node(dev->vid);
253 eth_dev->data->numa_node = newnode;
256 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
257 vq = eth_dev->data->rx_queues[i];
261 vq->internal = internal;
262 vq->port = eth_dev->data->port_id;
264 for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
265 vq = eth_dev->data->tx_queues[i];
269 vq->internal = internal;
270 vq->port = eth_dev->data->port_id;
273 for (i = 0; i < rte_vhost_get_queue_num(dev->vid) * VIRTIO_QNUM; i++)
274 rte_vhost_enable_guest_notification(dev, i, 0);
277 eth_dev->data->dev_link.link_status = ETH_LINK_UP;
279 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
280 vq = eth_dev->data->rx_queues[i];
283 rte_atomic32_set(&vq->allow_queuing, 1);
285 for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
286 vq = eth_dev->data->tx_queues[i];
289 rte_atomic32_set(&vq->allow_queuing, 1);
292 RTE_LOG(INFO, PMD, "New connection established\n");
294 _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC);
300 destroy_device(volatile struct virtio_net *dev)
302 struct rte_eth_dev *eth_dev;
303 struct vhost_queue *vq;
307 RTE_LOG(INFO, PMD, "Invalid argument\n");
311 eth_dev = (struct rte_eth_dev *)dev->priv;
312 if (eth_dev == NULL) {
313 RTE_LOG(INFO, PMD, "Failed to find a ethdev\n");
317 /* Wait until rx/tx_pkt_burst stops accessing vhost device */
318 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
319 vq = eth_dev->data->rx_queues[i];
322 rte_atomic32_set(&vq->allow_queuing, 0);
323 while (rte_atomic32_read(&vq->while_queuing))
326 for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
327 vq = eth_dev->data->tx_queues[i];
330 rte_atomic32_set(&vq->allow_queuing, 0);
331 while (rte_atomic32_read(&vq->while_queuing))
335 eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
339 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
340 vq = eth_dev->data->rx_queues[i];
345 for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
346 vq = eth_dev->data->tx_queues[i];
352 RTE_LOG(INFO, PMD, "Connection closed\n");
354 _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC);
358 vring_state_changed(struct virtio_net *dev, uint16_t vring, int enable)
360 struct rte_vhost_vring_state *state;
361 struct rte_eth_dev *eth_dev;
362 struct internal_list *list;
365 RTE_LOG(ERR, PMD, "Invalid argument\n");
369 list = find_internal_resource(dev->ifname);
371 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", dev->ifname);
375 eth_dev = list->eth_dev;
377 state = vring_states[eth_dev->data->port_id];
378 rte_spinlock_lock(&state->lock);
379 state->cur[vring] = enable;
380 state->max_vring = RTE_MAX(vring, state->max_vring);
381 rte_spinlock_unlock(&state->lock);
383 RTE_LOG(INFO, PMD, "vring%u is %s\n",
384 vring, enable ? "enabled" : "disabled");
386 _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE);
392 rte_eth_vhost_get_queue_event(uint8_t port_id,
393 struct rte_eth_vhost_queue_event *event)
395 struct rte_vhost_vring_state *state;
399 if (port_id >= RTE_MAX_ETHPORTS) {
400 RTE_LOG(ERR, PMD, "Invalid port id\n");
404 state = vring_states[port_id];
406 RTE_LOG(ERR, PMD, "Unused port\n");
410 rte_spinlock_lock(&state->lock);
411 for (i = 0; i <= state->max_vring; i++) {
412 idx = state->index++ % (state->max_vring + 1);
414 if (state->cur[idx] != state->seen[idx]) {
415 state->seen[idx] = state->cur[idx];
416 event->queue_id = idx / 2;
418 event->enable = state->cur[idx];
419 rte_spinlock_unlock(&state->lock);
423 rte_spinlock_unlock(&state->lock);
429 vhost_driver_session(void *param __rte_unused)
431 static struct virtio_net_device_ops vhost_ops;
433 /* set vhost arguments */
434 vhost_ops.new_device = new_device;
435 vhost_ops.destroy_device = destroy_device;
436 vhost_ops.vring_state_changed = vring_state_changed;
437 if (rte_vhost_driver_callback_register(&vhost_ops) < 0)
438 RTE_LOG(ERR, PMD, "Can't register callbacks\n");
440 /* start event handling */
441 rte_vhost_driver_session_start();
447 vhost_driver_session_start(void)
451 ret = pthread_create(&session_th,
452 NULL, vhost_driver_session, NULL);
454 RTE_LOG(ERR, PMD, "Can't create a thread\n");
460 vhost_driver_session_stop(void)
464 ret = pthread_cancel(session_th);
466 RTE_LOG(ERR, PMD, "Can't cancel the thread\n");
468 ret = pthread_join(session_th, NULL);
470 RTE_LOG(ERR, PMD, "Can't join the thread\n");
474 eth_dev_start(struct rte_eth_dev *dev)
476 struct pmd_internal *internal = dev->data->dev_private;
479 if (rte_atomic16_cmpset(&internal->once, 0, 1)) {
480 ret = rte_vhost_driver_register(internal->iface_name);
485 /* We need only one message handling thread */
486 if (rte_atomic16_add_return(&nb_started_ports, 1) == 1)
487 ret = vhost_driver_session_start();
493 eth_dev_stop(struct rte_eth_dev *dev)
495 struct pmd_internal *internal = dev->data->dev_private;
497 if (rte_atomic16_cmpset(&internal->once, 1, 0))
498 rte_vhost_driver_unregister(internal->iface_name);
500 if (rte_atomic16_sub_return(&nb_started_ports, 1) == 0)
501 vhost_driver_session_stop();
505 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
506 uint16_t nb_rx_desc __rte_unused,
507 unsigned int socket_id,
508 const struct rte_eth_rxconf *rx_conf __rte_unused,
509 struct rte_mempool *mb_pool)
511 struct vhost_queue *vq;
513 vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
514 RTE_CACHE_LINE_SIZE, socket_id);
516 RTE_LOG(ERR, PMD, "Failed to allocate memory for rx queue\n");
520 vq->mb_pool = mb_pool;
521 vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
522 dev->data->rx_queues[rx_queue_id] = vq;
528 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
529 uint16_t nb_tx_desc __rte_unused,
530 unsigned int socket_id,
531 const struct rte_eth_txconf *tx_conf __rte_unused)
533 struct vhost_queue *vq;
535 vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
536 RTE_CACHE_LINE_SIZE, socket_id);
538 RTE_LOG(ERR, PMD, "Failed to allocate memory for tx queue\n");
542 vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
543 dev->data->tx_queues[tx_queue_id] = vq;
549 eth_dev_info(struct rte_eth_dev *dev,
550 struct rte_eth_dev_info *dev_info)
552 struct pmd_internal *internal;
554 internal = dev->data->dev_private;
555 if (internal == NULL) {
556 RTE_LOG(ERR, PMD, "Invalid device specified\n");
560 dev_info->driver_name = drivername;
561 dev_info->max_mac_addrs = 1;
562 dev_info->max_rx_pktlen = (uint32_t)-1;
563 dev_info->max_rx_queues = internal->max_queues;
564 dev_info->max_tx_queues = internal->max_queues;
565 dev_info->min_rx_bufsize = 0;
569 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
572 unsigned long rx_total = 0, tx_total = 0, tx_missed_total = 0;
573 unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
574 struct vhost_queue *vq;
576 for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
577 i < dev->data->nb_rx_queues; i++) {
578 if (dev->data->rx_queues[i] == NULL)
580 vq = dev->data->rx_queues[i];
581 stats->q_ipackets[i] = vq->rx_pkts;
582 rx_total += stats->q_ipackets[i];
584 stats->q_ibytes[i] = vq->rx_bytes;
585 rx_total_bytes += stats->q_ibytes[i];
588 for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
589 i < dev->data->nb_tx_queues; i++) {
590 if (dev->data->tx_queues[i] == NULL)
592 vq = dev->data->tx_queues[i];
593 stats->q_opackets[i] = vq->tx_pkts;
594 tx_missed_total += vq->missed_pkts;
595 tx_total += stats->q_opackets[i];
597 stats->q_obytes[i] = vq->tx_bytes;
598 tx_total_bytes += stats->q_obytes[i];
601 stats->ipackets = rx_total;
602 stats->opackets = tx_total;
603 stats->imissed = tx_missed_total;
604 stats->ibytes = rx_total_bytes;
605 stats->obytes = tx_total_bytes;
609 eth_stats_reset(struct rte_eth_dev *dev)
611 struct vhost_queue *vq;
614 for (i = 0; i < dev->data->nb_rx_queues; i++) {
615 if (dev->data->rx_queues[i] == NULL)
617 vq = dev->data->rx_queues[i];
621 for (i = 0; i < dev->data->nb_tx_queues; i++) {
622 if (dev->data->tx_queues[i] == NULL)
624 vq = dev->data->tx_queues[i];
632 eth_queue_release(void *q)
638 eth_link_update(struct rte_eth_dev *dev __rte_unused,
639 int wait_to_complete __rte_unused)
645 * Disable features in feature_mask. Returns 0 on success.
648 rte_eth_vhost_feature_disable(uint64_t feature_mask)
650 return rte_vhost_feature_disable(feature_mask);
654 * Enable features in feature_mask. Returns 0 on success.
657 rte_eth_vhost_feature_enable(uint64_t feature_mask)
659 return rte_vhost_feature_enable(feature_mask);
662 /* Returns currently supported vhost features */
664 rte_eth_vhost_feature_get(void)
666 return rte_vhost_feature_get();
669 static const struct eth_dev_ops ops = {
670 .dev_start = eth_dev_start,
671 .dev_stop = eth_dev_stop,
672 .dev_configure = eth_dev_configure,
673 .dev_infos_get = eth_dev_info,
674 .rx_queue_setup = eth_rx_queue_setup,
675 .tx_queue_setup = eth_tx_queue_setup,
676 .rx_queue_release = eth_queue_release,
677 .tx_queue_release = eth_queue_release,
678 .link_update = eth_link_update,
679 .stats_get = eth_stats_get,
680 .stats_reset = eth_stats_reset,
684 eth_dev_vhost_create(const char *name, char *iface_name, int16_t queues,
685 const unsigned numa_node)
687 struct rte_eth_dev_data *data = NULL;
688 struct pmd_internal *internal = NULL;
689 struct rte_eth_dev *eth_dev = NULL;
690 struct ether_addr *eth_addr = NULL;
691 struct rte_vhost_vring_state *vring_state = NULL;
692 struct internal_list *list = NULL;
694 RTE_LOG(INFO, PMD, "Creating VHOST-USER backend on numa socket %u\n",
697 /* now do all data allocation - for eth_dev structure, dummy pci driver
698 * and internal (private) data
700 data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
704 internal = rte_zmalloc_socket(name, sizeof(*internal), 0, numa_node);
705 if (internal == NULL)
708 list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
712 /* reserve an ethdev entry */
713 eth_dev = rte_eth_dev_allocate(name, RTE_ETH_DEV_VIRTUAL);
717 eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
718 if (eth_addr == NULL)
720 *eth_addr = base_eth_addr;
721 eth_addr->addr_bytes[5] = eth_dev->data->port_id;
723 vring_state = rte_zmalloc_socket(name,
724 sizeof(*vring_state), 0, numa_node);
725 if (vring_state == NULL)
728 TAILQ_INIT(ð_dev->link_intr_cbs);
730 /* now put it all together
731 * - store queue data in internal,
732 * - store numa_node info in ethdev data
733 * - point eth_dev_data to internals
734 * - and point eth_dev structure to new eth_dev_data structure
736 internal->dev_name = strdup(name);
737 if (internal->dev_name == NULL)
739 internal->iface_name = strdup(iface_name);
740 if (internal->iface_name == NULL)
743 list->eth_dev = eth_dev;
744 pthread_mutex_lock(&internal_list_lock);
745 TAILQ_INSERT_TAIL(&internal_list, list, next);
746 pthread_mutex_unlock(&internal_list_lock);
748 rte_spinlock_init(&vring_state->lock);
749 vring_states[eth_dev->data->port_id] = vring_state;
751 data->dev_private = internal;
752 data->port_id = eth_dev->data->port_id;
753 memmove(data->name, eth_dev->data->name, sizeof(data->name));
754 data->nb_rx_queues = queues;
755 data->nb_tx_queues = queues;
756 internal->max_queues = queues;
757 data->dev_link = pmd_link;
758 data->mac_addrs = eth_addr;
760 /* We'll replace the 'data' originally allocated by eth_dev. So the
761 * vhost PMD resources won't be shared between multi processes.
763 eth_dev->data = data;
764 eth_dev->dev_ops = &ops;
765 eth_dev->driver = NULL;
767 RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
768 data->kdrv = RTE_KDRV_NONE;
769 data->drv_name = internal->dev_name;
770 data->numa_node = numa_node;
772 /* finally assign rx and tx ops */
773 eth_dev->rx_pkt_burst = eth_vhost_rx;
774 eth_dev->tx_pkt_burst = eth_vhost_tx;
776 return data->port_id;
780 free(internal->dev_name);
781 rte_free(vring_state);
784 rte_eth_dev_release_port(eth_dev);
793 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
795 const char **iface_name = extra_args;
806 open_queues(const char *key __rte_unused, const char *value, void *extra_args)
808 uint16_t *q = extra_args;
810 if (value == NULL || extra_args == NULL)
813 *q = (uint16_t)strtoul(value, NULL, 0);
814 if (*q == USHRT_MAX && errno == ERANGE)
817 if (*q > RTE_MAX_QUEUES_PER_PORT)
824 rte_pmd_vhost_devinit(const char *name, const char *params)
826 struct rte_kvargs *kvlist = NULL;
831 RTE_LOG(INFO, PMD, "Initializing pmd_vhost for %s\n", name);
833 kvlist = rte_kvargs_parse(params, valid_arguments);
837 if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
838 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
839 &open_iface, &iface_name);
847 if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
848 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
849 &open_queues, &queues);
856 eth_dev_vhost_create(name, iface_name, queues, rte_socket_id());
859 rte_kvargs_free(kvlist);
864 rte_pmd_vhost_devuninit(const char *name)
866 struct rte_eth_dev *eth_dev = NULL;
867 struct pmd_internal *internal;
868 struct internal_list *list;
871 RTE_LOG(INFO, PMD, "Un-Initializing pmd_vhost for %s\n", name);
873 /* find an ethdev entry */
874 eth_dev = rte_eth_dev_allocated(name);
878 internal = eth_dev->data->dev_private;
879 if (internal == NULL)
882 list = find_internal_resource(internal->iface_name);
886 pthread_mutex_lock(&internal_list_lock);
887 TAILQ_REMOVE(&internal_list, list, next);
888 pthread_mutex_unlock(&internal_list_lock);
891 eth_dev_stop(eth_dev);
893 rte_free(vring_states[eth_dev->data->port_id]);
894 vring_states[eth_dev->data->port_id] = NULL;
896 free(internal->dev_name);
897 free(internal->iface_name);
899 for (i = 0; i < eth_dev->data->nb_rx_queues; i++)
900 rte_free(eth_dev->data->rx_queues[i]);
901 for (i = 0; i < eth_dev->data->nb_tx_queues; i++)
902 rte_free(eth_dev->data->tx_queues[i]);
904 rte_free(eth_dev->data->mac_addrs);
905 rte_free(eth_dev->data);
908 rte_eth_dev_release_port(eth_dev);
913 static struct rte_driver pmd_vhost_drv = {
916 .init = rte_pmd_vhost_devinit,
917 .uninit = rte_pmd_vhost_devuninit,
920 PMD_REGISTER_DRIVER(pmd_vhost_drv);