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;
92 volatile uint16_t once;
95 struct internal_list {
96 TAILQ_ENTRY(internal_list) next;
97 struct rte_eth_dev *eth_dev;
100 TAILQ_HEAD(internal_list_head, internal_list);
101 static struct internal_list_head internal_list =
102 TAILQ_HEAD_INITIALIZER(internal_list);
104 static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
106 static rte_atomic16_t nb_started_ports;
107 static pthread_t session_th;
109 static struct rte_eth_link pmd_link = {
111 .link_duplex = ETH_LINK_FULL_DUPLEX,
115 struct rte_vhost_vring_state {
118 bool cur[RTE_MAX_QUEUES_PER_PORT * 2];
119 bool seen[RTE_MAX_QUEUES_PER_PORT * 2];
121 unsigned int max_vring;
124 static struct rte_vhost_vring_state *vring_states[RTE_MAX_ETHPORTS];
127 eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
129 struct vhost_queue *r = q;
130 uint16_t i, nb_rx = 0;
132 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
135 rte_atomic32_set(&r->while_queuing, 1);
137 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
140 /* Dequeue packets from guest TX queue */
141 nb_rx = rte_vhost_dequeue_burst(r->device,
142 r->virtqueue_id, r->mb_pool, bufs, nb_bufs);
146 for (i = 0; likely(i < nb_rx); i++) {
147 bufs[i]->port = r->port;
148 r->rx_bytes += bufs[i]->pkt_len;
152 rte_atomic32_set(&r->while_queuing, 0);
158 eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
160 struct vhost_queue *r = q;
161 uint16_t i, nb_tx = 0;
163 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
166 rte_atomic32_set(&r->while_queuing, 1);
168 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
171 /* Enqueue packets to guest RX queue */
172 nb_tx = rte_vhost_enqueue_burst(r->device,
173 r->virtqueue_id, bufs, nb_bufs);
176 r->missed_pkts += nb_bufs - nb_tx;
178 for (i = 0; likely(i < nb_tx); i++)
179 r->tx_bytes += bufs[i]->pkt_len;
181 for (i = 0; likely(i < nb_tx); i++)
182 rte_pktmbuf_free(bufs[i]);
184 rte_atomic32_set(&r->while_queuing, 0);
190 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
195 static inline struct internal_list *
196 find_internal_resource(char *ifname)
199 struct internal_list *list;
200 struct pmd_internal *internal;
205 pthread_mutex_lock(&internal_list_lock);
207 TAILQ_FOREACH(list, &internal_list, next) {
208 internal = list->eth_dev->data->dev_private;
209 if (!strcmp(internal->iface_name, ifname)) {
215 pthread_mutex_unlock(&internal_list_lock);
224 new_device(struct virtio_net *dev)
226 struct rte_eth_dev *eth_dev;
227 struct internal_list *list;
228 struct pmd_internal *internal;
229 struct vhost_queue *vq;
233 RTE_LOG(INFO, PMD, "Invalid argument\n");
237 list = find_internal_resource(dev->ifname);
239 RTE_LOG(INFO, PMD, "Invalid device name\n");
243 eth_dev = list->eth_dev;
244 internal = eth_dev->data->dev_private;
246 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
247 vq = eth_dev->data->rx_queues[i];
251 vq->internal = internal;
252 vq->port = eth_dev->data->port_id;
253 rte_vhost_enable_guest_notification(dev, vq->virtqueue_id, 0);
255 for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
256 vq = eth_dev->data->tx_queues[i];
260 vq->internal = internal;
261 vq->port = eth_dev->data->port_id;
262 rte_vhost_enable_guest_notification(dev, vq->virtqueue_id, 0);
265 dev->flags |= VIRTIO_DEV_RUNNING;
267 eth_dev->data->dev_link.link_status = 1;
269 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
270 vq = eth_dev->data->rx_queues[i];
273 rte_atomic32_set(&vq->allow_queuing, 1);
275 for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
276 vq = eth_dev->data->tx_queues[i];
279 rte_atomic32_set(&vq->allow_queuing, 1);
282 RTE_LOG(INFO, PMD, "New connection established\n");
284 _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC);
290 destroy_device(volatile struct virtio_net *dev)
292 struct rte_eth_dev *eth_dev;
293 struct vhost_queue *vq;
297 RTE_LOG(INFO, PMD, "Invalid argument\n");
301 eth_dev = (struct rte_eth_dev *)dev->priv;
302 if (eth_dev == NULL) {
303 RTE_LOG(INFO, PMD, "Failed to find a ethdev\n");
307 /* Wait until rx/tx_pkt_burst stops accessing vhost device */
308 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
309 vq = eth_dev->data->rx_queues[i];
312 rte_atomic32_set(&vq->allow_queuing, 0);
313 while (rte_atomic32_read(&vq->while_queuing))
316 for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
317 vq = eth_dev->data->tx_queues[i];
320 rte_atomic32_set(&vq->allow_queuing, 0);
321 while (rte_atomic32_read(&vq->while_queuing))
325 eth_dev->data->dev_link.link_status = 0;
328 dev->flags &= ~VIRTIO_DEV_RUNNING;
330 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
331 vq = eth_dev->data->rx_queues[i];
336 for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
337 vq = eth_dev->data->tx_queues[i];
343 RTE_LOG(INFO, PMD, "Connection closed\n");
345 _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC);
349 vring_state_changed(struct virtio_net *dev, uint16_t vring, int enable)
351 struct rte_vhost_vring_state *state;
352 struct rte_eth_dev *eth_dev;
353 struct internal_list *list;
354 #ifdef RTE_LIBRTE_VHOST_NUMA
359 RTE_LOG(ERR, PMD, "Invalid argument\n");
363 list = find_internal_resource(dev->ifname);
365 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", dev->ifname);
369 eth_dev = list->eth_dev;
371 state = vring_states[eth_dev->data->port_id];
373 #ifdef RTE_LIBRTE_VHOST_NUMA
374 ret = get_mempolicy(&newnode, NULL, 0, dev,
375 MPOL_F_NODE | MPOL_F_ADDR);
377 RTE_LOG(ERR, PMD, "Unknown numa node\n");
381 eth_dev->data->numa_node = newnode;
383 rte_spinlock_lock(&state->lock);
384 state->cur[vring] = enable;
385 state->max_vring = RTE_MAX(vring, state->max_vring);
386 rte_spinlock_unlock(&state->lock);
388 RTE_LOG(INFO, PMD, "vring%u is %s\n",
389 vring, enable ? "enabled" : "disabled");
391 _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE);
397 rte_eth_vhost_get_queue_event(uint8_t port_id,
398 struct rte_eth_vhost_queue_event *event)
400 struct rte_vhost_vring_state *state;
404 if (port_id >= RTE_MAX_ETHPORTS) {
405 RTE_LOG(ERR, PMD, "Invalid port id\n");
409 state = vring_states[port_id];
411 RTE_LOG(ERR, PMD, "Unused port\n");
415 rte_spinlock_lock(&state->lock);
416 for (i = 0; i <= state->max_vring; i++) {
417 idx = state->index++ % (state->max_vring + 1);
419 if (state->cur[idx] != state->seen[idx]) {
420 state->seen[idx] = state->cur[idx];
421 event->queue_id = idx / 2;
423 event->enable = state->cur[idx];
424 rte_spinlock_unlock(&state->lock);
428 rte_spinlock_unlock(&state->lock);
434 vhost_driver_session(void *param __rte_unused)
436 static struct virtio_net_device_ops vhost_ops;
438 /* set vhost arguments */
439 vhost_ops.new_device = new_device;
440 vhost_ops.destroy_device = destroy_device;
441 vhost_ops.vring_state_changed = vring_state_changed;
442 if (rte_vhost_driver_callback_register(&vhost_ops) < 0)
443 RTE_LOG(ERR, PMD, "Can't register callbacks\n");
445 /* start event handling */
446 rte_vhost_driver_session_start();
452 vhost_driver_session_start(void)
456 ret = pthread_create(&session_th,
457 NULL, vhost_driver_session, NULL);
459 RTE_LOG(ERR, PMD, "Can't create a thread\n");
465 vhost_driver_session_stop(void)
469 ret = pthread_cancel(session_th);
471 RTE_LOG(ERR, PMD, "Can't cancel the thread\n");
473 ret = pthread_join(session_th, NULL);
475 RTE_LOG(ERR, PMD, "Can't join the thread\n");
479 eth_dev_start(struct rte_eth_dev *dev)
481 struct pmd_internal *internal = dev->data->dev_private;
484 if (rte_atomic16_cmpset(&internal->once, 0, 1)) {
485 ret = rte_vhost_driver_register(internal->iface_name);
490 /* We need only one message handling thread */
491 if (rte_atomic16_add_return(&nb_started_ports, 1) == 1)
492 ret = vhost_driver_session_start();
498 eth_dev_stop(struct rte_eth_dev *dev)
500 struct pmd_internal *internal = dev->data->dev_private;
502 if (rte_atomic16_cmpset(&internal->once, 1, 0))
503 rte_vhost_driver_unregister(internal->iface_name);
505 if (rte_atomic16_sub_return(&nb_started_ports, 1) == 0)
506 vhost_driver_session_stop();
510 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
511 uint16_t nb_rx_desc __rte_unused,
512 unsigned int socket_id,
513 const struct rte_eth_rxconf *rx_conf __rte_unused,
514 struct rte_mempool *mb_pool)
516 struct vhost_queue *vq;
518 vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
519 RTE_CACHE_LINE_SIZE, socket_id);
521 RTE_LOG(ERR, PMD, "Failed to allocate memory for rx queue\n");
525 vq->mb_pool = mb_pool;
526 vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
527 dev->data->rx_queues[rx_queue_id] = vq;
533 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
534 uint16_t nb_tx_desc __rte_unused,
535 unsigned int socket_id,
536 const struct rte_eth_txconf *tx_conf __rte_unused)
538 struct vhost_queue *vq;
540 vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
541 RTE_CACHE_LINE_SIZE, socket_id);
543 RTE_LOG(ERR, PMD, "Failed to allocate memory for tx queue\n");
547 vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
548 dev->data->tx_queues[tx_queue_id] = vq;
554 eth_dev_info(struct rte_eth_dev *dev,
555 struct rte_eth_dev_info *dev_info)
557 dev_info->driver_name = drivername;
558 dev_info->max_mac_addrs = 1;
559 dev_info->max_rx_pktlen = (uint32_t)-1;
560 dev_info->max_rx_queues = dev->data->nb_rx_queues;
561 dev_info->max_tx_queues = dev->data->nb_tx_queues;
562 dev_info->min_rx_bufsize = 0;
566 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
569 unsigned long rx_total = 0, tx_total = 0, tx_missed_total = 0;
570 unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
571 struct vhost_queue *vq;
573 for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
574 i < dev->data->nb_rx_queues; i++) {
575 if (dev->data->rx_queues[i] == NULL)
577 vq = dev->data->rx_queues[i];
578 stats->q_ipackets[i] = vq->rx_pkts;
579 rx_total += stats->q_ipackets[i];
581 stats->q_ibytes[i] = vq->rx_bytes;
582 rx_total_bytes += stats->q_ibytes[i];
585 for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
586 i < dev->data->nb_tx_queues; i++) {
587 if (dev->data->tx_queues[i] == NULL)
589 vq = dev->data->tx_queues[i];
590 stats->q_opackets[i] = vq->tx_pkts;
591 tx_missed_total += vq->missed_pkts;
592 tx_total += stats->q_opackets[i];
594 stats->q_obytes[i] = vq->tx_bytes;
595 tx_total_bytes += stats->q_obytes[i];
598 stats->ipackets = rx_total;
599 stats->opackets = tx_total;
600 stats->imissed = tx_missed_total;
601 stats->ibytes = rx_total_bytes;
602 stats->obytes = tx_total_bytes;
606 eth_stats_reset(struct rte_eth_dev *dev)
608 struct vhost_queue *vq;
611 for (i = 0; i < dev->data->nb_rx_queues; i++) {
612 if (dev->data->rx_queues[i] == NULL)
614 vq = dev->data->rx_queues[i];
618 for (i = 0; i < dev->data->nb_tx_queues; i++) {
619 if (dev->data->tx_queues[i] == NULL)
621 vq = dev->data->tx_queues[i];
629 eth_queue_release(void *q)
635 eth_link_update(struct rte_eth_dev *dev __rte_unused,
636 int wait_to_complete __rte_unused)
642 * Disable features in feature_mask. Returns 0 on success.
645 rte_eth_vhost_feature_disable(uint64_t feature_mask)
647 return rte_vhost_feature_disable(feature_mask);
651 * Enable features in feature_mask. Returns 0 on success.
654 rte_eth_vhost_feature_enable(uint64_t feature_mask)
656 return rte_vhost_feature_enable(feature_mask);
659 /* Returns currently supported vhost features */
661 rte_eth_vhost_feature_get(void)
663 return rte_vhost_feature_get();
666 static const struct eth_dev_ops ops = {
667 .dev_start = eth_dev_start,
668 .dev_stop = eth_dev_stop,
669 .dev_configure = eth_dev_configure,
670 .dev_infos_get = eth_dev_info,
671 .rx_queue_setup = eth_rx_queue_setup,
672 .tx_queue_setup = eth_tx_queue_setup,
673 .rx_queue_release = eth_queue_release,
674 .tx_queue_release = eth_queue_release,
675 .link_update = eth_link_update,
676 .stats_get = eth_stats_get,
677 .stats_reset = eth_stats_reset,
681 eth_dev_vhost_create(const char *name, char *iface_name, int16_t queues,
682 const unsigned numa_node)
684 struct rte_eth_dev_data *data = NULL;
685 struct pmd_internal *internal = NULL;
686 struct rte_eth_dev *eth_dev = NULL;
687 struct ether_addr *eth_addr = NULL;
688 struct rte_vhost_vring_state *vring_state = NULL;
689 struct internal_list *list = NULL;
691 RTE_LOG(INFO, PMD, "Creating VHOST-USER backend on numa socket %u\n",
694 /* now do all data allocation - for eth_dev structure, dummy pci driver
695 * and internal (private) data
697 data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
701 internal = rte_zmalloc_socket(name, sizeof(*internal), 0, numa_node);
702 if (internal == NULL)
705 list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
709 /* reserve an ethdev entry */
710 eth_dev = rte_eth_dev_allocate(name, RTE_ETH_DEV_VIRTUAL);
714 eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
715 if (eth_addr == NULL)
717 *eth_addr = base_eth_addr;
718 eth_addr->addr_bytes[5] = eth_dev->data->port_id;
720 vring_state = rte_zmalloc_socket(name,
721 sizeof(*vring_state), 0, numa_node);
722 if (vring_state == NULL)
725 TAILQ_INIT(ð_dev->link_intr_cbs);
727 /* now put it all together
728 * - store queue data in internal,
729 * - store numa_node info in ethdev data
730 * - point eth_dev_data to internals
731 * - and point eth_dev structure to new eth_dev_data structure
733 internal->dev_name = strdup(name);
734 if (internal->dev_name == NULL)
736 internal->iface_name = strdup(iface_name);
737 if (internal->iface_name == NULL)
740 list->eth_dev = eth_dev;
741 pthread_mutex_lock(&internal_list_lock);
742 TAILQ_INSERT_TAIL(&internal_list, list, next);
743 pthread_mutex_unlock(&internal_list_lock);
745 rte_spinlock_init(&vring_state->lock);
746 vring_states[eth_dev->data->port_id] = vring_state;
748 data->dev_private = internal;
749 data->port_id = eth_dev->data->port_id;
750 memmove(data->name, eth_dev->data->name, sizeof(data->name));
751 data->nb_rx_queues = queues;
752 data->nb_tx_queues = queues;
753 data->dev_link = pmd_link;
754 data->mac_addrs = eth_addr;
756 /* We'll replace the 'data' originally allocated by eth_dev. So the
757 * vhost PMD resources won't be shared between multi processes.
759 eth_dev->data = data;
760 eth_dev->dev_ops = &ops;
761 eth_dev->driver = NULL;
763 RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
764 data->kdrv = RTE_KDRV_NONE;
765 data->drv_name = internal->dev_name;
766 data->numa_node = numa_node;
768 /* finally assign rx and tx ops */
769 eth_dev->rx_pkt_burst = eth_vhost_rx;
770 eth_dev->tx_pkt_burst = eth_vhost_tx;
772 return data->port_id;
776 free(internal->dev_name);
777 rte_free(vring_state);
780 rte_eth_dev_release_port(eth_dev);
789 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
791 const char **iface_name = extra_args;
802 open_queues(const char *key __rte_unused, const char *value, void *extra_args)
804 uint16_t *q = extra_args;
806 if (value == NULL || extra_args == NULL)
809 *q = (uint16_t)strtoul(value, NULL, 0);
810 if (*q == USHRT_MAX && errno == ERANGE)
813 if (*q > RTE_MAX_QUEUES_PER_PORT)
820 rte_pmd_vhost_devinit(const char *name, const char *params)
822 struct rte_kvargs *kvlist = NULL;
827 RTE_LOG(INFO, PMD, "Initializing pmd_vhost for %s\n", name);
829 kvlist = rte_kvargs_parse(params, valid_arguments);
833 if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
834 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
835 &open_iface, &iface_name);
843 if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
844 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
845 &open_queues, &queues);
852 eth_dev_vhost_create(name, iface_name, queues, rte_socket_id());
855 rte_kvargs_free(kvlist);
860 rte_pmd_vhost_devuninit(const char *name)
862 struct rte_eth_dev *eth_dev = NULL;
863 struct pmd_internal *internal;
864 struct internal_list *list;
867 RTE_LOG(INFO, PMD, "Un-Initializing pmd_vhost for %s\n", name);
869 /* find an ethdev entry */
870 eth_dev = rte_eth_dev_allocated(name);
874 internal = eth_dev->data->dev_private;
875 if (internal == NULL)
878 list = find_internal_resource(internal->iface_name);
882 pthread_mutex_lock(&internal_list_lock);
883 TAILQ_REMOVE(&internal_list, list, next);
884 pthread_mutex_unlock(&internal_list_lock);
887 eth_dev_stop(eth_dev);
889 rte_free(vring_states[eth_dev->data->port_id]);
890 vring_states[eth_dev->data->port_id] = NULL;
892 free(internal->dev_name);
893 free(internal->iface_name);
895 for (i = 0; i < eth_dev->data->nb_rx_queues; i++)
896 rte_free(eth_dev->data->rx_queues[i]);
897 for (i = 0; i < eth_dev->data->nb_tx_queues; i++)
898 rte_free(eth_dev->data->tx_queues[i]);
900 rte_free(eth_dev->data->mac_addrs);
901 rte_free(eth_dev->data);
904 rte_eth_dev_release_port(eth_dev);
909 static struct rte_driver pmd_vhost_drv = {
912 .init = rte_pmd_vhost_devinit,
913 .uninit = rte_pmd_vhost_devuninit,
916 PMD_REGISTER_DRIVER(pmd_vhost_drv);