1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2016 IGEL Co., Ltd.
3 * Copyright(c) 2016-2018 Intel Corporation
10 #include <rte_ethdev_driver.h>
11 #include <rte_ethdev_vdev.h>
12 #include <rte_malloc.h>
13 #include <rte_memcpy.h>
14 #include <rte_bus_vdev.h>
15 #include <rte_kvargs.h>
16 #include <rte_vhost.h>
17 #include <rte_spinlock.h>
19 #include "rte_eth_vhost.h"
21 static int vhost_logtype;
23 #define VHOST_LOG(level, ...) \
24 rte_log(RTE_LOG_ ## level, vhost_logtype, __VA_ARGS__)
26 enum {VIRTIO_RXQ, VIRTIO_TXQ, VIRTIO_QNUM};
28 #define ETH_VHOST_IFACE_ARG "iface"
29 #define ETH_VHOST_QUEUES_ARG "queues"
30 #define ETH_VHOST_CLIENT_ARG "client"
31 #define ETH_VHOST_DEQUEUE_ZERO_COPY "dequeue-zero-copy"
32 #define ETH_VHOST_IOMMU_SUPPORT "iommu-support"
33 #define ETH_VHOST_POSTCOPY_SUPPORT "postcopy-support"
34 #define ETH_VHOST_VIRTIO_NET_F_HOST_TSO "tso"
35 #define VHOST_MAX_PKT_BURST 32
37 static const char *valid_arguments[] = {
41 ETH_VHOST_DEQUEUE_ZERO_COPY,
42 ETH_VHOST_IOMMU_SUPPORT,
43 ETH_VHOST_POSTCOPY_SUPPORT,
44 ETH_VHOST_VIRTIO_NET_F_HOST_TSO,
48 static struct rte_ether_addr base_eth_addr = {
59 enum vhost_xstats_pkts {
60 VHOST_UNDERSIZE_PKT = 0,
65 VHOST_512_TO_1023_PKT,
66 VHOST_1024_TO_1522_PKT,
67 VHOST_1523_TO_MAX_PKT,
72 VHOST_ERRORS_FRAGMENTED,
74 VHOST_UNKNOWN_PROTOCOL,
82 uint64_t xstats[VHOST_XSTATS_MAX];
87 rte_atomic32_t allow_queuing;
88 rte_atomic32_t while_queuing;
89 struct pmd_internal *internal;
90 struct rte_mempool *mb_pool;
92 uint16_t virtqueue_id;
93 struct vhost_stats stats;
97 rte_atomic32_t dev_attached;
102 rte_atomic32_t started;
106 struct internal_list {
107 TAILQ_ENTRY(internal_list) next;
108 struct rte_eth_dev *eth_dev;
111 TAILQ_HEAD(internal_list_head, internal_list);
112 static struct internal_list_head internal_list =
113 TAILQ_HEAD_INITIALIZER(internal_list);
115 static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
117 static struct rte_eth_link pmd_link = {
119 .link_duplex = ETH_LINK_FULL_DUPLEX,
120 .link_status = ETH_LINK_DOWN
123 struct rte_vhost_vring_state {
126 bool cur[RTE_MAX_QUEUES_PER_PORT * 2];
127 bool seen[RTE_MAX_QUEUES_PER_PORT * 2];
129 unsigned int max_vring;
132 static struct rte_vhost_vring_state *vring_states[RTE_MAX_ETHPORTS];
134 #define VHOST_XSTATS_NAME_SIZE 64
136 struct vhost_xstats_name_off {
137 char name[VHOST_XSTATS_NAME_SIZE];
141 /* [rx]_is prepended to the name string here */
142 static const struct vhost_xstats_name_off vhost_rxport_stat_strings[] = {
144 offsetof(struct vhost_queue, stats.pkts)},
146 offsetof(struct vhost_queue, stats.bytes)},
148 offsetof(struct vhost_queue, stats.missed_pkts)},
149 {"broadcast_packets",
150 offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
151 {"multicast_packets",
152 offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
154 offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
155 {"undersize_packets",
156 offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
158 offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
159 {"size_65_to_127_packets",
160 offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
161 {"size_128_to_255_packets",
162 offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
163 {"size_256_to_511_packets",
164 offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
165 {"size_512_to_1023_packets",
166 offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
167 {"size_1024_to_1522_packets",
168 offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
169 {"size_1523_to_max_packets",
170 offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
171 {"errors_with_bad_CRC",
172 offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
173 {"fragmented_errors",
174 offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_FRAGMENTED])},
176 offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_JABBER])},
177 {"unknown_protos_packets",
178 offsetof(struct vhost_queue, stats.xstats[VHOST_UNKNOWN_PROTOCOL])},
181 /* [tx]_ is prepended to the name string here */
182 static const struct vhost_xstats_name_off vhost_txport_stat_strings[] = {
184 offsetof(struct vhost_queue, stats.pkts)},
186 offsetof(struct vhost_queue, stats.bytes)},
188 offsetof(struct vhost_queue, stats.missed_pkts)},
189 {"broadcast_packets",
190 offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
191 {"multicast_packets",
192 offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
194 offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
195 {"undersize_packets",
196 offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
198 offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
199 {"size_65_to_127_packets",
200 offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
201 {"size_128_to_255_packets",
202 offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
203 {"size_256_to_511_packets",
204 offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
205 {"size_512_to_1023_packets",
206 offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
207 {"size_1024_to_1522_packets",
208 offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
209 {"size_1523_to_max_packets",
210 offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
211 {"errors_with_bad_CRC",
212 offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
215 #define VHOST_NB_XSTATS_RXPORT (sizeof(vhost_rxport_stat_strings) / \
216 sizeof(vhost_rxport_stat_strings[0]))
218 #define VHOST_NB_XSTATS_TXPORT (sizeof(vhost_txport_stat_strings) / \
219 sizeof(vhost_txport_stat_strings[0]))
222 vhost_dev_xstats_reset(struct rte_eth_dev *dev)
224 struct vhost_queue *vq = NULL;
227 for (i = 0; i < dev->data->nb_rx_queues; i++) {
228 vq = dev->data->rx_queues[i];
231 memset(&vq->stats, 0, sizeof(vq->stats));
233 for (i = 0; i < dev->data->nb_tx_queues; i++) {
234 vq = dev->data->tx_queues[i];
237 memset(&vq->stats, 0, sizeof(vq->stats));
244 vhost_dev_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
245 struct rte_eth_xstat_name *xstats_names,
246 unsigned int limit __rte_unused)
250 int nstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
254 for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
255 snprintf(xstats_names[count].name,
256 sizeof(xstats_names[count].name),
257 "rx_%s", vhost_rxport_stat_strings[t].name);
260 for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
261 snprintf(xstats_names[count].name,
262 sizeof(xstats_names[count].name),
263 "tx_%s", vhost_txport_stat_strings[t].name);
270 vhost_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
275 unsigned int count = 0;
276 struct vhost_queue *vq = NULL;
277 unsigned int nxstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
282 for (i = 0; i < dev->data->nb_rx_queues; i++) {
283 vq = dev->data->rx_queues[i];
286 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
287 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
288 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
290 for (i = 0; i < dev->data->nb_tx_queues; i++) {
291 vq = dev->data->tx_queues[i];
294 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
295 + vq->stats.missed_pkts
296 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
297 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
299 for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
300 xstats[count].value = 0;
301 for (i = 0; i < dev->data->nb_rx_queues; i++) {
302 vq = dev->data->rx_queues[i];
305 xstats[count].value +=
306 *(uint64_t *)(((char *)vq)
307 + vhost_rxport_stat_strings[t].offset);
309 xstats[count].id = count;
312 for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
313 xstats[count].value = 0;
314 for (i = 0; i < dev->data->nb_tx_queues; i++) {
315 vq = dev->data->tx_queues[i];
318 xstats[count].value +=
319 *(uint64_t *)(((char *)vq)
320 + vhost_txport_stat_strings[t].offset);
322 xstats[count].id = count;
329 vhost_count_multicast_broadcast(struct vhost_queue *vq,
330 struct rte_mbuf *mbuf)
332 struct rte_ether_addr *ea = NULL;
333 struct vhost_stats *pstats = &vq->stats;
335 ea = rte_pktmbuf_mtod(mbuf, struct rte_ether_addr *);
336 if (rte_is_multicast_ether_addr(ea)) {
337 if (rte_is_broadcast_ether_addr(ea))
338 pstats->xstats[VHOST_BROADCAST_PKT]++;
340 pstats->xstats[VHOST_MULTICAST_PKT]++;
345 vhost_update_packet_xstats(struct vhost_queue *vq,
346 struct rte_mbuf **bufs,
349 uint32_t pkt_len = 0;
352 struct vhost_stats *pstats = &vq->stats;
354 for (i = 0; i < count ; i++) {
355 pkt_len = bufs[i]->pkt_len;
357 pstats->xstats[VHOST_64_PKT]++;
358 } else if (pkt_len > 64 && pkt_len < 1024) {
359 index = (sizeof(pkt_len) * 8)
360 - __builtin_clz(pkt_len) - 5;
361 pstats->xstats[index]++;
364 pstats->xstats[VHOST_UNDERSIZE_PKT]++;
365 else if (pkt_len <= 1522)
366 pstats->xstats[VHOST_1024_TO_1522_PKT]++;
367 else if (pkt_len > 1522)
368 pstats->xstats[VHOST_1523_TO_MAX_PKT]++;
370 vhost_count_multicast_broadcast(vq, bufs[i]);
375 eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
377 struct vhost_queue *r = q;
378 uint16_t i, nb_rx = 0;
379 uint16_t nb_receive = nb_bufs;
381 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
384 rte_atomic32_set(&r->while_queuing, 1);
386 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
389 /* Dequeue packets from guest TX queue */
392 uint16_t num = (uint16_t)RTE_MIN(nb_receive,
393 VHOST_MAX_PKT_BURST);
395 nb_pkts = rte_vhost_dequeue_burst(r->vid, r->virtqueue_id,
396 r->mb_pool, &bufs[nb_rx],
400 nb_receive -= nb_pkts;
405 r->stats.pkts += nb_rx;
407 for (i = 0; likely(i < nb_rx); i++) {
408 bufs[i]->port = r->port;
409 bufs[i]->vlan_tci = 0;
411 if (r->internal->vlan_strip)
412 rte_vlan_strip(bufs[i]);
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;
430 uint16_t nb_send = 0;
432 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
435 rte_atomic32_set(&r->while_queuing, 1);
437 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
440 for (i = 0; i < nb_bufs; i++) {
441 struct rte_mbuf *m = bufs[i];
443 /* Do VLAN tag insertion */
444 if (m->ol_flags & PKT_TX_VLAN_PKT) {
445 int error = rte_vlan_insert(&m);
446 if (unlikely(error)) {
456 /* Enqueue packets to guest RX queue */
459 uint16_t num = (uint16_t)RTE_MIN(nb_send,
460 VHOST_MAX_PKT_BURST);
462 nb_pkts = rte_vhost_enqueue_burst(r->vid, r->virtqueue_id,
471 r->stats.pkts += nb_tx;
472 r->stats.missed_pkts += nb_bufs - nb_tx;
474 for (i = 0; likely(i < nb_tx); i++)
475 r->stats.bytes += bufs[i]->pkt_len;
477 vhost_update_packet_xstats(r, bufs, nb_tx);
479 /* According to RFC2863 page42 section ifHCOutMulticastPkts and
480 * ifHCOutBroadcastPkts, the counters "multicast" and "broadcast"
481 * are increased when packets are not transmitted successfully.
483 for (i = nb_tx; i < nb_bufs; i++)
484 vhost_count_multicast_broadcast(r, bufs[i]);
486 for (i = 0; likely(i < nb_tx); i++)
487 rte_pktmbuf_free(bufs[i]);
489 rte_atomic32_set(&r->while_queuing, 0);
495 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
497 struct pmd_internal *internal = dev->data->dev_private;
498 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
500 internal->vlan_strip = !!(rxmode->offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
505 static inline struct internal_list *
506 find_internal_resource(char *ifname)
509 struct internal_list *list;
510 struct pmd_internal *internal;
515 pthread_mutex_lock(&internal_list_lock);
517 TAILQ_FOREACH(list, &internal_list, next) {
518 internal = list->eth_dev->data->dev_private;
519 if (!strcmp(internal->iface_name, ifname)) {
525 pthread_mutex_unlock(&internal_list_lock);
534 eth_rxq_intr_enable(struct rte_eth_dev *dev, uint16_t qid)
536 struct rte_vhost_vring vring;
537 struct vhost_queue *vq;
540 vq = dev->data->rx_queues[qid];
542 VHOST_LOG(ERR, "rxq%d is not setup yet\n", qid);
546 ret = rte_vhost_get_vhost_vring(vq->vid, (qid << 1) + 1, &vring);
548 VHOST_LOG(ERR, "Failed to get rxq%d's vring\n", qid);
551 VHOST_LOG(INFO, "Enable interrupt for rxq%d\n", qid);
552 rte_vhost_enable_guest_notification(vq->vid, (qid << 1) + 1, 1);
559 eth_rxq_intr_disable(struct rte_eth_dev *dev, uint16_t qid)
561 struct rte_vhost_vring vring;
562 struct vhost_queue *vq;
565 vq = dev->data->rx_queues[qid];
567 VHOST_LOG(ERR, "rxq%d is not setup yet\n", qid);
571 ret = rte_vhost_get_vhost_vring(vq->vid, (qid << 1) + 1, &vring);
573 VHOST_LOG(ERR, "Failed to get rxq%d's vring", qid);
576 VHOST_LOG(INFO, "Disable interrupt for rxq%d\n", qid);
577 rte_vhost_enable_guest_notification(vq->vid, (qid << 1) + 1, 0);
584 eth_vhost_uninstall_intr(struct rte_eth_dev *dev)
586 struct rte_intr_handle *intr_handle = dev->intr_handle;
589 if (intr_handle->intr_vec)
590 free(intr_handle->intr_vec);
594 dev->intr_handle = NULL;
598 eth_vhost_install_intr(struct rte_eth_dev *dev)
600 struct rte_vhost_vring vring;
601 struct vhost_queue *vq;
603 int nb_rxq = dev->data->nb_rx_queues;
607 /* uninstall firstly if we are reconnecting */
608 if (dev->intr_handle)
609 eth_vhost_uninstall_intr(dev);
611 dev->intr_handle = malloc(sizeof(*dev->intr_handle));
612 if (!dev->intr_handle) {
613 VHOST_LOG(ERR, "Fail to allocate intr_handle\n");
616 memset(dev->intr_handle, 0, sizeof(*dev->intr_handle));
618 dev->intr_handle->efd_counter_size = sizeof(uint64_t);
620 dev->intr_handle->intr_vec =
621 malloc(nb_rxq * sizeof(dev->intr_handle->intr_vec[0]));
623 if (!dev->intr_handle->intr_vec) {
625 "Failed to allocate memory for interrupt vector\n");
626 free(dev->intr_handle);
630 VHOST_LOG(INFO, "Prepare intr vec\n");
631 for (i = 0; i < nb_rxq; i++) {
632 vq = dev->data->rx_queues[i];
634 VHOST_LOG(INFO, "rxq-%d not setup yet, skip!\n", i);
638 ret = rte_vhost_get_vhost_vring(vq->vid, (i << 1) + 1, &vring);
641 "Failed to get rxq-%d's vring, skip!\n", i);
645 if (vring.kickfd < 0) {
647 "rxq-%d's kickfd is invalid, skip!\n", i);
650 dev->intr_handle->intr_vec[i] = RTE_INTR_VEC_RXTX_OFFSET + i;
651 dev->intr_handle->efds[i] = vring.kickfd;
653 VHOST_LOG(INFO, "Installed intr vec for rxq-%d\n", i);
656 dev->intr_handle->nb_efd = count;
657 dev->intr_handle->max_intr = count + 1;
658 dev->intr_handle->type = RTE_INTR_HANDLE_VDEV;
664 update_queuing_status(struct rte_eth_dev *dev)
666 struct pmd_internal *internal = dev->data->dev_private;
667 struct vhost_queue *vq;
669 int allow_queuing = 1;
671 if (!dev->data->rx_queues || !dev->data->tx_queues)
674 if (rte_atomic32_read(&internal->started) == 0 ||
675 rte_atomic32_read(&internal->dev_attached) == 0)
678 /* Wait until rx/tx_pkt_burst stops accessing vhost device */
679 for (i = 0; i < dev->data->nb_rx_queues; i++) {
680 vq = dev->data->rx_queues[i];
683 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
684 while (rte_atomic32_read(&vq->while_queuing))
688 for (i = 0; i < dev->data->nb_tx_queues; i++) {
689 vq = dev->data->tx_queues[i];
692 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
693 while (rte_atomic32_read(&vq->while_queuing))
699 queue_setup(struct rte_eth_dev *eth_dev, struct pmd_internal *internal)
701 struct vhost_queue *vq;
704 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
705 vq = eth_dev->data->rx_queues[i];
708 vq->vid = internal->vid;
709 vq->internal = internal;
710 vq->port = eth_dev->data->port_id;
712 for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
713 vq = eth_dev->data->tx_queues[i];
716 vq->vid = internal->vid;
717 vq->internal = internal;
718 vq->port = eth_dev->data->port_id;
725 struct rte_eth_dev *eth_dev;
726 struct internal_list *list;
727 struct pmd_internal *internal;
728 struct rte_eth_conf *dev_conf;
730 char ifname[PATH_MAX];
731 #ifdef RTE_LIBRTE_VHOST_NUMA
735 rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
736 list = find_internal_resource(ifname);
738 VHOST_LOG(INFO, "Invalid device name: %s\n", ifname);
742 eth_dev = list->eth_dev;
743 internal = eth_dev->data->dev_private;
744 dev_conf = ð_dev->data->dev_conf;
746 #ifdef RTE_LIBRTE_VHOST_NUMA
747 newnode = rte_vhost_get_numa_node(vid);
749 eth_dev->data->numa_node = newnode;
753 if (rte_atomic32_read(&internal->started) == 1) {
754 queue_setup(eth_dev, internal);
756 if (dev_conf->intr_conf.rxq) {
757 if (eth_vhost_install_intr(eth_dev) < 0) {
759 "Failed to install interrupt handler.");
764 VHOST_LOG(INFO, "RX/TX queues not exist yet\n");
767 for (i = 0; i < rte_vhost_get_vring_num(vid); i++)
768 rte_vhost_enable_guest_notification(vid, i, 0);
770 rte_vhost_get_mtu(vid, ð_dev->data->mtu);
772 eth_dev->data->dev_link.link_status = ETH_LINK_UP;
774 rte_atomic32_set(&internal->dev_attached, 1);
775 update_queuing_status(eth_dev);
777 VHOST_LOG(INFO, "Vhost device %d created\n", vid);
779 _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
785 destroy_device(int vid)
787 struct rte_eth_dev *eth_dev;
788 struct pmd_internal *internal;
789 struct vhost_queue *vq;
790 struct internal_list *list;
791 char ifname[PATH_MAX];
793 struct rte_vhost_vring_state *state;
795 rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
796 list = find_internal_resource(ifname);
798 VHOST_LOG(ERR, "Invalid interface name: %s\n", ifname);
801 eth_dev = list->eth_dev;
802 internal = eth_dev->data->dev_private;
804 rte_atomic32_set(&internal->dev_attached, 0);
805 update_queuing_status(eth_dev);
807 eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
809 if (eth_dev->data->rx_queues && eth_dev->data->tx_queues) {
810 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
811 vq = eth_dev->data->rx_queues[i];
816 for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
817 vq = eth_dev->data->tx_queues[i];
824 state = vring_states[eth_dev->data->port_id];
825 rte_spinlock_lock(&state->lock);
826 for (i = 0; i <= state->max_vring; i++) {
827 state->cur[i] = false;
828 state->seen[i] = false;
830 state->max_vring = 0;
831 rte_spinlock_unlock(&state->lock);
833 VHOST_LOG(INFO, "Vhost device %d destroyed\n", vid);
834 eth_vhost_uninstall_intr(eth_dev);
836 _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
840 vring_state_changed(int vid, uint16_t vring, int enable)
842 struct rte_vhost_vring_state *state;
843 struct rte_eth_dev *eth_dev;
844 struct internal_list *list;
845 char ifname[PATH_MAX];
847 rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
848 list = find_internal_resource(ifname);
850 VHOST_LOG(ERR, "Invalid interface name: %s\n", ifname);
854 eth_dev = list->eth_dev;
856 state = vring_states[eth_dev->data->port_id];
857 rte_spinlock_lock(&state->lock);
858 if (state->cur[vring] == enable) {
859 rte_spinlock_unlock(&state->lock);
862 state->cur[vring] = enable;
863 state->max_vring = RTE_MAX(vring, state->max_vring);
864 rte_spinlock_unlock(&state->lock);
866 VHOST_LOG(INFO, "vring%u is %s\n",
867 vring, enable ? "enabled" : "disabled");
869 _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE, NULL);
874 static struct vhost_device_ops vhost_ops = {
875 .new_device = new_device,
876 .destroy_device = destroy_device,
877 .vring_state_changed = vring_state_changed,
881 rte_eth_vhost_get_queue_event(uint16_t port_id,
882 struct rte_eth_vhost_queue_event *event)
884 struct rte_vhost_vring_state *state;
888 if (port_id >= RTE_MAX_ETHPORTS) {
889 VHOST_LOG(ERR, "Invalid port id\n");
893 state = vring_states[port_id];
895 VHOST_LOG(ERR, "Unused port\n");
899 rte_spinlock_lock(&state->lock);
900 for (i = 0; i <= state->max_vring; i++) {
901 idx = state->index++ % (state->max_vring + 1);
903 if (state->cur[idx] != state->seen[idx]) {
904 state->seen[idx] = state->cur[idx];
905 event->queue_id = idx / 2;
907 event->enable = state->cur[idx];
908 rte_spinlock_unlock(&state->lock);
912 rte_spinlock_unlock(&state->lock);
918 rte_eth_vhost_get_vid_from_port_id(uint16_t port_id)
920 struct internal_list *list;
921 struct rte_eth_dev *eth_dev;
922 struct vhost_queue *vq;
925 if (!rte_eth_dev_is_valid_port(port_id))
928 pthread_mutex_lock(&internal_list_lock);
930 TAILQ_FOREACH(list, &internal_list, next) {
931 eth_dev = list->eth_dev;
932 if (eth_dev->data->port_id == port_id) {
933 vq = eth_dev->data->rx_queues[0];
941 pthread_mutex_unlock(&internal_list_lock);
947 eth_dev_start(struct rte_eth_dev *eth_dev)
949 struct pmd_internal *internal = eth_dev->data->dev_private;
950 struct rte_eth_conf *dev_conf = ð_dev->data->dev_conf;
952 queue_setup(eth_dev, internal);
954 if (rte_atomic32_read(&internal->dev_attached) == 1) {
955 if (dev_conf->intr_conf.rxq) {
956 if (eth_vhost_install_intr(eth_dev) < 0) {
958 "Failed to install interrupt handler.");
964 rte_atomic32_set(&internal->started, 1);
965 update_queuing_status(eth_dev);
971 eth_dev_stop(struct rte_eth_dev *dev)
973 struct pmd_internal *internal = dev->data->dev_private;
975 rte_atomic32_set(&internal->started, 0);
976 update_queuing_status(dev);
980 eth_dev_close(struct rte_eth_dev *dev)
982 struct pmd_internal *internal;
983 struct internal_list *list;
986 internal = dev->data->dev_private;
992 rte_vhost_driver_unregister(internal->iface_name);
994 list = find_internal_resource(internal->iface_name);
998 pthread_mutex_lock(&internal_list_lock);
999 TAILQ_REMOVE(&internal_list, list, next);
1000 pthread_mutex_unlock(&internal_list_lock);
1003 if (dev->data->rx_queues)
1004 for (i = 0; i < dev->data->nb_rx_queues; i++)
1005 rte_free(dev->data->rx_queues[i]);
1007 if (dev->data->tx_queues)
1008 for (i = 0; i < dev->data->nb_tx_queues; i++)
1009 rte_free(dev->data->tx_queues[i]);
1011 free(internal->dev_name);
1012 free(internal->iface_name);
1015 dev->data->dev_private = NULL;
1017 rte_free(vring_states[dev->data->port_id]);
1018 vring_states[dev->data->port_id] = NULL;
1022 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1023 uint16_t nb_rx_desc __rte_unused,
1024 unsigned int socket_id,
1025 const struct rte_eth_rxconf *rx_conf __rte_unused,
1026 struct rte_mempool *mb_pool)
1028 struct vhost_queue *vq;
1030 vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
1031 RTE_CACHE_LINE_SIZE, socket_id);
1033 VHOST_LOG(ERR, "Failed to allocate memory for rx queue\n");
1037 vq->mb_pool = mb_pool;
1038 vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
1039 dev->data->rx_queues[rx_queue_id] = vq;
1045 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1046 uint16_t nb_tx_desc __rte_unused,
1047 unsigned int socket_id,
1048 const struct rte_eth_txconf *tx_conf __rte_unused)
1050 struct vhost_queue *vq;
1052 vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
1053 RTE_CACHE_LINE_SIZE, socket_id);
1055 VHOST_LOG(ERR, "Failed to allocate memory for tx queue\n");
1059 vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
1060 dev->data->tx_queues[tx_queue_id] = vq;
1066 eth_dev_info(struct rte_eth_dev *dev,
1067 struct rte_eth_dev_info *dev_info)
1069 struct pmd_internal *internal;
1071 internal = dev->data->dev_private;
1072 if (internal == NULL) {
1073 VHOST_LOG(ERR, "Invalid device specified\n");
1077 dev_info->max_mac_addrs = 1;
1078 dev_info->max_rx_pktlen = (uint32_t)-1;
1079 dev_info->max_rx_queues = internal->max_queues;
1080 dev_info->max_tx_queues = internal->max_queues;
1081 dev_info->min_rx_bufsize = 0;
1083 dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
1084 DEV_TX_OFFLOAD_VLAN_INSERT;
1085 dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
1091 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1094 unsigned long rx_total = 0, tx_total = 0;
1095 unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
1096 struct vhost_queue *vq;
1098 for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
1099 i < dev->data->nb_rx_queues; i++) {
1100 if (dev->data->rx_queues[i] == NULL)
1102 vq = dev->data->rx_queues[i];
1103 stats->q_ipackets[i] = vq->stats.pkts;
1104 rx_total += stats->q_ipackets[i];
1106 stats->q_ibytes[i] = vq->stats.bytes;
1107 rx_total_bytes += stats->q_ibytes[i];
1110 for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
1111 i < dev->data->nb_tx_queues; i++) {
1112 if (dev->data->tx_queues[i] == NULL)
1114 vq = dev->data->tx_queues[i];
1115 stats->q_opackets[i] = vq->stats.pkts;
1116 tx_total += stats->q_opackets[i];
1118 stats->q_obytes[i] = vq->stats.bytes;
1119 tx_total_bytes += stats->q_obytes[i];
1122 stats->ipackets = rx_total;
1123 stats->opackets = tx_total;
1124 stats->ibytes = rx_total_bytes;
1125 stats->obytes = tx_total_bytes;
1131 eth_stats_reset(struct rte_eth_dev *dev)
1133 struct vhost_queue *vq;
1136 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1137 if (dev->data->rx_queues[i] == NULL)
1139 vq = dev->data->rx_queues[i];
1141 vq->stats.bytes = 0;
1143 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1144 if (dev->data->tx_queues[i] == NULL)
1146 vq = dev->data->tx_queues[i];
1148 vq->stats.bytes = 0;
1149 vq->stats.missed_pkts = 0;
1156 eth_queue_release(void *q)
1162 eth_tx_done_cleanup(void *txq __rte_unused, uint32_t free_cnt __rte_unused)
1165 * vHost does not hang onto mbuf. eth_vhost_tx() copies packet data
1166 * and releases mbuf, so nothing to cleanup.
1172 eth_link_update(struct rte_eth_dev *dev __rte_unused,
1173 int wait_to_complete __rte_unused)
1179 eth_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1181 struct vhost_queue *vq;
1183 vq = dev->data->rx_queues[rx_queue_id];
1187 return rte_vhost_rx_queue_count(vq->vid, vq->virtqueue_id);
1190 static const struct eth_dev_ops ops = {
1191 .dev_start = eth_dev_start,
1192 .dev_stop = eth_dev_stop,
1193 .dev_close = eth_dev_close,
1194 .dev_configure = eth_dev_configure,
1195 .dev_infos_get = eth_dev_info,
1196 .rx_queue_setup = eth_rx_queue_setup,
1197 .tx_queue_setup = eth_tx_queue_setup,
1198 .rx_queue_release = eth_queue_release,
1199 .tx_queue_release = eth_queue_release,
1200 .tx_done_cleanup = eth_tx_done_cleanup,
1201 .rx_queue_count = eth_rx_queue_count,
1202 .link_update = eth_link_update,
1203 .stats_get = eth_stats_get,
1204 .stats_reset = eth_stats_reset,
1205 .xstats_reset = vhost_dev_xstats_reset,
1206 .xstats_get = vhost_dev_xstats_get,
1207 .xstats_get_names = vhost_dev_xstats_get_names,
1208 .rx_queue_intr_enable = eth_rxq_intr_enable,
1209 .rx_queue_intr_disable = eth_rxq_intr_disable,
1213 eth_dev_vhost_create(struct rte_vdev_device *dev, char *iface_name,
1214 int16_t queues, const unsigned int numa_node, uint64_t flags,
1215 uint64_t disable_flags)
1217 const char *name = rte_vdev_device_name(dev);
1218 struct rte_eth_dev_data *data;
1219 struct pmd_internal *internal = NULL;
1220 struct rte_eth_dev *eth_dev = NULL;
1221 struct rte_ether_addr *eth_addr = NULL;
1222 struct rte_vhost_vring_state *vring_state = NULL;
1223 struct internal_list *list = NULL;
1225 VHOST_LOG(INFO, "Creating VHOST-USER backend on numa socket %u\n",
1228 list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
1232 /* reserve an ethdev entry */
1233 eth_dev = rte_eth_vdev_allocate(dev, sizeof(*internal));
1234 if (eth_dev == NULL)
1236 data = eth_dev->data;
1238 eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
1239 if (eth_addr == NULL)
1241 data->mac_addrs = eth_addr;
1242 *eth_addr = base_eth_addr;
1243 eth_addr->addr_bytes[5] = eth_dev->data->port_id;
1245 vring_state = rte_zmalloc_socket(name,
1246 sizeof(*vring_state), 0, numa_node);
1247 if (vring_state == NULL)
1250 /* now put it all together
1251 * - store queue data in internal,
1252 * - point eth_dev_data to internals
1253 * - and point eth_dev structure to new eth_dev_data structure
1255 internal = eth_dev->data->dev_private;
1256 internal->dev_name = strdup(name);
1257 if (internal->dev_name == NULL)
1259 internal->iface_name = strdup(iface_name);
1260 if (internal->iface_name == NULL)
1263 list->eth_dev = eth_dev;
1264 pthread_mutex_lock(&internal_list_lock);
1265 TAILQ_INSERT_TAIL(&internal_list, list, next);
1266 pthread_mutex_unlock(&internal_list_lock);
1268 rte_spinlock_init(&vring_state->lock);
1269 vring_states[eth_dev->data->port_id] = vring_state;
1271 data->nb_rx_queues = queues;
1272 data->nb_tx_queues = queues;
1273 internal->max_queues = queues;
1275 data->dev_link = pmd_link;
1276 data->dev_flags = RTE_ETH_DEV_INTR_LSC | RTE_ETH_DEV_CLOSE_REMOVE;
1278 eth_dev->dev_ops = &ops;
1280 /* finally assign rx and tx ops */
1281 eth_dev->rx_pkt_burst = eth_vhost_rx;
1282 eth_dev->tx_pkt_burst = eth_vhost_tx;
1284 if (rte_vhost_driver_register(iface_name, flags))
1287 if (disable_flags) {
1288 if (rte_vhost_driver_disable_features(iface_name,
1293 if (rte_vhost_driver_callback_register(iface_name, &vhost_ops) < 0) {
1294 VHOST_LOG(ERR, "Can't register callbacks\n");
1298 if (rte_vhost_driver_start(iface_name) < 0) {
1299 VHOST_LOG(ERR, "Failed to start driver for %s\n",
1304 rte_eth_dev_probing_finish(eth_dev);
1309 free(internal->iface_name);
1310 free(internal->dev_name);
1312 rte_free(vring_state);
1313 rte_eth_dev_release_port(eth_dev);
1320 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
1322 const char **iface_name = extra_args;
1327 *iface_name = value;
1333 open_int(const char *key __rte_unused, const char *value, void *extra_args)
1335 uint16_t *n = extra_args;
1337 if (value == NULL || extra_args == NULL)
1340 *n = (uint16_t)strtoul(value, NULL, 0);
1341 if (*n == USHRT_MAX && errno == ERANGE)
1348 rte_pmd_vhost_probe(struct rte_vdev_device *dev)
1350 struct rte_kvargs *kvlist = NULL;
1355 uint64_t disable_flags = 0;
1356 int client_mode = 0;
1357 int dequeue_zero_copy = 0;
1358 int iommu_support = 0;
1359 int postcopy_support = 0;
1361 struct rte_eth_dev *eth_dev;
1362 const char *name = rte_vdev_device_name(dev);
1364 VHOST_LOG(INFO, "Initializing pmd_vhost for %s\n", name);
1366 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1367 eth_dev = rte_eth_dev_attach_secondary(name);
1369 VHOST_LOG(ERR, "Failed to probe %s\n", name);
1372 /* TODO: request info from primary to set up Rx and Tx */
1373 eth_dev->dev_ops = &ops;
1374 eth_dev->device = &dev->device;
1375 rte_eth_dev_probing_finish(eth_dev);
1379 kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
1383 if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
1384 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
1385 &open_iface, &iface_name);
1393 if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
1394 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
1395 &open_int, &queues);
1396 if (ret < 0 || queues > RTE_MAX_QUEUES_PER_PORT)
1402 if (rte_kvargs_count(kvlist, ETH_VHOST_CLIENT_ARG) == 1) {
1403 ret = rte_kvargs_process(kvlist, ETH_VHOST_CLIENT_ARG,
1404 &open_int, &client_mode);
1409 flags |= RTE_VHOST_USER_CLIENT;
1412 if (rte_kvargs_count(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY) == 1) {
1413 ret = rte_kvargs_process(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY,
1414 &open_int, &dequeue_zero_copy);
1418 if (dequeue_zero_copy)
1419 flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
1422 if (rte_kvargs_count(kvlist, ETH_VHOST_IOMMU_SUPPORT) == 1) {
1423 ret = rte_kvargs_process(kvlist, ETH_VHOST_IOMMU_SUPPORT,
1424 &open_int, &iommu_support);
1429 flags |= RTE_VHOST_USER_IOMMU_SUPPORT;
1432 if (rte_kvargs_count(kvlist, ETH_VHOST_POSTCOPY_SUPPORT) == 1) {
1433 ret = rte_kvargs_process(kvlist, ETH_VHOST_POSTCOPY_SUPPORT,
1434 &open_int, &postcopy_support);
1438 if (postcopy_support)
1439 flags |= RTE_VHOST_USER_POSTCOPY_SUPPORT;
1442 if (rte_kvargs_count(kvlist, ETH_VHOST_VIRTIO_NET_F_HOST_TSO) == 1) {
1443 ret = rte_kvargs_process(kvlist,
1444 ETH_VHOST_VIRTIO_NET_F_HOST_TSO,
1450 disable_flags |= (1ULL << VIRTIO_NET_F_HOST_TSO4);
1451 disable_flags |= (1ULL << VIRTIO_NET_F_HOST_TSO6);
1455 if (dev->device.numa_node == SOCKET_ID_ANY)
1456 dev->device.numa_node = rte_socket_id();
1458 ret = eth_dev_vhost_create(dev, iface_name, queues,
1459 dev->device.numa_node, flags, disable_flags);
1461 VHOST_LOG(ERR, "Failed to create %s\n", name);
1464 rte_kvargs_free(kvlist);
1469 rte_pmd_vhost_remove(struct rte_vdev_device *dev)
1472 struct rte_eth_dev *eth_dev = NULL;
1474 name = rte_vdev_device_name(dev);
1475 VHOST_LOG(INFO, "Un-Initializing pmd_vhost for %s\n", name);
1477 /* find an ethdev entry */
1478 eth_dev = rte_eth_dev_allocated(name);
1479 if (eth_dev == NULL)
1482 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1483 return rte_eth_dev_release_port(eth_dev);
1485 eth_dev_close(eth_dev);
1487 rte_eth_dev_release_port(eth_dev);
1492 static struct rte_vdev_driver pmd_vhost_drv = {
1493 .probe = rte_pmd_vhost_probe,
1494 .remove = rte_pmd_vhost_remove,
1497 RTE_PMD_REGISTER_VDEV(net_vhost, pmd_vhost_drv);
1498 RTE_PMD_REGISTER_ALIAS(net_vhost, eth_vhost);
1499 RTE_PMD_REGISTER_PARAM_STRING(net_vhost,
1503 "dequeue-zero-copy=<0|1> "
1504 "iommu-support=<0|1> "
1505 "postcopy-support=<0|1> "
1508 RTE_INIT(vhost_init_log)
1510 vhost_logtype = rte_log_register("pmd.net.vhost");
1511 if (vhost_logtype >= 0)
1512 rte_log_set_level(vhost_logtype, RTE_LOG_NOTICE);