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 VHOST_MAX_PKT_BURST 32
35 static const char *valid_arguments[] = {
39 ETH_VHOST_DEQUEUE_ZERO_COPY,
40 ETH_VHOST_IOMMU_SUPPORT,
44 static struct ether_addr base_eth_addr = {
55 enum vhost_xstats_pkts {
56 VHOST_UNDERSIZE_PKT = 0,
61 VHOST_512_TO_1023_PKT,
62 VHOST_1024_TO_1522_PKT,
63 VHOST_1523_TO_MAX_PKT,
68 VHOST_ERRORS_FRAGMENTED,
70 VHOST_UNKNOWN_PROTOCOL,
78 uint64_t xstats[VHOST_XSTATS_MAX];
83 rte_atomic32_t allow_queuing;
84 rte_atomic32_t while_queuing;
85 struct pmd_internal *internal;
86 struct rte_mempool *mb_pool;
88 uint16_t virtqueue_id;
89 struct vhost_stats stats;
93 rte_atomic32_t dev_attached;
98 rte_atomic32_t started;
102 struct internal_list {
103 TAILQ_ENTRY(internal_list) next;
104 struct rte_eth_dev *eth_dev;
107 TAILQ_HEAD(internal_list_head, internal_list);
108 static struct internal_list_head internal_list =
109 TAILQ_HEAD_INITIALIZER(internal_list);
111 static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
113 static struct rte_eth_link pmd_link = {
115 .link_duplex = ETH_LINK_FULL_DUPLEX,
116 .link_status = ETH_LINK_DOWN
119 struct rte_vhost_vring_state {
122 bool cur[RTE_MAX_QUEUES_PER_PORT * 2];
123 bool seen[RTE_MAX_QUEUES_PER_PORT * 2];
125 unsigned int max_vring;
128 static struct rte_vhost_vring_state *vring_states[RTE_MAX_ETHPORTS];
130 #define VHOST_XSTATS_NAME_SIZE 64
132 struct vhost_xstats_name_off {
133 char name[VHOST_XSTATS_NAME_SIZE];
137 /* [rx]_is prepended to the name string here */
138 static const struct vhost_xstats_name_off vhost_rxport_stat_strings[] = {
140 offsetof(struct vhost_queue, stats.pkts)},
142 offsetof(struct vhost_queue, stats.bytes)},
144 offsetof(struct vhost_queue, stats.missed_pkts)},
145 {"broadcast_packets",
146 offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
147 {"multicast_packets",
148 offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
150 offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
151 {"undersize_packets",
152 offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
154 offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
155 {"size_65_to_127_packets",
156 offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
157 {"size_128_to_255_packets",
158 offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
159 {"size_256_to_511_packets",
160 offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
161 {"size_512_to_1023_packets",
162 offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
163 {"size_1024_to_1522_packets",
164 offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
165 {"size_1523_to_max_packets",
166 offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
167 {"errors_with_bad_CRC",
168 offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
169 {"fragmented_errors",
170 offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_FRAGMENTED])},
172 offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_JABBER])},
173 {"unknown_protos_packets",
174 offsetof(struct vhost_queue, stats.xstats[VHOST_UNKNOWN_PROTOCOL])},
177 /* [tx]_ is prepended to the name string here */
178 static const struct vhost_xstats_name_off vhost_txport_stat_strings[] = {
180 offsetof(struct vhost_queue, stats.pkts)},
182 offsetof(struct vhost_queue, stats.bytes)},
184 offsetof(struct vhost_queue, stats.missed_pkts)},
185 {"broadcast_packets",
186 offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
187 {"multicast_packets",
188 offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
190 offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
191 {"undersize_packets",
192 offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
194 offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
195 {"size_65_to_127_packets",
196 offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
197 {"size_128_to_255_packets",
198 offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
199 {"size_256_to_511_packets",
200 offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
201 {"size_512_to_1023_packets",
202 offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
203 {"size_1024_to_1522_packets",
204 offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
205 {"size_1523_to_max_packets",
206 offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
207 {"errors_with_bad_CRC",
208 offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
211 #define VHOST_NB_XSTATS_RXPORT (sizeof(vhost_rxport_stat_strings) / \
212 sizeof(vhost_rxport_stat_strings[0]))
214 #define VHOST_NB_XSTATS_TXPORT (sizeof(vhost_txport_stat_strings) / \
215 sizeof(vhost_txport_stat_strings[0]))
218 vhost_dev_xstats_reset(struct rte_eth_dev *dev)
220 struct vhost_queue *vq = NULL;
223 for (i = 0; i < dev->data->nb_rx_queues; i++) {
224 vq = dev->data->rx_queues[i];
227 memset(&vq->stats, 0, sizeof(vq->stats));
229 for (i = 0; i < dev->data->nb_tx_queues; i++) {
230 vq = dev->data->tx_queues[i];
233 memset(&vq->stats, 0, sizeof(vq->stats));
238 vhost_dev_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
239 struct rte_eth_xstat_name *xstats_names,
240 unsigned int limit __rte_unused)
244 int nstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
248 for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
249 snprintf(xstats_names[count].name,
250 sizeof(xstats_names[count].name),
251 "rx_%s", vhost_rxport_stat_strings[t].name);
254 for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
255 snprintf(xstats_names[count].name,
256 sizeof(xstats_names[count].name),
257 "tx_%s", vhost_txport_stat_strings[t].name);
264 vhost_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
269 unsigned int count = 0;
270 struct vhost_queue *vq = NULL;
271 unsigned int nxstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
276 for (i = 0; i < dev->data->nb_rx_queues; i++) {
277 vq = dev->data->rx_queues[i];
280 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
281 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
282 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
284 for (i = 0; i < dev->data->nb_tx_queues; i++) {
285 vq = dev->data->tx_queues[i];
288 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
289 + vq->stats.missed_pkts
290 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
291 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
293 for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
294 xstats[count].value = 0;
295 for (i = 0; i < dev->data->nb_rx_queues; i++) {
296 vq = dev->data->rx_queues[i];
299 xstats[count].value +=
300 *(uint64_t *)(((char *)vq)
301 + vhost_rxport_stat_strings[t].offset);
303 xstats[count].id = count;
306 for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
307 xstats[count].value = 0;
308 for (i = 0; i < dev->data->nb_tx_queues; i++) {
309 vq = dev->data->tx_queues[i];
312 xstats[count].value +=
313 *(uint64_t *)(((char *)vq)
314 + vhost_txport_stat_strings[t].offset);
316 xstats[count].id = count;
323 vhost_count_multicast_broadcast(struct vhost_queue *vq,
324 struct rte_mbuf *mbuf)
326 struct ether_addr *ea = NULL;
327 struct vhost_stats *pstats = &vq->stats;
329 ea = rte_pktmbuf_mtod(mbuf, struct ether_addr *);
330 if (is_multicast_ether_addr(ea)) {
331 if (is_broadcast_ether_addr(ea))
332 pstats->xstats[VHOST_BROADCAST_PKT]++;
334 pstats->xstats[VHOST_MULTICAST_PKT]++;
339 vhost_update_packet_xstats(struct vhost_queue *vq,
340 struct rte_mbuf **bufs,
343 uint32_t pkt_len = 0;
346 struct vhost_stats *pstats = &vq->stats;
348 for (i = 0; i < count ; i++) {
349 pkt_len = bufs[i]->pkt_len;
351 pstats->xstats[VHOST_64_PKT]++;
352 } else if (pkt_len > 64 && pkt_len < 1024) {
353 index = (sizeof(pkt_len) * 8)
354 - __builtin_clz(pkt_len) - 5;
355 pstats->xstats[index]++;
358 pstats->xstats[VHOST_UNDERSIZE_PKT]++;
359 else if (pkt_len <= 1522)
360 pstats->xstats[VHOST_1024_TO_1522_PKT]++;
361 else if (pkt_len > 1522)
362 pstats->xstats[VHOST_1523_TO_MAX_PKT]++;
364 vhost_count_multicast_broadcast(vq, bufs[i]);
369 eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
371 struct vhost_queue *r = q;
372 uint16_t i, nb_rx = 0;
373 uint16_t nb_receive = nb_bufs;
375 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
378 rte_atomic32_set(&r->while_queuing, 1);
380 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
383 /* Dequeue packets from guest TX queue */
386 uint16_t num = (uint16_t)RTE_MIN(nb_receive,
387 VHOST_MAX_PKT_BURST);
389 nb_pkts = rte_vhost_dequeue_burst(r->vid, r->virtqueue_id,
390 r->mb_pool, &bufs[nb_rx],
394 nb_receive -= nb_pkts;
399 r->stats.pkts += nb_rx;
401 for (i = 0; likely(i < nb_rx); i++) {
402 bufs[i]->port = r->port;
403 bufs[i]->ol_flags = 0;
404 bufs[i]->vlan_tci = 0;
406 if (r->internal->vlan_strip)
407 rte_vlan_strip(bufs[i]);
409 r->stats.bytes += bufs[i]->pkt_len;
412 vhost_update_packet_xstats(r, bufs, nb_rx);
415 rte_atomic32_set(&r->while_queuing, 0);
421 eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
423 struct vhost_queue *r = q;
424 uint16_t i, nb_tx = 0;
425 uint16_t nb_send = 0;
427 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
430 rte_atomic32_set(&r->while_queuing, 1);
432 if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
435 for (i = 0; i < nb_bufs; i++) {
436 struct rte_mbuf *m = bufs[i];
438 /* Do VLAN tag insertion */
439 if (m->ol_flags & PKT_TX_VLAN_PKT) {
440 int error = rte_vlan_insert(&m);
441 if (unlikely(error)) {
451 /* Enqueue packets to guest RX queue */
454 uint16_t num = (uint16_t)RTE_MIN(nb_send,
455 VHOST_MAX_PKT_BURST);
457 nb_pkts = rte_vhost_enqueue_burst(r->vid, r->virtqueue_id,
466 r->stats.pkts += nb_tx;
467 r->stats.missed_pkts += nb_bufs - nb_tx;
469 for (i = 0; likely(i < nb_tx); i++)
470 r->stats.bytes += bufs[i]->pkt_len;
472 vhost_update_packet_xstats(r, bufs, nb_tx);
474 /* According to RFC2863 page42 section ifHCOutMulticastPkts and
475 * ifHCOutBroadcastPkts, the counters "multicast" and "broadcast"
476 * are increased when packets are not transmitted successfully.
478 for (i = nb_tx; i < nb_bufs; i++)
479 vhost_count_multicast_broadcast(r, bufs[i]);
481 for (i = 0; likely(i < nb_tx); i++)
482 rte_pktmbuf_free(bufs[i]);
484 rte_atomic32_set(&r->while_queuing, 0);
490 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
492 struct pmd_internal *internal = dev->data->dev_private;
493 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
495 internal->vlan_strip = rxmode->hw_vlan_strip;
497 if (rxmode->hw_vlan_filter)
499 "vhost(%s): vlan filtering not available\n",
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 state->cur[vring] = enable;
859 state->max_vring = RTE_MAX(vring, state->max_vring);
860 rte_spinlock_unlock(&state->lock);
862 VHOST_LOG(INFO, "vring%u is %s\n",
863 vring, enable ? "enabled" : "disabled");
865 _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE, NULL);
870 static struct vhost_device_ops vhost_ops = {
871 .new_device = new_device,
872 .destroy_device = destroy_device,
873 .vring_state_changed = vring_state_changed,
877 rte_eth_vhost_get_queue_event(uint16_t port_id,
878 struct rte_eth_vhost_queue_event *event)
880 struct rte_vhost_vring_state *state;
884 if (port_id >= RTE_MAX_ETHPORTS) {
885 VHOST_LOG(ERR, "Invalid port id\n");
889 state = vring_states[port_id];
891 VHOST_LOG(ERR, "Unused port\n");
895 rte_spinlock_lock(&state->lock);
896 for (i = 0; i <= state->max_vring; i++) {
897 idx = state->index++ % (state->max_vring + 1);
899 if (state->cur[idx] != state->seen[idx]) {
900 state->seen[idx] = state->cur[idx];
901 event->queue_id = idx / 2;
903 event->enable = state->cur[idx];
904 rte_spinlock_unlock(&state->lock);
908 rte_spinlock_unlock(&state->lock);
914 rte_eth_vhost_get_vid_from_port_id(uint16_t port_id)
916 struct internal_list *list;
917 struct rte_eth_dev *eth_dev;
918 struct vhost_queue *vq;
921 if (!rte_eth_dev_is_valid_port(port_id))
924 pthread_mutex_lock(&internal_list_lock);
926 TAILQ_FOREACH(list, &internal_list, next) {
927 eth_dev = list->eth_dev;
928 if (eth_dev->data->port_id == port_id) {
929 vq = eth_dev->data->rx_queues[0];
937 pthread_mutex_unlock(&internal_list_lock);
943 eth_dev_start(struct rte_eth_dev *eth_dev)
945 struct pmd_internal *internal = eth_dev->data->dev_private;
946 struct rte_eth_conf *dev_conf = ð_dev->data->dev_conf;
948 queue_setup(eth_dev, internal);
950 if (rte_atomic32_read(&internal->dev_attached) == 1) {
951 if (dev_conf->intr_conf.rxq) {
952 if (eth_vhost_install_intr(eth_dev) < 0) {
954 "Failed to install interrupt handler.");
960 rte_atomic32_set(&internal->started, 1);
961 update_queuing_status(eth_dev);
967 eth_dev_stop(struct rte_eth_dev *dev)
969 struct pmd_internal *internal = dev->data->dev_private;
971 rte_atomic32_set(&internal->started, 0);
972 update_queuing_status(dev);
976 eth_dev_close(struct rte_eth_dev *dev)
978 struct pmd_internal *internal;
979 struct internal_list *list;
982 internal = dev->data->dev_private;
988 rte_vhost_driver_unregister(internal->iface_name);
990 list = find_internal_resource(internal->iface_name);
994 pthread_mutex_lock(&internal_list_lock);
995 TAILQ_REMOVE(&internal_list, list, next);
996 pthread_mutex_unlock(&internal_list_lock);
999 if (dev->data->rx_queues)
1000 for (i = 0; i < dev->data->nb_rx_queues; i++)
1001 rte_free(dev->data->rx_queues[i]);
1003 if (dev->data->tx_queues)
1004 for (i = 0; i < dev->data->nb_tx_queues; i++)
1005 rte_free(dev->data->tx_queues[i]);
1007 rte_free(dev->data->mac_addrs);
1008 free(internal->dev_name);
1009 free(internal->iface_name);
1012 dev->data->dev_private = NULL;
1016 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1017 uint16_t nb_rx_desc __rte_unused,
1018 unsigned int socket_id,
1019 const struct rte_eth_rxconf *rx_conf __rte_unused,
1020 struct rte_mempool *mb_pool)
1022 struct vhost_queue *vq;
1024 vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
1025 RTE_CACHE_LINE_SIZE, socket_id);
1027 VHOST_LOG(ERR, "Failed to allocate memory for rx queue\n");
1031 vq->mb_pool = mb_pool;
1032 vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
1033 dev->data->rx_queues[rx_queue_id] = vq;
1039 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1040 uint16_t nb_tx_desc __rte_unused,
1041 unsigned int socket_id,
1042 const struct rte_eth_txconf *tx_conf __rte_unused)
1044 struct vhost_queue *vq;
1046 vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
1047 RTE_CACHE_LINE_SIZE, socket_id);
1049 VHOST_LOG(ERR, "Failed to allocate memory for tx queue\n");
1053 vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
1054 dev->data->tx_queues[tx_queue_id] = vq;
1060 eth_dev_info(struct rte_eth_dev *dev,
1061 struct rte_eth_dev_info *dev_info)
1063 struct pmd_internal *internal;
1065 internal = dev->data->dev_private;
1066 if (internal == NULL) {
1067 VHOST_LOG(ERR, "Invalid device specified\n");
1071 dev_info->max_mac_addrs = 1;
1072 dev_info->max_rx_pktlen = (uint32_t)-1;
1073 dev_info->max_rx_queues = internal->max_queues;
1074 dev_info->max_tx_queues = internal->max_queues;
1075 dev_info->min_rx_bufsize = 0;
1079 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1082 unsigned long rx_total = 0, tx_total = 0, tx_missed_total = 0;
1083 unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
1084 struct vhost_queue *vq;
1086 for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
1087 i < dev->data->nb_rx_queues; i++) {
1088 if (dev->data->rx_queues[i] == NULL)
1090 vq = dev->data->rx_queues[i];
1091 stats->q_ipackets[i] = vq->stats.pkts;
1092 rx_total += stats->q_ipackets[i];
1094 stats->q_ibytes[i] = vq->stats.bytes;
1095 rx_total_bytes += stats->q_ibytes[i];
1098 for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
1099 i < dev->data->nb_tx_queues; i++) {
1100 if (dev->data->tx_queues[i] == NULL)
1102 vq = dev->data->tx_queues[i];
1103 stats->q_opackets[i] = vq->stats.pkts;
1104 tx_missed_total += vq->stats.missed_pkts;
1105 tx_total += stats->q_opackets[i];
1107 stats->q_obytes[i] = vq->stats.bytes;
1108 tx_total_bytes += stats->q_obytes[i];
1111 stats->ipackets = rx_total;
1112 stats->opackets = tx_total;
1113 stats->oerrors = tx_missed_total;
1114 stats->ibytes = rx_total_bytes;
1115 stats->obytes = tx_total_bytes;
1121 eth_stats_reset(struct rte_eth_dev *dev)
1123 struct vhost_queue *vq;
1126 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1127 if (dev->data->rx_queues[i] == NULL)
1129 vq = dev->data->rx_queues[i];
1131 vq->stats.bytes = 0;
1133 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1134 if (dev->data->tx_queues[i] == NULL)
1136 vq = dev->data->tx_queues[i];
1138 vq->stats.bytes = 0;
1139 vq->stats.missed_pkts = 0;
1144 eth_queue_release(void *q)
1150 eth_tx_done_cleanup(void *txq __rte_unused, uint32_t free_cnt __rte_unused)
1153 * vHost does not hang onto mbuf. eth_vhost_tx() copies packet data
1154 * and releases mbuf, so nothing to cleanup.
1160 eth_link_update(struct rte_eth_dev *dev __rte_unused,
1161 int wait_to_complete __rte_unused)
1167 eth_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1169 struct vhost_queue *vq;
1171 vq = dev->data->rx_queues[rx_queue_id];
1175 return rte_vhost_rx_queue_count(vq->vid, vq->virtqueue_id);
1178 static const struct eth_dev_ops ops = {
1179 .dev_start = eth_dev_start,
1180 .dev_stop = eth_dev_stop,
1181 .dev_close = eth_dev_close,
1182 .dev_configure = eth_dev_configure,
1183 .dev_infos_get = eth_dev_info,
1184 .rx_queue_setup = eth_rx_queue_setup,
1185 .tx_queue_setup = eth_tx_queue_setup,
1186 .rx_queue_release = eth_queue_release,
1187 .tx_queue_release = eth_queue_release,
1188 .tx_done_cleanup = eth_tx_done_cleanup,
1189 .rx_queue_count = eth_rx_queue_count,
1190 .link_update = eth_link_update,
1191 .stats_get = eth_stats_get,
1192 .stats_reset = eth_stats_reset,
1193 .xstats_reset = vhost_dev_xstats_reset,
1194 .xstats_get = vhost_dev_xstats_get,
1195 .xstats_get_names = vhost_dev_xstats_get_names,
1196 .rx_queue_intr_enable = eth_rxq_intr_enable,
1197 .rx_queue_intr_disable = eth_rxq_intr_disable,
1200 static struct rte_vdev_driver pmd_vhost_drv;
1203 eth_dev_vhost_create(struct rte_vdev_device *dev, char *iface_name,
1204 int16_t queues, const unsigned int numa_node, uint64_t flags)
1206 const char *name = rte_vdev_device_name(dev);
1207 struct rte_eth_dev_data *data;
1208 struct pmd_internal *internal = NULL;
1209 struct rte_eth_dev *eth_dev = NULL;
1210 struct ether_addr *eth_addr = NULL;
1211 struct rte_vhost_vring_state *vring_state = NULL;
1212 struct internal_list *list = NULL;
1214 VHOST_LOG(INFO, "Creating VHOST-USER backend on numa socket %u\n",
1217 list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
1221 /* reserve an ethdev entry */
1222 eth_dev = rte_eth_vdev_allocate(dev, sizeof(*internal));
1223 if (eth_dev == NULL)
1226 eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
1227 if (eth_addr == NULL)
1229 *eth_addr = base_eth_addr;
1230 eth_addr->addr_bytes[5] = eth_dev->data->port_id;
1232 vring_state = rte_zmalloc_socket(name,
1233 sizeof(*vring_state), 0, numa_node);
1234 if (vring_state == NULL)
1237 /* now put it all together
1238 * - store queue data in internal,
1239 * - point eth_dev_data to internals
1240 * - and point eth_dev structure to new eth_dev_data structure
1242 internal = eth_dev->data->dev_private;
1243 internal->dev_name = strdup(name);
1244 if (internal->dev_name == NULL)
1246 internal->iface_name = strdup(iface_name);
1247 if (internal->iface_name == NULL)
1250 list->eth_dev = eth_dev;
1251 pthread_mutex_lock(&internal_list_lock);
1252 TAILQ_INSERT_TAIL(&internal_list, list, next);
1253 pthread_mutex_unlock(&internal_list_lock);
1255 rte_spinlock_init(&vring_state->lock);
1256 vring_states[eth_dev->data->port_id] = vring_state;
1258 data = eth_dev->data;
1259 data->nb_rx_queues = queues;
1260 data->nb_tx_queues = queues;
1261 internal->max_queues = queues;
1263 data->dev_link = pmd_link;
1264 data->mac_addrs = eth_addr;
1265 data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1267 eth_dev->dev_ops = &ops;
1269 /* finally assign rx and tx ops */
1270 eth_dev->rx_pkt_burst = eth_vhost_rx;
1271 eth_dev->tx_pkt_burst = eth_vhost_tx;
1273 if (rte_vhost_driver_register(iface_name, flags))
1276 if (rte_vhost_driver_callback_register(iface_name, &vhost_ops) < 0) {
1277 VHOST_LOG(ERR, "Can't register callbacks\n");
1281 if (rte_vhost_driver_start(iface_name) < 0) {
1282 VHOST_LOG(ERR, "Failed to start driver for %s\n",
1287 rte_eth_dev_probing_finish(eth_dev);
1288 return data->port_id;
1292 free(internal->iface_name);
1293 free(internal->dev_name);
1295 rte_free(vring_state);
1298 rte_eth_dev_release_port(eth_dev);
1306 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
1308 const char **iface_name = extra_args;
1313 *iface_name = value;
1319 open_int(const char *key __rte_unused, const char *value, void *extra_args)
1321 uint16_t *n = extra_args;
1323 if (value == NULL || extra_args == NULL)
1326 *n = (uint16_t)strtoul(value, NULL, 0);
1327 if (*n == USHRT_MAX && errno == ERANGE)
1334 rte_pmd_vhost_probe(struct rte_vdev_device *dev)
1336 struct rte_kvargs *kvlist = NULL;
1341 int client_mode = 0;
1342 int dequeue_zero_copy = 0;
1343 int iommu_support = 0;
1344 struct rte_eth_dev *eth_dev;
1345 const char *name = rte_vdev_device_name(dev);
1347 VHOST_LOG(INFO, "Initializing pmd_vhost for %s\n", name);
1349 if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1350 strlen(rte_vdev_device_args(dev)) == 0) {
1351 eth_dev = rte_eth_dev_attach_secondary(name);
1353 VHOST_LOG(ERR, "Failed to probe %s\n", name);
1356 /* TODO: request info from primary to set up Rx and Tx */
1357 eth_dev->dev_ops = &ops;
1358 rte_eth_dev_probing_finish(eth_dev);
1362 kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
1366 if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
1367 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
1368 &open_iface, &iface_name);
1376 if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
1377 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
1378 &open_int, &queues);
1379 if (ret < 0 || queues > RTE_MAX_QUEUES_PER_PORT)
1385 if (rte_kvargs_count(kvlist, ETH_VHOST_CLIENT_ARG) == 1) {
1386 ret = rte_kvargs_process(kvlist, ETH_VHOST_CLIENT_ARG,
1387 &open_int, &client_mode);
1392 flags |= RTE_VHOST_USER_CLIENT;
1395 if (rte_kvargs_count(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY) == 1) {
1396 ret = rte_kvargs_process(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY,
1397 &open_int, &dequeue_zero_copy);
1401 if (dequeue_zero_copy)
1402 flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
1405 if (rte_kvargs_count(kvlist, ETH_VHOST_IOMMU_SUPPORT) == 1) {
1406 ret = rte_kvargs_process(kvlist, ETH_VHOST_IOMMU_SUPPORT,
1407 &open_int, &iommu_support);
1412 flags |= RTE_VHOST_USER_IOMMU_SUPPORT;
1415 if (dev->device.numa_node == SOCKET_ID_ANY)
1416 dev->device.numa_node = rte_socket_id();
1418 eth_dev_vhost_create(dev, iface_name, queues, dev->device.numa_node,
1422 rte_kvargs_free(kvlist);
1427 rte_pmd_vhost_remove(struct rte_vdev_device *dev)
1430 struct rte_eth_dev *eth_dev = NULL;
1432 name = rte_vdev_device_name(dev);
1433 VHOST_LOG(INFO, "Un-Initializing pmd_vhost for %s\n", name);
1435 /* find an ethdev entry */
1436 eth_dev = rte_eth_dev_allocated(name);
1437 if (eth_dev == NULL)
1440 eth_dev_close(eth_dev);
1442 rte_free(vring_states[eth_dev->data->port_id]);
1443 vring_states[eth_dev->data->port_id] = NULL;
1445 rte_eth_dev_release_port(eth_dev);
1450 static struct rte_vdev_driver pmd_vhost_drv = {
1451 .probe = rte_pmd_vhost_probe,
1452 .remove = rte_pmd_vhost_remove,
1455 RTE_PMD_REGISTER_VDEV(net_vhost, pmd_vhost_drv);
1456 RTE_PMD_REGISTER_ALIAS(net_vhost, eth_vhost);
1457 RTE_PMD_REGISTER_PARAM_STRING(net_vhost,
1461 RTE_INIT(vhost_init_log);
1463 vhost_init_log(void)
1465 vhost_logtype = rte_log_register("pmd.net.vhost");
1466 if (vhost_logtype >= 0)
1467 rte_log_set_level(vhost_logtype, RTE_LOG_NOTICE);