1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
12 #include <sys/eventfd.h>
13 #include <sys/types.h>
16 #include <rte_eal_memconfig.h>
19 #include "virtio_user_dev.h"
20 #include "../virtio_ethdev.h"
22 #define VIRTIO_USER_MEM_EVENT_CLB_NAME "virtio_user_mem_event_clb"
25 virtio_user_create_queue(struct virtio_user_dev *dev, uint32_t queue_sel)
27 /* Of all per virtqueue MSGs, make sure VHOST_SET_VRING_CALL come
28 * firstly because vhost depends on this msg to allocate virtqueue
31 struct vhost_vring_file file;
33 file.index = queue_sel;
34 file.fd = dev->callfds[queue_sel];
35 dev->ops->send_request(dev, VHOST_USER_SET_VRING_CALL, &file);
41 virtio_user_kick_queue(struct virtio_user_dev *dev, uint32_t queue_sel)
43 struct vhost_vring_file file;
44 struct vhost_vring_state state;
45 struct vring *vring = &dev->vrings[queue_sel];
46 struct vhost_vring_addr addr = {
48 .desc_user_addr = (uint64_t)(uintptr_t)vring->desc,
49 .avail_user_addr = (uint64_t)(uintptr_t)vring->avail,
50 .used_user_addr = (uint64_t)(uintptr_t)vring->used,
52 .flags = 0, /* disable log */
55 state.index = queue_sel;
56 state.num = vring->num;
57 dev->ops->send_request(dev, VHOST_USER_SET_VRING_NUM, &state);
59 state.index = queue_sel;
60 state.num = 0; /* no reservation */
61 dev->ops->send_request(dev, VHOST_USER_SET_VRING_BASE, &state);
63 dev->ops->send_request(dev, VHOST_USER_SET_VRING_ADDR, &addr);
65 /* Of all per virtqueue MSGs, make sure VHOST_USER_SET_VRING_KICK comes
66 * lastly because vhost depends on this msg to judge if
69 file.index = queue_sel;
70 file.fd = dev->kickfds[queue_sel];
71 dev->ops->send_request(dev, VHOST_USER_SET_VRING_KICK, &file);
77 virtio_user_queue_setup(struct virtio_user_dev *dev,
78 int (*fn)(struct virtio_user_dev *, uint32_t))
80 uint32_t i, queue_sel;
82 for (i = 0; i < dev->max_queue_pairs; ++i) {
83 queue_sel = 2 * i + VTNET_SQ_RQ_QUEUE_IDX;
84 if (fn(dev, queue_sel) < 0) {
85 PMD_DRV_LOG(INFO, "setup rx vq fails: %u", i);
89 for (i = 0; i < dev->max_queue_pairs; ++i) {
90 queue_sel = 2 * i + VTNET_SQ_TQ_QUEUE_IDX;
91 if (fn(dev, queue_sel) < 0) {
92 PMD_DRV_LOG(INFO, "setup tx vq fails: %u", i);
101 is_vhost_user_by_type(const char *path)
105 if (stat(path, &sb) == -1)
108 return S_ISSOCK(sb.st_mode);
112 virtio_user_start_device(struct virtio_user_dev *dev)
114 struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
121 * We need to make sure that the locks will be
122 * taken in the correct order to avoid deadlocks.
124 * Before releasing this lock, this thread should
125 * not trigger any memory hotplug events.
127 * This is a temporary workaround, and should be
128 * replaced when we get proper supports from the
129 * memory subsystem in the future.
131 rte_rwlock_read_lock(&mcfg->memory_hotplug_lock);
132 pthread_mutex_lock(&dev->mutex);
134 if (is_vhost_user_by_type(dev->path) && dev->vhostfd < 0)
137 /* Do not check return as already done in init, or reset in stop */
138 dev->ops->send_request(dev, VHOST_USER_SET_OWNER, NULL);
140 /* Step 0: tell vhost to create queues */
141 if (virtio_user_queue_setup(dev, virtio_user_create_queue) < 0)
144 /* Step 1: set features */
145 features = dev->features;
146 /* Strip VIRTIO_NET_F_MAC, as MAC address is handled in vdev init */
147 features &= ~(1ull << VIRTIO_NET_F_MAC);
148 /* Strip VIRTIO_NET_F_CTRL_VQ, as devices do not really need to know */
149 features &= ~(1ull << VIRTIO_NET_F_CTRL_VQ);
150 features &= ~(1ull << VIRTIO_NET_F_STATUS);
151 ret = dev->ops->send_request(dev, VHOST_USER_SET_FEATURES, &features);
154 PMD_DRV_LOG(INFO, "set features: %" PRIx64, features);
156 /* Step 2: share memory regions */
157 ret = dev->ops->send_request(dev, VHOST_USER_SET_MEM_TABLE, NULL);
161 /* Step 3: kick queues */
162 if (virtio_user_queue_setup(dev, virtio_user_kick_queue) < 0)
165 /* Step 4: enable queues
166 * we enable the 1st queue pair by default.
168 dev->ops->enable_qp(dev, 0, 1);
171 pthread_mutex_unlock(&dev->mutex);
172 rte_rwlock_read_unlock(&mcfg->memory_hotplug_lock);
176 pthread_mutex_unlock(&dev->mutex);
177 rte_rwlock_read_unlock(&mcfg->memory_hotplug_lock);
178 /* TODO: free resource here or caller to check */
182 int virtio_user_stop_device(struct virtio_user_dev *dev)
186 pthread_mutex_lock(&dev->mutex);
187 for (i = 0; i < dev->max_queue_pairs; ++i)
188 dev->ops->enable_qp(dev, i, 0);
190 if (dev->ops->send_request(dev, VHOST_USER_RESET_OWNER, NULL) < 0) {
191 PMD_DRV_LOG(INFO, "Failed to reset the device\n");
192 pthread_mutex_unlock(&dev->mutex);
195 dev->started = false;
196 pthread_mutex_unlock(&dev->mutex);
202 parse_mac(struct virtio_user_dev *dev, const char *mac)
205 uint32_t tmp[ETHER_ADDR_LEN];
210 r = sscanf(mac, "%x:%x:%x:%x:%x:%x", &tmp[0],
211 &tmp[1], &tmp[2], &tmp[3], &tmp[4], &tmp[5]);
212 if (r == ETHER_ADDR_LEN) {
213 for (i = 0; i < ETHER_ADDR_LEN; ++i)
214 dev->mac_addr[i] = (uint8_t)tmp[i];
215 dev->mac_specified = 1;
217 /* ignore the wrong mac, use random mac */
218 PMD_DRV_LOG(ERR, "wrong format of mac: %s", mac);
223 virtio_user_dev_init_notify(struct virtio_user_dev *dev)
229 for (i = 0; i < VIRTIO_MAX_VIRTQUEUES; ++i) {
230 if (i >= dev->max_queue_pairs * 2) {
231 dev->kickfds[i] = -1;
232 dev->callfds[i] = -1;
236 /* May use invalid flag, but some backend uses kickfd and
237 * callfd as criteria to judge if dev is alive. so finally we
240 callfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK);
242 PMD_DRV_LOG(ERR, "callfd error, %s", strerror(errno));
245 kickfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK);
247 PMD_DRV_LOG(ERR, "kickfd error, %s", strerror(errno));
250 dev->callfds[i] = callfd;
251 dev->kickfds[i] = kickfd;
254 if (i < VIRTIO_MAX_VIRTQUEUES) {
255 for (j = 0; j <= i; ++j) {
256 close(dev->callfds[j]);
257 close(dev->kickfds[j]);
267 virtio_user_fill_intr_handle(struct virtio_user_dev *dev)
270 struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->port_id];
272 if (!eth_dev->intr_handle) {
273 eth_dev->intr_handle = malloc(sizeof(*eth_dev->intr_handle));
274 if (!eth_dev->intr_handle) {
275 PMD_DRV_LOG(ERR, "fail to allocate intr_handle");
278 memset(eth_dev->intr_handle, 0, sizeof(*eth_dev->intr_handle));
281 for (i = 0; i < dev->max_queue_pairs; ++i)
282 eth_dev->intr_handle->efds[i] = dev->callfds[i];
283 eth_dev->intr_handle->nb_efd = dev->max_queue_pairs;
284 eth_dev->intr_handle->max_intr = dev->max_queue_pairs + 1;
285 eth_dev->intr_handle->type = RTE_INTR_HANDLE_VDEV;
286 /* For virtio vdev, no need to read counter for clean */
287 eth_dev->intr_handle->efd_counter_size = 0;
288 eth_dev->intr_handle->fd = -1;
289 if (dev->vhostfd >= 0)
290 eth_dev->intr_handle->fd = dev->vhostfd;
291 else if (dev->is_server)
292 eth_dev->intr_handle->fd = dev->listenfd;
298 virtio_user_mem_event_cb(enum rte_mem_event type __rte_unused,
299 const void *addr __rte_unused,
300 size_t len __rte_unused,
303 struct virtio_user_dev *dev = arg;
304 struct rte_memseg_list *msl;
307 /* ignore externally allocated memory */
308 msl = rte_mem_virt2memseg_list(addr);
312 pthread_mutex_lock(&dev->mutex);
314 if (dev->started == false)
317 /* Step 1: pause the active queues */
318 for (i = 0; i < dev->queue_pairs; i++)
319 dev->ops->enable_qp(dev, i, 0);
321 /* Step 2: update memory regions */
322 dev->ops->send_request(dev, VHOST_USER_SET_MEM_TABLE, NULL);
324 /* Step 3: resume the active queues */
325 for (i = 0; i < dev->queue_pairs; i++)
326 dev->ops->enable_qp(dev, i, 1);
329 pthread_mutex_unlock(&dev->mutex);
333 virtio_user_dev_setup(struct virtio_user_dev *dev)
338 dev->vhostfds = NULL;
341 if (dev->is_server) {
342 if (access(dev->path, F_OK) == 0 &&
343 !is_vhost_user_by_type(dev->path)) {
344 PMD_DRV_LOG(ERR, "Server mode doesn't support vhost-kernel!");
347 dev->ops = &virtio_ops_user;
349 if (is_vhost_user_by_type(dev->path)) {
350 dev->ops = &virtio_ops_user;
352 dev->ops = &virtio_ops_kernel;
354 dev->vhostfds = malloc(dev->max_queue_pairs *
356 dev->tapfds = malloc(dev->max_queue_pairs *
358 if (!dev->vhostfds || !dev->tapfds) {
359 PMD_INIT_LOG(ERR, "Failed to malloc");
363 for (q = 0; q < dev->max_queue_pairs; ++q) {
364 dev->vhostfds[q] = -1;
370 if (dev->ops->setup(dev) < 0)
373 if (virtio_user_dev_init_notify(dev) < 0)
376 if (virtio_user_fill_intr_handle(dev) < 0)
382 /* Use below macro to filter features from vhost backend */
383 #define VIRTIO_USER_SUPPORTED_FEATURES \
384 (1ULL << VIRTIO_NET_F_MAC | \
385 1ULL << VIRTIO_NET_F_STATUS | \
386 1ULL << VIRTIO_NET_F_MQ | \
387 1ULL << VIRTIO_NET_F_CTRL_MAC_ADDR | \
388 1ULL << VIRTIO_NET_F_CTRL_VQ | \
389 1ULL << VIRTIO_NET_F_CTRL_RX | \
390 1ULL << VIRTIO_NET_F_CTRL_VLAN | \
391 1ULL << VIRTIO_NET_F_CSUM | \
392 1ULL << VIRTIO_NET_F_HOST_TSO4 | \
393 1ULL << VIRTIO_NET_F_HOST_TSO6 | \
394 1ULL << VIRTIO_NET_F_MRG_RXBUF | \
395 1ULL << VIRTIO_RING_F_INDIRECT_DESC | \
396 1ULL << VIRTIO_NET_F_GUEST_CSUM | \
397 1ULL << VIRTIO_NET_F_GUEST_TSO4 | \
398 1ULL << VIRTIO_NET_F_GUEST_TSO6 | \
399 1ULL << VIRTIO_F_IN_ORDER | \
400 1ULL << VIRTIO_F_VERSION_1)
403 virtio_user_dev_init(struct virtio_user_dev *dev, char *path, int queues,
404 int cq, int queue_size, const char *mac, char **ifname,
405 int mrg_rxbuf, int in_order)
407 pthread_mutex_init(&dev->mutex, NULL);
408 snprintf(dev->path, PATH_MAX, "%s", path);
410 dev->max_queue_pairs = queues;
411 dev->queue_pairs = 1; /* mq disabled by default */
412 dev->queue_size = queue_size;
413 dev->mac_specified = 0;
414 dev->unsupported_features = 0;
418 dev->ifname = *ifname;
422 if (virtio_user_dev_setup(dev) < 0) {
423 PMD_INIT_LOG(ERR, "backend set up fails");
427 if (!dev->is_server) {
428 if (dev->ops->send_request(dev, VHOST_USER_SET_OWNER,
430 PMD_INIT_LOG(ERR, "set_owner fails: %s",
435 if (dev->ops->send_request(dev, VHOST_USER_GET_FEATURES,
436 &dev->device_features) < 0) {
437 PMD_INIT_LOG(ERR, "get_features failed: %s",
442 /* We just pretend vhost-user can support all these features.
443 * Note that this could be problematic that if some feature is
444 * negotiated but not supported by the vhost-user which comes
447 dev->device_features = VIRTIO_USER_SUPPORTED_FEATURES;
451 dev->device_features &= ~(1ull << VIRTIO_NET_F_MRG_RXBUF);
452 dev->unsupported_features |= (1ull << VIRTIO_NET_F_MRG_RXBUF);
456 dev->device_features &= ~(1ull << VIRTIO_F_IN_ORDER);
457 dev->unsupported_features |= (1ull << VIRTIO_F_IN_ORDER);
460 if (dev->mac_specified) {
461 dev->device_features |= (1ull << VIRTIO_NET_F_MAC);
463 dev->device_features &= ~(1ull << VIRTIO_NET_F_MAC);
464 dev->unsupported_features |= (1ull << VIRTIO_NET_F_MAC);
468 /* device does not really need to know anything about CQ,
469 * so if necessary, we just claim to support CQ
471 dev->device_features |= (1ull << VIRTIO_NET_F_CTRL_VQ);
473 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_VQ);
474 /* Also disable features depends on VIRTIO_NET_F_CTRL_VQ */
475 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_RX);
476 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_VLAN);
477 dev->device_features &= ~(1ull << VIRTIO_NET_F_GUEST_ANNOUNCE);
478 dev->device_features &= ~(1ull << VIRTIO_NET_F_MQ);
479 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_MAC_ADDR);
480 dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_VQ);
481 dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_RX);
482 dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_VLAN);
483 dev->unsupported_features |=
484 (1ull << VIRTIO_NET_F_GUEST_ANNOUNCE);
485 dev->unsupported_features |= (1ull << VIRTIO_NET_F_MQ);
486 dev->unsupported_features |=
487 (1ull << VIRTIO_NET_F_CTRL_MAC_ADDR);
490 /* The backend will not report this feature, we add it explicitly */
491 if (is_vhost_user_by_type(dev->path))
492 dev->device_features |= (1ull << VIRTIO_NET_F_STATUS);
494 dev->device_features &= VIRTIO_USER_SUPPORTED_FEATURES;
495 dev->unsupported_features |= ~VIRTIO_USER_SUPPORTED_FEATURES;
497 if (rte_mem_event_callback_register(VIRTIO_USER_MEM_EVENT_CLB_NAME,
498 virtio_user_mem_event_cb, dev)) {
499 if (rte_errno != ENOTSUP) {
500 PMD_INIT_LOG(ERR, "Failed to register mem event"
510 virtio_user_dev_uninit(struct virtio_user_dev *dev)
514 virtio_user_stop_device(dev);
516 rte_mem_event_callback_unregister(VIRTIO_USER_MEM_EVENT_CLB_NAME, dev);
518 for (i = 0; i < dev->max_queue_pairs * 2; ++i) {
519 close(dev->callfds[i]);
520 close(dev->kickfds[i]);
525 if (dev->is_server && dev->listenfd >= 0) {
526 close(dev->listenfd);
531 for (i = 0; i < dev->max_queue_pairs; ++i)
532 close(dev->vhostfds[i]);
544 virtio_user_handle_mq(struct virtio_user_dev *dev, uint16_t q_pairs)
549 if (q_pairs > dev->max_queue_pairs) {
550 PMD_INIT_LOG(ERR, "multi-q config %u, but only %u supported",
551 q_pairs, dev->max_queue_pairs);
555 /* Server mode can't enable queue pairs if vhostfd is invalid,
556 * always return 0 in this case.
558 if (!dev->is_server || dev->vhostfd >= 0) {
559 for (i = 0; i < q_pairs; ++i)
560 ret |= dev->ops->enable_qp(dev, i, 1);
561 for (i = q_pairs; i < dev->max_queue_pairs; ++i)
562 ret |= dev->ops->enable_qp(dev, i, 0);
564 dev->queue_pairs = q_pairs;
570 virtio_user_handle_ctrl_msg(struct virtio_user_dev *dev, struct vring *vring,
573 struct virtio_net_ctrl_hdr *hdr;
574 virtio_net_ctrl_ack status = ~0;
575 uint16_t i, idx_data, idx_status;
576 uint32_t n_descs = 0;
578 /* locate desc for header, data, and status */
579 idx_data = vring->desc[idx_hdr].next;
583 while (vring->desc[i].flags == VRING_DESC_F_NEXT) {
584 i = vring->desc[i].next;
588 /* locate desc for status */
592 hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr;
593 if (hdr->class == VIRTIO_NET_CTRL_MQ &&
594 hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
597 queues = *(uint16_t *)(uintptr_t)vring->desc[idx_data].addr;
598 status = virtio_user_handle_mq(dev, queues);
602 *(virtio_net_ctrl_ack *)(uintptr_t)vring->desc[idx_status].addr = status;
608 virtio_user_handle_cq(struct virtio_user_dev *dev, uint16_t queue_idx)
610 uint16_t avail_idx, desc_idx;
611 struct vring_used_elem *uep;
613 struct vring *vring = &dev->vrings[queue_idx];
615 /* Consume avail ring, using used ring idx as first one */
616 while (vring->used->idx != vring->avail->idx) {
617 avail_idx = (vring->used->idx) & (vring->num - 1);
618 desc_idx = vring->avail->ring[avail_idx];
620 n_descs = virtio_user_handle_ctrl_msg(dev, vring, desc_idx);
622 /* Update used ring */
623 uep = &vring->used->ring[avail_idx];