1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
12 #include <sys/eventfd.h>
13 #include <sys/types.h>
17 #include "virtio_user_dev.h"
18 #include "../virtio_ethdev.h"
21 virtio_user_create_queue(struct virtio_user_dev *dev, uint32_t queue_sel)
23 /* Of all per virtqueue MSGs, make sure VHOST_SET_VRING_CALL come
24 * firstly because vhost depends on this msg to allocate virtqueue
27 struct vhost_vring_file file;
29 file.index = queue_sel;
30 file.fd = dev->callfds[queue_sel];
31 dev->ops->send_request(dev, VHOST_USER_SET_VRING_CALL, &file);
37 virtio_user_kick_queue(struct virtio_user_dev *dev, uint32_t queue_sel)
39 struct vhost_vring_file file;
40 struct vhost_vring_state state;
41 struct vring *vring = &dev->vrings[queue_sel];
42 struct vhost_vring_addr addr = {
44 .desc_user_addr = (uint64_t)(uintptr_t)vring->desc,
45 .avail_user_addr = (uint64_t)(uintptr_t)vring->avail,
46 .used_user_addr = (uint64_t)(uintptr_t)vring->used,
48 .flags = 0, /* disable log */
51 state.index = queue_sel;
52 state.num = vring->num;
53 dev->ops->send_request(dev, VHOST_USER_SET_VRING_NUM, &state);
55 state.index = queue_sel;
56 state.num = 0; /* no reservation */
57 dev->ops->send_request(dev, VHOST_USER_SET_VRING_BASE, &state);
59 dev->ops->send_request(dev, VHOST_USER_SET_VRING_ADDR, &addr);
61 /* Of all per virtqueue MSGs, make sure VHOST_USER_SET_VRING_KICK comes
62 * lastly because vhost depends on this msg to judge if
65 file.index = queue_sel;
66 file.fd = dev->kickfds[queue_sel];
67 dev->ops->send_request(dev, VHOST_USER_SET_VRING_KICK, &file);
73 virtio_user_queue_setup(struct virtio_user_dev *dev,
74 int (*fn)(struct virtio_user_dev *, uint32_t))
76 uint32_t i, queue_sel;
78 for (i = 0; i < dev->max_queue_pairs; ++i) {
79 queue_sel = 2 * i + VTNET_SQ_RQ_QUEUE_IDX;
80 if (fn(dev, queue_sel) < 0) {
81 PMD_DRV_LOG(INFO, "setup rx vq fails: %u", i);
85 for (i = 0; i < dev->max_queue_pairs; ++i) {
86 queue_sel = 2 * i + VTNET_SQ_TQ_QUEUE_IDX;
87 if (fn(dev, queue_sel) < 0) {
88 PMD_DRV_LOG(INFO, "setup tx vq fails: %u", i);
97 is_vhost_user_by_type(const char *path)
101 if (stat(path, &sb) == -1)
104 return S_ISSOCK(sb.st_mode);
108 virtio_user_start_device(struct virtio_user_dev *dev)
113 if (is_vhost_user_by_type(dev->path) && dev->vhostfd < 0)
116 /* Do not check return as already done in init, or reset in stop */
117 dev->ops->send_request(dev, VHOST_USER_SET_OWNER, NULL);
119 /* Step 0: tell vhost to create queues */
120 if (virtio_user_queue_setup(dev, virtio_user_create_queue) < 0)
123 /* Step 1: set features */
124 features = dev->features;
125 /* Strip VIRTIO_NET_F_MAC, as MAC address is handled in vdev init */
126 features &= ~(1ull << VIRTIO_NET_F_MAC);
127 /* Strip VIRTIO_NET_F_CTRL_VQ, as devices do not really need to know */
128 features &= ~(1ull << VIRTIO_NET_F_CTRL_VQ);
129 features &= ~(1ull << VIRTIO_NET_F_STATUS);
130 ret = dev->ops->send_request(dev, VHOST_USER_SET_FEATURES, &features);
133 PMD_DRV_LOG(INFO, "set features: %" PRIx64, features);
135 /* Step 2: share memory regions */
136 ret = dev->ops->send_request(dev, VHOST_USER_SET_MEM_TABLE, NULL);
140 /* Step 3: kick queues */
141 if (virtio_user_queue_setup(dev, virtio_user_kick_queue) < 0)
144 /* Step 4: enable queues
145 * we enable the 1st queue pair by default.
147 dev->ops->enable_qp(dev, 0, 1);
151 /* TODO: free resource here or caller to check */
155 int virtio_user_stop_device(struct virtio_user_dev *dev)
159 for (i = 0; i < dev->max_queue_pairs; ++i)
160 dev->ops->enable_qp(dev, i, 0);
162 if (dev->ops->send_request(dev, VHOST_USER_RESET_OWNER, NULL) < 0) {
163 PMD_DRV_LOG(INFO, "Failed to reset the device\n");
171 parse_mac(struct virtio_user_dev *dev, const char *mac)
174 uint32_t tmp[ETHER_ADDR_LEN];
179 r = sscanf(mac, "%x:%x:%x:%x:%x:%x", &tmp[0],
180 &tmp[1], &tmp[2], &tmp[3], &tmp[4], &tmp[5]);
181 if (r == ETHER_ADDR_LEN) {
182 for (i = 0; i < ETHER_ADDR_LEN; ++i)
183 dev->mac_addr[i] = (uint8_t)tmp[i];
184 dev->mac_specified = 1;
186 /* ignore the wrong mac, use random mac */
187 PMD_DRV_LOG(ERR, "wrong format of mac: %s", mac);
192 virtio_user_dev_init_notify(struct virtio_user_dev *dev)
198 for (i = 0; i < VIRTIO_MAX_VIRTQUEUES; ++i) {
199 if (i >= dev->max_queue_pairs * 2) {
200 dev->kickfds[i] = -1;
201 dev->callfds[i] = -1;
205 /* May use invalid flag, but some backend uses kickfd and
206 * callfd as criteria to judge if dev is alive. so finally we
209 callfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK);
211 PMD_DRV_LOG(ERR, "callfd error, %s", strerror(errno));
214 kickfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK);
216 PMD_DRV_LOG(ERR, "kickfd error, %s", strerror(errno));
219 dev->callfds[i] = callfd;
220 dev->kickfds[i] = kickfd;
223 if (i < VIRTIO_MAX_VIRTQUEUES) {
224 for (j = 0; j <= i; ++j) {
225 close(dev->callfds[j]);
226 close(dev->kickfds[j]);
236 virtio_user_fill_intr_handle(struct virtio_user_dev *dev)
239 struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->port_id];
241 if (!eth_dev->intr_handle) {
242 eth_dev->intr_handle = malloc(sizeof(*eth_dev->intr_handle));
243 if (!eth_dev->intr_handle) {
244 PMD_DRV_LOG(ERR, "fail to allocate intr_handle");
247 memset(eth_dev->intr_handle, 0, sizeof(*eth_dev->intr_handle));
250 for (i = 0; i < dev->max_queue_pairs; ++i)
251 eth_dev->intr_handle->efds[i] = dev->callfds[i];
252 eth_dev->intr_handle->nb_efd = dev->max_queue_pairs;
253 eth_dev->intr_handle->max_intr = dev->max_queue_pairs + 1;
254 eth_dev->intr_handle->type = RTE_INTR_HANDLE_VDEV;
255 /* For virtio vdev, no need to read counter for clean */
256 eth_dev->intr_handle->efd_counter_size = 0;
257 eth_dev->intr_handle->fd = -1;
258 if (dev->vhostfd >= 0)
259 eth_dev->intr_handle->fd = dev->vhostfd;
260 else if (dev->is_server)
261 eth_dev->intr_handle->fd = dev->listenfd;
267 virtio_user_dev_setup(struct virtio_user_dev *dev)
272 dev->vhostfds = NULL;
275 if (dev->is_server) {
276 if (access(dev->path, F_OK) == 0 &&
277 !is_vhost_user_by_type(dev->path)) {
278 PMD_DRV_LOG(ERR, "Server mode doesn't support vhost-kernel!");
281 dev->ops = &ops_user;
283 if (is_vhost_user_by_type(dev->path)) {
284 dev->ops = &ops_user;
286 dev->ops = &ops_kernel;
288 dev->vhostfds = malloc(dev->max_queue_pairs *
290 dev->tapfds = malloc(dev->max_queue_pairs *
292 if (!dev->vhostfds || !dev->tapfds) {
293 PMD_INIT_LOG(ERR, "Failed to malloc");
297 for (q = 0; q < dev->max_queue_pairs; ++q) {
298 dev->vhostfds[q] = -1;
304 if (dev->ops->setup(dev) < 0)
307 if (virtio_user_dev_init_notify(dev) < 0)
310 if (virtio_user_fill_intr_handle(dev) < 0)
316 /* Use below macro to filter features from vhost backend */
317 #define VIRTIO_USER_SUPPORTED_FEATURES \
318 (1ULL << VIRTIO_NET_F_MAC | \
319 1ULL << VIRTIO_NET_F_STATUS | \
320 1ULL << VIRTIO_NET_F_MQ | \
321 1ULL << VIRTIO_NET_F_CTRL_MAC_ADDR | \
322 1ULL << VIRTIO_NET_F_CTRL_VQ | \
323 1ULL << VIRTIO_NET_F_CTRL_RX | \
324 1ULL << VIRTIO_NET_F_CTRL_VLAN | \
325 1ULL << VIRTIO_NET_F_CSUM | \
326 1ULL << VIRTIO_NET_F_HOST_TSO4 | \
327 1ULL << VIRTIO_NET_F_HOST_TSO6 | \
328 1ULL << VIRTIO_NET_F_MRG_RXBUF | \
329 1ULL << VIRTIO_RING_F_INDIRECT_DESC | \
330 1ULL << VIRTIO_NET_F_GUEST_CSUM | \
331 1ULL << VIRTIO_NET_F_GUEST_TSO4 | \
332 1ULL << VIRTIO_NET_F_GUEST_TSO6 | \
333 1ULL << VIRTIO_F_VERSION_1)
336 virtio_user_dev_init(struct virtio_user_dev *dev, char *path, int queues,
337 int cq, int queue_size, const char *mac, char **ifname)
339 snprintf(dev->path, PATH_MAX, "%s", path);
340 dev->max_queue_pairs = queues;
341 dev->queue_pairs = 1; /* mq disabled by default */
342 dev->queue_size = queue_size;
343 dev->mac_specified = 0;
347 dev->ifname = *ifname;
351 if (virtio_user_dev_setup(dev) < 0) {
352 PMD_INIT_LOG(ERR, "backend set up fails");
356 if (dev->vhostfd >= 0) {
357 if (dev->ops->send_request(dev, VHOST_USER_SET_OWNER,
359 PMD_INIT_LOG(ERR, "set_owner fails: %s",
364 if (dev->ops->send_request(dev, VHOST_USER_GET_FEATURES,
365 &dev->device_features) < 0) {
366 PMD_INIT_LOG(ERR, "get_features failed: %s",
371 /* We just pretend vhost-user can support all these features.
372 * Note that this could be problematic that if some feature is
373 * negotiated but not supported by the vhost-user which comes
376 dev->device_features = VIRTIO_USER_SUPPORTED_FEATURES;
379 if (dev->mac_specified)
380 dev->device_features |= (1ull << VIRTIO_NET_F_MAC);
383 /* device does not really need to know anything about CQ,
384 * so if necessary, we just claim to support CQ
386 dev->device_features |= (1ull << VIRTIO_NET_F_CTRL_VQ);
388 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_VQ);
389 /* Also disable features depends on VIRTIO_NET_F_CTRL_VQ */
390 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_RX);
391 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_VLAN);
392 dev->device_features &= ~(1ull << VIRTIO_NET_F_GUEST_ANNOUNCE);
393 dev->device_features &= ~(1ull << VIRTIO_NET_F_MQ);
394 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_MAC_ADDR);
397 /* The backend will not report this feature, we add it explicitly */
398 if (is_vhost_user_by_type(dev->path))
399 dev->device_features |= (1ull << VIRTIO_NET_F_STATUS);
401 dev->device_features &= VIRTIO_USER_SUPPORTED_FEATURES;
407 virtio_user_dev_uninit(struct virtio_user_dev *dev)
411 virtio_user_stop_device(dev);
413 for (i = 0; i < dev->max_queue_pairs * 2; ++i) {
414 close(dev->callfds[i]);
415 close(dev->kickfds[i]);
420 if (dev->is_server && dev->listenfd >= 0) {
421 close(dev->listenfd);
426 for (i = 0; i < dev->max_queue_pairs; ++i)
427 close(dev->vhostfds[i]);
439 virtio_user_handle_mq(struct virtio_user_dev *dev, uint16_t q_pairs)
444 if (q_pairs > dev->max_queue_pairs) {
445 PMD_INIT_LOG(ERR, "multi-q config %u, but only %u supported",
446 q_pairs, dev->max_queue_pairs);
450 for (i = 0; i < q_pairs; ++i)
451 ret |= dev->ops->enable_qp(dev, i, 1);
452 for (i = q_pairs; i < dev->max_queue_pairs; ++i)
453 ret |= dev->ops->enable_qp(dev, i, 0);
455 dev->queue_pairs = q_pairs;
461 virtio_user_handle_ctrl_msg(struct virtio_user_dev *dev, struct vring *vring,
464 struct virtio_net_ctrl_hdr *hdr;
465 virtio_net_ctrl_ack status = ~0;
466 uint16_t i, idx_data, idx_status;
467 uint32_t n_descs = 0;
469 /* locate desc for header, data, and status */
470 idx_data = vring->desc[idx_hdr].next;
474 while (vring->desc[i].flags == VRING_DESC_F_NEXT) {
475 i = vring->desc[i].next;
479 /* locate desc for status */
483 hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr;
484 if (hdr->class == VIRTIO_NET_CTRL_MQ &&
485 hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
488 queues = *(uint16_t *)(uintptr_t)vring->desc[idx_data].addr;
489 status = virtio_user_handle_mq(dev, queues);
493 *(virtio_net_ctrl_ack *)(uintptr_t)vring->desc[idx_status].addr = status;
499 virtio_user_handle_cq(struct virtio_user_dev *dev, uint16_t queue_idx)
501 uint16_t avail_idx, desc_idx;
502 struct vring_used_elem *uep;
504 struct vring *vring = &dev->vrings[queue_idx];
506 /* Consume avail ring, using used ring idx as first one */
507 while (vring->used->idx != vring->avail->idx) {
508 avail_idx = (vring->used->idx) & (vring->num - 1);
509 desc_idx = vring->avail->ring[avail_idx];
511 n_descs = virtio_user_handle_ctrl_msg(dev, vring, desc_idx);
513 /* Update used ring */
514 uep = &vring->used->ring[avail_idx];