1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright(c) 2010-2016 Intel Corporation 3 */ 4 5 #include <stdint.h> 6 #include <stdio.h> 7 #include <fcntl.h> 8 #include <string.h> 9 #include <errno.h> 10 #include <sys/mman.h> 11 #include <unistd.h> 12 #include <sys/eventfd.h> 13 #include <sys/types.h> 14 #include <sys/stat.h> 15 16 #include "vhost.h" 17 #include "virtio_user_dev.h" 18 #include "../virtio_ethdev.h" 19 20 static int 21 virtio_user_create_queue(struct virtio_user_dev *dev, uint32_t queue_sel) 22 { 23 /* Of all per virtqueue MSGs, make sure VHOST_SET_VRING_CALL come 24 * firstly because vhost depends on this msg to allocate virtqueue 25 * pair. 26 */ 27 struct vhost_vring_file file; 28 29 file.index = queue_sel; 30 file.fd = dev->callfds[queue_sel]; 31 dev->ops->send_request(dev, VHOST_USER_SET_VRING_CALL, &file); 32 33 return 0; 34 } 35 36 static int 37 virtio_user_kick_queue(struct virtio_user_dev *dev, uint32_t queue_sel) 38 { 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 = { 43 .index = queue_sel, 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, 47 .log_guest_addr = 0, 48 .flags = 0, /* disable log */ 49 }; 50 51 state.index = queue_sel; 52 state.num = vring->num; 53 dev->ops->send_request(dev, VHOST_USER_SET_VRING_NUM, &state); 54 55 state.index = queue_sel; 56 state.num = 0; /* no reservation */ 57 dev->ops->send_request(dev, VHOST_USER_SET_VRING_BASE, &state); 58 59 dev->ops->send_request(dev, VHOST_USER_SET_VRING_ADDR, &addr); 60 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 63 * virtio is ready. 64 */ 65 file.index = queue_sel; 66 file.fd = dev->kickfds[queue_sel]; 67 dev->ops->send_request(dev, VHOST_USER_SET_VRING_KICK, &file); 68 69 return 0; 70 } 71 72 static int 73 virtio_user_queue_setup(struct virtio_user_dev *dev, 74 int (*fn)(struct virtio_user_dev *, uint32_t)) 75 { 76 uint32_t i, queue_sel; 77 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); 82 return -1; 83 } 84 } 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); 89 return -1; 90 } 91 } 92 93 return 0; 94 } 95 96 int 97 virtio_user_start_device(struct virtio_user_dev *dev) 98 { 99 uint64_t features; 100 int ret; 101 102 /* Step 0: tell vhost to create queues */ 103 if (virtio_user_queue_setup(dev, virtio_user_create_queue) < 0) 104 goto error; 105 106 /* Step 1: set features */ 107 features = dev->features; 108 /* Strip VIRTIO_NET_F_MAC, as MAC address is handled in vdev init */ 109 features &= ~(1ull << VIRTIO_NET_F_MAC); 110 /* Strip VIRTIO_NET_F_CTRL_VQ, as devices do not really need to know */ 111 features &= ~(1ull << VIRTIO_NET_F_CTRL_VQ); 112 features &= ~(1ull << VIRTIO_NET_F_STATUS); 113 ret = dev->ops->send_request(dev, VHOST_USER_SET_FEATURES, &features); 114 if (ret < 0) 115 goto error; 116 PMD_DRV_LOG(INFO, "set features: %" PRIx64, features); 117 118 /* Step 2: share memory regions */ 119 ret = dev->ops->send_request(dev, VHOST_USER_SET_MEM_TABLE, NULL); 120 if (ret < 0) 121 goto error; 122 123 /* Step 3: kick queues */ 124 if (virtio_user_queue_setup(dev, virtio_user_kick_queue) < 0) 125 goto error; 126 127 /* Step 4: enable queues 128 * we enable the 1st queue pair by default. 129 */ 130 dev->ops->enable_qp(dev, 0, 1); 131 132 return 0; 133 error: 134 /* TODO: free resource here or caller to check */ 135 return -1; 136 } 137 138 int virtio_user_stop_device(struct virtio_user_dev *dev) 139 { 140 uint32_t i; 141 142 for (i = 0; i < dev->max_queue_pairs; ++i) 143 dev->ops->enable_qp(dev, i, 0); 144 145 return 0; 146 } 147 148 static inline void 149 parse_mac(struct virtio_user_dev *dev, const char *mac) 150 { 151 int i, r; 152 uint32_t tmp[ETHER_ADDR_LEN]; 153 154 if (!mac) 155 return; 156 157 r = sscanf(mac, "%x:%x:%x:%x:%x:%x", &tmp[0], 158 &tmp[1], &tmp[2], &tmp[3], &tmp[4], &tmp[5]); 159 if (r == ETHER_ADDR_LEN) { 160 for (i = 0; i < ETHER_ADDR_LEN; ++i) 161 dev->mac_addr[i] = (uint8_t)tmp[i]; 162 dev->mac_specified = 1; 163 } else { 164 /* ignore the wrong mac, use random mac */ 165 PMD_DRV_LOG(ERR, "wrong format of mac: %s", mac); 166 } 167 } 168 169 int 170 is_vhost_user_by_type(const char *path) 171 { 172 struct stat sb; 173 174 if (stat(path, &sb) == -1) 175 return 0; 176 177 return S_ISSOCK(sb.st_mode); 178 } 179 180 static int 181 virtio_user_dev_init_notify(struct virtio_user_dev *dev) 182 { 183 uint32_t i, j; 184 int callfd; 185 int kickfd; 186 187 for (i = 0; i < VIRTIO_MAX_VIRTQUEUES; ++i) { 188 if (i >= dev->max_queue_pairs * 2) { 189 dev->kickfds[i] = -1; 190 dev->callfds[i] = -1; 191 continue; 192 } 193 194 /* May use invalid flag, but some backend uses kickfd and 195 * callfd as criteria to judge if dev is alive. so finally we 196 * use real event_fd. 197 */ 198 callfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK); 199 if (callfd < 0) { 200 PMD_DRV_LOG(ERR, "callfd error, %s", strerror(errno)); 201 break; 202 } 203 kickfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK); 204 if (kickfd < 0) { 205 PMD_DRV_LOG(ERR, "kickfd error, %s", strerror(errno)); 206 break; 207 } 208 dev->callfds[i] = callfd; 209 dev->kickfds[i] = kickfd; 210 } 211 212 if (i < VIRTIO_MAX_VIRTQUEUES) { 213 for (j = 0; j <= i; ++j) { 214 close(dev->callfds[j]); 215 close(dev->kickfds[j]); 216 } 217 218 return -1; 219 } 220 221 return 0; 222 } 223 224 static int 225 virtio_user_fill_intr_handle(struct virtio_user_dev *dev) 226 { 227 uint32_t i; 228 struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->port_id]; 229 230 if (!eth_dev->intr_handle) { 231 eth_dev->intr_handle = malloc(sizeof(*eth_dev->intr_handle)); 232 if (!eth_dev->intr_handle) { 233 PMD_DRV_LOG(ERR, "fail to allocate intr_handle"); 234 return -1; 235 } 236 memset(eth_dev->intr_handle, 0, sizeof(*eth_dev->intr_handle)); 237 } 238 239 for (i = 0; i < dev->max_queue_pairs; ++i) 240 eth_dev->intr_handle->efds[i] = dev->callfds[i]; 241 eth_dev->intr_handle->nb_efd = dev->max_queue_pairs; 242 eth_dev->intr_handle->max_intr = dev->max_queue_pairs + 1; 243 eth_dev->intr_handle->type = RTE_INTR_HANDLE_VDEV; 244 /* For virtio vdev, no need to read counter for clean */ 245 eth_dev->intr_handle->efd_counter_size = 0; 246 if (dev->vhostfd >= 0) 247 eth_dev->intr_handle->fd = dev->vhostfd; 248 249 return 0; 250 } 251 252 static int 253 virtio_user_dev_setup(struct virtio_user_dev *dev) 254 { 255 uint32_t q; 256 257 dev->vhostfd = -1; 258 dev->vhostfds = NULL; 259 dev->tapfds = NULL; 260 261 if (is_vhost_user_by_type(dev->path)) { 262 dev->ops = &ops_user; 263 } else { 264 dev->ops = &ops_kernel; 265 266 dev->vhostfds = malloc(dev->max_queue_pairs * sizeof(int)); 267 dev->tapfds = malloc(dev->max_queue_pairs * sizeof(int)); 268 if (!dev->vhostfds || !dev->tapfds) { 269 PMD_INIT_LOG(ERR, "Failed to malloc"); 270 return -1; 271 } 272 273 for (q = 0; q < dev->max_queue_pairs; ++q) { 274 dev->vhostfds[q] = -1; 275 dev->tapfds[q] = -1; 276 } 277 } 278 279 if (dev->ops->setup(dev) < 0) 280 return -1; 281 282 if (virtio_user_dev_init_notify(dev) < 0) 283 return -1; 284 285 if (virtio_user_fill_intr_handle(dev) < 0) 286 return -1; 287 288 return 0; 289 } 290 291 /* Use below macro to filter features from vhost backend */ 292 #define VIRTIO_USER_SUPPORTED_FEATURES \ 293 (1ULL << VIRTIO_NET_F_MAC | \ 294 1ULL << VIRTIO_NET_F_STATUS | \ 295 1ULL << VIRTIO_NET_F_MQ | \ 296 1ULL << VIRTIO_NET_F_CTRL_MAC_ADDR | \ 297 1ULL << VIRTIO_NET_F_CTRL_VQ | \ 298 1ULL << VIRTIO_NET_F_CTRL_RX | \ 299 1ULL << VIRTIO_NET_F_CTRL_VLAN | \ 300 1ULL << VIRTIO_NET_F_CSUM | \ 301 1ULL << VIRTIO_NET_F_HOST_TSO4 | \ 302 1ULL << VIRTIO_NET_F_HOST_TSO6 | \ 303 1ULL << VIRTIO_NET_F_MRG_RXBUF | \ 304 1ULL << VIRTIO_RING_F_INDIRECT_DESC | \ 305 1ULL << VIRTIO_NET_F_GUEST_CSUM | \ 306 1ULL << VIRTIO_NET_F_GUEST_TSO4 | \ 307 1ULL << VIRTIO_NET_F_GUEST_TSO6 | \ 308 1ULL << VIRTIO_F_VERSION_1) 309 310 int 311 virtio_user_dev_init(struct virtio_user_dev *dev, char *path, int queues, 312 int cq, int queue_size, const char *mac, char **ifname) 313 { 314 snprintf(dev->path, PATH_MAX, "%s", path); 315 dev->max_queue_pairs = queues; 316 dev->queue_pairs = 1; /* mq disabled by default */ 317 dev->queue_size = queue_size; 318 dev->mac_specified = 0; 319 parse_mac(dev, mac); 320 321 if (*ifname) { 322 dev->ifname = *ifname; 323 *ifname = NULL; 324 } 325 326 if (virtio_user_dev_setup(dev) < 0) { 327 PMD_INIT_LOG(ERR, "backend set up fails"); 328 return -1; 329 } 330 if (dev->ops->send_request(dev, VHOST_USER_SET_OWNER, NULL) < 0) { 331 PMD_INIT_LOG(ERR, "set_owner fails: %s", strerror(errno)); 332 return -1; 333 } 334 335 if (dev->ops->send_request(dev, VHOST_USER_GET_FEATURES, 336 &dev->device_features) < 0) { 337 PMD_INIT_LOG(ERR, "get_features failed: %s", strerror(errno)); 338 return -1; 339 } 340 if (dev->mac_specified) 341 dev->device_features |= (1ull << VIRTIO_NET_F_MAC); 342 343 if (cq) { 344 /* device does not really need to know anything about CQ, 345 * so if necessary, we just claim to support CQ 346 */ 347 dev->device_features |= (1ull << VIRTIO_NET_F_CTRL_VQ); 348 } else { 349 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_VQ); 350 /* Also disable features depends on VIRTIO_NET_F_CTRL_VQ */ 351 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_RX); 352 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_VLAN); 353 dev->device_features &= ~(1ull << VIRTIO_NET_F_GUEST_ANNOUNCE); 354 dev->device_features &= ~(1ull << VIRTIO_NET_F_MQ); 355 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_MAC_ADDR); 356 } 357 358 /* The backend will not report this feature, we add it explicitly */ 359 if (is_vhost_user_by_type(dev->path)) 360 dev->device_features |= (1ull << VIRTIO_NET_F_STATUS); 361 362 dev->device_features &= VIRTIO_USER_SUPPORTED_FEATURES; 363 364 return 0; 365 } 366 367 void 368 virtio_user_dev_uninit(struct virtio_user_dev *dev) 369 { 370 uint32_t i; 371 372 virtio_user_stop_device(dev); 373 374 for (i = 0; i < dev->max_queue_pairs * 2; ++i) { 375 close(dev->callfds[i]); 376 close(dev->kickfds[i]); 377 } 378 379 close(dev->vhostfd); 380 381 if (dev->vhostfds) { 382 for (i = 0; i < dev->max_queue_pairs; ++i) 383 close(dev->vhostfds[i]); 384 free(dev->vhostfds); 385 free(dev->tapfds); 386 } 387 388 free(dev->ifname); 389 } 390 391 static uint8_t 392 virtio_user_handle_mq(struct virtio_user_dev *dev, uint16_t q_pairs) 393 { 394 uint16_t i; 395 uint8_t ret = 0; 396 397 if (q_pairs > dev->max_queue_pairs) { 398 PMD_INIT_LOG(ERR, "multi-q config %u, but only %u supported", 399 q_pairs, dev->max_queue_pairs); 400 return -1; 401 } 402 403 for (i = 0; i < q_pairs; ++i) 404 ret |= dev->ops->enable_qp(dev, i, 1); 405 for (i = q_pairs; i < dev->max_queue_pairs; ++i) 406 ret |= dev->ops->enable_qp(dev, i, 0); 407 408 dev->queue_pairs = q_pairs; 409 410 return ret; 411 } 412 413 static uint32_t 414 virtio_user_handle_ctrl_msg(struct virtio_user_dev *dev, struct vring *vring, 415 uint16_t idx_hdr) 416 { 417 struct virtio_net_ctrl_hdr *hdr; 418 virtio_net_ctrl_ack status = ~0; 419 uint16_t i, idx_data, idx_status; 420 uint32_t n_descs = 0; 421 422 /* locate desc for header, data, and status */ 423 idx_data = vring->desc[idx_hdr].next; 424 n_descs++; 425 426 i = idx_data; 427 while (vring->desc[i].flags == VRING_DESC_F_NEXT) { 428 i = vring->desc[i].next; 429 n_descs++; 430 } 431 432 /* locate desc for status */ 433 idx_status = i; 434 n_descs++; 435 436 hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr; 437 if (hdr->class == VIRTIO_NET_CTRL_MQ && 438 hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) { 439 uint16_t queues; 440 441 queues = *(uint16_t *)(uintptr_t)vring->desc[idx_data].addr; 442 status = virtio_user_handle_mq(dev, queues); 443 } 444 445 /* Update status */ 446 *(virtio_net_ctrl_ack *)(uintptr_t)vring->desc[idx_status].addr = status; 447 448 return n_descs; 449 } 450 451 void 452 virtio_user_handle_cq(struct virtio_user_dev *dev, uint16_t queue_idx) 453 { 454 uint16_t avail_idx, desc_idx; 455 struct vring_used_elem *uep; 456 uint32_t n_descs; 457 struct vring *vring = &dev->vrings[queue_idx]; 458 459 /* Consume avail ring, using used ring idx as first one */ 460 while (vring->used->idx != vring->avail->idx) { 461 avail_idx = (vring->used->idx) & (vring->num - 1); 462 desc_idx = vring->avail->ring[avail_idx]; 463 464 n_descs = virtio_user_handle_ctrl_msg(dev, vring, desc_idx); 465 466 /* Update used ring */ 467 uep = &vring->used->ring[avail_idx]; 468 uep->id = avail_idx; 469 uep->len = n_descs; 470 471 vring->used->idx++; 472 } 473 } 474