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 if (dev->ops->send_request(dev, VHOST_USER_RESET_OWNER, NULL) < 0) { 146 PMD_DRV_LOG(INFO, "Failed to reset the device\n"); 147 return -1; 148 } 149 150 return 0; 151 } 152 153 static inline void 154 parse_mac(struct virtio_user_dev *dev, const char *mac) 155 { 156 int i, r; 157 uint32_t tmp[ETHER_ADDR_LEN]; 158 159 if (!mac) 160 return; 161 162 r = sscanf(mac, "%x:%x:%x:%x:%x:%x", &tmp[0], 163 &tmp[1], &tmp[2], &tmp[3], &tmp[4], &tmp[5]); 164 if (r == ETHER_ADDR_LEN) { 165 for (i = 0; i < ETHER_ADDR_LEN; ++i) 166 dev->mac_addr[i] = (uint8_t)tmp[i]; 167 dev->mac_specified = 1; 168 } else { 169 /* ignore the wrong mac, use random mac */ 170 PMD_DRV_LOG(ERR, "wrong format of mac: %s", mac); 171 } 172 } 173 174 int 175 is_vhost_user_by_type(const char *path) 176 { 177 struct stat sb; 178 179 if (stat(path, &sb) == -1) 180 return 0; 181 182 return S_ISSOCK(sb.st_mode); 183 } 184 185 static int 186 virtio_user_dev_init_notify(struct virtio_user_dev *dev) 187 { 188 uint32_t i, j; 189 int callfd; 190 int kickfd; 191 192 for (i = 0; i < VIRTIO_MAX_VIRTQUEUES; ++i) { 193 if (i >= dev->max_queue_pairs * 2) { 194 dev->kickfds[i] = -1; 195 dev->callfds[i] = -1; 196 continue; 197 } 198 199 /* May use invalid flag, but some backend uses kickfd and 200 * callfd as criteria to judge if dev is alive. so finally we 201 * use real event_fd. 202 */ 203 callfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK); 204 if (callfd < 0) { 205 PMD_DRV_LOG(ERR, "callfd error, %s", strerror(errno)); 206 break; 207 } 208 kickfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK); 209 if (kickfd < 0) { 210 PMD_DRV_LOG(ERR, "kickfd error, %s", strerror(errno)); 211 break; 212 } 213 dev->callfds[i] = callfd; 214 dev->kickfds[i] = kickfd; 215 } 216 217 if (i < VIRTIO_MAX_VIRTQUEUES) { 218 for (j = 0; j <= i; ++j) { 219 close(dev->callfds[j]); 220 close(dev->kickfds[j]); 221 } 222 223 return -1; 224 } 225 226 return 0; 227 } 228 229 static int 230 virtio_user_fill_intr_handle(struct virtio_user_dev *dev) 231 { 232 uint32_t i; 233 struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->port_id]; 234 235 if (!eth_dev->intr_handle) { 236 eth_dev->intr_handle = malloc(sizeof(*eth_dev->intr_handle)); 237 if (!eth_dev->intr_handle) { 238 PMD_DRV_LOG(ERR, "fail to allocate intr_handle"); 239 return -1; 240 } 241 memset(eth_dev->intr_handle, 0, sizeof(*eth_dev->intr_handle)); 242 } 243 244 for (i = 0; i < dev->max_queue_pairs; ++i) 245 eth_dev->intr_handle->efds[i] = dev->callfds[i]; 246 eth_dev->intr_handle->nb_efd = dev->max_queue_pairs; 247 eth_dev->intr_handle->max_intr = dev->max_queue_pairs + 1; 248 eth_dev->intr_handle->type = RTE_INTR_HANDLE_VDEV; 249 /* For virtio vdev, no need to read counter for clean */ 250 eth_dev->intr_handle->efd_counter_size = 0; 251 if (dev->vhostfd >= 0) 252 eth_dev->intr_handle->fd = dev->vhostfd; 253 254 return 0; 255 } 256 257 static int 258 virtio_user_dev_setup(struct virtio_user_dev *dev) 259 { 260 uint32_t q; 261 262 dev->vhostfd = -1; 263 dev->vhostfds = NULL; 264 dev->tapfds = NULL; 265 266 if (is_vhost_user_by_type(dev->path)) { 267 dev->ops = &ops_user; 268 } else { 269 dev->ops = &ops_kernel; 270 271 dev->vhostfds = malloc(dev->max_queue_pairs * sizeof(int)); 272 dev->tapfds = malloc(dev->max_queue_pairs * sizeof(int)); 273 if (!dev->vhostfds || !dev->tapfds) { 274 PMD_INIT_LOG(ERR, "Failed to malloc"); 275 return -1; 276 } 277 278 for (q = 0; q < dev->max_queue_pairs; ++q) { 279 dev->vhostfds[q] = -1; 280 dev->tapfds[q] = -1; 281 } 282 } 283 284 if (dev->ops->setup(dev) < 0) 285 return -1; 286 287 if (virtio_user_dev_init_notify(dev) < 0) 288 return -1; 289 290 if (virtio_user_fill_intr_handle(dev) < 0) 291 return -1; 292 293 return 0; 294 } 295 296 /* Use below macro to filter features from vhost backend */ 297 #define VIRTIO_USER_SUPPORTED_FEATURES \ 298 (1ULL << VIRTIO_NET_F_MAC | \ 299 1ULL << VIRTIO_NET_F_STATUS | \ 300 1ULL << VIRTIO_NET_F_MQ | \ 301 1ULL << VIRTIO_NET_F_CTRL_MAC_ADDR | \ 302 1ULL << VIRTIO_NET_F_CTRL_VQ | \ 303 1ULL << VIRTIO_NET_F_CTRL_RX | \ 304 1ULL << VIRTIO_NET_F_CTRL_VLAN | \ 305 1ULL << VIRTIO_NET_F_CSUM | \ 306 1ULL << VIRTIO_NET_F_HOST_TSO4 | \ 307 1ULL << VIRTIO_NET_F_HOST_TSO6 | \ 308 1ULL << VIRTIO_NET_F_MRG_RXBUF | \ 309 1ULL << VIRTIO_RING_F_INDIRECT_DESC | \ 310 1ULL << VIRTIO_NET_F_GUEST_CSUM | \ 311 1ULL << VIRTIO_NET_F_GUEST_TSO4 | \ 312 1ULL << VIRTIO_NET_F_GUEST_TSO6 | \ 313 1ULL << VIRTIO_F_VERSION_1) 314 315 int 316 virtio_user_dev_init(struct virtio_user_dev *dev, char *path, int queues, 317 int cq, int queue_size, const char *mac, char **ifname) 318 { 319 snprintf(dev->path, PATH_MAX, "%s", path); 320 dev->max_queue_pairs = queues; 321 dev->queue_pairs = 1; /* mq disabled by default */ 322 dev->queue_size = queue_size; 323 dev->mac_specified = 0; 324 parse_mac(dev, mac); 325 326 if (*ifname) { 327 dev->ifname = *ifname; 328 *ifname = NULL; 329 } 330 331 if (virtio_user_dev_setup(dev) < 0) { 332 PMD_INIT_LOG(ERR, "backend set up fails"); 333 return -1; 334 } 335 if (dev->ops->send_request(dev, VHOST_USER_SET_OWNER, NULL) < 0) { 336 PMD_INIT_LOG(ERR, "set_owner fails: %s", strerror(errno)); 337 return -1; 338 } 339 340 if (dev->ops->send_request(dev, VHOST_USER_GET_FEATURES, 341 &dev->device_features) < 0) { 342 PMD_INIT_LOG(ERR, "get_features failed: %s", strerror(errno)); 343 return -1; 344 } 345 if (dev->mac_specified) 346 dev->device_features |= (1ull << VIRTIO_NET_F_MAC); 347 348 if (cq) { 349 /* device does not really need to know anything about CQ, 350 * so if necessary, we just claim to support CQ 351 */ 352 dev->device_features |= (1ull << VIRTIO_NET_F_CTRL_VQ); 353 } else { 354 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_VQ); 355 /* Also disable features depends on VIRTIO_NET_F_CTRL_VQ */ 356 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_RX); 357 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_VLAN); 358 dev->device_features &= ~(1ull << VIRTIO_NET_F_GUEST_ANNOUNCE); 359 dev->device_features &= ~(1ull << VIRTIO_NET_F_MQ); 360 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_MAC_ADDR); 361 } 362 363 /* The backend will not report this feature, we add it explicitly */ 364 if (is_vhost_user_by_type(dev->path)) 365 dev->device_features |= (1ull << VIRTIO_NET_F_STATUS); 366 367 dev->device_features &= VIRTIO_USER_SUPPORTED_FEATURES; 368 369 return 0; 370 } 371 372 void 373 virtio_user_dev_uninit(struct virtio_user_dev *dev) 374 { 375 uint32_t i; 376 377 virtio_user_stop_device(dev); 378 379 for (i = 0; i < dev->max_queue_pairs * 2; ++i) { 380 close(dev->callfds[i]); 381 close(dev->kickfds[i]); 382 } 383 384 close(dev->vhostfd); 385 386 if (dev->vhostfds) { 387 for (i = 0; i < dev->max_queue_pairs; ++i) 388 close(dev->vhostfds[i]); 389 free(dev->vhostfds); 390 free(dev->tapfds); 391 } 392 393 free(dev->ifname); 394 } 395 396 static uint8_t 397 virtio_user_handle_mq(struct virtio_user_dev *dev, uint16_t q_pairs) 398 { 399 uint16_t i; 400 uint8_t ret = 0; 401 402 if (q_pairs > dev->max_queue_pairs) { 403 PMD_INIT_LOG(ERR, "multi-q config %u, but only %u supported", 404 q_pairs, dev->max_queue_pairs); 405 return -1; 406 } 407 408 for (i = 0; i < q_pairs; ++i) 409 ret |= dev->ops->enable_qp(dev, i, 1); 410 for (i = q_pairs; i < dev->max_queue_pairs; ++i) 411 ret |= dev->ops->enable_qp(dev, i, 0); 412 413 dev->queue_pairs = q_pairs; 414 415 return ret; 416 } 417 418 static uint32_t 419 virtio_user_handle_ctrl_msg(struct virtio_user_dev *dev, struct vring *vring, 420 uint16_t idx_hdr) 421 { 422 struct virtio_net_ctrl_hdr *hdr; 423 virtio_net_ctrl_ack status = ~0; 424 uint16_t i, idx_data, idx_status; 425 uint32_t n_descs = 0; 426 427 /* locate desc for header, data, and status */ 428 idx_data = vring->desc[idx_hdr].next; 429 n_descs++; 430 431 i = idx_data; 432 while (vring->desc[i].flags == VRING_DESC_F_NEXT) { 433 i = vring->desc[i].next; 434 n_descs++; 435 } 436 437 /* locate desc for status */ 438 idx_status = i; 439 n_descs++; 440 441 hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr; 442 if (hdr->class == VIRTIO_NET_CTRL_MQ && 443 hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) { 444 uint16_t queues; 445 446 queues = *(uint16_t *)(uintptr_t)vring->desc[idx_data].addr; 447 status = virtio_user_handle_mq(dev, queues); 448 } 449 450 /* Update status */ 451 *(virtio_net_ctrl_ack *)(uintptr_t)vring->desc[idx_status].addr = status; 452 453 return n_descs; 454 } 455 456 void 457 virtio_user_handle_cq(struct virtio_user_dev *dev, uint16_t queue_idx) 458 { 459 uint16_t avail_idx, desc_idx; 460 struct vring_used_elem *uep; 461 uint32_t n_descs; 462 struct vring *vring = &dev->vrings[queue_idx]; 463 464 /* Consume avail ring, using used ring idx as first one */ 465 while (vring->used->idx != vring->avail->idx) { 466 avail_idx = (vring->used->idx) & (vring->num - 1); 467 desc_idx = vring->avail->ring[avail_idx]; 468 469 n_descs = virtio_user_handle_ctrl_msg(dev, vring, desc_idx); 470 471 /* Update used ring */ 472 uep = &vring->used->ring[avail_idx]; 473 uep->id = avail_idx; 474 uep->len = n_descs; 475 476 vring->used->idx++; 477 } 478 } 479