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 /* Do not check return as already done in init, or reset in stop */ 103 dev->ops->send_request(dev, VHOST_USER_SET_OWNER, NULL); 104 105 /* Step 0: tell vhost to create queues */ 106 if (virtio_user_queue_setup(dev, virtio_user_create_queue) < 0) 107 goto error; 108 109 /* Step 1: set features */ 110 features = dev->features; 111 /* Strip VIRTIO_NET_F_MAC, as MAC address is handled in vdev init */ 112 features &= ~(1ull << VIRTIO_NET_F_MAC); 113 /* Strip VIRTIO_NET_F_CTRL_VQ, as devices do not really need to know */ 114 features &= ~(1ull << VIRTIO_NET_F_CTRL_VQ); 115 features &= ~(1ull << VIRTIO_NET_F_STATUS); 116 ret = dev->ops->send_request(dev, VHOST_USER_SET_FEATURES, &features); 117 if (ret < 0) 118 goto error; 119 PMD_DRV_LOG(INFO, "set features: %" PRIx64, features); 120 121 /* Step 2: share memory regions */ 122 ret = dev->ops->send_request(dev, VHOST_USER_SET_MEM_TABLE, NULL); 123 if (ret < 0) 124 goto error; 125 126 /* Step 3: kick queues */ 127 if (virtio_user_queue_setup(dev, virtio_user_kick_queue) < 0) 128 goto error; 129 130 /* Step 4: enable queues 131 * we enable the 1st queue pair by default. 132 */ 133 dev->ops->enable_qp(dev, 0, 1); 134 135 return 0; 136 error: 137 /* TODO: free resource here or caller to check */ 138 return -1; 139 } 140 141 int virtio_user_stop_device(struct virtio_user_dev *dev) 142 { 143 uint32_t i; 144 145 for (i = 0; i < dev->max_queue_pairs; ++i) 146 dev->ops->enable_qp(dev, i, 0); 147 148 if (dev->ops->send_request(dev, VHOST_USER_RESET_OWNER, NULL) < 0) { 149 PMD_DRV_LOG(INFO, "Failed to reset the device\n"); 150 return -1; 151 } 152 153 return 0; 154 } 155 156 static inline void 157 parse_mac(struct virtio_user_dev *dev, const char *mac) 158 { 159 int i, r; 160 uint32_t tmp[ETHER_ADDR_LEN]; 161 162 if (!mac) 163 return; 164 165 r = sscanf(mac, "%x:%x:%x:%x:%x:%x", &tmp[0], 166 &tmp[1], &tmp[2], &tmp[3], &tmp[4], &tmp[5]); 167 if (r == ETHER_ADDR_LEN) { 168 for (i = 0; i < ETHER_ADDR_LEN; ++i) 169 dev->mac_addr[i] = (uint8_t)tmp[i]; 170 dev->mac_specified = 1; 171 } else { 172 /* ignore the wrong mac, use random mac */ 173 PMD_DRV_LOG(ERR, "wrong format of mac: %s", mac); 174 } 175 } 176 177 int 178 is_vhost_user_by_type(const char *path) 179 { 180 struct stat sb; 181 182 if (stat(path, &sb) == -1) 183 return 0; 184 185 return S_ISSOCK(sb.st_mode); 186 } 187 188 static int 189 virtio_user_dev_init_notify(struct virtio_user_dev *dev) 190 { 191 uint32_t i, j; 192 int callfd; 193 int kickfd; 194 195 for (i = 0; i < VIRTIO_MAX_VIRTQUEUES; ++i) { 196 if (i >= dev->max_queue_pairs * 2) { 197 dev->kickfds[i] = -1; 198 dev->callfds[i] = -1; 199 continue; 200 } 201 202 /* May use invalid flag, but some backend uses kickfd and 203 * callfd as criteria to judge if dev is alive. so finally we 204 * use real event_fd. 205 */ 206 callfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK); 207 if (callfd < 0) { 208 PMD_DRV_LOG(ERR, "callfd error, %s", strerror(errno)); 209 break; 210 } 211 kickfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK); 212 if (kickfd < 0) { 213 PMD_DRV_LOG(ERR, "kickfd error, %s", strerror(errno)); 214 break; 215 } 216 dev->callfds[i] = callfd; 217 dev->kickfds[i] = kickfd; 218 } 219 220 if (i < VIRTIO_MAX_VIRTQUEUES) { 221 for (j = 0; j <= i; ++j) { 222 close(dev->callfds[j]); 223 close(dev->kickfds[j]); 224 } 225 226 return -1; 227 } 228 229 return 0; 230 } 231 232 static int 233 virtio_user_fill_intr_handle(struct virtio_user_dev *dev) 234 { 235 uint32_t i; 236 struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->port_id]; 237 238 if (!eth_dev->intr_handle) { 239 eth_dev->intr_handle = malloc(sizeof(*eth_dev->intr_handle)); 240 if (!eth_dev->intr_handle) { 241 PMD_DRV_LOG(ERR, "fail to allocate intr_handle"); 242 return -1; 243 } 244 memset(eth_dev->intr_handle, 0, sizeof(*eth_dev->intr_handle)); 245 } 246 247 for (i = 0; i < dev->max_queue_pairs; ++i) 248 eth_dev->intr_handle->efds[i] = dev->callfds[i]; 249 eth_dev->intr_handle->nb_efd = dev->max_queue_pairs; 250 eth_dev->intr_handle->max_intr = dev->max_queue_pairs + 1; 251 eth_dev->intr_handle->type = RTE_INTR_HANDLE_VDEV; 252 /* For virtio vdev, no need to read counter for clean */ 253 eth_dev->intr_handle->efd_counter_size = 0; 254 eth_dev->intr_handle->fd = -1; 255 if (dev->vhostfd >= 0) 256 eth_dev->intr_handle->fd = dev->vhostfd; 257 258 return 0; 259 } 260 261 static int 262 virtio_user_dev_setup(struct virtio_user_dev *dev) 263 { 264 uint32_t q; 265 266 dev->vhostfd = -1; 267 dev->vhostfds = NULL; 268 dev->tapfds = NULL; 269 270 if (is_vhost_user_by_type(dev->path)) { 271 dev->ops = &ops_user; 272 } else { 273 dev->ops = &ops_kernel; 274 275 dev->vhostfds = malloc(dev->max_queue_pairs * sizeof(int)); 276 dev->tapfds = malloc(dev->max_queue_pairs * sizeof(int)); 277 if (!dev->vhostfds || !dev->tapfds) { 278 PMD_INIT_LOG(ERR, "Failed to malloc"); 279 return -1; 280 } 281 282 for (q = 0; q < dev->max_queue_pairs; ++q) { 283 dev->vhostfds[q] = -1; 284 dev->tapfds[q] = -1; 285 } 286 } 287 288 if (dev->ops->setup(dev) < 0) 289 return -1; 290 291 if (virtio_user_dev_init_notify(dev) < 0) 292 return -1; 293 294 if (virtio_user_fill_intr_handle(dev) < 0) 295 return -1; 296 297 return 0; 298 } 299 300 /* Use below macro to filter features from vhost backend */ 301 #define VIRTIO_USER_SUPPORTED_FEATURES \ 302 (1ULL << VIRTIO_NET_F_MAC | \ 303 1ULL << VIRTIO_NET_F_STATUS | \ 304 1ULL << VIRTIO_NET_F_MQ | \ 305 1ULL << VIRTIO_NET_F_CTRL_MAC_ADDR | \ 306 1ULL << VIRTIO_NET_F_CTRL_VQ | \ 307 1ULL << VIRTIO_NET_F_CTRL_RX | \ 308 1ULL << VIRTIO_NET_F_CTRL_VLAN | \ 309 1ULL << VIRTIO_NET_F_CSUM | \ 310 1ULL << VIRTIO_NET_F_HOST_TSO4 | \ 311 1ULL << VIRTIO_NET_F_HOST_TSO6 | \ 312 1ULL << VIRTIO_NET_F_MRG_RXBUF | \ 313 1ULL << VIRTIO_RING_F_INDIRECT_DESC | \ 314 1ULL << VIRTIO_NET_F_GUEST_CSUM | \ 315 1ULL << VIRTIO_NET_F_GUEST_TSO4 | \ 316 1ULL << VIRTIO_NET_F_GUEST_TSO6 | \ 317 1ULL << VIRTIO_F_VERSION_1) 318 319 int 320 virtio_user_dev_init(struct virtio_user_dev *dev, char *path, int queues, 321 int cq, int queue_size, const char *mac, char **ifname) 322 { 323 snprintf(dev->path, PATH_MAX, "%s", path); 324 dev->max_queue_pairs = queues; 325 dev->queue_pairs = 1; /* mq disabled by default */ 326 dev->queue_size = queue_size; 327 dev->mac_specified = 0; 328 parse_mac(dev, mac); 329 330 if (*ifname) { 331 dev->ifname = *ifname; 332 *ifname = NULL; 333 } 334 335 if (virtio_user_dev_setup(dev) < 0) { 336 PMD_INIT_LOG(ERR, "backend set up fails"); 337 return -1; 338 } 339 340 if (dev->ops->send_request(dev, VHOST_USER_SET_OWNER, NULL) < 0) { 341 PMD_INIT_LOG(ERR, "set_owner fails: %s", strerror(errno)); 342 return -1; 343 } 344 345 if (dev->ops->send_request(dev, VHOST_USER_GET_FEATURES, 346 &dev->device_features) < 0) { 347 PMD_INIT_LOG(ERR, "get_features failed: %s", strerror(errno)); 348 return -1; 349 } 350 if (dev->mac_specified) 351 dev->device_features |= (1ull << VIRTIO_NET_F_MAC); 352 353 if (cq) { 354 /* device does not really need to know anything about CQ, 355 * so if necessary, we just claim to support CQ 356 */ 357 dev->device_features |= (1ull << VIRTIO_NET_F_CTRL_VQ); 358 } else { 359 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_VQ); 360 /* Also disable features depends on VIRTIO_NET_F_CTRL_VQ */ 361 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_RX); 362 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_VLAN); 363 dev->device_features &= ~(1ull << VIRTIO_NET_F_GUEST_ANNOUNCE); 364 dev->device_features &= ~(1ull << VIRTIO_NET_F_MQ); 365 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_MAC_ADDR); 366 } 367 368 /* The backend will not report this feature, we add it explicitly */ 369 if (is_vhost_user_by_type(dev->path)) 370 dev->device_features |= (1ull << VIRTIO_NET_F_STATUS); 371 372 dev->device_features &= VIRTIO_USER_SUPPORTED_FEATURES; 373 374 return 0; 375 } 376 377 void 378 virtio_user_dev_uninit(struct virtio_user_dev *dev) 379 { 380 uint32_t i; 381 382 virtio_user_stop_device(dev); 383 384 for (i = 0; i < dev->max_queue_pairs * 2; ++i) { 385 close(dev->callfds[i]); 386 close(dev->kickfds[i]); 387 } 388 389 close(dev->vhostfd); 390 391 if (dev->vhostfds) { 392 for (i = 0; i < dev->max_queue_pairs; ++i) 393 close(dev->vhostfds[i]); 394 free(dev->vhostfds); 395 free(dev->tapfds); 396 } 397 398 free(dev->ifname); 399 } 400 401 static uint8_t 402 virtio_user_handle_mq(struct virtio_user_dev *dev, uint16_t q_pairs) 403 { 404 uint16_t i; 405 uint8_t ret = 0; 406 407 if (q_pairs > dev->max_queue_pairs) { 408 PMD_INIT_LOG(ERR, "multi-q config %u, but only %u supported", 409 q_pairs, dev->max_queue_pairs); 410 return -1; 411 } 412 413 for (i = 0; i < q_pairs; ++i) 414 ret |= dev->ops->enable_qp(dev, i, 1); 415 for (i = q_pairs; i < dev->max_queue_pairs; ++i) 416 ret |= dev->ops->enable_qp(dev, i, 0); 417 418 dev->queue_pairs = q_pairs; 419 420 return ret; 421 } 422 423 static uint32_t 424 virtio_user_handle_ctrl_msg(struct virtio_user_dev *dev, struct vring *vring, 425 uint16_t idx_hdr) 426 { 427 struct virtio_net_ctrl_hdr *hdr; 428 virtio_net_ctrl_ack status = ~0; 429 uint16_t i, idx_data, idx_status; 430 uint32_t n_descs = 0; 431 432 /* locate desc for header, data, and status */ 433 idx_data = vring->desc[idx_hdr].next; 434 n_descs++; 435 436 i = idx_data; 437 while (vring->desc[i].flags == VRING_DESC_F_NEXT) { 438 i = vring->desc[i].next; 439 n_descs++; 440 } 441 442 /* locate desc for status */ 443 idx_status = i; 444 n_descs++; 445 446 hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr; 447 if (hdr->class == VIRTIO_NET_CTRL_MQ && 448 hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) { 449 uint16_t queues; 450 451 queues = *(uint16_t *)(uintptr_t)vring->desc[idx_data].addr; 452 status = virtio_user_handle_mq(dev, queues); 453 } 454 455 /* Update status */ 456 *(virtio_net_ctrl_ack *)(uintptr_t)vring->desc[idx_status].addr = status; 457 458 return n_descs; 459 } 460 461 void 462 virtio_user_handle_cq(struct virtio_user_dev *dev, uint16_t queue_idx) 463 { 464 uint16_t avail_idx, desc_idx; 465 struct vring_used_elem *uep; 466 uint32_t n_descs; 467 struct vring *vring = &dev->vrings[queue_idx]; 468 469 /* Consume avail ring, using used ring idx as first one */ 470 while (vring->used->idx != vring->avail->idx) { 471 avail_idx = (vring->used->idx) & (vring->num - 1); 472 desc_idx = vring->avail->ring[avail_idx]; 473 474 n_descs = virtio_user_handle_ctrl_msg(dev, vring, desc_idx); 475 476 /* Update used ring */ 477 uep = &vring->used->ring[avail_idx]; 478 uep->id = avail_idx; 479 uep->len = n_descs; 480 481 vring->used->idx++; 482 } 483 } 484