1 /*- 2 * BSD LICENSE 3 * 4 * Copyright(c) 2010-2016 Intel Corporation. All rights reserved. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of Intel Corporation nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include <stdint.h> 35 #include <stdio.h> 36 #include <fcntl.h> 37 #include <string.h> 38 #include <errno.h> 39 #include <sys/mman.h> 40 #include <unistd.h> 41 #include <sys/eventfd.h> 42 #include <sys/types.h> 43 #include <sys/stat.h> 44 #include <unistd.h> 45 46 #include "vhost.h" 47 #include "virtio_user_dev.h" 48 #include "../virtio_ethdev.h" 49 50 static int 51 virtio_user_create_queue(struct virtio_user_dev *dev, uint32_t queue_sel) 52 { 53 /* Of all per virtqueue MSGs, make sure VHOST_SET_VRING_CALL come 54 * firstly because vhost depends on this msg to allocate virtqueue 55 * pair. 56 */ 57 struct vhost_vring_file file; 58 59 file.index = queue_sel; 60 file.fd = dev->callfds[queue_sel]; 61 dev->ops->send_request(dev, VHOST_USER_SET_VRING_CALL, &file); 62 63 return 0; 64 } 65 66 static int 67 virtio_user_kick_queue(struct virtio_user_dev *dev, uint32_t queue_sel) 68 { 69 struct vhost_vring_file file; 70 struct vhost_vring_state state; 71 struct vring *vring = &dev->vrings[queue_sel]; 72 struct vhost_vring_addr addr = { 73 .index = queue_sel, 74 .desc_user_addr = (uint64_t)(uintptr_t)vring->desc, 75 .avail_user_addr = (uint64_t)(uintptr_t)vring->avail, 76 .used_user_addr = (uint64_t)(uintptr_t)vring->used, 77 .log_guest_addr = 0, 78 .flags = 0, /* disable log */ 79 }; 80 81 state.index = queue_sel; 82 state.num = vring->num; 83 dev->ops->send_request(dev, VHOST_USER_SET_VRING_NUM, &state); 84 85 state.index = queue_sel; 86 state.num = 0; /* no reservation */ 87 dev->ops->send_request(dev, VHOST_USER_SET_VRING_BASE, &state); 88 89 dev->ops->send_request(dev, VHOST_USER_SET_VRING_ADDR, &addr); 90 91 /* Of all per virtqueue MSGs, make sure VHOST_USER_SET_VRING_KICK comes 92 * lastly because vhost depends on this msg to judge if 93 * virtio is ready. 94 */ 95 file.index = queue_sel; 96 file.fd = dev->kickfds[queue_sel]; 97 dev->ops->send_request(dev, VHOST_USER_SET_VRING_KICK, &file); 98 99 return 0; 100 } 101 102 static int 103 virtio_user_queue_setup(struct virtio_user_dev *dev, 104 int (*fn)(struct virtio_user_dev *, uint32_t)) 105 { 106 uint32_t i, queue_sel; 107 108 for (i = 0; i < dev->max_queue_pairs; ++i) { 109 queue_sel = 2 * i + VTNET_SQ_RQ_QUEUE_IDX; 110 if (fn(dev, queue_sel) < 0) { 111 PMD_DRV_LOG(INFO, "setup rx vq fails: %u", i); 112 return -1; 113 } 114 } 115 for (i = 0; i < dev->max_queue_pairs; ++i) { 116 queue_sel = 2 * i + VTNET_SQ_TQ_QUEUE_IDX; 117 if (fn(dev, queue_sel) < 0) { 118 PMD_DRV_LOG(INFO, "setup tx vq fails: %u", i); 119 return -1; 120 } 121 } 122 123 return 0; 124 } 125 126 int 127 virtio_user_start_device(struct virtio_user_dev *dev) 128 { 129 uint64_t features; 130 int ret; 131 132 /* Step 0: tell vhost to create queues */ 133 if (virtio_user_queue_setup(dev, virtio_user_create_queue) < 0) 134 goto error; 135 136 /* Step 1: set features */ 137 features = dev->features; 138 /* Strip VIRTIO_NET_F_MAC, as MAC address is handled in vdev init */ 139 features &= ~(1ull << VIRTIO_NET_F_MAC); 140 /* Strip VIRTIO_NET_F_CTRL_VQ, as devices do not really need to know */ 141 features &= ~(1ull << VIRTIO_NET_F_CTRL_VQ); 142 features &= ~(1ull << VIRTIO_NET_F_STATUS); 143 ret = dev->ops->send_request(dev, VHOST_USER_SET_FEATURES, &features); 144 if (ret < 0) 145 goto error; 146 PMD_DRV_LOG(INFO, "set features: %" PRIx64, features); 147 148 /* Step 2: share memory regions */ 149 ret = dev->ops->send_request(dev, VHOST_USER_SET_MEM_TABLE, NULL); 150 if (ret < 0) 151 goto error; 152 153 /* Step 3: kick queues */ 154 if (virtio_user_queue_setup(dev, virtio_user_kick_queue) < 0) 155 goto error; 156 157 /* Step 4: enable queues 158 * we enable the 1st queue pair by default. 159 */ 160 dev->ops->enable_qp(dev, 0, 1); 161 162 return 0; 163 error: 164 /* TODO: free resource here or caller to check */ 165 return -1; 166 } 167 168 int virtio_user_stop_device(struct virtio_user_dev *dev) 169 { 170 uint32_t i; 171 172 for (i = 0; i < dev->max_queue_pairs; ++i) 173 dev->ops->enable_qp(dev, i, 0); 174 175 return 0; 176 } 177 178 static inline void 179 parse_mac(struct virtio_user_dev *dev, const char *mac) 180 { 181 int i, r; 182 uint32_t tmp[ETHER_ADDR_LEN]; 183 184 if (!mac) 185 return; 186 187 r = sscanf(mac, "%x:%x:%x:%x:%x:%x", &tmp[0], 188 &tmp[1], &tmp[2], &tmp[3], &tmp[4], &tmp[5]); 189 if (r == ETHER_ADDR_LEN) { 190 for (i = 0; i < ETHER_ADDR_LEN; ++i) 191 dev->mac_addr[i] = (uint8_t)tmp[i]; 192 dev->mac_specified = 1; 193 } else { 194 /* ignore the wrong mac, use random mac */ 195 PMD_DRV_LOG(ERR, "wrong format of mac: %s", mac); 196 } 197 } 198 199 int 200 is_vhost_user_by_type(const char *path) 201 { 202 struct stat sb; 203 204 if (stat(path, &sb) == -1) 205 return 0; 206 207 return S_ISSOCK(sb.st_mode); 208 } 209 210 static int 211 virtio_user_dev_init_notify(struct virtio_user_dev *dev) 212 { 213 uint32_t i, j; 214 int callfd; 215 int kickfd; 216 217 for (i = 0; i < VIRTIO_MAX_VIRTQUEUES; ++i) { 218 if (i >= dev->max_queue_pairs * 2) { 219 dev->kickfds[i] = -1; 220 dev->callfds[i] = -1; 221 continue; 222 } 223 224 /* May use invalid flag, but some backend uses kickfd and 225 * callfd as criteria to judge if dev is alive. so finally we 226 * use real event_fd. 227 */ 228 callfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK); 229 if (callfd < 0) { 230 PMD_DRV_LOG(ERR, "callfd error, %s", strerror(errno)); 231 break; 232 } 233 kickfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK); 234 if (kickfd < 0) { 235 PMD_DRV_LOG(ERR, "kickfd error, %s", strerror(errno)); 236 break; 237 } 238 dev->callfds[i] = callfd; 239 dev->kickfds[i] = kickfd; 240 } 241 242 if (i < VIRTIO_MAX_VIRTQUEUES) { 243 for (j = 0; j <= i; ++j) { 244 close(dev->callfds[j]); 245 close(dev->kickfds[j]); 246 } 247 248 return -1; 249 } 250 251 return 0; 252 } 253 254 static int 255 virtio_user_fill_intr_handle(struct virtio_user_dev *dev) 256 { 257 uint32_t i; 258 struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->port_id]; 259 260 if (!eth_dev->intr_handle) { 261 eth_dev->intr_handle = malloc(sizeof(*eth_dev->intr_handle)); 262 if (!eth_dev->intr_handle) { 263 PMD_DRV_LOG(ERR, "fail to allocate intr_handle"); 264 return -1; 265 } 266 memset(eth_dev->intr_handle, 0, sizeof(*eth_dev->intr_handle)); 267 } 268 269 for (i = 0; i < dev->max_queue_pairs; ++i) 270 eth_dev->intr_handle->efds[i] = dev->callfds[i]; 271 eth_dev->intr_handle->nb_efd = dev->max_queue_pairs; 272 eth_dev->intr_handle->max_intr = dev->max_queue_pairs + 1; 273 eth_dev->intr_handle->type = RTE_INTR_HANDLE_VDEV; 274 if (dev->vhostfd >= 0) 275 eth_dev->intr_handle->fd = dev->vhostfd; 276 277 return 0; 278 } 279 280 static int 281 virtio_user_dev_setup(struct virtio_user_dev *dev) 282 { 283 uint32_t q; 284 285 dev->vhostfd = -1; 286 dev->vhostfds = NULL; 287 dev->tapfds = NULL; 288 289 if (is_vhost_user_by_type(dev->path)) { 290 dev->ops = &ops_user; 291 } else { 292 dev->ops = &ops_kernel; 293 294 dev->vhostfds = malloc(dev->max_queue_pairs * sizeof(int)); 295 dev->tapfds = malloc(dev->max_queue_pairs * sizeof(int)); 296 if (!dev->vhostfds || !dev->tapfds) { 297 PMD_INIT_LOG(ERR, "Failed to malloc"); 298 return -1; 299 } 300 301 for (q = 0; q < dev->max_queue_pairs; ++q) { 302 dev->vhostfds[q] = -1; 303 dev->tapfds[q] = -1; 304 } 305 } 306 307 if (dev->ops->setup(dev) < 0) 308 return -1; 309 310 if (virtio_user_dev_init_notify(dev) < 0) 311 return -1; 312 313 if (virtio_user_fill_intr_handle(dev) < 0) 314 return -1; 315 316 return 0; 317 } 318 319 /* Use below macro to filter features from vhost backend */ 320 #define VIRTIO_USER_SUPPORTED_FEATURES \ 321 (1ULL << VIRTIO_NET_F_MAC | \ 322 1ULL << VIRTIO_NET_F_STATUS | \ 323 1ULL << VIRTIO_NET_F_MQ | \ 324 1ULL << VIRTIO_NET_F_CTRL_MAC_ADDR | \ 325 1ULL << VIRTIO_NET_F_CTRL_VQ | \ 326 1ULL << VIRTIO_NET_F_CTRL_RX | \ 327 1ULL << VIRTIO_NET_F_CTRL_VLAN | \ 328 1ULL << VIRTIO_NET_F_CSUM | \ 329 1ULL << VIRTIO_NET_F_HOST_TSO4 | \ 330 1ULL << VIRTIO_NET_F_HOST_TSO6 | \ 331 1ULL << VIRTIO_NET_F_MRG_RXBUF | \ 332 1ULL << VIRTIO_RING_F_INDIRECT_DESC | \ 333 1ULL << VIRTIO_NET_F_GUEST_CSUM | \ 334 1ULL << VIRTIO_NET_F_GUEST_TSO4 | \ 335 1ULL << VIRTIO_NET_F_GUEST_TSO6 | \ 336 1ULL << VIRTIO_F_VERSION_1) 337 338 int 339 virtio_user_dev_init(struct virtio_user_dev *dev, char *path, int queues, 340 int cq, int queue_size, const char *mac, char **ifname) 341 { 342 snprintf(dev->path, PATH_MAX, "%s", path); 343 dev->max_queue_pairs = queues; 344 dev->queue_pairs = 1; /* mq disabled by default */ 345 dev->queue_size = queue_size; 346 dev->mac_specified = 0; 347 parse_mac(dev, mac); 348 349 if (*ifname) { 350 dev->ifname = *ifname; 351 *ifname = NULL; 352 } 353 354 if (virtio_user_dev_setup(dev) < 0) { 355 PMD_INIT_LOG(ERR, "backend set up fails"); 356 return -1; 357 } 358 if (dev->ops->send_request(dev, VHOST_USER_SET_OWNER, NULL) < 0) { 359 PMD_INIT_LOG(ERR, "set_owner fails: %s", strerror(errno)); 360 return -1; 361 } 362 363 if (dev->ops->send_request(dev, VHOST_USER_GET_FEATURES, 364 &dev->device_features) < 0) { 365 PMD_INIT_LOG(ERR, "get_features failed: %s", strerror(errno)); 366 return -1; 367 } 368 if (dev->mac_specified) 369 dev->device_features |= (1ull << VIRTIO_NET_F_MAC); 370 371 if (cq) { 372 /* device does not really need to know anything about CQ, 373 * so if necessary, we just claim to support CQ 374 */ 375 dev->device_features |= (1ull << VIRTIO_NET_F_CTRL_VQ); 376 } else { 377 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_VQ); 378 /* Also disable features depends on VIRTIO_NET_F_CTRL_VQ */ 379 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_RX); 380 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_VLAN); 381 dev->device_features &= ~(1ull << VIRTIO_NET_F_GUEST_ANNOUNCE); 382 dev->device_features &= ~(1ull << VIRTIO_NET_F_MQ); 383 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_MAC_ADDR); 384 } 385 386 /* The backend will not report this feature, we add it explicitly */ 387 if (is_vhost_user_by_type(dev->path)) 388 dev->device_features |= (1ull << VIRTIO_NET_F_STATUS); 389 390 dev->device_features &= VIRTIO_USER_SUPPORTED_FEATURES; 391 392 return 0; 393 } 394 395 void 396 virtio_user_dev_uninit(struct virtio_user_dev *dev) 397 { 398 uint32_t i; 399 400 virtio_user_stop_device(dev); 401 402 for (i = 0; i < dev->max_queue_pairs * 2; ++i) { 403 close(dev->callfds[i]); 404 close(dev->kickfds[i]); 405 } 406 407 close(dev->vhostfd); 408 409 if (dev->vhostfds) { 410 for (i = 0; i < dev->max_queue_pairs; ++i) 411 close(dev->vhostfds[i]); 412 free(dev->vhostfds); 413 free(dev->tapfds); 414 } 415 416 free(dev->ifname); 417 } 418 419 static uint8_t 420 virtio_user_handle_mq(struct virtio_user_dev *dev, uint16_t q_pairs) 421 { 422 uint16_t i; 423 uint8_t ret = 0; 424 425 if (q_pairs > dev->max_queue_pairs) { 426 PMD_INIT_LOG(ERR, "multi-q config %u, but only %u supported", 427 q_pairs, dev->max_queue_pairs); 428 return -1; 429 } 430 431 for (i = 0; i < q_pairs; ++i) 432 ret |= dev->ops->enable_qp(dev, i, 1); 433 for (i = q_pairs; i < dev->max_queue_pairs; ++i) 434 ret |= dev->ops->enable_qp(dev, i, 0); 435 436 dev->queue_pairs = q_pairs; 437 438 return ret; 439 } 440 441 static uint32_t 442 virtio_user_handle_ctrl_msg(struct virtio_user_dev *dev, struct vring *vring, 443 uint16_t idx_hdr) 444 { 445 struct virtio_net_ctrl_hdr *hdr; 446 virtio_net_ctrl_ack status = ~0; 447 uint16_t i, idx_data, idx_status; 448 uint32_t n_descs = 0; 449 450 /* locate desc for header, data, and status */ 451 idx_data = vring->desc[idx_hdr].next; 452 n_descs++; 453 454 i = idx_data; 455 while (vring->desc[i].flags == VRING_DESC_F_NEXT) { 456 i = vring->desc[i].next; 457 n_descs++; 458 } 459 460 /* locate desc for status */ 461 idx_status = i; 462 n_descs++; 463 464 hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr; 465 if (hdr->class == VIRTIO_NET_CTRL_MQ && 466 hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) { 467 uint16_t queues; 468 469 queues = *(uint16_t *)(uintptr_t)vring->desc[idx_data].addr; 470 status = virtio_user_handle_mq(dev, queues); 471 } 472 473 /* Update status */ 474 *(virtio_net_ctrl_ack *)(uintptr_t)vring->desc[idx_status].addr = status; 475 476 return n_descs; 477 } 478 479 void 480 virtio_user_handle_cq(struct virtio_user_dev *dev, uint16_t queue_idx) 481 { 482 uint16_t avail_idx, desc_idx; 483 struct vring_used_elem *uep; 484 uint32_t n_descs; 485 struct vring *vring = &dev->vrings[queue_idx]; 486 487 /* Consume avail ring, using used ring idx as first one */ 488 while (vring->used->idx != vring->avail->idx) { 489 avail_idx = (vring->used->idx) & (vring->num - 1); 490 desc_idx = vring->avail->ring[avail_idx]; 491 492 n_descs = virtio_user_handle_ctrl_msg(dev, vring, desc_idx); 493 494 /* Update used ring */ 495 uep = &vring->used->ring[avail_idx]; 496 uep->id = avail_idx; 497 uep->len = n_descs; 498 499 vring->used->idx++; 500 } 501 } 502