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 45 #include "vhost.h" 46 #include "virtio_user_dev.h" 47 #include "../virtio_ethdev.h" 48 49 static int 50 virtio_user_create_queue(struct virtio_user_dev *dev, uint32_t queue_sel) 51 { 52 /* Of all per virtqueue MSGs, make sure VHOST_SET_VRING_CALL come 53 * firstly because vhost depends on this msg to allocate virtqueue 54 * pair. 55 */ 56 struct vhost_vring_file file; 57 58 file.index = queue_sel; 59 file.fd = dev->callfds[queue_sel]; 60 dev->ops->send_request(dev, VHOST_USER_SET_VRING_CALL, &file); 61 62 return 0; 63 } 64 65 static int 66 virtio_user_kick_queue(struct virtio_user_dev *dev, uint32_t queue_sel) 67 { 68 struct vhost_vring_file file; 69 struct vhost_vring_state state; 70 struct vring *vring = &dev->vrings[queue_sel]; 71 struct vhost_vring_addr addr = { 72 .index = queue_sel, 73 .desc_user_addr = (uint64_t)(uintptr_t)vring->desc, 74 .avail_user_addr = (uint64_t)(uintptr_t)vring->avail, 75 .used_user_addr = (uint64_t)(uintptr_t)vring->used, 76 .log_guest_addr = 0, 77 .flags = 0, /* disable log */ 78 }; 79 80 state.index = queue_sel; 81 state.num = vring->num; 82 dev->ops->send_request(dev, VHOST_USER_SET_VRING_NUM, &state); 83 84 state.index = queue_sel; 85 state.num = 0; /* no reservation */ 86 dev->ops->send_request(dev, VHOST_USER_SET_VRING_BASE, &state); 87 88 dev->ops->send_request(dev, VHOST_USER_SET_VRING_ADDR, &addr); 89 90 /* Of all per virtqueue MSGs, make sure VHOST_USER_SET_VRING_KICK comes 91 * lastly because vhost depends on this msg to judge if 92 * virtio is ready. 93 */ 94 file.index = queue_sel; 95 file.fd = dev->kickfds[queue_sel]; 96 dev->ops->send_request(dev, VHOST_USER_SET_VRING_KICK, &file); 97 98 return 0; 99 } 100 101 static int 102 virtio_user_queue_setup(struct virtio_user_dev *dev, 103 int (*fn)(struct virtio_user_dev *, uint32_t)) 104 { 105 uint32_t i, queue_sel; 106 107 for (i = 0; i < dev->max_queue_pairs; ++i) { 108 queue_sel = 2 * i + VTNET_SQ_RQ_QUEUE_IDX; 109 if (fn(dev, queue_sel) < 0) { 110 PMD_DRV_LOG(INFO, "setup rx vq fails: %u", i); 111 return -1; 112 } 113 } 114 for (i = 0; i < dev->max_queue_pairs; ++i) { 115 queue_sel = 2 * i + VTNET_SQ_TQ_QUEUE_IDX; 116 if (fn(dev, queue_sel) < 0) { 117 PMD_DRV_LOG(INFO, "setup tx vq fails: %u", i); 118 return -1; 119 } 120 } 121 122 return 0; 123 } 124 125 int 126 virtio_user_start_device(struct virtio_user_dev *dev) 127 { 128 uint64_t features; 129 int ret; 130 131 /* Step 0: tell vhost to create queues */ 132 if (virtio_user_queue_setup(dev, virtio_user_create_queue) < 0) 133 goto error; 134 135 /* Step 1: set features */ 136 features = dev->features; 137 /* Strip VIRTIO_NET_F_MAC, as MAC address is handled in vdev init */ 138 features &= ~(1ull << VIRTIO_NET_F_MAC); 139 /* Strip VIRTIO_NET_F_CTRL_VQ, as devices do not really need to know */ 140 features &= ~(1ull << VIRTIO_NET_F_CTRL_VQ); 141 features &= ~(1ull << VIRTIO_NET_F_STATUS); 142 ret = dev->ops->send_request(dev, VHOST_USER_SET_FEATURES, &features); 143 if (ret < 0) 144 goto error; 145 PMD_DRV_LOG(INFO, "set features: %" PRIx64, features); 146 147 /* Step 2: share memory regions */ 148 ret = dev->ops->send_request(dev, VHOST_USER_SET_MEM_TABLE, NULL); 149 if (ret < 0) 150 goto error; 151 152 /* Step 3: kick queues */ 153 if (virtio_user_queue_setup(dev, virtio_user_kick_queue) < 0) 154 goto error; 155 156 /* Step 4: enable queues 157 * we enable the 1st queue pair by default. 158 */ 159 dev->ops->enable_qp(dev, 0, 1); 160 161 return 0; 162 error: 163 /* TODO: free resource here or caller to check */ 164 return -1; 165 } 166 167 int virtio_user_stop_device(struct virtio_user_dev *dev) 168 { 169 uint32_t i; 170 171 for (i = 0; i < dev->max_queue_pairs; ++i) 172 dev->ops->enable_qp(dev, i, 0); 173 174 return 0; 175 } 176 177 static inline void 178 parse_mac(struct virtio_user_dev *dev, const char *mac) 179 { 180 int i, r; 181 uint32_t tmp[ETHER_ADDR_LEN]; 182 183 if (!mac) 184 return; 185 186 r = sscanf(mac, "%x:%x:%x:%x:%x:%x", &tmp[0], 187 &tmp[1], &tmp[2], &tmp[3], &tmp[4], &tmp[5]); 188 if (r == ETHER_ADDR_LEN) { 189 for (i = 0; i < ETHER_ADDR_LEN; ++i) 190 dev->mac_addr[i] = (uint8_t)tmp[i]; 191 dev->mac_specified = 1; 192 } else { 193 /* ignore the wrong mac, use random mac */ 194 PMD_DRV_LOG(ERR, "wrong format of mac: %s", mac); 195 } 196 } 197 198 int 199 is_vhost_user_by_type(const char *path) 200 { 201 struct stat sb; 202 203 if (stat(path, &sb) == -1) 204 return 0; 205 206 return S_ISSOCK(sb.st_mode); 207 } 208 209 static int 210 virtio_user_dev_init_notify(struct virtio_user_dev *dev) 211 { 212 uint32_t i, j; 213 int callfd; 214 int kickfd; 215 216 for (i = 0; i < VIRTIO_MAX_VIRTQUEUES; ++i) { 217 if (i >= dev->max_queue_pairs * 2) { 218 dev->kickfds[i] = -1; 219 dev->callfds[i] = -1; 220 continue; 221 } 222 223 /* May use invalid flag, but some backend uses kickfd and 224 * callfd as criteria to judge if dev is alive. so finally we 225 * use real event_fd. 226 */ 227 callfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK); 228 if (callfd < 0) { 229 PMD_DRV_LOG(ERR, "callfd error, %s", strerror(errno)); 230 break; 231 } 232 kickfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK); 233 if (kickfd < 0) { 234 PMD_DRV_LOG(ERR, "kickfd error, %s", strerror(errno)); 235 break; 236 } 237 dev->callfds[i] = callfd; 238 dev->kickfds[i] = kickfd; 239 } 240 241 if (i < VIRTIO_MAX_VIRTQUEUES) { 242 for (j = 0; j <= i; ++j) { 243 close(dev->callfds[j]); 244 close(dev->kickfds[j]); 245 } 246 247 return -1; 248 } 249 250 return 0; 251 } 252 253 static int 254 virtio_user_fill_intr_handle(struct virtio_user_dev *dev) 255 { 256 uint32_t i; 257 struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->port_id]; 258 259 if (!eth_dev->intr_handle) { 260 eth_dev->intr_handle = malloc(sizeof(*eth_dev->intr_handle)); 261 if (!eth_dev->intr_handle) { 262 PMD_DRV_LOG(ERR, "fail to allocate intr_handle"); 263 return -1; 264 } 265 memset(eth_dev->intr_handle, 0, sizeof(*eth_dev->intr_handle)); 266 } 267 268 for (i = 0; i < dev->max_queue_pairs; ++i) 269 eth_dev->intr_handle->efds[i] = dev->callfds[i]; 270 eth_dev->intr_handle->nb_efd = dev->max_queue_pairs; 271 eth_dev->intr_handle->max_intr = dev->max_queue_pairs + 1; 272 eth_dev->intr_handle->type = RTE_INTR_HANDLE_VDEV; 273 /* For virtio vdev, no need to read counter for clean */ 274 eth_dev->intr_handle->efd_counter_size = 0; 275 if (dev->vhostfd >= 0) 276 eth_dev->intr_handle->fd = dev->vhostfd; 277 278 return 0; 279 } 280 281 static int 282 virtio_user_dev_setup(struct virtio_user_dev *dev) 283 { 284 uint32_t q; 285 286 dev->vhostfd = -1; 287 dev->vhostfds = NULL; 288 dev->tapfds = NULL; 289 290 if (is_vhost_user_by_type(dev->path)) { 291 dev->ops = &ops_user; 292 } else { 293 dev->ops = &ops_kernel; 294 295 dev->vhostfds = malloc(dev->max_queue_pairs * sizeof(int)); 296 dev->tapfds = malloc(dev->max_queue_pairs * sizeof(int)); 297 if (!dev->vhostfds || !dev->tapfds) { 298 PMD_INIT_LOG(ERR, "Failed to malloc"); 299 return -1; 300 } 301 302 for (q = 0; q < dev->max_queue_pairs; ++q) { 303 dev->vhostfds[q] = -1; 304 dev->tapfds[q] = -1; 305 } 306 } 307 308 if (dev->ops->setup(dev) < 0) 309 return -1; 310 311 if (virtio_user_dev_init_notify(dev) < 0) 312 return -1; 313 314 if (virtio_user_fill_intr_handle(dev) < 0) 315 return -1; 316 317 return 0; 318 } 319 320 /* Use below macro to filter features from vhost backend */ 321 #define VIRTIO_USER_SUPPORTED_FEATURES \ 322 (1ULL << VIRTIO_NET_F_MAC | \ 323 1ULL << VIRTIO_NET_F_STATUS | \ 324 1ULL << VIRTIO_NET_F_MQ | \ 325 1ULL << VIRTIO_NET_F_CTRL_MAC_ADDR | \ 326 1ULL << VIRTIO_NET_F_CTRL_VQ | \ 327 1ULL << VIRTIO_NET_F_CTRL_RX | \ 328 1ULL << VIRTIO_NET_F_CTRL_VLAN | \ 329 1ULL << VIRTIO_NET_F_CSUM | \ 330 1ULL << VIRTIO_NET_F_HOST_TSO4 | \ 331 1ULL << VIRTIO_NET_F_HOST_TSO6 | \ 332 1ULL << VIRTIO_NET_F_MRG_RXBUF | \ 333 1ULL << VIRTIO_RING_F_INDIRECT_DESC | \ 334 1ULL << VIRTIO_NET_F_GUEST_CSUM | \ 335 1ULL << VIRTIO_NET_F_GUEST_TSO4 | \ 336 1ULL << VIRTIO_NET_F_GUEST_TSO6 | \ 337 1ULL << VIRTIO_F_VERSION_1) 338 339 int 340 virtio_user_dev_init(struct virtio_user_dev *dev, char *path, int queues, 341 int cq, int queue_size, const char *mac, char **ifname) 342 { 343 snprintf(dev->path, PATH_MAX, "%s", path); 344 dev->max_queue_pairs = queues; 345 dev->queue_pairs = 1; /* mq disabled by default */ 346 dev->queue_size = queue_size; 347 dev->mac_specified = 0; 348 parse_mac(dev, mac); 349 350 if (*ifname) { 351 dev->ifname = *ifname; 352 *ifname = NULL; 353 } 354 355 if (virtio_user_dev_setup(dev) < 0) { 356 PMD_INIT_LOG(ERR, "backend set up fails"); 357 return -1; 358 } 359 if (dev->ops->send_request(dev, VHOST_USER_SET_OWNER, NULL) < 0) { 360 PMD_INIT_LOG(ERR, "set_owner fails: %s", strerror(errno)); 361 return -1; 362 } 363 364 if (dev->ops->send_request(dev, VHOST_USER_GET_FEATURES, 365 &dev->device_features) < 0) { 366 PMD_INIT_LOG(ERR, "get_features failed: %s", strerror(errno)); 367 return -1; 368 } 369 if (dev->mac_specified) 370 dev->device_features |= (1ull << VIRTIO_NET_F_MAC); 371 372 if (cq) { 373 /* device does not really need to know anything about CQ, 374 * so if necessary, we just claim to support CQ 375 */ 376 dev->device_features |= (1ull << VIRTIO_NET_F_CTRL_VQ); 377 } else { 378 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_VQ); 379 /* Also disable features depends on VIRTIO_NET_F_CTRL_VQ */ 380 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_RX); 381 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_VLAN); 382 dev->device_features &= ~(1ull << VIRTIO_NET_F_GUEST_ANNOUNCE); 383 dev->device_features &= ~(1ull << VIRTIO_NET_F_MQ); 384 dev->device_features &= ~(1ull << VIRTIO_NET_F_CTRL_MAC_ADDR); 385 } 386 387 /* The backend will not report this feature, we add it explicitly */ 388 if (is_vhost_user_by_type(dev->path)) 389 dev->device_features |= (1ull << VIRTIO_NET_F_STATUS); 390 391 dev->device_features &= VIRTIO_USER_SUPPORTED_FEATURES; 392 393 return 0; 394 } 395 396 void 397 virtio_user_dev_uninit(struct virtio_user_dev *dev) 398 { 399 uint32_t i; 400 401 virtio_user_stop_device(dev); 402 403 for (i = 0; i < dev->max_queue_pairs * 2; ++i) { 404 close(dev->callfds[i]); 405 close(dev->kickfds[i]); 406 } 407 408 close(dev->vhostfd); 409 410 if (dev->vhostfds) { 411 for (i = 0; i < dev->max_queue_pairs; ++i) 412 close(dev->vhostfds[i]); 413 free(dev->vhostfds); 414 free(dev->tapfds); 415 } 416 417 free(dev->ifname); 418 } 419 420 static uint8_t 421 virtio_user_handle_mq(struct virtio_user_dev *dev, uint16_t q_pairs) 422 { 423 uint16_t i; 424 uint8_t ret = 0; 425 426 if (q_pairs > dev->max_queue_pairs) { 427 PMD_INIT_LOG(ERR, "multi-q config %u, but only %u supported", 428 q_pairs, dev->max_queue_pairs); 429 return -1; 430 } 431 432 for (i = 0; i < q_pairs; ++i) 433 ret |= dev->ops->enable_qp(dev, i, 1); 434 for (i = q_pairs; i < dev->max_queue_pairs; ++i) 435 ret |= dev->ops->enable_qp(dev, i, 0); 436 437 dev->queue_pairs = q_pairs; 438 439 return ret; 440 } 441 442 static uint32_t 443 virtio_user_handle_ctrl_msg(struct virtio_user_dev *dev, struct vring *vring, 444 uint16_t idx_hdr) 445 { 446 struct virtio_net_ctrl_hdr *hdr; 447 virtio_net_ctrl_ack status = ~0; 448 uint16_t i, idx_data, idx_status; 449 uint32_t n_descs = 0; 450 451 /* locate desc for header, data, and status */ 452 idx_data = vring->desc[idx_hdr].next; 453 n_descs++; 454 455 i = idx_data; 456 while (vring->desc[i].flags == VRING_DESC_F_NEXT) { 457 i = vring->desc[i].next; 458 n_descs++; 459 } 460 461 /* locate desc for status */ 462 idx_status = i; 463 n_descs++; 464 465 hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr; 466 if (hdr->class == VIRTIO_NET_CTRL_MQ && 467 hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) { 468 uint16_t queues; 469 470 queues = *(uint16_t *)(uintptr_t)vring->desc[idx_data].addr; 471 status = virtio_user_handle_mq(dev, queues); 472 } 473 474 /* Update status */ 475 *(virtio_net_ctrl_ack *)(uintptr_t)vring->desc[idx_status].addr = status; 476 477 return n_descs; 478 } 479 480 void 481 virtio_user_handle_cq(struct virtio_user_dev *dev, uint16_t queue_idx) 482 { 483 uint16_t avail_idx, desc_idx; 484 struct vring_used_elem *uep; 485 uint32_t n_descs; 486 struct vring *vring = &dev->vrings[queue_idx]; 487 488 /* Consume avail ring, using used ring idx as first one */ 489 while (vring->used->idx != vring->avail->idx) { 490 avail_idx = (vring->used->idx) & (vring->num - 1); 491 desc_idx = vring->avail->ring[avail_idx]; 492 493 n_descs = virtio_user_handle_ctrl_msg(dev, vring, desc_idx); 494 495 /* Update used ring */ 496 uep = &vring->used->ring[avail_idx]; 497 uep->id = avail_idx; 498 uep->len = n_descs; 499 500 vring->used->idx++; 501 } 502 } 503