drivers/net/virtio/virtio_user/virtio_user_dev.c

   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 <rte_string_fns.h>
  17 #include <rte_eal_memconfig.h>
  18
  19 #include "vhost.h"
  20 #include "virtio_user_dev.h"
  21 #include "../virtio_ethdev.h"
  22
  23 #define VIRTIO_USER_MEM_EVENT_CLB_NAME "virtio_user_mem_event_clb"
  24
  25 const char * const virtio_user_backend_strings[] = {
  26         [VIRTIO_USER_BACKEND_UNKNOWN] = "VIRTIO_USER_BACKEND_UNKNOWN",
  27         [VIRTIO_USER_BACKEND_VHOST_USER] = "VHOST_USER",
  28         [VIRTIO_USER_BACKEND_VHOST_KERNEL] = "VHOST_NET",
  29         [VIRTIO_USER_BACKEND_VHOST_VDPA] = "VHOST_VDPA",
  30 };
  31
  32 static int
  33 virtio_user_create_queue(struct virtio_user_dev *dev, uint32_t queue_sel)
  34 {
  35         /* Of all per virtqueue MSGs, make sure VHOST_SET_VRING_CALL come
  36          * firstly because vhost depends on this msg to allocate virtqueue
  37          * pair.
  38          */
  39         struct vhost_vring_file file;
  40         int ret;
  41
  42         file.index = queue_sel;
  43         file.fd = dev->callfds[queue_sel];
  44         ret = dev->ops->set_vring_call(dev, &file);
  45         if (ret < 0) {
  46                 PMD_INIT_LOG(ERR, "(%s) Failed to create queue %u\n", dev->path, queue_sel);
  47                 return -1;
  48         }
  49
  50         return 0;
  51 }
  52
  53 static int
  54 virtio_user_kick_queue(struct virtio_user_dev *dev, uint32_t queue_sel)
  55 {
  56         int ret;
  57         struct vhost_vring_file file;
  58         struct vhost_vring_state state;
  59         struct vring *vring = &dev->vrings[queue_sel];
  60         struct vring_packed *pq_vring = &dev->packed_vrings[queue_sel];
  61         struct vhost_vring_addr addr = {
  62                 .index = queue_sel,
  63                 .log_guest_addr = 0,
  64                 .flags = 0, /* disable log */
  65         };
  66
  67         if (dev->features & (1ULL << VIRTIO_F_RING_PACKED)) {
  68                 addr.desc_user_addr =
  69                         (uint64_t)(uintptr_t)pq_vring->desc;
  70                 addr.avail_user_addr =
  71                         (uint64_t)(uintptr_t)pq_vring->driver;
  72                 addr.used_user_addr =
  73                         (uint64_t)(uintptr_t)pq_vring->device;
  74         } else {
  75                 addr.desc_user_addr = (uint64_t)(uintptr_t)vring->desc;
  76                 addr.avail_user_addr = (uint64_t)(uintptr_t)vring->avail;
  77                 addr.used_user_addr = (uint64_t)(uintptr_t)vring->used;
  78         }
  79
  80         state.index = queue_sel;
  81         state.num = vring->num;
  82         ret = dev->ops->set_vring_num(dev, &state);
  83         if (ret < 0)
  84                 goto err;
  85
  86         state.index = queue_sel;
  87         state.num = 0; /* no reservation */
  88         if (dev->features & (1ULL << VIRTIO_F_RING_PACKED))
  89                 state.num |= (1 << 15);
  90         ret = dev->ops->set_vring_base(dev, &state);
  91         if (ret < 0)
  92                 goto err;
  93
  94         ret = dev->ops->set_vring_addr(dev, &addr);
  95         if (ret < 0)
  96                 goto err;
  97
  98         /* Of all per virtqueue MSGs, make sure VHOST_USER_SET_VRING_KICK comes
  99          * lastly because vhost depends on this msg to judge if
 100          * virtio is ready.
 101          */
 102         file.index = queue_sel;
 103         file.fd = dev->kickfds[queue_sel];
 104         ret = dev->ops->set_vring_kick(dev, &file);
 105         if (ret < 0)
 106                 goto err;
 107
 108         return 0;
 109 err:
 110         PMD_INIT_LOG(ERR, "(%s) Failed to kick queue %u\n", dev->path, queue_sel);
 111
 112         return -1;
 113 }
 114
 115 static int
 116 virtio_user_queue_setup(struct virtio_user_dev *dev,
 117                         int (*fn)(struct virtio_user_dev *, uint32_t))
 118 {
 119         uint32_t i, queue_sel;
 120
 121         for (i = 0; i < dev->max_queue_pairs; ++i) {
 122                 queue_sel = 2 * i + VTNET_SQ_RQ_QUEUE_IDX;
 123                 if (fn(dev, queue_sel) < 0) {
 124                         PMD_DRV_LOG(ERR, "(%s) setup rx vq %u failed", dev->path, i);
 125                         return -1;
 126                 }
 127         }
 128         for (i = 0; i < dev->max_queue_pairs; ++i) {
 129                 queue_sel = 2 * i + VTNET_SQ_TQ_QUEUE_IDX;
 130                 if (fn(dev, queue_sel) < 0) {
 131                         PMD_DRV_LOG(INFO, "(%s) setup tx vq %u failed", dev->path, i);
 132                         return -1;
 133                 }
 134         }
 135
 136         return 0;
 137 }
 138
 139 int
 140 virtio_user_dev_set_features(struct virtio_user_dev *dev)
 141 {
 142         uint64_t features;
 143         int ret = -1;
 144
 145         pthread_mutex_lock(&dev->mutex);
 146
 147         if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER &&
 148                         dev->vhostfd < 0)
 149                 goto error;
 150
 151         /* Step 0: tell vhost to create queues */
 152         if (virtio_user_queue_setup(dev, virtio_user_create_queue) < 0)
 153                 goto error;
 154
 155         features = dev->features;
 156
 157         /* Strip VIRTIO_NET_F_MAC, as MAC address is handled in vdev init */
 158         features &= ~(1ull << VIRTIO_NET_F_MAC);
 159         /* Strip VIRTIO_NET_F_CTRL_VQ, as devices do not really need to know */
 160         features &= ~(1ull << VIRTIO_NET_F_CTRL_VQ);
 161         features &= ~(1ull << VIRTIO_NET_F_STATUS);
 162         ret = dev->ops->set_features(dev, features);
 163         if (ret < 0)
 164                 goto error;
 165         PMD_DRV_LOG(INFO, "(%s) set features: 0x%" PRIx64, dev->path, features);
 166 error:
 167         pthread_mutex_unlock(&dev->mutex);
 168
 169         return ret;
 170 }
 171
 172 int
 173 virtio_user_start_device(struct virtio_user_dev *dev)
 174 {
 175         int ret;
 176
 177         /*
 178          * XXX workaround!
 179          *
 180          * We need to make sure that the locks will be
 181          * taken in the correct order to avoid deadlocks.
 182          *
 183          * Before releasing this lock, this thread should
 184          * not trigger any memory hotplug events.
 185          *
 186          * This is a temporary workaround, and should be
 187          * replaced when we get proper supports from the
 188          * memory subsystem in the future.
 189          */
 190         rte_mcfg_mem_read_lock();
 191         pthread_mutex_lock(&dev->mutex);
 192
 193         /* Vhost-user client not connected yet, will start later */
 194         if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER &&
 195                         dev->vhostfd < 0)
 196                 goto out;
 197
 198         /* Step 2: share memory regions */
 199         ret = dev->ops->set_memory_table(dev);
 200         if (ret < 0)
 201                 goto error;
 202
 203         /* Step 3: kick queues */
 204         ret = virtio_user_queue_setup(dev, virtio_user_kick_queue);
 205         if (ret < 0)
 206                 goto error;
 207
 208         /* Step 4: enable queues
 209          * we enable the 1st queue pair by default.
 210          */
 211         ret = dev->ops->enable_qp(dev, 0, 1);
 212         if (ret < 0)
 213                 goto error;
 214
 215         dev->started = true;
 216 out:
 217         pthread_mutex_unlock(&dev->mutex);
 218         rte_mcfg_mem_read_unlock();
 219
 220         return 0;
 221 error:
 222         pthread_mutex_unlock(&dev->mutex);
 223         rte_mcfg_mem_read_unlock();
 224
 225         PMD_INIT_LOG(ERR, "(%s) Failed to start device\n", dev->path);
 226
 227         /* TODO: free resource here or caller to check */
 228         return -1;
 229 }
 230
 231 int virtio_user_stop_device(struct virtio_user_dev *dev)
 232 {
 233         struct vhost_vring_state state;
 234         uint32_t i;
 235         int ret;
 236
 237         pthread_mutex_lock(&dev->mutex);
 238         if (!dev->started)
 239                 goto out;
 240
 241         for (i = 0; i < dev->max_queue_pairs; ++i) {
 242                 ret = dev->ops->enable_qp(dev, i, 0);
 243                 if (ret < 0)
 244                         goto err;
 245         }
 246
 247         /* Stop the backend. */
 248         for (i = 0; i < dev->max_queue_pairs * 2; ++i) {
 249                 state.index = i;
 250                 ret = dev->ops->get_vring_base(dev, &state);
 251                 if (ret < 0) {
 252                         PMD_DRV_LOG(ERR, "(%s) get_vring_base failed, index=%u", dev->path, i);
 253                         goto err;
 254                 }
 255         }
 256
 257         dev->started = false;
 258
 259 out:
 260         pthread_mutex_unlock(&dev->mutex);
 261
 262         return 0;
 263 err:
 264         pthread_mutex_unlock(&dev->mutex);
 265
 266         PMD_INIT_LOG(ERR, "(%s) Failed to stop device\n", dev->path);
 267
 268         return -1;
 269 }
 270
 271 static inline void
 272 parse_mac(struct virtio_user_dev *dev, const char *mac)
 273 {
 274         struct rte_ether_addr tmp;
 275
 276         if (!mac)
 277                 return;
 278
 279         if (rte_ether_unformat_addr(mac, &tmp) == 0) {
 280                 memcpy(dev->mac_addr, &tmp, RTE_ETHER_ADDR_LEN);
 281                 dev->mac_specified = 1;
 282         } else {
 283                 /* ignore the wrong mac, use random mac */
 284                 PMD_DRV_LOG(ERR, "wrong format of mac: %s", mac);
 285         }
 286 }
 287
 288 static int
 289 virtio_user_dev_init_notify(struct virtio_user_dev *dev)
 290 {
 291         uint32_t i, j;
 292         int callfd;
 293         int kickfd;
 294
 295         for (i = 0; i < VIRTIO_MAX_VIRTQUEUES; ++i) {
 296                 if (i >= dev->max_queue_pairs * 2) {
 297                         dev->kickfds[i] = -1;
 298                         dev->callfds[i] = -1;
 299                         continue;
 300                 }
 301
 302                 /* May use invalid flag, but some backend uses kickfd and
 303                  * callfd as criteria to judge if dev is alive. so finally we
 304                  * use real event_fd.
 305                  */
 306                 callfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK);
 307                 if (callfd < 0) {
 308                         PMD_DRV_LOG(ERR, "(%s) callfd error, %s", dev->path, strerror(errno));
 309                         break;
 310                 }
 311                 kickfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK);
 312                 if (kickfd < 0) {
 313                         close(callfd);
 314                         PMD_DRV_LOG(ERR, "(%s) kickfd error, %s", dev->path, strerror(errno));
 315                         break;
 316                 }
 317                 dev->callfds[i] = callfd;
 318                 dev->kickfds[i] = kickfd;
 319         }
 320
 321         if (i < VIRTIO_MAX_VIRTQUEUES) {
 322                 for (j = 0; j < i; ++j) {
 323                         close(dev->callfds[j]);
 324                         close(dev->kickfds[j]);
 325                 }
 326
 327                 return -1;
 328         }
 329
 330         return 0;
 331 }
 332
 333 static int
 334 virtio_user_fill_intr_handle(struct virtio_user_dev *dev)
 335 {
 336         uint32_t i;
 337         struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->port_id];
 338
 339         if (!eth_dev->intr_handle) {
 340                 eth_dev->intr_handle = malloc(sizeof(*eth_dev->intr_handle));
 341                 if (!eth_dev->intr_handle) {
 342                         PMD_DRV_LOG(ERR, "(%s) failed to allocate intr_handle", dev->path);
 343                         return -1;
 344                 }
 345                 memset(eth_dev->intr_handle, 0, sizeof(*eth_dev->intr_handle));
 346         }
 347
 348         for (i = 0; i < dev->max_queue_pairs; ++i)
 349                 eth_dev->intr_handle->efds[i] = dev->callfds[i];
 350         eth_dev->intr_handle->nb_efd = dev->max_queue_pairs;
 351         eth_dev->intr_handle->max_intr = dev->max_queue_pairs + 1;
 352         eth_dev->intr_handle->type = RTE_INTR_HANDLE_VDEV;
 353         /* For virtio vdev, no need to read counter for clean */
 354         eth_dev->intr_handle->efd_counter_size = 0;
 355         eth_dev->intr_handle->fd = -1;
 356         if (dev->vhostfd >= 0)
 357                 eth_dev->intr_handle->fd = dev->vhostfd;
 358         else if (dev->is_server)
 359                 eth_dev->intr_handle->fd = dev->listenfd;
 360
 361         return 0;
 362 }
 363
 364 static void
 365 virtio_user_mem_event_cb(enum rte_mem_event type __rte_unused,
 366                          const void *addr,
 367                          size_t len __rte_unused,
 368                          void *arg)
 369 {
 370         struct virtio_user_dev *dev = arg;
 371         struct rte_memseg_list *msl;
 372         uint16_t i;
 373         int ret = 0;
 374
 375         /* ignore externally allocated memory */
 376         msl = rte_mem_virt2memseg_list(addr);
 377         if (msl->external)
 378                 return;
 379
 380         pthread_mutex_lock(&dev->mutex);
 381
 382         if (dev->started == false)
 383                 goto exit;
 384
 385         /* Step 1: pause the active queues */
 386         for (i = 0; i < dev->queue_pairs; i++) {
 387                 ret = dev->ops->enable_qp(dev, i, 0);
 388                 if (ret < 0)
 389                         goto exit;
 390         }
 391
 392         /* Step 2: update memory regions */
 393         ret = dev->ops->set_memory_table(dev);
 394         if (ret < 0)
 395                 goto exit;
 396
 397         /* Step 3: resume the active queues */
 398         for (i = 0; i < dev->queue_pairs; i++) {
 399                 ret = dev->ops->enable_qp(dev, i, 1);
 400                 if (ret < 0)
 401                         goto exit;
 402         }
 403
 404 exit:
 405         pthread_mutex_unlock(&dev->mutex);
 406
 407         if (ret < 0)
 408                 PMD_DRV_LOG(ERR, "(%s) Failed to update memory table\n", dev->path);
 409 }
 410
 411 static int
 412 virtio_user_dev_setup(struct virtio_user_dev *dev)
 413 {
 414         uint32_t q;
 415
 416         dev->vhostfd = -1;
 417         dev->vhostfds = NULL;
 418         dev->tapfds = NULL;
 419
 420         if (dev->is_server) {
 421                 if (dev->backend_type != VIRTIO_USER_BACKEND_VHOST_USER) {
 422                         PMD_DRV_LOG(ERR, "Server mode only supports vhost-user!");
 423                         return -1;
 424                 }
 425                 dev->ops = &virtio_ops_user;
 426         } else {
 427                 if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER) {
 428                         dev->ops = &virtio_ops_user;
 429                 } else if (dev->backend_type ==
 430                                         VIRTIO_USER_BACKEND_VHOST_KERNEL) {
 431                         dev->ops = &virtio_ops_kernel;
 432
 433                         dev->vhostfds = malloc(dev->max_queue_pairs *
 434                                                sizeof(int));
 435                         dev->tapfds = malloc(dev->max_queue_pairs *
 436                                              sizeof(int));
 437                         if (!dev->vhostfds || !dev->tapfds) {
 438                                 PMD_INIT_LOG(ERR, "(%s) Failed to allocate FDs", dev->path);
 439                                 return -1;
 440                         }
 441
 442                         for (q = 0; q < dev->max_queue_pairs; ++q) {
 443                                 dev->vhostfds[q] = -1;
 444                                 dev->tapfds[q] = -1;
 445                         }
 446                 } else if (dev->backend_type ==
 447                                 VIRTIO_USER_BACKEND_VHOST_VDPA) {
 448                         dev->ops = &virtio_ops_vdpa;
 449                 } else {
 450                         PMD_DRV_LOG(ERR, "(%s) Unknown backend type", dev->path);
 451                         return -1;
 452                 }
 453         }
 454
 455         if (dev->ops->setup(dev) < 0) {
 456                 PMD_INIT_LOG(ERR, "(%s) Failed to setup backend\n", dev->path);
 457                 return -1;
 458         }
 459
 460         if (virtio_user_dev_init_notify(dev) < 0) {
 461                 PMD_INIT_LOG(ERR, "(%s) Failed to init notifiers\n", dev->path);
 462                 return -1;
 463         }
 464
 465         if (virtio_user_fill_intr_handle(dev) < 0) {
 466                 PMD_INIT_LOG(ERR, "(%s) Failed to init interrupt handler\n", dev->path);
 467                 return -1;
 468         }
 469
 470         return 0;
 471 }
 472
 473 /* Use below macro to filter features from vhost backend */
 474 #define VIRTIO_USER_SUPPORTED_FEATURES                  \
 475         (1ULL << VIRTIO_NET_F_MAC               |       \
 476          1ULL << VIRTIO_NET_F_STATUS            |       \
 477          1ULL << VIRTIO_NET_F_MQ                |       \
 478          1ULL << VIRTIO_NET_F_CTRL_MAC_ADDR     |       \
 479          1ULL << VIRTIO_NET_F_CTRL_VQ           |       \
 480          1ULL << VIRTIO_NET_F_CTRL_RX           |       \
 481          1ULL << VIRTIO_NET_F_CTRL_VLAN         |       \
 482          1ULL << VIRTIO_NET_F_CSUM              |       \
 483          1ULL << VIRTIO_NET_F_HOST_TSO4         |       \
 484          1ULL << VIRTIO_NET_F_HOST_TSO6         |       \
 485          1ULL << VIRTIO_NET_F_MRG_RXBUF         |       \
 486          1ULL << VIRTIO_RING_F_INDIRECT_DESC    |       \
 487          1ULL << VIRTIO_NET_F_GUEST_CSUM        |       \
 488          1ULL << VIRTIO_NET_F_GUEST_TSO4        |       \
 489          1ULL << VIRTIO_NET_F_GUEST_TSO6        |       \
 490          1ULL << VIRTIO_F_IN_ORDER              |       \
 491          1ULL << VIRTIO_F_VERSION_1             |       \
 492          1ULL << VIRTIO_F_RING_PACKED           |       \
 493          1ULL << VHOST_USER_F_PROTOCOL_FEATURES)
 494
 495 #define VHOST_USER_SUPPORTED_PROTOCOL_FEATURES          \
 496         (1ULL << VHOST_USER_PROTOCOL_F_MQ |             \
 497          1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK |      \
 498          1ULL << VHOST_USER_PROTOCOL_F_STATUS)
 499
 500 #define VHOST_VDPA_SUPPORTED_PROTOCOL_FEATURES          \
 501         (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2   |       \
 502         1ULL << VHOST_BACKEND_F_IOTLB_BATCH)
 503 int
 504 virtio_user_dev_init(struct virtio_user_dev *dev, char *path, int queues,
 505                      int cq, int queue_size, const char *mac, char **ifname,
 506                      int server, int mrg_rxbuf, int in_order, int packed_vq,
 507                      enum virtio_user_backend_type backend_type)
 508 {
 509         uint64_t protocol_features = 0;
 510
 511         pthread_mutex_init(&dev->mutex, NULL);
 512         strlcpy(dev->path, path, PATH_MAX);
 513         dev->started = 0;
 514         dev->max_queue_pairs = queues;
 515         dev->queue_pairs = 1; /* mq disabled by default */
 516         dev->queue_size = queue_size;
 517         dev->is_server = server;
 518         dev->mac_specified = 0;
 519         dev->frontend_features = 0;
 520         dev->unsupported_features = ~VIRTIO_USER_SUPPORTED_FEATURES;
 521         dev->backend_type = backend_type;
 522
 523         if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER)
 524                 dev->protocol_features = VHOST_USER_SUPPORTED_PROTOCOL_FEATURES;
 525         else if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_VDPA)
 526                 dev->protocol_features = VHOST_VDPA_SUPPORTED_PROTOCOL_FEATURES;
 527
 528         parse_mac(dev, mac);
 529
 530         if (*ifname) {
 531                 dev->ifname = *ifname;
 532                 *ifname = NULL;
 533         }
 534
 535         if (virtio_user_dev_setup(dev) < 0) {
 536                 PMD_INIT_LOG(ERR, "(%s) backend set up fails", dev->path);
 537                 return -1;
 538         }
 539
 540         if (dev->backend_type != VIRTIO_USER_BACKEND_VHOST_USER)
 541                 dev->unsupported_features |=
 542                         (1ULL << VHOST_USER_F_PROTOCOL_FEATURES);
 543
 544         if (!dev->is_server) {
 545                 if (dev->ops->set_owner(dev) < 0) {
 546                         PMD_INIT_LOG(ERR, "(%s) Failed to set backend owner", dev->path);
 547                         return -1;
 548                 }
 549
 550                 if (dev->ops->get_features(dev, &dev->device_features) < 0) {
 551                         PMD_INIT_LOG(ERR, "(%s) Failed to get backend features", dev->path);
 552                         return -1;
 553                 }
 554
 555
 556                 if ((dev->device_features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
 557                                 (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_VDPA)) {
 558                         if (dev->ops->get_protocol_features(dev, &protocol_features)) {
 559                                 PMD_INIT_LOG(ERR, "(%s) Failed to get backend protocol features",
 560                                                 dev->path);
 561                                 return -1;
 562                         }
 563
 564                         dev->protocol_features &= protocol_features;
 565
 566                         if (dev->ops->set_protocol_features(dev, dev->protocol_features)) {
 567                                 PMD_INIT_LOG(ERR, "(%s) Failed to set backend protocol features",
 568                                                 dev->path);
 569                                 return -1;
 570                         }
 571
 572                         if (!(dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_MQ)))
 573                                 dev->unsupported_features |= (1ull << VIRTIO_NET_F_MQ);
 574                 }
 575         } else {
 576                 /* We just pretend vhost-user can support all these features.
 577                  * Note that this could be problematic that if some feature is
 578                  * negotiated but not supported by the vhost-user which comes
 579                  * later.
 580                  */
 581                 dev->device_features = VIRTIO_USER_SUPPORTED_FEATURES;
 582
 583                 /* We cannot assume VHOST_USER_PROTOCOL_F_STATUS is supported
 584                  * until it's negotiated
 585                  */
 586                 dev->protocol_features &=
 587                         ~(1ULL << VHOST_USER_PROTOCOL_F_STATUS);
 588         }
 589
 590
 591
 592         if (!mrg_rxbuf)
 593                 dev->unsupported_features |= (1ull << VIRTIO_NET_F_MRG_RXBUF);
 594
 595         if (!in_order)
 596                 dev->unsupported_features |= (1ull << VIRTIO_F_IN_ORDER);
 597
 598         if (!packed_vq)
 599                 dev->unsupported_features |= (1ull << VIRTIO_F_RING_PACKED);
 600
 601         if (dev->mac_specified)
 602                 dev->frontend_features |= (1ull << VIRTIO_NET_F_MAC);
 603         else
 604                 dev->unsupported_features |= (1ull << VIRTIO_NET_F_MAC);
 605
 606         if (cq) {
 607                 /* device does not really need to know anything about CQ,
 608                  * so if necessary, we just claim to support CQ
 609                  */
 610                 dev->frontend_features |= (1ull << VIRTIO_NET_F_CTRL_VQ);
 611         } else {
 612                 dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_VQ);
 613                 /* Also disable features that depend on VIRTIO_NET_F_CTRL_VQ */
 614                 dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_RX);
 615                 dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_VLAN);
 616                 dev->unsupported_features |=
 617                         (1ull << VIRTIO_NET_F_GUEST_ANNOUNCE);
 618                 dev->unsupported_features |= (1ull << VIRTIO_NET_F_MQ);
 619                 dev->unsupported_features |=
 620                         (1ull << VIRTIO_NET_F_CTRL_MAC_ADDR);
 621         }
 622
 623         /* The backend will not report this feature, we add it explicitly */
 624         if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER)
 625                 dev->frontend_features |= (1ull << VIRTIO_NET_F_STATUS);
 626
 627         /*
 628          * Device features =
 629          *     (frontend_features | backend_features) & ~unsupported_features;
 630          */
 631         dev->device_features |= dev->frontend_features;
 632         dev->device_features &= ~dev->unsupported_features;
 633
 634         if (rte_mem_event_callback_register(VIRTIO_USER_MEM_EVENT_CLB_NAME,
 635                                 virtio_user_mem_event_cb, dev)) {
 636                 if (rte_errno != ENOTSUP) {
 637                         PMD_INIT_LOG(ERR, "(%s) Failed to register mem event callback\n",
 638                                         dev->path);
 639                         return -1;
 640                 }
 641         }
 642
 643         return 0;
 644 }
 645
 646 void
 647 virtio_user_dev_uninit(struct virtio_user_dev *dev)
 648 {
 649         uint32_t i;
 650
 651         virtio_user_stop_device(dev);
 652
 653         rte_mem_event_callback_unregister(VIRTIO_USER_MEM_EVENT_CLB_NAME, dev);
 654
 655         for (i = 0; i < dev->max_queue_pairs * 2; ++i) {
 656                 close(dev->callfds[i]);
 657                 close(dev->kickfds[i]);
 658         }
 659
 660         if (dev->vhostfd >= 0)
 661                 close(dev->vhostfd);
 662
 663         if (dev->is_server && dev->listenfd >= 0) {
 664                 close(dev->listenfd);
 665                 dev->listenfd = -1;
 666         }
 667
 668         if (dev->vhostfds) {
 669                 for (i = 0; i < dev->max_queue_pairs; ++i) {
 670                         close(dev->vhostfds[i]);
 671                         if (dev->tapfds[i] >= 0)
 672                                 close(dev->tapfds[i]);
 673                 }
 674                 free(dev->vhostfds);
 675                 free(dev->tapfds);
 676         }
 677
 678         free(dev->ifname);
 679
 680         if (dev->is_server)
 681                 unlink(dev->path);
 682 }
 683
 684 uint8_t
 685 virtio_user_handle_mq(struct virtio_user_dev *dev, uint16_t q_pairs)
 686 {
 687         uint16_t i;
 688         uint8_t ret = 0;
 689
 690         if (q_pairs > dev->max_queue_pairs) {
 691                 PMD_INIT_LOG(ERR, "(%s) multi-q config %u, but only %u supported",
 692                              dev->path, q_pairs, dev->max_queue_pairs);
 693                 return -1;
 694         }
 695
 696         /* Server mode can't enable queue pairs if vhostfd is invalid,
 697          * always return 0 in this case.
 698          */
 699         if (!dev->is_server || dev->vhostfd >= 0) {
 700                 for (i = 0; i < q_pairs; ++i)
 701                         ret |= dev->ops->enable_qp(dev, i, 1);
 702                 for (i = q_pairs; i < dev->max_queue_pairs; ++i)
 703                         ret |= dev->ops->enable_qp(dev, i, 0);
 704         }
 705         dev->queue_pairs = q_pairs;
 706
 707         return ret;
 708 }
 709
 710 static uint32_t
 711 virtio_user_handle_ctrl_msg(struct virtio_user_dev *dev, struct vring *vring,
 712                             uint16_t idx_hdr)
 713 {
 714         struct virtio_net_ctrl_hdr *hdr;
 715         virtio_net_ctrl_ack status = ~0;
 716         uint16_t i, idx_data, idx_status;
 717         uint32_t n_descs = 0;
 718
 719         /* locate desc for header, data, and status */
 720         idx_data = vring->desc[idx_hdr].next;
 721         n_descs++;
 722
 723         i = idx_data;
 724         while (vring->desc[i].flags == VRING_DESC_F_NEXT) {
 725                 i = vring->desc[i].next;
 726                 n_descs++;
 727         }
 728
 729         /* locate desc for status */
 730         idx_status = i;
 731         n_descs++;
 732
 733         hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr;
 734         if (hdr->class == VIRTIO_NET_CTRL_MQ &&
 735             hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
 736                 uint16_t queues;
 737
 738                 queues = *(uint16_t *)(uintptr_t)vring->desc[idx_data].addr;
 739                 status = virtio_user_handle_mq(dev, queues);
 740         } else if (hdr->class == VIRTIO_NET_CTRL_RX  ||
 741                    hdr->class == VIRTIO_NET_CTRL_MAC ||
 742                    hdr->class == VIRTIO_NET_CTRL_VLAN) {
 743                 status = 0;
 744         }
 745
 746         /* Update status */
 747         *(virtio_net_ctrl_ack *)(uintptr_t)vring->desc[idx_status].addr = status;
 748
 749         return n_descs;
 750 }
 751
 752 static inline int
 753 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
 754 {
 755         uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
 756
 757         return wrap_counter == !!(flags & VRING_PACKED_DESC_F_AVAIL) &&
 758                 wrap_counter != !!(flags & VRING_PACKED_DESC_F_USED);
 759 }
 760
 761 static uint32_t
 762 virtio_user_handle_ctrl_msg_packed(struct virtio_user_dev *dev,
 763                                    struct vring_packed *vring,
 764                                    uint16_t idx_hdr)
 765 {
 766         struct virtio_net_ctrl_hdr *hdr;
 767         virtio_net_ctrl_ack status = ~0;
 768         uint16_t idx_data, idx_status;
 769         /* initialize to one, header is first */
 770         uint32_t n_descs = 1;
 771
 772         /* locate desc for header, data, and status */
 773         idx_data = idx_hdr + 1;
 774         if (idx_data >= dev->queue_size)
 775                 idx_data -= dev->queue_size;
 776
 777         n_descs++;
 778
 779         idx_status = idx_data;
 780         while (vring->desc[idx_status].flags & VRING_DESC_F_NEXT) {
 781                 idx_status++;
 782                 if (idx_status >= dev->queue_size)
 783                         idx_status -= dev->queue_size;
 784                 n_descs++;
 785         }
 786
 787         hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr;
 788         if (hdr->class == VIRTIO_NET_CTRL_MQ &&
 789             hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
 790                 uint16_t queues;
 791
 792                 queues = *(uint16_t *)(uintptr_t)
 793                                 vring->desc[idx_data].addr;
 794                 status = virtio_user_handle_mq(dev, queues);
 795         } else if (hdr->class == VIRTIO_NET_CTRL_RX  ||
 796                    hdr->class == VIRTIO_NET_CTRL_MAC ||
 797                    hdr->class == VIRTIO_NET_CTRL_VLAN) {
 798                 status = 0;
 799         }
 800
 801         /* Update status */
 802         *(virtio_net_ctrl_ack *)(uintptr_t)
 803                 vring->desc[idx_status].addr = status;
 804
 805         /* Update used descriptor */
 806         vring->desc[idx_hdr].id = vring->desc[idx_status].id;
 807         vring->desc[idx_hdr].len = sizeof(status);
 808
 809         return n_descs;
 810 }
 811
 812 void
 813 virtio_user_handle_cq_packed(struct virtio_user_dev *dev, uint16_t queue_idx)
 814 {
 815         struct virtio_user_queue *vq = &dev->packed_queues[queue_idx];
 816         struct vring_packed *vring = &dev->packed_vrings[queue_idx];
 817         uint16_t n_descs, flags;
 818
 819         /* Perform a load-acquire barrier in desc_is_avail to
 820          * enforce the ordering between desc flags and desc
 821          * content.
 822          */
 823         while (desc_is_avail(&vring->desc[vq->used_idx],
 824                              vq->used_wrap_counter)) {
 825
 826                 n_descs = virtio_user_handle_ctrl_msg_packed(dev, vring,
 827                                 vq->used_idx);
 828
 829                 flags = VRING_DESC_F_WRITE;
 830                 if (vq->used_wrap_counter)
 831                         flags |= VRING_PACKED_DESC_F_AVAIL_USED;
 832
 833                 __atomic_store_n(&vring->desc[vq->used_idx].flags, flags,
 834                                  __ATOMIC_RELEASE);
 835
 836                 vq->used_idx += n_descs;
 837                 if (vq->used_idx >= dev->queue_size) {
 838                         vq->used_idx -= dev->queue_size;
 839                         vq->used_wrap_counter ^= 1;
 840                 }
 841         }
 842 }
 843
 844 void
 845 virtio_user_handle_cq(struct virtio_user_dev *dev, uint16_t queue_idx)
 846 {
 847         uint16_t avail_idx, desc_idx;
 848         struct vring_used_elem *uep;
 849         uint32_t n_descs;
 850         struct vring *vring = &dev->vrings[queue_idx];
 851
 852         /* Consume avail ring, using used ring idx as first one */
 853         while (__atomic_load_n(&vring->used->idx, __ATOMIC_RELAXED)
 854                != vring->avail->idx) {
 855                 avail_idx = __atomic_load_n(&vring->used->idx, __ATOMIC_RELAXED)
 856                             & (vring->num - 1);
 857                 desc_idx = vring->avail->ring[avail_idx];
 858
 859                 n_descs = virtio_user_handle_ctrl_msg(dev, vring, desc_idx);
 860
 861                 /* Update used ring */
 862                 uep = &vring->used->ring[avail_idx];
 863                 uep->id = desc_idx;
 864                 uep->len = n_descs;
 865
 866                 __atomic_add_fetch(&vring->used->idx, 1, __ATOMIC_RELAXED);
 867         }
 868 }
 869
 870 int
 871 virtio_user_dev_set_status(struct virtio_user_dev *dev, uint8_t status)
 872 {
 873         int ret;
 874
 875         pthread_mutex_lock(&dev->mutex);
 876         dev->status = status;
 877         ret = dev->ops->set_status(dev, status);
 878         if (ret && ret != -ENOTSUP)
 879                 PMD_INIT_LOG(ERR, "(%s) Failed to set backend status\n", dev->path);
 880
 881         pthread_mutex_unlock(&dev->mutex);
 882         return ret;
 883 }
 884
 885 int
 886 virtio_user_dev_update_status(struct virtio_user_dev *dev)
 887 {
 888         int ret;
 889         uint8_t status;
 890
 891         pthread_mutex_lock(&dev->mutex);
 892
 893         ret = dev->ops->get_status(dev, &status);
 894         if (!ret) {
 895                 dev->status = status;
 896                 PMD_INIT_LOG(DEBUG, "Updated Device Status(0x%08x):\n"
 897                         "\t-RESET: %u\n"
 898                         "\t-ACKNOWLEDGE: %u\n"
 899                         "\t-DRIVER: %u\n"
 900                         "\t-DRIVER_OK: %u\n"
 901                         "\t-FEATURES_OK: %u\n"
 902                         "\t-DEVICE_NEED_RESET: %u\n"
 903                         "\t-FAILED: %u\n",
 904                         dev->status,
 905                         (dev->status == VIRTIO_CONFIG_STATUS_RESET),
 906                         !!(dev->status & VIRTIO_CONFIG_STATUS_ACK),
 907                         !!(dev->status & VIRTIO_CONFIG_STATUS_DRIVER),
 908                         !!(dev->status & VIRTIO_CONFIG_STATUS_DRIVER_OK),
 909                         !!(dev->status & VIRTIO_CONFIG_STATUS_FEATURES_OK),
 910                         !!(dev->status & VIRTIO_CONFIG_STATUS_DEV_NEED_RESET),
 911                         !!(dev->status & VIRTIO_CONFIG_STATUS_FAILED));
 912         } else if (ret != -ENOTSUP) {
 913                 PMD_INIT_LOG(ERR, "(%s) Failed to get backend status\n", dev->path);
 914         }
 915
 916         pthread_mutex_unlock(&dev->mutex);
 917         return ret;
 918 }