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
  41         file.index = queue_sel;
  42         file.fd = dev->callfds[queue_sel];
  43         dev->ops->send_request(dev, VHOST_USER_SET_VRING_CALL, &file);
  44
  45         return 0;
  46 }
  47
  48 static int
  49 virtio_user_kick_queue(struct virtio_user_dev *dev, uint32_t queue_sel)
  50 {
  51         struct vhost_vring_file file;
  52         struct vhost_vring_state state;
  53         struct vring *vring = &dev->vrings[queue_sel];
  54         struct vring_packed *pq_vring = &dev->packed_vrings[queue_sel];
  55         struct vhost_vring_addr addr = {
  56                 .index = queue_sel,
  57                 .log_guest_addr = 0,
  58                 .flags = 0, /* disable log */
  59         };
  60
  61         if (dev->features & (1ULL << VIRTIO_F_RING_PACKED)) {
  62                 addr.desc_user_addr =
  63                         (uint64_t)(uintptr_t)pq_vring->desc;
  64                 addr.avail_user_addr =
  65                         (uint64_t)(uintptr_t)pq_vring->driver;
  66                 addr.used_user_addr =
  67                         (uint64_t)(uintptr_t)pq_vring->device;
  68         } else {
  69                 addr.desc_user_addr = (uint64_t)(uintptr_t)vring->desc;
  70                 addr.avail_user_addr = (uint64_t)(uintptr_t)vring->avail;
  71                 addr.used_user_addr = (uint64_t)(uintptr_t)vring->used;
  72         }
  73
  74         state.index = queue_sel;
  75         state.num = vring->num;
  76         dev->ops->send_request(dev, VHOST_USER_SET_VRING_NUM, &state);
  77
  78         state.index = queue_sel;
  79         state.num = 0; /* no reservation */
  80         if (dev->features & (1ULL << VIRTIO_F_RING_PACKED))
  81                 state.num |= (1 << 15);
  82         dev->ops->send_request(dev, VHOST_USER_SET_VRING_BASE, &state);
  83
  84         dev->ops->send_request(dev, VHOST_USER_SET_VRING_ADDR, &addr);
  85
  86         /* Of all per virtqueue MSGs, make sure VHOST_USER_SET_VRING_KICK comes
  87          * lastly because vhost depends on this msg to judge if
  88          * virtio is ready.
  89          */
  90         file.index = queue_sel;
  91         file.fd = dev->kickfds[queue_sel];
  92         dev->ops->send_request(dev, VHOST_USER_SET_VRING_KICK, &file);
  93
  94         return 0;
  95 }
  96
  97 static int
  98 virtio_user_queue_setup(struct virtio_user_dev *dev,
  99                         int (*fn)(struct virtio_user_dev *, uint32_t))
 100 {
 101         uint32_t i, queue_sel;
 102
 103         for (i = 0; i < dev->max_queue_pairs; ++i) {
 104                 queue_sel = 2 * i + VTNET_SQ_RQ_QUEUE_IDX;
 105                 if (fn(dev, queue_sel) < 0) {
 106                         PMD_DRV_LOG(INFO, "setup rx vq fails: %u", i);
 107                         return -1;
 108                 }
 109         }
 110         for (i = 0; i < dev->max_queue_pairs; ++i) {
 111                 queue_sel = 2 * i + VTNET_SQ_TQ_QUEUE_IDX;
 112                 if (fn(dev, queue_sel) < 0) {
 113                         PMD_DRV_LOG(INFO, "setup tx vq fails: %u", i);
 114                         return -1;
 115                 }
 116         }
 117
 118         return 0;
 119 }
 120
 121 int
 122 virtio_user_dev_set_features(struct virtio_user_dev *dev)
 123 {
 124         uint64_t features;
 125         int ret = -1;
 126
 127         pthread_mutex_lock(&dev->mutex);
 128
 129         if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER &&
 130                         dev->vhostfd < 0)
 131                 goto error;
 132
 133         /* Step 0: tell vhost to create queues */
 134         if (virtio_user_queue_setup(dev, virtio_user_create_queue) < 0)
 135                 goto error;
 136
 137         features = dev->features;
 138
 139         /* Strip VIRTIO_NET_F_MAC, as MAC address is handled in vdev init */
 140         features &= ~(1ull << VIRTIO_NET_F_MAC);
 141         /* Strip VIRTIO_NET_F_CTRL_VQ, as devices do not really need to know */
 142         features &= ~(1ull << VIRTIO_NET_F_CTRL_VQ);
 143         features &= ~(1ull << VIRTIO_NET_F_STATUS);
 144         ret = dev->ops->send_request(dev, VHOST_USER_SET_FEATURES, &features);
 145         if (ret < 0)
 146                 goto error;
 147         PMD_DRV_LOG(INFO, "set features: %" PRIx64, features);
 148 error:
 149         pthread_mutex_unlock(&dev->mutex);
 150
 151         return ret;
 152 }
 153
 154 int
 155 virtio_user_start_device(struct virtio_user_dev *dev)
 156 {
 157         int ret;
 158
 159         /*
 160          * XXX workaround!
 161          *
 162          * We need to make sure that the locks will be
 163          * taken in the correct order to avoid deadlocks.
 164          *
 165          * Before releasing this lock, this thread should
 166          * not trigger any memory hotplug events.
 167          *
 168          * This is a temporary workaround, and should be
 169          * replaced when we get proper supports from the
 170          * memory subsystem in the future.
 171          */
 172         rte_mcfg_mem_read_lock();
 173         pthread_mutex_lock(&dev->mutex);
 174
 175         if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER &&
 176                         dev->vhostfd < 0)
 177                 goto error;
 178
 179         /* Step 2: share memory regions */
 180         ret = dev->ops->send_request(dev, VHOST_USER_SET_MEM_TABLE, NULL);
 181         if (ret < 0)
 182                 goto error;
 183
 184         /* Step 3: kick queues */
 185         if (virtio_user_queue_setup(dev, virtio_user_kick_queue) < 0)
 186                 goto error;
 187
 188         /* Step 4: enable queues
 189          * we enable the 1st queue pair by default.
 190          */
 191         dev->ops->enable_qp(dev, 0, 1);
 192
 193         dev->started = true;
 194         pthread_mutex_unlock(&dev->mutex);
 195         rte_mcfg_mem_read_unlock();
 196
 197         return 0;
 198 error:
 199         pthread_mutex_unlock(&dev->mutex);
 200         rte_mcfg_mem_read_unlock();
 201         /* TODO: free resource here or caller to check */
 202         return -1;
 203 }
 204
 205 int virtio_user_stop_device(struct virtio_user_dev *dev)
 206 {
 207         struct vhost_vring_state state;
 208         uint32_t i;
 209         int error = 0;
 210
 211         pthread_mutex_lock(&dev->mutex);
 212         if (!dev->started)
 213                 goto out;
 214
 215         for (i = 0; i < dev->max_queue_pairs; ++i)
 216                 dev->ops->enable_qp(dev, i, 0);
 217
 218         /* Stop the backend. */
 219         for (i = 0; i < dev->max_queue_pairs * 2; ++i) {
 220                 state.index = i;
 221                 if (dev->ops->send_request(dev, VHOST_USER_GET_VRING_BASE,
 222                                            &state) < 0) {
 223                         PMD_DRV_LOG(ERR, "get_vring_base failed, index=%u\n",
 224                                     i);
 225                         error = -1;
 226                         goto out;
 227                 }
 228         }
 229
 230         dev->started = false;
 231 out:
 232         pthread_mutex_unlock(&dev->mutex);
 233
 234         return error;
 235 }
 236
 237 static inline void
 238 parse_mac(struct virtio_user_dev *dev, const char *mac)
 239 {
 240         struct rte_ether_addr tmp;
 241
 242         if (!mac)
 243                 return;
 244
 245         if (rte_ether_unformat_addr(mac, &tmp) == 0) {
 246                 memcpy(dev->mac_addr, &tmp, RTE_ETHER_ADDR_LEN);
 247                 dev->mac_specified = 1;
 248         } else {
 249                 /* ignore the wrong mac, use random mac */
 250                 PMD_DRV_LOG(ERR, "wrong format of mac: %s", mac);
 251         }
 252 }
 253
 254 static int
 255 virtio_user_dev_init_notify(struct virtio_user_dev *dev)
 256 {
 257         uint32_t i, j;
 258         int callfd;
 259         int kickfd;
 260
 261         for (i = 0; i < VIRTIO_MAX_VIRTQUEUES; ++i) {
 262                 if (i >= dev->max_queue_pairs * 2) {
 263                         dev->kickfds[i] = -1;
 264                         dev->callfds[i] = -1;
 265                         continue;
 266                 }
 267
 268                 /* May use invalid flag, but some backend uses kickfd and
 269                  * callfd as criteria to judge if dev is alive. so finally we
 270                  * use real event_fd.
 271                  */
 272                 callfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK);
 273                 if (callfd < 0) {
 274                         PMD_DRV_LOG(ERR, "callfd error, %s", strerror(errno));
 275                         break;
 276                 }
 277                 kickfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK);
 278                 if (kickfd < 0) {
 279                         close(callfd);
 280                         PMD_DRV_LOG(ERR, "kickfd error, %s", strerror(errno));
 281                         break;
 282                 }
 283                 dev->callfds[i] = callfd;
 284                 dev->kickfds[i] = kickfd;
 285         }
 286
 287         if (i < VIRTIO_MAX_VIRTQUEUES) {
 288                 for (j = 0; j < i; ++j) {
 289                         close(dev->callfds[j]);
 290                         close(dev->kickfds[j]);
 291                 }
 292
 293                 return -1;
 294         }
 295
 296         return 0;
 297 }
 298
 299 static int
 300 virtio_user_fill_intr_handle(struct virtio_user_dev *dev)
 301 {
 302         uint32_t i;
 303         struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->port_id];
 304
 305         if (!eth_dev->intr_handle) {
 306                 eth_dev->intr_handle = malloc(sizeof(*eth_dev->intr_handle));
 307                 if (!eth_dev->intr_handle) {
 308                         PMD_DRV_LOG(ERR, "fail to allocate intr_handle");
 309                         return -1;
 310                 }
 311                 memset(eth_dev->intr_handle, 0, sizeof(*eth_dev->intr_handle));
 312         }
 313
 314         for (i = 0; i < dev->max_queue_pairs; ++i)
 315                 eth_dev->intr_handle->efds[i] = dev->callfds[i];
 316         eth_dev->intr_handle->nb_efd = dev->max_queue_pairs;
 317         eth_dev->intr_handle->max_intr = dev->max_queue_pairs + 1;
 318         eth_dev->intr_handle->type = RTE_INTR_HANDLE_VDEV;
 319         /* For virtio vdev, no need to read counter for clean */
 320         eth_dev->intr_handle->efd_counter_size = 0;
 321         eth_dev->intr_handle->fd = -1;
 322         if (dev->vhostfd >= 0)
 323                 eth_dev->intr_handle->fd = dev->vhostfd;
 324         else if (dev->is_server)
 325                 eth_dev->intr_handle->fd = dev->listenfd;
 326
 327         return 0;
 328 }
 329
 330 static void
 331 virtio_user_mem_event_cb(enum rte_mem_event type __rte_unused,
 332                          const void *addr,
 333                          size_t len __rte_unused,
 334                          void *arg)
 335 {
 336         struct virtio_user_dev *dev = arg;
 337         struct rte_memseg_list *msl;
 338         uint16_t i;
 339
 340         /* ignore externally allocated memory */
 341         msl = rte_mem_virt2memseg_list(addr);
 342         if (msl->external)
 343                 return;
 344
 345         pthread_mutex_lock(&dev->mutex);
 346
 347         if (dev->started == false)
 348                 goto exit;
 349
 350         /* Step 1: pause the active queues */
 351         for (i = 0; i < dev->queue_pairs; i++)
 352                 dev->ops->enable_qp(dev, i, 0);
 353
 354         /* Step 2: update memory regions */
 355         dev->ops->send_request(dev, VHOST_USER_SET_MEM_TABLE, NULL);
 356
 357         /* Step 3: resume the active queues */
 358         for (i = 0; i < dev->queue_pairs; i++)
 359                 dev->ops->enable_qp(dev, i, 1);
 360
 361 exit:
 362         pthread_mutex_unlock(&dev->mutex);
 363 }
 364
 365 static int
 366 virtio_user_dev_setup(struct virtio_user_dev *dev)
 367 {
 368         uint32_t q;
 369
 370         dev->vhostfd = -1;
 371         dev->vhostfds = NULL;
 372         dev->tapfds = NULL;
 373
 374         if (dev->is_server) {
 375                 if (dev->backend_type != VIRTIO_USER_BACKEND_VHOST_USER) {
 376                         PMD_DRV_LOG(ERR, "Server mode only supports vhost-user!");
 377                         return -1;
 378                 }
 379                 dev->ops = &virtio_ops_user;
 380         } else {
 381                 if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER) {
 382                         dev->ops = &virtio_ops_user;
 383                 } else if (dev->backend_type ==
 384                                         VIRTIO_USER_BACKEND_VHOST_KERNEL) {
 385                         dev->ops = &virtio_ops_kernel;
 386
 387                         dev->vhostfds = malloc(dev->max_queue_pairs *
 388                                                sizeof(int));
 389                         dev->tapfds = malloc(dev->max_queue_pairs *
 390                                              sizeof(int));
 391                         if (!dev->vhostfds || !dev->tapfds) {
 392                                 PMD_INIT_LOG(ERR, "Failed to malloc");
 393                                 return -1;
 394                         }
 395
 396                         for (q = 0; q < dev->max_queue_pairs; ++q) {
 397                                 dev->vhostfds[q] = -1;
 398                                 dev->tapfds[q] = -1;
 399                         }
 400                 } else if (dev->backend_type ==
 401                                 VIRTIO_USER_BACKEND_VHOST_VDPA) {
 402                         dev->ops = &virtio_ops_vdpa;
 403                 } else {
 404                         PMD_DRV_LOG(ERR, "Unknown backend type");
 405                         return -1;
 406                 }
 407         }
 408
 409         if (dev->ops->setup(dev) < 0)
 410                 return -1;
 411
 412         if (virtio_user_dev_init_notify(dev) < 0)
 413                 return -1;
 414
 415         if (virtio_user_fill_intr_handle(dev) < 0)
 416                 return -1;
 417
 418         return 0;
 419 }
 420
 421 /* Use below macro to filter features from vhost backend */
 422 #define VIRTIO_USER_SUPPORTED_FEATURES                  \
 423         (1ULL << VIRTIO_NET_F_MAC               |       \
 424          1ULL << VIRTIO_NET_F_STATUS            |       \
 425          1ULL << VIRTIO_NET_F_MQ                |       \
 426          1ULL << VIRTIO_NET_F_CTRL_MAC_ADDR     |       \
 427          1ULL << VIRTIO_NET_F_CTRL_VQ           |       \
 428          1ULL << VIRTIO_NET_F_CTRL_RX           |       \
 429          1ULL << VIRTIO_NET_F_CTRL_VLAN         |       \
 430          1ULL << VIRTIO_NET_F_CSUM              |       \
 431          1ULL << VIRTIO_NET_F_HOST_TSO4         |       \
 432          1ULL << VIRTIO_NET_F_HOST_TSO6         |       \
 433          1ULL << VIRTIO_NET_F_MRG_RXBUF         |       \
 434          1ULL << VIRTIO_RING_F_INDIRECT_DESC    |       \
 435          1ULL << VIRTIO_NET_F_GUEST_CSUM        |       \
 436          1ULL << VIRTIO_NET_F_GUEST_TSO4        |       \
 437          1ULL << VIRTIO_NET_F_GUEST_TSO6        |       \
 438          1ULL << VIRTIO_F_IN_ORDER              |       \
 439          1ULL << VIRTIO_F_VERSION_1             |       \
 440          1ULL << VIRTIO_F_RING_PACKED           |       \
 441          1ULL << VHOST_USER_F_PROTOCOL_FEATURES)
 442
 443 #define VHOST_USER_SUPPORTED_PROTOCOL_FEATURES          \
 444         (1ULL << VHOST_USER_PROTOCOL_F_MQ |             \
 445          1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK |      \
 446          1ULL << VHOST_USER_PROTOCOL_F_STATUS)
 447
 448 #define VHOST_VDPA_SUPPORTED_PROTOCOL_FEATURES          \
 449         (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2)
 450 int
 451 virtio_user_dev_init(struct virtio_user_dev *dev, char *path, int queues,
 452                      int cq, int queue_size, const char *mac, char **ifname,
 453                      int server, int mrg_rxbuf, int in_order, int packed_vq,
 454                      enum virtio_user_backend_type backend_type)
 455 {
 456         uint64_t protocol_features = 0;
 457
 458         pthread_mutex_init(&dev->mutex, NULL);
 459         strlcpy(dev->path, path, PATH_MAX);
 460         dev->started = 0;
 461         dev->max_queue_pairs = queues;
 462         dev->queue_pairs = 1; /* mq disabled by default */
 463         dev->queue_size = queue_size;
 464         dev->is_server = server;
 465         dev->mac_specified = 0;
 466         dev->frontend_features = 0;
 467         dev->unsupported_features = ~VIRTIO_USER_SUPPORTED_FEATURES;
 468         dev->backend_type = backend_type;
 469
 470         if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER)
 471                 dev->protocol_features = VHOST_USER_SUPPORTED_PROTOCOL_FEATURES;
 472         else if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_VDPA)
 473                 dev->protocol_features = VHOST_VDPA_SUPPORTED_PROTOCOL_FEATURES;
 474
 475         parse_mac(dev, mac);
 476
 477         if (*ifname) {
 478                 dev->ifname = *ifname;
 479                 *ifname = NULL;
 480         }
 481
 482         if (virtio_user_dev_setup(dev) < 0) {
 483                 PMD_INIT_LOG(ERR, "backend set up fails");
 484                 return -1;
 485         }
 486
 487         if (dev->backend_type != VIRTIO_USER_BACKEND_VHOST_USER)
 488                 dev->unsupported_features |=
 489                         (1ULL << VHOST_USER_F_PROTOCOL_FEATURES);
 490
 491         if (!dev->is_server) {
 492                 if (dev->ops->send_request(dev, VHOST_USER_SET_OWNER,
 493                                            NULL) < 0) {
 494                         PMD_INIT_LOG(ERR, "set_owner fails: %s",
 495                                      strerror(errno));
 496                         return -1;
 497                 }
 498
 499                 if (dev->ops->send_request(dev, VHOST_USER_GET_FEATURES,
 500                                            &dev->device_features) < 0) {
 501                         PMD_INIT_LOG(ERR, "get_features failed: %s",
 502                                      strerror(errno));
 503                         return -1;
 504                 }
 505
 506
 507                 if ((dev->device_features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
 508                                 (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_VDPA)) {
 509                         if (dev->ops->send_request(dev,
 510                                         VHOST_USER_GET_PROTOCOL_FEATURES,
 511                                         &protocol_features))
 512                                 return -1;
 513
 514                         dev->protocol_features &= protocol_features;
 515
 516                         if (dev->ops->send_request(dev,
 517                                         VHOST_USER_SET_PROTOCOL_FEATURES,
 518                                         &dev->protocol_features))
 519                                 return -1;
 520
 521                         if (!(dev->protocol_features &
 522                                         (1ULL << VHOST_USER_PROTOCOL_F_MQ)))
 523                                 dev->unsupported_features |=
 524                                         (1ull << VIRTIO_NET_F_MQ);
 525                 }
 526         } else {
 527                 /* We just pretend vhost-user can support all these features.
 528                  * Note that this could be problematic that if some feature is
 529                  * negotiated but not supported by the vhost-user which comes
 530                  * later.
 531                  */
 532                 dev->device_features = VIRTIO_USER_SUPPORTED_FEATURES;
 533
 534                 /* We cannot assume VHOST_USER_PROTOCOL_F_STATUS is supported
 535                  * until it's negotiated
 536                  */
 537                 dev->protocol_features &=
 538                         ~(1ULL << VHOST_USER_PROTOCOL_F_STATUS);
 539         }
 540
 541
 542
 543         if (!mrg_rxbuf)
 544                 dev->unsupported_features |= (1ull << VIRTIO_NET_F_MRG_RXBUF);
 545
 546         if (!in_order)
 547                 dev->unsupported_features |= (1ull << VIRTIO_F_IN_ORDER);
 548
 549         if (!packed_vq)
 550                 dev->unsupported_features |= (1ull << VIRTIO_F_RING_PACKED);
 551
 552         if (dev->mac_specified)
 553                 dev->frontend_features |= (1ull << VIRTIO_NET_F_MAC);
 554         else
 555                 dev->unsupported_features |= (1ull << VIRTIO_NET_F_MAC);
 556
 557         if (cq) {
 558                 /* device does not really need to know anything about CQ,
 559                  * so if necessary, we just claim to support CQ
 560                  */
 561                 dev->frontend_features |= (1ull << VIRTIO_NET_F_CTRL_VQ);
 562         } else {
 563                 dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_VQ);
 564                 /* Also disable features that depend on VIRTIO_NET_F_CTRL_VQ */
 565                 dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_RX);
 566                 dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_VLAN);
 567                 dev->unsupported_features |=
 568                         (1ull << VIRTIO_NET_F_GUEST_ANNOUNCE);
 569                 dev->unsupported_features |= (1ull << VIRTIO_NET_F_MQ);
 570                 dev->unsupported_features |=
 571                         (1ull << VIRTIO_NET_F_CTRL_MAC_ADDR);
 572         }
 573
 574         /* The backend will not report this feature, we add it explicitly */
 575         if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER)
 576                 dev->frontend_features |= (1ull << VIRTIO_NET_F_STATUS);
 577
 578         /*
 579          * Device features =
 580          *     (frontend_features | backend_features) & ~unsupported_features;
 581          */
 582         dev->device_features |= dev->frontend_features;
 583         dev->device_features &= ~dev->unsupported_features;
 584
 585         if (rte_mem_event_callback_register(VIRTIO_USER_MEM_EVENT_CLB_NAME,
 586                                 virtio_user_mem_event_cb, dev)) {
 587                 if (rte_errno != ENOTSUP) {
 588                         PMD_INIT_LOG(ERR, "Failed to register mem event"
 589                                         " callback\n");
 590                         return -1;
 591                 }
 592         }
 593
 594         return 0;
 595 }
 596
 597 void
 598 virtio_user_dev_uninit(struct virtio_user_dev *dev)
 599 {
 600         uint32_t i;
 601
 602         virtio_user_stop_device(dev);
 603
 604         rte_mem_event_callback_unregister(VIRTIO_USER_MEM_EVENT_CLB_NAME, dev);
 605
 606         for (i = 0; i < dev->max_queue_pairs * 2; ++i) {
 607                 close(dev->callfds[i]);
 608                 close(dev->kickfds[i]);
 609         }
 610
 611         if (dev->vhostfd >= 0)
 612                 close(dev->vhostfd);
 613
 614         if (dev->is_server && dev->listenfd >= 0) {
 615                 close(dev->listenfd);
 616                 dev->listenfd = -1;
 617         }
 618
 619         if (dev->vhostfds) {
 620                 for (i = 0; i < dev->max_queue_pairs; ++i) {
 621                         close(dev->vhostfds[i]);
 622                         if (dev->tapfds[i] >= 0)
 623                                 close(dev->tapfds[i]);
 624                 }
 625                 free(dev->vhostfds);
 626                 free(dev->tapfds);
 627         }
 628
 629         free(dev->ifname);
 630
 631         if (dev->is_server)
 632                 unlink(dev->path);
 633 }
 634
 635 uint8_t
 636 virtio_user_handle_mq(struct virtio_user_dev *dev, uint16_t q_pairs)
 637 {
 638         uint16_t i;
 639         uint8_t ret = 0;
 640
 641         if (q_pairs > dev->max_queue_pairs) {
 642                 PMD_INIT_LOG(ERR, "multi-q config %u, but only %u supported",
 643                              q_pairs, dev->max_queue_pairs);
 644                 return -1;
 645         }
 646
 647         /* Server mode can't enable queue pairs if vhostfd is invalid,
 648          * always return 0 in this case.
 649          */
 650         if (!dev->is_server || dev->vhostfd >= 0) {
 651                 for (i = 0; i < q_pairs; ++i)
 652                         ret |= dev->ops->enable_qp(dev, i, 1);
 653                 for (i = q_pairs; i < dev->max_queue_pairs; ++i)
 654                         ret |= dev->ops->enable_qp(dev, i, 0);
 655         }
 656         dev->queue_pairs = q_pairs;
 657
 658         return ret;
 659 }
 660
 661 static uint32_t
 662 virtio_user_handle_ctrl_msg(struct virtio_user_dev *dev, struct vring *vring,
 663                             uint16_t idx_hdr)
 664 {
 665         struct virtio_net_ctrl_hdr *hdr;
 666         virtio_net_ctrl_ack status = ~0;
 667         uint16_t i, idx_data, idx_status;
 668         uint32_t n_descs = 0;
 669
 670         /* locate desc for header, data, and status */
 671         idx_data = vring->desc[idx_hdr].next;
 672         n_descs++;
 673
 674         i = idx_data;
 675         while (vring->desc[i].flags == VRING_DESC_F_NEXT) {
 676                 i = vring->desc[i].next;
 677                 n_descs++;
 678         }
 679
 680         /* locate desc for status */
 681         idx_status = i;
 682         n_descs++;
 683
 684         hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr;
 685         if (hdr->class == VIRTIO_NET_CTRL_MQ &&
 686             hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
 687                 uint16_t queues;
 688
 689                 queues = *(uint16_t *)(uintptr_t)vring->desc[idx_data].addr;
 690                 status = virtio_user_handle_mq(dev, queues);
 691         } else if (hdr->class == VIRTIO_NET_CTRL_RX  ||
 692                    hdr->class == VIRTIO_NET_CTRL_MAC ||
 693                    hdr->class == VIRTIO_NET_CTRL_VLAN) {
 694                 status = 0;
 695         }
 696
 697         /* Update status */
 698         *(virtio_net_ctrl_ack *)(uintptr_t)vring->desc[idx_status].addr = status;
 699
 700         return n_descs;
 701 }
 702
 703 static inline int
 704 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
 705 {
 706         uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
 707
 708         return wrap_counter == !!(flags & VRING_PACKED_DESC_F_AVAIL) &&
 709                 wrap_counter != !!(flags & VRING_PACKED_DESC_F_USED);
 710 }
 711
 712 static uint32_t
 713 virtio_user_handle_ctrl_msg_packed(struct virtio_user_dev *dev,
 714                                    struct vring_packed *vring,
 715                                    uint16_t idx_hdr)
 716 {
 717         struct virtio_net_ctrl_hdr *hdr;
 718         virtio_net_ctrl_ack status = ~0;
 719         uint16_t idx_data, idx_status;
 720         /* initialize to one, header is first */
 721         uint32_t n_descs = 1;
 722
 723         /* locate desc for header, data, and status */
 724         idx_data = idx_hdr + 1;
 725         if (idx_data >= dev->queue_size)
 726                 idx_data -= dev->queue_size;
 727
 728         n_descs++;
 729
 730         idx_status = idx_data;
 731         while (vring->desc[idx_status].flags & VRING_DESC_F_NEXT) {
 732                 idx_status++;
 733                 if (idx_status >= dev->queue_size)
 734                         idx_status -= dev->queue_size;
 735                 n_descs++;
 736         }
 737
 738         hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr;
 739         if (hdr->class == VIRTIO_NET_CTRL_MQ &&
 740             hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
 741                 uint16_t queues;
 742
 743                 queues = *(uint16_t *)(uintptr_t)
 744                                 vring->desc[idx_data].addr;
 745                 status = virtio_user_handle_mq(dev, queues);
 746         } else if (hdr->class == VIRTIO_NET_CTRL_RX  ||
 747                    hdr->class == VIRTIO_NET_CTRL_MAC ||
 748                    hdr->class == VIRTIO_NET_CTRL_VLAN) {
 749                 status = 0;
 750         }
 751
 752         /* Update status */
 753         *(virtio_net_ctrl_ack *)(uintptr_t)
 754                 vring->desc[idx_status].addr = status;
 755
 756         /* Update used descriptor */
 757         vring->desc[idx_hdr].id = vring->desc[idx_status].id;
 758         vring->desc[idx_hdr].len = sizeof(status);
 759
 760         return n_descs;
 761 }
 762
 763 void
 764 virtio_user_handle_cq_packed(struct virtio_user_dev *dev, uint16_t queue_idx)
 765 {
 766         struct virtio_user_queue *vq = &dev->packed_queues[queue_idx];
 767         struct vring_packed *vring = &dev->packed_vrings[queue_idx];
 768         uint16_t n_descs, flags;
 769
 770         /* Perform a load-acquire barrier in desc_is_avail to
 771          * enforce the ordering between desc flags and desc
 772          * content.
 773          */
 774         while (desc_is_avail(&vring->desc[vq->used_idx],
 775                              vq->used_wrap_counter)) {
 776
 777                 n_descs = virtio_user_handle_ctrl_msg_packed(dev, vring,
 778                                 vq->used_idx);
 779
 780                 flags = VRING_DESC_F_WRITE;
 781                 if (vq->used_wrap_counter)
 782                         flags |= VRING_PACKED_DESC_F_AVAIL_USED;
 783
 784                 __atomic_store_n(&vring->desc[vq->used_idx].flags, flags,
 785                                  __ATOMIC_RELEASE);
 786
 787                 vq->used_idx += n_descs;
 788                 if (vq->used_idx >= dev->queue_size) {
 789                         vq->used_idx -= dev->queue_size;
 790                         vq->used_wrap_counter ^= 1;
 791                 }
 792         }
 793 }
 794
 795 void
 796 virtio_user_handle_cq(struct virtio_user_dev *dev, uint16_t queue_idx)
 797 {
 798         uint16_t avail_idx, desc_idx;
 799         struct vring_used_elem *uep;
 800         uint32_t n_descs;
 801         struct vring *vring = &dev->vrings[queue_idx];
 802
 803         /* Consume avail ring, using used ring idx as first one */
 804         while (__atomic_load_n(&vring->used->idx, __ATOMIC_RELAXED)
 805                != vring->avail->idx) {
 806                 avail_idx = __atomic_load_n(&vring->used->idx, __ATOMIC_RELAXED)
 807                             & (vring->num - 1);
 808                 desc_idx = vring->avail->ring[avail_idx];
 809
 810                 n_descs = virtio_user_handle_ctrl_msg(dev, vring, desc_idx);
 811
 812                 /* Update used ring */
 813                 uep = &vring->used->ring[avail_idx];
 814                 uep->id = desc_idx;
 815                 uep->len = n_descs;
 816
 817                 __atomic_add_fetch(&vring->used->idx, 1, __ATOMIC_RELAXED);
 818         }
 819 }
 820
 821 int
 822 virtio_user_dev_set_status(struct virtio_user_dev *dev, uint8_t status)
 823 {
 824         int ret;
 825         uint64_t arg = status;
 826
 827         pthread_mutex_lock(&dev->mutex);
 828         dev->status = status;
 829         if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER)
 830                 ret = dev->ops->send_request(dev,
 831                                 VHOST_USER_SET_STATUS, &arg);
 832         else if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_VDPA)
 833                 ret = dev->ops->send_request(dev,
 834                                 VHOST_USER_SET_STATUS, &status);
 835         else
 836                 ret = -ENOTSUP;
 837
 838         if (ret && ret != -ENOTSUP) {
 839                 PMD_INIT_LOG(ERR, "VHOST_USER_SET_STATUS failed (%d): %s", ret,
 840                              strerror(errno));
 841         }
 842
 843         pthread_mutex_unlock(&dev->mutex);
 844         return ret;
 845 }
 846
 847 int
 848 virtio_user_dev_update_status(struct virtio_user_dev *dev)
 849 {
 850         uint64_t ret;
 851         uint8_t status;
 852         int err;
 853
 854         pthread_mutex_lock(&dev->mutex);
 855         if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER) {
 856                 err = dev->ops->send_request(dev, VHOST_USER_GET_STATUS, &ret);
 857                 if (!err && ret > UINT8_MAX) {
 858                         PMD_INIT_LOG(ERR, "Invalid VHOST_USER_GET_STATUS "
 859                                         "response 0x%" PRIx64 "\n", ret);
 860                         err = -1;
 861                         goto error;
 862                 }
 863
 864                 status = ret;
 865         } else if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_VDPA) {
 866                 err = dev->ops->send_request(dev, VHOST_USER_GET_STATUS,
 867                                 &status);
 868         } else {
 869                 err = -ENOTSUP;
 870         }
 871
 872         if (!err) {
 873                 dev->status = status;
 874                 PMD_INIT_LOG(DEBUG, "Updated Device Status(0x%08x):\n"
 875                         "\t-RESET: %u\n"
 876                         "\t-ACKNOWLEDGE: %u\n"
 877                         "\t-DRIVER: %u\n"
 878                         "\t-DRIVER_OK: %u\n"
 879                         "\t-FEATURES_OK: %u\n"
 880                         "\t-DEVICE_NEED_RESET: %u\n"
 881                         "\t-FAILED: %u\n",
 882                         dev->status,
 883                         (dev->status == VIRTIO_CONFIG_STATUS_RESET),
 884                         !!(dev->status & VIRTIO_CONFIG_STATUS_ACK),
 885                         !!(dev->status & VIRTIO_CONFIG_STATUS_DRIVER),
 886                         !!(dev->status & VIRTIO_CONFIG_STATUS_DRIVER_OK),
 887                         !!(dev->status & VIRTIO_CONFIG_STATUS_FEATURES_OK),
 888                         !!(dev->status & VIRTIO_CONFIG_STATUS_DEV_NEED_RESET),
 889                         !!(dev->status & VIRTIO_CONFIG_STATUS_FAILED));
 890         } else if (err != -ENOTSUP) {
 891                 PMD_INIT_LOG(ERR, "VHOST_USER_GET_STATUS failed (%d): %s", err,
 892                              strerror(errno));
 893         }
 894
 895 error:
 896         pthread_mutex_unlock(&dev->mutex);
 897         return err;
 898 }