drivers/net/vhost/rte_eth_vhost.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright (c) 2016 IGEL Co., Ltd.
   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 IGEL Co.,Ltd. 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 #include <unistd.h>
  34 #include <pthread.h>
  35 #include <stdbool.h>
  36 #ifdef RTE_LIBRTE_VHOST_NUMA
  37 #include <numaif.h>
  38 #endif
  39
  40 #include <rte_mbuf.h>
  41 #include <rte_ethdev.h>
  42 #include <rte_malloc.h>
  43 #include <rte_memcpy.h>
  44 #include <rte_dev.h>
  45 #include <rte_kvargs.h>
  46 #include <rte_virtio_net.h>
  47 #include <rte_spinlock.h>
  48
  49 #include "rte_eth_vhost.h"
  50
  51 #define ETH_VHOST_IFACE_ARG             "iface"
  52 #define ETH_VHOST_QUEUES_ARG            "queues"
  53
  54 static const char *drivername = "VHOST PMD";
  55
  56 static const char *valid_arguments[] = {
  57         ETH_VHOST_IFACE_ARG,
  58         ETH_VHOST_QUEUES_ARG,
  59         NULL
  60 };
  61
  62 static struct ether_addr base_eth_addr = {
  63         .addr_bytes = {
  64                 0x56 /* V */,
  65                 0x48 /* H */,
  66                 0x4F /* O */,
  67                 0x53 /* S */,
  68                 0x54 /* T */,
  69                 0x00
  70         }
  71 };
  72
  73 struct vhost_queue {
  74         rte_atomic32_t allow_queuing;
  75         rte_atomic32_t while_queuing;
  76         struct virtio_net *device;
  77         struct pmd_internal *internal;
  78         struct rte_mempool *mb_pool;
  79         uint8_t port;
  80         uint16_t virtqueue_id;
  81         uint64_t rx_pkts;
  82         uint64_t tx_pkts;
  83         uint64_t missed_pkts;
  84         uint64_t rx_bytes;
  85         uint64_t tx_bytes;
  86 };
  87
  88 struct pmd_internal {
  89         char *dev_name;
  90         char *iface_name;
  91         uint16_t max_queues;
  92
  93         volatile uint16_t once;
  94 };
  95
  96 struct internal_list {
  97         TAILQ_ENTRY(internal_list) next;
  98         struct rte_eth_dev *eth_dev;
  99 };
 100
 101 TAILQ_HEAD(internal_list_head, internal_list);
 102 static struct internal_list_head internal_list =
 103         TAILQ_HEAD_INITIALIZER(internal_list);
 104
 105 static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
 106
 107 static rte_atomic16_t nb_started_ports;
 108 static pthread_t session_th;
 109
 110 static struct rte_eth_link pmd_link = {
 111                 .link_speed = 10000,
 112                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 113                 .link_status = ETH_LINK_DOWN
 114 };
 115
 116 struct rte_vhost_vring_state {
 117         rte_spinlock_t lock;
 118
 119         bool cur[RTE_MAX_QUEUES_PER_PORT * 2];
 120         bool seen[RTE_MAX_QUEUES_PER_PORT * 2];
 121         unsigned int index;
 122         unsigned int max_vring;
 123 };
 124
 125 static struct rte_vhost_vring_state *vring_states[RTE_MAX_ETHPORTS];
 126
 127 static uint16_t
 128 eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 129 {
 130         struct vhost_queue *r = q;
 131         uint16_t i, nb_rx = 0;
 132
 133         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 134                 return 0;
 135
 136         rte_atomic32_set(&r->while_queuing, 1);
 137
 138         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 139                 goto out;
 140
 141         /* Dequeue packets from guest TX queue */
 142         nb_rx = rte_vhost_dequeue_burst(r->device,
 143                         r->virtqueue_id, r->mb_pool, bufs, nb_bufs);
 144
 145         r->rx_pkts += nb_rx;
 146
 147         for (i = 0; likely(i < nb_rx); i++) {
 148                 bufs[i]->port = r->port;
 149                 r->rx_bytes += bufs[i]->pkt_len;
 150         }
 151
 152 out:
 153         rte_atomic32_set(&r->while_queuing, 0);
 154
 155         return nb_rx;
 156 }
 157
 158 static uint16_t
 159 eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 160 {
 161         struct vhost_queue *r = q;
 162         uint16_t i, nb_tx = 0;
 163
 164         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 165                 return 0;
 166
 167         rte_atomic32_set(&r->while_queuing, 1);
 168
 169         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 170                 goto out;
 171
 172         /* Enqueue packets to guest RX queue */
 173         nb_tx = rte_vhost_enqueue_burst(r->device,
 174                         r->virtqueue_id, bufs, nb_bufs);
 175
 176         r->tx_pkts += nb_tx;
 177         r->missed_pkts += nb_bufs - nb_tx;
 178
 179         for (i = 0; likely(i < nb_tx); i++)
 180                 r->tx_bytes += bufs[i]->pkt_len;
 181
 182         for (i = 0; likely(i < nb_tx); i++)
 183                 rte_pktmbuf_free(bufs[i]);
 184 out:
 185         rte_atomic32_set(&r->while_queuing, 0);
 186
 187         return nb_tx;
 188 }
 189
 190 static int
 191 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 192 {
 193         return 0;
 194 }
 195
 196 static inline struct internal_list *
 197 find_internal_resource(char *ifname)
 198 {
 199         int found = 0;
 200         struct internal_list *list;
 201         struct pmd_internal *internal;
 202
 203         if (ifname == NULL)
 204                 return NULL;
 205
 206         pthread_mutex_lock(&internal_list_lock);
 207
 208         TAILQ_FOREACH(list, &internal_list, next) {
 209                 internal = list->eth_dev->data->dev_private;
 210                 if (!strcmp(internal->iface_name, ifname)) {
 211                         found = 1;
 212                         break;
 213                 }
 214         }
 215
 216         pthread_mutex_unlock(&internal_list_lock);
 217
 218         if (!found)
 219                 return NULL;
 220
 221         return list;
 222 }
 223
 224 static int
 225 new_device(struct virtio_net *dev)
 226 {
 227         struct rte_eth_dev *eth_dev;
 228         struct internal_list *list;
 229         struct pmd_internal *internal;
 230         struct vhost_queue *vq;
 231         unsigned i;
 232         char ifname[PATH_MAX];
 233 #ifdef RTE_LIBRTE_VHOST_NUMA
 234         int newnode;
 235 #endif
 236
 237         if (dev == NULL) {
 238                 RTE_LOG(INFO, PMD, "Invalid argument\n");
 239                 return -1;
 240         }
 241
 242         rte_vhost_get_ifname(dev->vid, ifname, sizeof(ifname));
 243         list = find_internal_resource(ifname);
 244         if (list == NULL) {
 245                 RTE_LOG(INFO, PMD, "Invalid device name: %s\n", ifname);
 246                 return -1;
 247         }
 248
 249         eth_dev = list->eth_dev;
 250         internal = eth_dev->data->dev_private;
 251
 252 #ifdef RTE_LIBRTE_VHOST_NUMA
 253         newnode = rte_vhost_get_numa_node(dev->vid);
 254         if (newnode >= 0)
 255                 eth_dev->data->numa_node = newnode;
 256 #endif
 257
 258         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 259                 vq = eth_dev->data->rx_queues[i];
 260                 if (vq == NULL)
 261                         continue;
 262                 vq->device = dev;
 263                 vq->internal = internal;
 264                 vq->port = eth_dev->data->port_id;
 265         }
 266         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 267                 vq = eth_dev->data->tx_queues[i];
 268                 if (vq == NULL)
 269                         continue;
 270                 vq->device = dev;
 271                 vq->internal = internal;
 272                 vq->port = eth_dev->data->port_id;
 273         }
 274
 275         for (i = 0; i < rte_vhost_get_queue_num(dev->vid) * VIRTIO_QNUM; i++)
 276                 rte_vhost_enable_guest_notification(dev, i, 0);
 277
 278         eth_dev->data->dev_link.link_status = ETH_LINK_UP;
 279
 280         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 281                 vq = eth_dev->data->rx_queues[i];
 282                 if (vq == NULL)
 283                         continue;
 284                 rte_atomic32_set(&vq->allow_queuing, 1);
 285         }
 286         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 287                 vq = eth_dev->data->tx_queues[i];
 288                 if (vq == NULL)
 289                         continue;
 290                 rte_atomic32_set(&vq->allow_queuing, 1);
 291         }
 292
 293         RTE_LOG(INFO, PMD, "New connection established\n");
 294
 295         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC);
 296
 297         return 0;
 298 }
 299
 300 static void
 301 destroy_device(volatile struct virtio_net *dev)
 302 {
 303         struct rte_eth_dev *eth_dev;
 304         struct vhost_queue *vq;
 305         struct internal_list *list;
 306         char ifname[PATH_MAX];
 307         unsigned i;
 308
 309         if (dev == NULL) {
 310                 RTE_LOG(INFO, PMD, "Invalid argument\n");
 311                 return;
 312         }
 313
 314         rte_vhost_get_ifname(dev->vid, ifname, sizeof(ifname));
 315         list = find_internal_resource(ifname);
 316         if (list == NULL) {
 317                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
 318                 return;
 319         }
 320         eth_dev = list->eth_dev;
 321
 322         /* Wait until rx/tx_pkt_burst stops accessing vhost device */
 323         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 324                 vq = eth_dev->data->rx_queues[i];
 325                 if (vq == NULL)
 326                         continue;
 327                 rte_atomic32_set(&vq->allow_queuing, 0);
 328                 while (rte_atomic32_read(&vq->while_queuing))
 329                         rte_pause();
 330         }
 331         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 332                 vq = eth_dev->data->tx_queues[i];
 333                 if (vq == NULL)
 334                         continue;
 335                 rte_atomic32_set(&vq->allow_queuing, 0);
 336                 while (rte_atomic32_read(&vq->while_queuing))
 337                         rte_pause();
 338         }
 339
 340         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
 341
 342         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 343                 vq = eth_dev->data->rx_queues[i];
 344                 if (vq == NULL)
 345                         continue;
 346                 vq->device = NULL;
 347         }
 348         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 349                 vq = eth_dev->data->tx_queues[i];
 350                 if (vq == NULL)
 351                         continue;
 352                 vq->device = NULL;
 353         }
 354
 355         RTE_LOG(INFO, PMD, "Connection closed\n");
 356
 357         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC);
 358 }
 359
 360 static int
 361 vring_state_changed(struct virtio_net *dev, uint16_t vring, int enable)
 362 {
 363         struct rte_vhost_vring_state *state;
 364         struct rte_eth_dev *eth_dev;
 365         struct internal_list *list;
 366         char ifname[PATH_MAX];
 367
 368         if (dev == NULL) {
 369                 RTE_LOG(ERR, PMD, "Invalid argument\n");
 370                 return -1;
 371         }
 372
 373         rte_vhost_get_ifname(dev->vid, ifname, sizeof(ifname));
 374         list = find_internal_resource(ifname);
 375         if (list == NULL) {
 376                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
 377                 return -1;
 378         }
 379
 380         eth_dev = list->eth_dev;
 381         /* won't be NULL */
 382         state = vring_states[eth_dev->data->port_id];
 383         rte_spinlock_lock(&state->lock);
 384         state->cur[vring] = enable;
 385         state->max_vring = RTE_MAX(vring, state->max_vring);
 386         rte_spinlock_unlock(&state->lock);
 387
 388         RTE_LOG(INFO, PMD, "vring%u is %s\n",
 389                         vring, enable ? "enabled" : "disabled");
 390
 391         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE);
 392
 393         return 0;
 394 }
 395
 396 int
 397 rte_eth_vhost_get_queue_event(uint8_t port_id,
 398                 struct rte_eth_vhost_queue_event *event)
 399 {
 400         struct rte_vhost_vring_state *state;
 401         unsigned int i;
 402         int idx;
 403
 404         if (port_id >= RTE_MAX_ETHPORTS) {
 405                 RTE_LOG(ERR, PMD, "Invalid port id\n");
 406                 return -1;
 407         }
 408
 409         state = vring_states[port_id];
 410         if (!state) {
 411                 RTE_LOG(ERR, PMD, "Unused port\n");
 412                 return -1;
 413         }
 414
 415         rte_spinlock_lock(&state->lock);
 416         for (i = 0; i <= state->max_vring; i++) {
 417                 idx = state->index++ % (state->max_vring + 1);
 418
 419                 if (state->cur[idx] != state->seen[idx]) {
 420                         state->seen[idx] = state->cur[idx];
 421                         event->queue_id = idx / 2;
 422                         event->rx = idx & 1;
 423                         event->enable = state->cur[idx];
 424                         rte_spinlock_unlock(&state->lock);
 425                         return 0;
 426                 }
 427         }
 428         rte_spinlock_unlock(&state->lock);
 429
 430         return -1;
 431 }
 432
 433 static void *
 434 vhost_driver_session(void *param __rte_unused)
 435 {
 436         static struct virtio_net_device_ops vhost_ops;
 437
 438         /* set vhost arguments */
 439         vhost_ops.new_device = new_device;
 440         vhost_ops.destroy_device = destroy_device;
 441         vhost_ops.vring_state_changed = vring_state_changed;
 442         if (rte_vhost_driver_callback_register(&vhost_ops) < 0)
 443                 RTE_LOG(ERR, PMD, "Can't register callbacks\n");
 444
 445         /* start event handling */
 446         rte_vhost_driver_session_start();
 447
 448         return NULL;
 449 }
 450
 451 static int
 452 vhost_driver_session_start(void)
 453 {
 454         int ret;
 455
 456         ret = pthread_create(&session_th,
 457                         NULL, vhost_driver_session, NULL);
 458         if (ret)
 459                 RTE_LOG(ERR, PMD, "Can't create a thread\n");
 460
 461         return ret;
 462 }
 463
 464 static void
 465 vhost_driver_session_stop(void)
 466 {
 467         int ret;
 468
 469         ret = pthread_cancel(session_th);
 470         if (ret)
 471                 RTE_LOG(ERR, PMD, "Can't cancel the thread\n");
 472
 473         ret = pthread_join(session_th, NULL);
 474         if (ret)
 475                 RTE_LOG(ERR, PMD, "Can't join the thread\n");
 476 }
 477
 478 static int
 479 eth_dev_start(struct rte_eth_dev *dev)
 480 {
 481         struct pmd_internal *internal = dev->data->dev_private;
 482         int ret = 0;
 483
 484         if (rte_atomic16_cmpset(&internal->once, 0, 1)) {
 485                 ret = rte_vhost_driver_register(internal->iface_name);
 486                 if (ret)
 487                         return ret;
 488         }
 489
 490         /* We need only one message handling thread */
 491         if (rte_atomic16_add_return(&nb_started_ports, 1) == 1)
 492                 ret = vhost_driver_session_start();
 493
 494         return ret;
 495 }
 496
 497 static void
 498 eth_dev_stop(struct rte_eth_dev *dev)
 499 {
 500         struct pmd_internal *internal = dev->data->dev_private;
 501
 502         if (rte_atomic16_cmpset(&internal->once, 1, 0))
 503                 rte_vhost_driver_unregister(internal->iface_name);
 504
 505         if (rte_atomic16_sub_return(&nb_started_ports, 1) == 0)
 506                 vhost_driver_session_stop();
 507 }
 508
 509 static int
 510 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 511                    uint16_t nb_rx_desc __rte_unused,
 512                    unsigned int socket_id,
 513                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 514                    struct rte_mempool *mb_pool)
 515 {
 516         struct vhost_queue *vq;
 517
 518         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
 519                         RTE_CACHE_LINE_SIZE, socket_id);
 520         if (vq == NULL) {
 521                 RTE_LOG(ERR, PMD, "Failed to allocate memory for rx queue\n");
 522                 return -ENOMEM;
 523         }
 524
 525         vq->mb_pool = mb_pool;
 526         vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
 527         dev->data->rx_queues[rx_queue_id] = vq;
 528
 529         return 0;
 530 }
 531
 532 static int
 533 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 534                    uint16_t nb_tx_desc __rte_unused,
 535                    unsigned int socket_id,
 536                    const struct rte_eth_txconf *tx_conf __rte_unused)
 537 {
 538         struct vhost_queue *vq;
 539
 540         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
 541                         RTE_CACHE_LINE_SIZE, socket_id);
 542         if (vq == NULL) {
 543                 RTE_LOG(ERR, PMD, "Failed to allocate memory for tx queue\n");
 544                 return -ENOMEM;
 545         }
 546
 547         vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
 548         dev->data->tx_queues[tx_queue_id] = vq;
 549
 550         return 0;
 551 }
 552
 553 static void
 554 eth_dev_info(struct rte_eth_dev *dev,
 555              struct rte_eth_dev_info *dev_info)
 556 {
 557         struct pmd_internal *internal;
 558
 559         internal = dev->data->dev_private;
 560         if (internal == NULL) {
 561                 RTE_LOG(ERR, PMD, "Invalid device specified\n");
 562                 return;
 563         }
 564
 565         dev_info->driver_name = drivername;
 566         dev_info->max_mac_addrs = 1;
 567         dev_info->max_rx_pktlen = (uint32_t)-1;
 568         dev_info->max_rx_queues = internal->max_queues;
 569         dev_info->max_tx_queues = internal->max_queues;
 570         dev_info->min_rx_bufsize = 0;
 571 }
 572
 573 static void
 574 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 575 {
 576         unsigned i;
 577         unsigned long rx_total = 0, tx_total = 0, tx_missed_total = 0;
 578         unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
 579         struct vhost_queue *vq;
 580
 581         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 582                         i < dev->data->nb_rx_queues; i++) {
 583                 if (dev->data->rx_queues[i] == NULL)
 584                         continue;
 585                 vq = dev->data->rx_queues[i];
 586                 stats->q_ipackets[i] = vq->rx_pkts;
 587                 rx_total += stats->q_ipackets[i];
 588
 589                 stats->q_ibytes[i] = vq->rx_bytes;
 590                 rx_total_bytes += stats->q_ibytes[i];
 591         }
 592
 593         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 594                         i < dev->data->nb_tx_queues; i++) {
 595                 if (dev->data->tx_queues[i] == NULL)
 596                         continue;
 597                 vq = dev->data->tx_queues[i];
 598                 stats->q_opackets[i] = vq->tx_pkts;
 599                 tx_missed_total += vq->missed_pkts;
 600                 tx_total += stats->q_opackets[i];
 601
 602                 stats->q_obytes[i] = vq->tx_bytes;
 603                 tx_total_bytes += stats->q_obytes[i];
 604         }
 605
 606         stats->ipackets = rx_total;
 607         stats->opackets = tx_total;
 608         stats->imissed = tx_missed_total;
 609         stats->ibytes = rx_total_bytes;
 610         stats->obytes = tx_total_bytes;
 611 }
 612
 613 static void
 614 eth_stats_reset(struct rte_eth_dev *dev)
 615 {
 616         struct vhost_queue *vq;
 617         unsigned i;
 618
 619         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 620                 if (dev->data->rx_queues[i] == NULL)
 621                         continue;
 622                 vq = dev->data->rx_queues[i];
 623                 vq->rx_pkts = 0;
 624                 vq->rx_bytes = 0;
 625         }
 626         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 627                 if (dev->data->tx_queues[i] == NULL)
 628                         continue;
 629                 vq = dev->data->tx_queues[i];
 630                 vq->tx_pkts = 0;
 631                 vq->tx_bytes = 0;
 632                 vq->missed_pkts = 0;
 633         }
 634 }
 635
 636 static void
 637 eth_queue_release(void *q)
 638 {
 639         rte_free(q);
 640 }
 641
 642 static int
 643 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 644                 int wait_to_complete __rte_unused)
 645 {
 646         return 0;
 647 }
 648
 649 /**
 650  * Disable features in feature_mask. Returns 0 on success.
 651  */
 652 int
 653 rte_eth_vhost_feature_disable(uint64_t feature_mask)
 654 {
 655         return rte_vhost_feature_disable(feature_mask);
 656 }
 657
 658 /**
 659  * Enable features in feature_mask. Returns 0 on success.
 660  */
 661 int
 662 rte_eth_vhost_feature_enable(uint64_t feature_mask)
 663 {
 664         return rte_vhost_feature_enable(feature_mask);
 665 }
 666
 667 /* Returns currently supported vhost features */
 668 uint64_t
 669 rte_eth_vhost_feature_get(void)
 670 {
 671         return rte_vhost_feature_get();
 672 }
 673
 674 static const struct eth_dev_ops ops = {
 675         .dev_start = eth_dev_start,
 676         .dev_stop = eth_dev_stop,
 677         .dev_configure = eth_dev_configure,
 678         .dev_infos_get = eth_dev_info,
 679         .rx_queue_setup = eth_rx_queue_setup,
 680         .tx_queue_setup = eth_tx_queue_setup,
 681         .rx_queue_release = eth_queue_release,
 682         .tx_queue_release = eth_queue_release,
 683         .link_update = eth_link_update,
 684         .stats_get = eth_stats_get,
 685         .stats_reset = eth_stats_reset,
 686 };
 687
 688 static int
 689 eth_dev_vhost_create(const char *name, char *iface_name, int16_t queues,
 690                      const unsigned numa_node)
 691 {
 692         struct rte_eth_dev_data *data = NULL;
 693         struct pmd_internal *internal = NULL;
 694         struct rte_eth_dev *eth_dev = NULL;
 695         struct ether_addr *eth_addr = NULL;
 696         struct rte_vhost_vring_state *vring_state = NULL;
 697         struct internal_list *list = NULL;
 698
 699         RTE_LOG(INFO, PMD, "Creating VHOST-USER backend on numa socket %u\n",
 700                 numa_node);
 701
 702         /* now do all data allocation - for eth_dev structure, dummy pci driver
 703          * and internal (private) data
 704          */
 705         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
 706         if (data == NULL)
 707                 goto error;
 708
 709         internal = rte_zmalloc_socket(name, sizeof(*internal), 0, numa_node);
 710         if (internal == NULL)
 711                 goto error;
 712
 713         list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
 714         if (list == NULL)
 715                 goto error;
 716
 717         /* reserve an ethdev entry */
 718         eth_dev = rte_eth_dev_allocate(name, RTE_ETH_DEV_VIRTUAL);
 719         if (eth_dev == NULL)
 720                 goto error;
 721
 722         eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
 723         if (eth_addr == NULL)
 724                 goto error;
 725         *eth_addr = base_eth_addr;
 726         eth_addr->addr_bytes[5] = eth_dev->data->port_id;
 727
 728         vring_state = rte_zmalloc_socket(name,
 729                         sizeof(*vring_state), 0, numa_node);
 730         if (vring_state == NULL)
 731                 goto error;
 732
 733         TAILQ_INIT(&eth_dev->link_intr_cbs);
 734
 735         /* now put it all together
 736          * - store queue data in internal,
 737          * - store numa_node info in ethdev data
 738          * - point eth_dev_data to internals
 739          * - and point eth_dev structure to new eth_dev_data structure
 740          */
 741         internal->dev_name = strdup(name);
 742         if (internal->dev_name == NULL)
 743                 goto error;
 744         internal->iface_name = strdup(iface_name);
 745         if (internal->iface_name == NULL)
 746                 goto error;
 747
 748         list->eth_dev = eth_dev;
 749         pthread_mutex_lock(&internal_list_lock);
 750         TAILQ_INSERT_TAIL(&internal_list, list, next);
 751         pthread_mutex_unlock(&internal_list_lock);
 752
 753         rte_spinlock_init(&vring_state->lock);
 754         vring_states[eth_dev->data->port_id] = vring_state;
 755
 756         data->dev_private = internal;
 757         data->port_id = eth_dev->data->port_id;
 758         memmove(data->name, eth_dev->data->name, sizeof(data->name));
 759         data->nb_rx_queues = queues;
 760         data->nb_tx_queues = queues;
 761         internal->max_queues = queues;
 762         data->dev_link = pmd_link;
 763         data->mac_addrs = eth_addr;
 764
 765         /* We'll replace the 'data' originally allocated by eth_dev. So the
 766          * vhost PMD resources won't be shared between multi processes.
 767          */
 768         eth_dev->data = data;
 769         eth_dev->dev_ops = &ops;
 770         eth_dev->driver = NULL;
 771         data->dev_flags =
 772                 RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
 773         data->kdrv = RTE_KDRV_NONE;
 774         data->drv_name = internal->dev_name;
 775         data->numa_node = numa_node;
 776
 777         /* finally assign rx and tx ops */
 778         eth_dev->rx_pkt_burst = eth_vhost_rx;
 779         eth_dev->tx_pkt_burst = eth_vhost_tx;
 780
 781         return data->port_id;
 782
 783 error:
 784         if (internal)
 785                 free(internal->dev_name);
 786         rte_free(vring_state);
 787         rte_free(eth_addr);
 788         if (eth_dev)
 789                 rte_eth_dev_release_port(eth_dev);
 790         rte_free(internal);
 791         rte_free(list);
 792         rte_free(data);
 793
 794         return -1;
 795 }
 796
 797 static inline int
 798 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
 799 {
 800         const char **iface_name = extra_args;
 801
 802         if (value == NULL)
 803                 return -1;
 804
 805         *iface_name = value;
 806
 807         return 0;
 808 }
 809
 810 static inline int
 811 open_queues(const char *key __rte_unused, const char *value, void *extra_args)
 812 {
 813         uint16_t *q = extra_args;
 814
 815         if (value == NULL || extra_args == NULL)
 816                 return -EINVAL;
 817
 818         *q = (uint16_t)strtoul(value, NULL, 0);
 819         if (*q == USHRT_MAX && errno == ERANGE)
 820                 return -1;
 821
 822         if (*q > RTE_MAX_QUEUES_PER_PORT)
 823                 return -1;
 824
 825         return 0;
 826 }
 827
 828 static int
 829 rte_pmd_vhost_devinit(const char *name, const char *params)
 830 {
 831         struct rte_kvargs *kvlist = NULL;
 832         int ret = 0;
 833         char *iface_name;
 834         uint16_t queues;
 835
 836         RTE_LOG(INFO, PMD, "Initializing pmd_vhost for %s\n", name);
 837
 838         kvlist = rte_kvargs_parse(params, valid_arguments);
 839         if (kvlist == NULL)
 840                 return -1;
 841
 842         if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
 843                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
 844                                          &open_iface, &iface_name);
 845                 if (ret < 0)
 846                         goto out_free;
 847         } else {
 848                 ret = -1;
 849                 goto out_free;
 850         }
 851
 852         if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
 853                 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
 854                                          &open_queues, &queues);
 855                 if (ret < 0)
 856                         goto out_free;
 857
 858         } else
 859                 queues = 1;
 860
 861         eth_dev_vhost_create(name, iface_name, queues, rte_socket_id());
 862
 863 out_free:
 864         rte_kvargs_free(kvlist);
 865         return ret;
 866 }
 867
 868 static int
 869 rte_pmd_vhost_devuninit(const char *name)
 870 {
 871         struct rte_eth_dev *eth_dev = NULL;
 872         struct pmd_internal *internal;
 873         struct internal_list *list;
 874         unsigned int i;
 875
 876         RTE_LOG(INFO, PMD, "Un-Initializing pmd_vhost for %s\n", name);
 877
 878         /* find an ethdev entry */
 879         eth_dev = rte_eth_dev_allocated(name);
 880         if (eth_dev == NULL)
 881                 return -ENODEV;
 882
 883         internal = eth_dev->data->dev_private;
 884         if (internal == NULL)
 885                 return -ENODEV;
 886
 887         list = find_internal_resource(internal->iface_name);
 888         if (list == NULL)
 889                 return -ENODEV;
 890
 891         pthread_mutex_lock(&internal_list_lock);
 892         TAILQ_REMOVE(&internal_list, list, next);
 893         pthread_mutex_unlock(&internal_list_lock);
 894         rte_free(list);
 895
 896         eth_dev_stop(eth_dev);
 897
 898         rte_free(vring_states[eth_dev->data->port_id]);
 899         vring_states[eth_dev->data->port_id] = NULL;
 900
 901         free(internal->dev_name);
 902         free(internal->iface_name);
 903
 904         for (i = 0; i < eth_dev->data->nb_rx_queues; i++)
 905                 rte_free(eth_dev->data->rx_queues[i]);
 906         for (i = 0; i < eth_dev->data->nb_tx_queues; i++)
 907                 rte_free(eth_dev->data->tx_queues[i]);
 908
 909         rte_free(eth_dev->data->mac_addrs);
 910         rte_free(eth_dev->data);
 911         rte_free(internal);
 912
 913         rte_eth_dev_release_port(eth_dev);
 914
 915         return 0;
 916 }
 917
 918 static struct rte_driver pmd_vhost_drv = {
 919         .name = "eth_vhost",
 920         .type = PMD_VDEV,
 921         .init = rte_pmd_vhost_devinit,
 922         .uninit = rte_pmd_vhost_devuninit,
 923 };
 924
 925 PMD_REGISTER_DRIVER(pmd_vhost_drv);