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         dev->priv = eth_dev;
 279         eth_dev->data->dev_link.link_status = ETH_LINK_UP;
 280
 281         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 282                 vq = eth_dev->data->rx_queues[i];
 283                 if (vq == NULL)
 284                         continue;
 285                 rte_atomic32_set(&vq->allow_queuing, 1);
 286         }
 287         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 288                 vq = eth_dev->data->tx_queues[i];
 289                 if (vq == NULL)
 290                         continue;
 291                 rte_atomic32_set(&vq->allow_queuing, 1);
 292         }
 293
 294         RTE_LOG(INFO, PMD, "New connection established\n");
 295
 296         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC);
 297
 298         return 0;
 299 }
 300
 301 static void
 302 destroy_device(volatile struct virtio_net *dev)
 303 {
 304         struct rte_eth_dev *eth_dev;
 305         struct vhost_queue *vq;
 306         unsigned i;
 307
 308         if (dev == NULL) {
 309                 RTE_LOG(INFO, PMD, "Invalid argument\n");
 310                 return;
 311         }
 312
 313         eth_dev = (struct rte_eth_dev *)dev->priv;
 314         if (eth_dev == NULL) {
 315                 RTE_LOG(INFO, PMD, "Failed to find a ethdev\n");
 316                 return;
 317         }
 318
 319         /* Wait until rx/tx_pkt_burst stops accessing vhost device */
 320         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 321                 vq = eth_dev->data->rx_queues[i];
 322                 if (vq == NULL)
 323                         continue;
 324                 rte_atomic32_set(&vq->allow_queuing, 0);
 325                 while (rte_atomic32_read(&vq->while_queuing))
 326                         rte_pause();
 327         }
 328         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 329                 vq = eth_dev->data->tx_queues[i];
 330                 if (vq == NULL)
 331                         continue;
 332                 rte_atomic32_set(&vq->allow_queuing, 0);
 333                 while (rte_atomic32_read(&vq->while_queuing))
 334                         rte_pause();
 335         }
 336
 337         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
 338
 339         dev->priv = NULL;
 340
 341         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 342                 vq = eth_dev->data->rx_queues[i];
 343                 if (vq == NULL)
 344                         continue;
 345                 vq->device = NULL;
 346         }
 347         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 348                 vq = eth_dev->data->tx_queues[i];
 349                 if (vq == NULL)
 350                         continue;
 351                 vq->device = NULL;
 352         }
 353
 354         RTE_LOG(INFO, PMD, "Connection closed\n");
 355
 356         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC);
 357 }
 358
 359 static int
 360 vring_state_changed(struct virtio_net *dev, uint16_t vring, int enable)
 361 {
 362         struct rte_vhost_vring_state *state;
 363         struct rte_eth_dev *eth_dev;
 364         struct internal_list *list;
 365         char ifname[PATH_MAX];
 366
 367         if (dev == NULL) {
 368                 RTE_LOG(ERR, PMD, "Invalid argument\n");
 369                 return -1;
 370         }
 371
 372         rte_vhost_get_ifname(dev->vid, ifname, sizeof(ifname));
 373         list = find_internal_resource(ifname);
 374         if (list == NULL) {
 375                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
 376                 return -1;
 377         }
 378
 379         eth_dev = list->eth_dev;
 380         /* won't be NULL */
 381         state = vring_states[eth_dev->data->port_id];
 382         rte_spinlock_lock(&state->lock);
 383         state->cur[vring] = enable;
 384         state->max_vring = RTE_MAX(vring, state->max_vring);
 385         rte_spinlock_unlock(&state->lock);
 386
 387         RTE_LOG(INFO, PMD, "vring%u is %s\n",
 388                         vring, enable ? "enabled" : "disabled");
 389
 390         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE);
 391
 392         return 0;
 393 }
 394
 395 int
 396 rte_eth_vhost_get_queue_event(uint8_t port_id,
 397                 struct rte_eth_vhost_queue_event *event)
 398 {
 399         struct rte_vhost_vring_state *state;
 400         unsigned int i;
 401         int idx;
 402
 403         if (port_id >= RTE_MAX_ETHPORTS) {
 404                 RTE_LOG(ERR, PMD, "Invalid port id\n");
 405                 return -1;
 406         }
 407
 408         state = vring_states[port_id];
 409         if (!state) {
 410                 RTE_LOG(ERR, PMD, "Unused port\n");
 411                 return -1;
 412         }
 413
 414         rte_spinlock_lock(&state->lock);
 415         for (i = 0; i <= state->max_vring; i++) {
 416                 idx = state->index++ % (state->max_vring + 1);
 417
 418                 if (state->cur[idx] != state->seen[idx]) {
 419                         state->seen[idx] = state->cur[idx];
 420                         event->queue_id = idx / 2;
 421                         event->rx = idx & 1;
 422                         event->enable = state->cur[idx];
 423                         rte_spinlock_unlock(&state->lock);
 424                         return 0;
 425                 }
 426         }
 427         rte_spinlock_unlock(&state->lock);
 428
 429         return -1;
 430 }
 431
 432 static void *
 433 vhost_driver_session(void *param __rte_unused)
 434 {
 435         static struct virtio_net_device_ops vhost_ops;
 436
 437         /* set vhost arguments */
 438         vhost_ops.new_device = new_device;
 439         vhost_ops.destroy_device = destroy_device;
 440         vhost_ops.vring_state_changed = vring_state_changed;
 441         if (rte_vhost_driver_callback_register(&vhost_ops) < 0)
 442                 RTE_LOG(ERR, PMD, "Can't register callbacks\n");
 443
 444         /* start event handling */
 445         rte_vhost_driver_session_start();
 446
 447         return NULL;
 448 }
 449
 450 static int
 451 vhost_driver_session_start(void)
 452 {
 453         int ret;
 454
 455         ret = pthread_create(&session_th,
 456                         NULL, vhost_driver_session, NULL);
 457         if (ret)
 458                 RTE_LOG(ERR, PMD, "Can't create a thread\n");
 459
 460         return ret;
 461 }
 462
 463 static void
 464 vhost_driver_session_stop(void)
 465 {
 466         int ret;
 467
 468         ret = pthread_cancel(session_th);
 469         if (ret)
 470                 RTE_LOG(ERR, PMD, "Can't cancel the thread\n");
 471
 472         ret = pthread_join(session_th, NULL);
 473         if (ret)
 474                 RTE_LOG(ERR, PMD, "Can't join the thread\n");
 475 }
 476
 477 static int
 478 eth_dev_start(struct rte_eth_dev *dev)
 479 {
 480         struct pmd_internal *internal = dev->data->dev_private;
 481         int ret = 0;
 482
 483         if (rte_atomic16_cmpset(&internal->once, 0, 1)) {
 484                 ret = rte_vhost_driver_register(internal->iface_name);
 485                 if (ret)
 486                         return ret;
 487         }
 488
 489         /* We need only one message handling thread */
 490         if (rte_atomic16_add_return(&nb_started_ports, 1) == 1)
 491                 ret = vhost_driver_session_start();
 492
 493         return ret;
 494 }
 495
 496 static void
 497 eth_dev_stop(struct rte_eth_dev *dev)
 498 {
 499         struct pmd_internal *internal = dev->data->dev_private;
 500
 501         if (rte_atomic16_cmpset(&internal->once, 1, 0))
 502                 rte_vhost_driver_unregister(internal->iface_name);
 503
 504         if (rte_atomic16_sub_return(&nb_started_ports, 1) == 0)
 505                 vhost_driver_session_stop();
 506 }
 507
 508 static int
 509 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 510                    uint16_t nb_rx_desc __rte_unused,
 511                    unsigned int socket_id,
 512                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 513                    struct rte_mempool *mb_pool)
 514 {
 515         struct vhost_queue *vq;
 516
 517         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
 518                         RTE_CACHE_LINE_SIZE, socket_id);
 519         if (vq == NULL) {
 520                 RTE_LOG(ERR, PMD, "Failed to allocate memory for rx queue\n");
 521                 return -ENOMEM;
 522         }
 523
 524         vq->mb_pool = mb_pool;
 525         vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
 526         dev->data->rx_queues[rx_queue_id] = vq;
 527
 528         return 0;
 529 }
 530
 531 static int
 532 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 533                    uint16_t nb_tx_desc __rte_unused,
 534                    unsigned int socket_id,
 535                    const struct rte_eth_txconf *tx_conf __rte_unused)
 536 {
 537         struct vhost_queue *vq;
 538
 539         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
 540                         RTE_CACHE_LINE_SIZE, socket_id);
 541         if (vq == NULL) {
 542                 RTE_LOG(ERR, PMD, "Failed to allocate memory for tx queue\n");
 543                 return -ENOMEM;
 544         }
 545
 546         vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
 547         dev->data->tx_queues[tx_queue_id] = vq;
 548
 549         return 0;
 550 }
 551
 552 static void
 553 eth_dev_info(struct rte_eth_dev *dev,
 554              struct rte_eth_dev_info *dev_info)
 555 {
 556         struct pmd_internal *internal;
 557
 558         internal = dev->data->dev_private;
 559         if (internal == NULL) {
 560                 RTE_LOG(ERR, PMD, "Invalid device specified\n");
 561                 return;
 562         }
 563
 564         dev_info->driver_name = drivername;
 565         dev_info->max_mac_addrs = 1;
 566         dev_info->max_rx_pktlen = (uint32_t)-1;
 567         dev_info->max_rx_queues = internal->max_queues;
 568         dev_info->max_tx_queues = internal->max_queues;
 569         dev_info->min_rx_bufsize = 0;
 570 }
 571
 572 static void
 573 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 574 {
 575         unsigned i;
 576         unsigned long rx_total = 0, tx_total = 0, tx_missed_total = 0;
 577         unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
 578         struct vhost_queue *vq;
 579
 580         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 581                         i < dev->data->nb_rx_queues; i++) {
 582                 if (dev->data->rx_queues[i] == NULL)
 583                         continue;
 584                 vq = dev->data->rx_queues[i];
 585                 stats->q_ipackets[i] = vq->rx_pkts;
 586                 rx_total += stats->q_ipackets[i];
 587
 588                 stats->q_ibytes[i] = vq->rx_bytes;
 589                 rx_total_bytes += stats->q_ibytes[i];
 590         }
 591
 592         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 593                         i < dev->data->nb_tx_queues; i++) {
 594                 if (dev->data->tx_queues[i] == NULL)
 595                         continue;
 596                 vq = dev->data->tx_queues[i];
 597                 stats->q_opackets[i] = vq->tx_pkts;
 598                 tx_missed_total += vq->missed_pkts;
 599                 tx_total += stats->q_opackets[i];
 600
 601                 stats->q_obytes[i] = vq->tx_bytes;
 602                 tx_total_bytes += stats->q_obytes[i];
 603         }
 604
 605         stats->ipackets = rx_total;
 606         stats->opackets = tx_total;
 607         stats->imissed = tx_missed_total;
 608         stats->ibytes = rx_total_bytes;
 609         stats->obytes = tx_total_bytes;
 610 }
 611
 612 static void
 613 eth_stats_reset(struct rte_eth_dev *dev)
 614 {
 615         struct vhost_queue *vq;
 616         unsigned i;
 617
 618         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 619                 if (dev->data->rx_queues[i] == NULL)
 620                         continue;
 621                 vq = dev->data->rx_queues[i];
 622                 vq->rx_pkts = 0;
 623                 vq->rx_bytes = 0;
 624         }
 625         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 626                 if (dev->data->tx_queues[i] == NULL)
 627                         continue;
 628                 vq = dev->data->tx_queues[i];
 629                 vq->tx_pkts = 0;
 630                 vq->tx_bytes = 0;
 631                 vq->missed_pkts = 0;
 632         }
 633 }
 634
 635 static void
 636 eth_queue_release(void *q)
 637 {
 638         rte_free(q);
 639 }
 640
 641 static int
 642 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 643                 int wait_to_complete __rte_unused)
 644 {
 645         return 0;
 646 }
 647
 648 /**
 649  * Disable features in feature_mask. Returns 0 on success.
 650  */
 651 int
 652 rte_eth_vhost_feature_disable(uint64_t feature_mask)
 653 {
 654         return rte_vhost_feature_disable(feature_mask);
 655 }
 656
 657 /**
 658  * Enable features in feature_mask. Returns 0 on success.
 659  */
 660 int
 661 rte_eth_vhost_feature_enable(uint64_t feature_mask)
 662 {
 663         return rte_vhost_feature_enable(feature_mask);
 664 }
 665
 666 /* Returns currently supported vhost features */
 667 uint64_t
 668 rte_eth_vhost_feature_get(void)
 669 {
 670         return rte_vhost_feature_get();
 671 }
 672
 673 static const struct eth_dev_ops ops = {
 674         .dev_start = eth_dev_start,
 675         .dev_stop = eth_dev_stop,
 676         .dev_configure = eth_dev_configure,
 677         .dev_infos_get = eth_dev_info,
 678         .rx_queue_setup = eth_rx_queue_setup,
 679         .tx_queue_setup = eth_tx_queue_setup,
 680         .rx_queue_release = eth_queue_release,
 681         .tx_queue_release = eth_queue_release,
 682         .link_update = eth_link_update,
 683         .stats_get = eth_stats_get,
 684         .stats_reset = eth_stats_reset,
 685 };
 686
 687 static int
 688 eth_dev_vhost_create(const char *name, char *iface_name, int16_t queues,
 689                      const unsigned numa_node)
 690 {
 691         struct rte_eth_dev_data *data = NULL;
 692         struct pmd_internal *internal = NULL;
 693         struct rte_eth_dev *eth_dev = NULL;
 694         struct ether_addr *eth_addr = NULL;
 695         struct rte_vhost_vring_state *vring_state = NULL;
 696         struct internal_list *list = NULL;
 697
 698         RTE_LOG(INFO, PMD, "Creating VHOST-USER backend on numa socket %u\n",
 699                 numa_node);
 700
 701         /* now do all data allocation - for eth_dev structure, dummy pci driver
 702          * and internal (private) data
 703          */
 704         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
 705         if (data == NULL)
 706                 goto error;
 707
 708         internal = rte_zmalloc_socket(name, sizeof(*internal), 0, numa_node);
 709         if (internal == NULL)
 710                 goto error;
 711
 712         list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
 713         if (list == NULL)
 714                 goto error;
 715
 716         /* reserve an ethdev entry */
 717         eth_dev = rte_eth_dev_allocate(name, RTE_ETH_DEV_VIRTUAL);
 718         if (eth_dev == NULL)
 719                 goto error;
 720
 721         eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
 722         if (eth_addr == NULL)
 723                 goto error;
 724         *eth_addr = base_eth_addr;
 725         eth_addr->addr_bytes[5] = eth_dev->data->port_id;
 726
 727         vring_state = rte_zmalloc_socket(name,
 728                         sizeof(*vring_state), 0, numa_node);
 729         if (vring_state == NULL)
 730                 goto error;
 731
 732         TAILQ_INIT(&eth_dev->link_intr_cbs);
 733
 734         /* now put it all together
 735          * - store queue data in internal,
 736          * - store numa_node info in ethdev data
 737          * - point eth_dev_data to internals
 738          * - and point eth_dev structure to new eth_dev_data structure
 739          */
 740         internal->dev_name = strdup(name);
 741         if (internal->dev_name == NULL)
 742                 goto error;
 743         internal->iface_name = strdup(iface_name);
 744         if (internal->iface_name == NULL)
 745                 goto error;
 746
 747         list->eth_dev = eth_dev;
 748         pthread_mutex_lock(&internal_list_lock);
 749         TAILQ_INSERT_TAIL(&internal_list, list, next);
 750         pthread_mutex_unlock(&internal_list_lock);
 751
 752         rte_spinlock_init(&vring_state->lock);
 753         vring_states[eth_dev->data->port_id] = vring_state;
 754
 755         data->dev_private = internal;
 756         data->port_id = eth_dev->data->port_id;
 757         memmove(data->name, eth_dev->data->name, sizeof(data->name));
 758         data->nb_rx_queues = queues;
 759         data->nb_tx_queues = queues;
 760         internal->max_queues = queues;
 761         data->dev_link = pmd_link;
 762         data->mac_addrs = eth_addr;
 763
 764         /* We'll replace the 'data' originally allocated by eth_dev. So the
 765          * vhost PMD resources won't be shared between multi processes.
 766          */
 767         eth_dev->data = data;
 768         eth_dev->dev_ops = &ops;
 769         eth_dev->driver = NULL;
 770         data->dev_flags =
 771                 RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
 772         data->kdrv = RTE_KDRV_NONE;
 773         data->drv_name = internal->dev_name;
 774         data->numa_node = numa_node;
 775
 776         /* finally assign rx and tx ops */
 777         eth_dev->rx_pkt_burst = eth_vhost_rx;
 778         eth_dev->tx_pkt_burst = eth_vhost_tx;
 779
 780         return data->port_id;
 781
 782 error:
 783         if (internal)
 784                 free(internal->dev_name);
 785         rte_free(vring_state);
 786         rte_free(eth_addr);
 787         if (eth_dev)
 788                 rte_eth_dev_release_port(eth_dev);
 789         rte_free(internal);
 790         rte_free(list);
 791         rte_free(data);
 792
 793         return -1;
 794 }
 795
 796 static inline int
 797 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
 798 {
 799         const char **iface_name = extra_args;
 800
 801         if (value == NULL)
 802                 return -1;
 803
 804         *iface_name = value;
 805
 806         return 0;
 807 }
 808
 809 static inline int
 810 open_queues(const char *key __rte_unused, const char *value, void *extra_args)
 811 {
 812         uint16_t *q = extra_args;
 813
 814         if (value == NULL || extra_args == NULL)
 815                 return -EINVAL;
 816
 817         *q = (uint16_t)strtoul(value, NULL, 0);
 818         if (*q == USHRT_MAX && errno == ERANGE)
 819                 return -1;
 820
 821         if (*q > RTE_MAX_QUEUES_PER_PORT)
 822                 return -1;
 823
 824         return 0;
 825 }
 826
 827 static int
 828 rte_pmd_vhost_devinit(const char *name, const char *params)
 829 {
 830         struct rte_kvargs *kvlist = NULL;
 831         int ret = 0;
 832         char *iface_name;
 833         uint16_t queues;
 834
 835         RTE_LOG(INFO, PMD, "Initializing pmd_vhost for %s\n", name);
 836
 837         kvlist = rte_kvargs_parse(params, valid_arguments);
 838         if (kvlist == NULL)
 839                 return -1;
 840
 841         if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
 842                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
 843                                          &open_iface, &iface_name);
 844                 if (ret < 0)
 845                         goto out_free;
 846         } else {
 847                 ret = -1;
 848                 goto out_free;
 849         }
 850
 851         if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
 852                 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
 853                                          &open_queues, &queues);
 854                 if (ret < 0)
 855                         goto out_free;
 856
 857         } else
 858                 queues = 1;
 859
 860         eth_dev_vhost_create(name, iface_name, queues, rte_socket_id());
 861
 862 out_free:
 863         rte_kvargs_free(kvlist);
 864         return ret;
 865 }
 866
 867 static int
 868 rte_pmd_vhost_devuninit(const char *name)
 869 {
 870         struct rte_eth_dev *eth_dev = NULL;
 871         struct pmd_internal *internal;
 872         struct internal_list *list;
 873         unsigned int i;
 874
 875         RTE_LOG(INFO, PMD, "Un-Initializing pmd_vhost for %s\n", name);
 876
 877         /* find an ethdev entry */
 878         eth_dev = rte_eth_dev_allocated(name);
 879         if (eth_dev == NULL)
 880                 return -ENODEV;
 881
 882         internal = eth_dev->data->dev_private;
 883         if (internal == NULL)
 884                 return -ENODEV;
 885
 886         list = find_internal_resource(internal->iface_name);
 887         if (list == NULL)
 888                 return -ENODEV;
 889
 890         pthread_mutex_lock(&internal_list_lock);
 891         TAILQ_REMOVE(&internal_list, list, next);
 892         pthread_mutex_unlock(&internal_list_lock);
 893         rte_free(list);
 894
 895         eth_dev_stop(eth_dev);
 896
 897         rte_free(vring_states[eth_dev->data->port_id]);
 898         vring_states[eth_dev->data->port_id] = NULL;
 899
 900         free(internal->dev_name);
 901         free(internal->iface_name);
 902
 903         for (i = 0; i < eth_dev->data->nb_rx_queues; i++)
 904                 rte_free(eth_dev->data->rx_queues[i]);
 905         for (i = 0; i < eth_dev->data->nb_tx_queues; i++)
 906                 rte_free(eth_dev->data->tx_queues[i]);
 907
 908         rte_free(eth_dev->data->mac_addrs);
 909         rte_free(eth_dev->data);
 910         rte_free(internal);
 911
 912         rte_eth_dev_release_port(eth_dev);
 913
 914         return 0;
 915 }
 916
 917 static struct rte_driver pmd_vhost_drv = {
 918         .name = "eth_vhost",
 919         .type = PMD_VDEV,
 920         .init = rte_pmd_vhost_devinit,
 921         .uninit = rte_pmd_vhost_devuninit,
 922 };
 923
 924 PMD_REGISTER_DRIVER(pmd_vhost_drv);