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 = 0
 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
 233         if (dev == NULL) {
 234                 RTE_LOG(INFO, PMD, "Invalid argument\n");
 235                 return -1;
 236         }
 237
 238         list = find_internal_resource(dev->ifname);
 239         if (list == NULL) {
 240                 RTE_LOG(INFO, PMD, "Invalid device name\n");
 241                 return -1;
 242         }
 243
 244         eth_dev = list->eth_dev;
 245         internal = eth_dev->data->dev_private;
 246
 247         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 248                 vq = eth_dev->data->rx_queues[i];
 249                 if (vq == NULL)
 250                         continue;
 251                 vq->device = dev;
 252                 vq->internal = internal;
 253                 vq->port = eth_dev->data->port_id;
 254                 rte_vhost_enable_guest_notification(dev, vq->virtqueue_id, 0);
 255         }
 256         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 257                 vq = eth_dev->data->tx_queues[i];
 258                 if (vq == NULL)
 259                         continue;
 260                 vq->device = dev;
 261                 vq->internal = internal;
 262                 vq->port = eth_dev->data->port_id;
 263                 rte_vhost_enable_guest_notification(dev, vq->virtqueue_id, 0);
 264         }
 265
 266         dev->flags |= VIRTIO_DEV_RUNNING;
 267         dev->priv = eth_dev;
 268         eth_dev->data->dev_link.link_status = 1;
 269
 270         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 271                 vq = eth_dev->data->rx_queues[i];
 272                 if (vq == NULL)
 273                         continue;
 274                 rte_atomic32_set(&vq->allow_queuing, 1);
 275         }
 276         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 277                 vq = eth_dev->data->tx_queues[i];
 278                 if (vq == NULL)
 279                         continue;
 280                 rte_atomic32_set(&vq->allow_queuing, 1);
 281         }
 282
 283         RTE_LOG(INFO, PMD, "New connection established\n");
 284
 285         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC);
 286
 287         return 0;
 288 }
 289
 290 static void
 291 destroy_device(volatile struct virtio_net *dev)
 292 {
 293         struct rte_eth_dev *eth_dev;
 294         struct vhost_queue *vq;
 295         unsigned i;
 296
 297         if (dev == NULL) {
 298                 RTE_LOG(INFO, PMD, "Invalid argument\n");
 299                 return;
 300         }
 301
 302         eth_dev = (struct rte_eth_dev *)dev->priv;
 303         if (eth_dev == NULL) {
 304                 RTE_LOG(INFO, PMD, "Failed to find a ethdev\n");
 305                 return;
 306         }
 307
 308         /* Wait until rx/tx_pkt_burst stops accessing vhost device */
 309         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 310                 vq = eth_dev->data->rx_queues[i];
 311                 if (vq == NULL)
 312                         continue;
 313                 rte_atomic32_set(&vq->allow_queuing, 0);
 314                 while (rte_atomic32_read(&vq->while_queuing))
 315                         rte_pause();
 316         }
 317         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 318                 vq = eth_dev->data->tx_queues[i];
 319                 if (vq == NULL)
 320                         continue;
 321                 rte_atomic32_set(&vq->allow_queuing, 0);
 322                 while (rte_atomic32_read(&vq->while_queuing))
 323                         rte_pause();
 324         }
 325
 326         eth_dev->data->dev_link.link_status = 0;
 327
 328         dev->priv = NULL;
 329         dev->flags &= ~VIRTIO_DEV_RUNNING;
 330
 331         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 332                 vq = eth_dev->data->rx_queues[i];
 333                 if (vq == NULL)
 334                         continue;
 335                 vq->device = NULL;
 336         }
 337         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 338                 vq = eth_dev->data->tx_queues[i];
 339                 if (vq == NULL)
 340                         continue;
 341                 vq->device = NULL;
 342         }
 343
 344         RTE_LOG(INFO, PMD, "Connection closed\n");
 345
 346         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC);
 347 }
 348
 349 static int
 350 vring_state_changed(struct virtio_net *dev, uint16_t vring, int enable)
 351 {
 352         struct rte_vhost_vring_state *state;
 353         struct rte_eth_dev *eth_dev;
 354         struct internal_list *list;
 355 #ifdef RTE_LIBRTE_VHOST_NUMA
 356         int newnode, ret;
 357 #endif
 358
 359         if (dev == NULL) {
 360                 RTE_LOG(ERR, PMD, "Invalid argument\n");
 361                 return -1;
 362         }
 363
 364         list = find_internal_resource(dev->ifname);
 365         if (list == NULL) {
 366                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", dev->ifname);
 367                 return -1;
 368         }
 369
 370         eth_dev = list->eth_dev;
 371         /* won't be NULL */
 372         state = vring_states[eth_dev->data->port_id];
 373
 374 #ifdef RTE_LIBRTE_VHOST_NUMA
 375         ret  = get_mempolicy(&newnode, NULL, 0, dev,
 376                         MPOL_F_NODE | MPOL_F_ADDR);
 377         if (ret < 0) {
 378                 RTE_LOG(ERR, PMD, "Unknown numa node\n");
 379                 return -1;
 380         }
 381
 382         eth_dev->data->numa_node = newnode;
 383 #endif
 384         rte_spinlock_lock(&state->lock);
 385         state->cur[vring] = enable;
 386         state->max_vring = RTE_MAX(vring, state->max_vring);
 387         rte_spinlock_unlock(&state->lock);
 388
 389         RTE_LOG(INFO, PMD, "vring%u is %s\n",
 390                         vring, enable ? "enabled" : "disabled");
 391
 392         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE);
 393
 394         return 0;
 395 }
 396
 397 int
 398 rte_eth_vhost_get_queue_event(uint8_t port_id,
 399                 struct rte_eth_vhost_queue_event *event)
 400 {
 401         struct rte_vhost_vring_state *state;
 402         unsigned int i;
 403         int idx;
 404
 405         if (port_id >= RTE_MAX_ETHPORTS) {
 406                 RTE_LOG(ERR, PMD, "Invalid port id\n");
 407                 return -1;
 408         }
 409
 410         state = vring_states[port_id];
 411         if (!state) {
 412                 RTE_LOG(ERR, PMD, "Unused port\n");
 413                 return -1;
 414         }
 415
 416         rte_spinlock_lock(&state->lock);
 417         for (i = 0; i <= state->max_vring; i++) {
 418                 idx = state->index++ % (state->max_vring + 1);
 419
 420                 if (state->cur[idx] != state->seen[idx]) {
 421                         state->seen[idx] = state->cur[idx];
 422                         event->queue_id = idx / 2;
 423                         event->rx = idx & 1;
 424                         event->enable = state->cur[idx];
 425                         rte_spinlock_unlock(&state->lock);
 426                         return 0;
 427                 }
 428         }
 429         rte_spinlock_unlock(&state->lock);
 430
 431         return -1;
 432 }
 433
 434 static void *
 435 vhost_driver_session(void *param __rte_unused)
 436 {
 437         static struct virtio_net_device_ops vhost_ops;
 438
 439         /* set vhost arguments */
 440         vhost_ops.new_device = new_device;
 441         vhost_ops.destroy_device = destroy_device;
 442         vhost_ops.vring_state_changed = vring_state_changed;
 443         if (rte_vhost_driver_callback_register(&vhost_ops) < 0)
 444                 RTE_LOG(ERR, PMD, "Can't register callbacks\n");
 445
 446         /* start event handling */
 447         rte_vhost_driver_session_start();
 448
 449         return NULL;
 450 }
 451
 452 static int
 453 vhost_driver_session_start(void)
 454 {
 455         int ret;
 456
 457         ret = pthread_create(&session_th,
 458                         NULL, vhost_driver_session, NULL);
 459         if (ret)
 460                 RTE_LOG(ERR, PMD, "Can't create a thread\n");
 461
 462         return ret;
 463 }
 464
 465 static void
 466 vhost_driver_session_stop(void)
 467 {
 468         int ret;
 469
 470         ret = pthread_cancel(session_th);
 471         if (ret)
 472                 RTE_LOG(ERR, PMD, "Can't cancel the thread\n");
 473
 474         ret = pthread_join(session_th, NULL);
 475         if (ret)
 476                 RTE_LOG(ERR, PMD, "Can't join the thread\n");
 477 }
 478
 479 static int
 480 eth_dev_start(struct rte_eth_dev *dev)
 481 {
 482         struct pmd_internal *internal = dev->data->dev_private;
 483         int ret = 0;
 484
 485         if (rte_atomic16_cmpset(&internal->once, 0, 1)) {
 486                 ret = rte_vhost_driver_register(internal->iface_name);
 487                 if (ret)
 488                         return ret;
 489         }
 490
 491         /* We need only one message handling thread */
 492         if (rte_atomic16_add_return(&nb_started_ports, 1) == 1)
 493                 ret = vhost_driver_session_start();
 494
 495         return ret;
 496 }
 497
 498 static void
 499 eth_dev_stop(struct rte_eth_dev *dev)
 500 {
 501         struct pmd_internal *internal = dev->data->dev_private;
 502
 503         if (rte_atomic16_cmpset(&internal->once, 1, 0))
 504                 rte_vhost_driver_unregister(internal->iface_name);
 505
 506         if (rte_atomic16_sub_return(&nb_started_ports, 1) == 0)
 507                 vhost_driver_session_stop();
 508 }
 509
 510 static int
 511 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 512                    uint16_t nb_rx_desc __rte_unused,
 513                    unsigned int socket_id,
 514                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 515                    struct rte_mempool *mb_pool)
 516 {
 517         struct vhost_queue *vq;
 518
 519         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
 520                         RTE_CACHE_LINE_SIZE, socket_id);
 521         if (vq == NULL) {
 522                 RTE_LOG(ERR, PMD, "Failed to allocate memory for rx queue\n");
 523                 return -ENOMEM;
 524         }
 525
 526         vq->mb_pool = mb_pool;
 527         vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
 528         dev->data->rx_queues[rx_queue_id] = vq;
 529
 530         return 0;
 531 }
 532
 533 static int
 534 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 535                    uint16_t nb_tx_desc __rte_unused,
 536                    unsigned int socket_id,
 537                    const struct rte_eth_txconf *tx_conf __rte_unused)
 538 {
 539         struct vhost_queue *vq;
 540
 541         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
 542                         RTE_CACHE_LINE_SIZE, socket_id);
 543         if (vq == NULL) {
 544                 RTE_LOG(ERR, PMD, "Failed to allocate memory for tx queue\n");
 545                 return -ENOMEM;
 546         }
 547
 548         vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
 549         dev->data->tx_queues[tx_queue_id] = vq;
 550
 551         return 0;
 552 }
 553
 554 static void
 555 eth_dev_info(struct rte_eth_dev *dev,
 556              struct rte_eth_dev_info *dev_info)
 557 {
 558         struct pmd_internal *internal;
 559
 560         internal = dev->data->dev_private;
 561         if (internal == NULL) {
 562                 RTE_LOG(ERR, PMD, "Invalid device specified\n");
 563                 return;
 564         }
 565
 566         dev_info->driver_name = drivername;
 567         dev_info->max_mac_addrs = 1;
 568         dev_info->max_rx_pktlen = (uint32_t)-1;
 569         dev_info->max_rx_queues = internal->max_queues;
 570         dev_info->max_tx_queues = internal->max_queues;
 571         dev_info->min_rx_bufsize = 0;
 572 }
 573
 574 static void
 575 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 576 {
 577         unsigned i;
 578         unsigned long rx_total = 0, tx_total = 0, tx_missed_total = 0;
 579         unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
 580         struct vhost_queue *vq;
 581
 582         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 583                         i < dev->data->nb_rx_queues; i++) {
 584                 if (dev->data->rx_queues[i] == NULL)
 585                         continue;
 586                 vq = dev->data->rx_queues[i];
 587                 stats->q_ipackets[i] = vq->rx_pkts;
 588                 rx_total += stats->q_ipackets[i];
 589
 590                 stats->q_ibytes[i] = vq->rx_bytes;
 591                 rx_total_bytes += stats->q_ibytes[i];
 592         }
 593
 594         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 595                         i < dev->data->nb_tx_queues; i++) {
 596                 if (dev->data->tx_queues[i] == NULL)
 597                         continue;
 598                 vq = dev->data->tx_queues[i];
 599                 stats->q_opackets[i] = vq->tx_pkts;
 600                 tx_missed_total += vq->missed_pkts;
 601                 tx_total += stats->q_opackets[i];
 602
 603                 stats->q_obytes[i] = vq->tx_bytes;
 604                 tx_total_bytes += stats->q_obytes[i];
 605         }
 606
 607         stats->ipackets = rx_total;
 608         stats->opackets = tx_total;
 609         stats->imissed = tx_missed_total;
 610         stats->ibytes = rx_total_bytes;
 611         stats->obytes = tx_total_bytes;
 612 }
 613
 614 static void
 615 eth_stats_reset(struct rte_eth_dev *dev)
 616 {
 617         struct vhost_queue *vq;
 618         unsigned i;
 619
 620         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 621                 if (dev->data->rx_queues[i] == NULL)
 622                         continue;
 623                 vq = dev->data->rx_queues[i];
 624                 vq->rx_pkts = 0;
 625                 vq->rx_bytes = 0;
 626         }
 627         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 628                 if (dev->data->tx_queues[i] == NULL)
 629                         continue;
 630                 vq = dev->data->tx_queues[i];
 631                 vq->tx_pkts = 0;
 632                 vq->tx_bytes = 0;
 633                 vq->missed_pkts = 0;
 634         }
 635 }
 636
 637 static void
 638 eth_queue_release(void *q)
 639 {
 640         rte_free(q);
 641 }
 642
 643 static int
 644 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 645                 int wait_to_complete __rte_unused)
 646 {
 647         return 0;
 648 }
 649
 650 /**
 651  * Disable features in feature_mask. Returns 0 on success.
 652  */
 653 int
 654 rte_eth_vhost_feature_disable(uint64_t feature_mask)
 655 {
 656         return rte_vhost_feature_disable(feature_mask);
 657 }
 658
 659 /**
 660  * Enable features in feature_mask. Returns 0 on success.
 661  */
 662 int
 663 rte_eth_vhost_feature_enable(uint64_t feature_mask)
 664 {
 665         return rte_vhost_feature_enable(feature_mask);
 666 }
 667
 668 /* Returns currently supported vhost features */
 669 uint64_t
 670 rte_eth_vhost_feature_get(void)
 671 {
 672         return rte_vhost_feature_get();
 673 }
 674
 675 static const struct eth_dev_ops ops = {
 676         .dev_start = eth_dev_start,
 677         .dev_stop = eth_dev_stop,
 678         .dev_configure = eth_dev_configure,
 679         .dev_infos_get = eth_dev_info,
 680         .rx_queue_setup = eth_rx_queue_setup,
 681         .tx_queue_setup = eth_tx_queue_setup,
 682         .rx_queue_release = eth_queue_release,
 683         .tx_queue_release = eth_queue_release,
 684         .link_update = eth_link_update,
 685         .stats_get = eth_stats_get,
 686         .stats_reset = eth_stats_reset,
 687 };
 688
 689 static int
 690 eth_dev_vhost_create(const char *name, char *iface_name, int16_t queues,
 691                      const unsigned numa_node)
 692 {
 693         struct rte_eth_dev_data *data = NULL;
 694         struct pmd_internal *internal = NULL;
 695         struct rte_eth_dev *eth_dev = NULL;
 696         struct ether_addr *eth_addr = NULL;
 697         struct rte_vhost_vring_state *vring_state = NULL;
 698         struct internal_list *list = NULL;
 699
 700         RTE_LOG(INFO, PMD, "Creating VHOST-USER backend on numa socket %u\n",
 701                 numa_node);
 702
 703         /* now do all data allocation - for eth_dev structure, dummy pci driver
 704          * and internal (private) data
 705          */
 706         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
 707         if (data == NULL)
 708                 goto error;
 709
 710         internal = rte_zmalloc_socket(name, sizeof(*internal), 0, numa_node);
 711         if (internal == NULL)
 712                 goto error;
 713
 714         list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
 715         if (list == NULL)
 716                 goto error;
 717
 718         /* reserve an ethdev entry */
 719         eth_dev = rte_eth_dev_allocate(name, RTE_ETH_DEV_VIRTUAL);
 720         if (eth_dev == NULL)
 721                 goto error;
 722
 723         eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
 724         if (eth_addr == NULL)
 725                 goto error;
 726         *eth_addr = base_eth_addr;
 727         eth_addr->addr_bytes[5] = eth_dev->data->port_id;
 728
 729         vring_state = rte_zmalloc_socket(name,
 730                         sizeof(*vring_state), 0, numa_node);
 731         if (vring_state == NULL)
 732                 goto error;
 733
 734         TAILQ_INIT(&eth_dev->link_intr_cbs);
 735
 736         /* now put it all together
 737          * - store queue data in internal,
 738          * - store numa_node info in ethdev data
 739          * - point eth_dev_data to internals
 740          * - and point eth_dev structure to new eth_dev_data structure
 741          */
 742         internal->dev_name = strdup(name);
 743         if (internal->dev_name == NULL)
 744                 goto error;
 745         internal->iface_name = strdup(iface_name);
 746         if (internal->iface_name == NULL)
 747                 goto error;
 748
 749         list->eth_dev = eth_dev;
 750         pthread_mutex_lock(&internal_list_lock);
 751         TAILQ_INSERT_TAIL(&internal_list, list, next);
 752         pthread_mutex_unlock(&internal_list_lock);
 753
 754         rte_spinlock_init(&vring_state->lock);
 755         vring_states[eth_dev->data->port_id] = vring_state;
 756
 757         data->dev_private = internal;
 758         data->port_id = eth_dev->data->port_id;
 759         memmove(data->name, eth_dev->data->name, sizeof(data->name));
 760         data->nb_rx_queues = queues;
 761         data->nb_tx_queues = queues;
 762         internal->max_queues = queues;
 763         data->dev_link = pmd_link;
 764         data->mac_addrs = eth_addr;
 765
 766         /* We'll replace the 'data' originally allocated by eth_dev. So the
 767          * vhost PMD resources won't be shared between multi processes.
 768          */
 769         eth_dev->data = data;
 770         eth_dev->dev_ops = &ops;
 771         eth_dev->driver = NULL;
 772         data->dev_flags =
 773                 RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
 774         data->kdrv = RTE_KDRV_NONE;
 775         data->drv_name = internal->dev_name;
 776         data->numa_node = numa_node;
 777
 778         /* finally assign rx and tx ops */
 779         eth_dev->rx_pkt_burst = eth_vhost_rx;
 780         eth_dev->tx_pkt_burst = eth_vhost_tx;
 781
 782         return data->port_id;
 783
 784 error:
 785         if (internal)
 786                 free(internal->dev_name);
 787         rte_free(vring_state);
 788         rte_free(eth_addr);
 789         if (eth_dev)
 790                 rte_eth_dev_release_port(eth_dev);
 791         rte_free(internal);
 792         rte_free(list);
 793         rte_free(data);
 794
 795         return -1;
 796 }
 797
 798 static inline int
 799 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
 800 {
 801         const char **iface_name = extra_args;
 802
 803         if (value == NULL)
 804                 return -1;
 805
 806         *iface_name = value;
 807
 808         return 0;
 809 }
 810
 811 static inline int
 812 open_queues(const char *key __rte_unused, const char *value, void *extra_args)
 813 {
 814         uint16_t *q = extra_args;
 815
 816         if (value == NULL || extra_args == NULL)
 817                 return -EINVAL;
 818
 819         *q = (uint16_t)strtoul(value, NULL, 0);
 820         if (*q == USHRT_MAX && errno == ERANGE)
 821                 return -1;
 822
 823         if (*q > RTE_MAX_QUEUES_PER_PORT)
 824                 return -1;
 825
 826         return 0;
 827 }
 828
 829 static int
 830 rte_pmd_vhost_devinit(const char *name, const char *params)
 831 {
 832         struct rte_kvargs *kvlist = NULL;
 833         int ret = 0;
 834         char *iface_name;
 835         uint16_t queues;
 836
 837         RTE_LOG(INFO, PMD, "Initializing pmd_vhost for %s\n", name);
 838
 839         kvlist = rte_kvargs_parse(params, valid_arguments);
 840         if (kvlist == NULL)
 841                 return -1;
 842
 843         if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
 844                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
 845                                          &open_iface, &iface_name);
 846                 if (ret < 0)
 847                         goto out_free;
 848         } else {
 849                 ret = -1;
 850                 goto out_free;
 851         }
 852
 853         if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
 854                 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
 855                                          &open_queues, &queues);
 856                 if (ret < 0)
 857                         goto out_free;
 858
 859         } else
 860                 queues = 1;
 861
 862         eth_dev_vhost_create(name, iface_name, queues, rte_socket_id());
 863
 864 out_free:
 865         rte_kvargs_free(kvlist);
 866         return ret;
 867 }
 868
 869 static int
 870 rte_pmd_vhost_devuninit(const char *name)
 871 {
 872         struct rte_eth_dev *eth_dev = NULL;
 873         struct pmd_internal *internal;
 874         struct internal_list *list;
 875         unsigned int i;
 876
 877         RTE_LOG(INFO, PMD, "Un-Initializing pmd_vhost for %s\n", name);
 878
 879         /* find an ethdev entry */
 880         eth_dev = rte_eth_dev_allocated(name);
 881         if (eth_dev == NULL)
 882                 return -ENODEV;
 883
 884         internal = eth_dev->data->dev_private;
 885         if (internal == NULL)
 886                 return -ENODEV;
 887
 888         list = find_internal_resource(internal->iface_name);
 889         if (list == NULL)
 890                 return -ENODEV;
 891
 892         pthread_mutex_lock(&internal_list_lock);
 893         TAILQ_REMOVE(&internal_list, list, next);
 894         pthread_mutex_unlock(&internal_list_lock);
 895         rte_free(list);
 896
 897         eth_dev_stop(eth_dev);
 898
 899         rte_free(vring_states[eth_dev->data->port_id]);
 900         vring_states[eth_dev->data->port_id] = NULL;
 901
 902         free(internal->dev_name);
 903         free(internal->iface_name);
 904
 905         for (i = 0; i < eth_dev->data->nb_rx_queues; i++)
 906                 rte_free(eth_dev->data->rx_queues[i]);
 907         for (i = 0; i < eth_dev->data->nb_tx_queues; i++)
 908                 rte_free(eth_dev->data->tx_queues[i]);
 909
 910         rte_free(eth_dev->data->mac_addrs);
 911         rte_free(eth_dev->data);
 912         rte_free(internal);
 913
 914         rte_eth_dev_release_port(eth_dev);
 915
 916         return 0;
 917 }
 918
 919 static struct rte_driver pmd_vhost_drv = {
 920         .name = "eth_vhost",
 921         .type = PMD_VDEV,
 922         .init = rte_pmd_vhost_devinit,
 923         .uninit = rte_pmd_vhost_devuninit,
 924 };
 925
 926 PMD_REGISTER_DRIVER(pmd_vhost_drv);