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