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