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         int vid;
  75         rte_atomic32_t allow_queuing;
  76         rte_atomic32_t while_queuing;
  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->vid,
 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->vid,
 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(int vid)
 226 {
 227         struct rte_eth_dev *eth_dev;
 228         struct internal_list *list;
 229         struct pmd_internal *internal;
 230         struct vhost_queue *vq;
 231         unsigned i;
 232         char ifname[PATH_MAX];
 233 #ifdef RTE_LIBRTE_VHOST_NUMA
 234         int newnode;
 235 #endif
 236
 237         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 238         list = find_internal_resource(ifname);
 239         if (list == NULL) {
 240                 RTE_LOG(INFO, PMD, "Invalid device name: %s\n", ifname);
 241                 return -1;
 242         }
 243
 244         eth_dev = list->eth_dev;
 245         internal = eth_dev->data->dev_private;
 246
 247 #ifdef RTE_LIBRTE_VHOST_NUMA
 248         newnode = rte_vhost_get_numa_node(vid);
 249         if (newnode >= 0)
 250                 eth_dev->data->numa_node = newnode;
 251 #endif
 252
 253         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 254                 vq = eth_dev->data->rx_queues[i];
 255                 if (vq == NULL)
 256                         continue;
 257                 vq->vid = vid;
 258                 vq->internal = internal;
 259                 vq->port = eth_dev->data->port_id;
 260         }
 261         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 262                 vq = eth_dev->data->tx_queues[i];
 263                 if (vq == NULL)
 264                         continue;
 265                 vq->vid = vid;
 266                 vq->internal = internal;
 267                 vq->port = eth_dev->data->port_id;
 268         }
 269
 270         for (i = 0; i < rte_vhost_get_queue_num(vid) * VIRTIO_QNUM; i++)
 271                 rte_vhost_enable_guest_notification(vid, i, 0);
 272
 273         eth_dev->data->dev_link.link_status = ETH_LINK_UP;
 274
 275         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 276                 vq = eth_dev->data->rx_queues[i];
 277                 if (vq == NULL)
 278                         continue;
 279                 rte_atomic32_set(&vq->allow_queuing, 1);
 280         }
 281         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 282                 vq = eth_dev->data->tx_queues[i];
 283                 if (vq == NULL)
 284                         continue;
 285                 rte_atomic32_set(&vq->allow_queuing, 1);
 286         }
 287
 288         RTE_LOG(INFO, PMD, "New connection established\n");
 289
 290         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC);
 291
 292         return 0;
 293 }
 294
 295 static void
 296 destroy_device(int vid)
 297 {
 298         struct rte_eth_dev *eth_dev;
 299         struct vhost_queue *vq;
 300         struct internal_list *list;
 301         char ifname[PATH_MAX];
 302         unsigned i;
 303
 304         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 305         list = find_internal_resource(ifname);
 306         if (list == NULL) {
 307                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
 308                 return;
 309         }
 310         eth_dev = list->eth_dev;
 311
 312         /* Wait until rx/tx_pkt_burst stops accessing vhost device */
 313         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 314                 vq = eth_dev->data->rx_queues[i];
 315                 if (vq == NULL)
 316                         continue;
 317                 rte_atomic32_set(&vq->allow_queuing, 0);
 318                 while (rte_atomic32_read(&vq->while_queuing))
 319                         rte_pause();
 320         }
 321         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 322                 vq = eth_dev->data->tx_queues[i];
 323                 if (vq == NULL)
 324                         continue;
 325                 rte_atomic32_set(&vq->allow_queuing, 0);
 326                 while (rte_atomic32_read(&vq->while_queuing))
 327                         rte_pause();
 328         }
 329
 330         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
 331
 332         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 333                 vq = eth_dev->data->rx_queues[i];
 334                 if (vq == NULL)
 335                         continue;
 336                 vq->vid = -1;
 337         }
 338         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 339                 vq = eth_dev->data->tx_queues[i];
 340                 if (vq == NULL)
 341                         continue;
 342                 vq->vid = -1;
 343         }
 344
 345         RTE_LOG(INFO, PMD, "Connection closed\n");
 346
 347         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC);
 348 }
 349
 350 static int
 351 vring_state_changed(int vid, uint16_t vring, int enable)
 352 {
 353         struct rte_vhost_vring_state *state;
 354         struct rte_eth_dev *eth_dev;
 355         struct internal_list *list;
 356         char ifname[PATH_MAX];
 357
 358         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 359         list = find_internal_resource(ifname);
 360         if (list == NULL) {
 361                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
 362                 return -1;
 363         }
 364
 365         eth_dev = list->eth_dev;
 366         /* won't be NULL */
 367         state = vring_states[eth_dev->data->port_id];
 368         rte_spinlock_lock(&state->lock);
 369         state->cur[vring] = enable;
 370         state->max_vring = RTE_MAX(vring, state->max_vring);
 371         rte_spinlock_unlock(&state->lock);
 372
 373         RTE_LOG(INFO, PMD, "vring%u is %s\n",
 374                         vring, enable ? "enabled" : "disabled");
 375
 376         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE);
 377
 378         return 0;
 379 }
 380
 381 int
 382 rte_eth_vhost_get_queue_event(uint8_t port_id,
 383                 struct rte_eth_vhost_queue_event *event)
 384 {
 385         struct rte_vhost_vring_state *state;
 386         unsigned int i;
 387         int idx;
 388
 389         if (port_id >= RTE_MAX_ETHPORTS) {
 390                 RTE_LOG(ERR, PMD, "Invalid port id\n");
 391                 return -1;
 392         }
 393
 394         state = vring_states[port_id];
 395         if (!state) {
 396                 RTE_LOG(ERR, PMD, "Unused port\n");
 397                 return -1;
 398         }
 399
 400         rte_spinlock_lock(&state->lock);
 401         for (i = 0; i <= state->max_vring; i++) {
 402                 idx = state->index++ % (state->max_vring + 1);
 403
 404                 if (state->cur[idx] != state->seen[idx]) {
 405                         state->seen[idx] = state->cur[idx];
 406                         event->queue_id = idx / 2;
 407                         event->rx = idx & 1;
 408                         event->enable = state->cur[idx];
 409                         rte_spinlock_unlock(&state->lock);
 410                         return 0;
 411                 }
 412         }
 413         rte_spinlock_unlock(&state->lock);
 414
 415         return -1;
 416 }
 417
 418 static void *
 419 vhost_driver_session(void *param __rte_unused)
 420 {
 421         static struct virtio_net_device_ops vhost_ops;
 422
 423         /* set vhost arguments */
 424         vhost_ops.new_device = new_device;
 425         vhost_ops.destroy_device = destroy_device;
 426         vhost_ops.vring_state_changed = vring_state_changed;
 427         if (rte_vhost_driver_callback_register(&vhost_ops) < 0)
 428                 RTE_LOG(ERR, PMD, "Can't register callbacks\n");
 429
 430         /* start event handling */
 431         rte_vhost_driver_session_start();
 432
 433         return NULL;
 434 }
 435
 436 static int
 437 vhost_driver_session_start(void)
 438 {
 439         int ret;
 440
 441         ret = pthread_create(&session_th,
 442                         NULL, vhost_driver_session, NULL);
 443         if (ret)
 444                 RTE_LOG(ERR, PMD, "Can't create a thread\n");
 445
 446         return ret;
 447 }
 448
 449 static void
 450 vhost_driver_session_stop(void)
 451 {
 452         int ret;
 453
 454         ret = pthread_cancel(session_th);
 455         if (ret)
 456                 RTE_LOG(ERR, PMD, "Can't cancel the thread\n");
 457
 458         ret = pthread_join(session_th, NULL);
 459         if (ret)
 460                 RTE_LOG(ERR, PMD, "Can't join the thread\n");
 461 }
 462
 463 static int
 464 eth_dev_start(struct rte_eth_dev *dev)
 465 {
 466         struct pmd_internal *internal = dev->data->dev_private;
 467         int ret = 0;
 468
 469         if (rte_atomic16_cmpset(&internal->once, 0, 1)) {
 470                 ret = rte_vhost_driver_register(internal->iface_name);
 471                 if (ret)
 472                         return ret;
 473         }
 474
 475         /* We need only one message handling thread */
 476         if (rte_atomic16_add_return(&nb_started_ports, 1) == 1)
 477                 ret = vhost_driver_session_start();
 478
 479         return ret;
 480 }
 481
 482 static void
 483 eth_dev_stop(struct rte_eth_dev *dev)
 484 {
 485         struct pmd_internal *internal = dev->data->dev_private;
 486
 487         if (rte_atomic16_cmpset(&internal->once, 1, 0))
 488                 rte_vhost_driver_unregister(internal->iface_name);
 489
 490         if (rte_atomic16_sub_return(&nb_started_ports, 1) == 0)
 491                 vhost_driver_session_stop();
 492 }
 493
 494 static int
 495 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 496                    uint16_t nb_rx_desc __rte_unused,
 497                    unsigned int socket_id,
 498                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 499                    struct rte_mempool *mb_pool)
 500 {
 501         struct vhost_queue *vq;
 502
 503         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
 504                         RTE_CACHE_LINE_SIZE, socket_id);
 505         if (vq == NULL) {
 506                 RTE_LOG(ERR, PMD, "Failed to allocate memory for rx queue\n");
 507                 return -ENOMEM;
 508         }
 509
 510         vq->mb_pool = mb_pool;
 511         vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
 512         dev->data->rx_queues[rx_queue_id] = vq;
 513
 514         return 0;
 515 }
 516
 517 static int
 518 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 519                    uint16_t nb_tx_desc __rte_unused,
 520                    unsigned int socket_id,
 521                    const struct rte_eth_txconf *tx_conf __rte_unused)
 522 {
 523         struct vhost_queue *vq;
 524
 525         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
 526                         RTE_CACHE_LINE_SIZE, socket_id);
 527         if (vq == NULL) {
 528                 RTE_LOG(ERR, PMD, "Failed to allocate memory for tx queue\n");
 529                 return -ENOMEM;
 530         }
 531
 532         vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
 533         dev->data->tx_queues[tx_queue_id] = vq;
 534
 535         return 0;
 536 }
 537
 538 static void
 539 eth_dev_info(struct rte_eth_dev *dev,
 540              struct rte_eth_dev_info *dev_info)
 541 {
 542         struct pmd_internal *internal;
 543
 544         internal = dev->data->dev_private;
 545         if (internal == NULL) {
 546                 RTE_LOG(ERR, PMD, "Invalid device specified\n");
 547                 return;
 548         }
 549
 550         dev_info->driver_name = drivername;
 551         dev_info->max_mac_addrs = 1;
 552         dev_info->max_rx_pktlen = (uint32_t)-1;
 553         dev_info->max_rx_queues = internal->max_queues;
 554         dev_info->max_tx_queues = internal->max_queues;
 555         dev_info->min_rx_bufsize = 0;
 556 }
 557
 558 static void
 559 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 560 {
 561         unsigned i;
 562         unsigned long rx_total = 0, tx_total = 0, tx_missed_total = 0;
 563         unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
 564         struct vhost_queue *vq;
 565
 566         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 567                         i < dev->data->nb_rx_queues; i++) {
 568                 if (dev->data->rx_queues[i] == NULL)
 569                         continue;
 570                 vq = dev->data->rx_queues[i];
 571                 stats->q_ipackets[i] = vq->rx_pkts;
 572                 rx_total += stats->q_ipackets[i];
 573
 574                 stats->q_ibytes[i] = vq->rx_bytes;
 575                 rx_total_bytes += stats->q_ibytes[i];
 576         }
 577
 578         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 579                         i < dev->data->nb_tx_queues; i++) {
 580                 if (dev->data->tx_queues[i] == NULL)
 581                         continue;
 582                 vq = dev->data->tx_queues[i];
 583                 stats->q_opackets[i] = vq->tx_pkts;
 584                 tx_missed_total += vq->missed_pkts;
 585                 tx_total += stats->q_opackets[i];
 586
 587                 stats->q_obytes[i] = vq->tx_bytes;
 588                 tx_total_bytes += stats->q_obytes[i];
 589         }
 590
 591         stats->ipackets = rx_total;
 592         stats->opackets = tx_total;
 593         stats->imissed = tx_missed_total;
 594         stats->ibytes = rx_total_bytes;
 595         stats->obytes = tx_total_bytes;
 596 }
 597
 598 static void
 599 eth_stats_reset(struct rte_eth_dev *dev)
 600 {
 601         struct vhost_queue *vq;
 602         unsigned i;
 603
 604         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 605                 if (dev->data->rx_queues[i] == NULL)
 606                         continue;
 607                 vq = dev->data->rx_queues[i];
 608                 vq->rx_pkts = 0;
 609                 vq->rx_bytes = 0;
 610         }
 611         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 612                 if (dev->data->tx_queues[i] == NULL)
 613                         continue;
 614                 vq = dev->data->tx_queues[i];
 615                 vq->tx_pkts = 0;
 616                 vq->tx_bytes = 0;
 617                 vq->missed_pkts = 0;
 618         }
 619 }
 620
 621 static void
 622 eth_queue_release(void *q)
 623 {
 624         rte_free(q);
 625 }
 626
 627 static int
 628 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 629                 int wait_to_complete __rte_unused)
 630 {
 631         return 0;
 632 }
 633
 634 /**
 635  * Disable features in feature_mask. Returns 0 on success.
 636  */
 637 int
 638 rte_eth_vhost_feature_disable(uint64_t feature_mask)
 639 {
 640         return rte_vhost_feature_disable(feature_mask);
 641 }
 642
 643 /**
 644  * Enable features in feature_mask. Returns 0 on success.
 645  */
 646 int
 647 rte_eth_vhost_feature_enable(uint64_t feature_mask)
 648 {
 649         return rte_vhost_feature_enable(feature_mask);
 650 }
 651
 652 /* Returns currently supported vhost features */
 653 uint64_t
 654 rte_eth_vhost_feature_get(void)
 655 {
 656         return rte_vhost_feature_get();
 657 }
 658
 659 static const struct eth_dev_ops ops = {
 660         .dev_start = eth_dev_start,
 661         .dev_stop = eth_dev_stop,
 662         .dev_configure = eth_dev_configure,
 663         .dev_infos_get = eth_dev_info,
 664         .rx_queue_setup = eth_rx_queue_setup,
 665         .tx_queue_setup = eth_tx_queue_setup,
 666         .rx_queue_release = eth_queue_release,
 667         .tx_queue_release = eth_queue_release,
 668         .link_update = eth_link_update,
 669         .stats_get = eth_stats_get,
 670         .stats_reset = eth_stats_reset,
 671 };
 672
 673 static int
 674 eth_dev_vhost_create(const char *name, char *iface_name, int16_t queues,
 675                      const unsigned numa_node)
 676 {
 677         struct rte_eth_dev_data *data = NULL;
 678         struct pmd_internal *internal = NULL;
 679         struct rte_eth_dev *eth_dev = NULL;
 680         struct ether_addr *eth_addr = NULL;
 681         struct rte_vhost_vring_state *vring_state = NULL;
 682         struct internal_list *list = NULL;
 683
 684         RTE_LOG(INFO, PMD, "Creating VHOST-USER backend on numa socket %u\n",
 685                 numa_node);
 686
 687         /* now do all data allocation - for eth_dev structure, dummy pci driver
 688          * and internal (private) data
 689          */
 690         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
 691         if (data == NULL)
 692                 goto error;
 693
 694         internal = rte_zmalloc_socket(name, sizeof(*internal), 0, numa_node);
 695         if (internal == NULL)
 696                 goto error;
 697
 698         list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
 699         if (list == NULL)
 700                 goto error;
 701
 702         /* reserve an ethdev entry */
 703         eth_dev = rte_eth_dev_allocate(name, RTE_ETH_DEV_VIRTUAL);
 704         if (eth_dev == NULL)
 705                 goto error;
 706
 707         eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
 708         if (eth_addr == NULL)
 709                 goto error;
 710         *eth_addr = base_eth_addr;
 711         eth_addr->addr_bytes[5] = eth_dev->data->port_id;
 712
 713         vring_state = rte_zmalloc_socket(name,
 714                         sizeof(*vring_state), 0, numa_node);
 715         if (vring_state == NULL)
 716                 goto error;
 717
 718         TAILQ_INIT(&eth_dev->link_intr_cbs);
 719
 720         /* now put it all together
 721          * - store queue data in internal,
 722          * - store numa_node info in ethdev data
 723          * - point eth_dev_data to internals
 724          * - and point eth_dev structure to new eth_dev_data structure
 725          */
 726         internal->dev_name = strdup(name);
 727         if (internal->dev_name == NULL)
 728                 goto error;
 729         internal->iface_name = strdup(iface_name);
 730         if (internal->iface_name == NULL)
 731                 goto error;
 732
 733         list->eth_dev = eth_dev;
 734         pthread_mutex_lock(&internal_list_lock);
 735         TAILQ_INSERT_TAIL(&internal_list, list, next);
 736         pthread_mutex_unlock(&internal_list_lock);
 737
 738         rte_spinlock_init(&vring_state->lock);
 739         vring_states[eth_dev->data->port_id] = vring_state;
 740
 741         data->dev_private = internal;
 742         data->port_id = eth_dev->data->port_id;
 743         memmove(data->name, eth_dev->data->name, sizeof(data->name));
 744         data->nb_rx_queues = queues;
 745         data->nb_tx_queues = queues;
 746         internal->max_queues = queues;
 747         data->dev_link = pmd_link;
 748         data->mac_addrs = eth_addr;
 749
 750         /* We'll replace the 'data' originally allocated by eth_dev. So the
 751          * vhost PMD resources won't be shared between multi processes.
 752          */
 753         eth_dev->data = data;
 754         eth_dev->dev_ops = &ops;
 755         eth_dev->driver = NULL;
 756         data->dev_flags =
 757                 RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
 758         data->kdrv = RTE_KDRV_NONE;
 759         data->drv_name = internal->dev_name;
 760         data->numa_node = numa_node;
 761
 762         /* finally assign rx and tx ops */
 763         eth_dev->rx_pkt_burst = eth_vhost_rx;
 764         eth_dev->tx_pkt_burst = eth_vhost_tx;
 765
 766         return data->port_id;
 767
 768 error:
 769         if (internal)
 770                 free(internal->dev_name);
 771         rte_free(vring_state);
 772         rte_free(eth_addr);
 773         if (eth_dev)
 774                 rte_eth_dev_release_port(eth_dev);
 775         rte_free(internal);
 776         rte_free(list);
 777         rte_free(data);
 778
 779         return -1;
 780 }
 781
 782 static inline int
 783 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
 784 {
 785         const char **iface_name = extra_args;
 786
 787         if (value == NULL)
 788                 return -1;
 789
 790         *iface_name = value;
 791
 792         return 0;
 793 }
 794
 795 static inline int
 796 open_queues(const char *key __rte_unused, const char *value, void *extra_args)
 797 {
 798         uint16_t *q = extra_args;
 799
 800         if (value == NULL || extra_args == NULL)
 801                 return -EINVAL;
 802
 803         *q = (uint16_t)strtoul(value, NULL, 0);
 804         if (*q == USHRT_MAX && errno == ERANGE)
 805                 return -1;
 806
 807         if (*q > RTE_MAX_QUEUES_PER_PORT)
 808                 return -1;
 809
 810         return 0;
 811 }
 812
 813 static int
 814 rte_pmd_vhost_devinit(const char *name, const char *params)
 815 {
 816         struct rte_kvargs *kvlist = NULL;
 817         int ret = 0;
 818         char *iface_name;
 819         uint16_t queues;
 820
 821         RTE_LOG(INFO, PMD, "Initializing pmd_vhost for %s\n", name);
 822
 823         kvlist = rte_kvargs_parse(params, valid_arguments);
 824         if (kvlist == NULL)
 825                 return -1;
 826
 827         if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
 828                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
 829                                          &open_iface, &iface_name);
 830                 if (ret < 0)
 831                         goto out_free;
 832         } else {
 833                 ret = -1;
 834                 goto out_free;
 835         }
 836
 837         if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
 838                 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
 839                                          &open_queues, &queues);
 840                 if (ret < 0)
 841                         goto out_free;
 842
 843         } else
 844                 queues = 1;
 845
 846         eth_dev_vhost_create(name, iface_name, queues, rte_socket_id());
 847
 848 out_free:
 849         rte_kvargs_free(kvlist);
 850         return ret;
 851 }
 852
 853 static int
 854 rte_pmd_vhost_devuninit(const char *name)
 855 {
 856         struct rte_eth_dev *eth_dev = NULL;
 857         struct pmd_internal *internal;
 858         struct internal_list *list;
 859         unsigned int i;
 860
 861         RTE_LOG(INFO, PMD, "Un-Initializing pmd_vhost for %s\n", name);
 862
 863         /* find an ethdev entry */
 864         eth_dev = rte_eth_dev_allocated(name);
 865         if (eth_dev == NULL)
 866                 return -ENODEV;
 867
 868         internal = eth_dev->data->dev_private;
 869         if (internal == NULL)
 870                 return -ENODEV;
 871
 872         list = find_internal_resource(internal->iface_name);
 873         if (list == NULL)
 874                 return -ENODEV;
 875
 876         pthread_mutex_lock(&internal_list_lock);
 877         TAILQ_REMOVE(&internal_list, list, next);
 878         pthread_mutex_unlock(&internal_list_lock);
 879         rte_free(list);
 880
 881         eth_dev_stop(eth_dev);
 882
 883         rte_free(vring_states[eth_dev->data->port_id]);
 884         vring_states[eth_dev->data->port_id] = NULL;
 885
 886         free(internal->dev_name);
 887         free(internal->iface_name);
 888
 889         for (i = 0; i < eth_dev->data->nb_rx_queues; i++)
 890                 rte_free(eth_dev->data->rx_queues[i]);
 891         for (i = 0; i < eth_dev->data->nb_tx_queues; i++)
 892                 rte_free(eth_dev->data->tx_queues[i]);
 893
 894         rte_free(eth_dev->data->mac_addrs);
 895         rte_free(eth_dev->data);
 896         rte_free(internal);
 897
 898         rte_eth_dev_release_port(eth_dev);
 899
 900         return 0;
 901 }
 902
 903 static struct rte_driver pmd_vhost_drv = {
 904         .name = "eth_vhost",
 905         .type = PMD_VDEV,
 906         .init = rte_pmd_vhost_devinit,
 907         .uninit = rte_pmd_vhost_devuninit,
 908 };
 909
 910 PMD_REGISTER_DRIVER(pmd_vhost_drv);