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_vdev.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 #define ETH_VHOST_CLIENT_ARG            "client"
  54 #define ETH_VHOST_DEQUEUE_ZERO_COPY     "dequeue-zero-copy"
  55
  56 static const char *drivername = "VHOST PMD";
  57
  58 static const char *valid_arguments[] = {
  59         ETH_VHOST_IFACE_ARG,
  60         ETH_VHOST_QUEUES_ARG,
  61         ETH_VHOST_CLIENT_ARG,
  62         ETH_VHOST_DEQUEUE_ZERO_COPY,
  63         NULL
  64 };
  65
  66 static struct ether_addr base_eth_addr = {
  67         .addr_bytes = {
  68                 0x56 /* V */,
  69                 0x48 /* H */,
  70                 0x4F /* O */,
  71                 0x53 /* S */,
  72                 0x54 /* T */,
  73                 0x00
  74         }
  75 };
  76
  77 enum vhost_xstats_pkts {
  78         VHOST_UNDERSIZE_PKT = 0,
  79         VHOST_64_PKT,
  80         VHOST_65_TO_127_PKT,
  81         VHOST_128_TO_255_PKT,
  82         VHOST_256_TO_511_PKT,
  83         VHOST_512_TO_1023_PKT,
  84         VHOST_1024_TO_1522_PKT,
  85         VHOST_1523_TO_MAX_PKT,
  86         VHOST_BROADCAST_PKT,
  87         VHOST_MULTICAST_PKT,
  88         VHOST_UNICAST_PKT,
  89         VHOST_ERRORS_PKT,
  90         VHOST_ERRORS_FRAGMENTED,
  91         VHOST_ERRORS_JABBER,
  92         VHOST_UNKNOWN_PROTOCOL,
  93         VHOST_XSTATS_MAX,
  94 };
  95
  96 struct vhost_stats {
  97         uint64_t pkts;
  98         uint64_t bytes;
  99         uint64_t missed_pkts;
 100         uint64_t xstats[VHOST_XSTATS_MAX];
 101 };
 102
 103 struct vhost_queue {
 104         int vid;
 105         rte_atomic32_t allow_queuing;
 106         rte_atomic32_t while_queuing;
 107         struct pmd_internal *internal;
 108         struct rte_mempool *mb_pool;
 109         uint8_t port;
 110         uint16_t virtqueue_id;
 111         struct vhost_stats stats;
 112 };
 113
 114 struct pmd_internal {
 115         rte_atomic32_t dev_attached;
 116         char *dev_name;
 117         char *iface_name;
 118         uint16_t max_queues;
 119         rte_atomic32_t started;
 120 };
 121
 122 struct internal_list {
 123         TAILQ_ENTRY(internal_list) next;
 124         struct rte_eth_dev *eth_dev;
 125 };
 126
 127 TAILQ_HEAD(internal_list_head, internal_list);
 128 static struct internal_list_head internal_list =
 129         TAILQ_HEAD_INITIALIZER(internal_list);
 130
 131 static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
 132
 133 static rte_atomic16_t nb_started_ports;
 134 static pthread_t session_th;
 135
 136 static struct rte_eth_link pmd_link = {
 137                 .link_speed = 10000,
 138                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 139                 .link_status = ETH_LINK_DOWN
 140 };
 141
 142 struct rte_vhost_vring_state {
 143         rte_spinlock_t lock;
 144
 145         bool cur[RTE_MAX_QUEUES_PER_PORT * 2];
 146         bool seen[RTE_MAX_QUEUES_PER_PORT * 2];
 147         unsigned int index;
 148         unsigned int max_vring;
 149 };
 150
 151 static struct rte_vhost_vring_state *vring_states[RTE_MAX_ETHPORTS];
 152
 153 #define VHOST_XSTATS_NAME_SIZE 64
 154
 155 struct vhost_xstats_name_off {
 156         char name[VHOST_XSTATS_NAME_SIZE];
 157         uint64_t offset;
 158 };
 159
 160 /* [rx]_is prepended to the name string here */
 161 static const struct vhost_xstats_name_off vhost_rxport_stat_strings[] = {
 162         {"good_packets",
 163          offsetof(struct vhost_queue, stats.pkts)},
 164         {"total_bytes",
 165          offsetof(struct vhost_queue, stats.bytes)},
 166         {"missed_pkts",
 167          offsetof(struct vhost_queue, stats.missed_pkts)},
 168         {"broadcast_packets",
 169          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
 170         {"multicast_packets",
 171          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
 172         {"unicast_packets",
 173          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
 174          {"undersize_packets",
 175          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
 176         {"size_64_packets",
 177          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
 178         {"size_65_to_127_packets",
 179          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
 180         {"size_128_to_255_packets",
 181          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
 182         {"size_256_to_511_packets",
 183          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
 184         {"size_512_to_1023_packets",
 185          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
 186         {"size_1024_to_1522_packets",
 187          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
 188         {"size_1523_to_max_packets",
 189          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
 190         {"errors_with_bad_CRC",
 191          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
 192         {"fragmented_errors",
 193          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_FRAGMENTED])},
 194         {"jabber_errors",
 195          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_JABBER])},
 196         {"unknown_protos_packets",
 197          offsetof(struct vhost_queue, stats.xstats[VHOST_UNKNOWN_PROTOCOL])},
 198 };
 199
 200 /* [tx]_ is prepended to the name string here */
 201 static const struct vhost_xstats_name_off vhost_txport_stat_strings[] = {
 202         {"good_packets",
 203          offsetof(struct vhost_queue, stats.pkts)},
 204         {"total_bytes",
 205          offsetof(struct vhost_queue, stats.bytes)},
 206         {"missed_pkts",
 207          offsetof(struct vhost_queue, stats.missed_pkts)},
 208         {"broadcast_packets",
 209          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
 210         {"multicast_packets",
 211          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
 212         {"unicast_packets",
 213          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
 214         {"undersize_packets",
 215          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
 216         {"size_64_packets",
 217          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
 218         {"size_65_to_127_packets",
 219          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
 220         {"size_128_to_255_packets",
 221          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
 222         {"size_256_to_511_packets",
 223          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
 224         {"size_512_to_1023_packets",
 225          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
 226         {"size_1024_to_1522_packets",
 227          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
 228         {"size_1523_to_max_packets",
 229          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
 230         {"errors_with_bad_CRC",
 231          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
 232 };
 233
 234 #define VHOST_NB_XSTATS_RXPORT (sizeof(vhost_rxport_stat_strings) / \
 235                                 sizeof(vhost_rxport_stat_strings[0]))
 236
 237 #define VHOST_NB_XSTATS_TXPORT (sizeof(vhost_txport_stat_strings) / \
 238                                 sizeof(vhost_txport_stat_strings[0]))
 239
 240 static void
 241 vhost_dev_xstats_reset(struct rte_eth_dev *dev)
 242 {
 243         struct vhost_queue *vq = NULL;
 244         unsigned int i = 0;
 245
 246         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 247                 vq = dev->data->rx_queues[i];
 248                 if (!vq)
 249                         continue;
 250                 memset(&vq->stats, 0, sizeof(vq->stats));
 251         }
 252         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 253                 vq = dev->data->tx_queues[i];
 254                 if (!vq)
 255                         continue;
 256                 memset(&vq->stats, 0, sizeof(vq->stats));
 257         }
 258 }
 259
 260 static int
 261 vhost_dev_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
 262                            struct rte_eth_xstat_name *xstats_names,
 263                            unsigned int limit __rte_unused)
 264 {
 265         unsigned int t = 0;
 266         int count = 0;
 267         int nstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
 268
 269         if (!xstats_names)
 270                 return nstats;
 271         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
 272                 snprintf(xstats_names[count].name,
 273                          sizeof(xstats_names[count].name),
 274                          "rx_%s", vhost_rxport_stat_strings[t].name);
 275                 count++;
 276         }
 277         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
 278                 snprintf(xstats_names[count].name,
 279                          sizeof(xstats_names[count].name),
 280                          "tx_%s", vhost_txport_stat_strings[t].name);
 281                 count++;
 282         }
 283         return count;
 284 }
 285
 286 static int
 287 vhost_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 288                      unsigned int n)
 289 {
 290         unsigned int i;
 291         unsigned int t;
 292         unsigned int count = 0;
 293         struct vhost_queue *vq = NULL;
 294         unsigned int nxstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
 295
 296         if (n < nxstats)
 297                 return nxstats;
 298
 299         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 300                 vq = dev->data->rx_queues[i];
 301                 if (!vq)
 302                         continue;
 303                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
 304                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
 305                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
 306         }
 307         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 308                 vq = dev->data->tx_queues[i];
 309                 if (!vq)
 310                         continue;
 311                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
 312                                 + vq->stats.missed_pkts
 313                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
 314                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
 315         }
 316         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
 317                 xstats[count].value = 0;
 318                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 319                         vq = dev->data->rx_queues[i];
 320                         if (!vq)
 321                                 continue;
 322                         xstats[count].value +=
 323                                 *(uint64_t *)(((char *)vq)
 324                                 + vhost_rxport_stat_strings[t].offset);
 325                 }
 326                 xstats[count].id = count;
 327                 count++;
 328         }
 329         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
 330                 xstats[count].value = 0;
 331                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
 332                         vq = dev->data->tx_queues[i];
 333                         if (!vq)
 334                                 continue;
 335                         xstats[count].value +=
 336                                 *(uint64_t *)(((char *)vq)
 337                                 + vhost_txport_stat_strings[t].offset);
 338                 }
 339                 xstats[count].id = count;
 340                 count++;
 341         }
 342         return count;
 343 }
 344
 345 static inline void
 346 vhost_count_multicast_broadcast(struct vhost_queue *vq,
 347                                 struct rte_mbuf *mbuf)
 348 {
 349         struct ether_addr *ea = NULL;
 350         struct vhost_stats *pstats = &vq->stats;
 351
 352         ea = rte_pktmbuf_mtod(mbuf, struct ether_addr *);
 353         if (is_multicast_ether_addr(ea)) {
 354                 if (is_broadcast_ether_addr(ea))
 355                         pstats->xstats[VHOST_BROADCAST_PKT]++;
 356                 else
 357                         pstats->xstats[VHOST_MULTICAST_PKT]++;
 358         }
 359 }
 360
 361 static void
 362 vhost_update_packet_xstats(struct vhost_queue *vq,
 363                            struct rte_mbuf **bufs,
 364                            uint16_t count)
 365 {
 366         uint32_t pkt_len = 0;
 367         uint64_t i = 0;
 368         uint64_t index;
 369         struct vhost_stats *pstats = &vq->stats;
 370
 371         for (i = 0; i < count ; i++) {
 372                 pkt_len = bufs[i]->pkt_len;
 373                 if (pkt_len == 64) {
 374                         pstats->xstats[VHOST_64_PKT]++;
 375                 } else if (pkt_len > 64 && pkt_len < 1024) {
 376                         index = (sizeof(pkt_len) * 8)
 377                                 - __builtin_clz(pkt_len) - 5;
 378                         pstats->xstats[index]++;
 379                 } else {
 380                         if (pkt_len < 64)
 381                                 pstats->xstats[VHOST_UNDERSIZE_PKT]++;
 382                         else if (pkt_len <= 1522)
 383                                 pstats->xstats[VHOST_1024_TO_1522_PKT]++;
 384                         else if (pkt_len > 1522)
 385                                 pstats->xstats[VHOST_1523_TO_MAX_PKT]++;
 386                 }
 387                 vhost_count_multicast_broadcast(vq, bufs[i]);
 388         }
 389 }
 390
 391 static uint16_t
 392 eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 393 {
 394         struct vhost_queue *r = q;
 395         uint16_t i, nb_rx = 0;
 396
 397         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 398                 return 0;
 399
 400         rte_atomic32_set(&r->while_queuing, 1);
 401
 402         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 403                 goto out;
 404
 405         /* Dequeue packets from guest TX queue */
 406         nb_rx = rte_vhost_dequeue_burst(r->vid,
 407                         r->virtqueue_id, r->mb_pool, bufs, nb_bufs);
 408
 409         r->stats.pkts += nb_rx;
 410
 411         for (i = 0; likely(i < nb_rx); i++) {
 412                 bufs[i]->port = r->port;
 413                 r->stats.bytes += bufs[i]->pkt_len;
 414         }
 415
 416         vhost_update_packet_xstats(r, bufs, nb_rx);
 417
 418 out:
 419         rte_atomic32_set(&r->while_queuing, 0);
 420
 421         return nb_rx;
 422 }
 423
 424 static uint16_t
 425 eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 426 {
 427         struct vhost_queue *r = q;
 428         uint16_t i, nb_tx = 0;
 429
 430         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 431                 return 0;
 432
 433         rte_atomic32_set(&r->while_queuing, 1);
 434
 435         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 436                 goto out;
 437
 438         /* Enqueue packets to guest RX queue */
 439         nb_tx = rte_vhost_enqueue_burst(r->vid,
 440                         r->virtqueue_id, bufs, nb_bufs);
 441
 442         r->stats.pkts += nb_tx;
 443         r->stats.missed_pkts += nb_bufs - nb_tx;
 444
 445         for (i = 0; likely(i < nb_tx); i++)
 446                 r->stats.bytes += bufs[i]->pkt_len;
 447
 448         vhost_update_packet_xstats(r, bufs, nb_tx);
 449
 450         /* According to RFC2863 page42 section ifHCOutMulticastPkts and
 451          * ifHCOutBroadcastPkts, the counters "multicast" and "broadcast"
 452          * are increased when packets are not transmitted successfully.
 453          */
 454         for (i = nb_tx; i < nb_bufs; i++)
 455                 vhost_count_multicast_broadcast(r, bufs[i]);
 456
 457         for (i = 0; likely(i < nb_tx); i++)
 458                 rte_pktmbuf_free(bufs[i]);
 459 out:
 460         rte_atomic32_set(&r->while_queuing, 0);
 461
 462         return nb_tx;
 463 }
 464
 465 static int
 466 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 467 {
 468         return 0;
 469 }
 470
 471 static inline struct internal_list *
 472 find_internal_resource(char *ifname)
 473 {
 474         int found = 0;
 475         struct internal_list *list;
 476         struct pmd_internal *internal;
 477
 478         if (ifname == NULL)
 479                 return NULL;
 480
 481         pthread_mutex_lock(&internal_list_lock);
 482
 483         TAILQ_FOREACH(list, &internal_list, next) {
 484                 internal = list->eth_dev->data->dev_private;
 485                 if (!strcmp(internal->iface_name, ifname)) {
 486                         found = 1;
 487                         break;
 488                 }
 489         }
 490
 491         pthread_mutex_unlock(&internal_list_lock);
 492
 493         if (!found)
 494                 return NULL;
 495
 496         return list;
 497 }
 498
 499 static void
 500 update_queuing_status(struct rte_eth_dev *dev)
 501 {
 502         struct pmd_internal *internal = dev->data->dev_private;
 503         struct vhost_queue *vq;
 504         unsigned int i;
 505         int allow_queuing = 1;
 506
 507         if (rte_atomic32_read(&internal->started) == 0 ||
 508             rte_atomic32_read(&internal->dev_attached) == 0)
 509                 allow_queuing = 0;
 510
 511         /* Wait until rx/tx_pkt_burst stops accessing vhost device */
 512         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 513                 vq = dev->data->rx_queues[i];
 514                 if (vq == NULL)
 515                         continue;
 516                 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
 517                 while (rte_atomic32_read(&vq->while_queuing))
 518                         rte_pause();
 519         }
 520
 521         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 522                 vq = dev->data->tx_queues[i];
 523                 if (vq == NULL)
 524                         continue;
 525                 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
 526                 while (rte_atomic32_read(&vq->while_queuing))
 527                         rte_pause();
 528         }
 529 }
 530
 531 static int
 532 new_device(int vid)
 533 {
 534         struct rte_eth_dev *eth_dev;
 535         struct internal_list *list;
 536         struct pmd_internal *internal;
 537         struct vhost_queue *vq;
 538         unsigned i;
 539         char ifname[PATH_MAX];
 540 #ifdef RTE_LIBRTE_VHOST_NUMA
 541         int newnode;
 542 #endif
 543
 544         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 545         list = find_internal_resource(ifname);
 546         if (list == NULL) {
 547                 RTE_LOG(INFO, PMD, "Invalid device name: %s\n", ifname);
 548                 return -1;
 549         }
 550
 551         eth_dev = list->eth_dev;
 552         internal = eth_dev->data->dev_private;
 553
 554 #ifdef RTE_LIBRTE_VHOST_NUMA
 555         newnode = rte_vhost_get_numa_node(vid);
 556         if (newnode >= 0)
 557                 eth_dev->data->numa_node = newnode;
 558 #endif
 559
 560         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 561                 vq = eth_dev->data->rx_queues[i];
 562                 if (vq == NULL)
 563                         continue;
 564                 vq->vid = vid;
 565                 vq->internal = internal;
 566                 vq->port = eth_dev->data->port_id;
 567         }
 568         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 569                 vq = eth_dev->data->tx_queues[i];
 570                 if (vq == NULL)
 571                         continue;
 572                 vq->vid = vid;
 573                 vq->internal = internal;
 574                 vq->port = eth_dev->data->port_id;
 575         }
 576
 577         for (i = 0; i < rte_vhost_get_queue_num(vid) * VIRTIO_QNUM; i++)
 578                 rte_vhost_enable_guest_notification(vid, i, 0);
 579
 580         eth_dev->data->dev_link.link_status = ETH_LINK_UP;
 581
 582         rte_atomic32_set(&internal->dev_attached, 1);
 583         update_queuing_status(eth_dev);
 584
 585         RTE_LOG(INFO, PMD, "New connection established\n");
 586
 587         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 588
 589         return 0;
 590 }
 591
 592 static void
 593 destroy_device(int vid)
 594 {
 595         struct rte_eth_dev *eth_dev;
 596         struct pmd_internal *internal;
 597         struct vhost_queue *vq;
 598         struct internal_list *list;
 599         char ifname[PATH_MAX];
 600         unsigned i;
 601         struct rte_vhost_vring_state *state;
 602
 603         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 604         list = find_internal_resource(ifname);
 605         if (list == NULL) {
 606                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
 607                 return;
 608         }
 609         eth_dev = list->eth_dev;
 610         internal = eth_dev->data->dev_private;
 611
 612         rte_atomic32_set(&internal->dev_attached, 0);
 613         update_queuing_status(eth_dev);
 614
 615         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
 616
 617         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 618                 vq = eth_dev->data->rx_queues[i];
 619                 if (vq == NULL)
 620                         continue;
 621                 vq->vid = -1;
 622         }
 623         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 624                 vq = eth_dev->data->tx_queues[i];
 625                 if (vq == NULL)
 626                         continue;
 627                 vq->vid = -1;
 628         }
 629
 630         state = vring_states[eth_dev->data->port_id];
 631         rte_spinlock_lock(&state->lock);
 632         for (i = 0; i <= state->max_vring; i++) {
 633                 state->cur[i] = false;
 634                 state->seen[i] = false;
 635         }
 636         state->max_vring = 0;
 637         rte_spinlock_unlock(&state->lock);
 638
 639         RTE_LOG(INFO, PMD, "Connection closed\n");
 640
 641         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 642 }
 643
 644 static int
 645 vring_state_changed(int vid, uint16_t vring, int enable)
 646 {
 647         struct rte_vhost_vring_state *state;
 648         struct rte_eth_dev *eth_dev;
 649         struct internal_list *list;
 650         char ifname[PATH_MAX];
 651
 652         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 653         list = find_internal_resource(ifname);
 654         if (list == NULL) {
 655                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
 656                 return -1;
 657         }
 658
 659         eth_dev = list->eth_dev;
 660         /* won't be NULL */
 661         state = vring_states[eth_dev->data->port_id];
 662         rte_spinlock_lock(&state->lock);
 663         state->cur[vring] = enable;
 664         state->max_vring = RTE_MAX(vring, state->max_vring);
 665         rte_spinlock_unlock(&state->lock);
 666
 667         RTE_LOG(INFO, PMD, "vring%u is %s\n",
 668                         vring, enable ? "enabled" : "disabled");
 669
 670         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE, NULL);
 671
 672         return 0;
 673 }
 674
 675 int
 676 rte_eth_vhost_get_queue_event(uint8_t port_id,
 677                 struct rte_eth_vhost_queue_event *event)
 678 {
 679         struct rte_vhost_vring_state *state;
 680         unsigned int i;
 681         int idx;
 682
 683         if (port_id >= RTE_MAX_ETHPORTS) {
 684                 RTE_LOG(ERR, PMD, "Invalid port id\n");
 685                 return -1;
 686         }
 687
 688         state = vring_states[port_id];
 689         if (!state) {
 690                 RTE_LOG(ERR, PMD, "Unused port\n");
 691                 return -1;
 692         }
 693
 694         rte_spinlock_lock(&state->lock);
 695         for (i = 0; i <= state->max_vring; i++) {
 696                 idx = state->index++ % (state->max_vring + 1);
 697
 698                 if (state->cur[idx] != state->seen[idx]) {
 699                         state->seen[idx] = state->cur[idx];
 700                         event->queue_id = idx / 2;
 701                         event->rx = idx & 1;
 702                         event->enable = state->cur[idx];
 703                         rte_spinlock_unlock(&state->lock);
 704                         return 0;
 705                 }
 706         }
 707         rte_spinlock_unlock(&state->lock);
 708
 709         return -1;
 710 }
 711
 712 int
 713 rte_eth_vhost_get_vid_from_port_id(uint8_t port_id)
 714 {
 715         struct internal_list *list;
 716         struct rte_eth_dev *eth_dev;
 717         struct vhost_queue *vq;
 718         int vid = -1;
 719
 720         if (!rte_eth_dev_is_valid_port(port_id))
 721                 return -1;
 722
 723         pthread_mutex_lock(&internal_list_lock);
 724
 725         TAILQ_FOREACH(list, &internal_list, next) {
 726                 eth_dev = list->eth_dev;
 727                 if (eth_dev->data->port_id == port_id) {
 728                         vq = eth_dev->data->rx_queues[0];
 729                         if (vq) {
 730                                 vid = vq->vid;
 731                         }
 732                         break;
 733                 }
 734         }
 735
 736         pthread_mutex_unlock(&internal_list_lock);
 737
 738         return vid;
 739 }
 740
 741 static void *
 742 vhost_driver_session(void *param __rte_unused)
 743 {
 744         static struct virtio_net_device_ops vhost_ops;
 745
 746         /* set vhost arguments */
 747         vhost_ops.new_device = new_device;
 748         vhost_ops.destroy_device = destroy_device;
 749         vhost_ops.vring_state_changed = vring_state_changed;
 750         if (rte_vhost_driver_callback_register(&vhost_ops) < 0)
 751                 RTE_LOG(ERR, PMD, "Can't register callbacks\n");
 752
 753         /* start event handling */
 754         rte_vhost_driver_session_start();
 755
 756         return NULL;
 757 }
 758
 759 static int
 760 vhost_driver_session_start(void)
 761 {
 762         int ret;
 763
 764         ret = pthread_create(&session_th,
 765                         NULL, vhost_driver_session, NULL);
 766         if (ret)
 767                 RTE_LOG(ERR, PMD, "Can't create a thread\n");
 768
 769         return ret;
 770 }
 771
 772 static void
 773 vhost_driver_session_stop(void)
 774 {
 775         int ret;
 776
 777         ret = pthread_cancel(session_th);
 778         if (ret)
 779                 RTE_LOG(ERR, PMD, "Can't cancel the thread\n");
 780
 781         ret = pthread_join(session_th, NULL);
 782         if (ret)
 783                 RTE_LOG(ERR, PMD, "Can't join the thread\n");
 784 }
 785
 786 static int
 787 eth_dev_start(struct rte_eth_dev *dev)
 788 {
 789         struct pmd_internal *internal = dev->data->dev_private;
 790
 791         rte_atomic32_set(&internal->started, 1);
 792         update_queuing_status(dev);
 793
 794         return 0;
 795 }
 796
 797 static void
 798 eth_dev_stop(struct rte_eth_dev *dev)
 799 {
 800         struct pmd_internal *internal = dev->data->dev_private;
 801
 802         rte_atomic32_set(&internal->started, 0);
 803         update_queuing_status(dev);
 804 }
 805
 806 static int
 807 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 808                    uint16_t nb_rx_desc __rte_unused,
 809                    unsigned int socket_id,
 810                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 811                    struct rte_mempool *mb_pool)
 812 {
 813         struct vhost_queue *vq;
 814
 815         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
 816                         RTE_CACHE_LINE_SIZE, socket_id);
 817         if (vq == NULL) {
 818                 RTE_LOG(ERR, PMD, "Failed to allocate memory for rx queue\n");
 819                 return -ENOMEM;
 820         }
 821
 822         vq->mb_pool = mb_pool;
 823         vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
 824         dev->data->rx_queues[rx_queue_id] = vq;
 825
 826         return 0;
 827 }
 828
 829 static int
 830 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 831                    uint16_t nb_tx_desc __rte_unused,
 832                    unsigned int socket_id,
 833                    const struct rte_eth_txconf *tx_conf __rte_unused)
 834 {
 835         struct vhost_queue *vq;
 836
 837         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
 838                         RTE_CACHE_LINE_SIZE, socket_id);
 839         if (vq == NULL) {
 840                 RTE_LOG(ERR, PMD, "Failed to allocate memory for tx queue\n");
 841                 return -ENOMEM;
 842         }
 843
 844         vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
 845         dev->data->tx_queues[tx_queue_id] = vq;
 846
 847         return 0;
 848 }
 849
 850 static void
 851 eth_dev_info(struct rte_eth_dev *dev,
 852              struct rte_eth_dev_info *dev_info)
 853 {
 854         struct pmd_internal *internal;
 855
 856         internal = dev->data->dev_private;
 857         if (internal == NULL) {
 858                 RTE_LOG(ERR, PMD, "Invalid device specified\n");
 859                 return;
 860         }
 861
 862         dev_info->driver_name = drivername;
 863         dev_info->max_mac_addrs = 1;
 864         dev_info->max_rx_pktlen = (uint32_t)-1;
 865         dev_info->max_rx_queues = internal->max_queues;
 866         dev_info->max_tx_queues = internal->max_queues;
 867         dev_info->min_rx_bufsize = 0;
 868 }
 869
 870 static void
 871 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 872 {
 873         unsigned i;
 874         unsigned long rx_total = 0, tx_total = 0, tx_missed_total = 0;
 875         unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
 876         struct vhost_queue *vq;
 877
 878         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 879                         i < dev->data->nb_rx_queues; i++) {
 880                 if (dev->data->rx_queues[i] == NULL)
 881                         continue;
 882                 vq = dev->data->rx_queues[i];
 883                 stats->q_ipackets[i] = vq->stats.pkts;
 884                 rx_total += stats->q_ipackets[i];
 885
 886                 stats->q_ibytes[i] = vq->stats.bytes;
 887                 rx_total_bytes += stats->q_ibytes[i];
 888         }
 889
 890         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 891                         i < dev->data->nb_tx_queues; i++) {
 892                 if (dev->data->tx_queues[i] == NULL)
 893                         continue;
 894                 vq = dev->data->tx_queues[i];
 895                 stats->q_opackets[i] = vq->stats.pkts;
 896                 tx_missed_total += vq->stats.missed_pkts;
 897                 tx_total += stats->q_opackets[i];
 898
 899                 stats->q_obytes[i] = vq->stats.bytes;
 900                 tx_total_bytes += stats->q_obytes[i];
 901         }
 902
 903         stats->ipackets = rx_total;
 904         stats->opackets = tx_total;
 905         stats->oerrors = tx_missed_total;
 906         stats->ibytes = rx_total_bytes;
 907         stats->obytes = tx_total_bytes;
 908 }
 909
 910 static void
 911 eth_stats_reset(struct rte_eth_dev *dev)
 912 {
 913         struct vhost_queue *vq;
 914         unsigned i;
 915
 916         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 917                 if (dev->data->rx_queues[i] == NULL)
 918                         continue;
 919                 vq = dev->data->rx_queues[i];
 920                 vq->stats.pkts = 0;
 921                 vq->stats.bytes = 0;
 922         }
 923         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 924                 if (dev->data->tx_queues[i] == NULL)
 925                         continue;
 926                 vq = dev->data->tx_queues[i];
 927                 vq->stats.pkts = 0;
 928                 vq->stats.bytes = 0;
 929                 vq->stats.missed_pkts = 0;
 930         }
 931 }
 932
 933 static void
 934 eth_queue_release(void *q)
 935 {
 936         rte_free(q);
 937 }
 938
 939 static int
 940 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 941                 int wait_to_complete __rte_unused)
 942 {
 943         return 0;
 944 }
 945
 946 /**
 947  * Disable features in feature_mask. Returns 0 on success.
 948  */
 949 int
 950 rte_eth_vhost_feature_disable(uint64_t feature_mask)
 951 {
 952         return rte_vhost_feature_disable(feature_mask);
 953 }
 954
 955 /**
 956  * Enable features in feature_mask. Returns 0 on success.
 957  */
 958 int
 959 rte_eth_vhost_feature_enable(uint64_t feature_mask)
 960 {
 961         return rte_vhost_feature_enable(feature_mask);
 962 }
 963
 964 /* Returns currently supported vhost features */
 965 uint64_t
 966 rte_eth_vhost_feature_get(void)
 967 {
 968         return rte_vhost_feature_get();
 969 }
 970
 971 static const struct eth_dev_ops ops = {
 972         .dev_start = eth_dev_start,
 973         .dev_stop = eth_dev_stop,
 974         .dev_configure = eth_dev_configure,
 975         .dev_infos_get = eth_dev_info,
 976         .rx_queue_setup = eth_rx_queue_setup,
 977         .tx_queue_setup = eth_tx_queue_setup,
 978         .rx_queue_release = eth_queue_release,
 979         .tx_queue_release = eth_queue_release,
 980         .link_update = eth_link_update,
 981         .stats_get = eth_stats_get,
 982         .stats_reset = eth_stats_reset,
 983         .xstats_reset = vhost_dev_xstats_reset,
 984         .xstats_get = vhost_dev_xstats_get,
 985         .xstats_get_names = vhost_dev_xstats_get_names,
 986 };
 987
 988 static int
 989 eth_dev_vhost_create(const char *name, char *iface_name, int16_t queues,
 990                      const unsigned numa_node, uint64_t flags)
 991 {
 992         struct rte_eth_dev_data *data = NULL;
 993         struct pmd_internal *internal = NULL;
 994         struct rte_eth_dev *eth_dev = NULL;
 995         struct ether_addr *eth_addr = NULL;
 996         struct rte_vhost_vring_state *vring_state = NULL;
 997         struct internal_list *list = NULL;
 998
 999         RTE_LOG(INFO, PMD, "Creating VHOST-USER backend on numa socket %u\n",
1000                 numa_node);
1001
1002         /* now do all data allocation - for eth_dev structure, dummy pci driver
1003          * and internal (private) data
1004          */
1005         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
1006         if (data == NULL)
1007                 goto error;
1008
1009         internal = rte_zmalloc_socket(name, sizeof(*internal), 0, numa_node);
1010         if (internal == NULL)
1011                 goto error;
1012
1013         list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
1014         if (list == NULL)
1015                 goto error;
1016
1017         /* reserve an ethdev entry */
1018         eth_dev = rte_eth_dev_allocate(name);
1019         if (eth_dev == NULL)
1020                 goto error;
1021
1022         eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
1023         if (eth_addr == NULL)
1024                 goto error;
1025         *eth_addr = base_eth_addr;
1026         eth_addr->addr_bytes[5] = eth_dev->data->port_id;
1027
1028         vring_state = rte_zmalloc_socket(name,
1029                         sizeof(*vring_state), 0, numa_node);
1030         if (vring_state == NULL)
1031                 goto error;
1032
1033         /* now put it all together
1034          * - store queue data in internal,
1035          * - store numa_node info in ethdev data
1036          * - point eth_dev_data to internals
1037          * - and point eth_dev structure to new eth_dev_data structure
1038          */
1039         internal->dev_name = strdup(name);
1040         if (internal->dev_name == NULL)
1041                 goto error;
1042         internal->iface_name = strdup(iface_name);
1043         if (internal->iface_name == NULL)
1044                 goto error;
1045
1046         list->eth_dev = eth_dev;
1047         pthread_mutex_lock(&internal_list_lock);
1048         TAILQ_INSERT_TAIL(&internal_list, list, next);
1049         pthread_mutex_unlock(&internal_list_lock);
1050
1051         rte_spinlock_init(&vring_state->lock);
1052         vring_states[eth_dev->data->port_id] = vring_state;
1053
1054         data->dev_private = internal;
1055         data->port_id = eth_dev->data->port_id;
1056         memmove(data->name, eth_dev->data->name, sizeof(data->name));
1057         data->nb_rx_queues = queues;
1058         data->nb_tx_queues = queues;
1059         internal->max_queues = queues;
1060         data->dev_link = pmd_link;
1061         data->mac_addrs = eth_addr;
1062
1063         /* We'll replace the 'data' originally allocated by eth_dev. So the
1064          * vhost PMD resources won't be shared between multi processes.
1065          */
1066         eth_dev->data = data;
1067         eth_dev->dev_ops = &ops;
1068         eth_dev->driver = NULL;
1069         data->dev_flags =
1070                 RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
1071         data->kdrv = RTE_KDRV_NONE;
1072         data->drv_name = internal->dev_name;
1073         data->numa_node = numa_node;
1074
1075         /* finally assign rx and tx ops */
1076         eth_dev->rx_pkt_burst = eth_vhost_rx;
1077         eth_dev->tx_pkt_burst = eth_vhost_tx;
1078
1079         if (rte_vhost_driver_register(iface_name, flags))
1080                 goto error;
1081
1082         /* We need only one message handling thread */
1083         if (rte_atomic16_add_return(&nb_started_ports, 1) == 1) {
1084                 if (vhost_driver_session_start())
1085                         goto error;
1086         }
1087
1088         return data->port_id;
1089
1090 error:
1091         if (internal)
1092                 free(internal->dev_name);
1093         rte_free(vring_state);
1094         rte_free(eth_addr);
1095         if (eth_dev)
1096                 rte_eth_dev_release_port(eth_dev);
1097         rte_free(internal);
1098         rte_free(list);
1099         rte_free(data);
1100
1101         return -1;
1102 }
1103
1104 static inline int
1105 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
1106 {
1107         const char **iface_name = extra_args;
1108
1109         if (value == NULL)
1110                 return -1;
1111
1112         *iface_name = value;
1113
1114         return 0;
1115 }
1116
1117 static inline int
1118 open_int(const char *key __rte_unused, const char *value, void *extra_args)
1119 {
1120         uint16_t *n = extra_args;
1121
1122         if (value == NULL || extra_args == NULL)
1123                 return -EINVAL;
1124
1125         *n = (uint16_t)strtoul(value, NULL, 0);
1126         if (*n == USHRT_MAX && errno == ERANGE)
1127                 return -1;
1128
1129         return 0;
1130 }
1131
1132 static int
1133 rte_pmd_vhost_probe(const char *name, const char *params)
1134 {
1135         struct rte_kvargs *kvlist = NULL;
1136         int ret = 0;
1137         char *iface_name;
1138         uint16_t queues;
1139         uint64_t flags = 0;
1140         int client_mode = 0;
1141         int dequeue_zero_copy = 0;
1142
1143         RTE_LOG(INFO, PMD, "Initializing pmd_vhost for %s\n", name);
1144
1145         kvlist = rte_kvargs_parse(params, valid_arguments);
1146         if (kvlist == NULL)
1147                 return -1;
1148
1149         if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
1150                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
1151                                          &open_iface, &iface_name);
1152                 if (ret < 0)
1153                         goto out_free;
1154         } else {
1155                 ret = -1;
1156                 goto out_free;
1157         }
1158
1159         if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
1160                 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
1161                                          &open_int, &queues);
1162                 if (ret < 0 || queues > RTE_MAX_QUEUES_PER_PORT)
1163                         goto out_free;
1164
1165         } else
1166                 queues = 1;
1167
1168         if (rte_kvargs_count(kvlist, ETH_VHOST_CLIENT_ARG) == 1) {
1169                 ret = rte_kvargs_process(kvlist, ETH_VHOST_CLIENT_ARG,
1170                                          &open_int, &client_mode);
1171                 if (ret < 0)
1172                         goto out_free;
1173
1174                 if (client_mode)
1175                         flags |= RTE_VHOST_USER_CLIENT;
1176         }
1177
1178         if (rte_kvargs_count(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY) == 1) {
1179                 ret = rte_kvargs_process(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY,
1180                                          &open_int, &dequeue_zero_copy);
1181                 if (ret < 0)
1182                         goto out_free;
1183
1184                 if (dequeue_zero_copy)
1185                         flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
1186         }
1187
1188         eth_dev_vhost_create(name, iface_name, queues, rte_socket_id(), flags);
1189
1190 out_free:
1191         rte_kvargs_free(kvlist);
1192         return ret;
1193 }
1194
1195 static int
1196 rte_pmd_vhost_remove(const char *name)
1197 {
1198         struct rte_eth_dev *eth_dev = NULL;
1199         struct pmd_internal *internal;
1200         struct internal_list *list;
1201         unsigned int i;
1202
1203         RTE_LOG(INFO, PMD, "Un-Initializing pmd_vhost for %s\n", name);
1204
1205         /* find an ethdev entry */
1206         eth_dev = rte_eth_dev_allocated(name);
1207         if (eth_dev == NULL)
1208                 return -ENODEV;
1209
1210         internal = eth_dev->data->dev_private;
1211         if (internal == NULL)
1212                 return -ENODEV;
1213
1214         list = find_internal_resource(internal->iface_name);
1215         if (list == NULL)
1216                 return -ENODEV;
1217
1218         pthread_mutex_lock(&internal_list_lock);
1219         TAILQ_REMOVE(&internal_list, list, next);
1220         pthread_mutex_unlock(&internal_list_lock);
1221         rte_free(list);
1222
1223         eth_dev_stop(eth_dev);
1224
1225         rte_vhost_driver_unregister(internal->iface_name);
1226
1227         if (rte_atomic16_sub_return(&nb_started_ports, 1) == 0)
1228                 vhost_driver_session_stop();
1229
1230         rte_free(vring_states[eth_dev->data->port_id]);
1231         vring_states[eth_dev->data->port_id] = NULL;
1232
1233         free(internal->dev_name);
1234         free(internal->iface_name);
1235
1236         for (i = 0; i < eth_dev->data->nb_rx_queues; i++)
1237                 rte_free(eth_dev->data->rx_queues[i]);
1238         for (i = 0; i < eth_dev->data->nb_tx_queues; i++)
1239                 rte_free(eth_dev->data->tx_queues[i]);
1240
1241         rte_free(eth_dev->data->mac_addrs);
1242         rte_free(eth_dev->data);
1243         rte_free(internal);
1244
1245         rte_eth_dev_release_port(eth_dev);
1246
1247         return 0;
1248 }
1249
1250 static struct rte_vdev_driver pmd_vhost_drv = {
1251         .probe = rte_pmd_vhost_probe,
1252         .remove = rte_pmd_vhost_remove,
1253 };
1254
1255 RTE_PMD_REGISTER_VDEV(net_vhost, pmd_vhost_drv);
1256 RTE_PMD_REGISTER_ALIAS(net_vhost, eth_vhost);
1257 RTE_PMD_REGISTER_PARAM_STRING(net_vhost,
1258         "iface=<ifc> "
1259         "queues=<int>");