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