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
  37 #include <rte_mbuf.h>
  38 #include <rte_ethdev_driver.h>
  39 #include <rte_ethdev_vdev.h>
  40 #include <rte_malloc.h>
  41 #include <rte_memcpy.h>
  42 #include <rte_bus_vdev.h>
  43 #include <rte_kvargs.h>
  44 #include <rte_vhost.h>
  45 #include <rte_spinlock.h>
  46
  47 #include "rte_eth_vhost.h"
  48
  49 enum {VIRTIO_RXQ, VIRTIO_TXQ, VIRTIO_QNUM};
  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 #define ETH_VHOST_IOMMU_SUPPORT         "iommu-support"
  56 #define VHOST_MAX_PKT_BURST 32
  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         ETH_VHOST_IOMMU_SUPPORT,
  64         NULL
  65 };
  66
  67 static struct ether_addr base_eth_addr = {
  68         .addr_bytes = {
  69                 0x56 /* V */,
  70                 0x48 /* H */,
  71                 0x4F /* O */,
  72                 0x53 /* S */,
  73                 0x54 /* T */,
  74                 0x00
  75         }
  76 };
  77
  78 enum vhost_xstats_pkts {
  79         VHOST_UNDERSIZE_PKT = 0,
  80         VHOST_64_PKT,
  81         VHOST_65_TO_127_PKT,
  82         VHOST_128_TO_255_PKT,
  83         VHOST_256_TO_511_PKT,
  84         VHOST_512_TO_1023_PKT,
  85         VHOST_1024_TO_1522_PKT,
  86         VHOST_1523_TO_MAX_PKT,
  87         VHOST_BROADCAST_PKT,
  88         VHOST_MULTICAST_PKT,
  89         VHOST_UNICAST_PKT,
  90         VHOST_ERRORS_PKT,
  91         VHOST_ERRORS_FRAGMENTED,
  92         VHOST_ERRORS_JABBER,
  93         VHOST_UNKNOWN_PROTOCOL,
  94         VHOST_XSTATS_MAX,
  95 };
  96
  97 struct vhost_stats {
  98         uint64_t pkts;
  99         uint64_t bytes;
 100         uint64_t missed_pkts;
 101         uint64_t xstats[VHOST_XSTATS_MAX];
 102 };
 103
 104 struct vhost_queue {
 105         int vid;
 106         rte_atomic32_t allow_queuing;
 107         rte_atomic32_t while_queuing;
 108         struct pmd_internal *internal;
 109         struct rte_mempool *mb_pool;
 110         uint16_t port;
 111         uint16_t virtqueue_id;
 112         struct vhost_stats stats;
 113 };
 114
 115 struct pmd_internal {
 116         rte_atomic32_t dev_attached;
 117         char *dev_name;
 118         char *iface_name;
 119         uint16_t max_queues;
 120         uint16_t vid;
 121         rte_atomic32_t started;
 122         uint8_t vlan_strip;
 123 };
 124
 125 struct internal_list {
 126         TAILQ_ENTRY(internal_list) next;
 127         struct rte_eth_dev *eth_dev;
 128 };
 129
 130 TAILQ_HEAD(internal_list_head, internal_list);
 131 static struct internal_list_head internal_list =
 132         TAILQ_HEAD_INITIALIZER(internal_list);
 133
 134 static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
 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         uint16_t nb_receive = nb_bufs;
 397
 398         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 399                 return 0;
 400
 401         rte_atomic32_set(&r->while_queuing, 1);
 402
 403         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 404                 goto out;
 405
 406         /* Dequeue packets from guest TX queue */
 407         while (nb_receive) {
 408                 uint16_t nb_pkts;
 409                 uint16_t num = (uint16_t)RTE_MIN(nb_receive,
 410                                                  VHOST_MAX_PKT_BURST);
 411
 412                 nb_pkts = rte_vhost_dequeue_burst(r->vid, r->virtqueue_id,
 413                                                   r->mb_pool, &bufs[nb_rx],
 414                                                   num);
 415
 416                 nb_rx += nb_pkts;
 417                 nb_receive -= nb_pkts;
 418                 if (nb_pkts < num)
 419                         break;
 420         }
 421
 422         r->stats.pkts += nb_rx;
 423
 424         for (i = 0; likely(i < nb_rx); i++) {
 425                 bufs[i]->port = r->port;
 426                 bufs[i]->ol_flags = 0;
 427                 bufs[i]->vlan_tci = 0;
 428
 429                 if (r->internal->vlan_strip)
 430                         rte_vlan_strip(bufs[i]);
 431
 432                 r->stats.bytes += bufs[i]->pkt_len;
 433         }
 434
 435         vhost_update_packet_xstats(r, bufs, nb_rx);
 436
 437 out:
 438         rte_atomic32_set(&r->while_queuing, 0);
 439
 440         return nb_rx;
 441 }
 442
 443 static uint16_t
 444 eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 445 {
 446         struct vhost_queue *r = q;
 447         uint16_t i, nb_tx = 0;
 448         uint16_t nb_send = 0;
 449
 450         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 451                 return 0;
 452
 453         rte_atomic32_set(&r->while_queuing, 1);
 454
 455         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 456                 goto out;
 457
 458         for (i = 0; i < nb_bufs; i++) {
 459                 struct rte_mbuf *m = bufs[i];
 460
 461                 /* Do VLAN tag insertion */
 462                 if (m->ol_flags & PKT_TX_VLAN_PKT) {
 463                         int error = rte_vlan_insert(&m);
 464                         if (unlikely(error)) {
 465                                 rte_pktmbuf_free(m);
 466                                 continue;
 467                         }
 468                 }
 469
 470                 bufs[nb_send] = m;
 471                 ++nb_send;
 472         }
 473
 474         /* Enqueue packets to guest RX queue */
 475         while (nb_send) {
 476                 uint16_t nb_pkts;
 477                 uint16_t num = (uint16_t)RTE_MIN(nb_send,
 478                                                  VHOST_MAX_PKT_BURST);
 479
 480                 nb_pkts = rte_vhost_enqueue_burst(r->vid, r->virtqueue_id,
 481                                                   &bufs[nb_tx], num);
 482
 483                 nb_tx += nb_pkts;
 484                 nb_send -= nb_pkts;
 485                 if (nb_pkts < num)
 486                         break;
 487         }
 488
 489         r->stats.pkts += nb_tx;
 490         r->stats.missed_pkts += nb_bufs - nb_tx;
 491
 492         for (i = 0; likely(i < nb_tx); i++)
 493                 r->stats.bytes += bufs[i]->pkt_len;
 494
 495         vhost_update_packet_xstats(r, bufs, nb_tx);
 496
 497         /* According to RFC2863 page42 section ifHCOutMulticastPkts and
 498          * ifHCOutBroadcastPkts, the counters "multicast" and "broadcast"
 499          * are increased when packets are not transmitted successfully.
 500          */
 501         for (i = nb_tx; i < nb_bufs; i++)
 502                 vhost_count_multicast_broadcast(r, bufs[i]);
 503
 504         for (i = 0; likely(i < nb_tx); i++)
 505                 rte_pktmbuf_free(bufs[i]);
 506 out:
 507         rte_atomic32_set(&r->while_queuing, 0);
 508
 509         return nb_tx;
 510 }
 511
 512 static int
 513 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 514 {
 515         struct pmd_internal *internal = dev->data->dev_private;
 516         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
 517
 518         internal->vlan_strip = rxmode->hw_vlan_strip;
 519
 520         if (rxmode->hw_vlan_filter)
 521                 RTE_LOG(WARNING, PMD,
 522                         "vhost(%s): vlan filtering not available\n",
 523                         internal->dev_name);
 524
 525         return 0;
 526 }
 527
 528 static inline struct internal_list *
 529 find_internal_resource(char *ifname)
 530 {
 531         int found = 0;
 532         struct internal_list *list;
 533         struct pmd_internal *internal;
 534
 535         if (ifname == NULL)
 536                 return NULL;
 537
 538         pthread_mutex_lock(&internal_list_lock);
 539
 540         TAILQ_FOREACH(list, &internal_list, next) {
 541                 internal = list->eth_dev->data->dev_private;
 542                 if (!strcmp(internal->iface_name, ifname)) {
 543                         found = 1;
 544                         break;
 545                 }
 546         }
 547
 548         pthread_mutex_unlock(&internal_list_lock);
 549
 550         if (!found)
 551                 return NULL;
 552
 553         return list;
 554 }
 555
 556 static void
 557 update_queuing_status(struct rte_eth_dev *dev)
 558 {
 559         struct pmd_internal *internal = dev->data->dev_private;
 560         struct vhost_queue *vq;
 561         unsigned int i;
 562         int allow_queuing = 1;
 563
 564         if (rte_atomic32_read(&internal->dev_attached) == 0)
 565                 return;
 566
 567         if (rte_atomic32_read(&internal->started) == 0)
 568                 allow_queuing = 0;
 569
 570         /* Wait until rx/tx_pkt_burst stops accessing vhost device */
 571         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 572                 vq = dev->data->rx_queues[i];
 573                 if (vq == NULL)
 574                         continue;
 575                 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
 576                 while (rte_atomic32_read(&vq->while_queuing))
 577                         rte_pause();
 578         }
 579
 580         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 581                 vq = dev->data->tx_queues[i];
 582                 if (vq == NULL)
 583                         continue;
 584                 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
 585                 while (rte_atomic32_read(&vq->while_queuing))
 586                         rte_pause();
 587         }
 588 }
 589
 590 static void
 591 queue_setup(struct rte_eth_dev *eth_dev, struct pmd_internal *internal)
 592 {
 593         struct vhost_queue *vq;
 594         int i;
 595
 596         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 597                 vq = eth_dev->data->rx_queues[i];
 598                 if (!vq)
 599                         continue;
 600                 vq->vid = internal->vid;
 601                 vq->internal = internal;
 602                 vq->port = eth_dev->data->port_id;
 603         }
 604         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 605                 vq = eth_dev->data->tx_queues[i];
 606                 if (!vq)
 607                         continue;
 608                 vq->vid = internal->vid;
 609                 vq->internal = internal;
 610                 vq->port = eth_dev->data->port_id;
 611         }
 612 }
 613
 614 static int
 615 new_device(int vid)
 616 {
 617         struct rte_eth_dev *eth_dev;
 618         struct internal_list *list;
 619         struct pmd_internal *internal;
 620         unsigned i;
 621         char ifname[PATH_MAX];
 622 #ifdef RTE_LIBRTE_VHOST_NUMA
 623         int newnode;
 624 #endif
 625
 626         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 627         list = find_internal_resource(ifname);
 628         if (list == NULL) {
 629                 RTE_LOG(INFO, PMD, "Invalid device name: %s\n", ifname);
 630                 return -1;
 631         }
 632
 633         eth_dev = list->eth_dev;
 634         internal = eth_dev->data->dev_private;
 635
 636 #ifdef RTE_LIBRTE_VHOST_NUMA
 637         newnode = rte_vhost_get_numa_node(vid);
 638         if (newnode >= 0)
 639                 eth_dev->data->numa_node = newnode;
 640 #endif
 641
 642         internal->vid = vid;
 643         if (eth_dev->data->rx_queues && eth_dev->data->tx_queues) {
 644                 queue_setup(eth_dev, internal);
 645                 rte_atomic32_set(&internal->dev_attached, 1);
 646         } else {
 647                 RTE_LOG(INFO, PMD, "RX/TX queues have not setup yet\n");
 648                 rte_atomic32_set(&internal->dev_attached, 0);
 649         }
 650
 651         for (i = 0; i < rte_vhost_get_vring_num(vid); i++)
 652                 rte_vhost_enable_guest_notification(vid, i, 0);
 653
 654         rte_vhost_get_mtu(vid, &eth_dev->data->mtu);
 655
 656         eth_dev->data->dev_link.link_status = ETH_LINK_UP;
 657
 658         update_queuing_status(eth_dev);
 659
 660         RTE_LOG(INFO, PMD, "Vhost device %d created\n", vid);
 661
 662         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 663
 664         return 0;
 665 }
 666
 667 static void
 668 destroy_device(int vid)
 669 {
 670         struct rte_eth_dev *eth_dev;
 671         struct pmd_internal *internal;
 672         struct vhost_queue *vq;
 673         struct internal_list *list;
 674         char ifname[PATH_MAX];
 675         unsigned i;
 676         struct rte_vhost_vring_state *state;
 677
 678         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 679         list = find_internal_resource(ifname);
 680         if (list == NULL) {
 681                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
 682                 return;
 683         }
 684         eth_dev = list->eth_dev;
 685         internal = eth_dev->data->dev_private;
 686
 687         rte_atomic32_set(&internal->started, 0);
 688         update_queuing_status(eth_dev);
 689         rte_atomic32_set(&internal->dev_attached, 0);
 690
 691         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
 692
 693         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 694                 vq = eth_dev->data->rx_queues[i];
 695                 if (vq == NULL)
 696                         continue;
 697                 vq->vid = -1;
 698         }
 699         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 700                 vq = eth_dev->data->tx_queues[i];
 701                 if (vq == NULL)
 702                         continue;
 703                 vq->vid = -1;
 704         }
 705
 706         state = vring_states[eth_dev->data->port_id];
 707         rte_spinlock_lock(&state->lock);
 708         for (i = 0; i <= state->max_vring; i++) {
 709                 state->cur[i] = false;
 710                 state->seen[i] = false;
 711         }
 712         state->max_vring = 0;
 713         rte_spinlock_unlock(&state->lock);
 714
 715         RTE_LOG(INFO, PMD, "Vhost device %d destroyed\n", vid);
 716
 717         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 718 }
 719
 720 static int
 721 vring_state_changed(int vid, uint16_t vring, int enable)
 722 {
 723         struct rte_vhost_vring_state *state;
 724         struct rte_eth_dev *eth_dev;
 725         struct internal_list *list;
 726         char ifname[PATH_MAX];
 727
 728         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 729         list = find_internal_resource(ifname);
 730         if (list == NULL) {
 731                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
 732                 return -1;
 733         }
 734
 735         eth_dev = list->eth_dev;
 736         /* won't be NULL */
 737         state = vring_states[eth_dev->data->port_id];
 738         rte_spinlock_lock(&state->lock);
 739         state->cur[vring] = enable;
 740         state->max_vring = RTE_MAX(vring, state->max_vring);
 741         rte_spinlock_unlock(&state->lock);
 742
 743         RTE_LOG(INFO, PMD, "vring%u is %s\n",
 744                         vring, enable ? "enabled" : "disabled");
 745
 746         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE, NULL);
 747
 748         return 0;
 749 }
 750
 751 static struct vhost_device_ops vhost_ops = {
 752         .new_device          = new_device,
 753         .destroy_device      = destroy_device,
 754         .vring_state_changed = vring_state_changed,
 755 };
 756
 757 int
 758 rte_eth_vhost_get_queue_event(uint16_t port_id,
 759                 struct rte_eth_vhost_queue_event *event)
 760 {
 761         struct rte_vhost_vring_state *state;
 762         unsigned int i;
 763         int idx;
 764
 765         if (port_id >= RTE_MAX_ETHPORTS) {
 766                 RTE_LOG(ERR, PMD, "Invalid port id\n");
 767                 return -1;
 768         }
 769
 770         state = vring_states[port_id];
 771         if (!state) {
 772                 RTE_LOG(ERR, PMD, "Unused port\n");
 773                 return -1;
 774         }
 775
 776         rte_spinlock_lock(&state->lock);
 777         for (i = 0; i <= state->max_vring; i++) {
 778                 idx = state->index++ % (state->max_vring + 1);
 779
 780                 if (state->cur[idx] != state->seen[idx]) {
 781                         state->seen[idx] = state->cur[idx];
 782                         event->queue_id = idx / 2;
 783                         event->rx = idx & 1;
 784                         event->enable = state->cur[idx];
 785                         rte_spinlock_unlock(&state->lock);
 786                         return 0;
 787                 }
 788         }
 789         rte_spinlock_unlock(&state->lock);
 790
 791         return -1;
 792 }
 793
 794 int
 795 rte_eth_vhost_get_vid_from_port_id(uint16_t port_id)
 796 {
 797         struct internal_list *list;
 798         struct rte_eth_dev *eth_dev;
 799         struct vhost_queue *vq;
 800         int vid = -1;
 801
 802         if (!rte_eth_dev_is_valid_port(port_id))
 803                 return -1;
 804
 805         pthread_mutex_lock(&internal_list_lock);
 806
 807         TAILQ_FOREACH(list, &internal_list, next) {
 808                 eth_dev = list->eth_dev;
 809                 if (eth_dev->data->port_id == port_id) {
 810                         vq = eth_dev->data->rx_queues[0];
 811                         if (vq) {
 812                                 vid = vq->vid;
 813                         }
 814                         break;
 815                 }
 816         }
 817
 818         pthread_mutex_unlock(&internal_list_lock);
 819
 820         return vid;
 821 }
 822
 823 static int
 824 eth_dev_start(struct rte_eth_dev *eth_dev)
 825 {
 826         struct pmd_internal *internal = eth_dev->data->dev_private;
 827
 828         if (unlikely(rte_atomic32_read(&internal->dev_attached) == 0)) {
 829                 queue_setup(eth_dev, internal);
 830                 rte_atomic32_set(&internal->dev_attached, 1);
 831         }
 832
 833         rte_atomic32_set(&internal->started, 1);
 834         update_queuing_status(eth_dev);
 835
 836         return 0;
 837 }
 838
 839 static void
 840 eth_dev_stop(struct rte_eth_dev *dev)
 841 {
 842         struct pmd_internal *internal = dev->data->dev_private;
 843
 844         rte_atomic32_set(&internal->started, 0);
 845         update_queuing_status(dev);
 846 }
 847
 848 static void
 849 eth_dev_close(struct rte_eth_dev *dev)
 850 {
 851         struct pmd_internal *internal;
 852         struct internal_list *list;
 853         unsigned int i;
 854
 855         internal = dev->data->dev_private;
 856         if (!internal)
 857                 return;
 858
 859         eth_dev_stop(dev);
 860
 861         rte_vhost_driver_unregister(internal->iface_name);
 862
 863         list = find_internal_resource(internal->iface_name);
 864         if (!list)
 865                 return;
 866
 867         pthread_mutex_lock(&internal_list_lock);
 868         TAILQ_REMOVE(&internal_list, list, next);
 869         pthread_mutex_unlock(&internal_list_lock);
 870         rte_free(list);
 871
 872         for (i = 0; i < dev->data->nb_rx_queues; i++)
 873                 rte_free(dev->data->rx_queues[i]);
 874         for (i = 0; i < dev->data->nb_tx_queues; i++)
 875                 rte_free(dev->data->tx_queues[i]);
 876
 877         rte_free(dev->data->mac_addrs);
 878         free(internal->dev_name);
 879         free(internal->iface_name);
 880         rte_free(internal);
 881
 882         dev->data->dev_private = NULL;
 883 }
 884
 885 static int
 886 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 887                    uint16_t nb_rx_desc __rte_unused,
 888                    unsigned int socket_id,
 889                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 890                    struct rte_mempool *mb_pool)
 891 {
 892         struct vhost_queue *vq;
 893
 894         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
 895                         RTE_CACHE_LINE_SIZE, socket_id);
 896         if (vq == NULL) {
 897                 RTE_LOG(ERR, PMD, "Failed to allocate memory for rx queue\n");
 898                 return -ENOMEM;
 899         }
 900
 901         vq->mb_pool = mb_pool;
 902         vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
 903         dev->data->rx_queues[rx_queue_id] = vq;
 904
 905         return 0;
 906 }
 907
 908 static int
 909 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 910                    uint16_t nb_tx_desc __rte_unused,
 911                    unsigned int socket_id,
 912                    const struct rte_eth_txconf *tx_conf __rte_unused)
 913 {
 914         struct vhost_queue *vq;
 915
 916         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
 917                         RTE_CACHE_LINE_SIZE, socket_id);
 918         if (vq == NULL) {
 919                 RTE_LOG(ERR, PMD, "Failed to allocate memory for tx queue\n");
 920                 return -ENOMEM;
 921         }
 922
 923         vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
 924         dev->data->tx_queues[tx_queue_id] = vq;
 925
 926         return 0;
 927 }
 928
 929 static void
 930 eth_dev_info(struct rte_eth_dev *dev,
 931              struct rte_eth_dev_info *dev_info)
 932 {
 933         struct pmd_internal *internal;
 934
 935         internal = dev->data->dev_private;
 936         if (internal == NULL) {
 937                 RTE_LOG(ERR, PMD, "Invalid device specified\n");
 938                 return;
 939         }
 940
 941         dev_info->max_mac_addrs = 1;
 942         dev_info->max_rx_pktlen = (uint32_t)-1;
 943         dev_info->max_rx_queues = internal->max_queues;
 944         dev_info->max_tx_queues = internal->max_queues;
 945         dev_info->min_rx_bufsize = 0;
 946 }
 947
 948 static int
 949 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 950 {
 951         unsigned i;
 952         unsigned long rx_total = 0, tx_total = 0, tx_missed_total = 0;
 953         unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
 954         struct vhost_queue *vq;
 955
 956         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 957                         i < dev->data->nb_rx_queues; i++) {
 958                 if (dev->data->rx_queues[i] == NULL)
 959                         continue;
 960                 vq = dev->data->rx_queues[i];
 961                 stats->q_ipackets[i] = vq->stats.pkts;
 962                 rx_total += stats->q_ipackets[i];
 963
 964                 stats->q_ibytes[i] = vq->stats.bytes;
 965                 rx_total_bytes += stats->q_ibytes[i];
 966         }
 967
 968         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 969                         i < dev->data->nb_tx_queues; i++) {
 970                 if (dev->data->tx_queues[i] == NULL)
 971                         continue;
 972                 vq = dev->data->tx_queues[i];
 973                 stats->q_opackets[i] = vq->stats.pkts;
 974                 tx_missed_total += vq->stats.missed_pkts;
 975                 tx_total += stats->q_opackets[i];
 976
 977                 stats->q_obytes[i] = vq->stats.bytes;
 978                 tx_total_bytes += stats->q_obytes[i];
 979         }
 980
 981         stats->ipackets = rx_total;
 982         stats->opackets = tx_total;
 983         stats->oerrors = tx_missed_total;
 984         stats->ibytes = rx_total_bytes;
 985         stats->obytes = tx_total_bytes;
 986
 987         return 0;
 988 }
 989
 990 static void
 991 eth_stats_reset(struct rte_eth_dev *dev)
 992 {
 993         struct vhost_queue *vq;
 994         unsigned i;
 995
 996         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 997                 if (dev->data->rx_queues[i] == NULL)
 998                         continue;
 999                 vq = dev->data->rx_queues[i];
1000                 vq->stats.pkts = 0;
1001                 vq->stats.bytes = 0;
1002         }
1003         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1004                 if (dev->data->tx_queues[i] == NULL)
1005                         continue;
1006                 vq = dev->data->tx_queues[i];
1007                 vq->stats.pkts = 0;
1008                 vq->stats.bytes = 0;
1009                 vq->stats.missed_pkts = 0;
1010         }
1011 }
1012
1013 static void
1014 eth_queue_release(void *q)
1015 {
1016         rte_free(q);
1017 }
1018
1019 static int
1020 eth_tx_done_cleanup(void *txq __rte_unused, uint32_t free_cnt __rte_unused)
1021 {
1022         /*
1023          * vHost does not hang onto mbuf. eth_vhost_tx() copies packet data
1024          * and releases mbuf, so nothing to cleanup.
1025          */
1026         return 0;
1027 }
1028
1029 static int
1030 eth_link_update(struct rte_eth_dev *dev __rte_unused,
1031                 int wait_to_complete __rte_unused)
1032 {
1033         return 0;
1034 }
1035
1036 static uint32_t
1037 eth_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1038 {
1039         struct vhost_queue *vq;
1040
1041         vq = dev->data->rx_queues[rx_queue_id];
1042         if (vq == NULL)
1043                 return 0;
1044
1045         return rte_vhost_rx_queue_count(vq->vid, vq->virtqueue_id);
1046 }
1047
1048 static const struct eth_dev_ops ops = {
1049         .dev_start = eth_dev_start,
1050         .dev_stop = eth_dev_stop,
1051         .dev_close = eth_dev_close,
1052         .dev_configure = eth_dev_configure,
1053         .dev_infos_get = eth_dev_info,
1054         .rx_queue_setup = eth_rx_queue_setup,
1055         .tx_queue_setup = eth_tx_queue_setup,
1056         .rx_queue_release = eth_queue_release,
1057         .tx_queue_release = eth_queue_release,
1058         .tx_done_cleanup = eth_tx_done_cleanup,
1059         .rx_queue_count = eth_rx_queue_count,
1060         .link_update = eth_link_update,
1061         .stats_get = eth_stats_get,
1062         .stats_reset = eth_stats_reset,
1063         .xstats_reset = vhost_dev_xstats_reset,
1064         .xstats_get = vhost_dev_xstats_get,
1065         .xstats_get_names = vhost_dev_xstats_get_names,
1066 };
1067
1068 static struct rte_vdev_driver pmd_vhost_drv;
1069
1070 static int
1071 eth_dev_vhost_create(struct rte_vdev_device *dev, char *iface_name,
1072         int16_t queues, const unsigned int numa_node, uint64_t flags)
1073 {
1074         const char *name = rte_vdev_device_name(dev);
1075         struct rte_eth_dev_data *data = NULL;
1076         struct pmd_internal *internal = NULL;
1077         struct rte_eth_dev *eth_dev = NULL;
1078         struct ether_addr *eth_addr = NULL;
1079         struct rte_vhost_vring_state *vring_state = NULL;
1080         struct internal_list *list = NULL;
1081
1082         RTE_LOG(INFO, PMD, "Creating VHOST-USER backend on numa socket %u\n",
1083                 numa_node);
1084
1085         /* now do all data allocation - for eth_dev structure and internal
1086          * (private) data
1087          */
1088         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
1089         if (data == NULL)
1090                 goto error;
1091
1092         list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
1093         if (list == NULL)
1094                 goto error;
1095
1096         /* reserve an ethdev entry */
1097         eth_dev = rte_eth_vdev_allocate(dev, sizeof(*internal));
1098         if (eth_dev == NULL)
1099                 goto error;
1100
1101         eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
1102         if (eth_addr == NULL)
1103                 goto error;
1104         *eth_addr = base_eth_addr;
1105         eth_addr->addr_bytes[5] = eth_dev->data->port_id;
1106
1107         vring_state = rte_zmalloc_socket(name,
1108                         sizeof(*vring_state), 0, numa_node);
1109         if (vring_state == NULL)
1110                 goto error;
1111
1112         /* now put it all together
1113          * - store queue data in internal,
1114          * - point eth_dev_data to internals
1115          * - and point eth_dev structure to new eth_dev_data structure
1116          */
1117         internal = eth_dev->data->dev_private;
1118         internal->dev_name = strdup(name);
1119         if (internal->dev_name == NULL)
1120                 goto error;
1121         internal->iface_name = strdup(iface_name);
1122         if (internal->iface_name == NULL)
1123                 goto error;
1124
1125         list->eth_dev = eth_dev;
1126         pthread_mutex_lock(&internal_list_lock);
1127         TAILQ_INSERT_TAIL(&internal_list, list, next);
1128         pthread_mutex_unlock(&internal_list_lock);
1129
1130         rte_spinlock_init(&vring_state->lock);
1131         vring_states[eth_dev->data->port_id] = vring_state;
1132
1133         /* We'll replace the 'data' originally allocated by eth_dev. So the
1134          * vhost PMD resources won't be shared between multi processes.
1135          */
1136         rte_memcpy(data, eth_dev->data, sizeof(*data));
1137         eth_dev->data = data;
1138
1139         data->nb_rx_queues = queues;
1140         data->nb_tx_queues = queues;
1141         internal->max_queues = queues;
1142         data->dev_link = pmd_link;
1143         data->mac_addrs = eth_addr;
1144         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1145
1146         eth_dev->dev_ops = &ops;
1147
1148         /* finally assign rx and tx ops */
1149         eth_dev->rx_pkt_burst = eth_vhost_rx;
1150         eth_dev->tx_pkt_burst = eth_vhost_tx;
1151
1152         if (rte_vhost_driver_register(iface_name, flags))
1153                 goto error;
1154
1155         if (rte_vhost_driver_callback_register(iface_name, &vhost_ops) < 0) {
1156                 RTE_LOG(ERR, PMD, "Can't register callbacks\n");
1157                 goto error;
1158         }
1159
1160         if (rte_vhost_driver_start(iface_name) < 0) {
1161                 RTE_LOG(ERR, PMD, "Failed to start driver for %s\n",
1162                         iface_name);
1163                 goto error;
1164         }
1165
1166         return data->port_id;
1167
1168 error:
1169         if (internal) {
1170                 free(internal->iface_name);
1171                 free(internal->dev_name);
1172         }
1173         rte_free(vring_state);
1174         rte_free(eth_addr);
1175         if (eth_dev)
1176                 rte_eth_dev_release_port(eth_dev);
1177         rte_free(internal);
1178         rte_free(list);
1179         rte_free(data);
1180
1181         return -1;
1182 }
1183
1184 static inline int
1185 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
1186 {
1187         const char **iface_name = extra_args;
1188
1189         if (value == NULL)
1190                 return -1;
1191
1192         *iface_name = value;
1193
1194         return 0;
1195 }
1196
1197 static inline int
1198 open_int(const char *key __rte_unused, const char *value, void *extra_args)
1199 {
1200         uint16_t *n = extra_args;
1201
1202         if (value == NULL || extra_args == NULL)
1203                 return -EINVAL;
1204
1205         *n = (uint16_t)strtoul(value, NULL, 0);
1206         if (*n == USHRT_MAX && errno == ERANGE)
1207                 return -1;
1208
1209         return 0;
1210 }
1211
1212 static int
1213 rte_pmd_vhost_probe(struct rte_vdev_device *dev)
1214 {
1215         struct rte_kvargs *kvlist = NULL;
1216         int ret = 0;
1217         char *iface_name;
1218         uint16_t queues;
1219         uint64_t flags = 0;
1220         int client_mode = 0;
1221         int dequeue_zero_copy = 0;
1222         int iommu_support = 0;
1223
1224         RTE_LOG(INFO, PMD, "Initializing pmd_vhost for %s\n",
1225                 rte_vdev_device_name(dev));
1226
1227         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
1228         if (kvlist == NULL)
1229                 return -1;
1230
1231         if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
1232                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
1233                                          &open_iface, &iface_name);
1234                 if (ret < 0)
1235                         goto out_free;
1236         } else {
1237                 ret = -1;
1238                 goto out_free;
1239         }
1240
1241         if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
1242                 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
1243                                          &open_int, &queues);
1244                 if (ret < 0 || queues > RTE_MAX_QUEUES_PER_PORT)
1245                         goto out_free;
1246
1247         } else
1248                 queues = 1;
1249
1250         if (rte_kvargs_count(kvlist, ETH_VHOST_CLIENT_ARG) == 1) {
1251                 ret = rte_kvargs_process(kvlist, ETH_VHOST_CLIENT_ARG,
1252                                          &open_int, &client_mode);
1253                 if (ret < 0)
1254                         goto out_free;
1255
1256                 if (client_mode)
1257                         flags |= RTE_VHOST_USER_CLIENT;
1258         }
1259
1260         if (rte_kvargs_count(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY) == 1) {
1261                 ret = rte_kvargs_process(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY,
1262                                          &open_int, &dequeue_zero_copy);
1263                 if (ret < 0)
1264                         goto out_free;
1265
1266                 if (dequeue_zero_copy)
1267                         flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
1268         }
1269
1270         if (rte_kvargs_count(kvlist, ETH_VHOST_IOMMU_SUPPORT) == 1) {
1271                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IOMMU_SUPPORT,
1272                                          &open_int, &iommu_support);
1273                 if (ret < 0)
1274                         goto out_free;
1275
1276                 if (iommu_support)
1277                         flags |= RTE_VHOST_USER_IOMMU_SUPPORT;
1278         }
1279
1280         if (dev->device.numa_node == SOCKET_ID_ANY)
1281                 dev->device.numa_node = rte_socket_id();
1282
1283         eth_dev_vhost_create(dev, iface_name, queues, dev->device.numa_node,
1284                 flags);
1285
1286 out_free:
1287         rte_kvargs_free(kvlist);
1288         return ret;
1289 }
1290
1291 static int
1292 rte_pmd_vhost_remove(struct rte_vdev_device *dev)
1293 {
1294         const char *name;
1295         struct rte_eth_dev *eth_dev = NULL;
1296
1297         name = rte_vdev_device_name(dev);
1298         RTE_LOG(INFO, PMD, "Un-Initializing pmd_vhost for %s\n", name);
1299
1300         /* find an ethdev entry */
1301         eth_dev = rte_eth_dev_allocated(name);
1302         if (eth_dev == NULL)
1303                 return -ENODEV;
1304
1305         eth_dev_close(eth_dev);
1306
1307         rte_free(vring_states[eth_dev->data->port_id]);
1308         vring_states[eth_dev->data->port_id] = NULL;
1309
1310         rte_free(eth_dev->data);
1311
1312         rte_eth_dev_release_port(eth_dev);
1313
1314         return 0;
1315 }
1316
1317 static struct rte_vdev_driver pmd_vhost_drv = {
1318         .probe = rte_pmd_vhost_probe,
1319         .remove = rte_pmd_vhost_remove,
1320 };
1321
1322 RTE_PMD_REGISTER_VDEV(net_vhost, pmd_vhost_drv);
1323 RTE_PMD_REGISTER_ALIAS(net_vhost, eth_vhost);
1324 RTE_PMD_REGISTER_PARAM_STRING(net_vhost,
1325         "iface=<ifc> "
1326         "queues=<int>");