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 };
 123
 124 struct internal_list {
 125         TAILQ_ENTRY(internal_list) next;
 126         struct rte_eth_dev *eth_dev;
 127 };
 128
 129 TAILQ_HEAD(internal_list_head, internal_list);
 130 static struct internal_list_head internal_list =
 131         TAILQ_HEAD_INITIALIZER(internal_list);
 132
 133 static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
 134
 135 static struct rte_eth_link pmd_link = {
 136                 .link_speed = 10000,
 137                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 138                 .link_status = ETH_LINK_DOWN
 139 };
 140
 141 struct rte_vhost_vring_state {
 142         rte_spinlock_t lock;
 143
 144         bool cur[RTE_MAX_QUEUES_PER_PORT * 2];
 145         bool seen[RTE_MAX_QUEUES_PER_PORT * 2];
 146         unsigned int index;
 147         unsigned int max_vring;
 148 };
 149
 150 static struct rte_vhost_vring_state *vring_states[RTE_MAX_ETHPORTS];
 151
 152 #define VHOST_XSTATS_NAME_SIZE 64
 153
 154 struct vhost_xstats_name_off {
 155         char name[VHOST_XSTATS_NAME_SIZE];
 156         uint64_t offset;
 157 };
 158
 159 /* [rx]_is prepended to the name string here */
 160 static const struct vhost_xstats_name_off vhost_rxport_stat_strings[] = {
 161         {"good_packets",
 162          offsetof(struct vhost_queue, stats.pkts)},
 163         {"total_bytes",
 164          offsetof(struct vhost_queue, stats.bytes)},
 165         {"missed_pkts",
 166          offsetof(struct vhost_queue, stats.missed_pkts)},
 167         {"broadcast_packets",
 168          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
 169         {"multicast_packets",
 170          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
 171         {"unicast_packets",
 172          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
 173          {"undersize_packets",
 174          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
 175         {"size_64_packets",
 176          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
 177         {"size_65_to_127_packets",
 178          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
 179         {"size_128_to_255_packets",
 180          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
 181         {"size_256_to_511_packets",
 182          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
 183         {"size_512_to_1023_packets",
 184          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
 185         {"size_1024_to_1522_packets",
 186          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
 187         {"size_1523_to_max_packets",
 188          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
 189         {"errors_with_bad_CRC",
 190          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
 191         {"fragmented_errors",
 192          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_FRAGMENTED])},
 193         {"jabber_errors",
 194          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_JABBER])},
 195         {"unknown_protos_packets",
 196          offsetof(struct vhost_queue, stats.xstats[VHOST_UNKNOWN_PROTOCOL])},
 197 };
 198
 199 /* [tx]_ is prepended to the name string here */
 200 static const struct vhost_xstats_name_off vhost_txport_stat_strings[] = {
 201         {"good_packets",
 202          offsetof(struct vhost_queue, stats.pkts)},
 203         {"total_bytes",
 204          offsetof(struct vhost_queue, stats.bytes)},
 205         {"missed_pkts",
 206          offsetof(struct vhost_queue, stats.missed_pkts)},
 207         {"broadcast_packets",
 208          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
 209         {"multicast_packets",
 210          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
 211         {"unicast_packets",
 212          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
 213         {"undersize_packets",
 214          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
 215         {"size_64_packets",
 216          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
 217         {"size_65_to_127_packets",
 218          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
 219         {"size_128_to_255_packets",
 220          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
 221         {"size_256_to_511_packets",
 222          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
 223         {"size_512_to_1023_packets",
 224          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
 225         {"size_1024_to_1522_packets",
 226          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
 227         {"size_1523_to_max_packets",
 228          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
 229         {"errors_with_bad_CRC",
 230          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
 231 };
 232
 233 #define VHOST_NB_XSTATS_RXPORT (sizeof(vhost_rxport_stat_strings) / \
 234                                 sizeof(vhost_rxport_stat_strings[0]))
 235
 236 #define VHOST_NB_XSTATS_TXPORT (sizeof(vhost_txport_stat_strings) / \
 237                                 sizeof(vhost_txport_stat_strings[0]))
 238
 239 static void
 240 vhost_dev_xstats_reset(struct rte_eth_dev *dev)
 241 {
 242         struct vhost_queue *vq = NULL;
 243         unsigned int i = 0;
 244
 245         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 246                 vq = dev->data->rx_queues[i];
 247                 if (!vq)
 248                         continue;
 249                 memset(&vq->stats, 0, sizeof(vq->stats));
 250         }
 251         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 252                 vq = dev->data->tx_queues[i];
 253                 if (!vq)
 254                         continue;
 255                 memset(&vq->stats, 0, sizeof(vq->stats));
 256         }
 257 }
 258
 259 static int
 260 vhost_dev_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
 261                            struct rte_eth_xstat_name *xstats_names,
 262                            unsigned int limit __rte_unused)
 263 {
 264         unsigned int t = 0;
 265         int count = 0;
 266         int nstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
 267
 268         if (!xstats_names)
 269                 return nstats;
 270         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
 271                 snprintf(xstats_names[count].name,
 272                          sizeof(xstats_names[count].name),
 273                          "rx_%s", vhost_rxport_stat_strings[t].name);
 274                 count++;
 275         }
 276         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
 277                 snprintf(xstats_names[count].name,
 278                          sizeof(xstats_names[count].name),
 279                          "tx_%s", vhost_txport_stat_strings[t].name);
 280                 count++;
 281         }
 282         return count;
 283 }
 284
 285 static int
 286 vhost_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 287                      unsigned int n)
 288 {
 289         unsigned int i;
 290         unsigned int t;
 291         unsigned int count = 0;
 292         struct vhost_queue *vq = NULL;
 293         unsigned int nxstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
 294
 295         if (n < nxstats)
 296                 return nxstats;
 297
 298         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 299                 vq = dev->data->rx_queues[i];
 300                 if (!vq)
 301                         continue;
 302                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
 303                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
 304                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
 305         }
 306         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 307                 vq = dev->data->tx_queues[i];
 308                 if (!vq)
 309                         continue;
 310                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
 311                                 + vq->stats.missed_pkts
 312                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
 313                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
 314         }
 315         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
 316                 xstats[count].value = 0;
 317                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 318                         vq = dev->data->rx_queues[i];
 319                         if (!vq)
 320                                 continue;
 321                         xstats[count].value +=
 322                                 *(uint64_t *)(((char *)vq)
 323                                 + vhost_rxport_stat_strings[t].offset);
 324                 }
 325                 xstats[count].id = count;
 326                 count++;
 327         }
 328         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
 329                 xstats[count].value = 0;
 330                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
 331                         vq = dev->data->tx_queues[i];
 332                         if (!vq)
 333                                 continue;
 334                         xstats[count].value +=
 335                                 *(uint64_t *)(((char *)vq)
 336                                 + vhost_txport_stat_strings[t].offset);
 337                 }
 338                 xstats[count].id = count;
 339                 count++;
 340         }
 341         return count;
 342 }
 343
 344 static inline void
 345 vhost_count_multicast_broadcast(struct vhost_queue *vq,
 346                                 struct rte_mbuf *mbuf)
 347 {
 348         struct ether_addr *ea = NULL;
 349         struct vhost_stats *pstats = &vq->stats;
 350
 351         ea = rte_pktmbuf_mtod(mbuf, struct ether_addr *);
 352         if (is_multicast_ether_addr(ea)) {
 353                 if (is_broadcast_ether_addr(ea))
 354                         pstats->xstats[VHOST_BROADCAST_PKT]++;
 355                 else
 356                         pstats->xstats[VHOST_MULTICAST_PKT]++;
 357         }
 358 }
 359
 360 static void
 361 vhost_update_packet_xstats(struct vhost_queue *vq,
 362                            struct rte_mbuf **bufs,
 363                            uint16_t count)
 364 {
 365         uint32_t pkt_len = 0;
 366         uint64_t i = 0;
 367         uint64_t index;
 368         struct vhost_stats *pstats = &vq->stats;
 369
 370         for (i = 0; i < count ; i++) {
 371                 pkt_len = bufs[i]->pkt_len;
 372                 if (pkt_len == 64) {
 373                         pstats->xstats[VHOST_64_PKT]++;
 374                 } else if (pkt_len > 64 && pkt_len < 1024) {
 375                         index = (sizeof(pkt_len) * 8)
 376                                 - __builtin_clz(pkt_len) - 5;
 377                         pstats->xstats[index]++;
 378                 } else {
 379                         if (pkt_len < 64)
 380                                 pstats->xstats[VHOST_UNDERSIZE_PKT]++;
 381                         else if (pkt_len <= 1522)
 382                                 pstats->xstats[VHOST_1024_TO_1522_PKT]++;
 383                         else if (pkt_len > 1522)
 384                                 pstats->xstats[VHOST_1523_TO_MAX_PKT]++;
 385                 }
 386                 vhost_count_multicast_broadcast(vq, bufs[i]);
 387         }
 388 }
 389
 390 static uint16_t
 391 eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 392 {
 393         struct vhost_queue *r = q;
 394         uint16_t i, nb_rx = 0;
 395         uint16_t nb_receive = nb_bufs;
 396
 397         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 398                 return 0;
 399
 400         rte_atomic32_set(&r->while_queuing, 1);
 401
 402         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 403                 goto out;
 404
 405         /* Dequeue packets from guest TX queue */
 406         while (nb_receive) {
 407                 uint16_t nb_pkts;
 408                 uint16_t num = (uint16_t)RTE_MIN(nb_receive,
 409                                                  VHOST_MAX_PKT_BURST);
 410
 411                 nb_pkts = rte_vhost_dequeue_burst(r->vid, r->virtqueue_id,
 412                                                   r->mb_pool, &bufs[nb_rx],
 413                                                   num);
 414
 415                 nb_rx += nb_pkts;
 416                 nb_receive -= nb_pkts;
 417                 if (nb_pkts < num)
 418                         break;
 419         }
 420
 421         r->stats.pkts += nb_rx;
 422
 423         for (i = 0; likely(i < nb_rx); i++) {
 424                 bufs[i]->port = r->port;
 425                 r->stats.bytes += bufs[i]->pkt_len;
 426         }
 427
 428         vhost_update_packet_xstats(r, bufs, nb_rx);
 429
 430 out:
 431         rte_atomic32_set(&r->while_queuing, 0);
 432
 433         return nb_rx;
 434 }
 435
 436 static uint16_t
 437 eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 438 {
 439         struct vhost_queue *r = q;
 440         uint16_t i, nb_tx = 0;
 441         uint16_t nb_send = nb_bufs;
 442
 443         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 444                 return 0;
 445
 446         rte_atomic32_set(&r->while_queuing, 1);
 447
 448         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 449                 goto out;
 450
 451         /* Enqueue packets to guest RX queue */
 452         while (nb_send) {
 453                 uint16_t nb_pkts;
 454                 uint16_t num = (uint16_t)RTE_MIN(nb_send,
 455                                                  VHOST_MAX_PKT_BURST);
 456
 457                 nb_pkts = rte_vhost_enqueue_burst(r->vid, r->virtqueue_id,
 458                                                   &bufs[nb_tx], num);
 459
 460                 nb_tx += nb_pkts;
 461                 nb_send -= nb_pkts;
 462                 if (nb_pkts < num)
 463                         break;
 464         }
 465
 466         r->stats.pkts += nb_tx;
 467         r->stats.missed_pkts += nb_bufs - nb_tx;
 468
 469         for (i = 0; likely(i < nb_tx); i++)
 470                 r->stats.bytes += bufs[i]->pkt_len;
 471
 472         vhost_update_packet_xstats(r, bufs, nb_tx);
 473
 474         /* According to RFC2863 page42 section ifHCOutMulticastPkts and
 475          * ifHCOutBroadcastPkts, the counters "multicast" and "broadcast"
 476          * are increased when packets are not transmitted successfully.
 477          */
 478         for (i = nb_tx; i < nb_bufs; i++)
 479                 vhost_count_multicast_broadcast(r, bufs[i]);
 480
 481         for (i = 0; likely(i < nb_tx); i++)
 482                 rte_pktmbuf_free(bufs[i]);
 483 out:
 484         rte_atomic32_set(&r->while_queuing, 0);
 485
 486         return nb_tx;
 487 }
 488
 489 static int
 490 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 491 {
 492         return 0;
 493 }
 494
 495 static inline struct internal_list *
 496 find_internal_resource(char *ifname)
 497 {
 498         int found = 0;
 499         struct internal_list *list;
 500         struct pmd_internal *internal;
 501
 502         if (ifname == NULL)
 503                 return NULL;
 504
 505         pthread_mutex_lock(&internal_list_lock);
 506
 507         TAILQ_FOREACH(list, &internal_list, next) {
 508                 internal = list->eth_dev->data->dev_private;
 509                 if (!strcmp(internal->iface_name, ifname)) {
 510                         found = 1;
 511                         break;
 512                 }
 513         }
 514
 515         pthread_mutex_unlock(&internal_list_lock);
 516
 517         if (!found)
 518                 return NULL;
 519
 520         return list;
 521 }
 522
 523 static void
 524 update_queuing_status(struct rte_eth_dev *dev)
 525 {
 526         struct pmd_internal *internal = dev->data->dev_private;
 527         struct vhost_queue *vq;
 528         unsigned int i;
 529         int allow_queuing = 1;
 530
 531         if (rte_atomic32_read(&internal->dev_attached) == 0)
 532                 return;
 533
 534         if (rte_atomic32_read(&internal->started) == 0)
 535                 allow_queuing = 0;
 536
 537         /* Wait until rx/tx_pkt_burst stops accessing vhost device */
 538         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 539                 vq = dev->data->rx_queues[i];
 540                 if (vq == NULL)
 541                         continue;
 542                 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
 543                 while (rte_atomic32_read(&vq->while_queuing))
 544                         rte_pause();
 545         }
 546
 547         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 548                 vq = dev->data->tx_queues[i];
 549                 if (vq == NULL)
 550                         continue;
 551                 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
 552                 while (rte_atomic32_read(&vq->while_queuing))
 553                         rte_pause();
 554         }
 555 }
 556
 557 static void
 558 queue_setup(struct rte_eth_dev *eth_dev, struct pmd_internal *internal)
 559 {
 560         struct vhost_queue *vq;
 561         int i;
 562
 563         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 564                 vq = eth_dev->data->rx_queues[i];
 565                 if (!vq)
 566                         continue;
 567                 vq->vid = internal->vid;
 568                 vq->internal = internal;
 569                 vq->port = eth_dev->data->port_id;
 570         }
 571         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 572                 vq = eth_dev->data->tx_queues[i];
 573                 if (!vq)
 574                         continue;
 575                 vq->vid = internal->vid;
 576                 vq->internal = internal;
 577                 vq->port = eth_dev->data->port_id;
 578         }
 579 }
 580
 581 static int
 582 new_device(int vid)
 583 {
 584         struct rte_eth_dev *eth_dev;
 585         struct internal_list *list;
 586         struct pmd_internal *internal;
 587         unsigned i;
 588         char ifname[PATH_MAX];
 589 #ifdef RTE_LIBRTE_VHOST_NUMA
 590         int newnode;
 591 #endif
 592
 593         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 594         list = find_internal_resource(ifname);
 595         if (list == NULL) {
 596                 RTE_LOG(INFO, PMD, "Invalid device name: %s\n", ifname);
 597                 return -1;
 598         }
 599
 600         eth_dev = list->eth_dev;
 601         internal = eth_dev->data->dev_private;
 602
 603 #ifdef RTE_LIBRTE_VHOST_NUMA
 604         newnode = rte_vhost_get_numa_node(vid);
 605         if (newnode >= 0)
 606                 eth_dev->data->numa_node = newnode;
 607 #endif
 608
 609         internal->vid = vid;
 610         if (eth_dev->data->rx_queues && eth_dev->data->tx_queues) {
 611                 queue_setup(eth_dev, internal);
 612                 rte_atomic32_set(&internal->dev_attached, 1);
 613         } else {
 614                 RTE_LOG(INFO, PMD, "RX/TX queues have not setup yet\n");
 615                 rte_atomic32_set(&internal->dev_attached, 0);
 616         }
 617
 618         for (i = 0; i < rte_vhost_get_vring_num(vid); i++)
 619                 rte_vhost_enable_guest_notification(vid, i, 0);
 620
 621         rte_vhost_get_mtu(vid, &eth_dev->data->mtu);
 622
 623         eth_dev->data->dev_link.link_status = ETH_LINK_UP;
 624
 625         update_queuing_status(eth_dev);
 626
 627         RTE_LOG(INFO, PMD, "Vhost device %d created\n", vid);
 628
 629         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 630
 631         return 0;
 632 }
 633
 634 static void
 635 destroy_device(int vid)
 636 {
 637         struct rte_eth_dev *eth_dev;
 638         struct pmd_internal *internal;
 639         struct vhost_queue *vq;
 640         struct internal_list *list;
 641         char ifname[PATH_MAX];
 642         unsigned i;
 643         struct rte_vhost_vring_state *state;
 644
 645         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 646         list = find_internal_resource(ifname);
 647         if (list == NULL) {
 648                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
 649                 return;
 650         }
 651         eth_dev = list->eth_dev;
 652         internal = eth_dev->data->dev_private;
 653
 654         rte_atomic32_set(&internal->started, 0);
 655         update_queuing_status(eth_dev);
 656         rte_atomic32_set(&internal->dev_attached, 0);
 657
 658         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
 659
 660         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 661                 vq = eth_dev->data->rx_queues[i];
 662                 if (vq == NULL)
 663                         continue;
 664                 vq->vid = -1;
 665         }
 666         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 667                 vq = eth_dev->data->tx_queues[i];
 668                 if (vq == NULL)
 669                         continue;
 670                 vq->vid = -1;
 671         }
 672
 673         state = vring_states[eth_dev->data->port_id];
 674         rte_spinlock_lock(&state->lock);
 675         for (i = 0; i <= state->max_vring; i++) {
 676                 state->cur[i] = false;
 677                 state->seen[i] = false;
 678         }
 679         state->max_vring = 0;
 680         rte_spinlock_unlock(&state->lock);
 681
 682         RTE_LOG(INFO, PMD, "Vhost device %d destroyed\n", vid);
 683
 684         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 685 }
 686
 687 static int
 688 vring_state_changed(int vid, uint16_t vring, int enable)
 689 {
 690         struct rte_vhost_vring_state *state;
 691         struct rte_eth_dev *eth_dev;
 692         struct internal_list *list;
 693         char ifname[PATH_MAX];
 694
 695         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 696         list = find_internal_resource(ifname);
 697         if (list == NULL) {
 698                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
 699                 return -1;
 700         }
 701
 702         eth_dev = list->eth_dev;
 703         /* won't be NULL */
 704         state = vring_states[eth_dev->data->port_id];
 705         rte_spinlock_lock(&state->lock);
 706         state->cur[vring] = enable;
 707         state->max_vring = RTE_MAX(vring, state->max_vring);
 708         rte_spinlock_unlock(&state->lock);
 709
 710         RTE_LOG(INFO, PMD, "vring%u is %s\n",
 711                         vring, enable ? "enabled" : "disabled");
 712
 713         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE, NULL);
 714
 715         return 0;
 716 }
 717
 718 static struct vhost_device_ops vhost_ops = {
 719         .new_device          = new_device,
 720         .destroy_device      = destroy_device,
 721         .vring_state_changed = vring_state_changed,
 722 };
 723
 724 int
 725 rte_eth_vhost_get_queue_event(uint16_t port_id,
 726                 struct rte_eth_vhost_queue_event *event)
 727 {
 728         struct rte_vhost_vring_state *state;
 729         unsigned int i;
 730         int idx;
 731
 732         if (port_id >= RTE_MAX_ETHPORTS) {
 733                 RTE_LOG(ERR, PMD, "Invalid port id\n");
 734                 return -1;
 735         }
 736
 737         state = vring_states[port_id];
 738         if (!state) {
 739                 RTE_LOG(ERR, PMD, "Unused port\n");
 740                 return -1;
 741         }
 742
 743         rte_spinlock_lock(&state->lock);
 744         for (i = 0; i <= state->max_vring; i++) {
 745                 idx = state->index++ % (state->max_vring + 1);
 746
 747                 if (state->cur[idx] != state->seen[idx]) {
 748                         state->seen[idx] = state->cur[idx];
 749                         event->queue_id = idx / 2;
 750                         event->rx = idx & 1;
 751                         event->enable = state->cur[idx];
 752                         rte_spinlock_unlock(&state->lock);
 753                         return 0;
 754                 }
 755         }
 756         rte_spinlock_unlock(&state->lock);
 757
 758         return -1;
 759 }
 760
 761 int
 762 rte_eth_vhost_get_vid_from_port_id(uint16_t port_id)
 763 {
 764         struct internal_list *list;
 765         struct rte_eth_dev *eth_dev;
 766         struct vhost_queue *vq;
 767         int vid = -1;
 768
 769         if (!rte_eth_dev_is_valid_port(port_id))
 770                 return -1;
 771
 772         pthread_mutex_lock(&internal_list_lock);
 773
 774         TAILQ_FOREACH(list, &internal_list, next) {
 775                 eth_dev = list->eth_dev;
 776                 if (eth_dev->data->port_id == port_id) {
 777                         vq = eth_dev->data->rx_queues[0];
 778                         if (vq) {
 779                                 vid = vq->vid;
 780                         }
 781                         break;
 782                 }
 783         }
 784
 785         pthread_mutex_unlock(&internal_list_lock);
 786
 787         return vid;
 788 }
 789
 790 static int
 791 eth_dev_start(struct rte_eth_dev *eth_dev)
 792 {
 793         struct pmd_internal *internal = eth_dev->data->dev_private;
 794
 795         if (unlikely(rte_atomic32_read(&internal->dev_attached) == 0)) {
 796                 queue_setup(eth_dev, internal);
 797                 rte_atomic32_set(&internal->dev_attached, 1);
 798         }
 799
 800         rte_atomic32_set(&internal->started, 1);
 801         update_queuing_status(eth_dev);
 802
 803         return 0;
 804 }
 805
 806 static void
 807 eth_dev_stop(struct rte_eth_dev *dev)
 808 {
 809         struct pmd_internal *internal = dev->data->dev_private;
 810
 811         rte_atomic32_set(&internal->started, 0);
 812         update_queuing_status(dev);
 813 }
 814
 815 static void
 816 eth_dev_close(struct rte_eth_dev *dev)
 817 {
 818         struct pmd_internal *internal;
 819         struct internal_list *list;
 820         unsigned int i;
 821
 822         internal = dev->data->dev_private;
 823         if (!internal)
 824                 return;
 825
 826         eth_dev_stop(dev);
 827
 828         rte_vhost_driver_unregister(internal->iface_name);
 829
 830         list = find_internal_resource(internal->iface_name);
 831         if (!list)
 832                 return;
 833
 834         pthread_mutex_lock(&internal_list_lock);
 835         TAILQ_REMOVE(&internal_list, list, next);
 836         pthread_mutex_unlock(&internal_list_lock);
 837         rte_free(list);
 838
 839         for (i = 0; i < dev->data->nb_rx_queues; i++)
 840                 rte_free(dev->data->rx_queues[i]);
 841         for (i = 0; i < dev->data->nb_tx_queues; i++)
 842                 rte_free(dev->data->tx_queues[i]);
 843
 844         rte_free(dev->data->mac_addrs);
 845         free(internal->dev_name);
 846         free(internal->iface_name);
 847         rte_free(internal);
 848
 849         dev->data->dev_private = NULL;
 850 }
 851
 852 static int
 853 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 854                    uint16_t nb_rx_desc __rte_unused,
 855                    unsigned int socket_id,
 856                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 857                    struct rte_mempool *mb_pool)
 858 {
 859         struct vhost_queue *vq;
 860
 861         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
 862                         RTE_CACHE_LINE_SIZE, socket_id);
 863         if (vq == NULL) {
 864                 RTE_LOG(ERR, PMD, "Failed to allocate memory for rx queue\n");
 865                 return -ENOMEM;
 866         }
 867
 868         vq->mb_pool = mb_pool;
 869         vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
 870         dev->data->rx_queues[rx_queue_id] = vq;
 871
 872         return 0;
 873 }
 874
 875 static int
 876 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 877                    uint16_t nb_tx_desc __rte_unused,
 878                    unsigned int socket_id,
 879                    const struct rte_eth_txconf *tx_conf __rte_unused)
 880 {
 881         struct vhost_queue *vq;
 882
 883         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
 884                         RTE_CACHE_LINE_SIZE, socket_id);
 885         if (vq == NULL) {
 886                 RTE_LOG(ERR, PMD, "Failed to allocate memory for tx queue\n");
 887                 return -ENOMEM;
 888         }
 889
 890         vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
 891         dev->data->tx_queues[tx_queue_id] = vq;
 892
 893         return 0;
 894 }
 895
 896 static void
 897 eth_dev_info(struct rte_eth_dev *dev,
 898              struct rte_eth_dev_info *dev_info)
 899 {
 900         struct pmd_internal *internal;
 901
 902         internal = dev->data->dev_private;
 903         if (internal == NULL) {
 904                 RTE_LOG(ERR, PMD, "Invalid device specified\n");
 905                 return;
 906         }
 907
 908         dev_info->max_mac_addrs = 1;
 909         dev_info->max_rx_pktlen = (uint32_t)-1;
 910         dev_info->max_rx_queues = internal->max_queues;
 911         dev_info->max_tx_queues = internal->max_queues;
 912         dev_info->min_rx_bufsize = 0;
 913 }
 914
 915 static int
 916 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 917 {
 918         unsigned i;
 919         unsigned long rx_total = 0, tx_total = 0, tx_missed_total = 0;
 920         unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
 921         struct vhost_queue *vq;
 922
 923         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 924                         i < dev->data->nb_rx_queues; i++) {
 925                 if (dev->data->rx_queues[i] == NULL)
 926                         continue;
 927                 vq = dev->data->rx_queues[i];
 928                 stats->q_ipackets[i] = vq->stats.pkts;
 929                 rx_total += stats->q_ipackets[i];
 930
 931                 stats->q_ibytes[i] = vq->stats.bytes;
 932                 rx_total_bytes += stats->q_ibytes[i];
 933         }
 934
 935         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 936                         i < dev->data->nb_tx_queues; i++) {
 937                 if (dev->data->tx_queues[i] == NULL)
 938                         continue;
 939                 vq = dev->data->tx_queues[i];
 940                 stats->q_opackets[i] = vq->stats.pkts;
 941                 tx_missed_total += vq->stats.missed_pkts;
 942                 tx_total += stats->q_opackets[i];
 943
 944                 stats->q_obytes[i] = vq->stats.bytes;
 945                 tx_total_bytes += stats->q_obytes[i];
 946         }
 947
 948         stats->ipackets = rx_total;
 949         stats->opackets = tx_total;
 950         stats->oerrors = tx_missed_total;
 951         stats->ibytes = rx_total_bytes;
 952         stats->obytes = tx_total_bytes;
 953
 954         return 0;
 955 }
 956
 957 static void
 958 eth_stats_reset(struct rte_eth_dev *dev)
 959 {
 960         struct vhost_queue *vq;
 961         unsigned i;
 962
 963         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 964                 if (dev->data->rx_queues[i] == NULL)
 965                         continue;
 966                 vq = dev->data->rx_queues[i];
 967                 vq->stats.pkts = 0;
 968                 vq->stats.bytes = 0;
 969         }
 970         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 971                 if (dev->data->tx_queues[i] == NULL)
 972                         continue;
 973                 vq = dev->data->tx_queues[i];
 974                 vq->stats.pkts = 0;
 975                 vq->stats.bytes = 0;
 976                 vq->stats.missed_pkts = 0;
 977         }
 978 }
 979
 980 static void
 981 eth_queue_release(void *q)
 982 {
 983         rte_free(q);
 984 }
 985
 986 static int
 987 eth_tx_done_cleanup(void *txq __rte_unused, uint32_t free_cnt __rte_unused)
 988 {
 989         /*
 990          * vHost does not hang onto mbuf. eth_vhost_tx() copies packet data
 991          * and releases mbuf, so nothing to cleanup.
 992          */
 993         return 0;
 994 }
 995
 996 static int
 997 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 998                 int wait_to_complete __rte_unused)
 999 {
1000         return 0;
1001 }
1002
1003 static uint32_t
1004 eth_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1005 {
1006         struct vhost_queue *vq;
1007
1008         vq = dev->data->rx_queues[rx_queue_id];
1009         if (vq == NULL)
1010                 return 0;
1011
1012         return rte_vhost_rx_queue_count(vq->vid, vq->virtqueue_id);
1013 }
1014
1015 static const struct eth_dev_ops ops = {
1016         .dev_start = eth_dev_start,
1017         .dev_stop = eth_dev_stop,
1018         .dev_close = eth_dev_close,
1019         .dev_configure = eth_dev_configure,
1020         .dev_infos_get = eth_dev_info,
1021         .rx_queue_setup = eth_rx_queue_setup,
1022         .tx_queue_setup = eth_tx_queue_setup,
1023         .rx_queue_release = eth_queue_release,
1024         .tx_queue_release = eth_queue_release,
1025         .tx_done_cleanup = eth_tx_done_cleanup,
1026         .rx_queue_count = eth_rx_queue_count,
1027         .link_update = eth_link_update,
1028         .stats_get = eth_stats_get,
1029         .stats_reset = eth_stats_reset,
1030         .xstats_reset = vhost_dev_xstats_reset,
1031         .xstats_get = vhost_dev_xstats_get,
1032         .xstats_get_names = vhost_dev_xstats_get_names,
1033 };
1034
1035 static struct rte_vdev_driver pmd_vhost_drv;
1036
1037 static int
1038 eth_dev_vhost_create(struct rte_vdev_device *dev, char *iface_name,
1039         int16_t queues, const unsigned int numa_node, uint64_t flags)
1040 {
1041         const char *name = rte_vdev_device_name(dev);
1042         struct rte_eth_dev_data *data = NULL;
1043         struct pmd_internal *internal = NULL;
1044         struct rte_eth_dev *eth_dev = NULL;
1045         struct ether_addr *eth_addr = NULL;
1046         struct rte_vhost_vring_state *vring_state = NULL;
1047         struct internal_list *list = NULL;
1048
1049         RTE_LOG(INFO, PMD, "Creating VHOST-USER backend on numa socket %u\n",
1050                 numa_node);
1051
1052         /* now do all data allocation - for eth_dev structure and internal
1053          * (private) data
1054          */
1055         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
1056         if (data == NULL)
1057                 goto error;
1058
1059         list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
1060         if (list == NULL)
1061                 goto error;
1062
1063         /* reserve an ethdev entry */
1064         eth_dev = rte_eth_vdev_allocate(dev, sizeof(*internal));
1065         if (eth_dev == NULL)
1066                 goto error;
1067
1068         eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
1069         if (eth_addr == NULL)
1070                 goto error;
1071         *eth_addr = base_eth_addr;
1072         eth_addr->addr_bytes[5] = eth_dev->data->port_id;
1073
1074         vring_state = rte_zmalloc_socket(name,
1075                         sizeof(*vring_state), 0, numa_node);
1076         if (vring_state == NULL)
1077                 goto error;
1078
1079         /* now put it all together
1080          * - store queue data in internal,
1081          * - point eth_dev_data to internals
1082          * - and point eth_dev structure to new eth_dev_data structure
1083          */
1084         internal = eth_dev->data->dev_private;
1085         internal->dev_name = strdup(name);
1086         if (internal->dev_name == NULL)
1087                 goto error;
1088         internal->iface_name = strdup(iface_name);
1089         if (internal->iface_name == NULL)
1090                 goto error;
1091
1092         list->eth_dev = eth_dev;
1093         pthread_mutex_lock(&internal_list_lock);
1094         TAILQ_INSERT_TAIL(&internal_list, list, next);
1095         pthread_mutex_unlock(&internal_list_lock);
1096
1097         rte_spinlock_init(&vring_state->lock);
1098         vring_states[eth_dev->data->port_id] = vring_state;
1099
1100         /* We'll replace the 'data' originally allocated by eth_dev. So the
1101          * vhost PMD resources won't be shared between multi processes.
1102          */
1103         rte_memcpy(data, eth_dev->data, sizeof(*data));
1104         eth_dev->data = data;
1105
1106         data->nb_rx_queues = queues;
1107         data->nb_tx_queues = queues;
1108         internal->max_queues = queues;
1109         data->dev_link = pmd_link;
1110         data->mac_addrs = eth_addr;
1111         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1112
1113         eth_dev->dev_ops = &ops;
1114
1115         /* finally assign rx and tx ops */
1116         eth_dev->rx_pkt_burst = eth_vhost_rx;
1117         eth_dev->tx_pkt_burst = eth_vhost_tx;
1118
1119         if (rte_vhost_driver_register(iface_name, flags))
1120                 goto error;
1121
1122         if (rte_vhost_driver_callback_register(iface_name, &vhost_ops) < 0) {
1123                 RTE_LOG(ERR, PMD, "Can't register callbacks\n");
1124                 goto error;
1125         }
1126
1127         if (rte_vhost_driver_start(iface_name) < 0) {
1128                 RTE_LOG(ERR, PMD, "Failed to start driver for %s\n",
1129                         iface_name);
1130                 goto error;
1131         }
1132
1133         return data->port_id;
1134
1135 error:
1136         if (internal) {
1137                 free(internal->iface_name);
1138                 free(internal->dev_name);
1139         }
1140         rte_free(vring_state);
1141         rte_free(eth_addr);
1142         if (eth_dev)
1143                 rte_eth_dev_release_port(eth_dev);
1144         rte_free(internal);
1145         rte_free(list);
1146         rte_free(data);
1147
1148         return -1;
1149 }
1150
1151 static inline int
1152 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
1153 {
1154         const char **iface_name = extra_args;
1155
1156         if (value == NULL)
1157                 return -1;
1158
1159         *iface_name = value;
1160
1161         return 0;
1162 }
1163
1164 static inline int
1165 open_int(const char *key __rte_unused, const char *value, void *extra_args)
1166 {
1167         uint16_t *n = extra_args;
1168
1169         if (value == NULL || extra_args == NULL)
1170                 return -EINVAL;
1171
1172         *n = (uint16_t)strtoul(value, NULL, 0);
1173         if (*n == USHRT_MAX && errno == ERANGE)
1174                 return -1;
1175
1176         return 0;
1177 }
1178
1179 static int
1180 rte_pmd_vhost_probe(struct rte_vdev_device *dev)
1181 {
1182         struct rte_kvargs *kvlist = NULL;
1183         int ret = 0;
1184         char *iface_name;
1185         uint16_t queues;
1186         uint64_t flags = 0;
1187         int client_mode = 0;
1188         int dequeue_zero_copy = 0;
1189         int iommu_support = 0;
1190
1191         RTE_LOG(INFO, PMD, "Initializing pmd_vhost for %s\n",
1192                 rte_vdev_device_name(dev));
1193
1194         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
1195         if (kvlist == NULL)
1196                 return -1;
1197
1198         if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
1199                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
1200                                          &open_iface, &iface_name);
1201                 if (ret < 0)
1202                         goto out_free;
1203         } else {
1204                 ret = -1;
1205                 goto out_free;
1206         }
1207
1208         if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
1209                 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
1210                                          &open_int, &queues);
1211                 if (ret < 0 || queues > RTE_MAX_QUEUES_PER_PORT)
1212                         goto out_free;
1213
1214         } else
1215                 queues = 1;
1216
1217         if (rte_kvargs_count(kvlist, ETH_VHOST_CLIENT_ARG) == 1) {
1218                 ret = rte_kvargs_process(kvlist, ETH_VHOST_CLIENT_ARG,
1219                                          &open_int, &client_mode);
1220                 if (ret < 0)
1221                         goto out_free;
1222
1223                 if (client_mode)
1224                         flags |= RTE_VHOST_USER_CLIENT;
1225         }
1226
1227         if (rte_kvargs_count(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY) == 1) {
1228                 ret = rte_kvargs_process(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY,
1229                                          &open_int, &dequeue_zero_copy);
1230                 if (ret < 0)
1231                         goto out_free;
1232
1233                 if (dequeue_zero_copy)
1234                         flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
1235         }
1236
1237         if (rte_kvargs_count(kvlist, ETH_VHOST_IOMMU_SUPPORT) == 1) {
1238                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IOMMU_SUPPORT,
1239                                          &open_int, &iommu_support);
1240                 if (ret < 0)
1241                         goto out_free;
1242
1243                 if (iommu_support)
1244                         flags |= RTE_VHOST_USER_IOMMU_SUPPORT;
1245         }
1246
1247         if (dev->device.numa_node == SOCKET_ID_ANY)
1248                 dev->device.numa_node = rte_socket_id();
1249
1250         eth_dev_vhost_create(dev, iface_name, queues, dev->device.numa_node,
1251                 flags);
1252
1253 out_free:
1254         rte_kvargs_free(kvlist);
1255         return ret;
1256 }
1257
1258 static int
1259 rte_pmd_vhost_remove(struct rte_vdev_device *dev)
1260 {
1261         const char *name;
1262         struct rte_eth_dev *eth_dev = NULL;
1263
1264         name = rte_vdev_device_name(dev);
1265         RTE_LOG(INFO, PMD, "Un-Initializing pmd_vhost for %s\n", name);
1266
1267         /* find an ethdev entry */
1268         eth_dev = rte_eth_dev_allocated(name);
1269         if (eth_dev == NULL)
1270                 return -ENODEV;
1271
1272         eth_dev_close(eth_dev);
1273
1274         rte_free(vring_states[eth_dev->data->port_id]);
1275         vring_states[eth_dev->data->port_id] = NULL;
1276
1277         rte_free(eth_dev->data);
1278
1279         rte_eth_dev_release_port(eth_dev);
1280
1281         return 0;
1282 }
1283
1284 static struct rte_vdev_driver pmd_vhost_drv = {
1285         .probe = rte_pmd_vhost_probe,
1286         .remove = rte_pmd_vhost_remove,
1287 };
1288
1289 RTE_PMD_REGISTER_VDEV(net_vhost, pmd_vhost_drv);
1290 RTE_PMD_REGISTER_ALIAS(net_vhost, eth_vhost);
1291 RTE_PMD_REGISTER_PARAM_STRING(net_vhost,
1292         "iface=<ifc> "
1293         "queues=<int>");