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.h>
  39 #include <rte_malloc.h>
  40 #include <rte_memcpy.h>
  41 #include <rte_vdev.h>
  42 #include <rte_kvargs.h>
  43 #include <rte_virtio_net.h>
  44 #include <rte_spinlock.h>
  45
  46 #include "rte_eth_vhost.h"
  47
  48 #define ETH_VHOST_IFACE_ARG             "iface"
  49 #define ETH_VHOST_QUEUES_ARG            "queues"
  50 #define ETH_VHOST_CLIENT_ARG            "client"
  51 #define ETH_VHOST_DEQUEUE_ZERO_COPY     "dequeue-zero-copy"
  52 #define VHOST_MAX_PKT_BURST 32
  53
  54 static const char *valid_arguments[] = {
  55         ETH_VHOST_IFACE_ARG,
  56         ETH_VHOST_QUEUES_ARG,
  57         ETH_VHOST_CLIENT_ARG,
  58         ETH_VHOST_DEQUEUE_ZERO_COPY,
  59         NULL
  60 };
  61
  62 static struct ether_addr base_eth_addr = {
  63         .addr_bytes = {
  64                 0x56 /* V */,
  65                 0x48 /* H */,
  66                 0x4F /* O */,
  67                 0x53 /* S */,
  68                 0x54 /* T */,
  69                 0x00
  70         }
  71 };
  72
  73 enum vhost_xstats_pkts {
  74         VHOST_UNDERSIZE_PKT = 0,
  75         VHOST_64_PKT,
  76         VHOST_65_TO_127_PKT,
  77         VHOST_128_TO_255_PKT,
  78         VHOST_256_TO_511_PKT,
  79         VHOST_512_TO_1023_PKT,
  80         VHOST_1024_TO_1522_PKT,
  81         VHOST_1523_TO_MAX_PKT,
  82         VHOST_BROADCAST_PKT,
  83         VHOST_MULTICAST_PKT,
  84         VHOST_UNICAST_PKT,
  85         VHOST_ERRORS_PKT,
  86         VHOST_ERRORS_FRAGMENTED,
  87         VHOST_ERRORS_JABBER,
  88         VHOST_UNKNOWN_PROTOCOL,
  89         VHOST_XSTATS_MAX,
  90 };
  91
  92 struct vhost_stats {
  93         uint64_t pkts;
  94         uint64_t bytes;
  95         uint64_t missed_pkts;
  96         uint64_t xstats[VHOST_XSTATS_MAX];
  97 };
  98
  99 struct vhost_queue {
 100         int vid;
 101         rte_atomic32_t allow_queuing;
 102         rte_atomic32_t while_queuing;
 103         struct pmd_internal *internal;
 104         struct rte_mempool *mb_pool;
 105         uint8_t port;
 106         uint16_t virtqueue_id;
 107         struct vhost_stats stats;
 108 };
 109
 110 struct pmd_internal {
 111         rte_atomic32_t dev_attached;
 112         char *dev_name;
 113         char *iface_name;
 114         uint16_t max_queues;
 115         rte_atomic32_t started;
 116 };
 117
 118 struct internal_list {
 119         TAILQ_ENTRY(internal_list) next;
 120         struct rte_eth_dev *eth_dev;
 121 };
 122
 123 TAILQ_HEAD(internal_list_head, internal_list);
 124 static struct internal_list_head internal_list =
 125         TAILQ_HEAD_INITIALIZER(internal_list);
 126
 127 static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
 128
 129 static rte_atomic16_t nb_started_ports;
 130 static pthread_t session_th;
 131
 132 static struct rte_eth_link pmd_link = {
 133                 .link_speed = 10000,
 134                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 135                 .link_status = ETH_LINK_DOWN
 136 };
 137
 138 struct rte_vhost_vring_state {
 139         rte_spinlock_t lock;
 140
 141         bool cur[RTE_MAX_QUEUES_PER_PORT * 2];
 142         bool seen[RTE_MAX_QUEUES_PER_PORT * 2];
 143         unsigned int index;
 144         unsigned int max_vring;
 145 };
 146
 147 static struct rte_vhost_vring_state *vring_states[RTE_MAX_ETHPORTS];
 148
 149 #define VHOST_XSTATS_NAME_SIZE 64
 150
 151 struct vhost_xstats_name_off {
 152         char name[VHOST_XSTATS_NAME_SIZE];
 153         uint64_t offset;
 154 };
 155
 156 /* [rx]_is prepended to the name string here */
 157 static const struct vhost_xstats_name_off vhost_rxport_stat_strings[] = {
 158         {"good_packets",
 159          offsetof(struct vhost_queue, stats.pkts)},
 160         {"total_bytes",
 161          offsetof(struct vhost_queue, stats.bytes)},
 162         {"missed_pkts",
 163          offsetof(struct vhost_queue, stats.missed_pkts)},
 164         {"broadcast_packets",
 165          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
 166         {"multicast_packets",
 167          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
 168         {"unicast_packets",
 169          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
 170          {"undersize_packets",
 171          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
 172         {"size_64_packets",
 173          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
 174         {"size_65_to_127_packets",
 175          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
 176         {"size_128_to_255_packets",
 177          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
 178         {"size_256_to_511_packets",
 179          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
 180         {"size_512_to_1023_packets",
 181          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
 182         {"size_1024_to_1522_packets",
 183          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
 184         {"size_1523_to_max_packets",
 185          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
 186         {"errors_with_bad_CRC",
 187          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
 188         {"fragmented_errors",
 189          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_FRAGMENTED])},
 190         {"jabber_errors",
 191          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_JABBER])},
 192         {"unknown_protos_packets",
 193          offsetof(struct vhost_queue, stats.xstats[VHOST_UNKNOWN_PROTOCOL])},
 194 };
 195
 196 /* [tx]_ is prepended to the name string here */
 197 static const struct vhost_xstats_name_off vhost_txport_stat_strings[] = {
 198         {"good_packets",
 199          offsetof(struct vhost_queue, stats.pkts)},
 200         {"total_bytes",
 201          offsetof(struct vhost_queue, stats.bytes)},
 202         {"missed_pkts",
 203          offsetof(struct vhost_queue, stats.missed_pkts)},
 204         {"broadcast_packets",
 205          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
 206         {"multicast_packets",
 207          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
 208         {"unicast_packets",
 209          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
 210         {"undersize_packets",
 211          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
 212         {"size_64_packets",
 213          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
 214         {"size_65_to_127_packets",
 215          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
 216         {"size_128_to_255_packets",
 217          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
 218         {"size_256_to_511_packets",
 219          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
 220         {"size_512_to_1023_packets",
 221          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
 222         {"size_1024_to_1522_packets",
 223          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
 224         {"size_1523_to_max_packets",
 225          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
 226         {"errors_with_bad_CRC",
 227          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
 228 };
 229
 230 #define VHOST_NB_XSTATS_RXPORT (sizeof(vhost_rxport_stat_strings) / \
 231                                 sizeof(vhost_rxport_stat_strings[0]))
 232
 233 #define VHOST_NB_XSTATS_TXPORT (sizeof(vhost_txport_stat_strings) / \
 234                                 sizeof(vhost_txport_stat_strings[0]))
 235
 236 static void
 237 vhost_dev_xstats_reset(struct rte_eth_dev *dev)
 238 {
 239         struct vhost_queue *vq = NULL;
 240         unsigned int i = 0;
 241
 242         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 243                 vq = dev->data->rx_queues[i];
 244                 if (!vq)
 245                         continue;
 246                 memset(&vq->stats, 0, sizeof(vq->stats));
 247         }
 248         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 249                 vq = dev->data->tx_queues[i];
 250                 if (!vq)
 251                         continue;
 252                 memset(&vq->stats, 0, sizeof(vq->stats));
 253         }
 254 }
 255
 256 static int
 257 vhost_dev_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
 258                            struct rte_eth_xstat_name *xstats_names,
 259                            unsigned int limit __rte_unused)
 260 {
 261         unsigned int t = 0;
 262         int count = 0;
 263         int nstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
 264
 265         if (!xstats_names)
 266                 return nstats;
 267         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
 268                 snprintf(xstats_names[count].name,
 269                          sizeof(xstats_names[count].name),
 270                          "rx_%s", vhost_rxport_stat_strings[t].name);
 271                 count++;
 272         }
 273         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
 274                 snprintf(xstats_names[count].name,
 275                          sizeof(xstats_names[count].name),
 276                          "tx_%s", vhost_txport_stat_strings[t].name);
 277                 count++;
 278         }
 279         return count;
 280 }
 281
 282 static int
 283 vhost_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 284                      unsigned int n)
 285 {
 286         unsigned int i;
 287         unsigned int t;
 288         unsigned int count = 0;
 289         struct vhost_queue *vq = NULL;
 290         unsigned int nxstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
 291
 292         if (n < nxstats)
 293                 return nxstats;
 294
 295         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 296                 vq = dev->data->rx_queues[i];
 297                 if (!vq)
 298                         continue;
 299                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
 300                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
 301                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
 302         }
 303         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 304                 vq = dev->data->tx_queues[i];
 305                 if (!vq)
 306                         continue;
 307                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
 308                                 + vq->stats.missed_pkts
 309                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
 310                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
 311         }
 312         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
 313                 xstats[count].value = 0;
 314                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 315                         vq = dev->data->rx_queues[i];
 316                         if (!vq)
 317                                 continue;
 318                         xstats[count].value +=
 319                                 *(uint64_t *)(((char *)vq)
 320                                 + vhost_rxport_stat_strings[t].offset);
 321                 }
 322                 xstats[count].id = count;
 323                 count++;
 324         }
 325         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
 326                 xstats[count].value = 0;
 327                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
 328                         vq = dev->data->tx_queues[i];
 329                         if (!vq)
 330                                 continue;
 331                         xstats[count].value +=
 332                                 *(uint64_t *)(((char *)vq)
 333                                 + vhost_txport_stat_strings[t].offset);
 334                 }
 335                 xstats[count].id = count;
 336                 count++;
 337         }
 338         return count;
 339 }
 340
 341 static inline void
 342 vhost_count_multicast_broadcast(struct vhost_queue *vq,
 343                                 struct rte_mbuf *mbuf)
 344 {
 345         struct ether_addr *ea = NULL;
 346         struct vhost_stats *pstats = &vq->stats;
 347
 348         ea = rte_pktmbuf_mtod(mbuf, struct ether_addr *);
 349         if (is_multicast_ether_addr(ea)) {
 350                 if (is_broadcast_ether_addr(ea))
 351                         pstats->xstats[VHOST_BROADCAST_PKT]++;
 352                 else
 353                         pstats->xstats[VHOST_MULTICAST_PKT]++;
 354         }
 355 }
 356
 357 static void
 358 vhost_update_packet_xstats(struct vhost_queue *vq,
 359                            struct rte_mbuf **bufs,
 360                            uint16_t count)
 361 {
 362         uint32_t pkt_len = 0;
 363         uint64_t i = 0;
 364         uint64_t index;
 365         struct vhost_stats *pstats = &vq->stats;
 366
 367         for (i = 0; i < count ; i++) {
 368                 pkt_len = bufs[i]->pkt_len;
 369                 if (pkt_len == 64) {
 370                         pstats->xstats[VHOST_64_PKT]++;
 371                 } else if (pkt_len > 64 && pkt_len < 1024) {
 372                         index = (sizeof(pkt_len) * 8)
 373                                 - __builtin_clz(pkt_len) - 5;
 374                         pstats->xstats[index]++;
 375                 } else {
 376                         if (pkt_len < 64)
 377                                 pstats->xstats[VHOST_UNDERSIZE_PKT]++;
 378                         else if (pkt_len <= 1522)
 379                                 pstats->xstats[VHOST_1024_TO_1522_PKT]++;
 380                         else if (pkt_len > 1522)
 381                                 pstats->xstats[VHOST_1523_TO_MAX_PKT]++;
 382                 }
 383                 vhost_count_multicast_broadcast(vq, bufs[i]);
 384         }
 385 }
 386
 387 static uint16_t
 388 eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 389 {
 390         struct vhost_queue *r = q;
 391         uint16_t i, nb_rx = 0;
 392         uint16_t nb_receive = nb_bufs;
 393
 394         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 395                 return 0;
 396
 397         rte_atomic32_set(&r->while_queuing, 1);
 398
 399         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 400                 goto out;
 401
 402         /* Dequeue packets from guest TX queue */
 403         while (nb_receive) {
 404                 uint16_t nb_pkts;
 405                 uint16_t num = (uint16_t)RTE_MIN(nb_receive,
 406                                                  VHOST_MAX_PKT_BURST);
 407
 408                 nb_pkts = rte_vhost_dequeue_burst(r->vid, r->virtqueue_id,
 409                                                   r->mb_pool, &bufs[nb_rx],
 410                                                   num);
 411
 412                 nb_rx += nb_pkts;
 413                 nb_receive -= nb_pkts;
 414                 if (nb_pkts < num)
 415                         break;
 416         }
 417
 418         r->stats.pkts += nb_rx;
 419
 420         for (i = 0; likely(i < nb_rx); i++) {
 421                 bufs[i]->port = r->port;
 422                 r->stats.bytes += bufs[i]->pkt_len;
 423         }
 424
 425         vhost_update_packet_xstats(r, bufs, nb_rx);
 426
 427 out:
 428         rte_atomic32_set(&r->while_queuing, 0);
 429
 430         return nb_rx;
 431 }
 432
 433 static uint16_t
 434 eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 435 {
 436         struct vhost_queue *r = q;
 437         uint16_t i, nb_tx = 0;
 438         uint16_t nb_send = nb_bufs;
 439
 440         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 441                 return 0;
 442
 443         rte_atomic32_set(&r->while_queuing, 1);
 444
 445         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 446                 goto out;
 447
 448         /* Enqueue packets to guest RX queue */
 449         while (nb_send) {
 450                 uint16_t nb_pkts;
 451                 uint16_t num = (uint16_t)RTE_MIN(nb_send,
 452                                                  VHOST_MAX_PKT_BURST);
 453
 454                 nb_pkts = rte_vhost_enqueue_burst(r->vid, r->virtqueue_id,
 455                                                   &bufs[nb_tx], num);
 456
 457                 nb_tx += nb_pkts;
 458                 nb_send -= nb_pkts;
 459                 if (nb_pkts < num)
 460                         break;
 461         }
 462
 463         r->stats.pkts += nb_tx;
 464         r->stats.missed_pkts += nb_bufs - nb_tx;
 465
 466         for (i = 0; likely(i < nb_tx); i++)
 467                 r->stats.bytes += bufs[i]->pkt_len;
 468
 469         vhost_update_packet_xstats(r, bufs, nb_tx);
 470
 471         /* According to RFC2863 page42 section ifHCOutMulticastPkts and
 472          * ifHCOutBroadcastPkts, the counters "multicast" and "broadcast"
 473          * are increased when packets are not transmitted successfully.
 474          */
 475         for (i = nb_tx; i < nb_bufs; i++)
 476                 vhost_count_multicast_broadcast(r, bufs[i]);
 477
 478         for (i = 0; likely(i < nb_tx); i++)
 479                 rte_pktmbuf_free(bufs[i]);
 480 out:
 481         rte_atomic32_set(&r->while_queuing, 0);
 482
 483         return nb_tx;
 484 }
 485
 486 static int
 487 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 488 {
 489         return 0;
 490 }
 491
 492 static inline struct internal_list *
 493 find_internal_resource(char *ifname)
 494 {
 495         int found = 0;
 496         struct internal_list *list;
 497         struct pmd_internal *internal;
 498
 499         if (ifname == NULL)
 500                 return NULL;
 501
 502         pthread_mutex_lock(&internal_list_lock);
 503
 504         TAILQ_FOREACH(list, &internal_list, next) {
 505                 internal = list->eth_dev->data->dev_private;
 506                 if (!strcmp(internal->iface_name, ifname)) {
 507                         found = 1;
 508                         break;
 509                 }
 510         }
 511
 512         pthread_mutex_unlock(&internal_list_lock);
 513
 514         if (!found)
 515                 return NULL;
 516
 517         return list;
 518 }
 519
 520 static void
 521 update_queuing_status(struct rte_eth_dev *dev)
 522 {
 523         struct pmd_internal *internal = dev->data->dev_private;
 524         struct vhost_queue *vq;
 525         unsigned int i;
 526         int allow_queuing = 1;
 527
 528         if (rte_atomic32_read(&internal->started) == 0 ||
 529             rte_atomic32_read(&internal->dev_attached) == 0)
 530                 allow_queuing = 0;
 531
 532         /* Wait until rx/tx_pkt_burst stops accessing vhost device */
 533         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 534                 vq = dev->data->rx_queues[i];
 535                 if (vq == NULL)
 536                         continue;
 537                 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
 538                 while (rte_atomic32_read(&vq->while_queuing))
 539                         rte_pause();
 540         }
 541
 542         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 543                 vq = dev->data->tx_queues[i];
 544                 if (vq == NULL)
 545                         continue;
 546                 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
 547                 while (rte_atomic32_read(&vq->while_queuing))
 548                         rte_pause();
 549         }
 550 }
 551
 552 static int
 553 new_device(int vid)
 554 {
 555         struct rte_eth_dev *eth_dev;
 556         struct internal_list *list;
 557         struct pmd_internal *internal;
 558         struct vhost_queue *vq;
 559         unsigned i;
 560         char ifname[PATH_MAX];
 561 #ifdef RTE_LIBRTE_VHOST_NUMA
 562         int newnode;
 563 #endif
 564
 565         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 566         list = find_internal_resource(ifname);
 567         if (list == NULL) {
 568                 RTE_LOG(INFO, PMD, "Invalid device name: %s\n", ifname);
 569                 return -1;
 570         }
 571
 572         eth_dev = list->eth_dev;
 573         internal = eth_dev->data->dev_private;
 574
 575 #ifdef RTE_LIBRTE_VHOST_NUMA
 576         newnode = rte_vhost_get_numa_node(vid);
 577         if (newnode >= 0)
 578                 eth_dev->data->numa_node = newnode;
 579 #endif
 580
 581         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 582                 vq = eth_dev->data->rx_queues[i];
 583                 if (vq == NULL)
 584                         continue;
 585                 vq->vid = vid;
 586                 vq->internal = internal;
 587                 vq->port = eth_dev->data->port_id;
 588         }
 589         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 590                 vq = eth_dev->data->tx_queues[i];
 591                 if (vq == NULL)
 592                         continue;
 593                 vq->vid = vid;
 594                 vq->internal = internal;
 595                 vq->port = eth_dev->data->port_id;
 596         }
 597
 598         for (i = 0; i < rte_vhost_get_queue_num(vid) * VIRTIO_QNUM; i++)
 599                 rte_vhost_enable_guest_notification(vid, i, 0);
 600
 601         eth_dev->data->dev_link.link_status = ETH_LINK_UP;
 602
 603         rte_atomic32_set(&internal->dev_attached, 1);
 604         update_queuing_status(eth_dev);
 605
 606         RTE_LOG(INFO, PMD, "New connection established\n");
 607
 608         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 609
 610         return 0;
 611 }
 612
 613 static void
 614 destroy_device(int vid)
 615 {
 616         struct rte_eth_dev *eth_dev;
 617         struct pmd_internal *internal;
 618         struct vhost_queue *vq;
 619         struct internal_list *list;
 620         char ifname[PATH_MAX];
 621         unsigned i;
 622         struct rte_vhost_vring_state *state;
 623
 624         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 625         list = find_internal_resource(ifname);
 626         if (list == NULL) {
 627                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
 628                 return;
 629         }
 630         eth_dev = list->eth_dev;
 631         internal = eth_dev->data->dev_private;
 632
 633         rte_atomic32_set(&internal->dev_attached, 0);
 634         update_queuing_status(eth_dev);
 635
 636         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
 637
 638         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 639                 vq = eth_dev->data->rx_queues[i];
 640                 if (vq == NULL)
 641                         continue;
 642                 vq->vid = -1;
 643         }
 644         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 645                 vq = eth_dev->data->tx_queues[i];
 646                 if (vq == NULL)
 647                         continue;
 648                 vq->vid = -1;
 649         }
 650
 651         state = vring_states[eth_dev->data->port_id];
 652         rte_spinlock_lock(&state->lock);
 653         for (i = 0; i <= state->max_vring; i++) {
 654                 state->cur[i] = false;
 655                 state->seen[i] = false;
 656         }
 657         state->max_vring = 0;
 658         rte_spinlock_unlock(&state->lock);
 659
 660         RTE_LOG(INFO, PMD, "Connection closed\n");
 661
 662         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 663 }
 664
 665 static int
 666 vring_state_changed(int vid, uint16_t vring, int enable)
 667 {
 668         struct rte_vhost_vring_state *state;
 669         struct rte_eth_dev *eth_dev;
 670         struct internal_list *list;
 671         char ifname[PATH_MAX];
 672
 673         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 674         list = find_internal_resource(ifname);
 675         if (list == NULL) {
 676                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
 677                 return -1;
 678         }
 679
 680         eth_dev = list->eth_dev;
 681         /* won't be NULL */
 682         state = vring_states[eth_dev->data->port_id];
 683         rte_spinlock_lock(&state->lock);
 684         state->cur[vring] = enable;
 685         state->max_vring = RTE_MAX(vring, state->max_vring);
 686         rte_spinlock_unlock(&state->lock);
 687
 688         RTE_LOG(INFO, PMD, "vring%u is %s\n",
 689                         vring, enable ? "enabled" : "disabled");
 690
 691         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE, NULL);
 692
 693         return 0;
 694 }
 695
 696 int
 697 rte_eth_vhost_get_queue_event(uint8_t port_id,
 698                 struct rte_eth_vhost_queue_event *event)
 699 {
 700         struct rte_vhost_vring_state *state;
 701         unsigned int i;
 702         int idx;
 703
 704         if (port_id >= RTE_MAX_ETHPORTS) {
 705                 RTE_LOG(ERR, PMD, "Invalid port id\n");
 706                 return -1;
 707         }
 708
 709         state = vring_states[port_id];
 710         if (!state) {
 711                 RTE_LOG(ERR, PMD, "Unused port\n");
 712                 return -1;
 713         }
 714
 715         rte_spinlock_lock(&state->lock);
 716         for (i = 0; i <= state->max_vring; i++) {
 717                 idx = state->index++ % (state->max_vring + 1);
 718
 719                 if (state->cur[idx] != state->seen[idx]) {
 720                         state->seen[idx] = state->cur[idx];
 721                         event->queue_id = idx / 2;
 722                         event->rx = idx & 1;
 723                         event->enable = state->cur[idx];
 724                         rte_spinlock_unlock(&state->lock);
 725                         return 0;
 726                 }
 727         }
 728         rte_spinlock_unlock(&state->lock);
 729
 730         return -1;
 731 }
 732
 733 int
 734 rte_eth_vhost_get_vid_from_port_id(uint8_t port_id)
 735 {
 736         struct internal_list *list;
 737         struct rte_eth_dev *eth_dev;
 738         struct vhost_queue *vq;
 739         int vid = -1;
 740
 741         if (!rte_eth_dev_is_valid_port(port_id))
 742                 return -1;
 743
 744         pthread_mutex_lock(&internal_list_lock);
 745
 746         TAILQ_FOREACH(list, &internal_list, next) {
 747                 eth_dev = list->eth_dev;
 748                 if (eth_dev->data->port_id == port_id) {
 749                         vq = eth_dev->data->rx_queues[0];
 750                         if (vq) {
 751                                 vid = vq->vid;
 752                         }
 753                         break;
 754                 }
 755         }
 756
 757         pthread_mutex_unlock(&internal_list_lock);
 758
 759         return vid;
 760 }
 761
 762 static void *
 763 vhost_driver_session(void *param __rte_unused)
 764 {
 765         static struct virtio_net_device_ops vhost_ops;
 766
 767         /* set vhost arguments */
 768         vhost_ops.new_device = new_device;
 769         vhost_ops.destroy_device = destroy_device;
 770         vhost_ops.vring_state_changed = vring_state_changed;
 771         if (rte_vhost_driver_callback_register(&vhost_ops) < 0)
 772                 RTE_LOG(ERR, PMD, "Can't register callbacks\n");
 773
 774         /* start event handling */
 775         rte_vhost_driver_session_start();
 776
 777         return NULL;
 778 }
 779
 780 static int
 781 vhost_driver_session_start(void)
 782 {
 783         int ret;
 784
 785         ret = pthread_create(&session_th,
 786                         NULL, vhost_driver_session, NULL);
 787         if (ret)
 788                 RTE_LOG(ERR, PMD, "Can't create a thread\n");
 789
 790         return ret;
 791 }
 792
 793 static void
 794 vhost_driver_session_stop(void)
 795 {
 796         int ret;
 797
 798         ret = pthread_cancel(session_th);
 799         if (ret)
 800                 RTE_LOG(ERR, PMD, "Can't cancel the thread\n");
 801
 802         ret = pthread_join(session_th, NULL);
 803         if (ret)
 804                 RTE_LOG(ERR, PMD, "Can't join the thread\n");
 805 }
 806
 807 static int
 808 eth_dev_start(struct rte_eth_dev *dev)
 809 {
 810         struct pmd_internal *internal = dev->data->dev_private;
 811
 812         rte_atomic32_set(&internal->started, 1);
 813         update_queuing_status(dev);
 814
 815         return 0;
 816 }
 817
 818 static void
 819 eth_dev_stop(struct rte_eth_dev *dev)
 820 {
 821         struct pmd_internal *internal = dev->data->dev_private;
 822
 823         rte_atomic32_set(&internal->started, 0);
 824         update_queuing_status(dev);
 825 }
 826
 827 static void
 828 eth_dev_close(struct rte_eth_dev *dev)
 829 {
 830         struct pmd_internal *internal;
 831         struct internal_list *list;
 832
 833         internal = dev->data->dev_private;
 834         if (!internal)
 835                 return;
 836
 837         rte_vhost_driver_unregister(internal->iface_name);
 838
 839         list = find_internal_resource(internal->iface_name);
 840         if (!list)
 841                 return;
 842
 843         pthread_mutex_lock(&internal_list_lock);
 844         TAILQ_REMOVE(&internal_list, list, next);
 845         pthread_mutex_unlock(&internal_list_lock);
 846         rte_free(list);
 847
 848         free(internal->dev_name);
 849         free(internal->iface_name);
 850         rte_free(internal);
 851 }
 852
 853 static int
 854 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 855                    uint16_t nb_rx_desc __rte_unused,
 856                    unsigned int socket_id,
 857                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 858                    struct rte_mempool *mb_pool)
 859 {
 860         struct vhost_queue *vq;
 861
 862         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
 863                         RTE_CACHE_LINE_SIZE, socket_id);
 864         if (vq == NULL) {
 865                 RTE_LOG(ERR, PMD, "Failed to allocate memory for rx queue\n");
 866                 return -ENOMEM;
 867         }
 868
 869         vq->mb_pool = mb_pool;
 870         vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
 871         dev->data->rx_queues[rx_queue_id] = vq;
 872
 873         return 0;
 874 }
 875
 876 static int
 877 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 878                    uint16_t nb_tx_desc __rte_unused,
 879                    unsigned int socket_id,
 880                    const struct rte_eth_txconf *tx_conf __rte_unused)
 881 {
 882         struct vhost_queue *vq;
 883
 884         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
 885                         RTE_CACHE_LINE_SIZE, socket_id);
 886         if (vq == NULL) {
 887                 RTE_LOG(ERR, PMD, "Failed to allocate memory for tx queue\n");
 888                 return -ENOMEM;
 889         }
 890
 891         vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
 892         dev->data->tx_queues[tx_queue_id] = vq;
 893
 894         return 0;
 895 }
 896
 897 static void
 898 eth_dev_info(struct rte_eth_dev *dev,
 899              struct rte_eth_dev_info *dev_info)
 900 {
 901         struct pmd_internal *internal;
 902
 903         internal = dev->data->dev_private;
 904         if (internal == NULL) {
 905                 RTE_LOG(ERR, PMD, "Invalid device specified\n");
 906                 return;
 907         }
 908
 909         dev_info->max_mac_addrs = 1;
 910         dev_info->max_rx_pktlen = (uint32_t)-1;
 911         dev_info->max_rx_queues = internal->max_queues;
 912         dev_info->max_tx_queues = internal->max_queues;
 913         dev_info->min_rx_bufsize = 0;
 914 }
 915
 916 static void
 917 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 918 {
 919         unsigned i;
 920         unsigned long rx_total = 0, tx_total = 0, tx_missed_total = 0;
 921         unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
 922         struct vhost_queue *vq;
 923
 924         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 925                         i < dev->data->nb_rx_queues; i++) {
 926                 if (dev->data->rx_queues[i] == NULL)
 927                         continue;
 928                 vq = dev->data->rx_queues[i];
 929                 stats->q_ipackets[i] = vq->stats.pkts;
 930                 rx_total += stats->q_ipackets[i];
 931
 932                 stats->q_ibytes[i] = vq->stats.bytes;
 933                 rx_total_bytes += stats->q_ibytes[i];
 934         }
 935
 936         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 937                         i < dev->data->nb_tx_queues; i++) {
 938                 if (dev->data->tx_queues[i] == NULL)
 939                         continue;
 940                 vq = dev->data->tx_queues[i];
 941                 stats->q_opackets[i] = vq->stats.pkts;
 942                 tx_missed_total += vq->stats.missed_pkts;
 943                 tx_total += stats->q_opackets[i];
 944
 945                 stats->q_obytes[i] = vq->stats.bytes;
 946                 tx_total_bytes += stats->q_obytes[i];
 947         }
 948
 949         stats->ipackets = rx_total;
 950         stats->opackets = tx_total;
 951         stats->oerrors = tx_missed_total;
 952         stats->ibytes = rx_total_bytes;
 953         stats->obytes = tx_total_bytes;
 954 }
 955
 956 static void
 957 eth_stats_reset(struct rte_eth_dev *dev)
 958 {
 959         struct vhost_queue *vq;
 960         unsigned i;
 961
 962         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 963                 if (dev->data->rx_queues[i] == NULL)
 964                         continue;
 965                 vq = dev->data->rx_queues[i];
 966                 vq->stats.pkts = 0;
 967                 vq->stats.bytes = 0;
 968         }
 969         for (i = 0; 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                 vq->stats.pkts = 0;
 974                 vq->stats.bytes = 0;
 975                 vq->stats.missed_pkts = 0;
 976         }
 977 }
 978
 979 static void
 980 eth_queue_release(void *q)
 981 {
 982         rte_free(q);
 983 }
 984
 985 static int
 986 eth_tx_done_cleanup(void *txq __rte_unused, uint32_t free_cnt __rte_unused)
 987 {
 988         /*
 989          * vHost does not hang onto mbuf. eth_vhost_tx() copies packet data
 990          * and releases mbuf, so nothing to cleanup.
 991          */
 992         return 0;
 993 }
 994
 995 static int
 996 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 997                 int wait_to_complete __rte_unused)
 998 {
 999         return 0;
1000 }
1001
1002 /**
1003  * Disable features in feature_mask. Returns 0 on success.
1004  */
1005 int
1006 rte_eth_vhost_feature_disable(uint64_t feature_mask)
1007 {
1008         return rte_vhost_feature_disable(feature_mask);
1009 }
1010
1011 /**
1012  * Enable features in feature_mask. Returns 0 on success.
1013  */
1014 int
1015 rte_eth_vhost_feature_enable(uint64_t feature_mask)
1016 {
1017         return rte_vhost_feature_enable(feature_mask);
1018 }
1019
1020 /* Returns currently supported vhost features */
1021 uint64_t
1022 rte_eth_vhost_feature_get(void)
1023 {
1024         return rte_vhost_feature_get();
1025 }
1026
1027 static const struct eth_dev_ops ops = {
1028         .dev_start = eth_dev_start,
1029         .dev_stop = eth_dev_stop,
1030         .dev_close = eth_dev_close,
1031         .dev_configure = eth_dev_configure,
1032         .dev_infos_get = eth_dev_info,
1033         .rx_queue_setup = eth_rx_queue_setup,
1034         .tx_queue_setup = eth_tx_queue_setup,
1035         .rx_queue_release = eth_queue_release,
1036         .tx_queue_release = eth_queue_release,
1037         .tx_done_cleanup = eth_tx_done_cleanup,
1038         .link_update = eth_link_update,
1039         .stats_get = eth_stats_get,
1040         .stats_reset = eth_stats_reset,
1041         .xstats_reset = vhost_dev_xstats_reset,
1042         .xstats_get = vhost_dev_xstats_get,
1043         .xstats_get_names = vhost_dev_xstats_get_names,
1044 };
1045
1046 static struct rte_vdev_driver pmd_vhost_drv;
1047
1048 static int
1049 eth_dev_vhost_create(const char *name, char *iface_name, int16_t queues,
1050                      const unsigned numa_node, uint64_t flags)
1051 {
1052         struct rte_eth_dev_data *data = NULL;
1053         struct pmd_internal *internal = NULL;
1054         struct rte_eth_dev *eth_dev = NULL;
1055         struct ether_addr *eth_addr = NULL;
1056         struct rte_vhost_vring_state *vring_state = NULL;
1057         struct internal_list *list = NULL;
1058
1059         RTE_LOG(INFO, PMD, "Creating VHOST-USER backend on numa socket %u\n",
1060                 numa_node);
1061
1062         /* now do all data allocation - for eth_dev structure, dummy pci driver
1063          * and internal (private) data
1064          */
1065         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
1066         if (data == NULL)
1067                 goto error;
1068
1069         internal = rte_zmalloc_socket(name, sizeof(*internal), 0, numa_node);
1070         if (internal == NULL)
1071                 goto error;
1072
1073         list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
1074         if (list == NULL)
1075                 goto error;
1076
1077         /* reserve an ethdev entry */
1078         eth_dev = rte_eth_dev_allocate(name);
1079         if (eth_dev == NULL)
1080                 goto error;
1081
1082         eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
1083         if (eth_addr == NULL)
1084                 goto error;
1085         *eth_addr = base_eth_addr;
1086         eth_addr->addr_bytes[5] = eth_dev->data->port_id;
1087
1088         vring_state = rte_zmalloc_socket(name,
1089                         sizeof(*vring_state), 0, numa_node);
1090         if (vring_state == NULL)
1091                 goto error;
1092
1093         /* now put it all together
1094          * - store queue data in internal,
1095          * - store numa_node info in ethdev data
1096          * - point eth_dev_data to internals
1097          * - and point eth_dev structure to new eth_dev_data structure
1098          */
1099         internal->dev_name = strdup(name);
1100         if (internal->dev_name == NULL)
1101                 goto error;
1102         internal->iface_name = strdup(iface_name);
1103         if (internal->iface_name == NULL)
1104                 goto error;
1105
1106         list->eth_dev = eth_dev;
1107         pthread_mutex_lock(&internal_list_lock);
1108         TAILQ_INSERT_TAIL(&internal_list, list, next);
1109         pthread_mutex_unlock(&internal_list_lock);
1110
1111         rte_spinlock_init(&vring_state->lock);
1112         vring_states[eth_dev->data->port_id] = vring_state;
1113
1114         data->dev_private = internal;
1115         data->port_id = eth_dev->data->port_id;
1116         memmove(data->name, eth_dev->data->name, sizeof(data->name));
1117         data->nb_rx_queues = queues;
1118         data->nb_tx_queues = queues;
1119         internal->max_queues = queues;
1120         data->dev_link = pmd_link;
1121         data->mac_addrs = eth_addr;
1122
1123         /* We'll replace the 'data' originally allocated by eth_dev. So the
1124          * vhost PMD resources won't be shared between multi processes.
1125          */
1126         eth_dev->data = data;
1127         eth_dev->dev_ops = &ops;
1128         eth_dev->driver = NULL;
1129         data->dev_flags =
1130                 RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
1131         data->kdrv = RTE_KDRV_NONE;
1132         data->drv_name = pmd_vhost_drv.driver.name;
1133         data->numa_node = numa_node;
1134
1135         /* finally assign rx and tx ops */
1136         eth_dev->rx_pkt_burst = eth_vhost_rx;
1137         eth_dev->tx_pkt_burst = eth_vhost_tx;
1138
1139         if (rte_vhost_driver_register(iface_name, flags))
1140                 goto error;
1141
1142         /* We need only one message handling thread */
1143         if (rte_atomic16_add_return(&nb_started_ports, 1) == 1) {
1144                 if (vhost_driver_session_start())
1145                         goto error;
1146         }
1147
1148         return data->port_id;
1149
1150 error:
1151         if (internal)
1152                 free(internal->dev_name);
1153         rte_free(vring_state);
1154         rte_free(eth_addr);
1155         if (eth_dev)
1156                 rte_eth_dev_release_port(eth_dev);
1157         rte_free(internal);
1158         rte_free(list);
1159         rte_free(data);
1160
1161         return -1;
1162 }
1163
1164 static inline int
1165 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
1166 {
1167         const char **iface_name = extra_args;
1168
1169         if (value == NULL)
1170                 return -1;
1171
1172         *iface_name = value;
1173
1174         return 0;
1175 }
1176
1177 static inline int
1178 open_int(const char *key __rte_unused, const char *value, void *extra_args)
1179 {
1180         uint16_t *n = extra_args;
1181
1182         if (value == NULL || extra_args == NULL)
1183                 return -EINVAL;
1184
1185         *n = (uint16_t)strtoul(value, NULL, 0);
1186         if (*n == USHRT_MAX && errno == ERANGE)
1187                 return -1;
1188
1189         return 0;
1190 }
1191
1192 static int
1193 rte_pmd_vhost_probe(const char *name, const char *params)
1194 {
1195         struct rte_kvargs *kvlist = NULL;
1196         int ret = 0;
1197         char *iface_name;
1198         uint16_t queues;
1199         uint64_t flags = 0;
1200         int client_mode = 0;
1201         int dequeue_zero_copy = 0;
1202
1203         RTE_LOG(INFO, PMD, "Initializing pmd_vhost for %s\n", name);
1204
1205         kvlist = rte_kvargs_parse(params, valid_arguments);
1206         if (kvlist == NULL)
1207                 return -1;
1208
1209         if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
1210                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
1211                                          &open_iface, &iface_name);
1212                 if (ret < 0)
1213                         goto out_free;
1214         } else {
1215                 ret = -1;
1216                 goto out_free;
1217         }
1218
1219         if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
1220                 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
1221                                          &open_int, &queues);
1222                 if (ret < 0 || queues > RTE_MAX_QUEUES_PER_PORT)
1223                         goto out_free;
1224
1225         } else
1226                 queues = 1;
1227
1228         if (rte_kvargs_count(kvlist, ETH_VHOST_CLIENT_ARG) == 1) {
1229                 ret = rte_kvargs_process(kvlist, ETH_VHOST_CLIENT_ARG,
1230                                          &open_int, &client_mode);
1231                 if (ret < 0)
1232                         goto out_free;
1233
1234                 if (client_mode)
1235                         flags |= RTE_VHOST_USER_CLIENT;
1236         }
1237
1238         if (rte_kvargs_count(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY) == 1) {
1239                 ret = rte_kvargs_process(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY,
1240                                          &open_int, &dequeue_zero_copy);
1241                 if (ret < 0)
1242                         goto out_free;
1243
1244                 if (dequeue_zero_copy)
1245                         flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
1246         }
1247
1248         eth_dev_vhost_create(name, iface_name, queues, rte_socket_id(), flags);
1249
1250 out_free:
1251         rte_kvargs_free(kvlist);
1252         return ret;
1253 }
1254
1255 static int
1256 rte_pmd_vhost_remove(const char *name)
1257 {
1258         struct rte_eth_dev *eth_dev = NULL;
1259         unsigned int i;
1260
1261         RTE_LOG(INFO, PMD, "Un-Initializing pmd_vhost for %s\n", name);
1262
1263         /* find an ethdev entry */
1264         eth_dev = rte_eth_dev_allocated(name);
1265         if (eth_dev == NULL)
1266                 return -ENODEV;
1267
1268         eth_dev_stop(eth_dev);
1269
1270         eth_dev_close(eth_dev);
1271
1272         if (rte_atomic16_sub_return(&nb_started_ports, 1) == 0)
1273                 vhost_driver_session_stop();
1274
1275         rte_free(vring_states[eth_dev->data->port_id]);
1276         vring_states[eth_dev->data->port_id] = NULL;
1277
1278         for (i = 0; i < eth_dev->data->nb_rx_queues; i++)
1279                 rte_free(eth_dev->data->rx_queues[i]);
1280         for (i = 0; i < eth_dev->data->nb_tx_queues; i++)
1281                 rte_free(eth_dev->data->tx_queues[i]);
1282
1283         rte_free(eth_dev->data->mac_addrs);
1284         rte_free(eth_dev->data);
1285
1286         rte_eth_dev_release_port(eth_dev);
1287
1288         return 0;
1289 }
1290
1291 static struct rte_vdev_driver pmd_vhost_drv = {
1292         .probe = rte_pmd_vhost_probe,
1293         .remove = rte_pmd_vhost_remove,
1294 };
1295
1296 RTE_PMD_REGISTER_VDEV(net_vhost, pmd_vhost_drv);
1297 RTE_PMD_REGISTER_ALIAS(net_vhost, eth_vhost);
1298 RTE_PMD_REGISTER_PARAM_STRING(net_vhost,
1299         "iface=<ifc> "
1300         "queues=<int>");