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         rte_vhost_get_mtu(vid, &eth_dev->data->mtu);
 602
 603         eth_dev->data->dev_link.link_status = ETH_LINK_UP;
 604
 605         rte_atomic32_set(&internal->dev_attached, 1);
 606         update_queuing_status(eth_dev);
 607
 608         RTE_LOG(INFO, PMD, "New connection established\n");
 609
 610         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 611
 612         return 0;
 613 }
 614
 615 static void
 616 destroy_device(int vid)
 617 {
 618         struct rte_eth_dev *eth_dev;
 619         struct pmd_internal *internal;
 620         struct vhost_queue *vq;
 621         struct internal_list *list;
 622         char ifname[PATH_MAX];
 623         unsigned i;
 624         struct rte_vhost_vring_state *state;
 625
 626         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 627         list = find_internal_resource(ifname);
 628         if (list == NULL) {
 629                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
 630                 return;
 631         }
 632         eth_dev = list->eth_dev;
 633         internal = eth_dev->data->dev_private;
 634
 635         rte_atomic32_set(&internal->dev_attached, 0);
 636         update_queuing_status(eth_dev);
 637
 638         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
 639
 640         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 641                 vq = eth_dev->data->rx_queues[i];
 642                 if (vq == NULL)
 643                         continue;
 644                 vq->vid = -1;
 645         }
 646         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 647                 vq = eth_dev->data->tx_queues[i];
 648                 if (vq == NULL)
 649                         continue;
 650                 vq->vid = -1;
 651         }
 652
 653         state = vring_states[eth_dev->data->port_id];
 654         rte_spinlock_lock(&state->lock);
 655         for (i = 0; i <= state->max_vring; i++) {
 656                 state->cur[i] = false;
 657                 state->seen[i] = false;
 658         }
 659         state->max_vring = 0;
 660         rte_spinlock_unlock(&state->lock);
 661
 662         RTE_LOG(INFO, PMD, "Connection closed\n");
 663
 664         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 665 }
 666
 667 static int
 668 vring_state_changed(int vid, uint16_t vring, int enable)
 669 {
 670         struct rte_vhost_vring_state *state;
 671         struct rte_eth_dev *eth_dev;
 672         struct internal_list *list;
 673         char ifname[PATH_MAX];
 674
 675         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 676         list = find_internal_resource(ifname);
 677         if (list == NULL) {
 678                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
 679                 return -1;
 680         }
 681
 682         eth_dev = list->eth_dev;
 683         /* won't be NULL */
 684         state = vring_states[eth_dev->data->port_id];
 685         rte_spinlock_lock(&state->lock);
 686         state->cur[vring] = enable;
 687         state->max_vring = RTE_MAX(vring, state->max_vring);
 688         rte_spinlock_unlock(&state->lock);
 689
 690         RTE_LOG(INFO, PMD, "vring%u is %s\n",
 691                         vring, enable ? "enabled" : "disabled");
 692
 693         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE, NULL);
 694
 695         return 0;
 696 }
 697
 698 int
 699 rte_eth_vhost_get_queue_event(uint8_t port_id,
 700                 struct rte_eth_vhost_queue_event *event)
 701 {
 702         struct rte_vhost_vring_state *state;
 703         unsigned int i;
 704         int idx;
 705
 706         if (port_id >= RTE_MAX_ETHPORTS) {
 707                 RTE_LOG(ERR, PMD, "Invalid port id\n");
 708                 return -1;
 709         }
 710
 711         state = vring_states[port_id];
 712         if (!state) {
 713                 RTE_LOG(ERR, PMD, "Unused port\n");
 714                 return -1;
 715         }
 716
 717         rte_spinlock_lock(&state->lock);
 718         for (i = 0; i <= state->max_vring; i++) {
 719                 idx = state->index++ % (state->max_vring + 1);
 720
 721                 if (state->cur[idx] != state->seen[idx]) {
 722                         state->seen[idx] = state->cur[idx];
 723                         event->queue_id = idx / 2;
 724                         event->rx = idx & 1;
 725                         event->enable = state->cur[idx];
 726                         rte_spinlock_unlock(&state->lock);
 727                         return 0;
 728                 }
 729         }
 730         rte_spinlock_unlock(&state->lock);
 731
 732         return -1;
 733 }
 734
 735 int
 736 rte_eth_vhost_get_vid_from_port_id(uint8_t port_id)
 737 {
 738         struct internal_list *list;
 739         struct rte_eth_dev *eth_dev;
 740         struct vhost_queue *vq;
 741         int vid = -1;
 742
 743         if (!rte_eth_dev_is_valid_port(port_id))
 744                 return -1;
 745
 746         pthread_mutex_lock(&internal_list_lock);
 747
 748         TAILQ_FOREACH(list, &internal_list, next) {
 749                 eth_dev = list->eth_dev;
 750                 if (eth_dev->data->port_id == port_id) {
 751                         vq = eth_dev->data->rx_queues[0];
 752                         if (vq) {
 753                                 vid = vq->vid;
 754                         }
 755                         break;
 756                 }
 757         }
 758
 759         pthread_mutex_unlock(&internal_list_lock);
 760
 761         return vid;
 762 }
 763
 764 static void *
 765 vhost_driver_session(void *param __rte_unused)
 766 {
 767         static struct virtio_net_device_ops vhost_ops;
 768
 769         /* set vhost arguments */
 770         vhost_ops.new_device = new_device;
 771         vhost_ops.destroy_device = destroy_device;
 772         vhost_ops.vring_state_changed = vring_state_changed;
 773         if (rte_vhost_driver_callback_register(&vhost_ops) < 0)
 774                 RTE_LOG(ERR, PMD, "Can't register callbacks\n");
 775
 776         /* start event handling */
 777         rte_vhost_driver_session_start();
 778
 779         return NULL;
 780 }
 781
 782 static int
 783 vhost_driver_session_start(void)
 784 {
 785         int ret;
 786
 787         ret = pthread_create(&session_th,
 788                         NULL, vhost_driver_session, NULL);
 789         if (ret)
 790                 RTE_LOG(ERR, PMD, "Can't create a thread\n");
 791
 792         return ret;
 793 }
 794
 795 static void
 796 vhost_driver_session_stop(void)
 797 {
 798         int ret;
 799
 800         ret = pthread_cancel(session_th);
 801         if (ret)
 802                 RTE_LOG(ERR, PMD, "Can't cancel the thread\n");
 803
 804         ret = pthread_join(session_th, NULL);
 805         if (ret)
 806                 RTE_LOG(ERR, PMD, "Can't join the thread\n");
 807 }
 808
 809 static int
 810 eth_dev_start(struct rte_eth_dev *dev)
 811 {
 812         struct pmd_internal *internal = dev->data->dev_private;
 813
 814         rte_atomic32_set(&internal->started, 1);
 815         update_queuing_status(dev);
 816
 817         return 0;
 818 }
 819
 820 static void
 821 eth_dev_stop(struct rte_eth_dev *dev)
 822 {
 823         struct pmd_internal *internal = dev->data->dev_private;
 824
 825         rte_atomic32_set(&internal->started, 0);
 826         update_queuing_status(dev);
 827 }
 828
 829 static void
 830 eth_dev_close(struct rte_eth_dev *dev)
 831 {
 832         struct pmd_internal *internal;
 833         struct internal_list *list;
 834
 835         internal = dev->data->dev_private;
 836         if (!internal)
 837                 return;
 838
 839         rte_vhost_driver_unregister(internal->iface_name);
 840
 841         list = find_internal_resource(internal->iface_name);
 842         if (!list)
 843                 return;
 844
 845         pthread_mutex_lock(&internal_list_lock);
 846         TAILQ_REMOVE(&internal_list, list, next);
 847         pthread_mutex_unlock(&internal_list_lock);
 848         rte_free(list);
 849
 850         free(internal->dev_name);
 851         free(internal->iface_name);
 852         rte_free(internal);
 853 }
 854
 855 static int
 856 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 857                    uint16_t nb_rx_desc __rte_unused,
 858                    unsigned int socket_id,
 859                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 860                    struct rte_mempool *mb_pool)
 861 {
 862         struct vhost_queue *vq;
 863
 864         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
 865                         RTE_CACHE_LINE_SIZE, socket_id);
 866         if (vq == NULL) {
 867                 RTE_LOG(ERR, PMD, "Failed to allocate memory for rx queue\n");
 868                 return -ENOMEM;
 869         }
 870
 871         vq->mb_pool = mb_pool;
 872         vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
 873         dev->data->rx_queues[rx_queue_id] = vq;
 874
 875         return 0;
 876 }
 877
 878 static int
 879 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 880                    uint16_t nb_tx_desc __rte_unused,
 881                    unsigned int socket_id,
 882                    const struct rte_eth_txconf *tx_conf __rte_unused)
 883 {
 884         struct vhost_queue *vq;
 885
 886         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
 887                         RTE_CACHE_LINE_SIZE, socket_id);
 888         if (vq == NULL) {
 889                 RTE_LOG(ERR, PMD, "Failed to allocate memory for tx queue\n");
 890                 return -ENOMEM;
 891         }
 892
 893         vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
 894         dev->data->tx_queues[tx_queue_id] = vq;
 895
 896         return 0;
 897 }
 898
 899 static void
 900 eth_dev_info(struct rte_eth_dev *dev,
 901              struct rte_eth_dev_info *dev_info)
 902 {
 903         struct pmd_internal *internal;
 904
 905         internal = dev->data->dev_private;
 906         if (internal == NULL) {
 907                 RTE_LOG(ERR, PMD, "Invalid device specified\n");
 908                 return;
 909         }
 910
 911         dev_info->max_mac_addrs = 1;
 912         dev_info->max_rx_pktlen = (uint32_t)-1;
 913         dev_info->max_rx_queues = internal->max_queues;
 914         dev_info->max_tx_queues = internal->max_queues;
 915         dev_info->min_rx_bufsize = 0;
 916 }
 917
 918 static void
 919 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 920 {
 921         unsigned i;
 922         unsigned long rx_total = 0, tx_total = 0, tx_missed_total = 0;
 923         unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
 924         struct vhost_queue *vq;
 925
 926         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 927                         i < dev->data->nb_rx_queues; i++) {
 928                 if (dev->data->rx_queues[i] == NULL)
 929                         continue;
 930                 vq = dev->data->rx_queues[i];
 931                 stats->q_ipackets[i] = vq->stats.pkts;
 932                 rx_total += stats->q_ipackets[i];
 933
 934                 stats->q_ibytes[i] = vq->stats.bytes;
 935                 rx_total_bytes += stats->q_ibytes[i];
 936         }
 937
 938         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 939                         i < dev->data->nb_tx_queues; i++) {
 940                 if (dev->data->tx_queues[i] == NULL)
 941                         continue;
 942                 vq = dev->data->tx_queues[i];
 943                 stats->q_opackets[i] = vq->stats.pkts;
 944                 tx_missed_total += vq->stats.missed_pkts;
 945                 tx_total += stats->q_opackets[i];
 946
 947                 stats->q_obytes[i] = vq->stats.bytes;
 948                 tx_total_bytes += stats->q_obytes[i];
 949         }
 950
 951         stats->ipackets = rx_total;
 952         stats->opackets = tx_total;
 953         stats->oerrors = tx_missed_total;
 954         stats->ibytes = rx_total_bytes;
 955         stats->obytes = tx_total_bytes;
 956 }
 957
 958 static void
 959 eth_stats_reset(struct rte_eth_dev *dev)
 960 {
 961         struct vhost_queue *vq;
 962         unsigned i;
 963
 964         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 965                 if (dev->data->rx_queues[i] == NULL)
 966                         continue;
 967                 vq = dev->data->rx_queues[i];
 968                 vq->stats.pkts = 0;
 969                 vq->stats.bytes = 0;
 970         }
 971         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 972                 if (dev->data->tx_queues[i] == NULL)
 973                         continue;
 974                 vq = dev->data->tx_queues[i];
 975                 vq->stats.pkts = 0;
 976                 vq->stats.bytes = 0;
 977                 vq->stats.missed_pkts = 0;
 978         }
 979 }
 980
 981 static void
 982 eth_queue_release(void *q)
 983 {
 984         rte_free(q);
 985 }
 986
 987 static int
 988 eth_tx_done_cleanup(void *txq __rte_unused, uint32_t free_cnt __rte_unused)
 989 {
 990         /*
 991          * vHost does not hang onto mbuf. eth_vhost_tx() copies packet data
 992          * and releases mbuf, so nothing to cleanup.
 993          */
 994         return 0;
 995 }
 996
 997 static int
 998 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 999                 int wait_to_complete __rte_unused)
1000 {
1001         return 0;
1002 }
1003
1004 static const struct eth_dev_ops ops = {
1005         .dev_start = eth_dev_start,
1006         .dev_stop = eth_dev_stop,
1007         .dev_close = eth_dev_close,
1008         .dev_configure = eth_dev_configure,
1009         .dev_infos_get = eth_dev_info,
1010         .rx_queue_setup = eth_rx_queue_setup,
1011         .tx_queue_setup = eth_tx_queue_setup,
1012         .rx_queue_release = eth_queue_release,
1013         .tx_queue_release = eth_queue_release,
1014         .tx_done_cleanup = eth_tx_done_cleanup,
1015         .link_update = eth_link_update,
1016         .stats_get = eth_stats_get,
1017         .stats_reset = eth_stats_reset,
1018         .xstats_reset = vhost_dev_xstats_reset,
1019         .xstats_get = vhost_dev_xstats_get,
1020         .xstats_get_names = vhost_dev_xstats_get_names,
1021 };
1022
1023 static struct rte_vdev_driver pmd_vhost_drv;
1024
1025 static int
1026 eth_dev_vhost_create(const char *name, char *iface_name, int16_t queues,
1027                      const unsigned numa_node, uint64_t flags)
1028 {
1029         struct rte_eth_dev_data *data = NULL;
1030         struct pmd_internal *internal = NULL;
1031         struct rte_eth_dev *eth_dev = NULL;
1032         struct ether_addr *eth_addr = NULL;
1033         struct rte_vhost_vring_state *vring_state = NULL;
1034         struct internal_list *list = NULL;
1035
1036         RTE_LOG(INFO, PMD, "Creating VHOST-USER backend on numa socket %u\n",
1037                 numa_node);
1038
1039         /* now do all data allocation - for eth_dev structure, dummy pci driver
1040          * and internal (private) data
1041          */
1042         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
1043         if (data == NULL)
1044                 goto error;
1045
1046         internal = rte_zmalloc_socket(name, sizeof(*internal), 0, numa_node);
1047         if (internal == NULL)
1048                 goto error;
1049
1050         list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
1051         if (list == NULL)
1052                 goto error;
1053
1054         /* reserve an ethdev entry */
1055         eth_dev = rte_eth_dev_allocate(name);
1056         if (eth_dev == NULL)
1057                 goto error;
1058
1059         eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
1060         if (eth_addr == NULL)
1061                 goto error;
1062         *eth_addr = base_eth_addr;
1063         eth_addr->addr_bytes[5] = eth_dev->data->port_id;
1064
1065         vring_state = rte_zmalloc_socket(name,
1066                         sizeof(*vring_state), 0, numa_node);
1067         if (vring_state == NULL)
1068                 goto error;
1069
1070         /* now put it all together
1071          * - store queue data in internal,
1072          * - store numa_node info in ethdev data
1073          * - point eth_dev_data to internals
1074          * - and point eth_dev structure to new eth_dev_data structure
1075          */
1076         internal->dev_name = strdup(name);
1077         if (internal->dev_name == NULL)
1078                 goto error;
1079         internal->iface_name = strdup(iface_name);
1080         if (internal->iface_name == NULL)
1081                 goto error;
1082
1083         list->eth_dev = eth_dev;
1084         pthread_mutex_lock(&internal_list_lock);
1085         TAILQ_INSERT_TAIL(&internal_list, list, next);
1086         pthread_mutex_unlock(&internal_list_lock);
1087
1088         rte_spinlock_init(&vring_state->lock);
1089         vring_states[eth_dev->data->port_id] = vring_state;
1090
1091         data->dev_private = internal;
1092         data->port_id = eth_dev->data->port_id;
1093         memmove(data->name, eth_dev->data->name, sizeof(data->name));
1094         data->nb_rx_queues = queues;
1095         data->nb_tx_queues = queues;
1096         internal->max_queues = queues;
1097         data->dev_link = pmd_link;
1098         data->mac_addrs = eth_addr;
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         eth_dev->data = data;
1104         eth_dev->dev_ops = &ops;
1105         eth_dev->driver = NULL;
1106         data->dev_flags =
1107                 RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
1108         data->kdrv = RTE_KDRV_NONE;
1109         data->drv_name = pmd_vhost_drv.driver.name;
1110         data->numa_node = numa_node;
1111
1112         /* finally assign rx and tx ops */
1113         eth_dev->rx_pkt_burst = eth_vhost_rx;
1114         eth_dev->tx_pkt_burst = eth_vhost_tx;
1115
1116         if (rte_vhost_driver_register(iface_name, flags))
1117                 goto error;
1118
1119         /* We need only one message handling thread */
1120         if (rte_atomic16_add_return(&nb_started_ports, 1) == 1) {
1121                 if (vhost_driver_session_start())
1122                         goto error;
1123         }
1124
1125         return data->port_id;
1126
1127 error:
1128         if (internal)
1129                 free(internal->dev_name);
1130         rte_free(vring_state);
1131         rte_free(eth_addr);
1132         if (eth_dev)
1133                 rte_eth_dev_release_port(eth_dev);
1134         rte_free(internal);
1135         rte_free(list);
1136         rte_free(data);
1137
1138         return -1;
1139 }
1140
1141 static inline int
1142 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
1143 {
1144         const char **iface_name = extra_args;
1145
1146         if (value == NULL)
1147                 return -1;
1148
1149         *iface_name = value;
1150
1151         return 0;
1152 }
1153
1154 static inline int
1155 open_int(const char *key __rte_unused, const char *value, void *extra_args)
1156 {
1157         uint16_t *n = extra_args;
1158
1159         if (value == NULL || extra_args == NULL)
1160                 return -EINVAL;
1161
1162         *n = (uint16_t)strtoul(value, NULL, 0);
1163         if (*n == USHRT_MAX && errno == ERANGE)
1164                 return -1;
1165
1166         return 0;
1167 }
1168
1169 static int
1170 rte_pmd_vhost_probe(const char *name, const char *params)
1171 {
1172         struct rte_kvargs *kvlist = NULL;
1173         int ret = 0;
1174         char *iface_name;
1175         uint16_t queues;
1176         uint64_t flags = 0;
1177         int client_mode = 0;
1178         int dequeue_zero_copy = 0;
1179
1180         RTE_LOG(INFO, PMD, "Initializing pmd_vhost for %s\n", name);
1181
1182         kvlist = rte_kvargs_parse(params, valid_arguments);
1183         if (kvlist == NULL)
1184                 return -1;
1185
1186         if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
1187                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
1188                                          &open_iface, &iface_name);
1189                 if (ret < 0)
1190                         goto out_free;
1191         } else {
1192                 ret = -1;
1193                 goto out_free;
1194         }
1195
1196         if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
1197                 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
1198                                          &open_int, &queues);
1199                 if (ret < 0 || queues > RTE_MAX_QUEUES_PER_PORT)
1200                         goto out_free;
1201
1202         } else
1203                 queues = 1;
1204
1205         if (rte_kvargs_count(kvlist, ETH_VHOST_CLIENT_ARG) == 1) {
1206                 ret = rte_kvargs_process(kvlist, ETH_VHOST_CLIENT_ARG,
1207                                          &open_int, &client_mode);
1208                 if (ret < 0)
1209                         goto out_free;
1210
1211                 if (client_mode)
1212                         flags |= RTE_VHOST_USER_CLIENT;
1213         }
1214
1215         if (rte_kvargs_count(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY) == 1) {
1216                 ret = rte_kvargs_process(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY,
1217                                          &open_int, &dequeue_zero_copy);
1218                 if (ret < 0)
1219                         goto out_free;
1220
1221                 if (dequeue_zero_copy)
1222                         flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
1223         }
1224
1225         eth_dev_vhost_create(name, iface_name, queues, rte_socket_id(), flags);
1226
1227 out_free:
1228         rte_kvargs_free(kvlist);
1229         return ret;
1230 }
1231
1232 static int
1233 rte_pmd_vhost_remove(const char *name)
1234 {
1235         struct rte_eth_dev *eth_dev = NULL;
1236         unsigned int i;
1237
1238         RTE_LOG(INFO, PMD, "Un-Initializing pmd_vhost for %s\n", name);
1239
1240         /* find an ethdev entry */
1241         eth_dev = rte_eth_dev_allocated(name);
1242         if (eth_dev == NULL)
1243                 return -ENODEV;
1244
1245         eth_dev_stop(eth_dev);
1246
1247         eth_dev_close(eth_dev);
1248
1249         if (rte_atomic16_sub_return(&nb_started_ports, 1) == 0)
1250                 vhost_driver_session_stop();
1251
1252         rte_free(vring_states[eth_dev->data->port_id]);
1253         vring_states[eth_dev->data->port_id] = NULL;
1254
1255         for (i = 0; i < eth_dev->data->nb_rx_queues; i++)
1256                 rte_free(eth_dev->data->rx_queues[i]);
1257         for (i = 0; i < eth_dev->data->nb_tx_queues; i++)
1258                 rte_free(eth_dev->data->tx_queues[i]);
1259
1260         rte_free(eth_dev->data->mac_addrs);
1261         rte_free(eth_dev->data);
1262
1263         rte_eth_dev_release_port(eth_dev);
1264
1265         return 0;
1266 }
1267
1268 static struct rte_vdev_driver pmd_vhost_drv = {
1269         .probe = rte_pmd_vhost_probe,
1270         .remove = rte_pmd_vhost_remove,
1271 };
1272
1273 RTE_PMD_REGISTER_VDEV(net_vhost, pmd_vhost_drv);
1274 RTE_PMD_REGISTER_ALIAS(net_vhost, eth_vhost);
1275 RTE_PMD_REGISTER_PARAM_STRING(net_vhost,
1276         "iface=<ifc> "
1277         "queues=<int>");