drivers/net/vhost/rte_eth_vhost.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright (c) 2016 IGEL Co., Ltd.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of IGEL Co.,Ltd. nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33 #include <unistd.h>
  34 #include <pthread.h>
  35 #include <stdbool.h>
  36
  37 #include <rte_mbuf.h>
  38 #include <rte_ethdev_driver.h>
  39 #include <rte_ethdev_vdev.h>
  40 #include <rte_malloc.h>
  41 #include <rte_memcpy.h>
  42 #include <rte_bus_vdev.h>
  43 #include <rte_kvargs.h>
  44 #include <rte_vhost.h>
  45 #include <rte_spinlock.h>
  46
  47 #include "rte_eth_vhost.h"
  48
  49 enum {VIRTIO_RXQ, VIRTIO_TXQ, VIRTIO_QNUM};
  50
  51 #define ETH_VHOST_IFACE_ARG             "iface"
  52 #define ETH_VHOST_QUEUES_ARG            "queues"
  53 #define ETH_VHOST_CLIENT_ARG            "client"
  54 #define ETH_VHOST_DEQUEUE_ZERO_COPY     "dequeue-zero-copy"
  55 #define ETH_VHOST_IOMMU_SUPPORT         "iommu-support"
  56 #define VHOST_MAX_PKT_BURST 32
  57
  58 static const char *valid_arguments[] = {
  59         ETH_VHOST_IFACE_ARG,
  60         ETH_VHOST_QUEUES_ARG,
  61         ETH_VHOST_CLIENT_ARG,
  62         ETH_VHOST_DEQUEUE_ZERO_COPY,
  63         ETH_VHOST_IOMMU_SUPPORT,
  64         NULL
  65 };
  66
  67 static struct ether_addr base_eth_addr = {
  68         .addr_bytes = {
  69                 0x56 /* V */,
  70                 0x48 /* H */,
  71                 0x4F /* O */,
  72                 0x53 /* S */,
  73                 0x54 /* T */,
  74                 0x00
  75         }
  76 };
  77
  78 enum vhost_xstats_pkts {
  79         VHOST_UNDERSIZE_PKT = 0,
  80         VHOST_64_PKT,
  81         VHOST_65_TO_127_PKT,
  82         VHOST_128_TO_255_PKT,
  83         VHOST_256_TO_511_PKT,
  84         VHOST_512_TO_1023_PKT,
  85         VHOST_1024_TO_1522_PKT,
  86         VHOST_1523_TO_MAX_PKT,
  87         VHOST_BROADCAST_PKT,
  88         VHOST_MULTICAST_PKT,
  89         VHOST_UNICAST_PKT,
  90         VHOST_ERRORS_PKT,
  91         VHOST_ERRORS_FRAGMENTED,
  92         VHOST_ERRORS_JABBER,
  93         VHOST_UNKNOWN_PROTOCOL,
  94         VHOST_XSTATS_MAX,
  95 };
  96
  97 struct vhost_stats {
  98         uint64_t pkts;
  99         uint64_t bytes;
 100         uint64_t missed_pkts;
 101         uint64_t xstats[VHOST_XSTATS_MAX];
 102 };
 103
 104 struct vhost_queue {
 105         int vid;
 106         rte_atomic32_t allow_queuing;
 107         rte_atomic32_t while_queuing;
 108         struct pmd_internal *internal;
 109         struct rte_mempool *mb_pool;
 110         uint16_t port;
 111         uint16_t virtqueue_id;
 112         struct vhost_stats stats;
 113 };
 114
 115 struct pmd_internal {
 116         rte_atomic32_t dev_attached;
 117         char *dev_name;
 118         char *iface_name;
 119         uint16_t max_queues;
 120         uint16_t vid;
 121         rte_atomic32_t started;
 122         uint8_t vlan_strip;
 123 };
 124
 125 struct internal_list {
 126         TAILQ_ENTRY(internal_list) next;
 127         struct rte_eth_dev *eth_dev;
 128 };
 129
 130 TAILQ_HEAD(internal_list_head, internal_list);
 131 static struct internal_list_head internal_list =
 132         TAILQ_HEAD_INITIALIZER(internal_list);
 133
 134 static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
 135
 136 static struct rte_eth_link pmd_link = {
 137                 .link_speed = 10000,
 138                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 139                 .link_status = ETH_LINK_DOWN
 140 };
 141
 142 struct rte_vhost_vring_state {
 143         rte_spinlock_t lock;
 144
 145         bool cur[RTE_MAX_QUEUES_PER_PORT * 2];
 146         bool seen[RTE_MAX_QUEUES_PER_PORT * 2];
 147         unsigned int index;
 148         unsigned int max_vring;
 149 };
 150
 151 static struct rte_vhost_vring_state *vring_states[RTE_MAX_ETHPORTS];
 152
 153 #define VHOST_XSTATS_NAME_SIZE 64
 154
 155 struct vhost_xstats_name_off {
 156         char name[VHOST_XSTATS_NAME_SIZE];
 157         uint64_t offset;
 158 };
 159
 160 /* [rx]_is prepended to the name string here */
 161 static const struct vhost_xstats_name_off vhost_rxport_stat_strings[] = {
 162         {"good_packets",
 163          offsetof(struct vhost_queue, stats.pkts)},
 164         {"total_bytes",
 165          offsetof(struct vhost_queue, stats.bytes)},
 166         {"missed_pkts",
 167          offsetof(struct vhost_queue, stats.missed_pkts)},
 168         {"broadcast_packets",
 169          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
 170         {"multicast_packets",
 171          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
 172         {"unicast_packets",
 173          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
 174          {"undersize_packets",
 175          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
 176         {"size_64_packets",
 177          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
 178         {"size_65_to_127_packets",
 179          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
 180         {"size_128_to_255_packets",
 181          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
 182         {"size_256_to_511_packets",
 183          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
 184         {"size_512_to_1023_packets",
 185          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
 186         {"size_1024_to_1522_packets",
 187          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
 188         {"size_1523_to_max_packets",
 189          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
 190         {"errors_with_bad_CRC",
 191          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
 192         {"fragmented_errors",
 193          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_FRAGMENTED])},
 194         {"jabber_errors",
 195          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_JABBER])},
 196         {"unknown_protos_packets",
 197          offsetof(struct vhost_queue, stats.xstats[VHOST_UNKNOWN_PROTOCOL])},
 198 };
 199
 200 /* [tx]_ is prepended to the name string here */
 201 static const struct vhost_xstats_name_off vhost_txport_stat_strings[] = {
 202         {"good_packets",
 203          offsetof(struct vhost_queue, stats.pkts)},
 204         {"total_bytes",
 205          offsetof(struct vhost_queue, stats.bytes)},
 206         {"missed_pkts",
 207          offsetof(struct vhost_queue, stats.missed_pkts)},
 208         {"broadcast_packets",
 209          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
 210         {"multicast_packets",
 211          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
 212         {"unicast_packets",
 213          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
 214         {"undersize_packets",
 215          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
 216         {"size_64_packets",
 217          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
 218         {"size_65_to_127_packets",
 219          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
 220         {"size_128_to_255_packets",
 221          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
 222         {"size_256_to_511_packets",
 223          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
 224         {"size_512_to_1023_packets",
 225          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
 226         {"size_1024_to_1522_packets",
 227          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
 228         {"size_1523_to_max_packets",
 229          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
 230         {"errors_with_bad_CRC",
 231          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
 232 };
 233
 234 #define VHOST_NB_XSTATS_RXPORT (sizeof(vhost_rxport_stat_strings) / \
 235                                 sizeof(vhost_rxport_stat_strings[0]))
 236
 237 #define VHOST_NB_XSTATS_TXPORT (sizeof(vhost_txport_stat_strings) / \
 238                                 sizeof(vhost_txport_stat_strings[0]))
 239
 240 static void
 241 vhost_dev_xstats_reset(struct rte_eth_dev *dev)
 242 {
 243         struct vhost_queue *vq = NULL;
 244         unsigned int i = 0;
 245
 246         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 247                 vq = dev->data->rx_queues[i];
 248                 if (!vq)
 249                         continue;
 250                 memset(&vq->stats, 0, sizeof(vq->stats));
 251         }
 252         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 253                 vq = dev->data->tx_queues[i];
 254                 if (!vq)
 255                         continue;
 256                 memset(&vq->stats, 0, sizeof(vq->stats));
 257         }
 258 }
 259
 260 static int
 261 vhost_dev_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
 262                            struct rte_eth_xstat_name *xstats_names,
 263                            unsigned int limit __rte_unused)
 264 {
 265         unsigned int t = 0;
 266         int count = 0;
 267         int nstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
 268
 269         if (!xstats_names)
 270                 return nstats;
 271         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
 272                 snprintf(xstats_names[count].name,
 273                          sizeof(xstats_names[count].name),
 274                          "rx_%s", vhost_rxport_stat_strings[t].name);
 275                 count++;
 276         }
 277         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
 278                 snprintf(xstats_names[count].name,
 279                          sizeof(xstats_names[count].name),
 280                          "tx_%s", vhost_txport_stat_strings[t].name);
 281                 count++;
 282         }
 283         return count;
 284 }
 285
 286 static int
 287 vhost_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 288                      unsigned int n)
 289 {
 290         unsigned int i;
 291         unsigned int t;
 292         unsigned int count = 0;
 293         struct vhost_queue *vq = NULL;
 294         unsigned int nxstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
 295
 296         if (n < nxstats)
 297                 return nxstats;
 298
 299         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 300                 vq = dev->data->rx_queues[i];
 301                 if (!vq)
 302                         continue;
 303                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
 304                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
 305                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
 306         }
 307         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 308                 vq = dev->data->tx_queues[i];
 309                 if (!vq)
 310                         continue;
 311                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
 312                                 + vq->stats.missed_pkts
 313                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
 314                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
 315         }
 316         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
 317                 xstats[count].value = 0;
 318                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 319                         vq = dev->data->rx_queues[i];
 320                         if (!vq)
 321                                 continue;
 322                         xstats[count].value +=
 323                                 *(uint64_t *)(((char *)vq)
 324                                 + vhost_rxport_stat_strings[t].offset);
 325                 }
 326                 xstats[count].id = count;
 327                 count++;
 328         }
 329         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
 330                 xstats[count].value = 0;
 331                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
 332                         vq = dev->data->tx_queues[i];
 333                         if (!vq)
 334                                 continue;
 335                         xstats[count].value +=
 336                                 *(uint64_t *)(((char *)vq)
 337                                 + vhost_txport_stat_strings[t].offset);
 338                 }
 339                 xstats[count].id = count;
 340                 count++;
 341         }
 342         return count;
 343 }
 344
 345 static inline void
 346 vhost_count_multicast_broadcast(struct vhost_queue *vq,
 347                                 struct rte_mbuf *mbuf)
 348 {
 349         struct ether_addr *ea = NULL;
 350         struct vhost_stats *pstats = &vq->stats;
 351
 352         ea = rte_pktmbuf_mtod(mbuf, struct ether_addr *);
 353         if (is_multicast_ether_addr(ea)) {
 354                 if (is_broadcast_ether_addr(ea))
 355                         pstats->xstats[VHOST_BROADCAST_PKT]++;
 356                 else
 357                         pstats->xstats[VHOST_MULTICAST_PKT]++;
 358         }
 359 }
 360
 361 static void
 362 vhost_update_packet_xstats(struct vhost_queue *vq,
 363                            struct rte_mbuf **bufs,
 364                            uint16_t count)
 365 {
 366         uint32_t pkt_len = 0;
 367         uint64_t i = 0;
 368         uint64_t index;
 369         struct vhost_stats *pstats = &vq->stats;
 370
 371         for (i = 0; i < count ; i++) {
 372                 pkt_len = bufs[i]->pkt_len;
 373                 if (pkt_len == 64) {
 374                         pstats->xstats[VHOST_64_PKT]++;
 375                 } else if (pkt_len > 64 && pkt_len < 1024) {
 376                         index = (sizeof(pkt_len) * 8)
 377                                 - __builtin_clz(pkt_len) - 5;
 378                         pstats->xstats[index]++;
 379                 } else {
 380                         if (pkt_len < 64)
 381                                 pstats->xstats[VHOST_UNDERSIZE_PKT]++;
 382                         else if (pkt_len <= 1522)
 383                                 pstats->xstats[VHOST_1024_TO_1522_PKT]++;
 384                         else if (pkt_len > 1522)
 385                                 pstats->xstats[VHOST_1523_TO_MAX_PKT]++;
 386                 }
 387                 vhost_count_multicast_broadcast(vq, bufs[i]);
 388         }
 389 }
 390
 391 static uint16_t
 392 eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 393 {
 394         struct vhost_queue *r = q;
 395         uint16_t i, nb_rx = 0;
 396         uint16_t nb_receive = nb_bufs;
 397
 398         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 399                 return 0;
 400
 401         rte_atomic32_set(&r->while_queuing, 1);
 402
 403         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 404                 goto out;
 405
 406         /* Dequeue packets from guest TX queue */
 407         while (nb_receive) {
 408                 uint16_t nb_pkts;
 409                 uint16_t num = (uint16_t)RTE_MIN(nb_receive,
 410                                                  VHOST_MAX_PKT_BURST);
 411
 412                 nb_pkts = rte_vhost_dequeue_burst(r->vid, r->virtqueue_id,
 413                                                   r->mb_pool, &bufs[nb_rx],
 414                                                   num);
 415
 416                 nb_rx += nb_pkts;
 417                 nb_receive -= nb_pkts;
 418                 if (nb_pkts < num)
 419                         break;
 420         }
 421
 422         r->stats.pkts += nb_rx;
 423
 424         for (i = 0; likely(i < nb_rx); i++) {
 425                 bufs[i]->port = r->port;
 426                 bufs[i]->ol_flags = 0;
 427                 bufs[i]->vlan_tci = 0;
 428
 429                 if (r->internal->vlan_strip)
 430                         rte_vlan_strip(bufs[i]);
 431
 432                 r->stats.bytes += bufs[i]->pkt_len;
 433         }
 434
 435         vhost_update_packet_xstats(r, bufs, nb_rx);
 436
 437 out:
 438         rte_atomic32_set(&r->while_queuing, 0);
 439
 440         return nb_rx;
 441 }
 442
 443 static uint16_t
 444 eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 445 {
 446         struct vhost_queue *r = q;
 447         uint16_t i, nb_tx = 0;
 448         uint16_t nb_send = 0;
 449
 450         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 451                 return 0;
 452
 453         rte_atomic32_set(&r->while_queuing, 1);
 454
 455         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 456                 goto out;
 457
 458         for (i = 0; i < nb_bufs; i++) {
 459                 struct rte_mbuf *m = bufs[i];
 460
 461                 /* Do VLAN tag insertion */
 462                 if (m->ol_flags & PKT_TX_VLAN_PKT) {
 463                         int error = rte_vlan_insert(&m);
 464                         if (unlikely(error)) {
 465                                 rte_pktmbuf_free(m);
 466                                 continue;
 467                         }
 468                 }
 469
 470                 bufs[nb_send] = m;
 471                 ++nb_send;
 472         }
 473
 474         /* Enqueue packets to guest RX queue */
 475         while (nb_send) {
 476                 uint16_t nb_pkts;
 477                 uint16_t num = (uint16_t)RTE_MIN(nb_send,
 478                                                  VHOST_MAX_PKT_BURST);
 479
 480                 nb_pkts = rte_vhost_enqueue_burst(r->vid, r->virtqueue_id,
 481                                                   &bufs[nb_tx], num);
 482
 483                 nb_tx += nb_pkts;
 484                 nb_send -= nb_pkts;
 485                 if (nb_pkts < num)
 486                         break;
 487         }
 488
 489         r->stats.pkts += nb_tx;
 490         r->stats.missed_pkts += nb_bufs - nb_tx;
 491
 492         for (i = 0; likely(i < nb_tx); i++)
 493                 r->stats.bytes += bufs[i]->pkt_len;
 494
 495         vhost_update_packet_xstats(r, bufs, nb_tx);
 496
 497         /* According to RFC2863 page42 section ifHCOutMulticastPkts and
 498          * ifHCOutBroadcastPkts, the counters "multicast" and "broadcast"
 499          * are increased when packets are not transmitted successfully.
 500          */
 501         for (i = nb_tx; i < nb_bufs; i++)
 502                 vhost_count_multicast_broadcast(r, bufs[i]);
 503
 504         for (i = 0; likely(i < nb_tx); i++)
 505                 rte_pktmbuf_free(bufs[i]);
 506 out:
 507         rte_atomic32_set(&r->while_queuing, 0);
 508
 509         return nb_tx;
 510 }
 511
 512 static int
 513 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 514 {
 515         struct pmd_internal *internal = dev->data->dev_private;
 516         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
 517
 518         internal->vlan_strip = rxmode->hw_vlan_strip;
 519
 520         if (rxmode->hw_vlan_filter)
 521                 RTE_LOG(WARNING, PMD,
 522                         "vhost(%s): vlan filtering not available\n",
 523                         internal->dev_name);
 524
 525         return 0;
 526 }
 527
 528 static inline struct internal_list *
 529 find_internal_resource(char *ifname)
 530 {
 531         int found = 0;
 532         struct internal_list *list;
 533         struct pmd_internal *internal;
 534
 535         if (ifname == NULL)
 536                 return NULL;
 537
 538         pthread_mutex_lock(&internal_list_lock);
 539
 540         TAILQ_FOREACH(list, &internal_list, next) {
 541                 internal = list->eth_dev->data->dev_private;
 542                 if (!strcmp(internal->iface_name, ifname)) {
 543                         found = 1;
 544                         break;
 545                 }
 546         }
 547
 548         pthread_mutex_unlock(&internal_list_lock);
 549
 550         if (!found)
 551                 return NULL;
 552
 553         return list;
 554 }
 555
 556 static int
 557 eth_rxq_intr_enable(struct rte_eth_dev *dev, uint16_t qid)
 558 {
 559         struct rte_vhost_vring vring;
 560         struct vhost_queue *vq;
 561         int ret = 0;
 562
 563         vq = dev->data->rx_queues[qid];
 564         if (!vq) {
 565                 RTE_LOG(ERR, PMD, "rxq%d is not setup yet\n", qid);
 566                 return -1;
 567         }
 568
 569         ret = rte_vhost_get_vhost_vring(vq->vid, (qid << 1) + 1, &vring);
 570         if (ret < 0) {
 571                 RTE_LOG(ERR, PMD, "Failed to get rxq%d's vring\n", qid);
 572                 return ret;
 573         }
 574         RTE_LOG(INFO, PMD, "Enable interrupt for rxq%d\n", qid);
 575         rte_vhost_enable_guest_notification(vq->vid, (qid << 1) + 1, 1);
 576         rte_wmb();
 577
 578         return ret;
 579 }
 580
 581 static int
 582 eth_rxq_intr_disable(struct rte_eth_dev *dev, uint16_t qid)
 583 {
 584         struct rte_vhost_vring vring;
 585         struct vhost_queue *vq;
 586         int ret = 0;
 587
 588         vq = dev->data->rx_queues[qid];
 589         if (!vq) {
 590                 RTE_LOG(ERR, PMD, "rxq%d is not setup yet\n", qid);
 591                 return -1;
 592         }
 593
 594         ret = rte_vhost_get_vhost_vring(vq->vid, (qid << 1) + 1, &vring);
 595         if (ret < 0) {
 596                 RTE_LOG(ERR, PMD, "Failed to get rxq%d's vring", qid);
 597                 return ret;
 598         }
 599         RTE_LOG(INFO, PMD, "Disable interrupt for rxq%d\n", qid);
 600         rte_vhost_enable_guest_notification(vq->vid, (qid << 1) + 1, 0);
 601         rte_wmb();
 602
 603         return 0;
 604 }
 605
 606 static void
 607 eth_vhost_uninstall_intr(struct rte_eth_dev *dev)
 608 {
 609         struct rte_intr_handle *intr_handle = dev->intr_handle;
 610
 611         if (intr_handle) {
 612                 if (intr_handle->intr_vec)
 613                         free(intr_handle->intr_vec);
 614                 free(intr_handle);
 615         }
 616
 617         dev->intr_handle = NULL;
 618 }
 619
 620 static int
 621 eth_vhost_install_intr(struct rte_eth_dev *dev)
 622 {
 623         struct rte_vhost_vring vring;
 624         struct vhost_queue *vq;
 625         int count = 0;
 626         int nb_rxq = dev->data->nb_rx_queues;
 627         int i;
 628         int ret;
 629
 630         /* uninstall firstly if we are reconnecting */
 631         if (dev->intr_handle)
 632                 eth_vhost_uninstall_intr(dev);
 633
 634         dev->intr_handle = malloc(sizeof(*dev->intr_handle));
 635         if (!dev->intr_handle) {
 636                 RTE_LOG(ERR, PMD, "Fail to allocate intr_handle\n");
 637                 return -ENOMEM;
 638         }
 639         memset(dev->intr_handle, 0, sizeof(*dev->intr_handle));
 640
 641         dev->intr_handle->efd_counter_size = sizeof(uint64_t);
 642
 643         dev->intr_handle->intr_vec =
 644                 malloc(nb_rxq * sizeof(dev->intr_handle->intr_vec[0]));
 645
 646         if (!dev->intr_handle->intr_vec) {
 647                 RTE_LOG(ERR, PMD,
 648                         "Failed to allocate memory for interrupt vector\n");
 649                 free(dev->intr_handle);
 650                 return -ENOMEM;
 651         }
 652
 653         RTE_LOG(INFO, PMD, "Prepare intr vec\n");
 654         for (i = 0; i < nb_rxq; i++) {
 655                 vq = dev->data->rx_queues[i];
 656                 if (!vq) {
 657                         RTE_LOG(INFO, PMD, "rxq-%d not setup yet, skip!\n", i);
 658                         continue;
 659                 }
 660
 661                 ret = rte_vhost_get_vhost_vring(vq->vid, (i << 1) + 1, &vring);
 662                 if (ret < 0) {
 663                         RTE_LOG(INFO, PMD,
 664                                 "Failed to get rxq-%d's vring, skip!\n", i);
 665                         continue;
 666                 }
 667
 668                 if (vring.kickfd < 0) {
 669                         RTE_LOG(INFO, PMD,
 670                                 "rxq-%d's kickfd is invalid, skip!\n", i);
 671                         continue;
 672                 }
 673                 dev->intr_handle->intr_vec[i] = RTE_INTR_VEC_RXTX_OFFSET + i;
 674                 dev->intr_handle->efds[i] = vring.kickfd;
 675                 count++;
 676                 RTE_LOG(INFO, PMD, "Installed intr vec for rxq-%d\n", i);
 677         }
 678
 679         dev->intr_handle->nb_efd = count;
 680         dev->intr_handle->max_intr = count + 1;
 681         dev->intr_handle->type = RTE_INTR_HANDLE_VDEV;
 682
 683         return 0;
 684 }
 685
 686 static void
 687 update_queuing_status(struct rte_eth_dev *dev)
 688 {
 689         struct pmd_internal *internal = dev->data->dev_private;
 690         struct vhost_queue *vq;
 691         unsigned int i;
 692         int allow_queuing = 1;
 693
 694         if (!dev->data->rx_queues || !dev->data->tx_queues)
 695                 return;
 696
 697         if (rte_atomic32_read(&internal->started) == 0 ||
 698             rte_atomic32_read(&internal->dev_attached) == 0)
 699                 allow_queuing = 0;
 700
 701         /* Wait until rx/tx_pkt_burst stops accessing vhost device */
 702         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 703                 vq = dev->data->rx_queues[i];
 704                 if (vq == NULL)
 705                         continue;
 706                 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
 707                 while (rte_atomic32_read(&vq->while_queuing))
 708                         rte_pause();
 709         }
 710
 711         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 712                 vq = dev->data->tx_queues[i];
 713                 if (vq == NULL)
 714                         continue;
 715                 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
 716                 while (rte_atomic32_read(&vq->while_queuing))
 717                         rte_pause();
 718         }
 719 }
 720
 721 static void
 722 queue_setup(struct rte_eth_dev *eth_dev, struct pmd_internal *internal)
 723 {
 724         struct vhost_queue *vq;
 725         int i;
 726
 727         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 728                 vq = eth_dev->data->rx_queues[i];
 729                 if (!vq)
 730                         continue;
 731                 vq->vid = internal->vid;
 732                 vq->internal = internal;
 733                 vq->port = eth_dev->data->port_id;
 734         }
 735         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 736                 vq = eth_dev->data->tx_queues[i];
 737                 if (!vq)
 738                         continue;
 739                 vq->vid = internal->vid;
 740                 vq->internal = internal;
 741                 vq->port = eth_dev->data->port_id;
 742         }
 743 }
 744
 745 static int
 746 new_device(int vid)
 747 {
 748         struct rte_eth_dev *eth_dev;
 749         struct internal_list *list;
 750         struct pmd_internal *internal;
 751         struct rte_eth_conf *dev_conf;
 752         unsigned i;
 753         char ifname[PATH_MAX];
 754 #ifdef RTE_LIBRTE_VHOST_NUMA
 755         int newnode;
 756 #endif
 757
 758         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 759         list = find_internal_resource(ifname);
 760         if (list == NULL) {
 761                 RTE_LOG(INFO, PMD, "Invalid device name: %s\n", ifname);
 762                 return -1;
 763         }
 764
 765         eth_dev = list->eth_dev;
 766         internal = eth_dev->data->dev_private;
 767         dev_conf = &eth_dev->data->dev_conf;
 768
 769 #ifdef RTE_LIBRTE_VHOST_NUMA
 770         newnode = rte_vhost_get_numa_node(vid);
 771         if (newnode >= 0)
 772                 eth_dev->data->numa_node = newnode;
 773 #endif
 774
 775         internal->vid = vid;
 776         if (rte_atomic32_read(&internal->started) == 1) {
 777                 queue_setup(eth_dev, internal);
 778
 779                 if (dev_conf->intr_conf.rxq) {
 780                         if (eth_vhost_install_intr(eth_dev) < 0) {
 781                                 RTE_LOG(INFO, PMD,
 782                                         "Failed to install interrupt handler.");
 783                                         return -1;
 784                         }
 785                 }
 786         } else {
 787                 RTE_LOG(INFO, PMD, "RX/TX queues not exist yet\n");
 788         }
 789
 790         for (i = 0; i < rte_vhost_get_vring_num(vid); i++)
 791                 rte_vhost_enable_guest_notification(vid, i, 0);
 792
 793         rte_vhost_get_mtu(vid, &eth_dev->data->mtu);
 794
 795         eth_dev->data->dev_link.link_status = ETH_LINK_UP;
 796
 797         rte_atomic32_set(&internal->dev_attached, 1);
 798         update_queuing_status(eth_dev);
 799
 800         RTE_LOG(INFO, PMD, "Vhost device %d created\n", vid);
 801
 802         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 803
 804         return 0;
 805 }
 806
 807 static void
 808 destroy_device(int vid)
 809 {
 810         struct rte_eth_dev *eth_dev;
 811         struct pmd_internal *internal;
 812         struct vhost_queue *vq;
 813         struct internal_list *list;
 814         char ifname[PATH_MAX];
 815         unsigned i;
 816         struct rte_vhost_vring_state *state;
 817
 818         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 819         list = find_internal_resource(ifname);
 820         if (list == NULL) {
 821                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
 822                 return;
 823         }
 824         eth_dev = list->eth_dev;
 825         internal = eth_dev->data->dev_private;
 826
 827         rte_atomic32_set(&internal->dev_attached, 0);
 828         update_queuing_status(eth_dev);
 829
 830         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
 831
 832         if (eth_dev->data->rx_queues && eth_dev->data->tx_queues) {
 833                 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 834                         vq = eth_dev->data->rx_queues[i];
 835                         if (!vq)
 836                                 continue;
 837                         vq->vid = -1;
 838                 }
 839                 for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 840                         vq = eth_dev->data->tx_queues[i];
 841                         if (!vq)
 842                                 continue;
 843                         vq->vid = -1;
 844                 }
 845         }
 846
 847         state = vring_states[eth_dev->data->port_id];
 848         rte_spinlock_lock(&state->lock);
 849         for (i = 0; i <= state->max_vring; i++) {
 850                 state->cur[i] = false;
 851                 state->seen[i] = false;
 852         }
 853         state->max_vring = 0;
 854         rte_spinlock_unlock(&state->lock);
 855
 856         RTE_LOG(INFO, PMD, "Vhost device %d destroyed\n", vid);
 857         eth_vhost_uninstall_intr(eth_dev);
 858
 859         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 860 }
 861
 862 static int
 863 vring_state_changed(int vid, uint16_t vring, int enable)
 864 {
 865         struct rte_vhost_vring_state *state;
 866         struct rte_eth_dev *eth_dev;
 867         struct internal_list *list;
 868         char ifname[PATH_MAX];
 869
 870         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 871         list = find_internal_resource(ifname);
 872         if (list == NULL) {
 873                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
 874                 return -1;
 875         }
 876
 877         eth_dev = list->eth_dev;
 878         /* won't be NULL */
 879         state = vring_states[eth_dev->data->port_id];
 880         rte_spinlock_lock(&state->lock);
 881         state->cur[vring] = enable;
 882         state->max_vring = RTE_MAX(vring, state->max_vring);
 883         rte_spinlock_unlock(&state->lock);
 884
 885         RTE_LOG(INFO, PMD, "vring%u is %s\n",
 886                         vring, enable ? "enabled" : "disabled");
 887
 888         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE, NULL);
 889
 890         return 0;
 891 }
 892
 893 static struct vhost_device_ops vhost_ops = {
 894         .new_device          = new_device,
 895         .destroy_device      = destroy_device,
 896         .vring_state_changed = vring_state_changed,
 897 };
 898
 899 int
 900 rte_eth_vhost_get_queue_event(uint16_t port_id,
 901                 struct rte_eth_vhost_queue_event *event)
 902 {
 903         struct rte_vhost_vring_state *state;
 904         unsigned int i;
 905         int idx;
 906
 907         if (port_id >= RTE_MAX_ETHPORTS) {
 908                 RTE_LOG(ERR, PMD, "Invalid port id\n");
 909                 return -1;
 910         }
 911
 912         state = vring_states[port_id];
 913         if (!state) {
 914                 RTE_LOG(ERR, PMD, "Unused port\n");
 915                 return -1;
 916         }
 917
 918         rte_spinlock_lock(&state->lock);
 919         for (i = 0; i <= state->max_vring; i++) {
 920                 idx = state->index++ % (state->max_vring + 1);
 921
 922                 if (state->cur[idx] != state->seen[idx]) {
 923                         state->seen[idx] = state->cur[idx];
 924                         event->queue_id = idx / 2;
 925                         event->rx = idx & 1;
 926                         event->enable = state->cur[idx];
 927                         rte_spinlock_unlock(&state->lock);
 928                         return 0;
 929                 }
 930         }
 931         rte_spinlock_unlock(&state->lock);
 932
 933         return -1;
 934 }
 935
 936 int
 937 rte_eth_vhost_get_vid_from_port_id(uint16_t port_id)
 938 {
 939         struct internal_list *list;
 940         struct rte_eth_dev *eth_dev;
 941         struct vhost_queue *vq;
 942         int vid = -1;
 943
 944         if (!rte_eth_dev_is_valid_port(port_id))
 945                 return -1;
 946
 947         pthread_mutex_lock(&internal_list_lock);
 948
 949         TAILQ_FOREACH(list, &internal_list, next) {
 950                 eth_dev = list->eth_dev;
 951                 if (eth_dev->data->port_id == port_id) {
 952                         vq = eth_dev->data->rx_queues[0];
 953                         if (vq) {
 954                                 vid = vq->vid;
 955                         }
 956                         break;
 957                 }
 958         }
 959
 960         pthread_mutex_unlock(&internal_list_lock);
 961
 962         return vid;
 963 }
 964
 965 static int
 966 eth_dev_start(struct rte_eth_dev *eth_dev)
 967 {
 968         struct pmd_internal *internal = eth_dev->data->dev_private;
 969         struct rte_eth_conf *dev_conf = &eth_dev->data->dev_conf;
 970
 971         queue_setup(eth_dev, internal);
 972
 973         if (rte_atomic32_read(&internal->dev_attached) == 1) {
 974                 if (dev_conf->intr_conf.rxq) {
 975                         if (eth_vhost_install_intr(eth_dev) < 0) {
 976                                 RTE_LOG(INFO, PMD,
 977                                         "Failed to install interrupt handler.");
 978                                         return -1;
 979                         }
 980                 }
 981         }
 982
 983         rte_atomic32_set(&internal->started, 1);
 984         update_queuing_status(eth_dev);
 985
 986         return 0;
 987 }
 988
 989 static void
 990 eth_dev_stop(struct rte_eth_dev *dev)
 991 {
 992         struct pmd_internal *internal = dev->data->dev_private;
 993
 994         rte_atomic32_set(&internal->started, 0);
 995         update_queuing_status(dev);
 996 }
 997
 998 static void
 999 eth_dev_close(struct rte_eth_dev *dev)
1000 {
1001         struct pmd_internal *internal;
1002         struct internal_list *list;
1003         unsigned int i;
1004
1005         internal = dev->data->dev_private;
1006         if (!internal)
1007                 return;
1008
1009         eth_dev_stop(dev);
1010
1011         rte_vhost_driver_unregister(internal->iface_name);
1012
1013         list = find_internal_resource(internal->iface_name);
1014         if (!list)
1015                 return;
1016
1017         pthread_mutex_lock(&internal_list_lock);
1018         TAILQ_REMOVE(&internal_list, list, next);
1019         pthread_mutex_unlock(&internal_list_lock);
1020         rte_free(list);
1021
1022         if (dev->data->rx_queues)
1023                 for (i = 0; i < dev->data->nb_rx_queues; i++)
1024                         rte_free(dev->data->rx_queues[i]);
1025
1026         if (dev->data->tx_queues)
1027                 for (i = 0; i < dev->data->nb_tx_queues; i++)
1028                         rte_free(dev->data->tx_queues[i]);
1029
1030         rte_free(dev->data->mac_addrs);
1031         free(internal->dev_name);
1032         free(internal->iface_name);
1033         rte_free(internal);
1034
1035         dev->data->dev_private = NULL;
1036 }
1037
1038 static int
1039 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1040                    uint16_t nb_rx_desc __rte_unused,
1041                    unsigned int socket_id,
1042                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1043                    struct rte_mempool *mb_pool)
1044 {
1045         struct vhost_queue *vq;
1046
1047         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
1048                         RTE_CACHE_LINE_SIZE, socket_id);
1049         if (vq == NULL) {
1050                 RTE_LOG(ERR, PMD, "Failed to allocate memory for rx queue\n");
1051                 return -ENOMEM;
1052         }
1053
1054         vq->mb_pool = mb_pool;
1055         vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
1056         dev->data->rx_queues[rx_queue_id] = vq;
1057
1058         return 0;
1059 }
1060
1061 static int
1062 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1063                    uint16_t nb_tx_desc __rte_unused,
1064                    unsigned int socket_id,
1065                    const struct rte_eth_txconf *tx_conf __rte_unused)
1066 {
1067         struct vhost_queue *vq;
1068
1069         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
1070                         RTE_CACHE_LINE_SIZE, socket_id);
1071         if (vq == NULL) {
1072                 RTE_LOG(ERR, PMD, "Failed to allocate memory for tx queue\n");
1073                 return -ENOMEM;
1074         }
1075
1076         vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
1077         dev->data->tx_queues[tx_queue_id] = vq;
1078
1079         return 0;
1080 }
1081
1082 static void
1083 eth_dev_info(struct rte_eth_dev *dev,
1084              struct rte_eth_dev_info *dev_info)
1085 {
1086         struct pmd_internal *internal;
1087
1088         internal = dev->data->dev_private;
1089         if (internal == NULL) {
1090                 RTE_LOG(ERR, PMD, "Invalid device specified\n");
1091                 return;
1092         }
1093
1094         dev_info->max_mac_addrs = 1;
1095         dev_info->max_rx_pktlen = (uint32_t)-1;
1096         dev_info->max_rx_queues = internal->max_queues;
1097         dev_info->max_tx_queues = internal->max_queues;
1098         dev_info->min_rx_bufsize = 0;
1099 }
1100
1101 static int
1102 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1103 {
1104         unsigned i;
1105         unsigned long rx_total = 0, tx_total = 0, tx_missed_total = 0;
1106         unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
1107         struct vhost_queue *vq;
1108
1109         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
1110                         i < dev->data->nb_rx_queues; i++) {
1111                 if (dev->data->rx_queues[i] == NULL)
1112                         continue;
1113                 vq = dev->data->rx_queues[i];
1114                 stats->q_ipackets[i] = vq->stats.pkts;
1115                 rx_total += stats->q_ipackets[i];
1116
1117                 stats->q_ibytes[i] = vq->stats.bytes;
1118                 rx_total_bytes += stats->q_ibytes[i];
1119         }
1120
1121         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
1122                         i < dev->data->nb_tx_queues; i++) {
1123                 if (dev->data->tx_queues[i] == NULL)
1124                         continue;
1125                 vq = dev->data->tx_queues[i];
1126                 stats->q_opackets[i] = vq->stats.pkts;
1127                 tx_missed_total += vq->stats.missed_pkts;
1128                 tx_total += stats->q_opackets[i];
1129
1130                 stats->q_obytes[i] = vq->stats.bytes;
1131                 tx_total_bytes += stats->q_obytes[i];
1132         }
1133
1134         stats->ipackets = rx_total;
1135         stats->opackets = tx_total;
1136         stats->oerrors = tx_missed_total;
1137         stats->ibytes = rx_total_bytes;
1138         stats->obytes = tx_total_bytes;
1139
1140         return 0;
1141 }
1142
1143 static void
1144 eth_stats_reset(struct rte_eth_dev *dev)
1145 {
1146         struct vhost_queue *vq;
1147         unsigned i;
1148
1149         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1150                 if (dev->data->rx_queues[i] == NULL)
1151                         continue;
1152                 vq = dev->data->rx_queues[i];
1153                 vq->stats.pkts = 0;
1154                 vq->stats.bytes = 0;
1155         }
1156         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1157                 if (dev->data->tx_queues[i] == NULL)
1158                         continue;
1159                 vq = dev->data->tx_queues[i];
1160                 vq->stats.pkts = 0;
1161                 vq->stats.bytes = 0;
1162                 vq->stats.missed_pkts = 0;
1163         }
1164 }
1165
1166 static void
1167 eth_queue_release(void *q)
1168 {
1169         rte_free(q);
1170 }
1171
1172 static int
1173 eth_tx_done_cleanup(void *txq __rte_unused, uint32_t free_cnt __rte_unused)
1174 {
1175         /*
1176          * vHost does not hang onto mbuf. eth_vhost_tx() copies packet data
1177          * and releases mbuf, so nothing to cleanup.
1178          */
1179         return 0;
1180 }
1181
1182 static int
1183 eth_link_update(struct rte_eth_dev *dev __rte_unused,
1184                 int wait_to_complete __rte_unused)
1185 {
1186         return 0;
1187 }
1188
1189 static uint32_t
1190 eth_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1191 {
1192         struct vhost_queue *vq;
1193
1194         vq = dev->data->rx_queues[rx_queue_id];
1195         if (vq == NULL)
1196                 return 0;
1197
1198         return rte_vhost_rx_queue_count(vq->vid, vq->virtqueue_id);
1199 }
1200
1201 static const struct eth_dev_ops ops = {
1202         .dev_start = eth_dev_start,
1203         .dev_stop = eth_dev_stop,
1204         .dev_close = eth_dev_close,
1205         .dev_configure = eth_dev_configure,
1206         .dev_infos_get = eth_dev_info,
1207         .rx_queue_setup = eth_rx_queue_setup,
1208         .tx_queue_setup = eth_tx_queue_setup,
1209         .rx_queue_release = eth_queue_release,
1210         .tx_queue_release = eth_queue_release,
1211         .tx_done_cleanup = eth_tx_done_cleanup,
1212         .rx_queue_count = eth_rx_queue_count,
1213         .link_update = eth_link_update,
1214         .stats_get = eth_stats_get,
1215         .stats_reset = eth_stats_reset,
1216         .xstats_reset = vhost_dev_xstats_reset,
1217         .xstats_get = vhost_dev_xstats_get,
1218         .xstats_get_names = vhost_dev_xstats_get_names,
1219         .rx_queue_intr_enable = eth_rxq_intr_enable,
1220         .rx_queue_intr_disable = eth_rxq_intr_disable,
1221 };
1222
1223 static struct rte_vdev_driver pmd_vhost_drv;
1224
1225 static int
1226 eth_dev_vhost_create(struct rte_vdev_device *dev, char *iface_name,
1227         int16_t queues, const unsigned int numa_node, uint64_t flags)
1228 {
1229         const char *name = rte_vdev_device_name(dev);
1230         struct rte_eth_dev_data *data;
1231         struct pmd_internal *internal = NULL;
1232         struct rte_eth_dev *eth_dev = NULL;
1233         struct ether_addr *eth_addr = NULL;
1234         struct rte_vhost_vring_state *vring_state = NULL;
1235         struct internal_list *list = NULL;
1236
1237         RTE_LOG(INFO, PMD, "Creating VHOST-USER backend on numa socket %u\n",
1238                 numa_node);
1239
1240         list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
1241         if (list == NULL)
1242                 goto error;
1243
1244         /* reserve an ethdev entry */
1245         eth_dev = rte_eth_vdev_allocate(dev, sizeof(*internal));
1246         if (eth_dev == NULL)
1247                 goto error;
1248
1249         eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
1250         if (eth_addr == NULL)
1251                 goto error;
1252         *eth_addr = base_eth_addr;
1253         eth_addr->addr_bytes[5] = eth_dev->data->port_id;
1254
1255         vring_state = rte_zmalloc_socket(name,
1256                         sizeof(*vring_state), 0, numa_node);
1257         if (vring_state == NULL)
1258                 goto error;
1259
1260         /* now put it all together
1261          * - store queue data in internal,
1262          * - point eth_dev_data to internals
1263          * - and point eth_dev structure to new eth_dev_data structure
1264          */
1265         internal = eth_dev->data->dev_private;
1266         internal->dev_name = strdup(name);
1267         if (internal->dev_name == NULL)
1268                 goto error;
1269         internal->iface_name = strdup(iface_name);
1270         if (internal->iface_name == NULL)
1271                 goto error;
1272
1273         list->eth_dev = eth_dev;
1274         pthread_mutex_lock(&internal_list_lock);
1275         TAILQ_INSERT_TAIL(&internal_list, list, next);
1276         pthread_mutex_unlock(&internal_list_lock);
1277
1278         rte_spinlock_init(&vring_state->lock);
1279         vring_states[eth_dev->data->port_id] = vring_state;
1280
1281         data = eth_dev->data;
1282         data->nb_rx_queues = queues;
1283         data->nb_tx_queues = queues;
1284         internal->max_queues = queues;
1285         data->dev_link = pmd_link;
1286         data->mac_addrs = eth_addr;
1287         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1288
1289         eth_dev->dev_ops = &ops;
1290
1291         /* finally assign rx and tx ops */
1292         eth_dev->rx_pkt_burst = eth_vhost_rx;
1293         eth_dev->tx_pkt_burst = eth_vhost_tx;
1294
1295         if (rte_vhost_driver_register(iface_name, flags))
1296                 goto error;
1297
1298         if (rte_vhost_driver_callback_register(iface_name, &vhost_ops) < 0) {
1299                 RTE_LOG(ERR, PMD, "Can't register callbacks\n");
1300                 goto error;
1301         }
1302
1303         if (rte_vhost_driver_start(iface_name) < 0) {
1304                 RTE_LOG(ERR, PMD, "Failed to start driver for %s\n",
1305                         iface_name);
1306                 goto error;
1307         }
1308
1309         return data->port_id;
1310
1311 error:
1312         if (internal) {
1313                 free(internal->iface_name);
1314                 free(internal->dev_name);
1315         }
1316         rte_free(vring_state);
1317         rte_free(eth_addr);
1318         if (eth_dev)
1319                 rte_eth_dev_release_port(eth_dev);
1320         rte_free(internal);
1321         rte_free(list);
1322
1323         return -1;
1324 }
1325
1326 static inline int
1327 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
1328 {
1329         const char **iface_name = extra_args;
1330
1331         if (value == NULL)
1332                 return -1;
1333
1334         *iface_name = value;
1335
1336         return 0;
1337 }
1338
1339 static inline int
1340 open_int(const char *key __rte_unused, const char *value, void *extra_args)
1341 {
1342         uint16_t *n = extra_args;
1343
1344         if (value == NULL || extra_args == NULL)
1345                 return -EINVAL;
1346
1347         *n = (uint16_t)strtoul(value, NULL, 0);
1348         if (*n == USHRT_MAX && errno == ERANGE)
1349                 return -1;
1350
1351         return 0;
1352 }
1353
1354 static int
1355 rte_pmd_vhost_probe(struct rte_vdev_device *dev)
1356 {
1357         struct rte_kvargs *kvlist = NULL;
1358         int ret = 0;
1359         char *iface_name;
1360         uint16_t queues;
1361         uint64_t flags = 0;
1362         int client_mode = 0;
1363         int dequeue_zero_copy = 0;
1364         int iommu_support = 0;
1365         struct rte_eth_dev *eth_dev;
1366         const char *name = rte_vdev_device_name(dev);
1367
1368         RTE_LOG(INFO, PMD, "Initializing pmd_vhost for %s\n", name);
1369
1370         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1371             strlen(rte_vdev_device_args(dev)) == 0) {
1372                 eth_dev = rte_eth_dev_attach_secondary(name);
1373                 if (!eth_dev) {
1374                         RTE_LOG(ERR, PMD, "Failed to probe %s\n", name);
1375                         return -1;
1376                 }
1377                 /* TODO: request info from primary to set up Rx and Tx */
1378                 eth_dev->dev_ops = &ops;
1379                 return 0;
1380         }
1381
1382         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
1383         if (kvlist == NULL)
1384                 return -1;
1385
1386         if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
1387                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
1388                                          &open_iface, &iface_name);
1389                 if (ret < 0)
1390                         goto out_free;
1391         } else {
1392                 ret = -1;
1393                 goto out_free;
1394         }
1395
1396         if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
1397                 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
1398                                          &open_int, &queues);
1399                 if (ret < 0 || queues > RTE_MAX_QUEUES_PER_PORT)
1400                         goto out_free;
1401
1402         } else
1403                 queues = 1;
1404
1405         if (rte_kvargs_count(kvlist, ETH_VHOST_CLIENT_ARG) == 1) {
1406                 ret = rte_kvargs_process(kvlist, ETH_VHOST_CLIENT_ARG,
1407                                          &open_int, &client_mode);
1408                 if (ret < 0)
1409                         goto out_free;
1410
1411                 if (client_mode)
1412                         flags |= RTE_VHOST_USER_CLIENT;
1413         }
1414
1415         if (rte_kvargs_count(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY) == 1) {
1416                 ret = rte_kvargs_process(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY,
1417                                          &open_int, &dequeue_zero_copy);
1418                 if (ret < 0)
1419                         goto out_free;
1420
1421                 if (dequeue_zero_copy)
1422                         flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
1423         }
1424
1425         if (rte_kvargs_count(kvlist, ETH_VHOST_IOMMU_SUPPORT) == 1) {
1426                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IOMMU_SUPPORT,
1427                                          &open_int, &iommu_support);
1428                 if (ret < 0)
1429                         goto out_free;
1430
1431                 if (iommu_support)
1432                         flags |= RTE_VHOST_USER_IOMMU_SUPPORT;
1433         }
1434
1435         if (dev->device.numa_node == SOCKET_ID_ANY)
1436                 dev->device.numa_node = rte_socket_id();
1437
1438         eth_dev_vhost_create(dev, iface_name, queues, dev->device.numa_node,
1439                 flags);
1440
1441 out_free:
1442         rte_kvargs_free(kvlist);
1443         return ret;
1444 }
1445
1446 static int
1447 rte_pmd_vhost_remove(struct rte_vdev_device *dev)
1448 {
1449         const char *name;
1450         struct rte_eth_dev *eth_dev = NULL;
1451
1452         name = rte_vdev_device_name(dev);
1453         RTE_LOG(INFO, PMD, "Un-Initializing pmd_vhost for %s\n", name);
1454
1455         /* find an ethdev entry */
1456         eth_dev = rte_eth_dev_allocated(name);
1457         if (eth_dev == NULL)
1458                 return -ENODEV;
1459
1460         eth_dev_close(eth_dev);
1461
1462         rte_free(vring_states[eth_dev->data->port_id]);
1463         vring_states[eth_dev->data->port_id] = NULL;
1464
1465         rte_eth_dev_release_port(eth_dev);
1466
1467         return 0;
1468 }
1469
1470 static struct rte_vdev_driver pmd_vhost_drv = {
1471         .probe = rte_pmd_vhost_probe,
1472         .remove = rte_pmd_vhost_remove,
1473 };
1474
1475 RTE_PMD_REGISTER_VDEV(net_vhost, pmd_vhost_drv);
1476 RTE_PMD_REGISTER_ALIAS(net_vhost, eth_vhost);
1477 RTE_PMD_REGISTER_PARAM_STRING(net_vhost,
1478         "iface=<ifc> "
1479         "queues=<int>");