drivers/net/vhost/rte_eth_vhost.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016 IGEL Co., Ltd.
   3  * Copyright(c) 2016-2018 Intel Corporation
   4  */
   5 #include <unistd.h>
   6 #include <pthread.h>
   7 #include <stdbool.h>
   8
   9 #include <rte_mbuf.h>
  10 #include <rte_ethdev_driver.h>
  11 #include <rte_ethdev_vdev.h>
  12 #include <rte_malloc.h>
  13 #include <rte_memcpy.h>
  14 #include <rte_bus_vdev.h>
  15 #include <rte_kvargs.h>
  16 #include <rte_vhost.h>
  17 #include <rte_spinlock.h>
  18
  19 #include "rte_eth_vhost.h"
  20
  21 RTE_LOG_REGISTER(vhost_logtype, pmd.net.vhost, NOTICE);
  22
  23 #define VHOST_LOG(level, ...) \
  24         rte_log(RTE_LOG_ ## level, vhost_logtype, __VA_ARGS__)
  25
  26 enum {VIRTIO_RXQ, VIRTIO_TXQ, VIRTIO_QNUM};
  27
  28 #define ETH_VHOST_IFACE_ARG             "iface"
  29 #define ETH_VHOST_QUEUES_ARG            "queues"
  30 #define ETH_VHOST_CLIENT_ARG            "client"
  31 #define ETH_VHOST_DEQUEUE_ZERO_COPY     "dequeue-zero-copy"
  32 #define ETH_VHOST_IOMMU_SUPPORT         "iommu-support"
  33 #define ETH_VHOST_POSTCOPY_SUPPORT      "postcopy-support"
  34 #define ETH_VHOST_VIRTIO_NET_F_HOST_TSO "tso"
  35 #define ETH_VHOST_LINEAR_BUF  "linear-buffer"
  36 #define ETH_VHOST_EXT_BUF  "ext-buffer"
  37 #define VHOST_MAX_PKT_BURST 32
  38
  39 static const char *valid_arguments[] = {
  40         ETH_VHOST_IFACE_ARG,
  41         ETH_VHOST_QUEUES_ARG,
  42         ETH_VHOST_CLIENT_ARG,
  43         ETH_VHOST_DEQUEUE_ZERO_COPY,
  44         ETH_VHOST_IOMMU_SUPPORT,
  45         ETH_VHOST_POSTCOPY_SUPPORT,
  46         ETH_VHOST_VIRTIO_NET_F_HOST_TSO,
  47         ETH_VHOST_LINEAR_BUF,
  48         ETH_VHOST_EXT_BUF,
  49         NULL
  50 };
  51
  52 static struct rte_ether_addr base_eth_addr = {
  53         .addr_bytes = {
  54                 0x56 /* V */,
  55                 0x48 /* H */,
  56                 0x4F /* O */,
  57                 0x53 /* S */,
  58                 0x54 /* T */,
  59                 0x00
  60         }
  61 };
  62
  63 enum vhost_xstats_pkts {
  64         VHOST_UNDERSIZE_PKT = 0,
  65         VHOST_64_PKT,
  66         VHOST_65_TO_127_PKT,
  67         VHOST_128_TO_255_PKT,
  68         VHOST_256_TO_511_PKT,
  69         VHOST_512_TO_1023_PKT,
  70         VHOST_1024_TO_1522_PKT,
  71         VHOST_1523_TO_MAX_PKT,
  72         VHOST_BROADCAST_PKT,
  73         VHOST_MULTICAST_PKT,
  74         VHOST_UNICAST_PKT,
  75         VHOST_ERRORS_PKT,
  76         VHOST_ERRORS_FRAGMENTED,
  77         VHOST_ERRORS_JABBER,
  78         VHOST_UNKNOWN_PROTOCOL,
  79         VHOST_XSTATS_MAX,
  80 };
  81
  82 struct vhost_stats {
  83         uint64_t pkts;
  84         uint64_t bytes;
  85         uint64_t missed_pkts;
  86         uint64_t xstats[VHOST_XSTATS_MAX];
  87 };
  88
  89 struct vhost_queue {
  90         int vid;
  91         rte_atomic32_t allow_queuing;
  92         rte_atomic32_t while_queuing;
  93         struct pmd_internal *internal;
  94         struct rte_mempool *mb_pool;
  95         uint16_t port;
  96         uint16_t virtqueue_id;
  97         bool intr_en;
  98         struct vhost_stats stats;
  99 };
 100
 101 struct pmd_internal {
 102         rte_atomic32_t dev_attached;
 103         char *iface_name;
 104         uint64_t flags;
 105         uint64_t disable_flags;
 106         uint16_t max_queues;
 107         int vid;
 108         rte_atomic32_t started;
 109         uint8_t vlan_strip;
 110 };
 111
 112 struct internal_list {
 113         TAILQ_ENTRY(internal_list) next;
 114         struct rte_eth_dev *eth_dev;
 115 };
 116
 117 TAILQ_HEAD(internal_list_head, internal_list);
 118 static struct internal_list_head internal_list =
 119         TAILQ_HEAD_INITIALIZER(internal_list);
 120
 121 static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
 122
 123 static struct rte_eth_link pmd_link = {
 124                 .link_speed = 10000,
 125                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 126                 .link_status = ETH_LINK_DOWN
 127 };
 128
 129 struct rte_vhost_vring_state {
 130         rte_spinlock_t lock;
 131
 132         bool cur[RTE_MAX_QUEUES_PER_PORT * 2];
 133         bool seen[RTE_MAX_QUEUES_PER_PORT * 2];
 134         unsigned int index;
 135         unsigned int max_vring;
 136 };
 137
 138 static struct rte_vhost_vring_state *vring_states[RTE_MAX_ETHPORTS];
 139
 140 #define VHOST_XSTATS_NAME_SIZE 64
 141
 142 struct vhost_xstats_name_off {
 143         char name[VHOST_XSTATS_NAME_SIZE];
 144         uint64_t offset;
 145 };
 146
 147 /* [rx]_is prepended to the name string here */
 148 static const struct vhost_xstats_name_off vhost_rxport_stat_strings[] = {
 149         {"good_packets",
 150          offsetof(struct vhost_queue, stats.pkts)},
 151         {"total_bytes",
 152          offsetof(struct vhost_queue, stats.bytes)},
 153         {"missed_pkts",
 154          offsetof(struct vhost_queue, stats.missed_pkts)},
 155         {"broadcast_packets",
 156          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
 157         {"multicast_packets",
 158          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
 159         {"unicast_packets",
 160          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
 161          {"undersize_packets",
 162          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
 163         {"size_64_packets",
 164          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
 165         {"size_65_to_127_packets",
 166          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
 167         {"size_128_to_255_packets",
 168          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
 169         {"size_256_to_511_packets",
 170          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
 171         {"size_512_to_1023_packets",
 172          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
 173         {"size_1024_to_1522_packets",
 174          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
 175         {"size_1523_to_max_packets",
 176          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
 177         {"errors_with_bad_CRC",
 178          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
 179         {"fragmented_errors",
 180          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_FRAGMENTED])},
 181         {"jabber_errors",
 182          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_JABBER])},
 183         {"unknown_protos_packets",
 184          offsetof(struct vhost_queue, stats.xstats[VHOST_UNKNOWN_PROTOCOL])},
 185 };
 186
 187 /* [tx]_ is prepended to the name string here */
 188 static const struct vhost_xstats_name_off vhost_txport_stat_strings[] = {
 189         {"good_packets",
 190          offsetof(struct vhost_queue, stats.pkts)},
 191         {"total_bytes",
 192          offsetof(struct vhost_queue, stats.bytes)},
 193         {"missed_pkts",
 194          offsetof(struct vhost_queue, stats.missed_pkts)},
 195         {"broadcast_packets",
 196          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
 197         {"multicast_packets",
 198          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
 199         {"unicast_packets",
 200          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
 201         {"undersize_packets",
 202          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
 203         {"size_64_packets",
 204          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
 205         {"size_65_to_127_packets",
 206          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
 207         {"size_128_to_255_packets",
 208          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
 209         {"size_256_to_511_packets",
 210          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
 211         {"size_512_to_1023_packets",
 212          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
 213         {"size_1024_to_1522_packets",
 214          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
 215         {"size_1523_to_max_packets",
 216          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
 217         {"errors_with_bad_CRC",
 218          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
 219 };
 220
 221 #define VHOST_NB_XSTATS_RXPORT (sizeof(vhost_rxport_stat_strings) / \
 222                                 sizeof(vhost_rxport_stat_strings[0]))
 223
 224 #define VHOST_NB_XSTATS_TXPORT (sizeof(vhost_txport_stat_strings) / \
 225                                 sizeof(vhost_txport_stat_strings[0]))
 226
 227 static int
 228 vhost_dev_xstats_reset(struct rte_eth_dev *dev)
 229 {
 230         struct vhost_queue *vq = NULL;
 231         unsigned int i = 0;
 232
 233         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 234                 vq = dev->data->rx_queues[i];
 235                 if (!vq)
 236                         continue;
 237                 memset(&vq->stats, 0, sizeof(vq->stats));
 238         }
 239         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 240                 vq = dev->data->tx_queues[i];
 241                 if (!vq)
 242                         continue;
 243                 memset(&vq->stats, 0, sizeof(vq->stats));
 244         }
 245
 246         return 0;
 247 }
 248
 249 static int
 250 vhost_dev_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
 251                            struct rte_eth_xstat_name *xstats_names,
 252                            unsigned int limit __rte_unused)
 253 {
 254         unsigned int t = 0;
 255         int count = 0;
 256         int nstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
 257
 258         if (!xstats_names)
 259                 return nstats;
 260         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
 261                 snprintf(xstats_names[count].name,
 262                          sizeof(xstats_names[count].name),
 263                          "rx_%s", vhost_rxport_stat_strings[t].name);
 264                 count++;
 265         }
 266         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
 267                 snprintf(xstats_names[count].name,
 268                          sizeof(xstats_names[count].name),
 269                          "tx_%s", vhost_txport_stat_strings[t].name);
 270                 count++;
 271         }
 272         return count;
 273 }
 274
 275 static int
 276 vhost_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 277                      unsigned int n)
 278 {
 279         unsigned int i;
 280         unsigned int t;
 281         unsigned int count = 0;
 282         struct vhost_queue *vq = NULL;
 283         unsigned int nxstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
 284
 285         if (n < nxstats)
 286                 return nxstats;
 287
 288         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 289                 vq = dev->data->rx_queues[i];
 290                 if (!vq)
 291                         continue;
 292                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
 293                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
 294                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
 295         }
 296         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 297                 vq = dev->data->tx_queues[i];
 298                 if (!vq)
 299                         continue;
 300                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
 301                                 + vq->stats.missed_pkts
 302                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
 303                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
 304         }
 305         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
 306                 xstats[count].value = 0;
 307                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 308                         vq = dev->data->rx_queues[i];
 309                         if (!vq)
 310                                 continue;
 311                         xstats[count].value +=
 312                                 *(uint64_t *)(((char *)vq)
 313                                 + vhost_rxport_stat_strings[t].offset);
 314                 }
 315                 xstats[count].id = count;
 316                 count++;
 317         }
 318         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
 319                 xstats[count].value = 0;
 320                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
 321                         vq = dev->data->tx_queues[i];
 322                         if (!vq)
 323                                 continue;
 324                         xstats[count].value +=
 325                                 *(uint64_t *)(((char *)vq)
 326                                 + vhost_txport_stat_strings[t].offset);
 327                 }
 328                 xstats[count].id = count;
 329                 count++;
 330         }
 331         return count;
 332 }
 333
 334 static inline void
 335 vhost_count_multicast_broadcast(struct vhost_queue *vq,
 336                                 struct rte_mbuf *mbuf)
 337 {
 338         struct rte_ether_addr *ea = NULL;
 339         struct vhost_stats *pstats = &vq->stats;
 340
 341         ea = rte_pktmbuf_mtod(mbuf, struct rte_ether_addr *);
 342         if (rte_is_multicast_ether_addr(ea)) {
 343                 if (rte_is_broadcast_ether_addr(ea))
 344                         pstats->xstats[VHOST_BROADCAST_PKT]++;
 345                 else
 346                         pstats->xstats[VHOST_MULTICAST_PKT]++;
 347         }
 348 }
 349
 350 static void
 351 vhost_update_packet_xstats(struct vhost_queue *vq,
 352                            struct rte_mbuf **bufs,
 353                            uint16_t count)
 354 {
 355         uint32_t pkt_len = 0;
 356         uint64_t i = 0;
 357         uint64_t index;
 358         struct vhost_stats *pstats = &vq->stats;
 359
 360         for (i = 0; i < count ; i++) {
 361                 pkt_len = bufs[i]->pkt_len;
 362                 if (pkt_len == 64) {
 363                         pstats->xstats[VHOST_64_PKT]++;
 364                 } else if (pkt_len > 64 && pkt_len < 1024) {
 365                         index = (sizeof(pkt_len) * 8)
 366                                 - __builtin_clz(pkt_len) - 5;
 367                         pstats->xstats[index]++;
 368                 } else {
 369                         if (pkt_len < 64)
 370                                 pstats->xstats[VHOST_UNDERSIZE_PKT]++;
 371                         else if (pkt_len <= 1522)
 372                                 pstats->xstats[VHOST_1024_TO_1522_PKT]++;
 373                         else if (pkt_len > 1522)
 374                                 pstats->xstats[VHOST_1523_TO_MAX_PKT]++;
 375                 }
 376                 vhost_count_multicast_broadcast(vq, bufs[i]);
 377         }
 378 }
 379
 380 static uint16_t
 381 eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 382 {
 383         struct vhost_queue *r = q;
 384         uint16_t i, nb_rx = 0;
 385         uint16_t nb_receive = nb_bufs;
 386
 387         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 388                 return 0;
 389
 390         rte_atomic32_set(&r->while_queuing, 1);
 391
 392         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 393                 goto out;
 394
 395         /* Dequeue packets from guest TX queue */
 396         while (nb_receive) {
 397                 uint16_t nb_pkts;
 398                 uint16_t num = (uint16_t)RTE_MIN(nb_receive,
 399                                                  VHOST_MAX_PKT_BURST);
 400
 401                 nb_pkts = rte_vhost_dequeue_burst(r->vid, r->virtqueue_id,
 402                                                   r->mb_pool, &bufs[nb_rx],
 403                                                   num);
 404
 405                 nb_rx += nb_pkts;
 406                 nb_receive -= nb_pkts;
 407                 if (nb_pkts < num)
 408                         break;
 409         }
 410
 411         r->stats.pkts += nb_rx;
 412
 413         for (i = 0; likely(i < nb_rx); i++) {
 414                 bufs[i]->port = r->port;
 415                 bufs[i]->vlan_tci = 0;
 416
 417                 if (r->internal->vlan_strip)
 418                         rte_vlan_strip(bufs[i]);
 419
 420                 r->stats.bytes += bufs[i]->pkt_len;
 421         }
 422
 423         vhost_update_packet_xstats(r, bufs, nb_rx);
 424
 425 out:
 426         rte_atomic32_set(&r->while_queuing, 0);
 427
 428         return nb_rx;
 429 }
 430
 431 static uint16_t
 432 eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 433 {
 434         struct vhost_queue *r = q;
 435         uint16_t i, nb_tx = 0;
 436         uint16_t nb_send = 0;
 437
 438         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 439                 return 0;
 440
 441         rte_atomic32_set(&r->while_queuing, 1);
 442
 443         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 444                 goto out;
 445
 446         for (i = 0; i < nb_bufs; i++) {
 447                 struct rte_mbuf *m = bufs[i];
 448
 449                 /* Do VLAN tag insertion */
 450                 if (m->ol_flags & PKT_TX_VLAN_PKT) {
 451                         int error = rte_vlan_insert(&m);
 452                         if (unlikely(error)) {
 453                                 rte_pktmbuf_free(m);
 454                                 continue;
 455                         }
 456                 }
 457
 458                 bufs[nb_send] = m;
 459                 ++nb_send;
 460         }
 461
 462         /* Enqueue packets to guest RX queue */
 463         while (nb_send) {
 464                 uint16_t nb_pkts;
 465                 uint16_t num = (uint16_t)RTE_MIN(nb_send,
 466                                                  VHOST_MAX_PKT_BURST);
 467
 468                 nb_pkts = rte_vhost_enqueue_burst(r->vid, r->virtqueue_id,
 469                                                   &bufs[nb_tx], num);
 470
 471                 nb_tx += nb_pkts;
 472                 nb_send -= nb_pkts;
 473                 if (nb_pkts < num)
 474                         break;
 475         }
 476
 477         r->stats.pkts += nb_tx;
 478         r->stats.missed_pkts += nb_bufs - nb_tx;
 479
 480         for (i = 0; likely(i < nb_tx); i++)
 481                 r->stats.bytes += bufs[i]->pkt_len;
 482
 483         vhost_update_packet_xstats(r, bufs, nb_tx);
 484
 485         /* According to RFC2863 page42 section ifHCOutMulticastPkts and
 486          * ifHCOutBroadcastPkts, the counters "multicast" and "broadcast"
 487          * are increased when packets are not transmitted successfully.
 488          */
 489         for (i = nb_tx; i < nb_bufs; i++)
 490                 vhost_count_multicast_broadcast(r, bufs[i]);
 491
 492         for (i = 0; likely(i < nb_tx); i++)
 493                 rte_pktmbuf_free(bufs[i]);
 494 out:
 495         rte_atomic32_set(&r->while_queuing, 0);
 496
 497         return nb_tx;
 498 }
 499
 500 static inline struct internal_list *
 501 find_internal_resource(char *ifname)
 502 {
 503         int found = 0;
 504         struct internal_list *list;
 505         struct pmd_internal *internal;
 506
 507         if (ifname == NULL)
 508                 return NULL;
 509
 510         pthread_mutex_lock(&internal_list_lock);
 511
 512         TAILQ_FOREACH(list, &internal_list, next) {
 513                 internal = list->eth_dev->data->dev_private;
 514                 if (!strcmp(internal->iface_name, ifname)) {
 515                         found = 1;
 516                         break;
 517                 }
 518         }
 519
 520         pthread_mutex_unlock(&internal_list_lock);
 521
 522         if (!found)
 523                 return NULL;
 524
 525         return list;
 526 }
 527
 528 static int
 529 eth_rxq_intr_enable(struct rte_eth_dev *dev, uint16_t qid)
 530 {
 531         struct rte_vhost_vring vring;
 532         struct vhost_queue *vq;
 533         int ret = 0;
 534
 535         vq = dev->data->rx_queues[qid];
 536         if (!vq) {
 537                 VHOST_LOG(ERR, "rxq%d is not setup yet\n", qid);
 538                 return -1;
 539         }
 540
 541         ret = rte_vhost_get_vhost_vring(vq->vid, (qid << 1) + 1, &vring);
 542         if (ret < 0) {
 543                 VHOST_LOG(ERR, "Failed to get rxq%d's vring\n", qid);
 544                 return ret;
 545         }
 546         VHOST_LOG(INFO, "Enable interrupt for rxq%d\n", qid);
 547         rte_vhost_enable_guest_notification(vq->vid, (qid << 1) + 1, 1);
 548         rte_wmb();
 549
 550         vq->intr_en = true;
 551
 552         return ret;
 553 }
 554
 555 static int
 556 eth_rxq_intr_disable(struct rte_eth_dev *dev, uint16_t qid)
 557 {
 558         struct rte_vhost_vring vring;
 559         struct vhost_queue *vq;
 560         int ret = 0;
 561
 562         vq = dev->data->rx_queues[qid];
 563         if (!vq) {
 564                 VHOST_LOG(ERR, "rxq%d is not setup yet\n", qid);
 565                 return -1;
 566         }
 567
 568         ret = rte_vhost_get_vhost_vring(vq->vid, (qid << 1) + 1, &vring);
 569         if (ret < 0) {
 570                 VHOST_LOG(ERR, "Failed to get rxq%d's vring", qid);
 571                 return ret;
 572         }
 573         VHOST_LOG(INFO, "Disable interrupt for rxq%d\n", qid);
 574         rte_vhost_enable_guest_notification(vq->vid, (qid << 1) + 1, 0);
 575         rte_wmb();
 576
 577         vq->intr_en = false;
 578
 579         return 0;
 580 }
 581
 582 static void
 583 eth_vhost_uninstall_intr(struct rte_eth_dev *dev)
 584 {
 585         struct rte_intr_handle *intr_handle = dev->intr_handle;
 586
 587         if (intr_handle) {
 588                 if (intr_handle->intr_vec)
 589                         free(intr_handle->intr_vec);
 590                 free(intr_handle);
 591         }
 592
 593         dev->intr_handle = NULL;
 594 }
 595
 596 static int
 597 eth_vhost_install_intr(struct rte_eth_dev *dev)
 598 {
 599         struct rte_vhost_vring vring;
 600         struct vhost_queue *vq;
 601         int count = 0;
 602         int nb_rxq = dev->data->nb_rx_queues;
 603         int i;
 604         int ret;
 605
 606         /* uninstall firstly if we are reconnecting */
 607         if (dev->intr_handle)
 608                 eth_vhost_uninstall_intr(dev);
 609
 610         dev->intr_handle = malloc(sizeof(*dev->intr_handle));
 611         if (!dev->intr_handle) {
 612                 VHOST_LOG(ERR, "Fail to allocate intr_handle\n");
 613                 return -ENOMEM;
 614         }
 615         memset(dev->intr_handle, 0, sizeof(*dev->intr_handle));
 616
 617         dev->intr_handle->efd_counter_size = sizeof(uint64_t);
 618
 619         dev->intr_handle->intr_vec =
 620                 malloc(nb_rxq * sizeof(dev->intr_handle->intr_vec[0]));
 621
 622         if (!dev->intr_handle->intr_vec) {
 623                 VHOST_LOG(ERR,
 624                         "Failed to allocate memory for interrupt vector\n");
 625                 free(dev->intr_handle);
 626                 return -ENOMEM;
 627         }
 628
 629         VHOST_LOG(INFO, "Prepare intr vec\n");
 630         for (i = 0; i < nb_rxq; i++) {
 631                 vq = dev->data->rx_queues[i];
 632                 if (!vq) {
 633                         VHOST_LOG(INFO, "rxq-%d not setup yet, skip!\n", i);
 634                         continue;
 635                 }
 636
 637                 ret = rte_vhost_get_vhost_vring(vq->vid, (i << 1) + 1, &vring);
 638                 if (ret < 0) {
 639                         VHOST_LOG(INFO,
 640                                 "Failed to get rxq-%d's vring, skip!\n", i);
 641                         continue;
 642                 }
 643
 644                 if (vring.kickfd < 0) {
 645                         VHOST_LOG(INFO,
 646                                 "rxq-%d's kickfd is invalid, skip!\n", i);
 647                         continue;
 648                 }
 649                 dev->intr_handle->intr_vec[i] = RTE_INTR_VEC_RXTX_OFFSET + i;
 650                 dev->intr_handle->efds[i] = vring.kickfd;
 651                 count++;
 652                 VHOST_LOG(INFO, "Installed intr vec for rxq-%d\n", i);
 653         }
 654
 655         dev->intr_handle->nb_efd = count;
 656         dev->intr_handle->max_intr = count + 1;
 657         dev->intr_handle->type = RTE_INTR_HANDLE_VDEV;
 658
 659         return 0;
 660 }
 661
 662 static void
 663 update_queuing_status(struct rte_eth_dev *dev)
 664 {
 665         struct pmd_internal *internal = dev->data->dev_private;
 666         struct vhost_queue *vq;
 667         unsigned int i;
 668         int allow_queuing = 1;
 669
 670         if (!dev->data->rx_queues || !dev->data->tx_queues)
 671                 return;
 672
 673         if (rte_atomic32_read(&internal->started) == 0 ||
 674             rte_atomic32_read(&internal->dev_attached) == 0)
 675                 allow_queuing = 0;
 676
 677         /* Wait until rx/tx_pkt_burst stops accessing vhost device */
 678         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 679                 vq = dev->data->rx_queues[i];
 680                 if (vq == NULL)
 681                         continue;
 682                 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
 683                 while (rte_atomic32_read(&vq->while_queuing))
 684                         rte_pause();
 685         }
 686
 687         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 688                 vq = dev->data->tx_queues[i];
 689                 if (vq == NULL)
 690                         continue;
 691                 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
 692                 while (rte_atomic32_read(&vq->while_queuing))
 693                         rte_pause();
 694         }
 695 }
 696
 697 static void
 698 queue_setup(struct rte_eth_dev *eth_dev, struct pmd_internal *internal)
 699 {
 700         struct vhost_queue *vq;
 701         int i;
 702
 703         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 704                 vq = eth_dev->data->rx_queues[i];
 705                 if (!vq)
 706                         continue;
 707                 vq->vid = internal->vid;
 708                 vq->internal = internal;
 709                 vq->port = eth_dev->data->port_id;
 710         }
 711         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 712                 vq = eth_dev->data->tx_queues[i];
 713                 if (!vq)
 714                         continue;
 715                 vq->vid = internal->vid;
 716                 vq->internal = internal;
 717                 vq->port = eth_dev->data->port_id;
 718         }
 719 }
 720
 721 static int
 722 new_device(int vid)
 723 {
 724         struct rte_eth_dev *eth_dev;
 725         struct internal_list *list;
 726         struct pmd_internal *internal;
 727         struct rte_eth_conf *dev_conf;
 728         unsigned i;
 729         char ifname[PATH_MAX];
 730 #ifdef RTE_LIBRTE_VHOST_NUMA
 731         int newnode;
 732 #endif
 733
 734         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 735         list = find_internal_resource(ifname);
 736         if (list == NULL) {
 737                 VHOST_LOG(INFO, "Invalid device name: %s\n", ifname);
 738                 return -1;
 739         }
 740
 741         eth_dev = list->eth_dev;
 742         internal = eth_dev->data->dev_private;
 743         dev_conf = &eth_dev->data->dev_conf;
 744
 745 #ifdef RTE_LIBRTE_VHOST_NUMA
 746         newnode = rte_vhost_get_numa_node(vid);
 747         if (newnode >= 0)
 748                 eth_dev->data->numa_node = newnode;
 749 #endif
 750
 751         internal->vid = vid;
 752         if (rte_atomic32_read(&internal->started) == 1) {
 753                 queue_setup(eth_dev, internal);
 754
 755                 if (dev_conf->intr_conf.rxq) {
 756                         if (eth_vhost_install_intr(eth_dev) < 0) {
 757                                 VHOST_LOG(INFO,
 758                                         "Failed to install interrupt handler.");
 759                                         return -1;
 760                         }
 761                 }
 762         } else {
 763                 VHOST_LOG(INFO, "RX/TX queues not exist yet\n");
 764         }
 765
 766         for (i = 0; i < rte_vhost_get_vring_num(vid); i++)
 767                 rte_vhost_enable_guest_notification(vid, i, 0);
 768
 769         rte_vhost_get_mtu(vid, &eth_dev->data->mtu);
 770
 771         eth_dev->data->dev_link.link_status = ETH_LINK_UP;
 772
 773         rte_atomic32_set(&internal->dev_attached, 1);
 774         update_queuing_status(eth_dev);
 775
 776         VHOST_LOG(INFO, "Vhost device %d created\n", vid);
 777
 778         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 779
 780         return 0;
 781 }
 782
 783 static void
 784 destroy_device(int vid)
 785 {
 786         struct rte_eth_dev *eth_dev;
 787         struct pmd_internal *internal;
 788         struct vhost_queue *vq;
 789         struct internal_list *list;
 790         char ifname[PATH_MAX];
 791         unsigned i;
 792         struct rte_vhost_vring_state *state;
 793
 794         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 795         list = find_internal_resource(ifname);
 796         if (list == NULL) {
 797                 VHOST_LOG(ERR, "Invalid interface name: %s\n", ifname);
 798                 return;
 799         }
 800         eth_dev = list->eth_dev;
 801         internal = eth_dev->data->dev_private;
 802
 803         rte_atomic32_set(&internal->dev_attached, 0);
 804         update_queuing_status(eth_dev);
 805
 806         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
 807
 808         if (eth_dev->data->rx_queues && eth_dev->data->tx_queues) {
 809                 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 810                         vq = eth_dev->data->rx_queues[i];
 811                         if (!vq)
 812                                 continue;
 813                         vq->vid = -1;
 814                 }
 815                 for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 816                         vq = eth_dev->data->tx_queues[i];
 817                         if (!vq)
 818                                 continue;
 819                         vq->vid = -1;
 820                 }
 821         }
 822
 823         state = vring_states[eth_dev->data->port_id];
 824         rte_spinlock_lock(&state->lock);
 825         for (i = 0; i <= state->max_vring; i++) {
 826                 state->cur[i] = false;
 827                 state->seen[i] = false;
 828         }
 829         state->max_vring = 0;
 830         rte_spinlock_unlock(&state->lock);
 831
 832         VHOST_LOG(INFO, "Vhost device %d destroyed\n", vid);
 833         eth_vhost_uninstall_intr(eth_dev);
 834
 835         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 836 }
 837
 838 static int
 839 vring_conf_update(int vid, struct rte_eth_dev *eth_dev, uint16_t vring_id)
 840 {
 841         struct rte_eth_conf *dev_conf = &eth_dev->data->dev_conf;
 842         struct pmd_internal *internal = eth_dev->data->dev_private;
 843         struct rte_vhost_vring vring;
 844         struct vhost_queue *vq;
 845         int rx_idx = vring_id % 2 ? (vring_id - 1) >> 1 : -1;
 846         int ret = 0;
 847
 848         /*
 849          * The vring kickfd may be changed after the new device notification.
 850          * Update it when the vring state is updated.
 851          */
 852         if (rx_idx >= 0 && rx_idx < eth_dev->data->nb_rx_queues &&
 853             rte_atomic32_read(&internal->dev_attached) &&
 854             rte_atomic32_read(&internal->started) &&
 855             dev_conf->intr_conf.rxq) {
 856                 vq = eth_dev->data->rx_queues[rx_idx];
 857                 ret = rte_vhost_get_vhost_vring(vid, vring_id, &vring);
 858                 if (!ret) {
 859                         if (vring.kickfd !=
 860                             eth_dev->intr_handle->efds[rx_idx]) {
 861                                 VHOST_LOG(INFO,
 862                                           "kickfd for rxq-%d was changed.\n",
 863                                           rx_idx);
 864                                 eth_dev->intr_handle->efds[rx_idx] =
 865                                                                    vring.kickfd;
 866                         }
 867
 868                         rte_vhost_enable_guest_notification(vid, vring_id,
 869                                                             vq->intr_en);
 870                         rte_wmb();
 871                 }
 872         }
 873
 874         return ret;
 875 }
 876
 877 static int
 878 vring_state_changed(int vid, uint16_t vring, int enable)
 879 {
 880         struct rte_vhost_vring_state *state;
 881         struct rte_eth_dev *eth_dev;
 882         struct internal_list *list;
 883         char ifname[PATH_MAX];
 884
 885         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 886         list = find_internal_resource(ifname);
 887         if (list == NULL) {
 888                 VHOST_LOG(ERR, "Invalid interface name: %s\n", ifname);
 889                 return -1;
 890         }
 891
 892         eth_dev = list->eth_dev;
 893         /* won't be NULL */
 894         state = vring_states[eth_dev->data->port_id];
 895
 896         if (enable && vring_conf_update(vid, eth_dev, vring))
 897                 VHOST_LOG(INFO, "Failed to update vring-%d configuration.\n",
 898                           (int)vring);
 899
 900         rte_spinlock_lock(&state->lock);
 901         if (state->cur[vring] == enable) {
 902                 rte_spinlock_unlock(&state->lock);
 903                 return 0;
 904         }
 905         state->cur[vring] = enable;
 906         state->max_vring = RTE_MAX(vring, state->max_vring);
 907         rte_spinlock_unlock(&state->lock);
 908
 909         VHOST_LOG(INFO, "vring%u is %s\n",
 910                         vring, enable ? "enabled" : "disabled");
 911
 912         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE, NULL);
 913
 914         return 0;
 915 }
 916
 917 static struct vhost_device_ops vhost_ops = {
 918         .new_device          = new_device,
 919         .destroy_device      = destroy_device,
 920         .vring_state_changed = vring_state_changed,
 921 };
 922
 923 static int
 924 vhost_driver_setup(struct rte_eth_dev *eth_dev)
 925 {
 926         struct pmd_internal *internal = eth_dev->data->dev_private;
 927         struct internal_list *list = NULL;
 928         struct rte_vhost_vring_state *vring_state = NULL;
 929         unsigned int numa_node = eth_dev->device->numa_node;
 930         const char *name = eth_dev->device->name;
 931
 932         /* Don't try to setup again if it has already been done. */
 933         list = find_internal_resource(internal->iface_name);
 934         if (list)
 935                 return 0;
 936
 937         list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
 938         if (list == NULL)
 939                 return -1;
 940
 941         vring_state = rte_zmalloc_socket(name, sizeof(*vring_state),
 942                                          0, numa_node);
 943         if (vring_state == NULL)
 944                 goto free_list;
 945
 946         list->eth_dev = eth_dev;
 947         pthread_mutex_lock(&internal_list_lock);
 948         TAILQ_INSERT_TAIL(&internal_list, list, next);
 949         pthread_mutex_unlock(&internal_list_lock);
 950
 951         rte_spinlock_init(&vring_state->lock);
 952         vring_states[eth_dev->data->port_id] = vring_state;
 953
 954         if (rte_vhost_driver_register(internal->iface_name, internal->flags))
 955                 goto list_remove;
 956
 957         if (internal->disable_flags) {
 958                 if (rte_vhost_driver_disable_features(internal->iface_name,
 959                                                       internal->disable_flags))
 960                         goto drv_unreg;
 961         }
 962
 963         if (rte_vhost_driver_callback_register(internal->iface_name,
 964                                                &vhost_ops) < 0) {
 965                 VHOST_LOG(ERR, "Can't register callbacks\n");
 966                 goto drv_unreg;
 967         }
 968
 969         if (rte_vhost_driver_start(internal->iface_name) < 0) {
 970                 VHOST_LOG(ERR, "Failed to start driver for %s\n",
 971                           internal->iface_name);
 972                 goto drv_unreg;
 973         }
 974
 975         return 0;
 976
 977 drv_unreg:
 978         rte_vhost_driver_unregister(internal->iface_name);
 979 list_remove:
 980         vring_states[eth_dev->data->port_id] = NULL;
 981         pthread_mutex_lock(&internal_list_lock);
 982         TAILQ_REMOVE(&internal_list, list, next);
 983         pthread_mutex_unlock(&internal_list_lock);
 984         rte_free(vring_state);
 985 free_list:
 986         rte_free(list);
 987
 988         return -1;
 989 }
 990
 991 int
 992 rte_eth_vhost_get_queue_event(uint16_t port_id,
 993                 struct rte_eth_vhost_queue_event *event)
 994 {
 995         struct rte_vhost_vring_state *state;
 996         unsigned int i;
 997         int idx;
 998
 999         if (port_id >= RTE_MAX_ETHPORTS) {
1000                 VHOST_LOG(ERR, "Invalid port id\n");
1001                 return -1;
1002         }
1003
1004         state = vring_states[port_id];
1005         if (!state) {
1006                 VHOST_LOG(ERR, "Unused port\n");
1007                 return -1;
1008         }
1009
1010         rte_spinlock_lock(&state->lock);
1011         for (i = 0; i <= state->max_vring; i++) {
1012                 idx = state->index++ % (state->max_vring + 1);
1013
1014                 if (state->cur[idx] != state->seen[idx]) {
1015                         state->seen[idx] = state->cur[idx];
1016                         event->queue_id = idx / 2;
1017                         event->rx = idx & 1;
1018                         event->enable = state->cur[idx];
1019                         rte_spinlock_unlock(&state->lock);
1020                         return 0;
1021                 }
1022         }
1023         rte_spinlock_unlock(&state->lock);
1024
1025         return -1;
1026 }
1027
1028 int
1029 rte_eth_vhost_get_vid_from_port_id(uint16_t port_id)
1030 {
1031         struct internal_list *list;
1032         struct rte_eth_dev *eth_dev;
1033         struct vhost_queue *vq;
1034         int vid = -1;
1035
1036         if (!rte_eth_dev_is_valid_port(port_id))
1037                 return -1;
1038
1039         pthread_mutex_lock(&internal_list_lock);
1040
1041         TAILQ_FOREACH(list, &internal_list, next) {
1042                 eth_dev = list->eth_dev;
1043                 if (eth_dev->data->port_id == port_id) {
1044                         vq = eth_dev->data->rx_queues[0];
1045                         if (vq) {
1046                                 vid = vq->vid;
1047                         }
1048                         break;
1049                 }
1050         }
1051
1052         pthread_mutex_unlock(&internal_list_lock);
1053
1054         return vid;
1055 }
1056
1057 static int
1058 eth_dev_configure(struct rte_eth_dev *dev)
1059 {
1060         struct pmd_internal *internal = dev->data->dev_private;
1061         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1062
1063         /* NOTE: the same process has to operate a vhost interface
1064          * from beginning to end (from eth_dev configure to eth_dev close).
1065          * It is user's responsibility at the moment.
1066          */
1067         if (vhost_driver_setup(dev) < 0)
1068                 return -1;
1069
1070         internal->vlan_strip = !!(rxmode->offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
1071
1072         return 0;
1073 }
1074
1075 static int
1076 eth_dev_start(struct rte_eth_dev *eth_dev)
1077 {
1078         struct pmd_internal *internal = eth_dev->data->dev_private;
1079         struct rte_eth_conf *dev_conf = &eth_dev->data->dev_conf;
1080
1081         queue_setup(eth_dev, internal);
1082
1083         if (rte_atomic32_read(&internal->dev_attached) == 1) {
1084                 if (dev_conf->intr_conf.rxq) {
1085                         if (eth_vhost_install_intr(eth_dev) < 0) {
1086                                 VHOST_LOG(INFO,
1087                                         "Failed to install interrupt handler.");
1088                                         return -1;
1089                         }
1090                 }
1091         }
1092
1093         rte_atomic32_set(&internal->started, 1);
1094         update_queuing_status(eth_dev);
1095
1096         return 0;
1097 }
1098
1099 static void
1100 eth_dev_stop(struct rte_eth_dev *dev)
1101 {
1102         struct pmd_internal *internal = dev->data->dev_private;
1103
1104         rte_atomic32_set(&internal->started, 0);
1105         update_queuing_status(dev);
1106 }
1107
1108 static void
1109 eth_dev_close(struct rte_eth_dev *dev)
1110 {
1111         struct pmd_internal *internal;
1112         struct internal_list *list;
1113         unsigned int i;
1114
1115         internal = dev->data->dev_private;
1116         if (!internal)
1117                 return;
1118
1119         eth_dev_stop(dev);
1120
1121         list = find_internal_resource(internal->iface_name);
1122         if (list) {
1123                 rte_vhost_driver_unregister(internal->iface_name);
1124                 pthread_mutex_lock(&internal_list_lock);
1125                 TAILQ_REMOVE(&internal_list, list, next);
1126                 pthread_mutex_unlock(&internal_list_lock);
1127                 rte_free(list);
1128         }
1129
1130         if (dev->data->rx_queues)
1131                 for (i = 0; i < dev->data->nb_rx_queues; i++)
1132                         rte_free(dev->data->rx_queues[i]);
1133
1134         if (dev->data->tx_queues)
1135                 for (i = 0; i < dev->data->nb_tx_queues; i++)
1136                         rte_free(dev->data->tx_queues[i]);
1137
1138         rte_free(internal->iface_name);
1139         rte_free(internal);
1140
1141         dev->data->dev_private = NULL;
1142
1143         rte_free(vring_states[dev->data->port_id]);
1144         vring_states[dev->data->port_id] = NULL;
1145 }
1146
1147 static int
1148 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1149                    uint16_t nb_rx_desc __rte_unused,
1150                    unsigned int socket_id,
1151                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1152                    struct rte_mempool *mb_pool)
1153 {
1154         struct vhost_queue *vq;
1155
1156         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
1157                         RTE_CACHE_LINE_SIZE, socket_id);
1158         if (vq == NULL) {
1159                 VHOST_LOG(ERR, "Failed to allocate memory for rx queue\n");
1160                 return -ENOMEM;
1161         }
1162
1163         vq->mb_pool = mb_pool;
1164         vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
1165         dev->data->rx_queues[rx_queue_id] = vq;
1166
1167         return 0;
1168 }
1169
1170 static int
1171 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1172                    uint16_t nb_tx_desc __rte_unused,
1173                    unsigned int socket_id,
1174                    const struct rte_eth_txconf *tx_conf __rte_unused)
1175 {
1176         struct vhost_queue *vq;
1177
1178         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
1179                         RTE_CACHE_LINE_SIZE, socket_id);
1180         if (vq == NULL) {
1181                 VHOST_LOG(ERR, "Failed to allocate memory for tx queue\n");
1182                 return -ENOMEM;
1183         }
1184
1185         vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
1186         dev->data->tx_queues[tx_queue_id] = vq;
1187
1188         return 0;
1189 }
1190
1191 static int
1192 eth_dev_info(struct rte_eth_dev *dev,
1193              struct rte_eth_dev_info *dev_info)
1194 {
1195         struct pmd_internal *internal;
1196
1197         internal = dev->data->dev_private;
1198         if (internal == NULL) {
1199                 VHOST_LOG(ERR, "Invalid device specified\n");
1200                 return -ENODEV;
1201         }
1202
1203         dev_info->max_mac_addrs = 1;
1204         dev_info->max_rx_pktlen = (uint32_t)-1;
1205         dev_info->max_rx_queues = internal->max_queues;
1206         dev_info->max_tx_queues = internal->max_queues;
1207         dev_info->min_rx_bufsize = 0;
1208
1209         dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
1210                                 DEV_TX_OFFLOAD_VLAN_INSERT;
1211         dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
1212
1213         return 0;
1214 }
1215
1216 static int
1217 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1218 {
1219         unsigned i;
1220         unsigned long rx_total = 0, tx_total = 0;
1221         unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
1222         struct vhost_queue *vq;
1223
1224         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
1225                         i < dev->data->nb_rx_queues; i++) {
1226                 if (dev->data->rx_queues[i] == NULL)
1227                         continue;
1228                 vq = dev->data->rx_queues[i];
1229                 stats->q_ipackets[i] = vq->stats.pkts;
1230                 rx_total += stats->q_ipackets[i];
1231
1232                 stats->q_ibytes[i] = vq->stats.bytes;
1233                 rx_total_bytes += stats->q_ibytes[i];
1234         }
1235
1236         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
1237                         i < dev->data->nb_tx_queues; i++) {
1238                 if (dev->data->tx_queues[i] == NULL)
1239                         continue;
1240                 vq = dev->data->tx_queues[i];
1241                 stats->q_opackets[i] = vq->stats.pkts;
1242                 tx_total += stats->q_opackets[i];
1243
1244                 stats->q_obytes[i] = vq->stats.bytes;
1245                 tx_total_bytes += stats->q_obytes[i];
1246         }
1247
1248         stats->ipackets = rx_total;
1249         stats->opackets = tx_total;
1250         stats->ibytes = rx_total_bytes;
1251         stats->obytes = tx_total_bytes;
1252
1253         return 0;
1254 }
1255
1256 static int
1257 eth_stats_reset(struct rte_eth_dev *dev)
1258 {
1259         struct vhost_queue *vq;
1260         unsigned i;
1261
1262         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1263                 if (dev->data->rx_queues[i] == NULL)
1264                         continue;
1265                 vq = dev->data->rx_queues[i];
1266                 vq->stats.pkts = 0;
1267                 vq->stats.bytes = 0;
1268         }
1269         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1270                 if (dev->data->tx_queues[i] == NULL)
1271                         continue;
1272                 vq = dev->data->tx_queues[i];
1273                 vq->stats.pkts = 0;
1274                 vq->stats.bytes = 0;
1275                 vq->stats.missed_pkts = 0;
1276         }
1277
1278         return 0;
1279 }
1280
1281 static void
1282 eth_queue_release(void *q)
1283 {
1284         rte_free(q);
1285 }
1286
1287 static int
1288 eth_tx_done_cleanup(void *txq __rte_unused, uint32_t free_cnt __rte_unused)
1289 {
1290         /*
1291          * vHost does not hang onto mbuf. eth_vhost_tx() copies packet data
1292          * and releases mbuf, so nothing to cleanup.
1293          */
1294         return 0;
1295 }
1296
1297 static int
1298 eth_link_update(struct rte_eth_dev *dev __rte_unused,
1299                 int wait_to_complete __rte_unused)
1300 {
1301         return 0;
1302 }
1303
1304 static uint32_t
1305 eth_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1306 {
1307         struct vhost_queue *vq;
1308
1309         vq = dev->data->rx_queues[rx_queue_id];
1310         if (vq == NULL)
1311                 return 0;
1312
1313         return rte_vhost_rx_queue_count(vq->vid, vq->virtqueue_id);
1314 }
1315
1316 static const struct eth_dev_ops ops = {
1317         .dev_start = eth_dev_start,
1318         .dev_stop = eth_dev_stop,
1319         .dev_close = eth_dev_close,
1320         .dev_configure = eth_dev_configure,
1321         .dev_infos_get = eth_dev_info,
1322         .rx_queue_setup = eth_rx_queue_setup,
1323         .tx_queue_setup = eth_tx_queue_setup,
1324         .rx_queue_release = eth_queue_release,
1325         .tx_queue_release = eth_queue_release,
1326         .tx_done_cleanup = eth_tx_done_cleanup,
1327         .rx_queue_count = eth_rx_queue_count,
1328         .link_update = eth_link_update,
1329         .stats_get = eth_stats_get,
1330         .stats_reset = eth_stats_reset,
1331         .xstats_reset = vhost_dev_xstats_reset,
1332         .xstats_get = vhost_dev_xstats_get,
1333         .xstats_get_names = vhost_dev_xstats_get_names,
1334         .rx_queue_intr_enable = eth_rxq_intr_enable,
1335         .rx_queue_intr_disable = eth_rxq_intr_disable,
1336 };
1337
1338 static int
1339 eth_dev_vhost_create(struct rte_vdev_device *dev, char *iface_name,
1340         int16_t queues, const unsigned int numa_node, uint64_t flags,
1341         uint64_t disable_flags)
1342 {
1343         const char *name = rte_vdev_device_name(dev);
1344         struct rte_eth_dev_data *data;
1345         struct pmd_internal *internal = NULL;
1346         struct rte_eth_dev *eth_dev = NULL;
1347         struct rte_ether_addr *eth_addr = NULL;
1348
1349         VHOST_LOG(INFO, "Creating VHOST-USER backend on numa socket %u\n",
1350                 numa_node);
1351
1352         /* reserve an ethdev entry */
1353         eth_dev = rte_eth_vdev_allocate(dev, sizeof(*internal));
1354         if (eth_dev == NULL)
1355                 goto error;
1356         data = eth_dev->data;
1357
1358         eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
1359         if (eth_addr == NULL)
1360                 goto error;
1361         data->mac_addrs = eth_addr;
1362         *eth_addr = base_eth_addr;
1363         eth_addr->addr_bytes[5] = eth_dev->data->port_id;
1364
1365         /* now put it all together
1366          * - store queue data in internal,
1367          * - point eth_dev_data to internals
1368          * - and point eth_dev structure to new eth_dev_data structure
1369          */
1370         internal = eth_dev->data->dev_private;
1371         internal->iface_name = rte_malloc_socket(name, strlen(iface_name) + 1,
1372                                                  0, numa_node);
1373         if (internal->iface_name == NULL)
1374                 goto error;
1375         strcpy(internal->iface_name, iface_name);
1376
1377         data->nb_rx_queues = queues;
1378         data->nb_tx_queues = queues;
1379         internal->max_queues = queues;
1380         internal->vid = -1;
1381         internal->flags = flags;
1382         internal->disable_flags = disable_flags;
1383         data->dev_link = pmd_link;
1384         data->dev_flags = RTE_ETH_DEV_INTR_LSC | RTE_ETH_DEV_CLOSE_REMOVE;
1385         data->promiscuous = 1;
1386         data->all_multicast = 1;
1387
1388         eth_dev->dev_ops = &ops;
1389
1390         /* finally assign rx and tx ops */
1391         eth_dev->rx_pkt_burst = eth_vhost_rx;
1392         eth_dev->tx_pkt_burst = eth_vhost_tx;
1393
1394         rte_eth_dev_probing_finish(eth_dev);
1395         return 0;
1396
1397 error:
1398         if (internal)
1399                 rte_free(internal->iface_name);
1400         rte_eth_dev_release_port(eth_dev);
1401
1402         return -1;
1403 }
1404
1405 static inline int
1406 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
1407 {
1408         const char **iface_name = extra_args;
1409
1410         if (value == NULL)
1411                 return -1;
1412
1413         *iface_name = value;
1414
1415         return 0;
1416 }
1417
1418 static inline int
1419 open_int(const char *key __rte_unused, const char *value, void *extra_args)
1420 {
1421         uint16_t *n = extra_args;
1422
1423         if (value == NULL || extra_args == NULL)
1424                 return -EINVAL;
1425
1426         *n = (uint16_t)strtoul(value, NULL, 0);
1427         if (*n == USHRT_MAX && errno == ERANGE)
1428                 return -1;
1429
1430         return 0;
1431 }
1432
1433 static int
1434 rte_pmd_vhost_probe(struct rte_vdev_device *dev)
1435 {
1436         struct rte_kvargs *kvlist = NULL;
1437         int ret = 0;
1438         char *iface_name;
1439         uint16_t queues;
1440         uint64_t flags = 0;
1441         uint64_t disable_flags = 0;
1442         int client_mode = 0;
1443         int dequeue_zero_copy = 0;
1444         int iommu_support = 0;
1445         int postcopy_support = 0;
1446         int tso = 0;
1447         int linear_buf = 0;
1448         int ext_buf = 0;
1449         struct rte_eth_dev *eth_dev;
1450         const char *name = rte_vdev_device_name(dev);
1451
1452         VHOST_LOG(INFO, "Initializing pmd_vhost for %s\n", name);
1453
1454         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1455                 eth_dev = rte_eth_dev_attach_secondary(name);
1456                 if (!eth_dev) {
1457                         VHOST_LOG(ERR, "Failed to probe %s\n", name);
1458                         return -1;
1459                 }
1460                 eth_dev->rx_pkt_burst = eth_vhost_rx;
1461                 eth_dev->tx_pkt_burst = eth_vhost_tx;
1462                 eth_dev->dev_ops = &ops;
1463                 if (dev->device.numa_node == SOCKET_ID_ANY)
1464                         dev->device.numa_node = rte_socket_id();
1465                 eth_dev->device = &dev->device;
1466                 rte_eth_dev_probing_finish(eth_dev);
1467                 return 0;
1468         }
1469
1470         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
1471         if (kvlist == NULL)
1472                 return -1;
1473
1474         if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
1475                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
1476                                          &open_iface, &iface_name);
1477                 if (ret < 0)
1478                         goto out_free;
1479         } else {
1480                 ret = -1;
1481                 goto out_free;
1482         }
1483
1484         if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
1485                 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
1486                                          &open_int, &queues);
1487                 if (ret < 0 || queues > RTE_MAX_QUEUES_PER_PORT)
1488                         goto out_free;
1489
1490         } else
1491                 queues = 1;
1492
1493         if (rte_kvargs_count(kvlist, ETH_VHOST_CLIENT_ARG) == 1) {
1494                 ret = rte_kvargs_process(kvlist, ETH_VHOST_CLIENT_ARG,
1495                                          &open_int, &client_mode);
1496                 if (ret < 0)
1497                         goto out_free;
1498
1499                 if (client_mode)
1500                         flags |= RTE_VHOST_USER_CLIENT;
1501         }
1502
1503         if (rte_kvargs_count(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY) == 1) {
1504                 ret = rte_kvargs_process(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY,
1505                                          &open_int, &dequeue_zero_copy);
1506                 if (ret < 0)
1507                         goto out_free;
1508
1509                 if (dequeue_zero_copy)
1510                         flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
1511         }
1512
1513         if (rte_kvargs_count(kvlist, ETH_VHOST_IOMMU_SUPPORT) == 1) {
1514                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IOMMU_SUPPORT,
1515                                          &open_int, &iommu_support);
1516                 if (ret < 0)
1517                         goto out_free;
1518
1519                 if (iommu_support)
1520                         flags |= RTE_VHOST_USER_IOMMU_SUPPORT;
1521         }
1522
1523         if (rte_kvargs_count(kvlist, ETH_VHOST_POSTCOPY_SUPPORT) == 1) {
1524                 ret = rte_kvargs_process(kvlist, ETH_VHOST_POSTCOPY_SUPPORT,
1525                                          &open_int, &postcopy_support);
1526                 if (ret < 0)
1527                         goto out_free;
1528
1529                 if (postcopy_support)
1530                         flags |= RTE_VHOST_USER_POSTCOPY_SUPPORT;
1531         }
1532
1533         if (rte_kvargs_count(kvlist, ETH_VHOST_VIRTIO_NET_F_HOST_TSO) == 1) {
1534                 ret = rte_kvargs_process(kvlist,
1535                                 ETH_VHOST_VIRTIO_NET_F_HOST_TSO,
1536                                 &open_int, &tso);
1537                 if (ret < 0)
1538                         goto out_free;
1539
1540                 if (tso == 0) {
1541                         disable_flags |= (1ULL << VIRTIO_NET_F_HOST_TSO4);
1542                         disable_flags |= (1ULL << VIRTIO_NET_F_HOST_TSO6);
1543                 }
1544         }
1545
1546         if (rte_kvargs_count(kvlist, ETH_VHOST_LINEAR_BUF) == 1) {
1547                 ret = rte_kvargs_process(kvlist,
1548                                 ETH_VHOST_LINEAR_BUF,
1549                                 &open_int, &linear_buf);
1550                 if (ret < 0)
1551                         goto out_free;
1552
1553                 if (linear_buf == 1)
1554                         flags |= RTE_VHOST_USER_LINEARBUF_SUPPORT;
1555         }
1556
1557         if (rte_kvargs_count(kvlist, ETH_VHOST_EXT_BUF) == 1) {
1558                 ret = rte_kvargs_process(kvlist,
1559                                 ETH_VHOST_EXT_BUF,
1560                                 &open_int, &ext_buf);
1561                 if (ret < 0)
1562                         goto out_free;
1563
1564                 if (ext_buf == 1)
1565                         flags |= RTE_VHOST_USER_EXTBUF_SUPPORT;
1566         }
1567
1568         if (dev->device.numa_node == SOCKET_ID_ANY)
1569                 dev->device.numa_node = rte_socket_id();
1570
1571         ret = eth_dev_vhost_create(dev, iface_name, queues,
1572                                    dev->device.numa_node, flags, disable_flags);
1573         if (ret == -1)
1574                 VHOST_LOG(ERR, "Failed to create %s\n", name);
1575
1576 out_free:
1577         rte_kvargs_free(kvlist);
1578         return ret;
1579 }
1580
1581 static int
1582 rte_pmd_vhost_remove(struct rte_vdev_device *dev)
1583 {
1584         const char *name;
1585         struct rte_eth_dev *eth_dev = NULL;
1586
1587         name = rte_vdev_device_name(dev);
1588         VHOST_LOG(INFO, "Un-Initializing pmd_vhost for %s\n", name);
1589
1590         /* find an ethdev entry */
1591         eth_dev = rte_eth_dev_allocated(name);
1592         if (eth_dev == NULL)
1593                 return 0;
1594
1595         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1596                 return rte_eth_dev_release_port(eth_dev);
1597
1598         eth_dev_close(eth_dev);
1599
1600         rte_eth_dev_release_port(eth_dev);
1601
1602         return 0;
1603 }
1604
1605 static struct rte_vdev_driver pmd_vhost_drv = {
1606         .probe = rte_pmd_vhost_probe,
1607         .remove = rte_pmd_vhost_remove,
1608 };
1609
1610 RTE_PMD_REGISTER_VDEV(net_vhost, pmd_vhost_drv);
1611 RTE_PMD_REGISTER_ALIAS(net_vhost, eth_vhost);
1612 RTE_PMD_REGISTER_PARAM_STRING(net_vhost,
1613         "iface=<ifc> "
1614         "queues=<int> "
1615         "client=<0|1> "
1616         "dequeue-zero-copy=<0|1> "
1617         "iommu-support=<0|1> "
1618         "postcopy-support=<0|1> "
1619         "tso=<0|1> "
1620         "linear-buffer=<0|1> "
1621         "ext-buffer=<0|1>");