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