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 VHOST_MAX_PKT_BURST 32
  34
  35 static const char *valid_arguments[] = {
  36         ETH_VHOST_IFACE_ARG,
  37         ETH_VHOST_QUEUES_ARG,
  38         ETH_VHOST_CLIENT_ARG,
  39         ETH_VHOST_DEQUEUE_ZERO_COPY,
  40         ETH_VHOST_IOMMU_SUPPORT,
  41         NULL
  42 };
  43
  44 static struct ether_addr base_eth_addr = {
  45         .addr_bytes = {
  46                 0x56 /* V */,
  47                 0x48 /* H */,
  48                 0x4F /* O */,
  49                 0x53 /* S */,
  50                 0x54 /* T */,
  51                 0x00
  52         }
  53 };
  54
  55 enum vhost_xstats_pkts {
  56         VHOST_UNDERSIZE_PKT = 0,
  57         VHOST_64_PKT,
  58         VHOST_65_TO_127_PKT,
  59         VHOST_128_TO_255_PKT,
  60         VHOST_256_TO_511_PKT,
  61         VHOST_512_TO_1023_PKT,
  62         VHOST_1024_TO_1522_PKT,
  63         VHOST_1523_TO_MAX_PKT,
  64         VHOST_BROADCAST_PKT,
  65         VHOST_MULTICAST_PKT,
  66         VHOST_UNICAST_PKT,
  67         VHOST_ERRORS_PKT,
  68         VHOST_ERRORS_FRAGMENTED,
  69         VHOST_ERRORS_JABBER,
  70         VHOST_UNKNOWN_PROTOCOL,
  71         VHOST_XSTATS_MAX,
  72 };
  73
  74 struct vhost_stats {
  75         uint64_t pkts;
  76         uint64_t bytes;
  77         uint64_t missed_pkts;
  78         uint64_t xstats[VHOST_XSTATS_MAX];
  79 };
  80
  81 struct vhost_queue {
  82         int vid;
  83         rte_atomic32_t allow_queuing;
  84         rte_atomic32_t while_queuing;
  85         struct pmd_internal *internal;
  86         struct rte_mempool *mb_pool;
  87         uint16_t port;
  88         uint16_t virtqueue_id;
  89         struct vhost_stats stats;
  90 };
  91
  92 struct pmd_internal {
  93         rte_atomic32_t dev_attached;
  94         char *dev_name;
  95         char *iface_name;
  96         uint16_t max_queues;
  97         int vid;
  98         rte_atomic32_t started;
  99         uint8_t vlan_strip;
 100 };
 101
 102 struct internal_list {
 103         TAILQ_ENTRY(internal_list) next;
 104         struct rte_eth_dev *eth_dev;
 105 };
 106
 107 TAILQ_HEAD(internal_list_head, internal_list);
 108 static struct internal_list_head internal_list =
 109         TAILQ_HEAD_INITIALIZER(internal_list);
 110
 111 static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
 112
 113 static struct rte_eth_link pmd_link = {
 114                 .link_speed = 10000,
 115                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 116                 .link_status = ETH_LINK_DOWN
 117 };
 118
 119 struct rte_vhost_vring_state {
 120         rte_spinlock_t lock;
 121
 122         bool cur[RTE_MAX_QUEUES_PER_PORT * 2];
 123         bool seen[RTE_MAX_QUEUES_PER_PORT * 2];
 124         unsigned int index;
 125         unsigned int max_vring;
 126 };
 127
 128 static struct rte_vhost_vring_state *vring_states[RTE_MAX_ETHPORTS];
 129
 130 #define VHOST_XSTATS_NAME_SIZE 64
 131
 132 struct vhost_xstats_name_off {
 133         char name[VHOST_XSTATS_NAME_SIZE];
 134         uint64_t offset;
 135 };
 136
 137 /* [rx]_is prepended to the name string here */
 138 static const struct vhost_xstats_name_off vhost_rxport_stat_strings[] = {
 139         {"good_packets",
 140          offsetof(struct vhost_queue, stats.pkts)},
 141         {"total_bytes",
 142          offsetof(struct vhost_queue, stats.bytes)},
 143         {"missed_pkts",
 144          offsetof(struct vhost_queue, stats.missed_pkts)},
 145         {"broadcast_packets",
 146          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
 147         {"multicast_packets",
 148          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
 149         {"unicast_packets",
 150          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
 151          {"undersize_packets",
 152          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
 153         {"size_64_packets",
 154          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
 155         {"size_65_to_127_packets",
 156          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
 157         {"size_128_to_255_packets",
 158          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
 159         {"size_256_to_511_packets",
 160          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
 161         {"size_512_to_1023_packets",
 162          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
 163         {"size_1024_to_1522_packets",
 164          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
 165         {"size_1523_to_max_packets",
 166          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
 167         {"errors_with_bad_CRC",
 168          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
 169         {"fragmented_errors",
 170          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_FRAGMENTED])},
 171         {"jabber_errors",
 172          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_JABBER])},
 173         {"unknown_protos_packets",
 174          offsetof(struct vhost_queue, stats.xstats[VHOST_UNKNOWN_PROTOCOL])},
 175 };
 176
 177 /* [tx]_ is prepended to the name string here */
 178 static const struct vhost_xstats_name_off vhost_txport_stat_strings[] = {
 179         {"good_packets",
 180          offsetof(struct vhost_queue, stats.pkts)},
 181         {"total_bytes",
 182          offsetof(struct vhost_queue, stats.bytes)},
 183         {"missed_pkts",
 184          offsetof(struct vhost_queue, stats.missed_pkts)},
 185         {"broadcast_packets",
 186          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
 187         {"multicast_packets",
 188          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
 189         {"unicast_packets",
 190          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
 191         {"undersize_packets",
 192          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
 193         {"size_64_packets",
 194          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
 195         {"size_65_to_127_packets",
 196          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
 197         {"size_128_to_255_packets",
 198          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
 199         {"size_256_to_511_packets",
 200          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
 201         {"size_512_to_1023_packets",
 202          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
 203         {"size_1024_to_1522_packets",
 204          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
 205         {"size_1523_to_max_packets",
 206          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
 207         {"errors_with_bad_CRC",
 208          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
 209 };
 210
 211 #define VHOST_NB_XSTATS_RXPORT (sizeof(vhost_rxport_stat_strings) / \
 212                                 sizeof(vhost_rxport_stat_strings[0]))
 213
 214 #define VHOST_NB_XSTATS_TXPORT (sizeof(vhost_txport_stat_strings) / \
 215                                 sizeof(vhost_txport_stat_strings[0]))
 216
 217 static void
 218 vhost_dev_xstats_reset(struct rte_eth_dev *dev)
 219 {
 220         struct vhost_queue *vq = NULL;
 221         unsigned int i = 0;
 222
 223         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 224                 vq = dev->data->rx_queues[i];
 225                 if (!vq)
 226                         continue;
 227                 memset(&vq->stats, 0, sizeof(vq->stats));
 228         }
 229         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 230                 vq = dev->data->tx_queues[i];
 231                 if (!vq)
 232                         continue;
 233                 memset(&vq->stats, 0, sizeof(vq->stats));
 234         }
 235 }
 236
 237 static int
 238 vhost_dev_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
 239                            struct rte_eth_xstat_name *xstats_names,
 240                            unsigned int limit __rte_unused)
 241 {
 242         unsigned int t = 0;
 243         int count = 0;
 244         int nstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
 245
 246         if (!xstats_names)
 247                 return nstats;
 248         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
 249                 snprintf(xstats_names[count].name,
 250                          sizeof(xstats_names[count].name),
 251                          "rx_%s", vhost_rxport_stat_strings[t].name);
 252                 count++;
 253         }
 254         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
 255                 snprintf(xstats_names[count].name,
 256                          sizeof(xstats_names[count].name),
 257                          "tx_%s", vhost_txport_stat_strings[t].name);
 258                 count++;
 259         }
 260         return count;
 261 }
 262
 263 static int
 264 vhost_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 265                      unsigned int n)
 266 {
 267         unsigned int i;
 268         unsigned int t;
 269         unsigned int count = 0;
 270         struct vhost_queue *vq = NULL;
 271         unsigned int nxstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
 272
 273         if (n < nxstats)
 274                 return nxstats;
 275
 276         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 277                 vq = dev->data->rx_queues[i];
 278                 if (!vq)
 279                         continue;
 280                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
 281                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
 282                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
 283         }
 284         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 285                 vq = dev->data->tx_queues[i];
 286                 if (!vq)
 287                         continue;
 288                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
 289                                 + vq->stats.missed_pkts
 290                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
 291                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
 292         }
 293         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
 294                 xstats[count].value = 0;
 295                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 296                         vq = dev->data->rx_queues[i];
 297                         if (!vq)
 298                                 continue;
 299                         xstats[count].value +=
 300                                 *(uint64_t *)(((char *)vq)
 301                                 + vhost_rxport_stat_strings[t].offset);
 302                 }
 303                 xstats[count].id = count;
 304                 count++;
 305         }
 306         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
 307                 xstats[count].value = 0;
 308                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
 309                         vq = dev->data->tx_queues[i];
 310                         if (!vq)
 311                                 continue;
 312                         xstats[count].value +=
 313                                 *(uint64_t *)(((char *)vq)
 314                                 + vhost_txport_stat_strings[t].offset);
 315                 }
 316                 xstats[count].id = count;
 317                 count++;
 318         }
 319         return count;
 320 }
 321
 322 static inline void
 323 vhost_count_multicast_broadcast(struct vhost_queue *vq,
 324                                 struct rte_mbuf *mbuf)
 325 {
 326         struct ether_addr *ea = NULL;
 327         struct vhost_stats *pstats = &vq->stats;
 328
 329         ea = rte_pktmbuf_mtod(mbuf, struct ether_addr *);
 330         if (is_multicast_ether_addr(ea)) {
 331                 if (is_broadcast_ether_addr(ea))
 332                         pstats->xstats[VHOST_BROADCAST_PKT]++;
 333                 else
 334                         pstats->xstats[VHOST_MULTICAST_PKT]++;
 335         }
 336 }
 337
 338 static void
 339 vhost_update_packet_xstats(struct vhost_queue *vq,
 340                            struct rte_mbuf **bufs,
 341                            uint16_t count)
 342 {
 343         uint32_t pkt_len = 0;
 344         uint64_t i = 0;
 345         uint64_t index;
 346         struct vhost_stats *pstats = &vq->stats;
 347
 348         for (i = 0; i < count ; i++) {
 349                 pkt_len = bufs[i]->pkt_len;
 350                 if (pkt_len == 64) {
 351                         pstats->xstats[VHOST_64_PKT]++;
 352                 } else if (pkt_len > 64 && pkt_len < 1024) {
 353                         index = (sizeof(pkt_len) * 8)
 354                                 - __builtin_clz(pkt_len) - 5;
 355                         pstats->xstats[index]++;
 356                 } else {
 357                         if (pkt_len < 64)
 358                                 pstats->xstats[VHOST_UNDERSIZE_PKT]++;
 359                         else if (pkt_len <= 1522)
 360                                 pstats->xstats[VHOST_1024_TO_1522_PKT]++;
 361                         else if (pkt_len > 1522)
 362                                 pstats->xstats[VHOST_1523_TO_MAX_PKT]++;
 363                 }
 364                 vhost_count_multicast_broadcast(vq, bufs[i]);
 365         }
 366 }
 367
 368 static uint16_t
 369 eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 370 {
 371         struct vhost_queue *r = q;
 372         uint16_t i, nb_rx = 0;
 373         uint16_t nb_receive = nb_bufs;
 374
 375         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 376                 return 0;
 377
 378         rte_atomic32_set(&r->while_queuing, 1);
 379
 380         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 381                 goto out;
 382
 383         /* Dequeue packets from guest TX queue */
 384         while (nb_receive) {
 385                 uint16_t nb_pkts;
 386                 uint16_t num = (uint16_t)RTE_MIN(nb_receive,
 387                                                  VHOST_MAX_PKT_BURST);
 388
 389                 nb_pkts = rte_vhost_dequeue_burst(r->vid, r->virtqueue_id,
 390                                                   r->mb_pool, &bufs[nb_rx],
 391                                                   num);
 392
 393                 nb_rx += nb_pkts;
 394                 nb_receive -= nb_pkts;
 395                 if (nb_pkts < num)
 396                         break;
 397         }
 398
 399         r->stats.pkts += nb_rx;
 400
 401         for (i = 0; likely(i < nb_rx); i++) {
 402                 bufs[i]->port = r->port;
 403                 bufs[i]->ol_flags = 0;
 404                 bufs[i]->vlan_tci = 0;
 405
 406                 if (r->internal->vlan_strip)
 407                         rte_vlan_strip(bufs[i]);
 408
 409                 r->stats.bytes += bufs[i]->pkt_len;
 410         }
 411
 412         vhost_update_packet_xstats(r, bufs, nb_rx);
 413
 414 out:
 415         rte_atomic32_set(&r->while_queuing, 0);
 416
 417         return nb_rx;
 418 }
 419
 420 static uint16_t
 421 eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 422 {
 423         struct vhost_queue *r = q;
 424         uint16_t i, nb_tx = 0;
 425         uint16_t nb_send = 0;
 426
 427         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 428                 return 0;
 429
 430         rte_atomic32_set(&r->while_queuing, 1);
 431
 432         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 433                 goto out;
 434
 435         for (i = 0; i < nb_bufs; i++) {
 436                 struct rte_mbuf *m = bufs[i];
 437
 438                 /* Do VLAN tag insertion */
 439                 if (m->ol_flags & PKT_TX_VLAN_PKT) {
 440                         int error = rte_vlan_insert(&m);
 441                         if (unlikely(error)) {
 442                                 rte_pktmbuf_free(m);
 443                                 continue;
 444                         }
 445                 }
 446
 447                 bufs[nb_send] = m;
 448                 ++nb_send;
 449         }
 450
 451         /* Enqueue packets to guest RX queue */
 452         while (nb_send) {
 453                 uint16_t nb_pkts;
 454                 uint16_t num = (uint16_t)RTE_MIN(nb_send,
 455                                                  VHOST_MAX_PKT_BURST);
 456
 457                 nb_pkts = rte_vhost_enqueue_burst(r->vid, r->virtqueue_id,
 458                                                   &bufs[nb_tx], num);
 459
 460                 nb_tx += nb_pkts;
 461                 nb_send -= nb_pkts;
 462                 if (nb_pkts < num)
 463                         break;
 464         }
 465
 466         r->stats.pkts += nb_tx;
 467         r->stats.missed_pkts += nb_bufs - nb_tx;
 468
 469         for (i = 0; likely(i < nb_tx); i++)
 470                 r->stats.bytes += bufs[i]->pkt_len;
 471
 472         vhost_update_packet_xstats(r, bufs, nb_tx);
 473
 474         /* According to RFC2863 page42 section ifHCOutMulticastPkts and
 475          * ifHCOutBroadcastPkts, the counters "multicast" and "broadcast"
 476          * are increased when packets are not transmitted successfully.
 477          */
 478         for (i = nb_tx; i < nb_bufs; i++)
 479                 vhost_count_multicast_broadcast(r, bufs[i]);
 480
 481         for (i = 0; likely(i < nb_tx); i++)
 482                 rte_pktmbuf_free(bufs[i]);
 483 out:
 484         rte_atomic32_set(&r->while_queuing, 0);
 485
 486         return nb_tx;
 487 }
 488
 489 static int
 490 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 491 {
 492         struct pmd_internal *internal = dev->data->dev_private;
 493         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
 494
 495         internal->vlan_strip = rxmode->hw_vlan_strip;
 496
 497         if (rxmode->hw_vlan_filter)
 498                 VHOST_LOG(WARNING,
 499                         "vhost(%s): vlan filtering not available\n",
 500                         internal->dev_name);
 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         state->cur[vring] = enable;
 859         state->max_vring = RTE_MAX(vring, state->max_vring);
 860         rte_spinlock_unlock(&state->lock);
 861
 862         VHOST_LOG(INFO, "vring%u is %s\n",
 863                         vring, enable ? "enabled" : "disabled");
 864
 865         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE, NULL);
 866
 867         return 0;
 868 }
 869
 870 static struct vhost_device_ops vhost_ops = {
 871         .new_device          = new_device,
 872         .destroy_device      = destroy_device,
 873         .vring_state_changed = vring_state_changed,
 874 };
 875
 876 int
 877 rte_eth_vhost_get_queue_event(uint16_t port_id,
 878                 struct rte_eth_vhost_queue_event *event)
 879 {
 880         struct rte_vhost_vring_state *state;
 881         unsigned int i;
 882         int idx;
 883
 884         if (port_id >= RTE_MAX_ETHPORTS) {
 885                 VHOST_LOG(ERR, "Invalid port id\n");
 886                 return -1;
 887         }
 888
 889         state = vring_states[port_id];
 890         if (!state) {
 891                 VHOST_LOG(ERR, "Unused port\n");
 892                 return -1;
 893         }
 894
 895         rte_spinlock_lock(&state->lock);
 896         for (i = 0; i <= state->max_vring; i++) {
 897                 idx = state->index++ % (state->max_vring + 1);
 898
 899                 if (state->cur[idx] != state->seen[idx]) {
 900                         state->seen[idx] = state->cur[idx];
 901                         event->queue_id = idx / 2;
 902                         event->rx = idx & 1;
 903                         event->enable = state->cur[idx];
 904                         rte_spinlock_unlock(&state->lock);
 905                         return 0;
 906                 }
 907         }
 908         rte_spinlock_unlock(&state->lock);
 909
 910         return -1;
 911 }
 912
 913 int
 914 rte_eth_vhost_get_vid_from_port_id(uint16_t port_id)
 915 {
 916         struct internal_list *list;
 917         struct rte_eth_dev *eth_dev;
 918         struct vhost_queue *vq;
 919         int vid = -1;
 920
 921         if (!rte_eth_dev_is_valid_port(port_id))
 922                 return -1;
 923
 924         pthread_mutex_lock(&internal_list_lock);
 925
 926         TAILQ_FOREACH(list, &internal_list, next) {
 927                 eth_dev = list->eth_dev;
 928                 if (eth_dev->data->port_id == port_id) {
 929                         vq = eth_dev->data->rx_queues[0];
 930                         if (vq) {
 931                                 vid = vq->vid;
 932                         }
 933                         break;
 934                 }
 935         }
 936
 937         pthread_mutex_unlock(&internal_list_lock);
 938
 939         return vid;
 940 }
 941
 942 static int
 943 eth_dev_start(struct rte_eth_dev *eth_dev)
 944 {
 945         struct pmd_internal *internal = eth_dev->data->dev_private;
 946         struct rte_eth_conf *dev_conf = &eth_dev->data->dev_conf;
 947
 948         queue_setup(eth_dev, internal);
 949
 950         if (rte_atomic32_read(&internal->dev_attached) == 1) {
 951                 if (dev_conf->intr_conf.rxq) {
 952                         if (eth_vhost_install_intr(eth_dev) < 0) {
 953                                 VHOST_LOG(INFO,
 954                                         "Failed to install interrupt handler.");
 955                                         return -1;
 956                         }
 957                 }
 958         }
 959
 960         rte_atomic32_set(&internal->started, 1);
 961         update_queuing_status(eth_dev);
 962
 963         return 0;
 964 }
 965
 966 static void
 967 eth_dev_stop(struct rte_eth_dev *dev)
 968 {
 969         struct pmd_internal *internal = dev->data->dev_private;
 970
 971         rte_atomic32_set(&internal->started, 0);
 972         update_queuing_status(dev);
 973 }
 974
 975 static void
 976 eth_dev_close(struct rte_eth_dev *dev)
 977 {
 978         struct pmd_internal *internal;
 979         struct internal_list *list;
 980         unsigned int i;
 981
 982         internal = dev->data->dev_private;
 983         if (!internal)
 984                 return;
 985
 986         eth_dev_stop(dev);
 987
 988         rte_vhost_driver_unregister(internal->iface_name);
 989
 990         list = find_internal_resource(internal->iface_name);
 991         if (!list)
 992                 return;
 993
 994         pthread_mutex_lock(&internal_list_lock);
 995         TAILQ_REMOVE(&internal_list, list, next);
 996         pthread_mutex_unlock(&internal_list_lock);
 997         rte_free(list);
 998
 999         if (dev->data->rx_queues)
1000                 for (i = 0; i < dev->data->nb_rx_queues; i++)
1001                         rte_free(dev->data->rx_queues[i]);
1002
1003         if (dev->data->tx_queues)
1004                 for (i = 0; i < dev->data->nb_tx_queues; i++)
1005                         rte_free(dev->data->tx_queues[i]);
1006
1007         rte_free(dev->data->mac_addrs);
1008         free(internal->dev_name);
1009         free(internal->iface_name);
1010         rte_free(internal);
1011
1012         dev->data->dev_private = NULL;
1013 }
1014
1015 static int
1016 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1017                    uint16_t nb_rx_desc __rte_unused,
1018                    unsigned int socket_id,
1019                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1020                    struct rte_mempool *mb_pool)
1021 {
1022         struct vhost_queue *vq;
1023
1024         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
1025                         RTE_CACHE_LINE_SIZE, socket_id);
1026         if (vq == NULL) {
1027                 VHOST_LOG(ERR, "Failed to allocate memory for rx queue\n");
1028                 return -ENOMEM;
1029         }
1030
1031         vq->mb_pool = mb_pool;
1032         vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
1033         dev->data->rx_queues[rx_queue_id] = vq;
1034
1035         return 0;
1036 }
1037
1038 static int
1039 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1040                    uint16_t nb_tx_desc __rte_unused,
1041                    unsigned int socket_id,
1042                    const struct rte_eth_txconf *tx_conf __rte_unused)
1043 {
1044         struct vhost_queue *vq;
1045
1046         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
1047                         RTE_CACHE_LINE_SIZE, socket_id);
1048         if (vq == NULL) {
1049                 VHOST_LOG(ERR, "Failed to allocate memory for tx queue\n");
1050                 return -ENOMEM;
1051         }
1052
1053         vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
1054         dev->data->tx_queues[tx_queue_id] = vq;
1055
1056         return 0;
1057 }
1058
1059 static void
1060 eth_dev_info(struct rte_eth_dev *dev,
1061              struct rte_eth_dev_info *dev_info)
1062 {
1063         struct pmd_internal *internal;
1064
1065         internal = dev->data->dev_private;
1066         if (internal == NULL) {
1067                 VHOST_LOG(ERR, "Invalid device specified\n");
1068                 return;
1069         }
1070
1071         dev_info->max_mac_addrs = 1;
1072         dev_info->max_rx_pktlen = (uint32_t)-1;
1073         dev_info->max_rx_queues = internal->max_queues;
1074         dev_info->max_tx_queues = internal->max_queues;
1075         dev_info->min_rx_bufsize = 0;
1076 }
1077
1078 static int
1079 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1080 {
1081         unsigned i;
1082         unsigned long rx_total = 0, tx_total = 0, tx_missed_total = 0;
1083         unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
1084         struct vhost_queue *vq;
1085
1086         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
1087                         i < dev->data->nb_rx_queues; i++) {
1088                 if (dev->data->rx_queues[i] == NULL)
1089                         continue;
1090                 vq = dev->data->rx_queues[i];
1091                 stats->q_ipackets[i] = vq->stats.pkts;
1092                 rx_total += stats->q_ipackets[i];
1093
1094                 stats->q_ibytes[i] = vq->stats.bytes;
1095                 rx_total_bytes += stats->q_ibytes[i];
1096         }
1097
1098         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
1099                         i < dev->data->nb_tx_queues; i++) {
1100                 if (dev->data->tx_queues[i] == NULL)
1101                         continue;
1102                 vq = dev->data->tx_queues[i];
1103                 stats->q_opackets[i] = vq->stats.pkts;
1104                 tx_missed_total += vq->stats.missed_pkts;
1105                 tx_total += stats->q_opackets[i];
1106
1107                 stats->q_obytes[i] = vq->stats.bytes;
1108                 tx_total_bytes += stats->q_obytes[i];
1109         }
1110
1111         stats->ipackets = rx_total;
1112         stats->opackets = tx_total;
1113         stats->oerrors = tx_missed_total;
1114         stats->ibytes = rx_total_bytes;
1115         stats->obytes = tx_total_bytes;
1116
1117         return 0;
1118 }
1119
1120 static void
1121 eth_stats_reset(struct rte_eth_dev *dev)
1122 {
1123         struct vhost_queue *vq;
1124         unsigned i;
1125
1126         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1127                 if (dev->data->rx_queues[i] == NULL)
1128                         continue;
1129                 vq = dev->data->rx_queues[i];
1130                 vq->stats.pkts = 0;
1131                 vq->stats.bytes = 0;
1132         }
1133         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1134                 if (dev->data->tx_queues[i] == NULL)
1135                         continue;
1136                 vq = dev->data->tx_queues[i];
1137                 vq->stats.pkts = 0;
1138                 vq->stats.bytes = 0;
1139                 vq->stats.missed_pkts = 0;
1140         }
1141 }
1142
1143 static void
1144 eth_queue_release(void *q)
1145 {
1146         rte_free(q);
1147 }
1148
1149 static int
1150 eth_tx_done_cleanup(void *txq __rte_unused, uint32_t free_cnt __rte_unused)
1151 {
1152         /*
1153          * vHost does not hang onto mbuf. eth_vhost_tx() copies packet data
1154          * and releases mbuf, so nothing to cleanup.
1155          */
1156         return 0;
1157 }
1158
1159 static int
1160 eth_link_update(struct rte_eth_dev *dev __rte_unused,
1161                 int wait_to_complete __rte_unused)
1162 {
1163         return 0;
1164 }
1165
1166 static uint32_t
1167 eth_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1168 {
1169         struct vhost_queue *vq;
1170
1171         vq = dev->data->rx_queues[rx_queue_id];
1172         if (vq == NULL)
1173                 return 0;
1174
1175         return rte_vhost_rx_queue_count(vq->vid, vq->virtqueue_id);
1176 }
1177
1178 static const struct eth_dev_ops ops = {
1179         .dev_start = eth_dev_start,
1180         .dev_stop = eth_dev_stop,
1181         .dev_close = eth_dev_close,
1182         .dev_configure = eth_dev_configure,
1183         .dev_infos_get = eth_dev_info,
1184         .rx_queue_setup = eth_rx_queue_setup,
1185         .tx_queue_setup = eth_tx_queue_setup,
1186         .rx_queue_release = eth_queue_release,
1187         .tx_queue_release = eth_queue_release,
1188         .tx_done_cleanup = eth_tx_done_cleanup,
1189         .rx_queue_count = eth_rx_queue_count,
1190         .link_update = eth_link_update,
1191         .stats_get = eth_stats_get,
1192         .stats_reset = eth_stats_reset,
1193         .xstats_reset = vhost_dev_xstats_reset,
1194         .xstats_get = vhost_dev_xstats_get,
1195         .xstats_get_names = vhost_dev_xstats_get_names,
1196         .rx_queue_intr_enable = eth_rxq_intr_enable,
1197         .rx_queue_intr_disable = eth_rxq_intr_disable,
1198 };
1199
1200 static struct rte_vdev_driver pmd_vhost_drv;
1201
1202 static int
1203 eth_dev_vhost_create(struct rte_vdev_device *dev, char *iface_name,
1204         int16_t queues, const unsigned int numa_node, uint64_t flags)
1205 {
1206         const char *name = rte_vdev_device_name(dev);
1207         struct rte_eth_dev_data *data;
1208         struct pmd_internal *internal = NULL;
1209         struct rte_eth_dev *eth_dev = NULL;
1210         struct ether_addr *eth_addr = NULL;
1211         struct rte_vhost_vring_state *vring_state = NULL;
1212         struct internal_list *list = NULL;
1213
1214         VHOST_LOG(INFO, "Creating VHOST-USER backend on numa socket %u\n",
1215                 numa_node);
1216
1217         list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
1218         if (list == NULL)
1219                 goto error;
1220
1221         /* reserve an ethdev entry */
1222         eth_dev = rte_eth_vdev_allocate(dev, sizeof(*internal));
1223         if (eth_dev == NULL)
1224                 goto error;
1225
1226         eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
1227         if (eth_addr == NULL)
1228                 goto error;
1229         *eth_addr = base_eth_addr;
1230         eth_addr->addr_bytes[5] = eth_dev->data->port_id;
1231
1232         vring_state = rte_zmalloc_socket(name,
1233                         sizeof(*vring_state), 0, numa_node);
1234         if (vring_state == NULL)
1235                 goto error;
1236
1237         /* now put it all together
1238          * - store queue data in internal,
1239          * - point eth_dev_data to internals
1240          * - and point eth_dev structure to new eth_dev_data structure
1241          */
1242         internal = eth_dev->data->dev_private;
1243         internal->dev_name = strdup(name);
1244         if (internal->dev_name == NULL)
1245                 goto error;
1246         internal->iface_name = strdup(iface_name);
1247         if (internal->iface_name == NULL)
1248                 goto error;
1249
1250         list->eth_dev = eth_dev;
1251         pthread_mutex_lock(&internal_list_lock);
1252         TAILQ_INSERT_TAIL(&internal_list, list, next);
1253         pthread_mutex_unlock(&internal_list_lock);
1254
1255         rte_spinlock_init(&vring_state->lock);
1256         vring_states[eth_dev->data->port_id] = vring_state;
1257
1258         data = eth_dev->data;
1259         data->nb_rx_queues = queues;
1260         data->nb_tx_queues = queues;
1261         internal->max_queues = queues;
1262         internal->vid = -1;
1263         data->dev_link = pmd_link;
1264         data->mac_addrs = eth_addr;
1265         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1266
1267         eth_dev->dev_ops = &ops;
1268
1269         /* finally assign rx and tx ops */
1270         eth_dev->rx_pkt_burst = eth_vhost_rx;
1271         eth_dev->tx_pkt_burst = eth_vhost_tx;
1272
1273         if (rte_vhost_driver_register(iface_name, flags))
1274                 goto error;
1275
1276         if (rte_vhost_driver_callback_register(iface_name, &vhost_ops) < 0) {
1277                 VHOST_LOG(ERR, "Can't register callbacks\n");
1278                 goto error;
1279         }
1280
1281         if (rte_vhost_driver_start(iface_name) < 0) {
1282                 VHOST_LOG(ERR, "Failed to start driver for %s\n",
1283                         iface_name);
1284                 goto error;
1285         }
1286
1287         rte_eth_dev_probing_finish(eth_dev);
1288         return data->port_id;
1289
1290 error:
1291         if (internal) {
1292                 free(internal->iface_name);
1293                 free(internal->dev_name);
1294         }
1295         rte_free(vring_state);
1296         rte_free(eth_addr);
1297         if (eth_dev)
1298                 rte_eth_dev_release_port(eth_dev);
1299         rte_free(internal);
1300         rte_free(list);
1301
1302         return -1;
1303 }
1304
1305 static inline int
1306 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
1307 {
1308         const char **iface_name = extra_args;
1309
1310         if (value == NULL)
1311                 return -1;
1312
1313         *iface_name = value;
1314
1315         return 0;
1316 }
1317
1318 static inline int
1319 open_int(const char *key __rte_unused, const char *value, void *extra_args)
1320 {
1321         uint16_t *n = extra_args;
1322
1323         if (value == NULL || extra_args == NULL)
1324                 return -EINVAL;
1325
1326         *n = (uint16_t)strtoul(value, NULL, 0);
1327         if (*n == USHRT_MAX && errno == ERANGE)
1328                 return -1;
1329
1330         return 0;
1331 }
1332
1333 static int
1334 rte_pmd_vhost_probe(struct rte_vdev_device *dev)
1335 {
1336         struct rte_kvargs *kvlist = NULL;
1337         int ret = 0;
1338         char *iface_name;
1339         uint16_t queues;
1340         uint64_t flags = 0;
1341         int client_mode = 0;
1342         int dequeue_zero_copy = 0;
1343         int iommu_support = 0;
1344         struct rte_eth_dev *eth_dev;
1345         const char *name = rte_vdev_device_name(dev);
1346
1347         VHOST_LOG(INFO, "Initializing pmd_vhost for %s\n", name);
1348
1349         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1350             strlen(rte_vdev_device_args(dev)) == 0) {
1351                 eth_dev = rte_eth_dev_attach_secondary(name);
1352                 if (!eth_dev) {
1353                         VHOST_LOG(ERR, "Failed to probe %s\n", name);
1354                         return -1;
1355                 }
1356                 /* TODO: request info from primary to set up Rx and Tx */
1357                 eth_dev->dev_ops = &ops;
1358                 rte_eth_dev_probing_finish(eth_dev);
1359                 return 0;
1360         }
1361
1362         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
1363         if (kvlist == NULL)
1364                 return -1;
1365
1366         if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
1367                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
1368                                          &open_iface, &iface_name);
1369                 if (ret < 0)
1370                         goto out_free;
1371         } else {
1372                 ret = -1;
1373                 goto out_free;
1374         }
1375
1376         if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
1377                 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
1378                                          &open_int, &queues);
1379                 if (ret < 0 || queues > RTE_MAX_QUEUES_PER_PORT)
1380                         goto out_free;
1381
1382         } else
1383                 queues = 1;
1384
1385         if (rte_kvargs_count(kvlist, ETH_VHOST_CLIENT_ARG) == 1) {
1386                 ret = rte_kvargs_process(kvlist, ETH_VHOST_CLIENT_ARG,
1387                                          &open_int, &client_mode);
1388                 if (ret < 0)
1389                         goto out_free;
1390
1391                 if (client_mode)
1392                         flags |= RTE_VHOST_USER_CLIENT;
1393         }
1394
1395         if (rte_kvargs_count(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY) == 1) {
1396                 ret = rte_kvargs_process(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY,
1397                                          &open_int, &dequeue_zero_copy);
1398                 if (ret < 0)
1399                         goto out_free;
1400
1401                 if (dequeue_zero_copy)
1402                         flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
1403         }
1404
1405         if (rte_kvargs_count(kvlist, ETH_VHOST_IOMMU_SUPPORT) == 1) {
1406                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IOMMU_SUPPORT,
1407                                          &open_int, &iommu_support);
1408                 if (ret < 0)
1409                         goto out_free;
1410
1411                 if (iommu_support)
1412                         flags |= RTE_VHOST_USER_IOMMU_SUPPORT;
1413         }
1414
1415         if (dev->device.numa_node == SOCKET_ID_ANY)
1416                 dev->device.numa_node = rte_socket_id();
1417
1418         eth_dev_vhost_create(dev, iface_name, queues, dev->device.numa_node,
1419                 flags);
1420
1421 out_free:
1422         rte_kvargs_free(kvlist);
1423         return ret;
1424 }
1425
1426 static int
1427 rte_pmd_vhost_remove(struct rte_vdev_device *dev)
1428 {
1429         const char *name;
1430         struct rte_eth_dev *eth_dev = NULL;
1431
1432         name = rte_vdev_device_name(dev);
1433         VHOST_LOG(INFO, "Un-Initializing pmd_vhost for %s\n", name);
1434
1435         /* find an ethdev entry */
1436         eth_dev = rte_eth_dev_allocated(name);
1437         if (eth_dev == NULL)
1438                 return -ENODEV;
1439
1440         eth_dev_close(eth_dev);
1441
1442         rte_free(vring_states[eth_dev->data->port_id]);
1443         vring_states[eth_dev->data->port_id] = NULL;
1444
1445         rte_eth_dev_release_port(eth_dev);
1446
1447         return 0;
1448 }
1449
1450 static struct rte_vdev_driver pmd_vhost_drv = {
1451         .probe = rte_pmd_vhost_probe,
1452         .remove = rte_pmd_vhost_remove,
1453 };
1454
1455 RTE_PMD_REGISTER_VDEV(net_vhost, pmd_vhost_drv);
1456 RTE_PMD_REGISTER_ALIAS(net_vhost, eth_vhost);
1457 RTE_PMD_REGISTER_PARAM_STRING(net_vhost,
1458         "iface=<ifc> "
1459         "queues=<int>");
1460
1461 RTE_INIT(vhost_init_log);
1462 static void
1463 vhost_init_log(void)
1464 {
1465         vhost_logtype = rte_log_register("pmd.net.vhost");
1466         if (vhost_logtype >= 0)
1467                 rte_log_set_level(vhost_logtype, RTE_LOG_NOTICE);
1468 }