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