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