drivers/net/vhost/rte_eth_vhost.c

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