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 enum {VIRTIO_RXQ, VIRTIO_TXQ, VIRTIO_QNUM};
  22
  23 #define ETH_VHOST_IFACE_ARG             "iface"
  24 #define ETH_VHOST_QUEUES_ARG            "queues"
  25 #define ETH_VHOST_CLIENT_ARG            "client"
  26 #define ETH_VHOST_DEQUEUE_ZERO_COPY     "dequeue-zero-copy"
  27 #define ETH_VHOST_IOMMU_SUPPORT         "iommu-support"
  28 #define VHOST_MAX_PKT_BURST 32
  29
  30 static const char *valid_arguments[] = {
  31         ETH_VHOST_IFACE_ARG,
  32         ETH_VHOST_QUEUES_ARG,
  33         ETH_VHOST_CLIENT_ARG,
  34         ETH_VHOST_DEQUEUE_ZERO_COPY,
  35         ETH_VHOST_IOMMU_SUPPORT,
  36         NULL
  37 };
  38
  39 static struct ether_addr base_eth_addr = {
  40         .addr_bytes = {
  41                 0x56 /* V */,
  42                 0x48 /* H */,
  43                 0x4F /* O */,
  44                 0x53 /* S */,
  45                 0x54 /* T */,
  46                 0x00
  47         }
  48 };
  49
  50 enum vhost_xstats_pkts {
  51         VHOST_UNDERSIZE_PKT = 0,
  52         VHOST_64_PKT,
  53         VHOST_65_TO_127_PKT,
  54         VHOST_128_TO_255_PKT,
  55         VHOST_256_TO_511_PKT,
  56         VHOST_512_TO_1023_PKT,
  57         VHOST_1024_TO_1522_PKT,
  58         VHOST_1523_TO_MAX_PKT,
  59         VHOST_BROADCAST_PKT,
  60         VHOST_MULTICAST_PKT,
  61         VHOST_UNICAST_PKT,
  62         VHOST_ERRORS_PKT,
  63         VHOST_ERRORS_FRAGMENTED,
  64         VHOST_ERRORS_JABBER,
  65         VHOST_UNKNOWN_PROTOCOL,
  66         VHOST_XSTATS_MAX,
  67 };
  68
  69 struct vhost_stats {
  70         uint64_t pkts;
  71         uint64_t bytes;
  72         uint64_t missed_pkts;
  73         uint64_t xstats[VHOST_XSTATS_MAX];
  74 };
  75
  76 struct vhost_queue {
  77         int vid;
  78         rte_atomic32_t allow_queuing;
  79         rte_atomic32_t while_queuing;
  80         struct pmd_internal *internal;
  81         struct rte_mempool *mb_pool;
  82         uint16_t port;
  83         uint16_t virtqueue_id;
  84         struct vhost_stats stats;
  85 };
  86
  87 struct pmd_internal {
  88         rte_atomic32_t dev_attached;
  89         char *dev_name;
  90         char *iface_name;
  91         uint16_t max_queues;
  92         uint16_t vid;
  93         rte_atomic32_t started;
  94         uint8_t vlan_strip;
  95 };
  96
  97 struct internal_list {
  98         TAILQ_ENTRY(internal_list) next;
  99         struct rte_eth_dev *eth_dev;
 100 };
 101
 102 TAILQ_HEAD(internal_list_head, internal_list);
 103 static struct internal_list_head internal_list =
 104         TAILQ_HEAD_INITIALIZER(internal_list);
 105
 106 static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
 107
 108 static struct rte_eth_link pmd_link = {
 109                 .link_speed = 10000,
 110                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 111                 .link_status = ETH_LINK_DOWN
 112 };
 113
 114 struct rte_vhost_vring_state {
 115         rte_spinlock_t lock;
 116
 117         bool cur[RTE_MAX_QUEUES_PER_PORT * 2];
 118         bool seen[RTE_MAX_QUEUES_PER_PORT * 2];
 119         unsigned int index;
 120         unsigned int max_vring;
 121 };
 122
 123 static struct rte_vhost_vring_state *vring_states[RTE_MAX_ETHPORTS];
 124
 125 #define VHOST_XSTATS_NAME_SIZE 64
 126
 127 struct vhost_xstats_name_off {
 128         char name[VHOST_XSTATS_NAME_SIZE];
 129         uint64_t offset;
 130 };
 131
 132 /* [rx]_is prepended to the name string here */
 133 static const struct vhost_xstats_name_off vhost_rxport_stat_strings[] = {
 134         {"good_packets",
 135          offsetof(struct vhost_queue, stats.pkts)},
 136         {"total_bytes",
 137          offsetof(struct vhost_queue, stats.bytes)},
 138         {"missed_pkts",
 139          offsetof(struct vhost_queue, stats.missed_pkts)},
 140         {"broadcast_packets",
 141          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
 142         {"multicast_packets",
 143          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
 144         {"unicast_packets",
 145          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
 146          {"undersize_packets",
 147          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
 148         {"size_64_packets",
 149          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
 150         {"size_65_to_127_packets",
 151          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
 152         {"size_128_to_255_packets",
 153          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
 154         {"size_256_to_511_packets",
 155          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
 156         {"size_512_to_1023_packets",
 157          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
 158         {"size_1024_to_1522_packets",
 159          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
 160         {"size_1523_to_max_packets",
 161          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
 162         {"errors_with_bad_CRC",
 163          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
 164         {"fragmented_errors",
 165          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_FRAGMENTED])},
 166         {"jabber_errors",
 167          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_JABBER])},
 168         {"unknown_protos_packets",
 169          offsetof(struct vhost_queue, stats.xstats[VHOST_UNKNOWN_PROTOCOL])},
 170 };
 171
 172 /* [tx]_ is prepended to the name string here */
 173 static const struct vhost_xstats_name_off vhost_txport_stat_strings[] = {
 174         {"good_packets",
 175          offsetof(struct vhost_queue, stats.pkts)},
 176         {"total_bytes",
 177          offsetof(struct vhost_queue, stats.bytes)},
 178         {"missed_pkts",
 179          offsetof(struct vhost_queue, stats.missed_pkts)},
 180         {"broadcast_packets",
 181          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
 182         {"multicast_packets",
 183          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
 184         {"unicast_packets",
 185          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
 186         {"undersize_packets",
 187          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
 188         {"size_64_packets",
 189          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
 190         {"size_65_to_127_packets",
 191          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
 192         {"size_128_to_255_packets",
 193          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
 194         {"size_256_to_511_packets",
 195          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
 196         {"size_512_to_1023_packets",
 197          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
 198         {"size_1024_to_1522_packets",
 199          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
 200         {"size_1523_to_max_packets",
 201          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
 202         {"errors_with_bad_CRC",
 203          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
 204 };
 205
 206 #define VHOST_NB_XSTATS_RXPORT (sizeof(vhost_rxport_stat_strings) / \
 207                                 sizeof(vhost_rxport_stat_strings[0]))
 208
 209 #define VHOST_NB_XSTATS_TXPORT (sizeof(vhost_txport_stat_strings) / \
 210                                 sizeof(vhost_txport_stat_strings[0]))
 211
 212 static void
 213 vhost_dev_xstats_reset(struct rte_eth_dev *dev)
 214 {
 215         struct vhost_queue *vq = NULL;
 216         unsigned int i = 0;
 217
 218         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 219                 vq = dev->data->rx_queues[i];
 220                 if (!vq)
 221                         continue;
 222                 memset(&vq->stats, 0, sizeof(vq->stats));
 223         }
 224         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 225                 vq = dev->data->tx_queues[i];
 226                 if (!vq)
 227                         continue;
 228                 memset(&vq->stats, 0, sizeof(vq->stats));
 229         }
 230 }
 231
 232 static int
 233 vhost_dev_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
 234                            struct rte_eth_xstat_name *xstats_names,
 235                            unsigned int limit __rte_unused)
 236 {
 237         unsigned int t = 0;
 238         int count = 0;
 239         int nstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
 240
 241         if (!xstats_names)
 242                 return nstats;
 243         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
 244                 snprintf(xstats_names[count].name,
 245                          sizeof(xstats_names[count].name),
 246                          "rx_%s", vhost_rxport_stat_strings[t].name);
 247                 count++;
 248         }
 249         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
 250                 snprintf(xstats_names[count].name,
 251                          sizeof(xstats_names[count].name),
 252                          "tx_%s", vhost_txport_stat_strings[t].name);
 253                 count++;
 254         }
 255         return count;
 256 }
 257
 258 static int
 259 vhost_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 260                      unsigned int n)
 261 {
 262         unsigned int i;
 263         unsigned int t;
 264         unsigned int count = 0;
 265         struct vhost_queue *vq = NULL;
 266         unsigned int nxstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
 267
 268         if (n < nxstats)
 269                 return nxstats;
 270
 271         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 272                 vq = dev->data->rx_queues[i];
 273                 if (!vq)
 274                         continue;
 275                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
 276                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
 277                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
 278         }
 279         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 280                 vq = dev->data->tx_queues[i];
 281                 if (!vq)
 282                         continue;
 283                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
 284                                 + vq->stats.missed_pkts
 285                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
 286                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
 287         }
 288         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
 289                 xstats[count].value = 0;
 290                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 291                         vq = dev->data->rx_queues[i];
 292                         if (!vq)
 293                                 continue;
 294                         xstats[count].value +=
 295                                 *(uint64_t *)(((char *)vq)
 296                                 + vhost_rxport_stat_strings[t].offset);
 297                 }
 298                 xstats[count].id = count;
 299                 count++;
 300         }
 301         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
 302                 xstats[count].value = 0;
 303                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
 304                         vq = dev->data->tx_queues[i];
 305                         if (!vq)
 306                                 continue;
 307                         xstats[count].value +=
 308                                 *(uint64_t *)(((char *)vq)
 309                                 + vhost_txport_stat_strings[t].offset);
 310                 }
 311                 xstats[count].id = count;
 312                 count++;
 313         }
 314         return count;
 315 }
 316
 317 static inline void
 318 vhost_count_multicast_broadcast(struct vhost_queue *vq,
 319                                 struct rte_mbuf *mbuf)
 320 {
 321         struct ether_addr *ea = NULL;
 322         struct vhost_stats *pstats = &vq->stats;
 323
 324         ea = rte_pktmbuf_mtod(mbuf, struct ether_addr *);
 325         if (is_multicast_ether_addr(ea)) {
 326                 if (is_broadcast_ether_addr(ea))
 327                         pstats->xstats[VHOST_BROADCAST_PKT]++;
 328                 else
 329                         pstats->xstats[VHOST_MULTICAST_PKT]++;
 330         }
 331 }
 332
 333 static void
 334 vhost_update_packet_xstats(struct vhost_queue *vq,
 335                            struct rte_mbuf **bufs,
 336                            uint16_t count)
 337 {
 338         uint32_t pkt_len = 0;
 339         uint64_t i = 0;
 340         uint64_t index;
 341         struct vhost_stats *pstats = &vq->stats;
 342
 343         for (i = 0; i < count ; i++) {
 344                 pkt_len = bufs[i]->pkt_len;
 345                 if (pkt_len == 64) {
 346                         pstats->xstats[VHOST_64_PKT]++;
 347                 } else if (pkt_len > 64 && pkt_len < 1024) {
 348                         index = (sizeof(pkt_len) * 8)
 349                                 - __builtin_clz(pkt_len) - 5;
 350                         pstats->xstats[index]++;
 351                 } else {
 352                         if (pkt_len < 64)
 353                                 pstats->xstats[VHOST_UNDERSIZE_PKT]++;
 354                         else if (pkt_len <= 1522)
 355                                 pstats->xstats[VHOST_1024_TO_1522_PKT]++;
 356                         else if (pkt_len > 1522)
 357                                 pstats->xstats[VHOST_1523_TO_MAX_PKT]++;
 358                 }
 359                 vhost_count_multicast_broadcast(vq, bufs[i]);
 360         }
 361 }
 362
 363 static uint16_t
 364 eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 365 {
 366         struct vhost_queue *r = q;
 367         uint16_t i, nb_rx = 0;
 368         uint16_t nb_receive = nb_bufs;
 369
 370         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 371                 return 0;
 372
 373         rte_atomic32_set(&r->while_queuing, 1);
 374
 375         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 376                 goto out;
 377
 378         /* Dequeue packets from guest TX queue */
 379         while (nb_receive) {
 380                 uint16_t nb_pkts;
 381                 uint16_t num = (uint16_t)RTE_MIN(nb_receive,
 382                                                  VHOST_MAX_PKT_BURST);
 383
 384                 nb_pkts = rte_vhost_dequeue_burst(r->vid, r->virtqueue_id,
 385                                                   r->mb_pool, &bufs[nb_rx],
 386                                                   num);
 387
 388                 nb_rx += nb_pkts;
 389                 nb_receive -= nb_pkts;
 390                 if (nb_pkts < num)
 391                         break;
 392         }
 393
 394         r->stats.pkts += nb_rx;
 395
 396         for (i = 0; likely(i < nb_rx); i++) {
 397                 bufs[i]->port = r->port;
 398                 bufs[i]->ol_flags = 0;
 399                 bufs[i]->vlan_tci = 0;
 400
 401                 if (r->internal->vlan_strip)
 402                         rte_vlan_strip(bufs[i]);
 403
 404                 r->stats.bytes += bufs[i]->pkt_len;
 405         }
 406
 407         vhost_update_packet_xstats(r, bufs, nb_rx);
 408
 409 out:
 410         rte_atomic32_set(&r->while_queuing, 0);
 411
 412         return nb_rx;
 413 }
 414
 415 static uint16_t
 416 eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 417 {
 418         struct vhost_queue *r = q;
 419         uint16_t i, nb_tx = 0;
 420         uint16_t nb_send = 0;
 421
 422         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 423                 return 0;
 424
 425         rte_atomic32_set(&r->while_queuing, 1);
 426
 427         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 428                 goto out;
 429
 430         for (i = 0; i < nb_bufs; i++) {
 431                 struct rte_mbuf *m = bufs[i];
 432
 433                 /* Do VLAN tag insertion */
 434                 if (m->ol_flags & PKT_TX_VLAN_PKT) {
 435                         int error = rte_vlan_insert(&m);
 436                         if (unlikely(error)) {
 437                                 rte_pktmbuf_free(m);
 438                                 continue;
 439                         }
 440                 }
 441
 442                 bufs[nb_send] = m;
 443                 ++nb_send;
 444         }
 445
 446         /* Enqueue packets to guest RX queue */
 447         while (nb_send) {
 448                 uint16_t nb_pkts;
 449                 uint16_t num = (uint16_t)RTE_MIN(nb_send,
 450                                                  VHOST_MAX_PKT_BURST);
 451
 452                 nb_pkts = rte_vhost_enqueue_burst(r->vid, r->virtqueue_id,
 453                                                   &bufs[nb_tx], num);
 454
 455                 nb_tx += nb_pkts;
 456                 nb_send -= nb_pkts;
 457                 if (nb_pkts < num)
 458                         break;
 459         }
 460
 461         r->stats.pkts += nb_tx;
 462         r->stats.missed_pkts += nb_bufs - nb_tx;
 463
 464         for (i = 0; likely(i < nb_tx); i++)
 465                 r->stats.bytes += bufs[i]->pkt_len;
 466
 467         vhost_update_packet_xstats(r, bufs, nb_tx);
 468
 469         /* According to RFC2863 page42 section ifHCOutMulticastPkts and
 470          * ifHCOutBroadcastPkts, the counters "multicast" and "broadcast"
 471          * are increased when packets are not transmitted successfully.
 472          */
 473         for (i = nb_tx; i < nb_bufs; i++)
 474                 vhost_count_multicast_broadcast(r, bufs[i]);
 475
 476         for (i = 0; likely(i < nb_tx); i++)
 477                 rte_pktmbuf_free(bufs[i]);
 478 out:
 479         rte_atomic32_set(&r->while_queuing, 0);
 480
 481         return nb_tx;
 482 }
 483
 484 static int
 485 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 486 {
 487         struct pmd_internal *internal = dev->data->dev_private;
 488         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
 489
 490         internal->vlan_strip = rxmode->hw_vlan_strip;
 491
 492         if (rxmode->hw_vlan_filter)
 493                 RTE_LOG(WARNING, PMD,
 494                         "vhost(%s): vlan filtering not available\n",
 495                         internal->dev_name);
 496
 497         return 0;
 498 }
 499
 500 static inline struct internal_list *
 501 find_internal_resource(char *ifname)
 502 {
 503         int found = 0;
 504         struct internal_list *list;
 505         struct pmd_internal *internal;
 506
 507         if (ifname == NULL)
 508                 return NULL;
 509
 510         pthread_mutex_lock(&internal_list_lock);
 511
 512         TAILQ_FOREACH(list, &internal_list, next) {
 513                 internal = list->eth_dev->data->dev_private;
 514                 if (!strcmp(internal->iface_name, ifname)) {
 515                         found = 1;
 516                         break;
 517                 }
 518         }
 519
 520         pthread_mutex_unlock(&internal_list_lock);
 521
 522         if (!found)
 523                 return NULL;
 524
 525         return list;
 526 }
 527
 528 static int
 529 eth_rxq_intr_enable(struct rte_eth_dev *dev, uint16_t qid)
 530 {
 531         struct rte_vhost_vring vring;
 532         struct vhost_queue *vq;
 533         int ret = 0;
 534
 535         vq = dev->data->rx_queues[qid];
 536         if (!vq) {
 537                 RTE_LOG(ERR, PMD, "rxq%d is not setup yet\n", qid);
 538                 return -1;
 539         }
 540
 541         ret = rte_vhost_get_vhost_vring(vq->vid, (qid << 1) + 1, &vring);
 542         if (ret < 0) {
 543                 RTE_LOG(ERR, PMD, "Failed to get rxq%d's vring\n", qid);
 544                 return ret;
 545         }
 546         RTE_LOG(INFO, PMD, "Enable interrupt for rxq%d\n", qid);
 547         rte_vhost_enable_guest_notification(vq->vid, (qid << 1) + 1, 1);
 548         rte_wmb();
 549
 550         return ret;
 551 }
 552
 553 static int
 554 eth_rxq_intr_disable(struct rte_eth_dev *dev, uint16_t qid)
 555 {
 556         struct rte_vhost_vring vring;
 557         struct vhost_queue *vq;
 558         int ret = 0;
 559
 560         vq = dev->data->rx_queues[qid];
 561         if (!vq) {
 562                 RTE_LOG(ERR, PMD, "rxq%d is not setup yet\n", qid);
 563                 return -1;
 564         }
 565
 566         ret = rte_vhost_get_vhost_vring(vq->vid, (qid << 1) + 1, &vring);
 567         if (ret < 0) {
 568                 RTE_LOG(ERR, PMD, "Failed to get rxq%d's vring", qid);
 569                 return ret;
 570         }
 571         RTE_LOG(INFO, PMD, "Disable interrupt for rxq%d\n", qid);
 572         rte_vhost_enable_guest_notification(vq->vid, (qid << 1) + 1, 0);
 573         rte_wmb();
 574
 575         return 0;
 576 }
 577
 578 static void
 579 eth_vhost_uninstall_intr(struct rte_eth_dev *dev)
 580 {
 581         struct rte_intr_handle *intr_handle = dev->intr_handle;
 582
 583         if (intr_handle) {
 584                 if (intr_handle->intr_vec)
 585                         free(intr_handle->intr_vec);
 586                 free(intr_handle);
 587         }
 588
 589         dev->intr_handle = NULL;
 590 }
 591
 592 static int
 593 eth_vhost_install_intr(struct rte_eth_dev *dev)
 594 {
 595         struct rte_vhost_vring vring;
 596         struct vhost_queue *vq;
 597         int count = 0;
 598         int nb_rxq = dev->data->nb_rx_queues;
 599         int i;
 600         int ret;
 601
 602         /* uninstall firstly if we are reconnecting */
 603         if (dev->intr_handle)
 604                 eth_vhost_uninstall_intr(dev);
 605
 606         dev->intr_handle = malloc(sizeof(*dev->intr_handle));
 607         if (!dev->intr_handle) {
 608                 RTE_LOG(ERR, PMD, "Fail to allocate intr_handle\n");
 609                 return -ENOMEM;
 610         }
 611         memset(dev->intr_handle, 0, sizeof(*dev->intr_handle));
 612
 613         dev->intr_handle->efd_counter_size = sizeof(uint64_t);
 614
 615         dev->intr_handle->intr_vec =
 616                 malloc(nb_rxq * sizeof(dev->intr_handle->intr_vec[0]));
 617
 618         if (!dev->intr_handle->intr_vec) {
 619                 RTE_LOG(ERR, PMD,
 620                         "Failed to allocate memory for interrupt vector\n");
 621                 free(dev->intr_handle);
 622                 return -ENOMEM;
 623         }
 624
 625         RTE_LOG(INFO, PMD, "Prepare intr vec\n");
 626         for (i = 0; i < nb_rxq; i++) {
 627                 vq = dev->data->rx_queues[i];
 628                 if (!vq) {
 629                         RTE_LOG(INFO, PMD, "rxq-%d not setup yet, skip!\n", i);
 630                         continue;
 631                 }
 632
 633                 ret = rte_vhost_get_vhost_vring(vq->vid, (i << 1) + 1, &vring);
 634                 if (ret < 0) {
 635                         RTE_LOG(INFO, PMD,
 636                                 "Failed to get rxq-%d's vring, skip!\n", i);
 637                         continue;
 638                 }
 639
 640                 if (vring.kickfd < 0) {
 641                         RTE_LOG(INFO, PMD,
 642                                 "rxq-%d's kickfd is invalid, skip!\n", i);
 643                         continue;
 644                 }
 645                 dev->intr_handle->intr_vec[i] = RTE_INTR_VEC_RXTX_OFFSET + i;
 646                 dev->intr_handle->efds[i] = vring.kickfd;
 647                 count++;
 648                 RTE_LOG(INFO, PMD, "Installed intr vec for rxq-%d\n", i);
 649         }
 650
 651         dev->intr_handle->nb_efd = count;
 652         dev->intr_handle->max_intr = count + 1;
 653         dev->intr_handle->type = RTE_INTR_HANDLE_VDEV;
 654
 655         return 0;
 656 }
 657
 658 static void
 659 update_queuing_status(struct rte_eth_dev *dev)
 660 {
 661         struct pmd_internal *internal = dev->data->dev_private;
 662         struct vhost_queue *vq;
 663         unsigned int i;
 664         int allow_queuing = 1;
 665
 666         if (!dev->data->rx_queues || !dev->data->tx_queues)
 667                 return;
 668
 669         if (rte_atomic32_read(&internal->started) == 0 ||
 670             rte_atomic32_read(&internal->dev_attached) == 0)
 671                 allow_queuing = 0;
 672
 673         /* Wait until rx/tx_pkt_burst stops accessing vhost device */
 674         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 675                 vq = dev->data->rx_queues[i];
 676                 if (vq == NULL)
 677                         continue;
 678                 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
 679                 while (rte_atomic32_read(&vq->while_queuing))
 680                         rte_pause();
 681         }
 682
 683         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 684                 vq = dev->data->tx_queues[i];
 685                 if (vq == NULL)
 686                         continue;
 687                 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
 688                 while (rte_atomic32_read(&vq->while_queuing))
 689                         rte_pause();
 690         }
 691 }
 692
 693 static void
 694 queue_setup(struct rte_eth_dev *eth_dev, struct pmd_internal *internal)
 695 {
 696         struct vhost_queue *vq;
 697         int i;
 698
 699         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 700                 vq = eth_dev->data->rx_queues[i];
 701                 if (!vq)
 702                         continue;
 703                 vq->vid = internal->vid;
 704                 vq->internal = internal;
 705                 vq->port = eth_dev->data->port_id;
 706         }
 707         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 708                 vq = eth_dev->data->tx_queues[i];
 709                 if (!vq)
 710                         continue;
 711                 vq->vid = internal->vid;
 712                 vq->internal = internal;
 713                 vq->port = eth_dev->data->port_id;
 714         }
 715 }
 716
 717 static int
 718 new_device(int vid)
 719 {
 720         struct rte_eth_dev *eth_dev;
 721         struct internal_list *list;
 722         struct pmd_internal *internal;
 723         struct rte_eth_conf *dev_conf;
 724         unsigned i;
 725         char ifname[PATH_MAX];
 726 #ifdef RTE_LIBRTE_VHOST_NUMA
 727         int newnode;
 728 #endif
 729
 730         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 731         list = find_internal_resource(ifname);
 732         if (list == NULL) {
 733                 RTE_LOG(INFO, PMD, "Invalid device name: %s\n", ifname);
 734                 return -1;
 735         }
 736
 737         eth_dev = list->eth_dev;
 738         internal = eth_dev->data->dev_private;
 739         dev_conf = &eth_dev->data->dev_conf;
 740
 741 #ifdef RTE_LIBRTE_VHOST_NUMA
 742         newnode = rte_vhost_get_numa_node(vid);
 743         if (newnode >= 0)
 744                 eth_dev->data->numa_node = newnode;
 745 #endif
 746
 747         internal->vid = vid;
 748         if (rte_atomic32_read(&internal->started) == 1) {
 749                 queue_setup(eth_dev, internal);
 750
 751                 if (dev_conf->intr_conf.rxq) {
 752                         if (eth_vhost_install_intr(eth_dev) < 0) {
 753                                 RTE_LOG(INFO, PMD,
 754                                         "Failed to install interrupt handler.");
 755                                         return -1;
 756                         }
 757                 }
 758         } else {
 759                 RTE_LOG(INFO, PMD, "RX/TX queues not exist yet\n");
 760         }
 761
 762         for (i = 0; i < rte_vhost_get_vring_num(vid); i++)
 763                 rte_vhost_enable_guest_notification(vid, i, 0);
 764
 765         rte_vhost_get_mtu(vid, &eth_dev->data->mtu);
 766
 767         eth_dev->data->dev_link.link_status = ETH_LINK_UP;
 768
 769         rte_atomic32_set(&internal->dev_attached, 1);
 770         update_queuing_status(eth_dev);
 771
 772         RTE_LOG(INFO, PMD, "Vhost device %d created\n", vid);
 773
 774         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 775
 776         return 0;
 777 }
 778
 779 static void
 780 destroy_device(int vid)
 781 {
 782         struct rte_eth_dev *eth_dev;
 783         struct pmd_internal *internal;
 784         struct vhost_queue *vq;
 785         struct internal_list *list;
 786         char ifname[PATH_MAX];
 787         unsigned i;
 788         struct rte_vhost_vring_state *state;
 789
 790         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 791         list = find_internal_resource(ifname);
 792         if (list == NULL) {
 793                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
 794                 return;
 795         }
 796         eth_dev = list->eth_dev;
 797         internal = eth_dev->data->dev_private;
 798
 799         rte_atomic32_set(&internal->dev_attached, 0);
 800         update_queuing_status(eth_dev);
 801
 802         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
 803
 804         if (eth_dev->data->rx_queues && eth_dev->data->tx_queues) {
 805                 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 806                         vq = eth_dev->data->rx_queues[i];
 807                         if (!vq)
 808                                 continue;
 809                         vq->vid = -1;
 810                 }
 811                 for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 812                         vq = eth_dev->data->tx_queues[i];
 813                         if (!vq)
 814                                 continue;
 815                         vq->vid = -1;
 816                 }
 817         }
 818
 819         state = vring_states[eth_dev->data->port_id];
 820         rte_spinlock_lock(&state->lock);
 821         for (i = 0; i <= state->max_vring; i++) {
 822                 state->cur[i] = false;
 823                 state->seen[i] = false;
 824         }
 825         state->max_vring = 0;
 826         rte_spinlock_unlock(&state->lock);
 827
 828         RTE_LOG(INFO, PMD, "Vhost device %d destroyed\n", vid);
 829         eth_vhost_uninstall_intr(eth_dev);
 830
 831         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 832 }
 833
 834 static int
 835 vring_state_changed(int vid, uint16_t vring, int enable)
 836 {
 837         struct rte_vhost_vring_state *state;
 838         struct rte_eth_dev *eth_dev;
 839         struct internal_list *list;
 840         char ifname[PATH_MAX];
 841
 842         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 843         list = find_internal_resource(ifname);
 844         if (list == NULL) {
 845                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
 846                 return -1;
 847         }
 848
 849         eth_dev = list->eth_dev;
 850         /* won't be NULL */
 851         state = vring_states[eth_dev->data->port_id];
 852         rte_spinlock_lock(&state->lock);
 853         state->cur[vring] = enable;
 854         state->max_vring = RTE_MAX(vring, state->max_vring);
 855         rte_spinlock_unlock(&state->lock);
 856
 857         RTE_LOG(INFO, PMD, "vring%u is %s\n",
 858                         vring, enable ? "enabled" : "disabled");
 859
 860         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE, NULL);
 861
 862         return 0;
 863 }
 864
 865 static struct vhost_device_ops vhost_ops = {
 866         .new_device          = new_device,
 867         .destroy_device      = destroy_device,
 868         .vring_state_changed = vring_state_changed,
 869 };
 870
 871 int
 872 rte_eth_vhost_get_queue_event(uint16_t port_id,
 873                 struct rte_eth_vhost_queue_event *event)
 874 {
 875         struct rte_vhost_vring_state *state;
 876         unsigned int i;
 877         int idx;
 878
 879         if (port_id >= RTE_MAX_ETHPORTS) {
 880                 RTE_LOG(ERR, PMD, "Invalid port id\n");
 881                 return -1;
 882         }
 883
 884         state = vring_states[port_id];
 885         if (!state) {
 886                 RTE_LOG(ERR, PMD, "Unused port\n");
 887                 return -1;
 888         }
 889
 890         rte_spinlock_lock(&state->lock);
 891         for (i = 0; i <= state->max_vring; i++) {
 892                 idx = state->index++ % (state->max_vring + 1);
 893
 894                 if (state->cur[idx] != state->seen[idx]) {
 895                         state->seen[idx] = state->cur[idx];
 896                         event->queue_id = idx / 2;
 897                         event->rx = idx & 1;
 898                         event->enable = state->cur[idx];
 899                         rte_spinlock_unlock(&state->lock);
 900                         return 0;
 901                 }
 902         }
 903         rte_spinlock_unlock(&state->lock);
 904
 905         return -1;
 906 }
 907
 908 int
 909 rte_eth_vhost_get_vid_from_port_id(uint16_t port_id)
 910 {
 911         struct internal_list *list;
 912         struct rte_eth_dev *eth_dev;
 913         struct vhost_queue *vq;
 914         int vid = -1;
 915
 916         if (!rte_eth_dev_is_valid_port(port_id))
 917                 return -1;
 918
 919         pthread_mutex_lock(&internal_list_lock);
 920
 921         TAILQ_FOREACH(list, &internal_list, next) {
 922                 eth_dev = list->eth_dev;
 923                 if (eth_dev->data->port_id == port_id) {
 924                         vq = eth_dev->data->rx_queues[0];
 925                         if (vq) {
 926                                 vid = vq->vid;
 927                         }
 928                         break;
 929                 }
 930         }
 931
 932         pthread_mutex_unlock(&internal_list_lock);
 933
 934         return vid;
 935 }
 936
 937 static int
 938 eth_dev_start(struct rte_eth_dev *eth_dev)
 939 {
 940         struct pmd_internal *internal = eth_dev->data->dev_private;
 941         struct rte_eth_conf *dev_conf = &eth_dev->data->dev_conf;
 942
 943         queue_setup(eth_dev, internal);
 944
 945         if (rte_atomic32_read(&internal->dev_attached) == 1) {
 946                 if (dev_conf->intr_conf.rxq) {
 947                         if (eth_vhost_install_intr(eth_dev) < 0) {
 948                                 RTE_LOG(INFO, PMD,
 949                                         "Failed to install interrupt handler.");
 950                                         return -1;
 951                         }
 952                 }
 953         }
 954
 955         rte_atomic32_set(&internal->started, 1);
 956         update_queuing_status(eth_dev);
 957
 958         return 0;
 959 }
 960
 961 static void
 962 eth_dev_stop(struct rte_eth_dev *dev)
 963 {
 964         struct pmd_internal *internal = dev->data->dev_private;
 965
 966         rte_atomic32_set(&internal->started, 0);
 967         update_queuing_status(dev);
 968 }
 969
 970 static void
 971 eth_dev_close(struct rte_eth_dev *dev)
 972 {
 973         struct pmd_internal *internal;
 974         struct internal_list *list;
 975         unsigned int i;
 976
 977         internal = dev->data->dev_private;
 978         if (!internal)
 979                 return;
 980
 981         eth_dev_stop(dev);
 982
 983         rte_vhost_driver_unregister(internal->iface_name);
 984
 985         list = find_internal_resource(internal->iface_name);
 986         if (!list)
 987                 return;
 988
 989         pthread_mutex_lock(&internal_list_lock);
 990         TAILQ_REMOVE(&internal_list, list, next);
 991         pthread_mutex_unlock(&internal_list_lock);
 992         rte_free(list);
 993
 994         if (dev->data->rx_queues)
 995                 for (i = 0; i < dev->data->nb_rx_queues; i++)
 996                         rte_free(dev->data->rx_queues[i]);
 997
 998         if (dev->data->tx_queues)
 999                 for (i = 0; i < dev->data->nb_tx_queues; i++)
1000                         rte_free(dev->data->tx_queues[i]);
1001
1002         rte_free(dev->data->mac_addrs);
1003         free(internal->dev_name);
1004         free(internal->iface_name);
1005         rte_free(internal);
1006
1007         dev->data->dev_private = NULL;
1008 }
1009
1010 static int
1011 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1012                    uint16_t nb_rx_desc __rte_unused,
1013                    unsigned int socket_id,
1014                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1015                    struct rte_mempool *mb_pool)
1016 {
1017         struct vhost_queue *vq;
1018
1019         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
1020                         RTE_CACHE_LINE_SIZE, socket_id);
1021         if (vq == NULL) {
1022                 RTE_LOG(ERR, PMD, "Failed to allocate memory for rx queue\n");
1023                 return -ENOMEM;
1024         }
1025
1026         vq->mb_pool = mb_pool;
1027         vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
1028         dev->data->rx_queues[rx_queue_id] = vq;
1029
1030         return 0;
1031 }
1032
1033 static int
1034 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1035                    uint16_t nb_tx_desc __rte_unused,
1036                    unsigned int socket_id,
1037                    const struct rte_eth_txconf *tx_conf __rte_unused)
1038 {
1039         struct vhost_queue *vq;
1040
1041         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
1042                         RTE_CACHE_LINE_SIZE, socket_id);
1043         if (vq == NULL) {
1044                 RTE_LOG(ERR, PMD, "Failed to allocate memory for tx queue\n");
1045                 return -ENOMEM;
1046         }
1047
1048         vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
1049         dev->data->tx_queues[tx_queue_id] = vq;
1050
1051         return 0;
1052 }
1053
1054 static void
1055 eth_dev_info(struct rte_eth_dev *dev,
1056              struct rte_eth_dev_info *dev_info)
1057 {
1058         struct pmd_internal *internal;
1059
1060         internal = dev->data->dev_private;
1061         if (internal == NULL) {
1062                 RTE_LOG(ERR, PMD, "Invalid device specified\n");
1063                 return;
1064         }
1065
1066         dev_info->max_mac_addrs = 1;
1067         dev_info->max_rx_pktlen = (uint32_t)-1;
1068         dev_info->max_rx_queues = internal->max_queues;
1069         dev_info->max_tx_queues = internal->max_queues;
1070         dev_info->min_rx_bufsize = 0;
1071 }
1072
1073 static int
1074 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1075 {
1076         unsigned i;
1077         unsigned long rx_total = 0, tx_total = 0, tx_missed_total = 0;
1078         unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
1079         struct vhost_queue *vq;
1080
1081         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
1082                         i < dev->data->nb_rx_queues; i++) {
1083                 if (dev->data->rx_queues[i] == NULL)
1084                         continue;
1085                 vq = dev->data->rx_queues[i];
1086                 stats->q_ipackets[i] = vq->stats.pkts;
1087                 rx_total += stats->q_ipackets[i];
1088
1089                 stats->q_ibytes[i] = vq->stats.bytes;
1090                 rx_total_bytes += stats->q_ibytes[i];
1091         }
1092
1093         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
1094                         i < dev->data->nb_tx_queues; i++) {
1095                 if (dev->data->tx_queues[i] == NULL)
1096                         continue;
1097                 vq = dev->data->tx_queues[i];
1098                 stats->q_opackets[i] = vq->stats.pkts;
1099                 tx_missed_total += vq->stats.missed_pkts;
1100                 tx_total += stats->q_opackets[i];
1101
1102                 stats->q_obytes[i] = vq->stats.bytes;
1103                 tx_total_bytes += stats->q_obytes[i];
1104         }
1105
1106         stats->ipackets = rx_total;
1107         stats->opackets = tx_total;
1108         stats->oerrors = tx_missed_total;
1109         stats->ibytes = rx_total_bytes;
1110         stats->obytes = tx_total_bytes;
1111
1112         return 0;
1113 }
1114
1115 static void
1116 eth_stats_reset(struct rte_eth_dev *dev)
1117 {
1118         struct vhost_queue *vq;
1119         unsigned i;
1120
1121         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1122                 if (dev->data->rx_queues[i] == NULL)
1123                         continue;
1124                 vq = dev->data->rx_queues[i];
1125                 vq->stats.pkts = 0;
1126                 vq->stats.bytes = 0;
1127         }
1128         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1129                 if (dev->data->tx_queues[i] == NULL)
1130                         continue;
1131                 vq = dev->data->tx_queues[i];
1132                 vq->stats.pkts = 0;
1133                 vq->stats.bytes = 0;
1134                 vq->stats.missed_pkts = 0;
1135         }
1136 }
1137
1138 static void
1139 eth_queue_release(void *q)
1140 {
1141         rte_free(q);
1142 }
1143
1144 static int
1145 eth_tx_done_cleanup(void *txq __rte_unused, uint32_t free_cnt __rte_unused)
1146 {
1147         /*
1148          * vHost does not hang onto mbuf. eth_vhost_tx() copies packet data
1149          * and releases mbuf, so nothing to cleanup.
1150          */
1151         return 0;
1152 }
1153
1154 static int
1155 eth_link_update(struct rte_eth_dev *dev __rte_unused,
1156                 int wait_to_complete __rte_unused)
1157 {
1158         return 0;
1159 }
1160
1161 static uint32_t
1162 eth_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1163 {
1164         struct vhost_queue *vq;
1165
1166         vq = dev->data->rx_queues[rx_queue_id];
1167         if (vq == NULL)
1168                 return 0;
1169
1170         return rte_vhost_rx_queue_count(vq->vid, vq->virtqueue_id);
1171 }
1172
1173 static const struct eth_dev_ops ops = {
1174         .dev_start = eth_dev_start,
1175         .dev_stop = eth_dev_stop,
1176         .dev_close = eth_dev_close,
1177         .dev_configure = eth_dev_configure,
1178         .dev_infos_get = eth_dev_info,
1179         .rx_queue_setup = eth_rx_queue_setup,
1180         .tx_queue_setup = eth_tx_queue_setup,
1181         .rx_queue_release = eth_queue_release,
1182         .tx_queue_release = eth_queue_release,
1183         .tx_done_cleanup = eth_tx_done_cleanup,
1184         .rx_queue_count = eth_rx_queue_count,
1185         .link_update = eth_link_update,
1186         .stats_get = eth_stats_get,
1187         .stats_reset = eth_stats_reset,
1188         .xstats_reset = vhost_dev_xstats_reset,
1189         .xstats_get = vhost_dev_xstats_get,
1190         .xstats_get_names = vhost_dev_xstats_get_names,
1191         .rx_queue_intr_enable = eth_rxq_intr_enable,
1192         .rx_queue_intr_disable = eth_rxq_intr_disable,
1193 };
1194
1195 static struct rte_vdev_driver pmd_vhost_drv;
1196
1197 static int
1198 eth_dev_vhost_create(struct rte_vdev_device *dev, char *iface_name,
1199         int16_t queues, const unsigned int numa_node, uint64_t flags)
1200 {
1201         const char *name = rte_vdev_device_name(dev);
1202         struct rte_eth_dev_data *data;
1203         struct pmd_internal *internal = NULL;
1204         struct rte_eth_dev *eth_dev = NULL;
1205         struct ether_addr *eth_addr = NULL;
1206         struct rte_vhost_vring_state *vring_state = NULL;
1207         struct internal_list *list = NULL;
1208
1209         RTE_LOG(INFO, PMD, "Creating VHOST-USER backend on numa socket %u\n",
1210                 numa_node);
1211
1212         list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
1213         if (list == NULL)
1214                 goto error;
1215
1216         /* reserve an ethdev entry */
1217         eth_dev = rte_eth_vdev_allocate(dev, sizeof(*internal));
1218         if (eth_dev == NULL)
1219                 goto error;
1220
1221         eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
1222         if (eth_addr == NULL)
1223                 goto error;
1224         *eth_addr = base_eth_addr;
1225         eth_addr->addr_bytes[5] = eth_dev->data->port_id;
1226
1227         vring_state = rte_zmalloc_socket(name,
1228                         sizeof(*vring_state), 0, numa_node);
1229         if (vring_state == NULL)
1230                 goto error;
1231
1232         /* now put it all together
1233          * - store queue data in internal,
1234          * - point eth_dev_data to internals
1235          * - and point eth_dev structure to new eth_dev_data structure
1236          */
1237         internal = eth_dev->data->dev_private;
1238         internal->dev_name = strdup(name);
1239         if (internal->dev_name == NULL)
1240                 goto error;
1241         internal->iface_name = strdup(iface_name);
1242         if (internal->iface_name == NULL)
1243                 goto error;
1244
1245         list->eth_dev = eth_dev;
1246         pthread_mutex_lock(&internal_list_lock);
1247         TAILQ_INSERT_TAIL(&internal_list, list, next);
1248         pthread_mutex_unlock(&internal_list_lock);
1249
1250         rte_spinlock_init(&vring_state->lock);
1251         vring_states[eth_dev->data->port_id] = vring_state;
1252
1253         data = eth_dev->data;
1254         data->nb_rx_queues = queues;
1255         data->nb_tx_queues = queues;
1256         internal->max_queues = queues;
1257         data->dev_link = pmd_link;
1258         data->mac_addrs = eth_addr;
1259         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1260
1261         eth_dev->dev_ops = &ops;
1262
1263         /* finally assign rx and tx ops */
1264         eth_dev->rx_pkt_burst = eth_vhost_rx;
1265         eth_dev->tx_pkt_burst = eth_vhost_tx;
1266
1267         if (rte_vhost_driver_register(iface_name, flags))
1268                 goto error;
1269
1270         if (rte_vhost_driver_callback_register(iface_name, &vhost_ops) < 0) {
1271                 RTE_LOG(ERR, PMD, "Can't register callbacks\n");
1272                 goto error;
1273         }
1274
1275         if (rte_vhost_driver_start(iface_name) < 0) {
1276                 RTE_LOG(ERR, PMD, "Failed to start driver for %s\n",
1277                         iface_name);
1278                 goto error;
1279         }
1280
1281         return data->port_id;
1282
1283 error:
1284         if (internal) {
1285                 free(internal->iface_name);
1286                 free(internal->dev_name);
1287         }
1288         rte_free(vring_state);
1289         rte_free(eth_addr);
1290         if (eth_dev)
1291                 rte_eth_dev_release_port(eth_dev);
1292         rte_free(internal);
1293         rte_free(list);
1294
1295         return -1;
1296 }
1297
1298 static inline int
1299 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
1300 {
1301         const char **iface_name = extra_args;
1302
1303         if (value == NULL)
1304                 return -1;
1305
1306         *iface_name = value;
1307
1308         return 0;
1309 }
1310
1311 static inline int
1312 open_int(const char *key __rte_unused, const char *value, void *extra_args)
1313 {
1314         uint16_t *n = extra_args;
1315
1316         if (value == NULL || extra_args == NULL)
1317                 return -EINVAL;
1318
1319         *n = (uint16_t)strtoul(value, NULL, 0);
1320         if (*n == USHRT_MAX && errno == ERANGE)
1321                 return -1;
1322
1323         return 0;
1324 }
1325
1326 static int
1327 rte_pmd_vhost_probe(struct rte_vdev_device *dev)
1328 {
1329         struct rte_kvargs *kvlist = NULL;
1330         int ret = 0;
1331         char *iface_name;
1332         uint16_t queues;
1333         uint64_t flags = 0;
1334         int client_mode = 0;
1335         int dequeue_zero_copy = 0;
1336         int iommu_support = 0;
1337         struct rte_eth_dev *eth_dev;
1338         const char *name = rte_vdev_device_name(dev);
1339
1340         RTE_LOG(INFO, PMD, "Initializing pmd_vhost for %s\n", name);
1341
1342         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1343             strlen(rte_vdev_device_args(dev)) == 0) {
1344                 eth_dev = rte_eth_dev_attach_secondary(name);
1345                 if (!eth_dev) {
1346                         RTE_LOG(ERR, PMD, "Failed to probe %s\n", name);
1347                         return -1;
1348                 }
1349                 /* TODO: request info from primary to set up Rx and Tx */
1350                 eth_dev->dev_ops = &ops;
1351                 return 0;
1352         }
1353
1354         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
1355         if (kvlist == NULL)
1356                 return -1;
1357
1358         if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
1359                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
1360                                          &open_iface, &iface_name);
1361                 if (ret < 0)
1362                         goto out_free;
1363         } else {
1364                 ret = -1;
1365                 goto out_free;
1366         }
1367
1368         if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
1369                 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
1370                                          &open_int, &queues);
1371                 if (ret < 0 || queues > RTE_MAX_QUEUES_PER_PORT)
1372                         goto out_free;
1373
1374         } else
1375                 queues = 1;
1376
1377         if (rte_kvargs_count(kvlist, ETH_VHOST_CLIENT_ARG) == 1) {
1378                 ret = rte_kvargs_process(kvlist, ETH_VHOST_CLIENT_ARG,
1379                                          &open_int, &client_mode);
1380                 if (ret < 0)
1381                         goto out_free;
1382
1383                 if (client_mode)
1384                         flags |= RTE_VHOST_USER_CLIENT;
1385         }
1386
1387         if (rte_kvargs_count(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY) == 1) {
1388                 ret = rte_kvargs_process(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY,
1389                                          &open_int, &dequeue_zero_copy);
1390                 if (ret < 0)
1391                         goto out_free;
1392
1393                 if (dequeue_zero_copy)
1394                         flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
1395         }
1396
1397         if (rte_kvargs_count(kvlist, ETH_VHOST_IOMMU_SUPPORT) == 1) {
1398                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IOMMU_SUPPORT,
1399                                          &open_int, &iommu_support);
1400                 if (ret < 0)
1401                         goto out_free;
1402
1403                 if (iommu_support)
1404                         flags |= RTE_VHOST_USER_IOMMU_SUPPORT;
1405         }
1406
1407         if (dev->device.numa_node == SOCKET_ID_ANY)
1408                 dev->device.numa_node = rte_socket_id();
1409
1410         eth_dev_vhost_create(dev, iface_name, queues, dev->device.numa_node,
1411                 flags);
1412
1413 out_free:
1414         rte_kvargs_free(kvlist);
1415         return ret;
1416 }
1417
1418 static int
1419 rte_pmd_vhost_remove(struct rte_vdev_device *dev)
1420 {
1421         const char *name;
1422         struct rte_eth_dev *eth_dev = NULL;
1423
1424         name = rte_vdev_device_name(dev);
1425         RTE_LOG(INFO, PMD, "Un-Initializing pmd_vhost for %s\n", name);
1426
1427         /* find an ethdev entry */
1428         eth_dev = rte_eth_dev_allocated(name);
1429         if (eth_dev == NULL)
1430                 return -ENODEV;
1431
1432         eth_dev_close(eth_dev);
1433
1434         rte_free(vring_states[eth_dev->data->port_id]);
1435         vring_states[eth_dev->data->port_id] = NULL;
1436
1437         rte_eth_dev_release_port(eth_dev);
1438
1439         return 0;
1440 }
1441
1442 static struct rte_vdev_driver pmd_vhost_drv = {
1443         .probe = rte_pmd_vhost_probe,
1444         .remove = rte_pmd_vhost_remove,
1445 };
1446
1447 RTE_PMD_REGISTER_VDEV(net_vhost, pmd_vhost_drv);
1448 RTE_PMD_REGISTER_ALIAS(net_vhost, eth_vhost);
1449 RTE_PMD_REGISTER_PARAM_STRING(net_vhost,
1450         "iface=<ifc> "
1451         "queues=<int>");