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 #include <sys/epoll.h>
   9
  10 #include <rte_mbuf.h>
  11 #include <ethdev_driver.h>
  12 #include <ethdev_vdev.h>
  13 #include <rte_malloc.h>
  14 #include <rte_memcpy.h>
  15 #include <rte_bus_vdev.h>
  16 #include <rte_kvargs.h>
  17 #include <rte_vhost.h>
  18 #include <rte_spinlock.h>
  19
  20 #include "rte_eth_vhost.h"
  21
  22 RTE_LOG_REGISTER_DEFAULT(vhost_logtype, NOTICE);
  23
  24 #define VHOST_LOG(level, ...) \
  25         rte_log(RTE_LOG_ ## level, vhost_logtype, __VA_ARGS__)
  26
  27 enum {VIRTIO_RXQ, VIRTIO_TXQ, VIRTIO_QNUM};
  28
  29 #define ETH_VHOST_IFACE_ARG             "iface"
  30 #define ETH_VHOST_QUEUES_ARG            "queues"
  31 #define ETH_VHOST_CLIENT_ARG            "client"
  32 #define ETH_VHOST_IOMMU_SUPPORT         "iommu-support"
  33 #define ETH_VHOST_POSTCOPY_SUPPORT      "postcopy-support"
  34 #define ETH_VHOST_VIRTIO_NET_F_HOST_TSO "tso"
  35 #define ETH_VHOST_LINEAR_BUF  "linear-buffer"
  36 #define ETH_VHOST_EXT_BUF  "ext-buffer"
  37 #define VHOST_MAX_PKT_BURST 32
  38
  39 static const char *valid_arguments[] = {
  40         ETH_VHOST_IFACE_ARG,
  41         ETH_VHOST_QUEUES_ARG,
  42         ETH_VHOST_CLIENT_ARG,
  43         ETH_VHOST_IOMMU_SUPPORT,
  44         ETH_VHOST_POSTCOPY_SUPPORT,
  45         ETH_VHOST_VIRTIO_NET_F_HOST_TSO,
  46         ETH_VHOST_LINEAR_BUF,
  47         ETH_VHOST_EXT_BUF,
  48         NULL
  49 };
  50
  51 static struct rte_ether_addr base_eth_addr = {
  52         .addr_bytes = {
  53                 0x56 /* V */,
  54                 0x48 /* H */,
  55                 0x4F /* O */,
  56                 0x53 /* S */,
  57                 0x54 /* T */,
  58                 0x00
  59         }
  60 };
  61
  62 enum vhost_xstats_pkts {
  63         VHOST_UNDERSIZE_PKT = 0,
  64         VHOST_64_PKT,
  65         VHOST_65_TO_127_PKT,
  66         VHOST_128_TO_255_PKT,
  67         VHOST_256_TO_511_PKT,
  68         VHOST_512_TO_1023_PKT,
  69         VHOST_1024_TO_1522_PKT,
  70         VHOST_1523_TO_MAX_PKT,
  71         VHOST_BROADCAST_PKT,
  72         VHOST_MULTICAST_PKT,
  73         VHOST_UNICAST_PKT,
  74         VHOST_PKT,
  75         VHOST_BYTE,
  76         VHOST_MISSED_PKT,
  77         VHOST_ERRORS_PKT,
  78         VHOST_ERRORS_FRAGMENTED,
  79         VHOST_ERRORS_JABBER,
  80         VHOST_UNKNOWN_PROTOCOL,
  81         VHOST_XSTATS_MAX,
  82 };
  83
  84 struct vhost_stats {
  85         uint64_t pkts;
  86         uint64_t bytes;
  87         uint64_t missed_pkts;
  88         uint64_t xstats[VHOST_XSTATS_MAX];
  89 };
  90
  91 struct vhost_queue {
  92         int vid;
  93         rte_atomic32_t allow_queuing;
  94         rte_atomic32_t while_queuing;
  95         struct pmd_internal *internal;
  96         struct rte_mempool *mb_pool;
  97         uint16_t port;
  98         uint16_t virtqueue_id;
  99         struct vhost_stats stats;
 100         int intr_enable;
 101         rte_spinlock_t intr_lock;
 102 };
 103
 104 struct pmd_internal {
 105         rte_atomic32_t dev_attached;
 106         char *iface_name;
 107         uint64_t flags;
 108         uint64_t disable_flags;
 109         uint16_t max_queues;
 110         int vid;
 111         rte_atomic32_t started;
 112         uint8_t vlan_strip;
 113 };
 114
 115 struct internal_list {
 116         TAILQ_ENTRY(internal_list) next;
 117         struct rte_eth_dev *eth_dev;
 118 };
 119
 120 TAILQ_HEAD(internal_list_head, internal_list);
 121 static struct internal_list_head internal_list =
 122         TAILQ_HEAD_INITIALIZER(internal_list);
 123
 124 static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
 125
 126 static struct rte_eth_link pmd_link = {
 127                 .link_speed = 10000,
 128                 .link_duplex = RTE_ETH_LINK_FULL_DUPLEX,
 129                 .link_status = RTE_ETH_LINK_DOWN
 130 };
 131
 132 struct rte_vhost_vring_state {
 133         rte_spinlock_t lock;
 134
 135         bool cur[RTE_MAX_QUEUES_PER_PORT * 2];
 136         bool seen[RTE_MAX_QUEUES_PER_PORT * 2];
 137         unsigned int index;
 138         unsigned int max_vring;
 139 };
 140
 141 static struct rte_vhost_vring_state *vring_states[RTE_MAX_ETHPORTS];
 142
 143 #define VHOST_XSTATS_NAME_SIZE 64
 144
 145 struct vhost_xstats_name_off {
 146         char name[VHOST_XSTATS_NAME_SIZE];
 147         uint64_t offset;
 148 };
 149
 150 /* [rx]_is prepended to the name string here */
 151 static const struct vhost_xstats_name_off vhost_rxport_stat_strings[] = {
 152         {"good_packets",
 153          offsetof(struct vhost_queue, stats.xstats[VHOST_PKT])},
 154         {"total_bytes",
 155          offsetof(struct vhost_queue, stats.xstats[VHOST_BYTE])},
 156         {"missed_pkts",
 157          offsetof(struct vhost_queue, stats.xstats[VHOST_MISSED_PKT])},
 158         {"broadcast_packets",
 159          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
 160         {"multicast_packets",
 161          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
 162         {"unicast_packets",
 163          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
 164          {"undersize_packets",
 165          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
 166         {"size_64_packets",
 167          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
 168         {"size_65_to_127_packets",
 169          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
 170         {"size_128_to_255_packets",
 171          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
 172         {"size_256_to_511_packets",
 173          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
 174         {"size_512_to_1023_packets",
 175          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
 176         {"size_1024_to_1522_packets",
 177          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
 178         {"size_1523_to_max_packets",
 179          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
 180         {"errors_with_bad_CRC",
 181          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
 182         {"fragmented_errors",
 183          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_FRAGMENTED])},
 184         {"jabber_errors",
 185          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_JABBER])},
 186         {"unknown_protos_packets",
 187          offsetof(struct vhost_queue, stats.xstats[VHOST_UNKNOWN_PROTOCOL])},
 188 };
 189
 190 /* [tx]_ is prepended to the name string here */
 191 static const struct vhost_xstats_name_off vhost_txport_stat_strings[] = {
 192         {"good_packets",
 193          offsetof(struct vhost_queue, stats.xstats[VHOST_PKT])},
 194         {"total_bytes",
 195          offsetof(struct vhost_queue, stats.xstats[VHOST_BYTE])},
 196         {"missed_pkts",
 197          offsetof(struct vhost_queue, stats.xstats[VHOST_MISSED_PKT])},
 198         {"broadcast_packets",
 199          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
 200         {"multicast_packets",
 201          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
 202         {"unicast_packets",
 203          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
 204         {"undersize_packets",
 205          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
 206         {"size_64_packets",
 207          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
 208         {"size_65_to_127_packets",
 209          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
 210         {"size_128_to_255_packets",
 211          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
 212         {"size_256_to_511_packets",
 213          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
 214         {"size_512_to_1023_packets",
 215          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
 216         {"size_1024_to_1522_packets",
 217          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
 218         {"size_1523_to_max_packets",
 219          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
 220         {"errors_with_bad_CRC",
 221          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
 222 };
 223
 224 #define VHOST_NB_XSTATS_RXPORT (sizeof(vhost_rxport_stat_strings) / \
 225                                 sizeof(vhost_rxport_stat_strings[0]))
 226
 227 #define VHOST_NB_XSTATS_TXPORT (sizeof(vhost_txport_stat_strings) / \
 228                                 sizeof(vhost_txport_stat_strings[0]))
 229
 230 static int
 231 vhost_dev_xstats_reset(struct rte_eth_dev *dev)
 232 {
 233         struct vhost_queue *vq = NULL;
 234         unsigned int i = 0;
 235
 236         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 237                 vq = dev->data->rx_queues[i];
 238                 if (!vq)
 239                         continue;
 240                 memset(&vq->stats, 0, sizeof(vq->stats));
 241         }
 242         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 243                 vq = dev->data->tx_queues[i];
 244                 if (!vq)
 245                         continue;
 246                 memset(&vq->stats, 0, sizeof(vq->stats));
 247         }
 248
 249         return 0;
 250 }
 251
 252 static int
 253 vhost_dev_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
 254                            struct rte_eth_xstat_name *xstats_names,
 255                            unsigned int limit __rte_unused)
 256 {
 257         unsigned int t = 0;
 258         int count = 0;
 259         int nstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
 260
 261         if (!xstats_names)
 262                 return nstats;
 263         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
 264                 snprintf(xstats_names[count].name,
 265                          sizeof(xstats_names[count].name),
 266                          "rx_%s", vhost_rxport_stat_strings[t].name);
 267                 count++;
 268         }
 269         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
 270                 snprintf(xstats_names[count].name,
 271                          sizeof(xstats_names[count].name),
 272                          "tx_%s", vhost_txport_stat_strings[t].name);
 273                 count++;
 274         }
 275         return count;
 276 }
 277
 278 static int
 279 vhost_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 280                      unsigned int n)
 281 {
 282         unsigned int i;
 283         unsigned int t;
 284         unsigned int count = 0;
 285         struct vhost_queue *vq = NULL;
 286         unsigned int nxstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
 287
 288         if (n < nxstats)
 289                 return nxstats;
 290
 291         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
 292                 xstats[count].value = 0;
 293                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 294                         vq = dev->data->rx_queues[i];
 295                         if (!vq)
 296                                 continue;
 297                         xstats[count].value +=
 298                                 *(uint64_t *)(((char *)vq)
 299                                 + vhost_rxport_stat_strings[t].offset);
 300                 }
 301                 xstats[count].id = count;
 302                 count++;
 303         }
 304         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
 305                 xstats[count].value = 0;
 306                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
 307                         vq = dev->data->tx_queues[i];
 308                         if (!vq)
 309                                 continue;
 310                         xstats[count].value +=
 311                                 *(uint64_t *)(((char *)vq)
 312                                 + vhost_txport_stat_strings[t].offset);
 313                 }
 314                 xstats[count].id = count;
 315                 count++;
 316         }
 317         return count;
 318 }
 319
 320 static inline void
 321 vhost_count_xcast_packets(struct vhost_queue *vq,
 322                                 struct rte_mbuf *mbuf)
 323 {
 324         struct rte_ether_addr *ea = NULL;
 325         struct vhost_stats *pstats = &vq->stats;
 326
 327         ea = rte_pktmbuf_mtod(mbuf, struct rte_ether_addr *);
 328         if (rte_is_multicast_ether_addr(ea)) {
 329                 if (rte_is_broadcast_ether_addr(ea))
 330                         pstats->xstats[VHOST_BROADCAST_PKT]++;
 331                 else
 332                         pstats->xstats[VHOST_MULTICAST_PKT]++;
 333         } else {
 334                 pstats->xstats[VHOST_UNICAST_PKT]++;
 335         }
 336 }
 337
 338 static __rte_always_inline void
 339 vhost_update_single_packet_xstats(struct vhost_queue *vq, struct rte_mbuf *buf)
 340 {
 341         uint32_t pkt_len = 0;
 342         uint64_t index;
 343         struct vhost_stats *pstats = &vq->stats;
 344
 345         pstats->xstats[VHOST_PKT]++;
 346         pkt_len = buf->pkt_len;
 347         if (pkt_len == 64) {
 348                 pstats->xstats[VHOST_64_PKT]++;
 349         } else if (pkt_len > 64 && pkt_len < 1024) {
 350                 index = (sizeof(pkt_len) * 8)
 351                         - __builtin_clz(pkt_len) - 5;
 352                 pstats->xstats[index]++;
 353         } else {
 354                 if (pkt_len < 64)
 355                         pstats->xstats[VHOST_UNDERSIZE_PKT]++;
 356                 else if (pkt_len <= 1522)
 357                         pstats->xstats[VHOST_1024_TO_1522_PKT]++;
 358                 else if (pkt_len > 1522)
 359                         pstats->xstats[VHOST_1523_TO_MAX_PKT]++;
 360         }
 361         vhost_count_xcast_packets(vq, buf);
 362 }
 363
 364 static uint16_t
 365 eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 366 {
 367         struct vhost_queue *r = q;
 368         uint16_t i, nb_rx = 0;
 369         uint16_t nb_receive = nb_bufs;
 370
 371         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 372                 return 0;
 373
 374         rte_atomic32_set(&r->while_queuing, 1);
 375
 376         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 377                 goto out;
 378
 379         /* Dequeue packets from guest TX queue */
 380         while (nb_receive) {
 381                 uint16_t nb_pkts;
 382                 uint16_t num = (uint16_t)RTE_MIN(nb_receive,
 383                                                  VHOST_MAX_PKT_BURST);
 384
 385                 nb_pkts = rte_vhost_dequeue_burst(r->vid, r->virtqueue_id,
 386                                                   r->mb_pool, &bufs[nb_rx],
 387                                                   num);
 388
 389                 nb_rx += nb_pkts;
 390                 nb_receive -= nb_pkts;
 391                 if (nb_pkts < num)
 392                         break;
 393         }
 394
 395         r->stats.pkts += nb_rx;
 396
 397         for (i = 0; likely(i < nb_rx); i++) {
 398                 bufs[i]->port = r->port;
 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                 r->stats.xstats[VHOST_BYTE] += bufs[i]->pkt_len;
 406
 407                 vhost_update_single_packet_xstats(r, bufs[i]);
 408         }
 409
 410 out:
 411         rte_atomic32_set(&r->while_queuing, 0);
 412
 413         return nb_rx;
 414 }
 415
 416 static uint16_t
 417 eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 418 {
 419         struct vhost_queue *r = q;
 420         uint16_t i, nb_tx = 0;
 421         uint16_t nb_send = 0;
 422         uint64_t nb_bytes = 0;
 423         uint64_t nb_missed = 0;
 424
 425         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 426                 return 0;
 427
 428         rte_atomic32_set(&r->while_queuing, 1);
 429
 430         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 431                 goto out;
 432
 433         for (i = 0; i < nb_bufs; i++) {
 434                 struct rte_mbuf *m = bufs[i];
 435
 436                 /* Do VLAN tag insertion */
 437                 if (m->ol_flags & RTE_MBUF_F_TX_VLAN) {
 438                         int error = rte_vlan_insert(&m);
 439                         if (unlikely(error)) {
 440                                 rte_pktmbuf_free(m);
 441                                 continue;
 442                         }
 443                 }
 444
 445                 bufs[nb_send] = m;
 446                 ++nb_send;
 447         }
 448
 449         /* Enqueue packets to guest RX queue */
 450         while (nb_send) {
 451                 uint16_t nb_pkts;
 452                 uint16_t num = (uint16_t)RTE_MIN(nb_send,
 453                                                  VHOST_MAX_PKT_BURST);
 454
 455                 nb_pkts = rte_vhost_enqueue_burst(r->vid, r->virtqueue_id,
 456                                                   &bufs[nb_tx], num);
 457
 458                 nb_tx += nb_pkts;
 459                 nb_send -= nb_pkts;
 460                 if (nb_pkts < num)
 461                         break;
 462         }
 463
 464         for (i = 0; likely(i < nb_tx); i++) {
 465                 nb_bytes += bufs[i]->pkt_len;
 466                 vhost_update_single_packet_xstats(r, bufs[i]);
 467         }
 468
 469         nb_missed = nb_bufs - nb_tx;
 470
 471         r->stats.pkts += nb_tx;
 472         r->stats.bytes += nb_bytes;
 473         r->stats.missed_pkts += nb_missed;
 474
 475         r->stats.xstats[VHOST_BYTE] += nb_bytes;
 476         r->stats.xstats[VHOST_MISSED_PKT] += nb_missed;
 477         r->stats.xstats[VHOST_UNICAST_PKT] += nb_missed;
 478
 479         /* According to RFC2863, ifHCOutUcastPkts, ifHCOutMulticastPkts and
 480          * ifHCOutBroadcastPkts counters are increased when packets are not
 481          * transmitted successfully.
 482          */
 483         for (i = nb_tx; i < nb_bufs; i++)
 484                 vhost_count_xcast_packets(r, bufs[i]);
 485
 486         for (i = 0; likely(i < nb_tx); i++)
 487                 rte_pktmbuf_free(bufs[i]);
 488 out:
 489         rte_atomic32_set(&r->while_queuing, 0);
 490
 491         return nb_tx;
 492 }
 493
 494 static inline struct internal_list *
 495 find_internal_resource(char *ifname)
 496 {
 497         int found = 0;
 498         struct internal_list *list;
 499         struct pmd_internal *internal;
 500
 501         if (ifname == NULL)
 502                 return NULL;
 503
 504         pthread_mutex_lock(&internal_list_lock);
 505
 506         TAILQ_FOREACH(list, &internal_list, next) {
 507                 internal = list->eth_dev->data->dev_private;
 508                 if (!strcmp(internal->iface_name, ifname)) {
 509                         found = 1;
 510                         break;
 511                 }
 512         }
 513
 514         pthread_mutex_unlock(&internal_list_lock);
 515
 516         if (!found)
 517                 return NULL;
 518
 519         return list;
 520 }
 521
 522 static int
 523 eth_vhost_update_intr(struct rte_eth_dev *eth_dev, uint16_t rxq_idx)
 524 {
 525         struct rte_intr_handle *handle = eth_dev->intr_handle;
 526         struct rte_epoll_event rev, *elist;
 527         int epfd, ret;
 528
 529         if (handle == NULL)
 530                 return 0;
 531
 532         elist = rte_intr_elist_index_get(handle, rxq_idx);
 533         if (rte_intr_efds_index_get(handle, rxq_idx) == elist->fd)
 534                 return 0;
 535
 536         VHOST_LOG(INFO, "kickfd for rxq-%d was changed, updating handler.\n",
 537                         rxq_idx);
 538
 539         if (elist->fd != -1)
 540                 VHOST_LOG(ERR, "Unexpected previous kickfd value (Got %d, expected -1).\n",
 541                         elist->fd);
 542
 543         /*
 544          * First remove invalid epoll event, and then install
 545          * the new one. May be solved with a proper API in the
 546          * future.
 547          */
 548         epfd = elist->epfd;
 549         rev = *elist;
 550         ret = rte_epoll_ctl(epfd, EPOLL_CTL_DEL, rev.fd,
 551                         elist);
 552         if (ret) {
 553                 VHOST_LOG(ERR, "Delete epoll event failed.\n");
 554                 return ret;
 555         }
 556
 557         rev.fd = rte_intr_efds_index_get(handle, rxq_idx);
 558         if (rte_intr_elist_index_set(handle, rxq_idx, rev))
 559                 return -rte_errno;
 560
 561         elist = rte_intr_elist_index_get(handle, rxq_idx);
 562         ret = rte_epoll_ctl(epfd, EPOLL_CTL_ADD, rev.fd, elist);
 563         if (ret) {
 564                 VHOST_LOG(ERR, "Add epoll event failed.\n");
 565                 return ret;
 566         }
 567
 568         return 0;
 569 }
 570
 571 static int
 572 eth_rxq_intr_enable(struct rte_eth_dev *dev, uint16_t qid)
 573 {
 574         struct rte_vhost_vring vring;
 575         struct vhost_queue *vq;
 576         int old_intr_enable, ret = 0;
 577
 578         vq = dev->data->rx_queues[qid];
 579         if (!vq) {
 580                 VHOST_LOG(ERR, "rxq%d is not setup yet\n", qid);
 581                 return -1;
 582         }
 583
 584         rte_spinlock_lock(&vq->intr_lock);
 585         old_intr_enable = vq->intr_enable;
 586         vq->intr_enable = 1;
 587         ret = eth_vhost_update_intr(dev, qid);
 588         rte_spinlock_unlock(&vq->intr_lock);
 589
 590         if (ret < 0) {
 591                 VHOST_LOG(ERR, "Failed to update rxq%d's intr\n", qid);
 592                 vq->intr_enable = old_intr_enable;
 593                 return ret;
 594         }
 595
 596         ret = rte_vhost_get_vhost_vring(vq->vid, (qid << 1) + 1, &vring);
 597         if (ret < 0) {
 598                 VHOST_LOG(ERR, "Failed to get rxq%d's vring\n", qid);
 599                 return ret;
 600         }
 601         VHOST_LOG(INFO, "Enable interrupt for rxq%d\n", qid);
 602         rte_vhost_enable_guest_notification(vq->vid, (qid << 1) + 1, 1);
 603         rte_wmb();
 604
 605         return ret;
 606 }
 607
 608 static int
 609 eth_rxq_intr_disable(struct rte_eth_dev *dev, uint16_t qid)
 610 {
 611         struct rte_vhost_vring vring;
 612         struct vhost_queue *vq;
 613         int ret = 0;
 614
 615         vq = dev->data->rx_queues[qid];
 616         if (!vq) {
 617                 VHOST_LOG(ERR, "rxq%d is not setup yet\n", qid);
 618                 return -1;
 619         }
 620
 621         ret = rte_vhost_get_vhost_vring(vq->vid, (qid << 1) + 1, &vring);
 622         if (ret < 0) {
 623                 VHOST_LOG(ERR, "Failed to get rxq%d's vring", qid);
 624                 return ret;
 625         }
 626         VHOST_LOG(INFO, "Disable interrupt for rxq%d\n", qid);
 627         rte_vhost_enable_guest_notification(vq->vid, (qid << 1) + 1, 0);
 628         rte_wmb();
 629
 630         vq->intr_enable = 0;
 631
 632         return 0;
 633 }
 634
 635 static void
 636 eth_vhost_uninstall_intr(struct rte_eth_dev *dev)
 637 {
 638         struct rte_intr_handle *intr_handle = dev->intr_handle;
 639
 640         if (intr_handle != NULL) {
 641                 rte_intr_vec_list_free(intr_handle);
 642                 rte_intr_instance_free(intr_handle);
 643         }
 644         dev->intr_handle = NULL;
 645 }
 646
 647 static int
 648 eth_vhost_install_intr(struct rte_eth_dev *dev)
 649 {
 650         struct rte_vhost_vring vring;
 651         struct vhost_queue *vq;
 652         int nb_rxq = dev->data->nb_rx_queues;
 653         int i;
 654         int ret;
 655
 656         /* uninstall firstly if we are reconnecting */
 657         if (dev->intr_handle != NULL)
 658                 eth_vhost_uninstall_intr(dev);
 659
 660         dev->intr_handle = rte_intr_instance_alloc(RTE_INTR_INSTANCE_F_PRIVATE);
 661         if (dev->intr_handle == NULL) {
 662                 VHOST_LOG(ERR, "Fail to allocate intr_handle\n");
 663                 return -ENOMEM;
 664         }
 665         if (rte_intr_efd_counter_size_set(dev->intr_handle, sizeof(uint64_t)))
 666                 return -rte_errno;
 667
 668         if (rte_intr_vec_list_alloc(dev->intr_handle, NULL, nb_rxq)) {
 669                 VHOST_LOG(ERR,
 670                         "Failed to allocate memory for interrupt vector\n");
 671                 rte_intr_instance_free(dev->intr_handle);
 672                 return -ENOMEM;
 673         }
 674
 675
 676         VHOST_LOG(INFO, "Prepare intr vec\n");
 677         for (i = 0; i < nb_rxq; i++) {
 678                 if (rte_intr_vec_list_index_set(dev->intr_handle, i, RTE_INTR_VEC_RXTX_OFFSET + i))
 679                         return -rte_errno;
 680                 if (rte_intr_efds_index_set(dev->intr_handle, i, -1))
 681                         return -rte_errno;
 682                 vq = dev->data->rx_queues[i];
 683                 if (!vq) {
 684                         VHOST_LOG(INFO, "rxq-%d not setup yet, skip!\n", i);
 685                         continue;
 686                 }
 687
 688                 ret = rte_vhost_get_vhost_vring(vq->vid, (i << 1) + 1, &vring);
 689                 if (ret < 0) {
 690                         VHOST_LOG(INFO,
 691                                 "Failed to get rxq-%d's vring, skip!\n", i);
 692                         continue;
 693                 }
 694
 695                 if (vring.kickfd < 0) {
 696                         VHOST_LOG(INFO,
 697                                 "rxq-%d's kickfd is invalid, skip!\n", i);
 698                         continue;
 699                 }
 700
 701                 if (rte_intr_efds_index_set(dev->intr_handle, i, vring.kickfd))
 702                         continue;
 703                 VHOST_LOG(INFO, "Installed intr vec for rxq-%d\n", i);
 704         }
 705
 706         if (rte_intr_nb_efd_set(dev->intr_handle, nb_rxq))
 707                 return -rte_errno;
 708
 709         if (rte_intr_max_intr_set(dev->intr_handle, nb_rxq + 1))
 710                 return -rte_errno;
 711
 712         if (rte_intr_type_set(dev->intr_handle, RTE_INTR_HANDLE_VDEV))
 713                 return -rte_errno;
 714
 715         return 0;
 716 }
 717
 718 static void
 719 update_queuing_status(struct rte_eth_dev *dev)
 720 {
 721         struct pmd_internal *internal = dev->data->dev_private;
 722         struct vhost_queue *vq;
 723         struct rte_vhost_vring_state *state;
 724         unsigned int i;
 725         int allow_queuing = 1;
 726
 727         if (!dev->data->rx_queues || !dev->data->tx_queues)
 728                 return;
 729
 730         if (rte_atomic32_read(&internal->started) == 0 ||
 731             rte_atomic32_read(&internal->dev_attached) == 0)
 732                 allow_queuing = 0;
 733
 734         state = vring_states[dev->data->port_id];
 735
 736         /* Wait until rx/tx_pkt_burst stops accessing vhost device */
 737         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 738                 vq = dev->data->rx_queues[i];
 739                 if (vq == NULL)
 740                         continue;
 741                 if (allow_queuing && state->cur[vq->virtqueue_id])
 742                         rte_atomic32_set(&vq->allow_queuing, 1);
 743                 else
 744                         rte_atomic32_set(&vq->allow_queuing, 0);
 745                 while (rte_atomic32_read(&vq->while_queuing))
 746                         rte_pause();
 747         }
 748
 749         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 750                 vq = dev->data->tx_queues[i];
 751                 if (vq == NULL)
 752                         continue;
 753                 if (allow_queuing && state->cur[vq->virtqueue_id])
 754                         rte_atomic32_set(&vq->allow_queuing, 1);
 755                 else
 756                         rte_atomic32_set(&vq->allow_queuing, 0);
 757                 while (rte_atomic32_read(&vq->while_queuing))
 758                         rte_pause();
 759         }
 760 }
 761
 762 static void
 763 queue_setup(struct rte_eth_dev *eth_dev, struct pmd_internal *internal)
 764 {
 765         struct vhost_queue *vq;
 766         int i;
 767
 768         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 769                 vq = eth_dev->data->rx_queues[i];
 770                 if (!vq)
 771                         continue;
 772                 vq->vid = internal->vid;
 773                 vq->internal = internal;
 774                 vq->port = eth_dev->data->port_id;
 775         }
 776         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 777                 vq = eth_dev->data->tx_queues[i];
 778                 if (!vq)
 779                         continue;
 780                 vq->vid = internal->vid;
 781                 vq->internal = internal;
 782                 vq->port = eth_dev->data->port_id;
 783         }
 784 }
 785
 786 static int
 787 new_device(int vid)
 788 {
 789         struct rte_eth_dev *eth_dev;
 790         struct internal_list *list;
 791         struct pmd_internal *internal;
 792         struct rte_eth_conf *dev_conf;
 793         unsigned i;
 794         char ifname[PATH_MAX];
 795 #ifdef RTE_LIBRTE_VHOST_NUMA
 796         int newnode;
 797 #endif
 798
 799         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 800         list = find_internal_resource(ifname);
 801         if (list == NULL) {
 802                 VHOST_LOG(INFO, "Invalid device name: %s\n", ifname);
 803                 return -1;
 804         }
 805
 806         eth_dev = list->eth_dev;
 807         internal = eth_dev->data->dev_private;
 808         dev_conf = &eth_dev->data->dev_conf;
 809
 810 #ifdef RTE_LIBRTE_VHOST_NUMA
 811         newnode = rte_vhost_get_numa_node(vid);
 812         if (newnode >= 0)
 813                 eth_dev->data->numa_node = newnode;
 814 #endif
 815
 816         internal->vid = vid;
 817         if (rte_atomic32_read(&internal->started) == 1) {
 818                 queue_setup(eth_dev, internal);
 819
 820                 if (dev_conf->intr_conf.rxq) {
 821                         if (eth_vhost_install_intr(eth_dev) < 0) {
 822                                 VHOST_LOG(INFO,
 823                                         "Failed to install interrupt handler.");
 824                                         return -1;
 825                         }
 826                 }
 827         } else {
 828                 VHOST_LOG(INFO, "RX/TX queues not exist yet\n");
 829         }
 830
 831         for (i = 0; i < rte_vhost_get_vring_num(vid); i++)
 832                 rte_vhost_enable_guest_notification(vid, i, 0);
 833
 834         rte_vhost_get_mtu(vid, &eth_dev->data->mtu);
 835
 836         eth_dev->data->dev_link.link_status = RTE_ETH_LINK_UP;
 837
 838         rte_atomic32_set(&internal->dev_attached, 1);
 839         update_queuing_status(eth_dev);
 840
 841         VHOST_LOG(INFO, "Vhost device %d created\n", vid);
 842
 843         rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 844
 845         return 0;
 846 }
 847
 848 static void
 849 destroy_device(int vid)
 850 {
 851         struct rte_eth_dev *eth_dev;
 852         struct pmd_internal *internal;
 853         struct vhost_queue *vq;
 854         struct internal_list *list;
 855         char ifname[PATH_MAX];
 856         unsigned i;
 857         struct rte_vhost_vring_state *state;
 858
 859         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 860         list = find_internal_resource(ifname);
 861         if (list == NULL) {
 862                 VHOST_LOG(ERR, "Invalid interface name: %s\n", ifname);
 863                 return;
 864         }
 865         eth_dev = list->eth_dev;
 866         internal = eth_dev->data->dev_private;
 867
 868         rte_atomic32_set(&internal->dev_attached, 0);
 869         update_queuing_status(eth_dev);
 870
 871         eth_dev->data->dev_link.link_status = RTE_ETH_LINK_DOWN;
 872
 873         if (eth_dev->data->rx_queues && eth_dev->data->tx_queues) {
 874                 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 875                         vq = eth_dev->data->rx_queues[i];
 876                         if (!vq)
 877                                 continue;
 878                         vq->vid = -1;
 879                 }
 880                 for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 881                         vq = eth_dev->data->tx_queues[i];
 882                         if (!vq)
 883                                 continue;
 884                         vq->vid = -1;
 885                 }
 886         }
 887
 888         state = vring_states[eth_dev->data->port_id];
 889         rte_spinlock_lock(&state->lock);
 890         for (i = 0; i <= state->max_vring; i++) {
 891                 state->cur[i] = false;
 892                 state->seen[i] = false;
 893         }
 894         state->max_vring = 0;
 895         rte_spinlock_unlock(&state->lock);
 896
 897         VHOST_LOG(INFO, "Vhost device %d destroyed\n", vid);
 898         eth_vhost_uninstall_intr(eth_dev);
 899
 900         rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 901 }
 902
 903 static int
 904 vring_conf_update(int vid, struct rte_eth_dev *eth_dev, uint16_t vring_id)
 905 {
 906         struct rte_eth_conf *dev_conf = &eth_dev->data->dev_conf;
 907         struct pmd_internal *internal = eth_dev->data->dev_private;
 908         struct vhost_queue *vq;
 909         struct rte_vhost_vring vring;
 910         int rx_idx = vring_id % 2 ? (vring_id - 1) >> 1 : -1;
 911         int ret = 0;
 912
 913         /*
 914          * The vring kickfd may be changed after the new device notification.
 915          * Update it when the vring state is updated.
 916          */
 917         if (rx_idx >= 0 && rx_idx < eth_dev->data->nb_rx_queues &&
 918             rte_atomic32_read(&internal->dev_attached) &&
 919             rte_atomic32_read(&internal->started) &&
 920             dev_conf->intr_conf.rxq) {
 921                 ret = rte_vhost_get_vhost_vring(vid, vring_id, &vring);
 922                 if (ret) {
 923                         VHOST_LOG(ERR, "Failed to get vring %d information.\n",
 924                                         vring_id);
 925                         return ret;
 926                 }
 927
 928                 if (rte_intr_efds_index_set(eth_dev->intr_handle, rx_idx,
 929                                                    vring.kickfd))
 930                         return -rte_errno;
 931
 932                 vq = eth_dev->data->rx_queues[rx_idx];
 933                 if (!vq) {
 934                         VHOST_LOG(ERR, "rxq%d is not setup yet\n", rx_idx);
 935                         return -1;
 936                 }
 937
 938                 rte_spinlock_lock(&vq->intr_lock);
 939                 if (vq->intr_enable)
 940                         ret = eth_vhost_update_intr(eth_dev, rx_idx);
 941                 rte_spinlock_unlock(&vq->intr_lock);
 942         }
 943
 944         return ret;
 945 }
 946
 947 static int
 948 vring_state_changed(int vid, uint16_t vring, int enable)
 949 {
 950         struct rte_vhost_vring_state *state;
 951         struct rte_eth_dev *eth_dev;
 952         struct internal_list *list;
 953         char ifname[PATH_MAX];
 954
 955         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 956         list = find_internal_resource(ifname);
 957         if (list == NULL) {
 958                 VHOST_LOG(ERR, "Invalid interface name: %s\n", ifname);
 959                 return -1;
 960         }
 961
 962         eth_dev = list->eth_dev;
 963         /* won't be NULL */
 964         state = vring_states[eth_dev->data->port_id];
 965
 966         if (enable && vring_conf_update(vid, eth_dev, vring))
 967                 VHOST_LOG(INFO, "Failed to update vring-%d configuration.\n",
 968                           (int)vring);
 969
 970         rte_spinlock_lock(&state->lock);
 971         if (state->cur[vring] == enable) {
 972                 rte_spinlock_unlock(&state->lock);
 973                 return 0;
 974         }
 975         state->cur[vring] = enable;
 976         state->max_vring = RTE_MAX(vring, state->max_vring);
 977         rte_spinlock_unlock(&state->lock);
 978
 979         update_queuing_status(eth_dev);
 980
 981         VHOST_LOG(INFO, "vring%u is %s\n",
 982                         vring, enable ? "enabled" : "disabled");
 983
 984         rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE, NULL);
 985
 986         return 0;
 987 }
 988
 989 static struct rte_vhost_device_ops vhost_ops = {
 990         .new_device          = new_device,
 991         .destroy_device      = destroy_device,
 992         .vring_state_changed = vring_state_changed,
 993 };
 994
 995 static int
 996 vhost_driver_setup(struct rte_eth_dev *eth_dev)
 997 {
 998         struct pmd_internal *internal = eth_dev->data->dev_private;
 999         struct internal_list *list = NULL;
1000         struct rte_vhost_vring_state *vring_state = NULL;
1001         unsigned int numa_node = eth_dev->device->numa_node;
1002         const char *name = eth_dev->device->name;
1003
1004         /* Don't try to setup again if it has already been done. */
1005         list = find_internal_resource(internal->iface_name);
1006         if (list)
1007                 return 0;
1008
1009         list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
1010         if (list == NULL)
1011                 return -1;
1012
1013         vring_state = rte_zmalloc_socket(name, sizeof(*vring_state),
1014                                          0, numa_node);
1015         if (vring_state == NULL)
1016                 goto free_list;
1017
1018         list->eth_dev = eth_dev;
1019         pthread_mutex_lock(&internal_list_lock);
1020         TAILQ_INSERT_TAIL(&internal_list, list, next);
1021         pthread_mutex_unlock(&internal_list_lock);
1022
1023         rte_spinlock_init(&vring_state->lock);
1024         vring_states[eth_dev->data->port_id] = vring_state;
1025
1026         if (rte_vhost_driver_register(internal->iface_name, internal->flags))
1027                 goto list_remove;
1028
1029         if (internal->disable_flags) {
1030                 if (rte_vhost_driver_disable_features(internal->iface_name,
1031                                                       internal->disable_flags))
1032                         goto drv_unreg;
1033         }
1034
1035         if (rte_vhost_driver_callback_register(internal->iface_name,
1036                                                &vhost_ops) < 0) {
1037                 VHOST_LOG(ERR, "Can't register callbacks\n");
1038                 goto drv_unreg;
1039         }
1040
1041         if (rte_vhost_driver_start(internal->iface_name) < 0) {
1042                 VHOST_LOG(ERR, "Failed to start driver for %s\n",
1043                           internal->iface_name);
1044                 goto drv_unreg;
1045         }
1046
1047         return 0;
1048
1049 drv_unreg:
1050         rte_vhost_driver_unregister(internal->iface_name);
1051 list_remove:
1052         vring_states[eth_dev->data->port_id] = NULL;
1053         pthread_mutex_lock(&internal_list_lock);
1054         TAILQ_REMOVE(&internal_list, list, next);
1055         pthread_mutex_unlock(&internal_list_lock);
1056         rte_free(vring_state);
1057 free_list:
1058         rte_free(list);
1059
1060         return -1;
1061 }
1062
1063 int
1064 rte_eth_vhost_get_queue_event(uint16_t port_id,
1065                 struct rte_eth_vhost_queue_event *event)
1066 {
1067         struct rte_vhost_vring_state *state;
1068         unsigned int i;
1069         int idx;
1070
1071         if (port_id >= RTE_MAX_ETHPORTS) {
1072                 VHOST_LOG(ERR, "Invalid port id\n");
1073                 return -1;
1074         }
1075
1076         state = vring_states[port_id];
1077         if (!state) {
1078                 VHOST_LOG(ERR, "Unused port\n");
1079                 return -1;
1080         }
1081
1082         rte_spinlock_lock(&state->lock);
1083         for (i = 0; i <= state->max_vring; i++) {
1084                 idx = state->index++ % (state->max_vring + 1);
1085
1086                 if (state->cur[idx] != state->seen[idx]) {
1087                         state->seen[idx] = state->cur[idx];
1088                         event->queue_id = idx / 2;
1089                         event->rx = idx & 1;
1090                         event->enable = state->cur[idx];
1091                         rte_spinlock_unlock(&state->lock);
1092                         return 0;
1093                 }
1094         }
1095         rte_spinlock_unlock(&state->lock);
1096
1097         return -1;
1098 }
1099
1100 int
1101 rte_eth_vhost_get_vid_from_port_id(uint16_t port_id)
1102 {
1103         struct internal_list *list;
1104         struct rte_eth_dev *eth_dev;
1105         struct vhost_queue *vq;
1106         int vid = -1;
1107
1108         if (!rte_eth_dev_is_valid_port(port_id))
1109                 return -1;
1110
1111         pthread_mutex_lock(&internal_list_lock);
1112
1113         TAILQ_FOREACH(list, &internal_list, next) {
1114                 eth_dev = list->eth_dev;
1115                 if (eth_dev->data->port_id == port_id) {
1116                         vq = eth_dev->data->rx_queues[0];
1117                         if (vq) {
1118                                 vid = vq->vid;
1119                         }
1120                         break;
1121                 }
1122         }
1123
1124         pthread_mutex_unlock(&internal_list_lock);
1125
1126         return vid;
1127 }
1128
1129 static int
1130 eth_dev_configure(struct rte_eth_dev *dev)
1131 {
1132         struct pmd_internal *internal = dev->data->dev_private;
1133         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1134
1135         /* NOTE: the same process has to operate a vhost interface
1136          * from beginning to end (from eth_dev configure to eth_dev close).
1137          * It is user's responsibility at the moment.
1138          */
1139         if (vhost_driver_setup(dev) < 0)
1140                 return -1;
1141
1142         internal->vlan_strip = !!(rxmode->offloads & RTE_ETH_RX_OFFLOAD_VLAN_STRIP);
1143
1144         return 0;
1145 }
1146
1147 static int
1148 eth_dev_start(struct rte_eth_dev *eth_dev)
1149 {
1150         struct pmd_internal *internal = eth_dev->data->dev_private;
1151         struct rte_eth_conf *dev_conf = &eth_dev->data->dev_conf;
1152
1153         queue_setup(eth_dev, internal);
1154
1155         if (rte_atomic32_read(&internal->dev_attached) == 1) {
1156                 if (dev_conf->intr_conf.rxq) {
1157                         if (eth_vhost_install_intr(eth_dev) < 0) {
1158                                 VHOST_LOG(INFO,
1159                                         "Failed to install interrupt handler.");
1160                                         return -1;
1161                         }
1162                 }
1163         }
1164
1165         rte_atomic32_set(&internal->started, 1);
1166         update_queuing_status(eth_dev);
1167
1168         return 0;
1169 }
1170
1171 static int
1172 eth_dev_stop(struct rte_eth_dev *dev)
1173 {
1174         struct pmd_internal *internal = dev->data->dev_private;
1175
1176         dev->data->dev_started = 0;
1177         rte_atomic32_set(&internal->started, 0);
1178         update_queuing_status(dev);
1179
1180         return 0;
1181 }
1182
1183 static int
1184 eth_dev_close(struct rte_eth_dev *dev)
1185 {
1186         struct pmd_internal *internal;
1187         struct internal_list *list;
1188         unsigned int i, ret;
1189
1190         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1191                 return 0;
1192
1193         internal = dev->data->dev_private;
1194         if (!internal)
1195                 return 0;
1196
1197         ret = eth_dev_stop(dev);
1198
1199         list = find_internal_resource(internal->iface_name);
1200         if (list) {
1201                 rte_vhost_driver_unregister(internal->iface_name);
1202                 pthread_mutex_lock(&internal_list_lock);
1203                 TAILQ_REMOVE(&internal_list, list, next);
1204                 pthread_mutex_unlock(&internal_list_lock);
1205                 rte_free(list);
1206         }
1207
1208         if (dev->data->rx_queues)
1209                 for (i = 0; i < dev->data->nb_rx_queues; i++)
1210                         rte_free(dev->data->rx_queues[i]);
1211
1212         if (dev->data->tx_queues)
1213                 for (i = 0; i < dev->data->nb_tx_queues; i++)
1214                         rte_free(dev->data->tx_queues[i]);
1215
1216         rte_free(internal->iface_name);
1217         rte_free(internal);
1218
1219         dev->data->dev_private = NULL;
1220
1221         rte_free(vring_states[dev->data->port_id]);
1222         vring_states[dev->data->port_id] = NULL;
1223
1224         return ret;
1225 }
1226
1227 static int
1228 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1229                    uint16_t nb_rx_desc __rte_unused,
1230                    unsigned int socket_id,
1231                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1232                    struct rte_mempool *mb_pool)
1233 {
1234         struct vhost_queue *vq;
1235
1236         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
1237                         RTE_CACHE_LINE_SIZE, socket_id);
1238         if (vq == NULL) {
1239                 VHOST_LOG(ERR, "Failed to allocate memory for rx queue\n");
1240                 return -ENOMEM;
1241         }
1242
1243         vq->mb_pool = mb_pool;
1244         vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
1245         rte_spinlock_init(&vq->intr_lock);
1246         dev->data->rx_queues[rx_queue_id] = vq;
1247
1248         return 0;
1249 }
1250
1251 static int
1252 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1253                    uint16_t nb_tx_desc __rte_unused,
1254                    unsigned int socket_id,
1255                    const struct rte_eth_txconf *tx_conf __rte_unused)
1256 {
1257         struct vhost_queue *vq;
1258
1259         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
1260                         RTE_CACHE_LINE_SIZE, socket_id);
1261         if (vq == NULL) {
1262                 VHOST_LOG(ERR, "Failed to allocate memory for tx queue\n");
1263                 return -ENOMEM;
1264         }
1265
1266         vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
1267         rte_spinlock_init(&vq->intr_lock);
1268         dev->data->tx_queues[tx_queue_id] = vq;
1269
1270         return 0;
1271 }
1272
1273 static int
1274 eth_dev_info(struct rte_eth_dev *dev,
1275              struct rte_eth_dev_info *dev_info)
1276 {
1277         struct pmd_internal *internal;
1278
1279         internal = dev->data->dev_private;
1280         if (internal == NULL) {
1281                 VHOST_LOG(ERR, "Invalid device specified\n");
1282                 return -ENODEV;
1283         }
1284
1285         dev_info->max_mac_addrs = 1;
1286         dev_info->max_rx_pktlen = (uint32_t)-1;
1287         dev_info->max_rx_queues = internal->max_queues;
1288         dev_info->max_tx_queues = internal->max_queues;
1289         dev_info->min_rx_bufsize = 0;
1290
1291         dev_info->tx_offload_capa = RTE_ETH_TX_OFFLOAD_MULTI_SEGS |
1292                                 RTE_ETH_TX_OFFLOAD_VLAN_INSERT;
1293         dev_info->rx_offload_capa = RTE_ETH_RX_OFFLOAD_VLAN_STRIP;
1294
1295         return 0;
1296 }
1297
1298 static int
1299 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1300 {
1301         unsigned i;
1302         unsigned long rx_total = 0, tx_total = 0;
1303         unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
1304         struct vhost_queue *vq;
1305
1306         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
1307                         i < dev->data->nb_rx_queues; i++) {
1308                 if (dev->data->rx_queues[i] == NULL)
1309                         continue;
1310                 vq = dev->data->rx_queues[i];
1311                 stats->q_ipackets[i] = vq->stats.pkts;
1312                 rx_total += stats->q_ipackets[i];
1313
1314                 stats->q_ibytes[i] = vq->stats.bytes;
1315                 rx_total_bytes += stats->q_ibytes[i];
1316         }
1317
1318         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
1319                         i < dev->data->nb_tx_queues; i++) {
1320                 if (dev->data->tx_queues[i] == NULL)
1321                         continue;
1322                 vq = dev->data->tx_queues[i];
1323                 stats->q_opackets[i] = vq->stats.pkts;
1324                 tx_total += stats->q_opackets[i];
1325
1326                 stats->q_obytes[i] = vq->stats.bytes;
1327                 tx_total_bytes += stats->q_obytes[i];
1328         }
1329
1330         stats->ipackets = rx_total;
1331         stats->opackets = tx_total;
1332         stats->ibytes = rx_total_bytes;
1333         stats->obytes = tx_total_bytes;
1334
1335         return 0;
1336 }
1337
1338 static int
1339 eth_stats_reset(struct rte_eth_dev *dev)
1340 {
1341         struct vhost_queue *vq;
1342         unsigned i;
1343
1344         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1345                 if (dev->data->rx_queues[i] == NULL)
1346                         continue;
1347                 vq = dev->data->rx_queues[i];
1348                 vq->stats.pkts = 0;
1349                 vq->stats.bytes = 0;
1350         }
1351         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1352                 if (dev->data->tx_queues[i] == NULL)
1353                         continue;
1354                 vq = dev->data->tx_queues[i];
1355                 vq->stats.pkts = 0;
1356                 vq->stats.bytes = 0;
1357                 vq->stats.missed_pkts = 0;
1358         }
1359
1360         return 0;
1361 }
1362
1363 static void
1364 eth_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
1365 {
1366         rte_free(dev->data->rx_queues[qid]);
1367 }
1368
1369 static void
1370 eth_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
1371 {
1372         rte_free(dev->data->tx_queues[qid]);
1373 }
1374
1375 static int
1376 eth_tx_done_cleanup(void *txq __rte_unused, uint32_t free_cnt __rte_unused)
1377 {
1378         /*
1379          * vHost does not hang onto mbuf. eth_vhost_tx() copies packet data
1380          * and releases mbuf, so nothing to cleanup.
1381          */
1382         return 0;
1383 }
1384
1385 static int
1386 eth_link_update(struct rte_eth_dev *dev __rte_unused,
1387                 int wait_to_complete __rte_unused)
1388 {
1389         return 0;
1390 }
1391
1392 static uint32_t
1393 eth_rx_queue_count(void *rx_queue)
1394 {
1395         struct vhost_queue *vq;
1396
1397         vq = rx_queue;
1398         if (vq == NULL)
1399                 return 0;
1400
1401         return rte_vhost_rx_queue_count(vq->vid, vq->virtqueue_id);
1402 }
1403
1404 #define CLB_VAL_IDX 0
1405 #define CLB_MSK_IDX 1
1406 #define CLB_MATCH_IDX 2
1407 static int
1408 vhost_monitor_callback(const uint64_t value,
1409                 const uint64_t opaque[RTE_POWER_MONITOR_OPAQUE_SZ])
1410 {
1411         const uint64_t m = opaque[CLB_MSK_IDX];
1412         const uint64_t v = opaque[CLB_VAL_IDX];
1413         const uint64_t c = opaque[CLB_MATCH_IDX];
1414
1415         if (c)
1416                 return (value & m) == v ? -1 : 0;
1417         else
1418                 return (value & m) == v ? 0 : -1;
1419 }
1420
1421 static int
1422 vhost_get_monitor_addr(void *rx_queue, struct rte_power_monitor_cond *pmc)
1423 {
1424         struct vhost_queue *vq = rx_queue;
1425         struct rte_vhost_power_monitor_cond vhost_pmc;
1426         int ret;
1427         if (vq == NULL)
1428                 return -EINVAL;
1429         ret = rte_vhost_get_monitor_addr(vq->vid, vq->virtqueue_id,
1430                         &vhost_pmc);
1431         if (ret < 0)
1432                 return -EINVAL;
1433         pmc->addr = vhost_pmc.addr;
1434         pmc->opaque[CLB_VAL_IDX] = vhost_pmc.val;
1435         pmc->opaque[CLB_MSK_IDX] = vhost_pmc.mask;
1436         pmc->opaque[CLB_MATCH_IDX] = vhost_pmc.match;
1437         pmc->size = vhost_pmc.size;
1438         pmc->fn = vhost_monitor_callback;
1439
1440         return 0;
1441 }
1442
1443 static const struct eth_dev_ops ops = {
1444         .dev_start = eth_dev_start,
1445         .dev_stop = eth_dev_stop,
1446         .dev_close = eth_dev_close,
1447         .dev_configure = eth_dev_configure,
1448         .dev_infos_get = eth_dev_info,
1449         .rx_queue_setup = eth_rx_queue_setup,
1450         .tx_queue_setup = eth_tx_queue_setup,
1451         .rx_queue_release = eth_rx_queue_release,
1452         .tx_queue_release = eth_tx_queue_release,
1453         .tx_done_cleanup = eth_tx_done_cleanup,
1454         .link_update = eth_link_update,
1455         .stats_get = eth_stats_get,
1456         .stats_reset = eth_stats_reset,
1457         .xstats_reset = vhost_dev_xstats_reset,
1458         .xstats_get = vhost_dev_xstats_get,
1459         .xstats_get_names = vhost_dev_xstats_get_names,
1460         .rx_queue_intr_enable = eth_rxq_intr_enable,
1461         .rx_queue_intr_disable = eth_rxq_intr_disable,
1462         .get_monitor_addr = vhost_get_monitor_addr,
1463 };
1464
1465 static int
1466 eth_dev_vhost_create(struct rte_vdev_device *dev, char *iface_name,
1467         int16_t queues, const unsigned int numa_node, uint64_t flags,
1468         uint64_t disable_flags)
1469 {
1470         const char *name = rte_vdev_device_name(dev);
1471         struct rte_eth_dev_data *data;
1472         struct pmd_internal *internal = NULL;
1473         struct rte_eth_dev *eth_dev = NULL;
1474         struct rte_ether_addr *eth_addr = NULL;
1475
1476         VHOST_LOG(INFO, "Creating VHOST-USER backend on numa socket %u\n",
1477                 numa_node);
1478
1479         /* reserve an ethdev entry */
1480         eth_dev = rte_eth_vdev_allocate(dev, sizeof(*internal));
1481         if (eth_dev == NULL)
1482                 goto error;
1483         data = eth_dev->data;
1484
1485         eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
1486         if (eth_addr == NULL)
1487                 goto error;
1488         data->mac_addrs = eth_addr;
1489         *eth_addr = base_eth_addr;
1490         eth_addr->addr_bytes[5] = eth_dev->data->port_id;
1491
1492         /* now put it all together
1493          * - store queue data in internal,
1494          * - point eth_dev_data to internals
1495          * - and point eth_dev structure to new eth_dev_data structure
1496          */
1497         internal = eth_dev->data->dev_private;
1498         internal->iface_name = rte_malloc_socket(name, strlen(iface_name) + 1,
1499                                                  0, numa_node);
1500         if (internal->iface_name == NULL)
1501                 goto error;
1502         strcpy(internal->iface_name, iface_name);
1503
1504         data->nb_rx_queues = queues;
1505         data->nb_tx_queues = queues;
1506         internal->max_queues = queues;
1507         internal->vid = -1;
1508         internal->flags = flags;
1509         internal->disable_flags = disable_flags;
1510         data->dev_link = pmd_link;
1511         data->dev_flags = RTE_ETH_DEV_INTR_LSC |
1512                                 RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1513         data->promiscuous = 1;
1514         data->all_multicast = 1;
1515
1516         eth_dev->dev_ops = &ops;
1517         eth_dev->rx_queue_count = eth_rx_queue_count;
1518
1519         /* finally assign rx and tx ops */
1520         eth_dev->rx_pkt_burst = eth_vhost_rx;
1521         eth_dev->tx_pkt_burst = eth_vhost_tx;
1522
1523         rte_eth_dev_probing_finish(eth_dev);
1524         return 0;
1525
1526 error:
1527         if (internal)
1528                 rte_free(internal->iface_name);
1529         rte_eth_dev_release_port(eth_dev);
1530
1531         return -1;
1532 }
1533
1534 static inline int
1535 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
1536 {
1537         const char **iface_name = extra_args;
1538
1539         if (value == NULL)
1540                 return -1;
1541
1542         *iface_name = value;
1543
1544         return 0;
1545 }
1546
1547 static inline int
1548 open_int(const char *key __rte_unused, const char *value, void *extra_args)
1549 {
1550         uint16_t *n = extra_args;
1551
1552         if (value == NULL || extra_args == NULL)
1553                 return -EINVAL;
1554
1555         *n = (uint16_t)strtoul(value, NULL, 0);
1556         if (*n == USHRT_MAX && errno == ERANGE)
1557                 return -1;
1558
1559         return 0;
1560 }
1561
1562 static int
1563 rte_pmd_vhost_probe(struct rte_vdev_device *dev)
1564 {
1565         struct rte_kvargs *kvlist = NULL;
1566         int ret = 0;
1567         char *iface_name;
1568         uint16_t queues;
1569         uint64_t flags = 0;
1570         uint64_t disable_flags = 0;
1571         int client_mode = 0;
1572         int iommu_support = 0;
1573         int postcopy_support = 0;
1574         int tso = 0;
1575         int linear_buf = 0;
1576         int ext_buf = 0;
1577         struct rte_eth_dev *eth_dev;
1578         const char *name = rte_vdev_device_name(dev);
1579
1580         VHOST_LOG(INFO, "Initializing pmd_vhost for %s\n", name);
1581
1582         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1583                 eth_dev = rte_eth_dev_attach_secondary(name);
1584                 if (!eth_dev) {
1585                         VHOST_LOG(ERR, "Failed to probe %s\n", name);
1586                         return -1;
1587                 }
1588                 eth_dev->rx_pkt_burst = eth_vhost_rx;
1589                 eth_dev->tx_pkt_burst = eth_vhost_tx;
1590                 eth_dev->dev_ops = &ops;
1591                 if (dev->device.numa_node == SOCKET_ID_ANY)
1592                         dev->device.numa_node = rte_socket_id();
1593                 eth_dev->device = &dev->device;
1594                 rte_eth_dev_probing_finish(eth_dev);
1595                 return 0;
1596         }
1597
1598         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
1599         if (kvlist == NULL)
1600                 return -1;
1601
1602         if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
1603                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
1604                                          &open_iface, &iface_name);
1605                 if (ret < 0)
1606                         goto out_free;
1607         } else {
1608                 ret = -1;
1609                 goto out_free;
1610         }
1611
1612         if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
1613                 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
1614                                          &open_int, &queues);
1615                 if (ret < 0 || queues > RTE_MAX_QUEUES_PER_PORT)
1616                         goto out_free;
1617
1618         } else
1619                 queues = 1;
1620
1621         if (rte_kvargs_count(kvlist, ETH_VHOST_CLIENT_ARG) == 1) {
1622                 ret = rte_kvargs_process(kvlist, ETH_VHOST_CLIENT_ARG,
1623                                          &open_int, &client_mode);
1624                 if (ret < 0)
1625                         goto out_free;
1626
1627                 if (client_mode)
1628                         flags |= RTE_VHOST_USER_CLIENT;
1629         }
1630
1631         if (rte_kvargs_count(kvlist, ETH_VHOST_IOMMU_SUPPORT) == 1) {
1632                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IOMMU_SUPPORT,
1633                                          &open_int, &iommu_support);
1634                 if (ret < 0)
1635                         goto out_free;
1636
1637                 if (iommu_support)
1638                         flags |= RTE_VHOST_USER_IOMMU_SUPPORT;
1639         }
1640
1641         if (rte_kvargs_count(kvlist, ETH_VHOST_POSTCOPY_SUPPORT) == 1) {
1642                 ret = rte_kvargs_process(kvlist, ETH_VHOST_POSTCOPY_SUPPORT,
1643                                          &open_int, &postcopy_support);
1644                 if (ret < 0)
1645                         goto out_free;
1646
1647                 if (postcopy_support)
1648                         flags |= RTE_VHOST_USER_POSTCOPY_SUPPORT;
1649         }
1650
1651         if (rte_kvargs_count(kvlist, ETH_VHOST_VIRTIO_NET_F_HOST_TSO) == 1) {
1652                 ret = rte_kvargs_process(kvlist,
1653                                 ETH_VHOST_VIRTIO_NET_F_HOST_TSO,
1654                                 &open_int, &tso);
1655                 if (ret < 0)
1656                         goto out_free;
1657         }
1658
1659         if (tso == 0) {
1660                 disable_flags |= (1ULL << VIRTIO_NET_F_HOST_TSO4);
1661                 disable_flags |= (1ULL << VIRTIO_NET_F_HOST_TSO6);
1662         }
1663
1664         if (rte_kvargs_count(kvlist, ETH_VHOST_LINEAR_BUF) == 1) {
1665                 ret = rte_kvargs_process(kvlist,
1666                                 ETH_VHOST_LINEAR_BUF,
1667                                 &open_int, &linear_buf);
1668                 if (ret < 0)
1669                         goto out_free;
1670
1671                 if (linear_buf == 1)
1672                         flags |= RTE_VHOST_USER_LINEARBUF_SUPPORT;
1673         }
1674
1675         if (rte_kvargs_count(kvlist, ETH_VHOST_EXT_BUF) == 1) {
1676                 ret = rte_kvargs_process(kvlist,
1677                                 ETH_VHOST_EXT_BUF,
1678                                 &open_int, &ext_buf);
1679                 if (ret < 0)
1680                         goto out_free;
1681
1682                 if (ext_buf == 1)
1683                         flags |= RTE_VHOST_USER_EXTBUF_SUPPORT;
1684         }
1685
1686         if (dev->device.numa_node == SOCKET_ID_ANY)
1687                 dev->device.numa_node = rte_socket_id();
1688
1689         ret = eth_dev_vhost_create(dev, iface_name, queues,
1690                                    dev->device.numa_node, flags, disable_flags);
1691         if (ret == -1)
1692                 VHOST_LOG(ERR, "Failed to create %s\n", name);
1693
1694 out_free:
1695         rte_kvargs_free(kvlist);
1696         return ret;
1697 }
1698
1699 static int
1700 rte_pmd_vhost_remove(struct rte_vdev_device *dev)
1701 {
1702         const char *name;
1703         struct rte_eth_dev *eth_dev = NULL;
1704
1705         name = rte_vdev_device_name(dev);
1706         VHOST_LOG(INFO, "Un-Initializing pmd_vhost for %s\n", name);
1707
1708         /* find an ethdev entry */
1709         eth_dev = rte_eth_dev_allocated(name);
1710         if (eth_dev == NULL)
1711                 return 0;
1712
1713         eth_dev_close(eth_dev);
1714         rte_eth_dev_release_port(eth_dev);
1715
1716         return 0;
1717 }
1718
1719 static struct rte_vdev_driver pmd_vhost_drv = {
1720         .probe = rte_pmd_vhost_probe,
1721         .remove = rte_pmd_vhost_remove,
1722 };
1723
1724 RTE_PMD_REGISTER_VDEV(net_vhost, pmd_vhost_drv);
1725 RTE_PMD_REGISTER_ALIAS(net_vhost, eth_vhost);
1726 RTE_PMD_REGISTER_PARAM_STRING(net_vhost,
1727         "iface=<ifc> "
1728         "queues=<int> "
1729         "client=<0|1> "
1730         "iommu-support=<0|1> "
1731         "postcopy-support=<0|1> "
1732         "tso=<0|1> "
1733         "linear-buffer=<0|1> "
1734         "ext-buffer=<0|1>");