drivers/net/vhost/rte_eth_vhost.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright (c) 2016 IGEL Co., Ltd.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of IGEL Co.,Ltd. nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33 #include <unistd.h>
  34 #include <pthread.h>
  35 #include <stdbool.h>
  36 #ifdef RTE_LIBRTE_VHOST_NUMA
  37 #include <numaif.h>
  38 #endif
  39
  40 #include <rte_mbuf.h>
  41 #include <rte_ethdev.h>
  42 #include <rte_malloc.h>
  43 #include <rte_memcpy.h>
  44 #include <rte_vdev.h>
  45 #include <rte_kvargs.h>
  46 #include <rte_virtio_net.h>
  47 #include <rte_spinlock.h>
  48
  49 #include "rte_eth_vhost.h"
  50
  51 #define ETH_VHOST_IFACE_ARG             "iface"
  52 #define ETH_VHOST_QUEUES_ARG            "queues"
  53 #define ETH_VHOST_CLIENT_ARG            "client"
  54
  55 static const char *drivername = "VHOST PMD";
  56
  57 static const char *valid_arguments[] = {
  58         ETH_VHOST_IFACE_ARG,
  59         ETH_VHOST_QUEUES_ARG,
  60         ETH_VHOST_CLIENT_ARG,
  61         NULL
  62 };
  63
  64 static struct ether_addr base_eth_addr = {
  65         .addr_bytes = {
  66                 0x56 /* V */,
  67                 0x48 /* H */,
  68                 0x4F /* O */,
  69                 0x53 /* S */,
  70                 0x54 /* T */,
  71                 0x00
  72         }
  73 };
  74
  75 enum vhost_xstats_pkts {
  76         VHOST_UNDERSIZE_PKT = 0,
  77         VHOST_64_PKT,
  78         VHOST_65_TO_127_PKT,
  79         VHOST_128_TO_255_PKT,
  80         VHOST_256_TO_511_PKT,
  81         VHOST_512_TO_1023_PKT,
  82         VHOST_1024_TO_1522_PKT,
  83         VHOST_1523_TO_MAX_PKT,
  84         VHOST_BROADCAST_PKT,
  85         VHOST_MULTICAST_PKT,
  86         VHOST_UNICAST_PKT,
  87         VHOST_ERRORS_PKT,
  88         VHOST_ERRORS_FRAGMENTED,
  89         VHOST_ERRORS_JABBER,
  90         VHOST_UNKNOWN_PROTOCOL,
  91         VHOST_XSTATS_MAX,
  92 };
  93
  94 struct vhost_stats {
  95         uint64_t pkts;
  96         uint64_t bytes;
  97         uint64_t missed_pkts;
  98         uint64_t xstats[VHOST_XSTATS_MAX];
  99 };
 100
 101 struct vhost_queue {
 102         int vid;
 103         rte_atomic32_t allow_queuing;
 104         rte_atomic32_t while_queuing;
 105         struct pmd_internal *internal;
 106         struct rte_mempool *mb_pool;
 107         uint8_t port;
 108         uint16_t virtqueue_id;
 109         struct vhost_stats stats;
 110 };
 111
 112 struct pmd_internal {
 113         char *dev_name;
 114         char *iface_name;
 115         uint16_t max_queues;
 116         uint64_t flags;
 117
 118         volatile uint16_t once;
 119 };
 120
 121 struct internal_list {
 122         TAILQ_ENTRY(internal_list) next;
 123         struct rte_eth_dev *eth_dev;
 124 };
 125
 126 TAILQ_HEAD(internal_list_head, internal_list);
 127 static struct internal_list_head internal_list =
 128         TAILQ_HEAD_INITIALIZER(internal_list);
 129
 130 static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
 131
 132 static rte_atomic16_t nb_started_ports;
 133 static pthread_t session_th;
 134
 135 static struct rte_eth_link pmd_link = {
 136                 .link_speed = 10000,
 137                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 138                 .link_status = ETH_LINK_DOWN
 139 };
 140
 141 struct rte_vhost_vring_state {
 142         rte_spinlock_t lock;
 143
 144         bool cur[RTE_MAX_QUEUES_PER_PORT * 2];
 145         bool seen[RTE_MAX_QUEUES_PER_PORT * 2];
 146         unsigned int index;
 147         unsigned int max_vring;
 148 };
 149
 150 static struct rte_vhost_vring_state *vring_states[RTE_MAX_ETHPORTS];
 151
 152 #define VHOST_XSTATS_NAME_SIZE 64
 153
 154 struct vhost_xstats_name_off {
 155         char name[VHOST_XSTATS_NAME_SIZE];
 156         uint64_t offset;
 157 };
 158
 159 /* [rx]_is prepended to the name string here */
 160 static const struct vhost_xstats_name_off vhost_rxport_stat_strings[] = {
 161         {"good_packets",
 162          offsetof(struct vhost_queue, stats.pkts)},
 163         {"total_bytes",
 164          offsetof(struct vhost_queue, stats.bytes)},
 165         {"missed_pkts",
 166          offsetof(struct vhost_queue, stats.missed_pkts)},
 167         {"broadcast_packets",
 168          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
 169         {"multicast_packets",
 170          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
 171         {"unicast_packets",
 172          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
 173          {"undersize_packets",
 174          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
 175         {"size_64_packets",
 176          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
 177         {"size_65_to_127_packets",
 178          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
 179         {"size_128_to_255_packets",
 180          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
 181         {"size_256_to_511_packets",
 182          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
 183         {"size_512_to_1023_packets",
 184          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
 185         {"size_1024_to_1522_packets",
 186          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
 187         {"size_1523_to_max_packets",
 188          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
 189         {"errors_with_bad_CRC",
 190          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
 191         {"fragmented_errors",
 192          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_FRAGMENTED])},
 193         {"jabber_errors",
 194          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_JABBER])},
 195         {"unknown_protos_packets",
 196          offsetof(struct vhost_queue, stats.xstats[VHOST_UNKNOWN_PROTOCOL])},
 197 };
 198
 199 /* [tx]_ is prepended to the name string here */
 200 static const struct vhost_xstats_name_off vhost_txport_stat_strings[] = {
 201         {"good_packets",
 202          offsetof(struct vhost_queue, stats.pkts)},
 203         {"total_bytes",
 204          offsetof(struct vhost_queue, stats.bytes)},
 205         {"missed_pkts",
 206          offsetof(struct vhost_queue, stats.missed_pkts)},
 207         {"broadcast_packets",
 208          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
 209         {"multicast_packets",
 210          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
 211         {"unicast_packets",
 212          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
 213         {"undersize_packets",
 214          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
 215         {"size_64_packets",
 216          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
 217         {"size_65_to_127_packets",
 218          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
 219         {"size_128_to_255_packets",
 220          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
 221         {"size_256_to_511_packets",
 222          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
 223         {"size_512_to_1023_packets",
 224          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
 225         {"size_1024_to_1522_packets",
 226          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
 227         {"size_1523_to_max_packets",
 228          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
 229         {"errors_with_bad_CRC",
 230          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
 231 };
 232
 233 #define VHOST_NB_XSTATS_RXPORT (sizeof(vhost_rxport_stat_strings) / \
 234                                 sizeof(vhost_rxport_stat_strings[0]))
 235
 236 #define VHOST_NB_XSTATS_TXPORT (sizeof(vhost_txport_stat_strings) / \
 237                                 sizeof(vhost_txport_stat_strings[0]))
 238
 239 static void
 240 vhost_dev_xstats_reset(struct rte_eth_dev *dev)
 241 {
 242         struct vhost_queue *vq = NULL;
 243         unsigned int i = 0;
 244
 245         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 246                 vq = dev->data->rx_queues[i];
 247                 if (!vq)
 248                         continue;
 249                 memset(&vq->stats, 0, sizeof(vq->stats));
 250         }
 251         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 252                 vq = dev->data->tx_queues[i];
 253                 if (!vq)
 254                         continue;
 255                 memset(&vq->stats, 0, sizeof(vq->stats));
 256         }
 257 }
 258
 259 static int
 260 vhost_dev_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
 261                            struct rte_eth_xstat_name *xstats_names,
 262                            unsigned int limit __rte_unused)
 263 {
 264         unsigned int t = 0;
 265         int count = 0;
 266         int nstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
 267
 268         if (!xstats_names)
 269                 return nstats;
 270         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
 271                 snprintf(xstats_names[count].name,
 272                          sizeof(xstats_names[count].name),
 273                          "rx_%s", vhost_rxport_stat_strings[t].name);
 274                 count++;
 275         }
 276         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
 277                 snprintf(xstats_names[count].name,
 278                          sizeof(xstats_names[count].name),
 279                          "tx_%s", vhost_txport_stat_strings[t].name);
 280                 count++;
 281         }
 282         return count;
 283 }
 284
 285 static int
 286 vhost_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 287                      unsigned int n)
 288 {
 289         unsigned int i;
 290         unsigned int t;
 291         unsigned int count = 0;
 292         struct vhost_queue *vq = NULL;
 293         unsigned int nxstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
 294
 295         if (n < nxstats)
 296                 return nxstats;
 297
 298         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 299                 vq = dev->data->rx_queues[i];
 300                 if (!vq)
 301                         continue;
 302                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
 303                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
 304                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
 305         }
 306         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 307                 vq = dev->data->tx_queues[i];
 308                 if (!vq)
 309                         continue;
 310                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
 311                                 + vq->stats.missed_pkts
 312                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
 313                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
 314         }
 315         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
 316                 xstats[count].value = 0;
 317                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 318                         vq = dev->data->rx_queues[i];
 319                         if (!vq)
 320                                 continue;
 321                         xstats[count].value +=
 322                                 *(uint64_t *)(((char *)vq)
 323                                 + vhost_rxport_stat_strings[t].offset);
 324                 }
 325                 count++;
 326         }
 327         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
 328                 xstats[count].value = 0;
 329                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
 330                         vq = dev->data->tx_queues[i];
 331                         if (!vq)
 332                                 continue;
 333                         xstats[count].value +=
 334                                 *(uint64_t *)(((char *)vq)
 335                                 + vhost_txport_stat_strings[t].offset);
 336                 }
 337                 count++;
 338         }
 339         return count;
 340 }
 341
 342 static inline void
 343 vhost_count_multicast_broadcast(struct vhost_queue *vq,
 344                                 struct rte_mbuf *mbuf)
 345 {
 346         struct ether_addr *ea = NULL;
 347         struct vhost_stats *pstats = &vq->stats;
 348
 349         ea = rte_pktmbuf_mtod(mbuf, struct ether_addr *);
 350         if (is_multicast_ether_addr(ea)) {
 351                 if (is_broadcast_ether_addr(ea))
 352                         pstats->xstats[VHOST_BROADCAST_PKT]++;
 353                 else
 354                         pstats->xstats[VHOST_MULTICAST_PKT]++;
 355         }
 356 }
 357
 358 static void
 359 vhost_update_packet_xstats(struct vhost_queue *vq,
 360                            struct rte_mbuf **bufs,
 361                            uint16_t count)
 362 {
 363         uint32_t pkt_len = 0;
 364         uint64_t i = 0;
 365         uint64_t index;
 366         struct vhost_stats *pstats = &vq->stats;
 367
 368         for (i = 0; i < count ; i++) {
 369                 pkt_len = bufs[i]->pkt_len;
 370                 if (pkt_len == 64) {
 371                         pstats->xstats[VHOST_64_PKT]++;
 372                 } else if (pkt_len > 64 && pkt_len < 1024) {
 373                         index = (sizeof(pkt_len) * 8)
 374                                 - __builtin_clz(pkt_len) - 5;
 375                         pstats->xstats[index]++;
 376                 } else {
 377                         if (pkt_len < 64)
 378                                 pstats->xstats[VHOST_UNDERSIZE_PKT]++;
 379                         else if (pkt_len <= 1522)
 380                                 pstats->xstats[VHOST_1024_TO_1522_PKT]++;
 381                         else if (pkt_len > 1522)
 382                                 pstats->xstats[VHOST_1523_TO_MAX_PKT]++;
 383                 }
 384                 vhost_count_multicast_broadcast(vq, bufs[i]);
 385         }
 386 }
 387
 388 static uint16_t
 389 eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 390 {
 391         struct vhost_queue *r = q;
 392         uint16_t i, nb_rx = 0;
 393
 394         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 395                 return 0;
 396
 397         rte_atomic32_set(&r->while_queuing, 1);
 398
 399         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 400                 goto out;
 401
 402         /* Dequeue packets from guest TX queue */
 403         nb_rx = rte_vhost_dequeue_burst(r->vid,
 404                         r->virtqueue_id, r->mb_pool, bufs, nb_bufs);
 405
 406         r->stats.pkts += nb_rx;
 407
 408         for (i = 0; likely(i < nb_rx); i++) {
 409                 bufs[i]->port = r->port;
 410                 r->stats.bytes += bufs[i]->pkt_len;
 411         }
 412
 413         vhost_update_packet_xstats(r, bufs, nb_rx);
 414
 415 out:
 416         rte_atomic32_set(&r->while_queuing, 0);
 417
 418         return nb_rx;
 419 }
 420
 421 static uint16_t
 422 eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 423 {
 424         struct vhost_queue *r = q;
 425         uint16_t i, nb_tx = 0;
 426
 427         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 428                 return 0;
 429
 430         rte_atomic32_set(&r->while_queuing, 1);
 431
 432         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
 433                 goto out;
 434
 435         /* Enqueue packets to guest RX queue */
 436         nb_tx = rte_vhost_enqueue_burst(r->vid,
 437                         r->virtqueue_id, bufs, nb_bufs);
 438
 439         r->stats.pkts += nb_tx;
 440         r->stats.missed_pkts += nb_bufs - nb_tx;
 441
 442         for (i = 0; likely(i < nb_tx); i++)
 443                 r->stats.bytes += bufs[i]->pkt_len;
 444
 445         vhost_update_packet_xstats(r, bufs, nb_tx);
 446
 447         /* According to RFC2863 page42 section ifHCOutMulticastPkts and
 448          * ifHCOutBroadcastPkts, the counters "multicast" and "broadcast"
 449          * are increased when packets are not transmitted successfully.
 450          */
 451         for (i = nb_tx; i < nb_bufs; i++)
 452                 vhost_count_multicast_broadcast(r, bufs[i]);
 453
 454         for (i = 0; likely(i < nb_tx); i++)
 455                 rte_pktmbuf_free(bufs[i]);
 456 out:
 457         rte_atomic32_set(&r->while_queuing, 0);
 458
 459         return nb_tx;
 460 }
 461
 462 static int
 463 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 464 {
 465         return 0;
 466 }
 467
 468 static inline struct internal_list *
 469 find_internal_resource(char *ifname)
 470 {
 471         int found = 0;
 472         struct internal_list *list;
 473         struct pmd_internal *internal;
 474
 475         if (ifname == NULL)
 476                 return NULL;
 477
 478         pthread_mutex_lock(&internal_list_lock);
 479
 480         TAILQ_FOREACH(list, &internal_list, next) {
 481                 internal = list->eth_dev->data->dev_private;
 482                 if (!strcmp(internal->iface_name, ifname)) {
 483                         found = 1;
 484                         break;
 485                 }
 486         }
 487
 488         pthread_mutex_unlock(&internal_list_lock);
 489
 490         if (!found)
 491                 return NULL;
 492
 493         return list;
 494 }
 495
 496 static int
 497 new_device(int vid)
 498 {
 499         struct rte_eth_dev *eth_dev;
 500         struct internal_list *list;
 501         struct pmd_internal *internal;
 502         struct vhost_queue *vq;
 503         unsigned i;
 504         char ifname[PATH_MAX];
 505 #ifdef RTE_LIBRTE_VHOST_NUMA
 506         int newnode;
 507 #endif
 508
 509         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 510         list = find_internal_resource(ifname);
 511         if (list == NULL) {
 512                 RTE_LOG(INFO, PMD, "Invalid device name: %s\n", ifname);
 513                 return -1;
 514         }
 515
 516         eth_dev = list->eth_dev;
 517         internal = eth_dev->data->dev_private;
 518
 519 #ifdef RTE_LIBRTE_VHOST_NUMA
 520         newnode = rte_vhost_get_numa_node(vid);
 521         if (newnode >= 0)
 522                 eth_dev->data->numa_node = newnode;
 523 #endif
 524
 525         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 526                 vq = eth_dev->data->rx_queues[i];
 527                 if (vq == NULL)
 528                         continue;
 529                 vq->vid = vid;
 530                 vq->internal = internal;
 531                 vq->port = eth_dev->data->port_id;
 532         }
 533         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 534                 vq = eth_dev->data->tx_queues[i];
 535                 if (vq == NULL)
 536                         continue;
 537                 vq->vid = vid;
 538                 vq->internal = internal;
 539                 vq->port = eth_dev->data->port_id;
 540         }
 541
 542         for (i = 0; i < rte_vhost_get_queue_num(vid) * VIRTIO_QNUM; i++)
 543                 rte_vhost_enable_guest_notification(vid, i, 0);
 544
 545         eth_dev->data->dev_link.link_status = ETH_LINK_UP;
 546
 547         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 548                 vq = eth_dev->data->rx_queues[i];
 549                 if (vq == NULL)
 550                         continue;
 551                 rte_atomic32_set(&vq->allow_queuing, 1);
 552         }
 553         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 554                 vq = eth_dev->data->tx_queues[i];
 555                 if (vq == NULL)
 556                         continue;
 557                 rte_atomic32_set(&vq->allow_queuing, 1);
 558         }
 559
 560         RTE_LOG(INFO, PMD, "New connection established\n");
 561
 562         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC);
 563
 564         return 0;
 565 }
 566
 567 static void
 568 destroy_device(int vid)
 569 {
 570         struct rte_eth_dev *eth_dev;
 571         struct vhost_queue *vq;
 572         struct internal_list *list;
 573         char ifname[PATH_MAX];
 574         unsigned i;
 575         struct rte_vhost_vring_state *state;
 576
 577         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 578         list = find_internal_resource(ifname);
 579         if (list == NULL) {
 580                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
 581                 return;
 582         }
 583         eth_dev = list->eth_dev;
 584
 585         /* Wait until rx/tx_pkt_burst stops accessing vhost device */
 586         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 587                 vq = eth_dev->data->rx_queues[i];
 588                 if (vq == NULL)
 589                         continue;
 590                 rte_atomic32_set(&vq->allow_queuing, 0);
 591                 while (rte_atomic32_read(&vq->while_queuing))
 592                         rte_pause();
 593         }
 594         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 595                 vq = eth_dev->data->tx_queues[i];
 596                 if (vq == NULL)
 597                         continue;
 598                 rte_atomic32_set(&vq->allow_queuing, 0);
 599                 while (rte_atomic32_read(&vq->while_queuing))
 600                         rte_pause();
 601         }
 602
 603         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
 604
 605         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 606                 vq = eth_dev->data->rx_queues[i];
 607                 if (vq == NULL)
 608                         continue;
 609                 vq->vid = -1;
 610         }
 611         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 612                 vq = eth_dev->data->tx_queues[i];
 613                 if (vq == NULL)
 614                         continue;
 615                 vq->vid = -1;
 616         }
 617
 618         state = vring_states[eth_dev->data->port_id];
 619         rte_spinlock_lock(&state->lock);
 620         for (i = 0; i <= state->max_vring; i++) {
 621                 state->cur[i] = false;
 622                 state->seen[i] = false;
 623         }
 624         state->max_vring = 0;
 625         rte_spinlock_unlock(&state->lock);
 626
 627         RTE_LOG(INFO, PMD, "Connection closed\n");
 628
 629         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC);
 630 }
 631
 632 static int
 633 vring_state_changed(int vid, uint16_t vring, int enable)
 634 {
 635         struct rte_vhost_vring_state *state;
 636         struct rte_eth_dev *eth_dev;
 637         struct internal_list *list;
 638         char ifname[PATH_MAX];
 639
 640         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
 641         list = find_internal_resource(ifname);
 642         if (list == NULL) {
 643                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
 644                 return -1;
 645         }
 646
 647         eth_dev = list->eth_dev;
 648         /* won't be NULL */
 649         state = vring_states[eth_dev->data->port_id];
 650         rte_spinlock_lock(&state->lock);
 651         state->cur[vring] = enable;
 652         state->max_vring = RTE_MAX(vring, state->max_vring);
 653         rte_spinlock_unlock(&state->lock);
 654
 655         RTE_LOG(INFO, PMD, "vring%u is %s\n",
 656                         vring, enable ? "enabled" : "disabled");
 657
 658         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE);
 659
 660         return 0;
 661 }
 662
 663 int
 664 rte_eth_vhost_get_queue_event(uint8_t port_id,
 665                 struct rte_eth_vhost_queue_event *event)
 666 {
 667         struct rte_vhost_vring_state *state;
 668         unsigned int i;
 669         int idx;
 670
 671         if (port_id >= RTE_MAX_ETHPORTS) {
 672                 RTE_LOG(ERR, PMD, "Invalid port id\n");
 673                 return -1;
 674         }
 675
 676         state = vring_states[port_id];
 677         if (!state) {
 678                 RTE_LOG(ERR, PMD, "Unused port\n");
 679                 return -1;
 680         }
 681
 682         rte_spinlock_lock(&state->lock);
 683         for (i = 0; i <= state->max_vring; i++) {
 684                 idx = state->index++ % (state->max_vring + 1);
 685
 686                 if (state->cur[idx] != state->seen[idx]) {
 687                         state->seen[idx] = state->cur[idx];
 688                         event->queue_id = idx / 2;
 689                         event->rx = idx & 1;
 690                         event->enable = state->cur[idx];
 691                         rte_spinlock_unlock(&state->lock);
 692                         return 0;
 693                 }
 694         }
 695         rte_spinlock_unlock(&state->lock);
 696
 697         return -1;
 698 }
 699
 700 int
 701 rte_eth_vhost_get_vid_from_port_id(uint8_t port_id)
 702 {
 703         struct internal_list *list;
 704         struct rte_eth_dev *eth_dev;
 705         struct vhost_queue *vq;
 706         int vid = -1;
 707
 708         if (!rte_eth_dev_is_valid_port(port_id))
 709                 return -1;
 710
 711         pthread_mutex_lock(&internal_list_lock);
 712
 713         TAILQ_FOREACH(list, &internal_list, next) {
 714                 eth_dev = list->eth_dev;
 715                 if (eth_dev->data->port_id == port_id) {
 716                         vq = eth_dev->data->rx_queues[0];
 717                         if (vq) {
 718                                 vid = vq->vid;
 719                         }
 720                         break;
 721                 }
 722         }
 723
 724         pthread_mutex_unlock(&internal_list_lock);
 725
 726         return vid;
 727 }
 728
 729 static void *
 730 vhost_driver_session(void *param __rte_unused)
 731 {
 732         static struct virtio_net_device_ops vhost_ops;
 733
 734         /* set vhost arguments */
 735         vhost_ops.new_device = new_device;
 736         vhost_ops.destroy_device = destroy_device;
 737         vhost_ops.vring_state_changed = vring_state_changed;
 738         if (rte_vhost_driver_callback_register(&vhost_ops) < 0)
 739                 RTE_LOG(ERR, PMD, "Can't register callbacks\n");
 740
 741         /* start event handling */
 742         rte_vhost_driver_session_start();
 743
 744         return NULL;
 745 }
 746
 747 static int
 748 vhost_driver_session_start(void)
 749 {
 750         int ret;
 751
 752         ret = pthread_create(&session_th,
 753                         NULL, vhost_driver_session, NULL);
 754         if (ret)
 755                 RTE_LOG(ERR, PMD, "Can't create a thread\n");
 756
 757         return ret;
 758 }
 759
 760 static void
 761 vhost_driver_session_stop(void)
 762 {
 763         int ret;
 764
 765         ret = pthread_cancel(session_th);
 766         if (ret)
 767                 RTE_LOG(ERR, PMD, "Can't cancel the thread\n");
 768
 769         ret = pthread_join(session_th, NULL);
 770         if (ret)
 771                 RTE_LOG(ERR, PMD, "Can't join the thread\n");
 772 }
 773
 774 static int
 775 eth_dev_start(struct rte_eth_dev *dev)
 776 {
 777         struct pmd_internal *internal = dev->data->dev_private;
 778         int ret = 0;
 779
 780         if (rte_atomic16_cmpset(&internal->once, 0, 1)) {
 781                 ret = rte_vhost_driver_register(internal->iface_name,
 782                                                 internal->flags);
 783                 if (ret)
 784                         return ret;
 785         }
 786
 787         /* We need only one message handling thread */
 788         if (rte_atomic16_add_return(&nb_started_ports, 1) == 1)
 789                 ret = vhost_driver_session_start();
 790
 791         return ret;
 792 }
 793
 794 static void
 795 eth_dev_stop(struct rte_eth_dev *dev)
 796 {
 797         struct pmd_internal *internal = dev->data->dev_private;
 798
 799         if (rte_atomic16_cmpset(&internal->once, 1, 0))
 800                 rte_vhost_driver_unregister(internal->iface_name);
 801
 802         if (rte_atomic16_sub_return(&nb_started_ports, 1) == 0)
 803                 vhost_driver_session_stop();
 804 }
 805
 806 static int
 807 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 808                    uint16_t nb_rx_desc __rte_unused,
 809                    unsigned int socket_id,
 810                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 811                    struct rte_mempool *mb_pool)
 812 {
 813         struct vhost_queue *vq;
 814
 815         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
 816                         RTE_CACHE_LINE_SIZE, socket_id);
 817         if (vq == NULL) {
 818                 RTE_LOG(ERR, PMD, "Failed to allocate memory for rx queue\n");
 819                 return -ENOMEM;
 820         }
 821
 822         vq->mb_pool = mb_pool;
 823         vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
 824         dev->data->rx_queues[rx_queue_id] = vq;
 825
 826         return 0;
 827 }
 828
 829 static int
 830 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 831                    uint16_t nb_tx_desc __rte_unused,
 832                    unsigned int socket_id,
 833                    const struct rte_eth_txconf *tx_conf __rte_unused)
 834 {
 835         struct vhost_queue *vq;
 836
 837         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
 838                         RTE_CACHE_LINE_SIZE, socket_id);
 839         if (vq == NULL) {
 840                 RTE_LOG(ERR, PMD, "Failed to allocate memory for tx queue\n");
 841                 return -ENOMEM;
 842         }
 843
 844         vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
 845         dev->data->tx_queues[tx_queue_id] = vq;
 846
 847         return 0;
 848 }
 849
 850 static void
 851 eth_dev_info(struct rte_eth_dev *dev,
 852              struct rte_eth_dev_info *dev_info)
 853 {
 854         struct pmd_internal *internal;
 855
 856         internal = dev->data->dev_private;
 857         if (internal == NULL) {
 858                 RTE_LOG(ERR, PMD, "Invalid device specified\n");
 859                 return;
 860         }
 861
 862         dev_info->driver_name = drivername;
 863         dev_info->max_mac_addrs = 1;
 864         dev_info->max_rx_pktlen = (uint32_t)-1;
 865         dev_info->max_rx_queues = internal->max_queues;
 866         dev_info->max_tx_queues = internal->max_queues;
 867         dev_info->min_rx_bufsize = 0;
 868 }
 869
 870 static void
 871 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 872 {
 873         unsigned i;
 874         unsigned long rx_total = 0, tx_total = 0, tx_missed_total = 0;
 875         unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
 876         struct vhost_queue *vq;
 877
 878         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 879                         i < dev->data->nb_rx_queues; i++) {
 880                 if (dev->data->rx_queues[i] == NULL)
 881                         continue;
 882                 vq = dev->data->rx_queues[i];
 883                 stats->q_ipackets[i] = vq->stats.pkts;
 884                 rx_total += stats->q_ipackets[i];
 885
 886                 stats->q_ibytes[i] = vq->stats.bytes;
 887                 rx_total_bytes += stats->q_ibytes[i];
 888         }
 889
 890         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 891                         i < dev->data->nb_tx_queues; i++) {
 892                 if (dev->data->tx_queues[i] == NULL)
 893                         continue;
 894                 vq = dev->data->tx_queues[i];
 895                 stats->q_opackets[i] = vq->stats.pkts;
 896                 tx_missed_total += vq->stats.missed_pkts;
 897                 tx_total += stats->q_opackets[i];
 898
 899                 stats->q_obytes[i] = vq->stats.bytes;
 900                 tx_total_bytes += stats->q_obytes[i];
 901         }
 902
 903         stats->ipackets = rx_total;
 904         stats->opackets = tx_total;
 905         stats->oerrors = tx_missed_total;
 906         stats->ibytes = rx_total_bytes;
 907         stats->obytes = tx_total_bytes;
 908 }
 909
 910 static void
 911 eth_stats_reset(struct rte_eth_dev *dev)
 912 {
 913         struct vhost_queue *vq;
 914         unsigned i;
 915
 916         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 917                 if (dev->data->rx_queues[i] == NULL)
 918                         continue;
 919                 vq = dev->data->rx_queues[i];
 920                 vq->stats.pkts = 0;
 921                 vq->stats.bytes = 0;
 922         }
 923         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 924                 if (dev->data->tx_queues[i] == NULL)
 925                         continue;
 926                 vq = dev->data->tx_queues[i];
 927                 vq->stats.pkts = 0;
 928                 vq->stats.bytes = 0;
 929                 vq->stats.missed_pkts = 0;
 930         }
 931 }
 932
 933 static void
 934 eth_queue_release(void *q)
 935 {
 936         rte_free(q);
 937 }
 938
 939 static int
 940 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 941                 int wait_to_complete __rte_unused)
 942 {
 943         return 0;
 944 }
 945
 946 /**
 947  * Disable features in feature_mask. Returns 0 on success.
 948  */
 949 int
 950 rte_eth_vhost_feature_disable(uint64_t feature_mask)
 951 {
 952         return rte_vhost_feature_disable(feature_mask);
 953 }
 954
 955 /**
 956  * Enable features in feature_mask. Returns 0 on success.
 957  */
 958 int
 959 rte_eth_vhost_feature_enable(uint64_t feature_mask)
 960 {
 961         return rte_vhost_feature_enable(feature_mask);
 962 }
 963
 964 /* Returns currently supported vhost features */
 965 uint64_t
 966 rte_eth_vhost_feature_get(void)
 967 {
 968         return rte_vhost_feature_get();
 969 }
 970
 971 static const struct eth_dev_ops ops = {
 972         .dev_start = eth_dev_start,
 973         .dev_stop = eth_dev_stop,
 974         .dev_configure = eth_dev_configure,
 975         .dev_infos_get = eth_dev_info,
 976         .rx_queue_setup = eth_rx_queue_setup,
 977         .tx_queue_setup = eth_tx_queue_setup,
 978         .rx_queue_release = eth_queue_release,
 979         .tx_queue_release = eth_queue_release,
 980         .link_update = eth_link_update,
 981         .stats_get = eth_stats_get,
 982         .stats_reset = eth_stats_reset,
 983         .xstats_reset = vhost_dev_xstats_reset,
 984         .xstats_get = vhost_dev_xstats_get,
 985         .xstats_get_names = vhost_dev_xstats_get_names,
 986 };
 987
 988 static int
 989 eth_dev_vhost_create(const char *name, char *iface_name, int16_t queues,
 990                      const unsigned numa_node, uint64_t flags)
 991 {
 992         struct rte_eth_dev_data *data = NULL;
 993         struct pmd_internal *internal = NULL;
 994         struct rte_eth_dev *eth_dev = NULL;
 995         struct ether_addr *eth_addr = NULL;
 996         struct rte_vhost_vring_state *vring_state = NULL;
 997         struct internal_list *list = NULL;
 998
 999         RTE_LOG(INFO, PMD, "Creating VHOST-USER backend on numa socket %u\n",
1000                 numa_node);
1001
1002         /* now do all data allocation - for eth_dev structure, dummy pci driver
1003          * and internal (private) data
1004          */
1005         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
1006         if (data == NULL)
1007                 goto error;
1008
1009         internal = rte_zmalloc_socket(name, sizeof(*internal), 0, numa_node);
1010         if (internal == NULL)
1011                 goto error;
1012
1013         list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
1014         if (list == NULL)
1015                 goto error;
1016
1017         /* reserve an ethdev entry */
1018         eth_dev = rte_eth_dev_allocate(name);
1019         if (eth_dev == NULL)
1020                 goto error;
1021
1022         eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
1023         if (eth_addr == NULL)
1024                 goto error;
1025         *eth_addr = base_eth_addr;
1026         eth_addr->addr_bytes[5] = eth_dev->data->port_id;
1027
1028         vring_state = rte_zmalloc_socket(name,
1029                         sizeof(*vring_state), 0, numa_node);
1030         if (vring_state == NULL)
1031                 goto error;
1032
1033         TAILQ_INIT(&eth_dev->link_intr_cbs);
1034
1035         /* now put it all together
1036          * - store queue data in internal,
1037          * - store numa_node info in ethdev data
1038          * - point eth_dev_data to internals
1039          * - and point eth_dev structure to new eth_dev_data structure
1040          */
1041         internal->dev_name = strdup(name);
1042         if (internal->dev_name == NULL)
1043                 goto error;
1044         internal->iface_name = strdup(iface_name);
1045         if (internal->iface_name == NULL)
1046                 goto error;
1047         internal->flags = flags;
1048
1049         list->eth_dev = eth_dev;
1050         pthread_mutex_lock(&internal_list_lock);
1051         TAILQ_INSERT_TAIL(&internal_list, list, next);
1052         pthread_mutex_unlock(&internal_list_lock);
1053
1054         rte_spinlock_init(&vring_state->lock);
1055         vring_states[eth_dev->data->port_id] = vring_state;
1056
1057         data->dev_private = internal;
1058         data->port_id = eth_dev->data->port_id;
1059         memmove(data->name, eth_dev->data->name, sizeof(data->name));
1060         data->nb_rx_queues = queues;
1061         data->nb_tx_queues = queues;
1062         internal->max_queues = queues;
1063         data->dev_link = pmd_link;
1064         data->mac_addrs = eth_addr;
1065
1066         /* We'll replace the 'data' originally allocated by eth_dev. So the
1067          * vhost PMD resources won't be shared between multi processes.
1068          */
1069         eth_dev->data = data;
1070         eth_dev->dev_ops = &ops;
1071         eth_dev->driver = NULL;
1072         data->dev_flags =
1073                 RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
1074         data->kdrv = RTE_KDRV_NONE;
1075         data->drv_name = internal->dev_name;
1076         data->numa_node = numa_node;
1077
1078         /* finally assign rx and tx ops */
1079         eth_dev->rx_pkt_burst = eth_vhost_rx;
1080         eth_dev->tx_pkt_burst = eth_vhost_tx;
1081
1082         return data->port_id;
1083
1084 error:
1085         if (internal)
1086                 free(internal->dev_name);
1087         rte_free(vring_state);
1088         rte_free(eth_addr);
1089         if (eth_dev)
1090                 rte_eth_dev_release_port(eth_dev);
1091         rte_free(internal);
1092         rte_free(list);
1093         rte_free(data);
1094
1095         return -1;
1096 }
1097
1098 static inline int
1099 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
1100 {
1101         const char **iface_name = extra_args;
1102
1103         if (value == NULL)
1104                 return -1;
1105
1106         *iface_name = value;
1107
1108         return 0;
1109 }
1110
1111 static inline int
1112 open_int(const char *key __rte_unused, const char *value, void *extra_args)
1113 {
1114         uint16_t *n = extra_args;
1115
1116         if (value == NULL || extra_args == NULL)
1117                 return -EINVAL;
1118
1119         *n = (uint16_t)strtoul(value, NULL, 0);
1120         if (*n == USHRT_MAX && errno == ERANGE)
1121                 return -1;
1122
1123         return 0;
1124 }
1125
1126 static int
1127 rte_pmd_vhost_devinit(const char *name, const char *params)
1128 {
1129         struct rte_kvargs *kvlist = NULL;
1130         int ret = 0;
1131         char *iface_name;
1132         uint16_t queues;
1133         uint64_t flags = 0;
1134         int client_mode = 0;
1135
1136         RTE_LOG(INFO, PMD, "Initializing pmd_vhost for %s\n", name);
1137
1138         kvlist = rte_kvargs_parse(params, valid_arguments);
1139         if (kvlist == NULL)
1140                 return -1;
1141
1142         if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
1143                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
1144                                          &open_iface, &iface_name);
1145                 if (ret < 0)
1146                         goto out_free;
1147         } else {
1148                 ret = -1;
1149                 goto out_free;
1150         }
1151
1152         if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
1153                 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
1154                                          &open_int, &queues);
1155                 if (ret < 0 || queues > RTE_MAX_QUEUES_PER_PORT)
1156                         goto out_free;
1157
1158         } else
1159                 queues = 1;
1160
1161         if (rte_kvargs_count(kvlist, ETH_VHOST_CLIENT_ARG) == 1) {
1162                 ret = rte_kvargs_process(kvlist, ETH_VHOST_CLIENT_ARG,
1163                                          &open_int, &client_mode);
1164                 if (ret < 0)
1165                         goto out_free;
1166
1167                 if (client_mode)
1168                         flags |= RTE_VHOST_USER_CLIENT;
1169         }
1170
1171         eth_dev_vhost_create(name, iface_name, queues, rte_socket_id(), flags);
1172
1173 out_free:
1174         rte_kvargs_free(kvlist);
1175         return ret;
1176 }
1177
1178 static int
1179 rte_pmd_vhost_devuninit(const char *name)
1180 {
1181         struct rte_eth_dev *eth_dev = NULL;
1182         struct pmd_internal *internal;
1183         struct internal_list *list;
1184         unsigned int i;
1185
1186         RTE_LOG(INFO, PMD, "Un-Initializing pmd_vhost for %s\n", name);
1187
1188         /* find an ethdev entry */
1189         eth_dev = rte_eth_dev_allocated(name);
1190         if (eth_dev == NULL)
1191                 return -ENODEV;
1192
1193         internal = eth_dev->data->dev_private;
1194         if (internal == NULL)
1195                 return -ENODEV;
1196
1197         list = find_internal_resource(internal->iface_name);
1198         if (list == NULL)
1199                 return -ENODEV;
1200
1201         pthread_mutex_lock(&internal_list_lock);
1202         TAILQ_REMOVE(&internal_list, list, next);
1203         pthread_mutex_unlock(&internal_list_lock);
1204         rte_free(list);
1205
1206         eth_dev_stop(eth_dev);
1207
1208         rte_free(vring_states[eth_dev->data->port_id]);
1209         vring_states[eth_dev->data->port_id] = NULL;
1210
1211         free(internal->dev_name);
1212         free(internal->iface_name);
1213
1214         for (i = 0; i < eth_dev->data->nb_rx_queues; i++)
1215                 rte_free(eth_dev->data->rx_queues[i]);
1216         for (i = 0; i < eth_dev->data->nb_tx_queues; i++)
1217                 rte_free(eth_dev->data->tx_queues[i]);
1218
1219         rte_free(eth_dev->data->mac_addrs);
1220         rte_free(eth_dev->data);
1221         rte_free(internal);
1222
1223         rte_eth_dev_release_port(eth_dev);
1224
1225         return 0;
1226 }
1227
1228 static struct rte_vdev_driver pmd_vhost_drv = {
1229         .init = rte_pmd_vhost_devinit,
1230         .uninit = rte_pmd_vhost_devuninit,
1231 };
1232
1233 DRIVER_REGISTER_VDEV(net_vhost, pmd_vhost_drv);
1234 DRIVER_REGISTER_PARAM_STRING(net_vhost,
1235         "iface=<ifc> "
1236         "queues=<int>");