lib/librte_vhost/vhost.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2017 Intel Corporation
   3  */
   4
   5 #include <linux/vhost.h>
   6 #include <linux/virtio_net.h>
   7 #include <stddef.h>
   8 #include <stdint.h>
   9 #include <stdlib.h>
  10 #ifdef RTE_LIBRTE_VHOST_NUMA
  11 #include <numa.h>
  12 #include <numaif.h>
  13 #endif
  14
  15 #include <rte_errno.h>
  16 #include <rte_ethdev.h>
  17 #include <rte_log.h>
  18 #include <rte_string_fns.h>
  19 #include <rte_memory.h>
  20 #include <rte_malloc.h>
  21 #include <rte_vhost.h>
  22 #include <rte_rwlock.h>
  23
  24 #include "iotlb.h"
  25 #include "vhost.h"
  26 #include "vhost_user.h"
  27
  28 struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
  29
  30 /* Called with iotlb_lock read-locked */
  31 uint64_t
  32 __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
  33                     uint64_t iova, uint64_t *size, uint8_t perm)
  34 {
  35         uint64_t vva, tmp_size;
  36
  37         if (unlikely(!*size))
  38                 return 0;
  39
  40         tmp_size = *size;
  41
  42         vva = vhost_user_iotlb_cache_find(vq, iova, &tmp_size, perm);
  43         if (tmp_size == *size)
  44                 return vva;
  45
  46         iova += tmp_size;
  47
  48         if (!vhost_user_iotlb_pending_miss(vq, iova, perm)) {
  49                 /*
  50                  * iotlb_lock is read-locked for a full burst,
  51                  * but it only protects the iotlb cache.
  52                  * In case of IOTLB miss, we might block on the socket,
  53                  * which could cause a deadlock with QEMU if an IOTLB update
  54                  * is being handled. We can safely unlock here to avoid it.
  55                  */
  56                 vhost_user_iotlb_rd_unlock(vq);
  57
  58                 vhost_user_iotlb_pending_insert(vq, iova, perm);
  59                 if (vhost_user_iotlb_miss(dev, iova, perm)) {
  60                         RTE_LOG(ERR, VHOST_CONFIG,
  61                                 "IOTLB miss req failed for IOVA 0x%" PRIx64 "\n",
  62                                 iova);
  63                         vhost_user_iotlb_pending_remove(vq, iova, 1, perm);
  64                 }
  65
  66                 vhost_user_iotlb_rd_lock(vq);
  67         }
  68
  69         return 0;
  70 }
  71
  72 #define VHOST_LOG_PAGE  4096
  73
  74 /*
  75  * Atomically set a bit in memory.
  76  */
  77 static __rte_always_inline void
  78 vhost_set_bit(unsigned int nr, volatile uint8_t *addr)
  79 {
  80 #if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION < 70100)
  81         /*
  82          * __sync_ built-ins are deprecated, but __atomic_ ones
  83          * are sub-optimized in older GCC versions.
  84          */
  85         __sync_fetch_and_or_1(addr, (1U << nr));
  86 #else
  87         __atomic_fetch_or(addr, (1U << nr), __ATOMIC_RELAXED);
  88 #endif
  89 }
  90
  91 static __rte_always_inline void
  92 vhost_log_page(uint8_t *log_base, uint64_t page)
  93 {
  94         vhost_set_bit(page % 8, &log_base[page / 8]);
  95 }
  96
  97 void
  98 __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
  99 {
 100         uint64_t page;
 101
 102         if (unlikely(!dev->log_base || !len))
 103                 return;
 104
 105         if (unlikely(dev->log_size <= ((addr + len - 1) / VHOST_LOG_PAGE / 8)))
 106                 return;
 107
 108         /* To make sure guest memory updates are committed before logging */
 109         rte_smp_wmb();
 110
 111         page = addr / VHOST_LOG_PAGE;
 112         while (page * VHOST_LOG_PAGE < addr + len) {
 113                 vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
 114                 page += 1;
 115         }
 116 }
 117
 118 void
 119 __vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
 120 {
 121         unsigned long *log_base;
 122         int i;
 123
 124         if (unlikely(!dev->log_base))
 125                 return;
 126
 127         rte_smp_wmb();
 128
 129         log_base = (unsigned long *)(uintptr_t)dev->log_base;
 130
 131         for (i = 0; i < vq->log_cache_nb_elem; i++) {
 132                 struct log_cache_entry *elem = vq->log_cache + i;
 133
 134 #if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION < 70100)
 135                 /*
 136                  * '__sync' builtins are deprecated, but '__atomic' ones
 137                  * are sub-optimized in older GCC versions.
 138                  */
 139                 __sync_fetch_and_or(log_base + elem->offset, elem->val);
 140 #else
 141                 __atomic_fetch_or(log_base + elem->offset, elem->val,
 142                                 __ATOMIC_RELAXED);
 143 #endif
 144         }
 145
 146         rte_smp_wmb();
 147
 148         vq->log_cache_nb_elem = 0;
 149 }
 150
 151 static __rte_always_inline void
 152 vhost_log_cache_page(struct virtio_net *dev, struct vhost_virtqueue *vq,
 153                         uint64_t page)
 154 {
 155         uint32_t bit_nr = page % (sizeof(unsigned long) << 3);
 156         uint32_t offset = page / (sizeof(unsigned long) << 3);
 157         int i;
 158
 159         for (i = 0; i < vq->log_cache_nb_elem; i++) {
 160                 struct log_cache_entry *elem = vq->log_cache + i;
 161
 162                 if (elem->offset == offset) {
 163                         elem->val |= (1UL << bit_nr);
 164                         return;
 165                 }
 166         }
 167
 168         if (unlikely(i >= VHOST_LOG_CACHE_NR)) {
 169                 /*
 170                  * No more room for a new log cache entry,
 171                  * so write the dirty log map directly.
 172                  */
 173                 rte_smp_wmb();
 174                 vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
 175
 176                 return;
 177         }
 178
 179         vq->log_cache[i].offset = offset;
 180         vq->log_cache[i].val = (1UL << bit_nr);
 181         vq->log_cache_nb_elem++;
 182 }
 183
 184 void
 185 __vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
 186                         uint64_t addr, uint64_t len)
 187 {
 188         uint64_t page;
 189
 190         if (unlikely(!dev->log_base || !len))
 191                 return;
 192
 193         if (unlikely(dev->log_size <= ((addr + len - 1) / VHOST_LOG_PAGE / 8)))
 194                 return;
 195
 196         page = addr / VHOST_LOG_PAGE;
 197         while (page * VHOST_LOG_PAGE < addr + len) {
 198                 vhost_log_cache_page(dev, vq, page);
 199                 page += 1;
 200         }
 201 }
 202
 203 void *
 204 vhost_alloc_copy_ind_table(struct virtio_net *dev, struct vhost_virtqueue *vq,
 205                 uint64_t desc_addr, uint64_t desc_len)
 206 {
 207         void *idesc;
 208         uint64_t src, dst;
 209         uint64_t len, remain = desc_len;
 210
 211         idesc = rte_malloc(__func__, desc_len, 0);
 212         if (unlikely(!idesc))
 213                 return NULL;
 214
 215         dst = (uint64_t)(uintptr_t)idesc;
 216
 217         while (remain) {
 218                 len = remain;
 219                 src = vhost_iova_to_vva(dev, vq, desc_addr, &len,
 220                                 VHOST_ACCESS_RO);
 221                 if (unlikely(!src || !len)) {
 222                         rte_free(idesc);
 223                         return NULL;
 224                 }
 225
 226                 rte_memcpy((void *)(uintptr_t)dst, (void *)(uintptr_t)src, len);
 227
 228                 remain -= len;
 229                 dst += len;
 230                 desc_addr += len;
 231         }
 232
 233         return idesc;
 234 }
 235
 236 void
 237 cleanup_vq(struct vhost_virtqueue *vq, int destroy)
 238 {
 239         if ((vq->callfd >= 0) && (destroy != 0))
 240                 close(vq->callfd);
 241         if (vq->kickfd >= 0)
 242                 close(vq->kickfd);
 243 }
 244
 245 /*
 246  * Unmap any memory, close any file descriptors and
 247  * free any memory owned by a device.
 248  */
 249 void
 250 cleanup_device(struct virtio_net *dev, int destroy)
 251 {
 252         uint32_t i;
 253
 254         vhost_backend_cleanup(dev);
 255
 256         for (i = 0; i < dev->nr_vring; i++)
 257                 cleanup_vq(dev->virtqueue[i], destroy);
 258 }
 259
 260 void
 261 free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq)
 262 {
 263         if (vq_is_packed(dev))
 264                 rte_free(vq->shadow_used_packed);
 265         else
 266                 rte_free(vq->shadow_used_split);
 267         rte_free(vq->batch_copy_elems);
 268         rte_mempool_free(vq->iotlb_pool);
 269         rte_free(vq);
 270 }
 271
 272 /*
 273  * Release virtqueues and device memory.
 274  */
 275 static void
 276 free_device(struct virtio_net *dev)
 277 {
 278         uint32_t i;
 279
 280         for (i = 0; i < dev->nr_vring; i++)
 281                 free_vq(dev, dev->virtqueue[i]);
 282
 283         rte_free(dev);
 284 }
 285
 286 static int
 287 vring_translate_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 288 {
 289         uint64_t req_size, size;
 290
 291         req_size = sizeof(struct vring_desc) * vq->size;
 292         size = req_size;
 293         vq->desc = (struct vring_desc *)(uintptr_t)vhost_iova_to_vva(dev, vq,
 294                                                 vq->ring_addrs.desc_user_addr,
 295                                                 &size, VHOST_ACCESS_RW);
 296         if (!vq->desc || size != req_size)
 297                 return -1;
 298
 299         req_size = sizeof(struct vring_avail);
 300         req_size += sizeof(uint16_t) * vq->size;
 301         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 302                 req_size += sizeof(uint16_t);
 303         size = req_size;
 304         vq->avail = (struct vring_avail *)(uintptr_t)vhost_iova_to_vva(dev, vq,
 305                                                 vq->ring_addrs.avail_user_addr,
 306                                                 &size, VHOST_ACCESS_RW);
 307         if (!vq->avail || size != req_size)
 308                 return -1;
 309
 310         req_size = sizeof(struct vring_used);
 311         req_size += sizeof(struct vring_used_elem) * vq->size;
 312         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 313                 req_size += sizeof(uint16_t);
 314         size = req_size;
 315         vq->used = (struct vring_used *)(uintptr_t)vhost_iova_to_vva(dev, vq,
 316                                                 vq->ring_addrs.used_user_addr,
 317                                                 &size, VHOST_ACCESS_RW);
 318         if (!vq->used || size != req_size)
 319                 return -1;
 320
 321         return 0;
 322 }
 323
 324 static int
 325 vring_translate_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
 326 {
 327         uint64_t req_size, size;
 328
 329         req_size = sizeof(struct vring_packed_desc) * vq->size;
 330         size = req_size;
 331         vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
 332                 vhost_iova_to_vva(dev, vq, vq->ring_addrs.desc_user_addr,
 333                                 &size, VHOST_ACCESS_RW);
 334         if (!vq->desc_packed || size != req_size)
 335                 return -1;
 336
 337         req_size = sizeof(struct vring_packed_desc_event);
 338         size = req_size;
 339         vq->driver_event = (struct vring_packed_desc_event *)(uintptr_t)
 340                 vhost_iova_to_vva(dev, vq, vq->ring_addrs.avail_user_addr,
 341                                 &size, VHOST_ACCESS_RW);
 342         if (!vq->driver_event || size != req_size)
 343                 return -1;
 344
 345         req_size = sizeof(struct vring_packed_desc_event);
 346         size = req_size;
 347         vq->device_event = (struct vring_packed_desc_event *)(uintptr_t)
 348                 vhost_iova_to_vva(dev, vq, vq->ring_addrs.used_user_addr,
 349                                 &size, VHOST_ACCESS_RW);
 350         if (!vq->device_event || size != req_size)
 351                 return -1;
 352
 353         return 0;
 354 }
 355
 356 int
 357 vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq)
 358 {
 359
 360         if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 361                 return -1;
 362
 363         if (vq_is_packed(dev)) {
 364                 if (vring_translate_packed(dev, vq) < 0)
 365                         return -1;
 366         } else {
 367                 if (vring_translate_split(dev, vq) < 0)
 368                         return -1;
 369         }
 370         vq->access_ok = 1;
 371
 372         return 0;
 373 }
 374
 375 void
 376 vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq)
 377 {
 378         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 379                 vhost_user_iotlb_wr_lock(vq);
 380
 381         vq->access_ok = 0;
 382         vq->desc = NULL;
 383         vq->avail = NULL;
 384         vq->used = NULL;
 385
 386         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 387                 vhost_user_iotlb_wr_unlock(vq);
 388 }
 389
 390 static void
 391 init_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 392 {
 393         struct vhost_virtqueue *vq;
 394
 395         if (vring_idx >= VHOST_MAX_VRING) {
 396                 RTE_LOG(ERR, VHOST_CONFIG,
 397                                 "Failed not init vring, out of bound (%d)\n",
 398                                 vring_idx);
 399                 return;
 400         }
 401
 402         vq = dev->virtqueue[vring_idx];
 403
 404         memset(vq, 0, sizeof(struct vhost_virtqueue));
 405
 406         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 407         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
 408
 409         vhost_user_iotlb_init(dev, vring_idx);
 410         /* Backends are set to -1 indicating an inactive device. */
 411         vq->backend = -1;
 412
 413         TAILQ_INIT(&vq->zmbuf_list);
 414 }
 415
 416 static void
 417 reset_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 418 {
 419         struct vhost_virtqueue *vq;
 420         int callfd;
 421
 422         if (vring_idx >= VHOST_MAX_VRING) {
 423                 RTE_LOG(ERR, VHOST_CONFIG,
 424                                 "Failed not init vring, out of bound (%d)\n",
 425                                 vring_idx);
 426                 return;
 427         }
 428
 429         vq = dev->virtqueue[vring_idx];
 430         callfd = vq->callfd;
 431         init_vring_queue(dev, vring_idx);
 432         vq->callfd = callfd;
 433 }
 434
 435 int
 436 alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 437 {
 438         struct vhost_virtqueue *vq;
 439
 440         vq = rte_malloc(NULL, sizeof(struct vhost_virtqueue), 0);
 441         if (vq == NULL) {
 442                 RTE_LOG(ERR, VHOST_CONFIG,
 443                         "Failed to allocate memory for vring:%u.\n", vring_idx);
 444                 return -1;
 445         }
 446
 447         dev->virtqueue[vring_idx] = vq;
 448         init_vring_queue(dev, vring_idx);
 449         rte_spinlock_init(&vq->access_lock);
 450         vq->avail_wrap_counter = 1;
 451         vq->used_wrap_counter = 1;
 452         vq->signalled_used_valid = false;
 453
 454         dev->nr_vring += 1;
 455
 456         return 0;
 457 }
 458
 459 /*
 460  * Reset some variables in device structure, while keeping few
 461  * others untouched, such as vid, ifname, nr_vring: they
 462  * should be same unless the device is removed.
 463  */
 464 void
 465 reset_device(struct virtio_net *dev)
 466 {
 467         uint32_t i;
 468
 469         dev->features = 0;
 470         dev->protocol_features = 0;
 471         dev->flags &= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 472
 473         for (i = 0; i < dev->nr_vring; i++)
 474                 reset_vring_queue(dev, i);
 475 }
 476
 477 /*
 478  * Invoked when there is a new vhost-user connection established (when
 479  * there is a new virtio device being attached).
 480  */
 481 int
 482 vhost_new_device(void)
 483 {
 484         struct virtio_net *dev;
 485         int i;
 486
 487         for (i = 0; i < MAX_VHOST_DEVICE; i++) {
 488                 if (vhost_devices[i] == NULL)
 489                         break;
 490         }
 491
 492         if (i == MAX_VHOST_DEVICE) {
 493                 RTE_LOG(ERR, VHOST_CONFIG,
 494                         "Failed to find a free slot for new device.\n");
 495                 return -1;
 496         }
 497
 498         dev = rte_zmalloc(NULL, sizeof(struct virtio_net), 0);
 499         if (dev == NULL) {
 500                 RTE_LOG(ERR, VHOST_CONFIG,
 501                         "Failed to allocate memory for new dev.\n");
 502                 return -1;
 503         }
 504
 505         vhost_devices[i] = dev;
 506         dev->vid = i;
 507         dev->flags = VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 508         dev->slave_req_fd = -1;
 509         dev->vdpa_dev_id = -1;
 510         dev->postcopy_ufd = -1;
 511         rte_spinlock_init(&dev->slave_req_lock);
 512
 513         return i;
 514 }
 515
 516 void
 517 vhost_destroy_device_notify(struct virtio_net *dev)
 518 {
 519         struct rte_vdpa_device *vdpa_dev;
 520         int did;
 521
 522         if (dev->flags & VIRTIO_DEV_RUNNING) {
 523                 did = dev->vdpa_dev_id;
 524                 vdpa_dev = rte_vdpa_get_device(did);
 525                 if (vdpa_dev && vdpa_dev->ops->dev_close)
 526                         vdpa_dev->ops->dev_close(dev->vid);
 527                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 528                 dev->notify_ops->destroy_device(dev->vid);
 529         }
 530 }
 531
 532 /*
 533  * Invoked when there is the vhost-user connection is broken (when
 534  * the virtio device is being detached).
 535  */
 536 void
 537 vhost_destroy_device(int vid)
 538 {
 539         struct virtio_net *dev = get_device(vid);
 540
 541         if (dev == NULL)
 542                 return;
 543
 544         vhost_destroy_device_notify(dev);
 545
 546         cleanup_device(dev, 1);
 547         free_device(dev);
 548
 549         vhost_devices[vid] = NULL;
 550 }
 551
 552 void
 553 vhost_attach_vdpa_device(int vid, int did)
 554 {
 555         struct virtio_net *dev = get_device(vid);
 556
 557         if (dev == NULL)
 558                 return;
 559
 560         if (rte_vdpa_get_device(did) == NULL)
 561                 return;
 562
 563         dev->vdpa_dev_id = did;
 564 }
 565
 566 void
 567 vhost_set_ifname(int vid, const char *if_name, unsigned int if_len)
 568 {
 569         struct virtio_net *dev;
 570         unsigned int len;
 571
 572         dev = get_device(vid);
 573         if (dev == NULL)
 574                 return;
 575
 576         len = if_len > sizeof(dev->ifname) ?
 577                 sizeof(dev->ifname) : if_len;
 578
 579         strncpy(dev->ifname, if_name, len);
 580         dev->ifname[sizeof(dev->ifname) - 1] = '\0';
 581 }
 582
 583 void
 584 vhost_enable_dequeue_zero_copy(int vid)
 585 {
 586         struct virtio_net *dev = get_device(vid);
 587
 588         if (dev == NULL)
 589                 return;
 590
 591         dev->dequeue_zero_copy = 1;
 592 }
 593
 594 void
 595 vhost_set_builtin_virtio_net(int vid, bool enable)
 596 {
 597         struct virtio_net *dev = get_device(vid);
 598
 599         if (dev == NULL)
 600                 return;
 601
 602         if (enable)
 603                 dev->flags |= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 604         else
 605                 dev->flags &= ~VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 606 }
 607
 608 int
 609 rte_vhost_get_mtu(int vid, uint16_t *mtu)
 610 {
 611         struct virtio_net *dev = get_device(vid);
 612
 613         if (dev == NULL || mtu == NULL)
 614                 return -ENODEV;
 615
 616         if (!(dev->flags & VIRTIO_DEV_READY))
 617                 return -EAGAIN;
 618
 619         if (!(dev->features & (1ULL << VIRTIO_NET_F_MTU)))
 620                 return -ENOTSUP;
 621
 622         *mtu = dev->mtu;
 623
 624         return 0;
 625 }
 626
 627 int
 628 rte_vhost_get_numa_node(int vid)
 629 {
 630 #ifdef RTE_LIBRTE_VHOST_NUMA
 631         struct virtio_net *dev = get_device(vid);
 632         int numa_node;
 633         int ret;
 634
 635         if (dev == NULL || numa_available() != 0)
 636                 return -1;
 637
 638         ret = get_mempolicy(&numa_node, NULL, 0, dev,
 639                             MPOL_F_NODE | MPOL_F_ADDR);
 640         if (ret < 0) {
 641                 RTE_LOG(ERR, VHOST_CONFIG,
 642                         "(%d) failed to query numa node: %s\n",
 643                         vid, rte_strerror(errno));
 644                 return -1;
 645         }
 646
 647         return numa_node;
 648 #else
 649         RTE_SET_USED(vid);
 650         return -1;
 651 #endif
 652 }
 653
 654 uint32_t
 655 rte_vhost_get_queue_num(int vid)
 656 {
 657         struct virtio_net *dev = get_device(vid);
 658
 659         if (dev == NULL)
 660                 return 0;
 661
 662         return dev->nr_vring / 2;
 663 }
 664
 665 uint16_t
 666 rte_vhost_get_vring_num(int vid)
 667 {
 668         struct virtio_net *dev = get_device(vid);
 669
 670         if (dev == NULL)
 671                 return 0;
 672
 673         return dev->nr_vring;
 674 }
 675
 676 int
 677 rte_vhost_get_ifname(int vid, char *buf, size_t len)
 678 {
 679         struct virtio_net *dev = get_device(vid);
 680
 681         if (dev == NULL || buf == NULL)
 682                 return -1;
 683
 684         len = RTE_MIN(len, sizeof(dev->ifname));
 685
 686         strncpy(buf, dev->ifname, len);
 687         buf[len - 1] = '\0';
 688
 689         return 0;
 690 }
 691
 692 int
 693 rte_vhost_get_negotiated_features(int vid, uint64_t *features)
 694 {
 695         struct virtio_net *dev;
 696
 697         dev = get_device(vid);
 698         if (dev == NULL || features == NULL)
 699                 return -1;
 700
 701         *features = dev->features;
 702         return 0;
 703 }
 704
 705 int
 706 rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem)
 707 {
 708         struct virtio_net *dev;
 709         struct rte_vhost_memory *m;
 710         size_t size;
 711
 712         dev = get_device(vid);
 713         if (dev == NULL || mem == NULL)
 714                 return -1;
 715
 716         size = dev->mem->nregions * sizeof(struct rte_vhost_mem_region);
 717         m = malloc(sizeof(struct rte_vhost_memory) + size);
 718         if (!m)
 719                 return -1;
 720
 721         m->nregions = dev->mem->nregions;
 722         memcpy(m->regions, dev->mem->regions, size);
 723         *mem = m;
 724
 725         return 0;
 726 }
 727
 728 int
 729 rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
 730                           struct rte_vhost_vring *vring)
 731 {
 732         struct virtio_net *dev;
 733         struct vhost_virtqueue *vq;
 734
 735         dev = get_device(vid);
 736         if (dev == NULL || vring == NULL)
 737                 return -1;
 738
 739         if (vring_idx >= VHOST_MAX_VRING)
 740                 return -1;
 741
 742         vq = dev->virtqueue[vring_idx];
 743         if (!vq)
 744                 return -1;
 745
 746         vring->desc  = vq->desc;
 747         vring->avail = vq->avail;
 748         vring->used  = vq->used;
 749         vring->log_guest_addr  = vq->log_guest_addr;
 750
 751         vring->callfd  = vq->callfd;
 752         vring->kickfd  = vq->kickfd;
 753         vring->size    = vq->size;
 754
 755         return 0;
 756 }
 757
 758 int
 759 rte_vhost_vring_call(int vid, uint16_t vring_idx)
 760 {
 761         struct virtio_net *dev;
 762         struct vhost_virtqueue *vq;
 763
 764         dev = get_device(vid);
 765         if (!dev)
 766                 return -1;
 767
 768         if (vring_idx >= VHOST_MAX_VRING)
 769                 return -1;
 770
 771         vq = dev->virtqueue[vring_idx];
 772         if (!vq)
 773                 return -1;
 774
 775         if (vq_is_packed(dev))
 776                 vhost_vring_call_packed(dev, vq);
 777         else
 778                 vhost_vring_call_split(dev, vq);
 779
 780         return 0;
 781 }
 782
 783 uint16_t
 784 rte_vhost_avail_entries(int vid, uint16_t queue_id)
 785 {
 786         struct virtio_net *dev;
 787         struct vhost_virtqueue *vq;
 788         uint16_t ret = 0;
 789
 790         dev = get_device(vid);
 791         if (!dev)
 792                 return 0;
 793
 794         vq = dev->virtqueue[queue_id];
 795
 796         rte_spinlock_lock(&vq->access_lock);
 797
 798         if (unlikely(!vq->enabled || vq->avail == NULL))
 799                 goto out;
 800
 801         ret = *(volatile uint16_t *)&vq->avail->idx - vq->last_used_idx;
 802
 803 out:
 804         rte_spinlock_unlock(&vq->access_lock);
 805         return ret;
 806 }
 807
 808 static inline int
 809 vhost_enable_notify_split(struct virtio_net *dev,
 810                 struct vhost_virtqueue *vq, int enable)
 811 {
 812         if (vq->used == NULL)
 813                 return -1;
 814
 815         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
 816                 if (enable)
 817                         vq->used->flags &= ~VRING_USED_F_NO_NOTIFY;
 818                 else
 819                         vq->used->flags |= VRING_USED_F_NO_NOTIFY;
 820         } else {
 821                 if (enable)
 822                         vhost_avail_event(vq) = vq->last_avail_idx;
 823         }
 824         return 0;
 825 }
 826
 827 static inline int
 828 vhost_enable_notify_packed(struct virtio_net *dev,
 829                 struct vhost_virtqueue *vq, int enable)
 830 {
 831         uint16_t flags;
 832
 833         if (vq->device_event == NULL)
 834                 return -1;
 835
 836         if (!enable) {
 837                 vq->device_event->flags = VRING_EVENT_F_DISABLE;
 838                 return 0;
 839         }
 840
 841         flags = VRING_EVENT_F_ENABLE;
 842         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
 843                 flags = VRING_EVENT_F_DESC;
 844                 vq->device_event->off_wrap = vq->last_avail_idx |
 845                         vq->avail_wrap_counter << 15;
 846         }
 847
 848         rte_smp_wmb();
 849
 850         vq->device_event->flags = flags;
 851         return 0;
 852 }
 853
 854 int
 855 rte_vhost_enable_guest_notification(int vid, uint16_t queue_id, int enable)
 856 {
 857         struct virtio_net *dev = get_device(vid);
 858         struct vhost_virtqueue *vq;
 859         int ret;
 860
 861         if (!dev)
 862                 return -1;
 863
 864         vq = dev->virtqueue[queue_id];
 865
 866         rte_spinlock_lock(&vq->access_lock);
 867
 868         if (vq_is_packed(dev))
 869                 ret = vhost_enable_notify_packed(dev, vq, enable);
 870         else
 871                 ret = vhost_enable_notify_split(dev, vq, enable);
 872
 873         rte_spinlock_unlock(&vq->access_lock);
 874
 875         return ret;
 876 }
 877
 878 void
 879 rte_vhost_log_write(int vid, uint64_t addr, uint64_t len)
 880 {
 881         struct virtio_net *dev = get_device(vid);
 882
 883         if (dev == NULL)
 884                 return;
 885
 886         vhost_log_write(dev, addr, len);
 887 }
 888
 889 void
 890 rte_vhost_log_used_vring(int vid, uint16_t vring_idx,
 891                          uint64_t offset, uint64_t len)
 892 {
 893         struct virtio_net *dev;
 894         struct vhost_virtqueue *vq;
 895
 896         dev = get_device(vid);
 897         if (dev == NULL)
 898                 return;
 899
 900         if (vring_idx >= VHOST_MAX_VRING)
 901                 return;
 902         vq = dev->virtqueue[vring_idx];
 903         if (!vq)
 904                 return;
 905
 906         vhost_log_used_vring(dev, vq, offset, len);
 907 }
 908
 909 uint32_t
 910 rte_vhost_rx_queue_count(int vid, uint16_t qid)
 911 {
 912         struct virtio_net *dev;
 913         struct vhost_virtqueue *vq;
 914         uint32_t ret = 0;
 915
 916         dev = get_device(vid);
 917         if (dev == NULL)
 918                 return 0;
 919
 920         if (unlikely(qid >= dev->nr_vring || (qid & 1) == 0)) {
 921                 RTE_LOG(ERR, VHOST_DATA, "(%d) %s: invalid virtqueue idx %d.\n",
 922                         dev->vid, __func__, qid);
 923                 return 0;
 924         }
 925
 926         vq = dev->virtqueue[qid];
 927         if (vq == NULL)
 928                 return 0;
 929
 930         rte_spinlock_lock(&vq->access_lock);
 931
 932         if (unlikely(vq->enabled == 0 || vq->avail == NULL))
 933                 goto out;
 934
 935         ret = *((volatile uint16_t *)&vq->avail->idx) - vq->last_avail_idx;
 936
 937 out:
 938         rte_spinlock_unlock(&vq->access_lock);
 939         return ret;
 940 }
 941
 942 int rte_vhost_get_vdpa_device_id(int vid)
 943 {
 944         struct virtio_net *dev = get_device(vid);
 945
 946         if (dev == NULL)
 947                 return -1;
 948
 949         return dev->vdpa_dev_id;
 950 }
 951
 952 int rte_vhost_get_log_base(int vid, uint64_t *log_base,
 953                 uint64_t *log_size)
 954 {
 955         struct virtio_net *dev = get_device(vid);
 956
 957         if (dev == NULL || log_base == NULL || log_size == NULL)
 958                 return -1;
 959
 960         *log_base = dev->log_base;
 961         *log_size = dev->log_size;
 962
 963         return 0;
 964 }
 965
 966 int rte_vhost_get_vring_base(int vid, uint16_t queue_id,
 967                 uint16_t *last_avail_idx, uint16_t *last_used_idx)
 968 {
 969         struct virtio_net *dev = get_device(vid);
 970
 971         if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
 972                 return -1;
 973
 974         *last_avail_idx = dev->virtqueue[queue_id]->last_avail_idx;
 975         *last_used_idx = dev->virtqueue[queue_id]->last_used_idx;
 976
 977         return 0;
 978 }
 979
 980 int rte_vhost_set_vring_base(int vid, uint16_t queue_id,
 981                 uint16_t last_avail_idx, uint16_t last_used_idx)
 982 {
 983         struct virtio_net *dev = get_device(vid);
 984
 985         if (!dev)
 986                 return -1;
 987
 988         dev->virtqueue[queue_id]->last_avail_idx = last_avail_idx;
 989         dev->virtqueue[queue_id]->last_used_idx = last_used_idx;
 990
 991         return 0;
 992 }
 993
 994 int rte_vhost_extern_callback_register(int vid,
 995                 struct rte_vhost_user_extern_ops const * const ops, void *ctx)
 996 {
 997         struct virtio_net *dev = get_device(vid);
 998
 999         if (dev == NULL || ops == NULL)
1000                 return -1;
1001
1002         dev->extern_ops = *ops;
1003         dev->extern_data = ctx;
1004         return 0;
1005 }