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                 goto out;
 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 out:
 371         vq->access_ok = 1;
 372
 373         return 0;
 374 }
 375
 376 void
 377 vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq)
 378 {
 379         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 380                 vhost_user_iotlb_wr_lock(vq);
 381
 382         vq->access_ok = 0;
 383         vq->desc = NULL;
 384         vq->avail = NULL;
 385         vq->used = NULL;
 386
 387         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 388                 vhost_user_iotlb_wr_unlock(vq);
 389 }
 390
 391 static void
 392 init_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 393 {
 394         struct vhost_virtqueue *vq;
 395
 396         if (vring_idx >= VHOST_MAX_VRING) {
 397                 RTE_LOG(ERR, VHOST_CONFIG,
 398                                 "Failed not init vring, out of bound (%d)\n",
 399                                 vring_idx);
 400                 return;
 401         }
 402
 403         vq = dev->virtqueue[vring_idx];
 404
 405         memset(vq, 0, sizeof(struct vhost_virtqueue));
 406
 407         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 408         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
 409
 410         vhost_user_iotlb_init(dev, vring_idx);
 411         /* Backends are set to -1 indicating an inactive device. */
 412         vq->backend = -1;
 413
 414         TAILQ_INIT(&vq->zmbuf_list);
 415 }
 416
 417 static void
 418 reset_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 419 {
 420         struct vhost_virtqueue *vq;
 421         int callfd;
 422
 423         if (vring_idx >= VHOST_MAX_VRING) {
 424                 RTE_LOG(ERR, VHOST_CONFIG,
 425                                 "Failed not init vring, out of bound (%d)\n",
 426                                 vring_idx);
 427                 return;
 428         }
 429
 430         vq = dev->virtqueue[vring_idx];
 431         callfd = vq->callfd;
 432         init_vring_queue(dev, vring_idx);
 433         vq->callfd = callfd;
 434 }
 435
 436 int
 437 alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 438 {
 439         struct vhost_virtqueue *vq;
 440
 441         vq = rte_malloc(NULL, sizeof(struct vhost_virtqueue), 0);
 442         if (vq == NULL) {
 443                 RTE_LOG(ERR, VHOST_CONFIG,
 444                         "Failed to allocate memory for vring:%u.\n", vring_idx);
 445                 return -1;
 446         }
 447
 448         dev->virtqueue[vring_idx] = vq;
 449         init_vring_queue(dev, vring_idx);
 450         rte_spinlock_init(&vq->access_lock);
 451         vq->avail_wrap_counter = 1;
 452         vq->used_wrap_counter = 1;
 453         vq->signalled_used_valid = false;
 454
 455         dev->nr_vring += 1;
 456
 457         return 0;
 458 }
 459
 460 /*
 461  * Reset some variables in device structure, while keeping few
 462  * others untouched, such as vid, ifname, nr_vring: they
 463  * should be same unless the device is removed.
 464  */
 465 void
 466 reset_device(struct virtio_net *dev)
 467 {
 468         uint32_t i;
 469
 470         dev->features = 0;
 471         dev->protocol_features = 0;
 472         dev->flags &= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 473
 474         for (i = 0; i < dev->nr_vring; i++)
 475                 reset_vring_queue(dev, i);
 476 }
 477
 478 /*
 479  * Invoked when there is a new vhost-user connection established (when
 480  * there is a new virtio device being attached).
 481  */
 482 int
 483 vhost_new_device(void)
 484 {
 485         struct virtio_net *dev;
 486         int i;
 487
 488         for (i = 0; i < MAX_VHOST_DEVICE; i++) {
 489                 if (vhost_devices[i] == NULL)
 490                         break;
 491         }
 492
 493         if (i == MAX_VHOST_DEVICE) {
 494                 RTE_LOG(ERR, VHOST_CONFIG,
 495                         "Failed to find a free slot for new device.\n");
 496                 return -1;
 497         }
 498
 499         dev = rte_zmalloc(NULL, sizeof(struct virtio_net), 0);
 500         if (dev == NULL) {
 501                 RTE_LOG(ERR, VHOST_CONFIG,
 502                         "Failed to allocate memory for new dev.\n");
 503                 return -1;
 504         }
 505
 506         vhost_devices[i] = dev;
 507         dev->vid = i;
 508         dev->flags = VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 509         dev->slave_req_fd = -1;
 510         dev->vdpa_dev_id = -1;
 511         dev->postcopy_ufd = -1;
 512         rte_spinlock_init(&dev->slave_req_lock);
 513
 514         return i;
 515 }
 516
 517 void
 518 vhost_destroy_device_notify(struct virtio_net *dev)
 519 {
 520         struct rte_vdpa_device *vdpa_dev;
 521         int did;
 522
 523         if (dev->flags & VIRTIO_DEV_RUNNING) {
 524                 did = dev->vdpa_dev_id;
 525                 vdpa_dev = rte_vdpa_get_device(did);
 526                 if (vdpa_dev && vdpa_dev->ops->dev_close)
 527                         vdpa_dev->ops->dev_close(dev->vid);
 528                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 529                 dev->notify_ops->destroy_device(dev->vid);
 530         }
 531 }
 532
 533 /*
 534  * Invoked when there is the vhost-user connection is broken (when
 535  * the virtio device is being detached).
 536  */
 537 void
 538 vhost_destroy_device(int vid)
 539 {
 540         struct virtio_net *dev = get_device(vid);
 541
 542         if (dev == NULL)
 543                 return;
 544
 545         vhost_destroy_device_notify(dev);
 546
 547         cleanup_device(dev, 1);
 548         free_device(dev);
 549
 550         vhost_devices[vid] = NULL;
 551 }
 552
 553 void
 554 vhost_attach_vdpa_device(int vid, int did)
 555 {
 556         struct virtio_net *dev = get_device(vid);
 557
 558         if (dev == NULL)
 559                 return;
 560
 561         if (rte_vdpa_get_device(did) == NULL)
 562                 return;
 563
 564         dev->vdpa_dev_id = did;
 565 }
 566
 567 void
 568 vhost_set_ifname(int vid, const char *if_name, unsigned int if_len)
 569 {
 570         struct virtio_net *dev;
 571         unsigned int len;
 572
 573         dev = get_device(vid);
 574         if (dev == NULL)
 575                 return;
 576
 577         len = if_len > sizeof(dev->ifname) ?
 578                 sizeof(dev->ifname) : if_len;
 579
 580         strncpy(dev->ifname, if_name, len);
 581         dev->ifname[sizeof(dev->ifname) - 1] = '\0';
 582 }
 583
 584 void
 585 vhost_enable_dequeue_zero_copy(int vid)
 586 {
 587         struct virtio_net *dev = get_device(vid);
 588
 589         if (dev == NULL)
 590                 return;
 591
 592         dev->dequeue_zero_copy = 1;
 593 }
 594
 595 void
 596 vhost_set_builtin_virtio_net(int vid, bool enable)
 597 {
 598         struct virtio_net *dev = get_device(vid);
 599
 600         if (dev == NULL)
 601                 return;
 602
 603         if (enable)
 604                 dev->flags |= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 605         else
 606                 dev->flags &= ~VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 607 }
 608
 609 int
 610 rte_vhost_get_mtu(int vid, uint16_t *mtu)
 611 {
 612         struct virtio_net *dev = get_device(vid);
 613
 614         if (dev == NULL || mtu == NULL)
 615                 return -ENODEV;
 616
 617         if (!(dev->flags & VIRTIO_DEV_READY))
 618                 return -EAGAIN;
 619
 620         if (!(dev->features & (1ULL << VIRTIO_NET_F_MTU)))
 621                 return -ENOTSUP;
 622
 623         *mtu = dev->mtu;
 624
 625         return 0;
 626 }
 627
 628 int
 629 rte_vhost_get_numa_node(int vid)
 630 {
 631 #ifdef RTE_LIBRTE_VHOST_NUMA
 632         struct virtio_net *dev = get_device(vid);
 633         int numa_node;
 634         int ret;
 635
 636         if (dev == NULL || numa_available() != 0)
 637                 return -1;
 638
 639         ret = get_mempolicy(&numa_node, NULL, 0, dev,
 640                             MPOL_F_NODE | MPOL_F_ADDR);
 641         if (ret < 0) {
 642                 RTE_LOG(ERR, VHOST_CONFIG,
 643                         "(%d) failed to query numa node: %s\n",
 644                         vid, rte_strerror(errno));
 645                 return -1;
 646         }
 647
 648         return numa_node;
 649 #else
 650         RTE_SET_USED(vid);
 651         return -1;
 652 #endif
 653 }
 654
 655 uint32_t
 656 rte_vhost_get_queue_num(int vid)
 657 {
 658         struct virtio_net *dev = get_device(vid);
 659
 660         if (dev == NULL)
 661                 return 0;
 662
 663         return dev->nr_vring / 2;
 664 }
 665
 666 uint16_t
 667 rte_vhost_get_vring_num(int vid)
 668 {
 669         struct virtio_net *dev = get_device(vid);
 670
 671         if (dev == NULL)
 672                 return 0;
 673
 674         return dev->nr_vring;
 675 }
 676
 677 int
 678 rte_vhost_get_ifname(int vid, char *buf, size_t len)
 679 {
 680         struct virtio_net *dev = get_device(vid);
 681
 682         if (dev == NULL || buf == NULL)
 683                 return -1;
 684
 685         len = RTE_MIN(len, sizeof(dev->ifname));
 686
 687         strncpy(buf, dev->ifname, len);
 688         buf[len - 1] = '\0';
 689
 690         return 0;
 691 }
 692
 693 int
 694 rte_vhost_get_negotiated_features(int vid, uint64_t *features)
 695 {
 696         struct virtio_net *dev;
 697
 698         dev = get_device(vid);
 699         if (dev == NULL || features == NULL)
 700                 return -1;
 701
 702         *features = dev->features;
 703         return 0;
 704 }
 705
 706 int
 707 rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem)
 708 {
 709         struct virtio_net *dev;
 710         struct rte_vhost_memory *m;
 711         size_t size;
 712
 713         dev = get_device(vid);
 714         if (dev == NULL || mem == NULL)
 715                 return -1;
 716
 717         size = dev->mem->nregions * sizeof(struct rte_vhost_mem_region);
 718         m = malloc(sizeof(struct rte_vhost_memory) + size);
 719         if (!m)
 720                 return -1;
 721
 722         m->nregions = dev->mem->nregions;
 723         memcpy(m->regions, dev->mem->regions, size);
 724         *mem = m;
 725
 726         return 0;
 727 }
 728
 729 int
 730 rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
 731                           struct rte_vhost_vring *vring)
 732 {
 733         struct virtio_net *dev;
 734         struct vhost_virtqueue *vq;
 735
 736         dev = get_device(vid);
 737         if (dev == NULL || vring == NULL)
 738                 return -1;
 739
 740         if (vring_idx >= VHOST_MAX_VRING)
 741                 return -1;
 742
 743         vq = dev->virtqueue[vring_idx];
 744         if (!vq)
 745                 return -1;
 746
 747         vring->desc  = vq->desc;
 748         vring->avail = vq->avail;
 749         vring->used  = vq->used;
 750         vring->log_guest_addr  = vq->log_guest_addr;
 751
 752         vring->callfd  = vq->callfd;
 753         vring->kickfd  = vq->kickfd;
 754         vring->size    = vq->size;
 755
 756         return 0;
 757 }
 758
 759 int
 760 rte_vhost_vring_call(int vid, uint16_t vring_idx)
 761 {
 762         struct virtio_net *dev;
 763         struct vhost_virtqueue *vq;
 764
 765         dev = get_device(vid);
 766         if (!dev)
 767                 return -1;
 768
 769         if (vring_idx >= VHOST_MAX_VRING)
 770                 return -1;
 771
 772         vq = dev->virtqueue[vring_idx];
 773         if (!vq)
 774                 return -1;
 775
 776         if (vq_is_packed(dev))
 777                 vhost_vring_call_packed(dev, vq);
 778         else
 779                 vhost_vring_call_split(dev, vq);
 780
 781         return 0;
 782 }
 783
 784 uint16_t
 785 rte_vhost_avail_entries(int vid, uint16_t queue_id)
 786 {
 787         struct virtio_net *dev;
 788         struct vhost_virtqueue *vq;
 789
 790         dev = get_device(vid);
 791         if (!dev)
 792                 return 0;
 793
 794         vq = dev->virtqueue[queue_id];
 795         if (!vq->enabled)
 796                 return 0;
 797
 798         return *(volatile uint16_t *)&vq->avail->idx - vq->last_used_idx;
 799 }
 800
 801 static inline void
 802 vhost_enable_notify_split(struct virtio_net *dev,
 803                 struct vhost_virtqueue *vq, int enable)
 804 {
 805         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
 806                 if (enable)
 807                         vq->used->flags &= ~VRING_USED_F_NO_NOTIFY;
 808                 else
 809                         vq->used->flags |= VRING_USED_F_NO_NOTIFY;
 810         } else {
 811                 if (enable)
 812                         vhost_avail_event(vq) = vq->last_avail_idx;
 813         }
 814 }
 815
 816 static inline void
 817 vhost_enable_notify_packed(struct virtio_net *dev,
 818                 struct vhost_virtqueue *vq, int enable)
 819 {
 820         uint16_t flags;
 821
 822         if (!enable) {
 823                 vq->device_event->flags = VRING_EVENT_F_DISABLE;
 824                 return;
 825         }
 826
 827         flags = VRING_EVENT_F_ENABLE;
 828         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
 829                 flags = VRING_EVENT_F_DESC;
 830                 vq->device_event->off_wrap = vq->last_avail_idx |
 831                         vq->avail_wrap_counter << 15;
 832         }
 833
 834         rte_smp_wmb();
 835
 836         vq->device_event->flags = flags;
 837 }
 838
 839 int
 840 rte_vhost_enable_guest_notification(int vid, uint16_t queue_id, int enable)
 841 {
 842         struct virtio_net *dev = get_device(vid);
 843         struct vhost_virtqueue *vq;
 844
 845         if (!dev)
 846                 return -1;
 847
 848         vq = dev->virtqueue[queue_id];
 849
 850         if (vq_is_packed(dev))
 851                 vhost_enable_notify_packed(dev, vq, enable);
 852         else
 853                 vhost_enable_notify_split(dev, vq, enable);
 854
 855         return 0;
 856 }
 857
 858 void
 859 rte_vhost_log_write(int vid, uint64_t addr, uint64_t len)
 860 {
 861         struct virtio_net *dev = get_device(vid);
 862
 863         if (dev == NULL)
 864                 return;
 865
 866         vhost_log_write(dev, addr, len);
 867 }
 868
 869 void
 870 rte_vhost_log_used_vring(int vid, uint16_t vring_idx,
 871                          uint64_t offset, uint64_t len)
 872 {
 873         struct virtio_net *dev;
 874         struct vhost_virtqueue *vq;
 875
 876         dev = get_device(vid);
 877         if (dev == NULL)
 878                 return;
 879
 880         if (vring_idx >= VHOST_MAX_VRING)
 881                 return;
 882         vq = dev->virtqueue[vring_idx];
 883         if (!vq)
 884                 return;
 885
 886         vhost_log_used_vring(dev, vq, offset, len);
 887 }
 888
 889 uint32_t
 890 rte_vhost_rx_queue_count(int vid, uint16_t qid)
 891 {
 892         struct virtio_net *dev;
 893         struct vhost_virtqueue *vq;
 894
 895         dev = get_device(vid);
 896         if (dev == NULL)
 897                 return 0;
 898
 899         if (unlikely(qid >= dev->nr_vring || (qid & 1) == 0)) {
 900                 RTE_LOG(ERR, VHOST_DATA, "(%d) %s: invalid virtqueue idx %d.\n",
 901                         dev->vid, __func__, qid);
 902                 return 0;
 903         }
 904
 905         vq = dev->virtqueue[qid];
 906         if (vq == NULL)
 907                 return 0;
 908
 909         if (unlikely(vq->enabled == 0 || vq->avail == NULL))
 910                 return 0;
 911
 912         return *((volatile uint16_t *)&vq->avail->idx) - vq->last_avail_idx;
 913 }
 914
 915 int rte_vhost_get_vdpa_device_id(int vid)
 916 {
 917         struct virtio_net *dev = get_device(vid);
 918
 919         if (dev == NULL)
 920                 return -1;
 921
 922         return dev->vdpa_dev_id;
 923 }
 924
 925 int rte_vhost_get_log_base(int vid, uint64_t *log_base,
 926                 uint64_t *log_size)
 927 {
 928         struct virtio_net *dev = get_device(vid);
 929
 930         if (dev == NULL || log_base == NULL || log_size == NULL)
 931                 return -1;
 932
 933         *log_base = dev->log_base;
 934         *log_size = dev->log_size;
 935
 936         return 0;
 937 }
 938
 939 int rte_vhost_get_vring_base(int vid, uint16_t queue_id,
 940                 uint16_t *last_avail_idx, uint16_t *last_used_idx)
 941 {
 942         struct virtio_net *dev = get_device(vid);
 943
 944         if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
 945                 return -1;
 946
 947         *last_avail_idx = dev->virtqueue[queue_id]->last_avail_idx;
 948         *last_used_idx = dev->virtqueue[queue_id]->last_used_idx;
 949
 950         return 0;
 951 }
 952
 953 int rte_vhost_set_vring_base(int vid, uint16_t queue_id,
 954                 uint16_t last_avail_idx, uint16_t last_used_idx)
 955 {
 956         struct virtio_net *dev = get_device(vid);
 957
 958         if (!dev)
 959                 return -1;
 960
 961         dev->virtqueue[queue_id]->last_avail_idx = last_avail_idx;
 962         dev->virtqueue[queue_id]->last_used_idx = last_used_idx;
 963
 964         return 0;
 965 }
 966
 967 int rte_vhost_extern_callback_register(int vid,
 968                 struct rte_vhost_user_extern_ops const * const ops, void *ctx)
 969 {
 970         struct virtio_net *dev = get_device(vid);
 971
 972         if (dev == NULL || ops == NULL)
 973                 return -1;
 974
 975         dev->extern_ops = *ops;
 976         dev->extern_data = ctx;
 977         return 0;
 978 }