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                         VHOST_LOG_CONFIG(ERR,
  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_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 120                              uint64_t iova, uint64_t len)
 121 {
 122         uint64_t hva, gpa, map_len;
 123         map_len = len;
 124
 125         hva = __vhost_iova_to_vva(dev, vq, iova, &map_len, VHOST_ACCESS_RW);
 126         if (map_len != len) {
 127                 VHOST_LOG_DATA(ERR,
 128                         "Failed to write log for IOVA 0x%" PRIx64 ". No IOTLB entry found\n",
 129                         iova);
 130                 return;
 131         }
 132
 133         gpa = hva_to_gpa(dev, hva, len);
 134         if (gpa)
 135                 __vhost_log_write(dev, gpa, len);
 136 }
 137
 138 void
 139 __vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
 140 {
 141         unsigned long *log_base;
 142         int i;
 143
 144         if (unlikely(!dev->log_base))
 145                 return;
 146
 147         rte_smp_wmb();
 148
 149         log_base = (unsigned long *)(uintptr_t)dev->log_base;
 150
 151         for (i = 0; i < vq->log_cache_nb_elem; i++) {
 152                 struct log_cache_entry *elem = vq->log_cache + i;
 153
 154 #if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION < 70100)
 155                 /*
 156                  * '__sync' builtins are deprecated, but '__atomic' ones
 157                  * are sub-optimized in older GCC versions.
 158                  */
 159                 __sync_fetch_and_or(log_base + elem->offset, elem->val);
 160 #else
 161                 __atomic_fetch_or(log_base + elem->offset, elem->val,
 162                                 __ATOMIC_RELAXED);
 163 #endif
 164         }
 165
 166         rte_smp_wmb();
 167
 168         vq->log_cache_nb_elem = 0;
 169 }
 170
 171 static __rte_always_inline void
 172 vhost_log_cache_page(struct virtio_net *dev, struct vhost_virtqueue *vq,
 173                         uint64_t page)
 174 {
 175         uint32_t bit_nr = page % (sizeof(unsigned long) << 3);
 176         uint32_t offset = page / (sizeof(unsigned long) << 3);
 177         int i;
 178
 179         for (i = 0; i < vq->log_cache_nb_elem; i++) {
 180                 struct log_cache_entry *elem = vq->log_cache + i;
 181
 182                 if (elem->offset == offset) {
 183                         elem->val |= (1UL << bit_nr);
 184                         return;
 185                 }
 186         }
 187
 188         if (unlikely(i >= VHOST_LOG_CACHE_NR)) {
 189                 /*
 190                  * No more room for a new log cache entry,
 191                  * so write the dirty log map directly.
 192                  */
 193                 rte_smp_wmb();
 194                 vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
 195
 196                 return;
 197         }
 198
 199         vq->log_cache[i].offset = offset;
 200         vq->log_cache[i].val = (1UL << bit_nr);
 201         vq->log_cache_nb_elem++;
 202 }
 203
 204 void
 205 __vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
 206                         uint64_t addr, uint64_t len)
 207 {
 208         uint64_t page;
 209
 210         if (unlikely(!dev->log_base || !len))
 211                 return;
 212
 213         if (unlikely(dev->log_size <= ((addr + len - 1) / VHOST_LOG_PAGE / 8)))
 214                 return;
 215
 216         page = addr / VHOST_LOG_PAGE;
 217         while (page * VHOST_LOG_PAGE < addr + len) {
 218                 vhost_log_cache_page(dev, vq, page);
 219                 page += 1;
 220         }
 221 }
 222
 223 void
 224 __vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 225                              uint64_t iova, uint64_t len)
 226 {
 227         uint64_t hva, gpa, map_len;
 228         map_len = len;
 229
 230         hva = __vhost_iova_to_vva(dev, vq, iova, &map_len, VHOST_ACCESS_RW);
 231         if (map_len != len) {
 232                 VHOST_LOG_DATA(ERR,
 233                         "Failed to write log for IOVA 0x%" PRIx64 ". No IOTLB entry found\n",
 234                         iova);
 235                 return;
 236         }
 237
 238         gpa = hva_to_gpa(dev, hva, len);
 239         if (gpa)
 240                 __vhost_log_cache_write(dev, vq, gpa, len);
 241 }
 242
 243 void *
 244 vhost_alloc_copy_ind_table(struct virtio_net *dev, struct vhost_virtqueue *vq,
 245                 uint64_t desc_addr, uint64_t desc_len)
 246 {
 247         void *idesc;
 248         uint64_t src, dst;
 249         uint64_t len, remain = desc_len;
 250
 251         idesc = rte_malloc(__func__, desc_len, 0);
 252         if (unlikely(!idesc))
 253                 return NULL;
 254
 255         dst = (uint64_t)(uintptr_t)idesc;
 256
 257         while (remain) {
 258                 len = remain;
 259                 src = vhost_iova_to_vva(dev, vq, desc_addr, &len,
 260                                 VHOST_ACCESS_RO);
 261                 if (unlikely(!src || !len)) {
 262                         rte_free(idesc);
 263                         return NULL;
 264                 }
 265
 266                 rte_memcpy((void *)(uintptr_t)dst, (void *)(uintptr_t)src, len);
 267
 268                 remain -= len;
 269                 dst += len;
 270                 desc_addr += len;
 271         }
 272
 273         return idesc;
 274 }
 275
 276 void
 277 cleanup_vq(struct vhost_virtqueue *vq, int destroy)
 278 {
 279         if ((vq->callfd >= 0) && (destroy != 0))
 280                 close(vq->callfd);
 281         if (vq->kickfd >= 0)
 282                 close(vq->kickfd);
 283 }
 284
 285 void
 286 cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq)
 287 {
 288         if (!(dev->protocol_features &
 289             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
 290                 return;
 291
 292         if (vq_is_packed(dev)) {
 293                 if (vq->inflight_packed)
 294                         vq->inflight_packed = NULL;
 295         } else {
 296                 if (vq->inflight_split)
 297                         vq->inflight_split = NULL;
 298         }
 299
 300         if (vq->resubmit_inflight) {
 301                 if (vq->resubmit_inflight->resubmit_list) {
 302                         free(vq->resubmit_inflight->resubmit_list);
 303                         vq->resubmit_inflight->resubmit_list = NULL;
 304                 }
 305                 free(vq->resubmit_inflight);
 306                 vq->resubmit_inflight = NULL;
 307         }
 308 }
 309
 310 /*
 311  * Unmap any memory, close any file descriptors and
 312  * free any memory owned by a device.
 313  */
 314 void
 315 cleanup_device(struct virtio_net *dev, int destroy)
 316 {
 317         uint32_t i;
 318
 319         vhost_backend_cleanup(dev);
 320
 321         for (i = 0; i < dev->nr_vring; i++) {
 322                 cleanup_vq(dev->virtqueue[i], destroy);
 323                 cleanup_vq_inflight(dev, dev->virtqueue[i]);
 324         }
 325 }
 326
 327 void
 328 free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq)
 329 {
 330         if (vq_is_packed(dev))
 331                 rte_free(vq->shadow_used_packed);
 332         else {
 333                 rte_free(vq->shadow_used_split);
 334                 if (vq->async_pkts_pending)
 335                         rte_free(vq->async_pkts_pending);
 336                 if (vq->async_pending_info)
 337                         rte_free(vq->async_pending_info);
 338         }
 339         rte_free(vq->batch_copy_elems);
 340         rte_mempool_free(vq->iotlb_pool);
 341         rte_free(vq);
 342 }
 343
 344 /*
 345  * Release virtqueues and device memory.
 346  */
 347 static void
 348 free_device(struct virtio_net *dev)
 349 {
 350         uint32_t i;
 351
 352         for (i = 0; i < dev->nr_vring; i++)
 353                 free_vq(dev, dev->virtqueue[i]);
 354
 355         rte_free(dev);
 356 }
 357
 358 static __rte_always_inline int
 359 log_translate(struct virtio_net *dev, struct vhost_virtqueue *vq)
 360 {
 361         if (likely(!(vq->ring_addrs.flags & (1 << VHOST_VRING_F_LOG))))
 362                 return 0;
 363
 364         vq->log_guest_addr = translate_log_addr(dev, vq,
 365                                                 vq->ring_addrs.log_guest_addr);
 366         if (vq->log_guest_addr == 0)
 367                 return -1;
 368
 369         return 0;
 370 }
 371
 372 /*
 373  * Converts vring log address to GPA
 374  * If IOMMU is enabled, the log address is IOVA
 375  * If IOMMU not enabled, the log address is already GPA
 376  *
 377  * Caller should have iotlb_lock read-locked
 378  */
 379 uint64_t
 380 translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
 381                 uint64_t log_addr)
 382 {
 383         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 384                 const uint64_t exp_size = sizeof(uint64_t);
 385                 uint64_t hva, gpa;
 386                 uint64_t size = exp_size;
 387
 388                 hva = vhost_iova_to_vva(dev, vq, log_addr,
 389                                         &size, VHOST_ACCESS_RW);
 390
 391                 if (size != exp_size)
 392                         return 0;
 393
 394                 gpa = hva_to_gpa(dev, hva, exp_size);
 395                 if (!gpa) {
 396                         VHOST_LOG_CONFIG(ERR,
 397                                 "VQ: Failed to find GPA for log_addr: 0x%"
 398                                 PRIx64 " hva: 0x%" PRIx64 "\n",
 399                                 log_addr, hva);
 400                         return 0;
 401                 }
 402                 return gpa;
 403
 404         } else
 405                 return log_addr;
 406 }
 407
 408 /* Caller should have iotlb_lock read-locked */
 409 static int
 410 vring_translate_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 411 {
 412         uint64_t req_size, size;
 413
 414         req_size = sizeof(struct vring_desc) * vq->size;
 415         size = req_size;
 416         vq->desc = (struct vring_desc *)(uintptr_t)vhost_iova_to_vva(dev, vq,
 417                                                 vq->ring_addrs.desc_user_addr,
 418                                                 &size, VHOST_ACCESS_RW);
 419         if (!vq->desc || size != req_size)
 420                 return -1;
 421
 422         req_size = sizeof(struct vring_avail);
 423         req_size += sizeof(uint16_t) * vq->size;
 424         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 425                 req_size += sizeof(uint16_t);
 426         size = req_size;
 427         vq->avail = (struct vring_avail *)(uintptr_t)vhost_iova_to_vva(dev, vq,
 428                                                 vq->ring_addrs.avail_user_addr,
 429                                                 &size, VHOST_ACCESS_RW);
 430         if (!vq->avail || size != req_size)
 431                 return -1;
 432
 433         req_size = sizeof(struct vring_used);
 434         req_size += sizeof(struct vring_used_elem) * vq->size;
 435         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 436                 req_size += sizeof(uint16_t);
 437         size = req_size;
 438         vq->used = (struct vring_used *)(uintptr_t)vhost_iova_to_vva(dev, vq,
 439                                                 vq->ring_addrs.used_user_addr,
 440                                                 &size, VHOST_ACCESS_RW);
 441         if (!vq->used || size != req_size)
 442                 return -1;
 443
 444         return 0;
 445 }
 446
 447 /* Caller should have iotlb_lock read-locked */
 448 static int
 449 vring_translate_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
 450 {
 451         uint64_t req_size, size;
 452
 453         req_size = sizeof(struct vring_packed_desc) * vq->size;
 454         size = req_size;
 455         vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
 456                 vhost_iova_to_vva(dev, vq, vq->ring_addrs.desc_user_addr,
 457                                 &size, VHOST_ACCESS_RW);
 458         if (!vq->desc_packed || size != req_size)
 459                 return -1;
 460
 461         req_size = sizeof(struct vring_packed_desc_event);
 462         size = req_size;
 463         vq->driver_event = (struct vring_packed_desc_event *)(uintptr_t)
 464                 vhost_iova_to_vva(dev, vq, vq->ring_addrs.avail_user_addr,
 465                                 &size, VHOST_ACCESS_RW);
 466         if (!vq->driver_event || size != req_size)
 467                 return -1;
 468
 469         req_size = sizeof(struct vring_packed_desc_event);
 470         size = req_size;
 471         vq->device_event = (struct vring_packed_desc_event *)(uintptr_t)
 472                 vhost_iova_to_vva(dev, vq, vq->ring_addrs.used_user_addr,
 473                                 &size, VHOST_ACCESS_RW);
 474         if (!vq->device_event || size != req_size)
 475                 return -1;
 476
 477         return 0;
 478 }
 479
 480 int
 481 vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq)
 482 {
 483
 484         if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 485                 return -1;
 486
 487         if (vq_is_packed(dev)) {
 488                 if (vring_translate_packed(dev, vq) < 0)
 489                         return -1;
 490         } else {
 491                 if (vring_translate_split(dev, vq) < 0)
 492                         return -1;
 493         }
 494
 495         if (log_translate(dev, vq) < 0)
 496                 return -1;
 497
 498         vq->access_ok = 1;
 499
 500         return 0;
 501 }
 502
 503 void
 504 vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq)
 505 {
 506         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 507                 vhost_user_iotlb_wr_lock(vq);
 508
 509         vq->access_ok = 0;
 510         vq->desc = NULL;
 511         vq->avail = NULL;
 512         vq->used = NULL;
 513         vq->log_guest_addr = 0;
 514
 515         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 516                 vhost_user_iotlb_wr_unlock(vq);
 517 }
 518
 519 static void
 520 init_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 521 {
 522         struct vhost_virtqueue *vq;
 523
 524         if (vring_idx >= VHOST_MAX_VRING) {
 525                 VHOST_LOG_CONFIG(ERR,
 526                                 "Failed not init vring, out of bound (%d)\n",
 527                                 vring_idx);
 528                 return;
 529         }
 530
 531         vq = dev->virtqueue[vring_idx];
 532
 533         memset(vq, 0, sizeof(struct vhost_virtqueue));
 534
 535         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 536         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
 537
 538         vhost_user_iotlb_init(dev, vring_idx);
 539         /* Backends are set to -1 indicating an inactive device. */
 540         vq->backend = -1;
 541
 542         TAILQ_INIT(&vq->zmbuf_list);
 543 }
 544
 545 static void
 546 reset_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 547 {
 548         struct vhost_virtqueue *vq;
 549         int callfd;
 550
 551         if (vring_idx >= VHOST_MAX_VRING) {
 552                 VHOST_LOG_CONFIG(ERR,
 553                                 "Failed not init vring, out of bound (%d)\n",
 554                                 vring_idx);
 555                 return;
 556         }
 557
 558         vq = dev->virtqueue[vring_idx];
 559         callfd = vq->callfd;
 560         init_vring_queue(dev, vring_idx);
 561         vq->callfd = callfd;
 562 }
 563
 564 int
 565 alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 566 {
 567         struct vhost_virtqueue *vq;
 568
 569         vq = rte_malloc(NULL, sizeof(struct vhost_virtqueue), 0);
 570         if (vq == NULL) {
 571                 VHOST_LOG_CONFIG(ERR,
 572                         "Failed to allocate memory for vring:%u.\n", vring_idx);
 573                 return -1;
 574         }
 575
 576         dev->virtqueue[vring_idx] = vq;
 577         init_vring_queue(dev, vring_idx);
 578         rte_spinlock_init(&vq->access_lock);
 579         vq->avail_wrap_counter = 1;
 580         vq->used_wrap_counter = 1;
 581         vq->signalled_used_valid = false;
 582
 583         dev->nr_vring += 1;
 584
 585         return 0;
 586 }
 587
 588 /*
 589  * Reset some variables in device structure, while keeping few
 590  * others untouched, such as vid, ifname, nr_vring: they
 591  * should be same unless the device is removed.
 592  */
 593 void
 594 reset_device(struct virtio_net *dev)
 595 {
 596         uint32_t i;
 597
 598         dev->features = 0;
 599         dev->protocol_features = 0;
 600         dev->flags &= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 601
 602         for (i = 0; i < dev->nr_vring; i++)
 603                 reset_vring_queue(dev, i);
 604 }
 605
 606 /*
 607  * Invoked when there is a new vhost-user connection established (when
 608  * there is a new virtio device being attached).
 609  */
 610 int
 611 vhost_new_device(void)
 612 {
 613         struct virtio_net *dev;
 614         int i;
 615
 616         for (i = 0; i < MAX_VHOST_DEVICE; i++) {
 617                 if (vhost_devices[i] == NULL)
 618                         break;
 619         }
 620
 621         if (i == MAX_VHOST_DEVICE) {
 622                 VHOST_LOG_CONFIG(ERR,
 623                         "Failed to find a free slot for new device.\n");
 624                 return -1;
 625         }
 626
 627         dev = rte_zmalloc(NULL, sizeof(struct virtio_net), 0);
 628         if (dev == NULL) {
 629                 VHOST_LOG_CONFIG(ERR,
 630                         "Failed to allocate memory for new dev.\n");
 631                 return -1;
 632         }
 633
 634         vhost_devices[i] = dev;
 635         dev->vid = i;
 636         dev->flags = VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 637         dev->slave_req_fd = -1;
 638         dev->postcopy_ufd = -1;
 639         rte_spinlock_init(&dev->slave_req_lock);
 640
 641         return i;
 642 }
 643
 644 void
 645 vhost_destroy_device_notify(struct virtio_net *dev)
 646 {
 647         struct rte_vdpa_device *vdpa_dev;
 648
 649         if (dev->flags & VIRTIO_DEV_RUNNING) {
 650                 vdpa_dev = dev->vdpa_dev;
 651                 if (vdpa_dev)
 652                         vdpa_dev->ops->dev_close(dev->vid);
 653                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 654                 dev->notify_ops->destroy_device(dev->vid);
 655         }
 656 }
 657
 658 /*
 659  * Invoked when there is the vhost-user connection is broken (when
 660  * the virtio device is being detached).
 661  */
 662 void
 663 vhost_destroy_device(int vid)
 664 {
 665         struct virtio_net *dev = get_device(vid);
 666
 667         if (dev == NULL)
 668                 return;
 669
 670         vhost_destroy_device_notify(dev);
 671
 672         cleanup_device(dev, 1);
 673         free_device(dev);
 674
 675         vhost_devices[vid] = NULL;
 676 }
 677
 678 void
 679 vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *vdpa_dev)
 680 {
 681         struct virtio_net *dev = get_device(vid);
 682
 683         if (dev == NULL)
 684                 return;
 685
 686         dev->vdpa_dev = vdpa_dev;
 687 }
 688
 689 void
 690 vhost_set_ifname(int vid, const char *if_name, unsigned int if_len)
 691 {
 692         struct virtio_net *dev;
 693         unsigned int len;
 694
 695         dev = get_device(vid);
 696         if (dev == NULL)
 697                 return;
 698
 699         len = if_len > sizeof(dev->ifname) ?
 700                 sizeof(dev->ifname) : if_len;
 701
 702         strncpy(dev->ifname, if_name, len);
 703         dev->ifname[sizeof(dev->ifname) - 1] = '\0';
 704 }
 705
 706 void
 707 vhost_enable_dequeue_zero_copy(int vid)
 708 {
 709         struct virtio_net *dev = get_device(vid);
 710
 711         if (dev == NULL)
 712                 return;
 713
 714         dev->dequeue_zero_copy = 1;
 715         VHOST_LOG_CONFIG(INFO, "dequeue zero copy is enabled\n");
 716 }
 717
 718 void
 719 vhost_set_builtin_virtio_net(int vid, bool enable)
 720 {
 721         struct virtio_net *dev = get_device(vid);
 722
 723         if (dev == NULL)
 724                 return;
 725
 726         if (enable)
 727                 dev->flags |= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 728         else
 729                 dev->flags &= ~VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 730 }
 731
 732 void
 733 vhost_enable_extbuf(int vid)
 734 {
 735         struct virtio_net *dev = get_device(vid);
 736
 737         if (dev == NULL)
 738                 return;
 739
 740         dev->extbuf = 1;
 741 }
 742
 743 void
 744 vhost_enable_linearbuf(int vid)
 745 {
 746         struct virtio_net *dev = get_device(vid);
 747
 748         if (dev == NULL)
 749                 return;
 750
 751         dev->linearbuf = 1;
 752 }
 753
 754 int
 755 rte_vhost_get_mtu(int vid, uint16_t *mtu)
 756 {
 757         struct virtio_net *dev = get_device(vid);
 758
 759         if (dev == NULL || mtu == NULL)
 760                 return -ENODEV;
 761
 762         if (!(dev->flags & VIRTIO_DEV_READY))
 763                 return -EAGAIN;
 764
 765         if (!(dev->features & (1ULL << VIRTIO_NET_F_MTU)))
 766                 return -ENOTSUP;
 767
 768         *mtu = dev->mtu;
 769
 770         return 0;
 771 }
 772
 773 int
 774 rte_vhost_get_numa_node(int vid)
 775 {
 776 #ifdef RTE_LIBRTE_VHOST_NUMA
 777         struct virtio_net *dev = get_device(vid);
 778         int numa_node;
 779         int ret;
 780
 781         if (dev == NULL || numa_available() != 0)
 782                 return -1;
 783
 784         ret = get_mempolicy(&numa_node, NULL, 0, dev,
 785                             MPOL_F_NODE | MPOL_F_ADDR);
 786         if (ret < 0) {
 787                 VHOST_LOG_CONFIG(ERR,
 788                         "(%d) failed to query numa node: %s\n",
 789                         vid, rte_strerror(errno));
 790                 return -1;
 791         }
 792
 793         return numa_node;
 794 #else
 795         RTE_SET_USED(vid);
 796         return -1;
 797 #endif
 798 }
 799
 800 uint32_t
 801 rte_vhost_get_queue_num(int vid)
 802 {
 803         struct virtio_net *dev = get_device(vid);
 804
 805         if (dev == NULL)
 806                 return 0;
 807
 808         return dev->nr_vring / 2;
 809 }
 810
 811 uint16_t
 812 rte_vhost_get_vring_num(int vid)
 813 {
 814         struct virtio_net *dev = get_device(vid);
 815
 816         if (dev == NULL)
 817                 return 0;
 818
 819         return dev->nr_vring;
 820 }
 821
 822 int
 823 rte_vhost_get_ifname(int vid, char *buf, size_t len)
 824 {
 825         struct virtio_net *dev = get_device(vid);
 826
 827         if (dev == NULL || buf == NULL)
 828                 return -1;
 829
 830         len = RTE_MIN(len, sizeof(dev->ifname));
 831
 832         strncpy(buf, dev->ifname, len);
 833         buf[len - 1] = '\0';
 834
 835         return 0;
 836 }
 837
 838 int
 839 rte_vhost_get_negotiated_features(int vid, uint64_t *features)
 840 {
 841         struct virtio_net *dev;
 842
 843         dev = get_device(vid);
 844         if (dev == NULL || features == NULL)
 845                 return -1;
 846
 847         *features = dev->features;
 848         return 0;
 849 }
 850
 851 int
 852 rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem)
 853 {
 854         struct virtio_net *dev;
 855         struct rte_vhost_memory *m;
 856         size_t size;
 857
 858         dev = get_device(vid);
 859         if (dev == NULL || mem == NULL)
 860                 return -1;
 861
 862         size = dev->mem->nregions * sizeof(struct rte_vhost_mem_region);
 863         m = malloc(sizeof(struct rte_vhost_memory) + size);
 864         if (!m)
 865                 return -1;
 866
 867         m->nregions = dev->mem->nregions;
 868         memcpy(m->regions, dev->mem->regions, size);
 869         *mem = m;
 870
 871         return 0;
 872 }
 873
 874 int
 875 rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
 876                           struct rte_vhost_vring *vring)
 877 {
 878         struct virtio_net *dev;
 879         struct vhost_virtqueue *vq;
 880
 881         dev = get_device(vid);
 882         if (dev == NULL || vring == NULL)
 883                 return -1;
 884
 885         if (vring_idx >= VHOST_MAX_VRING)
 886                 return -1;
 887
 888         vq = dev->virtqueue[vring_idx];
 889         if (!vq)
 890                 return -1;
 891
 892         if (vq_is_packed(dev)) {
 893                 vring->desc_packed = vq->desc_packed;
 894                 vring->driver_event = vq->driver_event;
 895                 vring->device_event = vq->device_event;
 896         } else {
 897                 vring->desc = vq->desc;
 898                 vring->avail = vq->avail;
 899                 vring->used = vq->used;
 900         }
 901         vring->log_guest_addr  = vq->log_guest_addr;
 902
 903         vring->callfd  = vq->callfd;
 904         vring->kickfd  = vq->kickfd;
 905         vring->size    = vq->size;
 906
 907         return 0;
 908 }
 909
 910 int
 911 rte_vhost_get_vhost_ring_inflight(int vid, uint16_t vring_idx,
 912                                   struct rte_vhost_ring_inflight *vring)
 913 {
 914         struct virtio_net *dev;
 915         struct vhost_virtqueue *vq;
 916
 917         dev = get_device(vid);
 918         if (unlikely(!dev))
 919                 return -1;
 920
 921         if (vring_idx >= VHOST_MAX_VRING)
 922                 return -1;
 923
 924         vq = dev->virtqueue[vring_idx];
 925         if (unlikely(!vq))
 926                 return -1;
 927
 928         if (vq_is_packed(dev)) {
 929                 if (unlikely(!vq->inflight_packed))
 930                         return -1;
 931
 932                 vring->inflight_packed = vq->inflight_packed;
 933         } else {
 934                 if (unlikely(!vq->inflight_split))
 935                         return -1;
 936
 937                 vring->inflight_split = vq->inflight_split;
 938         }
 939
 940         vring->resubmit_inflight = vq->resubmit_inflight;
 941
 942         return 0;
 943 }
 944
 945 int
 946 rte_vhost_set_inflight_desc_split(int vid, uint16_t vring_idx,
 947                                   uint16_t idx)
 948 {
 949         struct vhost_virtqueue *vq;
 950         struct virtio_net *dev;
 951
 952         dev = get_device(vid);
 953         if (unlikely(!dev))
 954                 return -1;
 955
 956         if (unlikely(!(dev->protocol_features &
 957             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
 958                 return 0;
 959
 960         if (unlikely(vq_is_packed(dev)))
 961                 return -1;
 962
 963         if (unlikely(vring_idx >= VHOST_MAX_VRING))
 964                 return -1;
 965
 966         vq = dev->virtqueue[vring_idx];
 967         if (unlikely(!vq))
 968                 return -1;
 969
 970         if (unlikely(!vq->inflight_split))
 971                 return -1;
 972
 973         if (unlikely(idx >= vq->size))
 974                 return -1;
 975
 976         vq->inflight_split->desc[idx].counter = vq->global_counter++;
 977         vq->inflight_split->desc[idx].inflight = 1;
 978         return 0;
 979 }
 980
 981 int
 982 rte_vhost_set_inflight_desc_packed(int vid, uint16_t vring_idx,
 983                                    uint16_t head, uint16_t last,
 984                                    uint16_t *inflight_entry)
 985 {
 986         struct rte_vhost_inflight_info_packed *inflight_info;
 987         struct virtio_net *dev;
 988         struct vhost_virtqueue *vq;
 989         struct vring_packed_desc *desc;
 990         uint16_t old_free_head, free_head;
 991
 992         dev = get_device(vid);
 993         if (unlikely(!dev))
 994                 return -1;
 995
 996         if (unlikely(!(dev->protocol_features &
 997             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
 998                 return 0;
 999
1000         if (unlikely(!vq_is_packed(dev)))
1001                 return -1;
1002
1003         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1004                 return -1;
1005
1006         vq = dev->virtqueue[vring_idx];
1007         if (unlikely(!vq))
1008                 return -1;
1009
1010         inflight_info = vq->inflight_packed;
1011         if (unlikely(!inflight_info))
1012                 return -1;
1013
1014         if (unlikely(head >= vq->size))
1015                 return -1;
1016
1017         desc = vq->desc_packed;
1018         old_free_head = inflight_info->old_free_head;
1019         if (unlikely(old_free_head >= vq->size))
1020                 return -1;
1021
1022         free_head = old_free_head;
1023
1024         /* init header descriptor */
1025         inflight_info->desc[old_free_head].num = 0;
1026         inflight_info->desc[old_free_head].counter = vq->global_counter++;
1027         inflight_info->desc[old_free_head].inflight = 1;
1028
1029         /* save desc entry in flight entry */
1030         while (head != ((last + 1) % vq->size)) {
1031                 inflight_info->desc[old_free_head].num++;
1032                 inflight_info->desc[free_head].addr = desc[head].addr;
1033                 inflight_info->desc[free_head].len = desc[head].len;
1034                 inflight_info->desc[free_head].flags = desc[head].flags;
1035                 inflight_info->desc[free_head].id = desc[head].id;
1036
1037                 inflight_info->desc[old_free_head].last = free_head;
1038                 free_head = inflight_info->desc[free_head].next;
1039                 inflight_info->free_head = free_head;
1040                 head = (head + 1) % vq->size;
1041         }
1042
1043         inflight_info->old_free_head = free_head;
1044         *inflight_entry = old_free_head;
1045
1046         return 0;
1047 }
1048
1049 int
1050 rte_vhost_clr_inflight_desc_split(int vid, uint16_t vring_idx,
1051                                   uint16_t last_used_idx, uint16_t idx)
1052 {
1053         struct virtio_net *dev;
1054         struct vhost_virtqueue *vq;
1055
1056         dev = get_device(vid);
1057         if (unlikely(!dev))
1058                 return -1;
1059
1060         if (unlikely(!(dev->protocol_features &
1061             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1062                 return 0;
1063
1064         if (unlikely(vq_is_packed(dev)))
1065                 return -1;
1066
1067         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1068                 return -1;
1069
1070         vq = dev->virtqueue[vring_idx];
1071         if (unlikely(!vq))
1072                 return -1;
1073
1074         if (unlikely(!vq->inflight_split))
1075                 return -1;
1076
1077         if (unlikely(idx >= vq->size))
1078                 return -1;
1079
1080         rte_smp_mb();
1081
1082         vq->inflight_split->desc[idx].inflight = 0;
1083
1084         rte_smp_mb();
1085
1086         vq->inflight_split->used_idx = last_used_idx;
1087         return 0;
1088 }
1089
1090 int
1091 rte_vhost_clr_inflight_desc_packed(int vid, uint16_t vring_idx,
1092                                    uint16_t head)
1093 {
1094         struct rte_vhost_inflight_info_packed *inflight_info;
1095         struct virtio_net *dev;
1096         struct vhost_virtqueue *vq;
1097
1098         dev = get_device(vid);
1099         if (unlikely(!dev))
1100                 return -1;
1101
1102         if (unlikely(!(dev->protocol_features &
1103             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1104                 return 0;
1105
1106         if (unlikely(!vq_is_packed(dev)))
1107                 return -1;
1108
1109         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1110                 return -1;
1111
1112         vq = dev->virtqueue[vring_idx];
1113         if (unlikely(!vq))
1114                 return -1;
1115
1116         inflight_info = vq->inflight_packed;
1117         if (unlikely(!inflight_info))
1118                 return -1;
1119
1120         if (unlikely(head >= vq->size))
1121                 return -1;
1122
1123         rte_smp_mb();
1124
1125         inflight_info->desc[head].inflight = 0;
1126
1127         rte_smp_mb();
1128
1129         inflight_info->old_free_head = inflight_info->free_head;
1130         inflight_info->old_used_idx = inflight_info->used_idx;
1131         inflight_info->old_used_wrap_counter = inflight_info->used_wrap_counter;
1132
1133         return 0;
1134 }
1135
1136 int
1137 rte_vhost_set_last_inflight_io_split(int vid, uint16_t vring_idx,
1138                                      uint16_t idx)
1139 {
1140         struct virtio_net *dev;
1141         struct vhost_virtqueue *vq;
1142
1143         dev = get_device(vid);
1144         if (unlikely(!dev))
1145                 return -1;
1146
1147         if (unlikely(!(dev->protocol_features &
1148             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1149                 return 0;
1150
1151         if (unlikely(vq_is_packed(dev)))
1152                 return -1;
1153
1154         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1155                 return -1;
1156
1157         vq = dev->virtqueue[vring_idx];
1158         if (unlikely(!vq))
1159                 return -1;
1160
1161         if (unlikely(!vq->inflight_split))
1162                 return -1;
1163
1164         vq->inflight_split->last_inflight_io = idx;
1165         return 0;
1166 }
1167
1168 int
1169 rte_vhost_set_last_inflight_io_packed(int vid, uint16_t vring_idx,
1170                                       uint16_t head)
1171 {
1172         struct rte_vhost_inflight_info_packed *inflight_info;
1173         struct virtio_net *dev;
1174         struct vhost_virtqueue *vq;
1175         uint16_t last;
1176
1177         dev = get_device(vid);
1178         if (unlikely(!dev))
1179                 return -1;
1180
1181         if (unlikely(!(dev->protocol_features &
1182             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1183                 return 0;
1184
1185         if (unlikely(!vq_is_packed(dev)))
1186                 return -1;
1187
1188         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1189                 return -1;
1190
1191         vq = dev->virtqueue[vring_idx];
1192         if (unlikely(!vq))
1193                 return -1;
1194
1195         inflight_info = vq->inflight_packed;
1196         if (unlikely(!inflight_info))
1197                 return -1;
1198
1199         if (unlikely(head >= vq->size))
1200                 return -1;
1201
1202         last = inflight_info->desc[head].last;
1203         if (unlikely(last >= vq->size))
1204                 return -1;
1205
1206         inflight_info->desc[last].next = inflight_info->free_head;
1207         inflight_info->free_head = head;
1208         inflight_info->used_idx += inflight_info->desc[head].num;
1209         if (inflight_info->used_idx >= inflight_info->desc_num) {
1210                 inflight_info->used_idx -= inflight_info->desc_num;
1211                 inflight_info->used_wrap_counter =
1212                         !inflight_info->used_wrap_counter;
1213         }
1214
1215         return 0;
1216 }
1217
1218 int
1219 rte_vhost_vring_call(int vid, uint16_t vring_idx)
1220 {
1221         struct virtio_net *dev;
1222         struct vhost_virtqueue *vq;
1223
1224         dev = get_device(vid);
1225         if (!dev)
1226                 return -1;
1227
1228         if (vring_idx >= VHOST_MAX_VRING)
1229                 return -1;
1230
1231         vq = dev->virtqueue[vring_idx];
1232         if (!vq)
1233                 return -1;
1234
1235         if (vq_is_packed(dev))
1236                 vhost_vring_call_packed(dev, vq);
1237         else
1238                 vhost_vring_call_split(dev, vq);
1239
1240         return 0;
1241 }
1242
1243 uint16_t
1244 rte_vhost_avail_entries(int vid, uint16_t queue_id)
1245 {
1246         struct virtio_net *dev;
1247         struct vhost_virtqueue *vq;
1248         uint16_t ret = 0;
1249
1250         dev = get_device(vid);
1251         if (!dev)
1252                 return 0;
1253
1254         vq = dev->virtqueue[queue_id];
1255
1256         rte_spinlock_lock(&vq->access_lock);
1257
1258         if (unlikely(!vq->enabled || vq->avail == NULL))
1259                 goto out;
1260
1261         ret = *(volatile uint16_t *)&vq->avail->idx - vq->last_used_idx;
1262
1263 out:
1264         rte_spinlock_unlock(&vq->access_lock);
1265         return ret;
1266 }
1267
1268 static inline int
1269 vhost_enable_notify_split(struct virtio_net *dev,
1270                 struct vhost_virtqueue *vq, int enable)
1271 {
1272         if (vq->used == NULL)
1273                 return -1;
1274
1275         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
1276                 if (enable)
1277                         vq->used->flags &= ~VRING_USED_F_NO_NOTIFY;
1278                 else
1279                         vq->used->flags |= VRING_USED_F_NO_NOTIFY;
1280         } else {
1281                 if (enable)
1282                         vhost_avail_event(vq) = vq->last_avail_idx;
1283         }
1284         return 0;
1285 }
1286
1287 static inline int
1288 vhost_enable_notify_packed(struct virtio_net *dev,
1289                 struct vhost_virtqueue *vq, int enable)
1290 {
1291         uint16_t flags;
1292
1293         if (vq->device_event == NULL)
1294                 return -1;
1295
1296         if (!enable) {
1297                 vq->device_event->flags = VRING_EVENT_F_DISABLE;
1298                 return 0;
1299         }
1300
1301         flags = VRING_EVENT_F_ENABLE;
1302         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
1303                 flags = VRING_EVENT_F_DESC;
1304                 vq->device_event->off_wrap = vq->last_avail_idx |
1305                         vq->avail_wrap_counter << 15;
1306         }
1307
1308         rte_smp_wmb();
1309
1310         vq->device_event->flags = flags;
1311         return 0;
1312 }
1313
1314 int
1315 rte_vhost_enable_guest_notification(int vid, uint16_t queue_id, int enable)
1316 {
1317         struct virtio_net *dev = get_device(vid);
1318         struct vhost_virtqueue *vq;
1319         int ret;
1320
1321         if (!dev)
1322                 return -1;
1323
1324         vq = dev->virtqueue[queue_id];
1325
1326         rte_spinlock_lock(&vq->access_lock);
1327
1328         if (vq_is_packed(dev))
1329                 ret = vhost_enable_notify_packed(dev, vq, enable);
1330         else
1331                 ret = vhost_enable_notify_split(dev, vq, enable);
1332
1333         rte_spinlock_unlock(&vq->access_lock);
1334
1335         return ret;
1336 }
1337
1338 void
1339 rte_vhost_log_write(int vid, uint64_t addr, uint64_t len)
1340 {
1341         struct virtio_net *dev = get_device(vid);
1342
1343         if (dev == NULL)
1344                 return;
1345
1346         vhost_log_write(dev, addr, len);
1347 }
1348
1349 void
1350 rte_vhost_log_used_vring(int vid, uint16_t vring_idx,
1351                          uint64_t offset, uint64_t len)
1352 {
1353         struct virtio_net *dev;
1354         struct vhost_virtqueue *vq;
1355
1356         dev = get_device(vid);
1357         if (dev == NULL)
1358                 return;
1359
1360         if (vring_idx >= VHOST_MAX_VRING)
1361                 return;
1362         vq = dev->virtqueue[vring_idx];
1363         if (!vq)
1364                 return;
1365
1366         vhost_log_used_vring(dev, vq, offset, len);
1367 }
1368
1369 uint32_t
1370 rte_vhost_rx_queue_count(int vid, uint16_t qid)
1371 {
1372         struct virtio_net *dev;
1373         struct vhost_virtqueue *vq;
1374         uint32_t ret = 0;
1375
1376         dev = get_device(vid);
1377         if (dev == NULL)
1378                 return 0;
1379
1380         if (unlikely(qid >= dev->nr_vring || (qid & 1) == 0)) {
1381                 VHOST_LOG_DATA(ERR, "(%d) %s: invalid virtqueue idx %d.\n",
1382                         dev->vid, __func__, qid);
1383                 return 0;
1384         }
1385
1386         vq = dev->virtqueue[qid];
1387         if (vq == NULL)
1388                 return 0;
1389
1390         rte_spinlock_lock(&vq->access_lock);
1391
1392         if (unlikely(vq->enabled == 0 || vq->avail == NULL))
1393                 goto out;
1394
1395         ret = *((volatile uint16_t *)&vq->avail->idx) - vq->last_avail_idx;
1396
1397 out:
1398         rte_spinlock_unlock(&vq->access_lock);
1399         return ret;
1400 }
1401
1402 struct rte_vdpa_device *
1403 rte_vhost_get_vdpa_device(int vid)
1404 {
1405         struct virtio_net *dev = get_device(vid);
1406
1407         if (dev == NULL)
1408                 return NULL;
1409
1410         return dev->vdpa_dev;
1411 }
1412
1413 int rte_vhost_get_log_base(int vid, uint64_t *log_base,
1414                 uint64_t *log_size)
1415 {
1416         struct virtio_net *dev = get_device(vid);
1417
1418         if (dev == NULL || log_base == NULL || log_size == NULL)
1419                 return -1;
1420
1421         *log_base = dev->log_base;
1422         *log_size = dev->log_size;
1423
1424         return 0;
1425 }
1426
1427 int rte_vhost_get_vring_base(int vid, uint16_t queue_id,
1428                 uint16_t *last_avail_idx, uint16_t *last_used_idx)
1429 {
1430         struct vhost_virtqueue *vq;
1431         struct virtio_net *dev = get_device(vid);
1432
1433         if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
1434                 return -1;
1435
1436         vq = dev->virtqueue[queue_id];
1437         if (!vq)
1438                 return -1;
1439
1440         if (vq_is_packed(dev)) {
1441                 *last_avail_idx = (vq->avail_wrap_counter << 15) |
1442                                   vq->last_avail_idx;
1443                 *last_used_idx = (vq->used_wrap_counter << 15) |
1444                                  vq->last_used_idx;
1445         } else {
1446                 *last_avail_idx = vq->last_avail_idx;
1447                 *last_used_idx = vq->last_used_idx;
1448         }
1449
1450         return 0;
1451 }
1452
1453 int rte_vhost_set_vring_base(int vid, uint16_t queue_id,
1454                 uint16_t last_avail_idx, uint16_t last_used_idx)
1455 {
1456         struct vhost_virtqueue *vq;
1457         struct virtio_net *dev = get_device(vid);
1458
1459         if (!dev)
1460                 return -1;
1461
1462         vq = dev->virtqueue[queue_id];
1463         if (!vq)
1464                 return -1;
1465
1466         if (vq_is_packed(dev)) {
1467                 vq->last_avail_idx = last_avail_idx & 0x7fff;
1468                 vq->avail_wrap_counter = !!(last_avail_idx & (1 << 15));
1469                 vq->last_used_idx = last_used_idx & 0x7fff;
1470                 vq->used_wrap_counter = !!(last_used_idx & (1 << 15));
1471         } else {
1472                 vq->last_avail_idx = last_avail_idx;
1473                 vq->last_used_idx = last_used_idx;
1474         }
1475
1476         return 0;
1477 }
1478
1479 int
1480 rte_vhost_get_vring_base_from_inflight(int vid,
1481                                        uint16_t queue_id,
1482                                        uint16_t *last_avail_idx,
1483                                        uint16_t *last_used_idx)
1484 {
1485         struct rte_vhost_inflight_info_packed *inflight_info;
1486         struct virtio_net *dev = get_device(vid);
1487
1488         if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
1489                 return -1;
1490
1491         if (!vq_is_packed(dev))
1492                 return -1;
1493
1494         inflight_info = dev->virtqueue[queue_id]->inflight_packed;
1495         if (!inflight_info)
1496                 return -1;
1497
1498         *last_avail_idx = (inflight_info->old_used_wrap_counter << 15) |
1499                           inflight_info->old_used_idx;
1500         *last_used_idx = *last_avail_idx;
1501
1502         return 0;
1503 }
1504
1505 int rte_vhost_extern_callback_register(int vid,
1506                 struct rte_vhost_user_extern_ops const * const ops, void *ctx)
1507 {
1508         struct virtio_net *dev = get_device(vid);
1509
1510         if (dev == NULL || ops == NULL)
1511                 return -1;
1512
1513         dev->extern_ops = *ops;
1514         dev->extern_data = ctx;
1515         return 0;
1516 }
1517
1518 int rte_vhost_async_channel_register(int vid, uint16_t queue_id,
1519                                         uint32_t features,
1520                                         struct rte_vhost_async_channel_ops *ops)
1521 {
1522         struct vhost_virtqueue *vq;
1523         struct virtio_net *dev = get_device(vid);
1524         struct rte_vhost_async_features f;
1525
1526         if (dev == NULL || ops == NULL)
1527                 return -1;
1528
1529         f.intval = features;
1530
1531         vq = dev->virtqueue[queue_id];
1532
1533         if (unlikely(vq == NULL || !dev->async_copy))
1534                 return -1;
1535
1536         /* packed queue is not supported */
1537         if (unlikely(vq_is_packed(dev) || !f.async_inorder)) {
1538                 VHOST_LOG_CONFIG(ERR,
1539                         "async copy is not supported on packed queue or non-inorder mode "
1540                         "(vid %d, qid: %d)\n", vid, queue_id);
1541                 return -1;
1542         }
1543
1544         if (unlikely(ops->check_completed_copies == NULL ||
1545                 ops->transfer_data == NULL))
1546                 return -1;
1547
1548         rte_spinlock_lock(&vq->access_lock);
1549
1550         if (unlikely(vq->async_registered)) {
1551                 VHOST_LOG_CONFIG(ERR,
1552                         "async register failed: channel already registered "
1553                         "(vid %d, qid: %d)\n", vid, queue_id);
1554                 goto reg_out;
1555         }
1556
1557         vq->async_pkts_pending = rte_malloc(NULL,
1558                         vq->size * sizeof(uintptr_t),
1559                         RTE_CACHE_LINE_SIZE);
1560         vq->async_pending_info = rte_malloc(NULL,
1561                         vq->size * sizeof(uint64_t),
1562                         RTE_CACHE_LINE_SIZE);
1563         if (!vq->async_pkts_pending || !vq->async_pending_info) {
1564                 if (vq->async_pkts_pending)
1565                         rte_free(vq->async_pkts_pending);
1566
1567                 if (vq->async_pending_info)
1568                         rte_free(vq->async_pending_info);
1569
1570                 VHOST_LOG_CONFIG(ERR,
1571                                 "async register failed: cannot allocate memory for vq data "
1572                                 "(vid %d, qid: %d)\n", vid, queue_id);
1573                 goto reg_out;
1574         }
1575
1576         vq->async_ops.check_completed_copies = ops->check_completed_copies;
1577         vq->async_ops.transfer_data = ops->transfer_data;
1578
1579         vq->async_inorder = f.async_inorder;
1580         vq->async_threshold = f.async_threshold;
1581
1582         vq->async_registered = true;
1583
1584 reg_out:
1585         rte_spinlock_unlock(&vq->access_lock);
1586
1587         return 0;
1588 }
1589
1590 int rte_vhost_async_channel_unregister(int vid, uint16_t queue_id)
1591 {
1592         struct vhost_virtqueue *vq;
1593         struct virtio_net *dev = get_device(vid);
1594         int ret = -1;
1595
1596         if (dev == NULL)
1597                 return ret;
1598
1599         vq = dev->virtqueue[queue_id];
1600
1601         if (vq == NULL)
1602                 return ret;
1603
1604         ret = 0;
1605         rte_spinlock_lock(&vq->access_lock);
1606
1607         if (!vq->async_registered)
1608                 goto out;
1609
1610         if (vq->async_pkts_inflight_n) {
1611                 VHOST_LOG_CONFIG(ERR, "Failed to unregister async channel. "
1612                         "async inflight packets must be completed before unregistration.\n");
1613                 ret = -1;
1614                 goto out;
1615         }
1616
1617         if (vq->async_pkts_pending) {
1618                 rte_free(vq->async_pkts_pending);
1619                 vq->async_pkts_pending = NULL;
1620         }
1621
1622         if (vq->async_pending_info) {
1623                 rte_free(vq->async_pending_info);
1624                 vq->async_pending_info = NULL;
1625         }
1626
1627         vq->async_ops.transfer_data = NULL;
1628         vq->async_ops.check_completed_copies = NULL;
1629         vq->async_registered = false;
1630
1631 out:
1632         rte_spinlock_unlock(&vq->access_lock);
1633
1634         return ret;
1635 }
1636
1637 RTE_LOG_REGISTER(vhost_config_log_level, lib.vhost.config, INFO);
1638 RTE_LOG_REGISTER(vhost_data_log_level, lib.vhost.data, WARNING);