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_pkts_info)
 337                         rte_free(vq->async_pkts_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         vq->notif_enable = VIRTIO_UNINITIALIZED_NOTIF;
 538
 539         vhost_user_iotlb_init(dev, vring_idx);
 540         /* Backends are set to -1 indicating an inactive device. */
 541         vq->backend = -1;
 542 }
 543
 544 static void
 545 reset_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 546 {
 547         struct vhost_virtqueue *vq;
 548         int callfd;
 549
 550         if (vring_idx >= VHOST_MAX_VRING) {
 551                 VHOST_LOG_CONFIG(ERR,
 552                                 "Failed not init vring, out of bound (%d)\n",
 553                                 vring_idx);
 554                 return;
 555         }
 556
 557         vq = dev->virtqueue[vring_idx];
 558         callfd = vq->callfd;
 559         init_vring_queue(dev, vring_idx);
 560         vq->callfd = callfd;
 561 }
 562
 563 int
 564 alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 565 {
 566         struct vhost_virtqueue *vq;
 567
 568         vq = rte_malloc(NULL, sizeof(struct vhost_virtqueue), 0);
 569         if (vq == NULL) {
 570                 VHOST_LOG_CONFIG(ERR,
 571                         "Failed to allocate memory for vring:%u.\n", vring_idx);
 572                 return -1;
 573         }
 574
 575         dev->virtqueue[vring_idx] = vq;
 576         init_vring_queue(dev, vring_idx);
 577         rte_spinlock_init(&vq->access_lock);
 578         vq->avail_wrap_counter = 1;
 579         vq->used_wrap_counter = 1;
 580         vq->signalled_used_valid = false;
 581
 582         dev->nr_vring += 1;
 583
 584         return 0;
 585 }
 586
 587 /*
 588  * Reset some variables in device structure, while keeping few
 589  * others untouched, such as vid, ifname, nr_vring: they
 590  * should be same unless the device is removed.
 591  */
 592 void
 593 reset_device(struct virtio_net *dev)
 594 {
 595         uint32_t i;
 596
 597         dev->features = 0;
 598         dev->protocol_features = 0;
 599         dev->flags &= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 600
 601         for (i = 0; i < dev->nr_vring; i++)
 602                 reset_vring_queue(dev, i);
 603 }
 604
 605 /*
 606  * Invoked when there is a new vhost-user connection established (when
 607  * there is a new virtio device being attached).
 608  */
 609 int
 610 vhost_new_device(void)
 611 {
 612         struct virtio_net *dev;
 613         int i;
 614
 615         for (i = 0; i < MAX_VHOST_DEVICE; i++) {
 616                 if (vhost_devices[i] == NULL)
 617                         break;
 618         }
 619
 620         if (i == MAX_VHOST_DEVICE) {
 621                 VHOST_LOG_CONFIG(ERR,
 622                         "Failed to find a free slot for new device.\n");
 623                 return -1;
 624         }
 625
 626         dev = rte_zmalloc(NULL, sizeof(struct virtio_net), 0);
 627         if (dev == NULL) {
 628                 VHOST_LOG_CONFIG(ERR,
 629                         "Failed to allocate memory for new dev.\n");
 630                 return -1;
 631         }
 632
 633         vhost_devices[i] = dev;
 634         dev->vid = i;
 635         dev->flags = VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 636         dev->slave_req_fd = -1;
 637         dev->postcopy_ufd = -1;
 638         rte_spinlock_init(&dev->slave_req_lock);
 639
 640         return i;
 641 }
 642
 643 void
 644 vhost_destroy_device_notify(struct virtio_net *dev)
 645 {
 646         struct rte_vdpa_device *vdpa_dev;
 647
 648         if (dev->flags & VIRTIO_DEV_RUNNING) {
 649                 vdpa_dev = dev->vdpa_dev;
 650                 if (vdpa_dev)
 651                         vdpa_dev->ops->dev_close(dev->vid);
 652                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 653                 dev->notify_ops->destroy_device(dev->vid);
 654         }
 655 }
 656
 657 /*
 658  * Invoked when there is the vhost-user connection is broken (when
 659  * the virtio device is being detached).
 660  */
 661 void
 662 vhost_destroy_device(int vid)
 663 {
 664         struct virtio_net *dev = get_device(vid);
 665
 666         if (dev == NULL)
 667                 return;
 668
 669         vhost_destroy_device_notify(dev);
 670
 671         cleanup_device(dev, 1);
 672         free_device(dev);
 673
 674         vhost_devices[vid] = NULL;
 675 }
 676
 677 void
 678 vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *vdpa_dev)
 679 {
 680         struct virtio_net *dev = get_device(vid);
 681
 682         if (dev == NULL)
 683                 return;
 684
 685         dev->vdpa_dev = vdpa_dev;
 686 }
 687
 688 void
 689 vhost_set_ifname(int vid, const char *if_name, unsigned int if_len)
 690 {
 691         struct virtio_net *dev;
 692         unsigned int len;
 693
 694         dev = get_device(vid);
 695         if (dev == NULL)
 696                 return;
 697
 698         len = if_len > sizeof(dev->ifname) ?
 699                 sizeof(dev->ifname) : if_len;
 700
 701         strncpy(dev->ifname, if_name, len);
 702         dev->ifname[sizeof(dev->ifname) - 1] = '\0';
 703 }
 704
 705 void
 706 vhost_set_builtin_virtio_net(int vid, bool enable)
 707 {
 708         struct virtio_net *dev = get_device(vid);
 709
 710         if (dev == NULL)
 711                 return;
 712
 713         if (enable)
 714                 dev->flags |= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 715         else
 716                 dev->flags &= ~VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 717 }
 718
 719 void
 720 vhost_enable_extbuf(int vid)
 721 {
 722         struct virtio_net *dev = get_device(vid);
 723
 724         if (dev == NULL)
 725                 return;
 726
 727         dev->extbuf = 1;
 728 }
 729
 730 void
 731 vhost_enable_linearbuf(int vid)
 732 {
 733         struct virtio_net *dev = get_device(vid);
 734
 735         if (dev == NULL)
 736                 return;
 737
 738         dev->linearbuf = 1;
 739 }
 740
 741 int
 742 rte_vhost_get_mtu(int vid, uint16_t *mtu)
 743 {
 744         struct virtio_net *dev = get_device(vid);
 745
 746         if (dev == NULL || mtu == NULL)
 747                 return -ENODEV;
 748
 749         if (!(dev->flags & VIRTIO_DEV_READY))
 750                 return -EAGAIN;
 751
 752         if (!(dev->features & (1ULL << VIRTIO_NET_F_MTU)))
 753                 return -ENOTSUP;
 754
 755         *mtu = dev->mtu;
 756
 757         return 0;
 758 }
 759
 760 int
 761 rte_vhost_get_numa_node(int vid)
 762 {
 763 #ifdef RTE_LIBRTE_VHOST_NUMA
 764         struct virtio_net *dev = get_device(vid);
 765         int numa_node;
 766         int ret;
 767
 768         if (dev == NULL || numa_available() != 0)
 769                 return -1;
 770
 771         ret = get_mempolicy(&numa_node, NULL, 0, dev,
 772                             MPOL_F_NODE | MPOL_F_ADDR);
 773         if (ret < 0) {
 774                 VHOST_LOG_CONFIG(ERR,
 775                         "(%d) failed to query numa node: %s\n",
 776                         vid, rte_strerror(errno));
 777                 return -1;
 778         }
 779
 780         return numa_node;
 781 #else
 782         RTE_SET_USED(vid);
 783         return -1;
 784 #endif
 785 }
 786
 787 uint32_t
 788 rte_vhost_get_queue_num(int vid)
 789 {
 790         struct virtio_net *dev = get_device(vid);
 791
 792         if (dev == NULL)
 793                 return 0;
 794
 795         return dev->nr_vring / 2;
 796 }
 797
 798 uint16_t
 799 rte_vhost_get_vring_num(int vid)
 800 {
 801         struct virtio_net *dev = get_device(vid);
 802
 803         if (dev == NULL)
 804                 return 0;
 805
 806         return dev->nr_vring;
 807 }
 808
 809 int
 810 rte_vhost_get_ifname(int vid, char *buf, size_t len)
 811 {
 812         struct virtio_net *dev = get_device(vid);
 813
 814         if (dev == NULL || buf == NULL)
 815                 return -1;
 816
 817         len = RTE_MIN(len, sizeof(dev->ifname));
 818
 819         strncpy(buf, dev->ifname, len);
 820         buf[len - 1] = '\0';
 821
 822         return 0;
 823 }
 824
 825 int
 826 rte_vhost_get_negotiated_features(int vid, uint64_t *features)
 827 {
 828         struct virtio_net *dev;
 829
 830         dev = get_device(vid);
 831         if (dev == NULL || features == NULL)
 832                 return -1;
 833
 834         *features = dev->features;
 835         return 0;
 836 }
 837
 838 int
 839 rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem)
 840 {
 841         struct virtio_net *dev;
 842         struct rte_vhost_memory *m;
 843         size_t size;
 844
 845         dev = get_device(vid);
 846         if (dev == NULL || mem == NULL)
 847                 return -1;
 848
 849         size = dev->mem->nregions * sizeof(struct rte_vhost_mem_region);
 850         m = malloc(sizeof(struct rte_vhost_memory) + size);
 851         if (!m)
 852                 return -1;
 853
 854         m->nregions = dev->mem->nregions;
 855         memcpy(m->regions, dev->mem->regions, size);
 856         *mem = m;
 857
 858         return 0;
 859 }
 860
 861 int
 862 rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
 863                           struct rte_vhost_vring *vring)
 864 {
 865         struct virtio_net *dev;
 866         struct vhost_virtqueue *vq;
 867
 868         dev = get_device(vid);
 869         if (dev == NULL || vring == NULL)
 870                 return -1;
 871
 872         if (vring_idx >= VHOST_MAX_VRING)
 873                 return -1;
 874
 875         vq = dev->virtqueue[vring_idx];
 876         if (!vq)
 877                 return -1;
 878
 879         if (vq_is_packed(dev)) {
 880                 vring->desc_packed = vq->desc_packed;
 881                 vring->driver_event = vq->driver_event;
 882                 vring->device_event = vq->device_event;
 883         } else {
 884                 vring->desc = vq->desc;
 885                 vring->avail = vq->avail;
 886                 vring->used = vq->used;
 887         }
 888         vring->log_guest_addr  = vq->log_guest_addr;
 889
 890         vring->callfd  = vq->callfd;
 891         vring->kickfd  = vq->kickfd;
 892         vring->size    = vq->size;
 893
 894         return 0;
 895 }
 896
 897 int
 898 rte_vhost_get_vhost_ring_inflight(int vid, uint16_t vring_idx,
 899                                   struct rte_vhost_ring_inflight *vring)
 900 {
 901         struct virtio_net *dev;
 902         struct vhost_virtqueue *vq;
 903
 904         dev = get_device(vid);
 905         if (unlikely(!dev))
 906                 return -1;
 907
 908         if (vring_idx >= VHOST_MAX_VRING)
 909                 return -1;
 910
 911         vq = dev->virtqueue[vring_idx];
 912         if (unlikely(!vq))
 913                 return -1;
 914
 915         if (vq_is_packed(dev)) {
 916                 if (unlikely(!vq->inflight_packed))
 917                         return -1;
 918
 919                 vring->inflight_packed = vq->inflight_packed;
 920         } else {
 921                 if (unlikely(!vq->inflight_split))
 922                         return -1;
 923
 924                 vring->inflight_split = vq->inflight_split;
 925         }
 926
 927         vring->resubmit_inflight = vq->resubmit_inflight;
 928
 929         return 0;
 930 }
 931
 932 int
 933 rte_vhost_set_inflight_desc_split(int vid, uint16_t vring_idx,
 934                                   uint16_t idx)
 935 {
 936         struct vhost_virtqueue *vq;
 937         struct virtio_net *dev;
 938
 939         dev = get_device(vid);
 940         if (unlikely(!dev))
 941                 return -1;
 942
 943         if (unlikely(!(dev->protocol_features &
 944             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
 945                 return 0;
 946
 947         if (unlikely(vq_is_packed(dev)))
 948                 return -1;
 949
 950         if (unlikely(vring_idx >= VHOST_MAX_VRING))
 951                 return -1;
 952
 953         vq = dev->virtqueue[vring_idx];
 954         if (unlikely(!vq))
 955                 return -1;
 956
 957         if (unlikely(!vq->inflight_split))
 958                 return -1;
 959
 960         if (unlikely(idx >= vq->size))
 961                 return -1;
 962
 963         vq->inflight_split->desc[idx].counter = vq->global_counter++;
 964         vq->inflight_split->desc[idx].inflight = 1;
 965         return 0;
 966 }
 967
 968 int
 969 rte_vhost_set_inflight_desc_packed(int vid, uint16_t vring_idx,
 970                                    uint16_t head, uint16_t last,
 971                                    uint16_t *inflight_entry)
 972 {
 973         struct rte_vhost_inflight_info_packed *inflight_info;
 974         struct virtio_net *dev;
 975         struct vhost_virtqueue *vq;
 976         struct vring_packed_desc *desc;
 977         uint16_t old_free_head, free_head;
 978
 979         dev = get_device(vid);
 980         if (unlikely(!dev))
 981                 return -1;
 982
 983         if (unlikely(!(dev->protocol_features &
 984             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
 985                 return 0;
 986
 987         if (unlikely(!vq_is_packed(dev)))
 988                 return -1;
 989
 990         if (unlikely(vring_idx >= VHOST_MAX_VRING))
 991                 return -1;
 992
 993         vq = dev->virtqueue[vring_idx];
 994         if (unlikely(!vq))
 995                 return -1;
 996
 997         inflight_info = vq->inflight_packed;
 998         if (unlikely(!inflight_info))
 999                 return -1;
1000
1001         if (unlikely(head >= vq->size))
1002                 return -1;
1003
1004         desc = vq->desc_packed;
1005         old_free_head = inflight_info->old_free_head;
1006         if (unlikely(old_free_head >= vq->size))
1007                 return -1;
1008
1009         free_head = old_free_head;
1010
1011         /* init header descriptor */
1012         inflight_info->desc[old_free_head].num = 0;
1013         inflight_info->desc[old_free_head].counter = vq->global_counter++;
1014         inflight_info->desc[old_free_head].inflight = 1;
1015
1016         /* save desc entry in flight entry */
1017         while (head != ((last + 1) % vq->size)) {
1018                 inflight_info->desc[old_free_head].num++;
1019                 inflight_info->desc[free_head].addr = desc[head].addr;
1020                 inflight_info->desc[free_head].len = desc[head].len;
1021                 inflight_info->desc[free_head].flags = desc[head].flags;
1022                 inflight_info->desc[free_head].id = desc[head].id;
1023
1024                 inflight_info->desc[old_free_head].last = free_head;
1025                 free_head = inflight_info->desc[free_head].next;
1026                 inflight_info->free_head = free_head;
1027                 head = (head + 1) % vq->size;
1028         }
1029
1030         inflight_info->old_free_head = free_head;
1031         *inflight_entry = old_free_head;
1032
1033         return 0;
1034 }
1035
1036 int
1037 rte_vhost_clr_inflight_desc_split(int vid, uint16_t vring_idx,
1038                                   uint16_t last_used_idx, uint16_t idx)
1039 {
1040         struct virtio_net *dev;
1041         struct vhost_virtqueue *vq;
1042
1043         dev = get_device(vid);
1044         if (unlikely(!dev))
1045                 return -1;
1046
1047         if (unlikely(!(dev->protocol_features &
1048             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1049                 return 0;
1050
1051         if (unlikely(vq_is_packed(dev)))
1052                 return -1;
1053
1054         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1055                 return -1;
1056
1057         vq = dev->virtqueue[vring_idx];
1058         if (unlikely(!vq))
1059                 return -1;
1060
1061         if (unlikely(!vq->inflight_split))
1062                 return -1;
1063
1064         if (unlikely(idx >= vq->size))
1065                 return -1;
1066
1067         rte_smp_mb();
1068
1069         vq->inflight_split->desc[idx].inflight = 0;
1070
1071         rte_smp_mb();
1072
1073         vq->inflight_split->used_idx = last_used_idx;
1074         return 0;
1075 }
1076
1077 int
1078 rte_vhost_clr_inflight_desc_packed(int vid, uint16_t vring_idx,
1079                                    uint16_t head)
1080 {
1081         struct rte_vhost_inflight_info_packed *inflight_info;
1082         struct virtio_net *dev;
1083         struct vhost_virtqueue *vq;
1084
1085         dev = get_device(vid);
1086         if (unlikely(!dev))
1087                 return -1;
1088
1089         if (unlikely(!(dev->protocol_features &
1090             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1091                 return 0;
1092
1093         if (unlikely(!vq_is_packed(dev)))
1094                 return -1;
1095
1096         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1097                 return -1;
1098
1099         vq = dev->virtqueue[vring_idx];
1100         if (unlikely(!vq))
1101                 return -1;
1102
1103         inflight_info = vq->inflight_packed;
1104         if (unlikely(!inflight_info))
1105                 return -1;
1106
1107         if (unlikely(head >= vq->size))
1108                 return -1;
1109
1110         rte_smp_mb();
1111
1112         inflight_info->desc[head].inflight = 0;
1113
1114         rte_smp_mb();
1115
1116         inflight_info->old_free_head = inflight_info->free_head;
1117         inflight_info->old_used_idx = inflight_info->used_idx;
1118         inflight_info->old_used_wrap_counter = inflight_info->used_wrap_counter;
1119
1120         return 0;
1121 }
1122
1123 int
1124 rte_vhost_set_last_inflight_io_split(int vid, uint16_t vring_idx,
1125                                      uint16_t idx)
1126 {
1127         struct virtio_net *dev;
1128         struct vhost_virtqueue *vq;
1129
1130         dev = get_device(vid);
1131         if (unlikely(!dev))
1132                 return -1;
1133
1134         if (unlikely(!(dev->protocol_features &
1135             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1136                 return 0;
1137
1138         if (unlikely(vq_is_packed(dev)))
1139                 return -1;
1140
1141         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1142                 return -1;
1143
1144         vq = dev->virtqueue[vring_idx];
1145         if (unlikely(!vq))
1146                 return -1;
1147
1148         if (unlikely(!vq->inflight_split))
1149                 return -1;
1150
1151         vq->inflight_split->last_inflight_io = idx;
1152         return 0;
1153 }
1154
1155 int
1156 rte_vhost_set_last_inflight_io_packed(int vid, uint16_t vring_idx,
1157                                       uint16_t head)
1158 {
1159         struct rte_vhost_inflight_info_packed *inflight_info;
1160         struct virtio_net *dev;
1161         struct vhost_virtqueue *vq;
1162         uint16_t last;
1163
1164         dev = get_device(vid);
1165         if (unlikely(!dev))
1166                 return -1;
1167
1168         if (unlikely(!(dev->protocol_features &
1169             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1170                 return 0;
1171
1172         if (unlikely(!vq_is_packed(dev)))
1173                 return -1;
1174
1175         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1176                 return -1;
1177
1178         vq = dev->virtqueue[vring_idx];
1179         if (unlikely(!vq))
1180                 return -1;
1181
1182         inflight_info = vq->inflight_packed;
1183         if (unlikely(!inflight_info))
1184                 return -1;
1185
1186         if (unlikely(head >= vq->size))
1187                 return -1;
1188
1189         last = inflight_info->desc[head].last;
1190         if (unlikely(last >= vq->size))
1191                 return -1;
1192
1193         inflight_info->desc[last].next = inflight_info->free_head;
1194         inflight_info->free_head = head;
1195         inflight_info->used_idx += inflight_info->desc[head].num;
1196         if (inflight_info->used_idx >= inflight_info->desc_num) {
1197                 inflight_info->used_idx -= inflight_info->desc_num;
1198                 inflight_info->used_wrap_counter =
1199                         !inflight_info->used_wrap_counter;
1200         }
1201
1202         return 0;
1203 }
1204
1205 int
1206 rte_vhost_vring_call(int vid, uint16_t vring_idx)
1207 {
1208         struct virtio_net *dev;
1209         struct vhost_virtqueue *vq;
1210
1211         dev = get_device(vid);
1212         if (!dev)
1213                 return -1;
1214
1215         if (vring_idx >= VHOST_MAX_VRING)
1216                 return -1;
1217
1218         vq = dev->virtqueue[vring_idx];
1219         if (!vq)
1220                 return -1;
1221
1222         if (vq_is_packed(dev))
1223                 vhost_vring_call_packed(dev, vq);
1224         else
1225                 vhost_vring_call_split(dev, vq);
1226
1227         return 0;
1228 }
1229
1230 uint16_t
1231 rte_vhost_avail_entries(int vid, uint16_t queue_id)
1232 {
1233         struct virtio_net *dev;
1234         struct vhost_virtqueue *vq;
1235         uint16_t ret = 0;
1236
1237         dev = get_device(vid);
1238         if (!dev)
1239                 return 0;
1240
1241         vq = dev->virtqueue[queue_id];
1242
1243         rte_spinlock_lock(&vq->access_lock);
1244
1245         if (unlikely(!vq->enabled || vq->avail == NULL))
1246                 goto out;
1247
1248         ret = *(volatile uint16_t *)&vq->avail->idx - vq->last_used_idx;
1249
1250 out:
1251         rte_spinlock_unlock(&vq->access_lock);
1252         return ret;
1253 }
1254
1255 static inline int
1256 vhost_enable_notify_split(struct virtio_net *dev,
1257                 struct vhost_virtqueue *vq, int enable)
1258 {
1259         if (vq->used == NULL)
1260                 return -1;
1261
1262         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
1263                 if (enable)
1264                         vq->used->flags &= ~VRING_USED_F_NO_NOTIFY;
1265                 else
1266                         vq->used->flags |= VRING_USED_F_NO_NOTIFY;
1267         } else {
1268                 if (enable)
1269                         vhost_avail_event(vq) = vq->last_avail_idx;
1270         }
1271         return 0;
1272 }
1273
1274 static inline int
1275 vhost_enable_notify_packed(struct virtio_net *dev,
1276                 struct vhost_virtqueue *vq, int enable)
1277 {
1278         uint16_t flags;
1279
1280         if (vq->device_event == NULL)
1281                 return -1;
1282
1283         if (!enable) {
1284                 vq->device_event->flags = VRING_EVENT_F_DISABLE;
1285                 return 0;
1286         }
1287
1288         flags = VRING_EVENT_F_ENABLE;
1289         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
1290                 flags = VRING_EVENT_F_DESC;
1291                 vq->device_event->off_wrap = vq->last_avail_idx |
1292                         vq->avail_wrap_counter << 15;
1293         }
1294
1295         rte_smp_wmb();
1296
1297         vq->device_event->flags = flags;
1298         return 0;
1299 }
1300
1301 int
1302 vhost_enable_guest_notification(struct virtio_net *dev,
1303                 struct vhost_virtqueue *vq, int enable)
1304 {
1305         /*
1306          * If the virtqueue is not ready yet, it will be applied
1307          * when it will become ready.
1308          */
1309         if (!vq->ready)
1310                 return 0;
1311
1312         if (vq_is_packed(dev))
1313                 return vhost_enable_notify_packed(dev, vq, enable);
1314         else
1315                 return vhost_enable_notify_split(dev, vq, enable);
1316 }
1317
1318 int
1319 rte_vhost_enable_guest_notification(int vid, uint16_t queue_id, int enable)
1320 {
1321         struct virtio_net *dev = get_device(vid);
1322         struct vhost_virtqueue *vq;
1323         int ret;
1324
1325         if (!dev)
1326                 return -1;
1327
1328         vq = dev->virtqueue[queue_id];
1329
1330         rte_spinlock_lock(&vq->access_lock);
1331
1332         vq->notif_enable = enable;
1333         ret = vhost_enable_guest_notification(dev, vq, enable);
1334
1335         rte_spinlock_unlock(&vq->access_lock);
1336
1337         return ret;
1338 }
1339
1340 void
1341 rte_vhost_log_write(int vid, uint64_t addr, uint64_t len)
1342 {
1343         struct virtio_net *dev = get_device(vid);
1344
1345         if (dev == NULL)
1346                 return;
1347
1348         vhost_log_write(dev, addr, len);
1349 }
1350
1351 void
1352 rte_vhost_log_used_vring(int vid, uint16_t vring_idx,
1353                          uint64_t offset, uint64_t len)
1354 {
1355         struct virtio_net *dev;
1356         struct vhost_virtqueue *vq;
1357
1358         dev = get_device(vid);
1359         if (dev == NULL)
1360                 return;
1361
1362         if (vring_idx >= VHOST_MAX_VRING)
1363                 return;
1364         vq = dev->virtqueue[vring_idx];
1365         if (!vq)
1366                 return;
1367
1368         vhost_log_used_vring(dev, vq, offset, len);
1369 }
1370
1371 uint32_t
1372 rte_vhost_rx_queue_count(int vid, uint16_t qid)
1373 {
1374         struct virtio_net *dev;
1375         struct vhost_virtqueue *vq;
1376         uint32_t ret = 0;
1377
1378         dev = get_device(vid);
1379         if (dev == NULL)
1380                 return 0;
1381
1382         if (unlikely(qid >= dev->nr_vring || (qid & 1) == 0)) {
1383                 VHOST_LOG_DATA(ERR, "(%d) %s: invalid virtqueue idx %d.\n",
1384                         dev->vid, __func__, qid);
1385                 return 0;
1386         }
1387
1388         vq = dev->virtqueue[qid];
1389         if (vq == NULL)
1390                 return 0;
1391
1392         rte_spinlock_lock(&vq->access_lock);
1393
1394         if (unlikely(vq->enabled == 0 || vq->avail == NULL))
1395                 goto out;
1396
1397         ret = *((volatile uint16_t *)&vq->avail->idx) - vq->last_avail_idx;
1398
1399 out:
1400         rte_spinlock_unlock(&vq->access_lock);
1401         return ret;
1402 }
1403
1404 struct rte_vdpa_device *
1405 rte_vhost_get_vdpa_device(int vid)
1406 {
1407         struct virtio_net *dev = get_device(vid);
1408
1409         if (dev == NULL)
1410                 return NULL;
1411
1412         return dev->vdpa_dev;
1413 }
1414
1415 int rte_vhost_get_log_base(int vid, uint64_t *log_base,
1416                 uint64_t *log_size)
1417 {
1418         struct virtio_net *dev = get_device(vid);
1419
1420         if (dev == NULL || log_base == NULL || log_size == NULL)
1421                 return -1;
1422
1423         *log_base = dev->log_base;
1424         *log_size = dev->log_size;
1425
1426         return 0;
1427 }
1428
1429 int rte_vhost_get_vring_base(int vid, uint16_t queue_id,
1430                 uint16_t *last_avail_idx, uint16_t *last_used_idx)
1431 {
1432         struct vhost_virtqueue *vq;
1433         struct virtio_net *dev = get_device(vid);
1434
1435         if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
1436                 return -1;
1437
1438         vq = dev->virtqueue[queue_id];
1439         if (!vq)
1440                 return -1;
1441
1442         if (vq_is_packed(dev)) {
1443                 *last_avail_idx = (vq->avail_wrap_counter << 15) |
1444                                   vq->last_avail_idx;
1445                 *last_used_idx = (vq->used_wrap_counter << 15) |
1446                                  vq->last_used_idx;
1447         } else {
1448                 *last_avail_idx = vq->last_avail_idx;
1449                 *last_used_idx = vq->last_used_idx;
1450         }
1451
1452         return 0;
1453 }
1454
1455 int rte_vhost_set_vring_base(int vid, uint16_t queue_id,
1456                 uint16_t last_avail_idx, uint16_t last_used_idx)
1457 {
1458         struct vhost_virtqueue *vq;
1459         struct virtio_net *dev = get_device(vid);
1460
1461         if (!dev)
1462                 return -1;
1463
1464         vq = dev->virtqueue[queue_id];
1465         if (!vq)
1466                 return -1;
1467
1468         if (vq_is_packed(dev)) {
1469                 vq->last_avail_idx = last_avail_idx & 0x7fff;
1470                 vq->avail_wrap_counter = !!(last_avail_idx & (1 << 15));
1471                 vq->last_used_idx = last_used_idx & 0x7fff;
1472                 vq->used_wrap_counter = !!(last_used_idx & (1 << 15));
1473         } else {
1474                 vq->last_avail_idx = last_avail_idx;
1475                 vq->last_used_idx = last_used_idx;
1476         }
1477
1478         return 0;
1479 }
1480
1481 int
1482 rte_vhost_get_vring_base_from_inflight(int vid,
1483                                        uint16_t queue_id,
1484                                        uint16_t *last_avail_idx,
1485                                        uint16_t *last_used_idx)
1486 {
1487         struct rte_vhost_inflight_info_packed *inflight_info;
1488         struct virtio_net *dev = get_device(vid);
1489
1490         if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
1491                 return -1;
1492
1493         if (!vq_is_packed(dev))
1494                 return -1;
1495
1496         inflight_info = dev->virtqueue[queue_id]->inflight_packed;
1497         if (!inflight_info)
1498                 return -1;
1499
1500         *last_avail_idx = (inflight_info->old_used_wrap_counter << 15) |
1501                           inflight_info->old_used_idx;
1502         *last_used_idx = *last_avail_idx;
1503
1504         return 0;
1505 }
1506
1507 int rte_vhost_extern_callback_register(int vid,
1508                 struct rte_vhost_user_extern_ops const * const ops, void *ctx)
1509 {
1510         struct virtio_net *dev = get_device(vid);
1511
1512         if (dev == NULL || ops == NULL)
1513                 return -1;
1514
1515         dev->extern_ops = *ops;
1516         dev->extern_data = ctx;
1517         return 0;
1518 }
1519
1520 int rte_vhost_async_channel_register(int vid, uint16_t queue_id,
1521                                         uint32_t features,
1522                                         struct rte_vhost_async_channel_ops *ops)
1523 {
1524         struct vhost_virtqueue *vq;
1525         struct virtio_net *dev = get_device(vid);
1526         struct rte_vhost_async_features f;
1527
1528         if (dev == NULL || ops == NULL)
1529                 return -1;
1530
1531         f.intval = features;
1532
1533         vq = dev->virtqueue[queue_id];
1534
1535         if (unlikely(vq == NULL || !dev->async_copy))
1536                 return -1;
1537
1538         /* packed queue is not supported */
1539         if (unlikely(vq_is_packed(dev) || !f.async_inorder)) {
1540                 VHOST_LOG_CONFIG(ERR,
1541                         "async copy is not supported on packed queue or non-inorder mode "
1542                         "(vid %d, qid: %d)\n", vid, queue_id);
1543                 return -1;
1544         }
1545
1546         if (unlikely(ops->check_completed_copies == NULL ||
1547                 ops->transfer_data == NULL))
1548                 return -1;
1549
1550         rte_spinlock_lock(&vq->access_lock);
1551
1552         if (unlikely(vq->async_registered)) {
1553                 VHOST_LOG_CONFIG(ERR,
1554                         "async register failed: channel already registered "
1555                         "(vid %d, qid: %d)\n", vid, queue_id);
1556                 goto reg_out;
1557         }
1558
1559         vq->async_pkts_pending = rte_malloc(NULL,
1560                         vq->size * sizeof(uintptr_t),
1561                         RTE_CACHE_LINE_SIZE);
1562         vq->async_pkts_info = rte_malloc(NULL,
1563                         vq->size * sizeof(struct async_inflight_info),
1564                         RTE_CACHE_LINE_SIZE);
1565         if (!vq->async_pkts_pending || !vq->async_pkts_info) {
1566                 if (vq->async_pkts_pending)
1567                         rte_free(vq->async_pkts_pending);
1568
1569                 if (vq->async_pkts_info)
1570                         rte_free(vq->async_pkts_info);
1571
1572                 VHOST_LOG_CONFIG(ERR,
1573                                 "async register failed: cannot allocate memory for vq data "
1574                                 "(vid %d, qid: %d)\n", vid, queue_id);
1575                 goto reg_out;
1576         }
1577
1578         vq->async_ops.check_completed_copies = ops->check_completed_copies;
1579         vq->async_ops.transfer_data = ops->transfer_data;
1580
1581         vq->async_inorder = f.async_inorder;
1582         vq->async_threshold = f.async_threshold;
1583
1584         vq->async_registered = true;
1585
1586 reg_out:
1587         rte_spinlock_unlock(&vq->access_lock);
1588
1589         return 0;
1590 }
1591
1592 int rte_vhost_async_channel_unregister(int vid, uint16_t queue_id)
1593 {
1594         struct vhost_virtqueue *vq;
1595         struct virtio_net *dev = get_device(vid);
1596         int ret = -1;
1597
1598         if (dev == NULL)
1599                 return ret;
1600
1601         vq = dev->virtqueue[queue_id];
1602
1603         if (vq == NULL)
1604                 return ret;
1605
1606         ret = 0;
1607         rte_spinlock_lock(&vq->access_lock);
1608
1609         if (!vq->async_registered)
1610                 goto out;
1611
1612         if (vq->async_pkts_inflight_n) {
1613                 VHOST_LOG_CONFIG(ERR, "Failed to unregister async channel. "
1614                         "async inflight packets must be completed before unregistration.\n");
1615                 ret = -1;
1616                 goto out;
1617         }
1618
1619         if (vq->async_pkts_pending) {
1620                 rte_free(vq->async_pkts_pending);
1621                 vq->async_pkts_pending = NULL;
1622         }
1623
1624         if (vq->async_pkts_info) {
1625                 rte_free(vq->async_pkts_info);
1626                 vq->async_pkts_info = NULL;
1627         }
1628
1629         vq->async_ops.transfer_data = NULL;
1630         vq->async_ops.check_completed_copies = NULL;
1631         vq->async_registered = false;
1632
1633 out:
1634         rte_spinlock_unlock(&vq->access_lock);
1635
1636         return ret;
1637 }
1638
1639 RTE_LOG_REGISTER(vhost_config_log_level, lib.vhost.config, INFO);
1640 RTE_LOG_REGISTER(vhost_data_log_level, lib.vhost.data, WARNING);