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