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         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         TAILQ_INIT(&vq->zmbuf_list);
 544 }
 545
 546 static void
 547 reset_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 548 {
 549         struct vhost_virtqueue *vq;
 550         int callfd;
 551
 552         if (vring_idx >= VHOST_MAX_VRING) {
 553                 VHOST_LOG_CONFIG(ERR,
 554                                 "Failed not init vring, out of bound (%d)\n",
 555                                 vring_idx);
 556                 return;
 557         }
 558
 559         vq = dev->virtqueue[vring_idx];
 560         callfd = vq->callfd;
 561         init_vring_queue(dev, vring_idx);
 562         vq->callfd = callfd;
 563 }
 564
 565 int
 566 alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 567 {
 568         struct vhost_virtqueue *vq;
 569
 570         vq = rte_malloc(NULL, sizeof(struct vhost_virtqueue), 0);
 571         if (vq == NULL) {
 572                 VHOST_LOG_CONFIG(ERR,
 573                         "Failed to allocate memory for vring:%u.\n", vring_idx);
 574                 return -1;
 575         }
 576
 577         dev->virtqueue[vring_idx] = vq;
 578         init_vring_queue(dev, vring_idx);
 579         rte_spinlock_init(&vq->access_lock);
 580         vq->avail_wrap_counter = 1;
 581         vq->used_wrap_counter = 1;
 582         vq->signalled_used_valid = false;
 583
 584         dev->nr_vring += 1;
 585
 586         return 0;
 587 }
 588
 589 /*
 590  * Reset some variables in device structure, while keeping few
 591  * others untouched, such as vid, ifname, nr_vring: they
 592  * should be same unless the device is removed.
 593  */
 594 void
 595 reset_device(struct virtio_net *dev)
 596 {
 597         uint32_t i;
 598
 599         dev->features = 0;
 600         dev->protocol_features = 0;
 601         dev->flags &= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 602
 603         for (i = 0; i < dev->nr_vring; i++)
 604                 reset_vring_queue(dev, i);
 605 }
 606
 607 /*
 608  * Invoked when there is a new vhost-user connection established (when
 609  * there is a new virtio device being attached).
 610  */
 611 int
 612 vhost_new_device(void)
 613 {
 614         struct virtio_net *dev;
 615         int i;
 616
 617         for (i = 0; i < MAX_VHOST_DEVICE; i++) {
 618                 if (vhost_devices[i] == NULL)
 619                         break;
 620         }
 621
 622         if (i == MAX_VHOST_DEVICE) {
 623                 VHOST_LOG_CONFIG(ERR,
 624                         "Failed to find a free slot for new device.\n");
 625                 return -1;
 626         }
 627
 628         dev = rte_zmalloc(NULL, sizeof(struct virtio_net), 0);
 629         if (dev == NULL) {
 630                 VHOST_LOG_CONFIG(ERR,
 631                         "Failed to allocate memory for new dev.\n");
 632                 return -1;
 633         }
 634
 635         vhost_devices[i] = dev;
 636         dev->vid = i;
 637         dev->flags = VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 638         dev->slave_req_fd = -1;
 639         dev->postcopy_ufd = -1;
 640         rte_spinlock_init(&dev->slave_req_lock);
 641
 642         return i;
 643 }
 644
 645 void
 646 vhost_destroy_device_notify(struct virtio_net *dev)
 647 {
 648         struct rte_vdpa_device *vdpa_dev;
 649
 650         if (dev->flags & VIRTIO_DEV_RUNNING) {
 651                 vdpa_dev = dev->vdpa_dev;
 652                 if (vdpa_dev)
 653                         vdpa_dev->ops->dev_close(dev->vid);
 654                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 655                 dev->notify_ops->destroy_device(dev->vid);
 656         }
 657 }
 658
 659 /*
 660  * Invoked when there is the vhost-user connection is broken (when
 661  * the virtio device is being detached).
 662  */
 663 void
 664 vhost_destroy_device(int vid)
 665 {
 666         struct virtio_net *dev = get_device(vid);
 667
 668         if (dev == NULL)
 669                 return;
 670
 671         vhost_destroy_device_notify(dev);
 672
 673         cleanup_device(dev, 1);
 674         free_device(dev);
 675
 676         vhost_devices[vid] = NULL;
 677 }
 678
 679 void
 680 vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *vdpa_dev)
 681 {
 682         struct virtio_net *dev = get_device(vid);
 683
 684         if (dev == NULL)
 685                 return;
 686
 687         dev->vdpa_dev = vdpa_dev;
 688 }
 689
 690 void
 691 vhost_set_ifname(int vid, const char *if_name, unsigned int if_len)
 692 {
 693         struct virtio_net *dev;
 694         unsigned int len;
 695
 696         dev = get_device(vid);
 697         if (dev == NULL)
 698                 return;
 699
 700         len = if_len > sizeof(dev->ifname) ?
 701                 sizeof(dev->ifname) : if_len;
 702
 703         strncpy(dev->ifname, if_name, len);
 704         dev->ifname[sizeof(dev->ifname) - 1] = '\0';
 705 }
 706
 707 void
 708 vhost_enable_dequeue_zero_copy(int vid)
 709 {
 710         struct virtio_net *dev = get_device(vid);
 711
 712         if (dev == NULL)
 713                 return;
 714
 715         dev->dequeue_zero_copy = 1;
 716         VHOST_LOG_CONFIG(INFO, "dequeue zero copy is enabled\n");
 717 }
 718
 719 void
 720 vhost_set_builtin_virtio_net(int vid, bool enable)
 721 {
 722         struct virtio_net *dev = get_device(vid);
 723
 724         if (dev == NULL)
 725                 return;
 726
 727         if (enable)
 728                 dev->flags |= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 729         else
 730                 dev->flags &= ~VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 731 }
 732
 733 void
 734 vhost_enable_extbuf(int vid)
 735 {
 736         struct virtio_net *dev = get_device(vid);
 737
 738         if (dev == NULL)
 739                 return;
 740
 741         dev->extbuf = 1;
 742 }
 743
 744 void
 745 vhost_enable_linearbuf(int vid)
 746 {
 747         struct virtio_net *dev = get_device(vid);
 748
 749         if (dev == NULL)
 750                 return;
 751
 752         dev->linearbuf = 1;
 753 }
 754
 755 int
 756 rte_vhost_get_mtu(int vid, uint16_t *mtu)
 757 {
 758         struct virtio_net *dev = get_device(vid);
 759
 760         if (dev == NULL || mtu == NULL)
 761                 return -ENODEV;
 762
 763         if (!(dev->flags & VIRTIO_DEV_READY))
 764                 return -EAGAIN;
 765
 766         if (!(dev->features & (1ULL << VIRTIO_NET_F_MTU)))
 767                 return -ENOTSUP;
 768
 769         *mtu = dev->mtu;
 770
 771         return 0;
 772 }
 773
 774 int
 775 rte_vhost_get_numa_node(int vid)
 776 {
 777 #ifdef RTE_LIBRTE_VHOST_NUMA
 778         struct virtio_net *dev = get_device(vid);
 779         int numa_node;
 780         int ret;
 781
 782         if (dev == NULL || numa_available() != 0)
 783                 return -1;
 784
 785         ret = get_mempolicy(&numa_node, NULL, 0, dev,
 786                             MPOL_F_NODE | MPOL_F_ADDR);
 787         if (ret < 0) {
 788                 VHOST_LOG_CONFIG(ERR,
 789                         "(%d) failed to query numa node: %s\n",
 790                         vid, rte_strerror(errno));
 791                 return -1;
 792         }
 793
 794         return numa_node;
 795 #else
 796         RTE_SET_USED(vid);
 797         return -1;
 798 #endif
 799 }
 800
 801 uint32_t
 802 rte_vhost_get_queue_num(int vid)
 803 {
 804         struct virtio_net *dev = get_device(vid);
 805
 806         if (dev == NULL)
 807                 return 0;
 808
 809         return dev->nr_vring / 2;
 810 }
 811
 812 uint16_t
 813 rte_vhost_get_vring_num(int vid)
 814 {
 815         struct virtio_net *dev = get_device(vid);
 816
 817         if (dev == NULL)
 818                 return 0;
 819
 820         return dev->nr_vring;
 821 }
 822
 823 int
 824 rte_vhost_get_ifname(int vid, char *buf, size_t len)
 825 {
 826         struct virtio_net *dev = get_device(vid);
 827
 828         if (dev == NULL || buf == NULL)
 829                 return -1;
 830
 831         len = RTE_MIN(len, sizeof(dev->ifname));
 832
 833         strncpy(buf, dev->ifname, len);
 834         buf[len - 1] = '\0';
 835
 836         return 0;
 837 }
 838
 839 int
 840 rte_vhost_get_negotiated_features(int vid, uint64_t *features)
 841 {
 842         struct virtio_net *dev;
 843
 844         dev = get_device(vid);
 845         if (dev == NULL || features == NULL)
 846                 return -1;
 847
 848         *features = dev->features;
 849         return 0;
 850 }
 851
 852 int
 853 rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem)
 854 {
 855         struct virtio_net *dev;
 856         struct rte_vhost_memory *m;
 857         size_t size;
 858
 859         dev = get_device(vid);
 860         if (dev == NULL || mem == NULL)
 861                 return -1;
 862
 863         size = dev->mem->nregions * sizeof(struct rte_vhost_mem_region);
 864         m = malloc(sizeof(struct rte_vhost_memory) + size);
 865         if (!m)
 866                 return -1;
 867
 868         m->nregions = dev->mem->nregions;
 869         memcpy(m->regions, dev->mem->regions, size);
 870         *mem = m;
 871
 872         return 0;
 873 }
 874
 875 int
 876 rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
 877                           struct rte_vhost_vring *vring)
 878 {
 879         struct virtio_net *dev;
 880         struct vhost_virtqueue *vq;
 881
 882         dev = get_device(vid);
 883         if (dev == NULL || vring == NULL)
 884                 return -1;
 885
 886         if (vring_idx >= VHOST_MAX_VRING)
 887                 return -1;
 888
 889         vq = dev->virtqueue[vring_idx];
 890         if (!vq)
 891                 return -1;
 892
 893         if (vq_is_packed(dev)) {
 894                 vring->desc_packed = vq->desc_packed;
 895                 vring->driver_event = vq->driver_event;
 896                 vring->device_event = vq->device_event;
 897         } else {
 898                 vring->desc = vq->desc;
 899                 vring->avail = vq->avail;
 900                 vring->used = vq->used;
 901         }
 902         vring->log_guest_addr  = vq->log_guest_addr;
 903
 904         vring->callfd  = vq->callfd;
 905         vring->kickfd  = vq->kickfd;
 906         vring->size    = vq->size;
 907
 908         return 0;
 909 }
 910
 911 int
 912 rte_vhost_get_vhost_ring_inflight(int vid, uint16_t vring_idx,
 913                                   struct rte_vhost_ring_inflight *vring)
 914 {
 915         struct virtio_net *dev;
 916         struct vhost_virtqueue *vq;
 917
 918         dev = get_device(vid);
 919         if (unlikely(!dev))
 920                 return -1;
 921
 922         if (vring_idx >= VHOST_MAX_VRING)
 923                 return -1;
 924
 925         vq = dev->virtqueue[vring_idx];
 926         if (unlikely(!vq))
 927                 return -1;
 928
 929         if (vq_is_packed(dev)) {
 930                 if (unlikely(!vq->inflight_packed))
 931                         return -1;
 932
 933                 vring->inflight_packed = vq->inflight_packed;
 934         } else {
 935                 if (unlikely(!vq->inflight_split))
 936                         return -1;
 937
 938                 vring->inflight_split = vq->inflight_split;
 939         }
 940
 941         vring->resubmit_inflight = vq->resubmit_inflight;
 942
 943         return 0;
 944 }
 945
 946 int
 947 rte_vhost_set_inflight_desc_split(int vid, uint16_t vring_idx,
 948                                   uint16_t idx)
 949 {
 950         struct vhost_virtqueue *vq;
 951         struct virtio_net *dev;
 952
 953         dev = get_device(vid);
 954         if (unlikely(!dev))
 955                 return -1;
 956
 957         if (unlikely(!(dev->protocol_features &
 958             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
 959                 return 0;
 960
 961         if (unlikely(vq_is_packed(dev)))
 962                 return -1;
 963
 964         if (unlikely(vring_idx >= VHOST_MAX_VRING))
 965                 return -1;
 966
 967         vq = dev->virtqueue[vring_idx];
 968         if (unlikely(!vq))
 969                 return -1;
 970
 971         if (unlikely(!vq->inflight_split))
 972                 return -1;
 973
 974         if (unlikely(idx >= vq->size))
 975                 return -1;
 976
 977         vq->inflight_split->desc[idx].counter = vq->global_counter++;
 978         vq->inflight_split->desc[idx].inflight = 1;
 979         return 0;
 980 }
 981
 982 int
 983 rte_vhost_set_inflight_desc_packed(int vid, uint16_t vring_idx,
 984                                    uint16_t head, uint16_t last,
 985                                    uint16_t *inflight_entry)
 986 {
 987         struct rte_vhost_inflight_info_packed *inflight_info;
 988         struct virtio_net *dev;
 989         struct vhost_virtqueue *vq;
 990         struct vring_packed_desc *desc;
 991         uint16_t old_free_head, free_head;
 992
 993         dev = get_device(vid);
 994         if (unlikely(!dev))
 995                 return -1;
 996
 997         if (unlikely(!(dev->protocol_features &
 998             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
 999                 return 0;
1000
1001         if (unlikely(!vq_is_packed(dev)))
1002                 return -1;
1003
1004         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1005                 return -1;
1006
1007         vq = dev->virtqueue[vring_idx];
1008         if (unlikely(!vq))
1009                 return -1;
1010
1011         inflight_info = vq->inflight_packed;
1012         if (unlikely(!inflight_info))
1013                 return -1;
1014
1015         if (unlikely(head >= vq->size))
1016                 return -1;
1017
1018         desc = vq->desc_packed;
1019         old_free_head = inflight_info->old_free_head;
1020         if (unlikely(old_free_head >= vq->size))
1021                 return -1;
1022
1023         free_head = old_free_head;
1024
1025         /* init header descriptor */
1026         inflight_info->desc[old_free_head].num = 0;
1027         inflight_info->desc[old_free_head].counter = vq->global_counter++;
1028         inflight_info->desc[old_free_head].inflight = 1;
1029
1030         /* save desc entry in flight entry */
1031         while (head != ((last + 1) % vq->size)) {
1032                 inflight_info->desc[old_free_head].num++;
1033                 inflight_info->desc[free_head].addr = desc[head].addr;
1034                 inflight_info->desc[free_head].len = desc[head].len;
1035                 inflight_info->desc[free_head].flags = desc[head].flags;
1036                 inflight_info->desc[free_head].id = desc[head].id;
1037
1038                 inflight_info->desc[old_free_head].last = free_head;
1039                 free_head = inflight_info->desc[free_head].next;
1040                 inflight_info->free_head = free_head;
1041                 head = (head + 1) % vq->size;
1042         }
1043
1044         inflight_info->old_free_head = free_head;
1045         *inflight_entry = old_free_head;
1046
1047         return 0;
1048 }
1049
1050 int
1051 rte_vhost_clr_inflight_desc_split(int vid, uint16_t vring_idx,
1052                                   uint16_t last_used_idx, uint16_t idx)
1053 {
1054         struct virtio_net *dev;
1055         struct vhost_virtqueue *vq;
1056
1057         dev = get_device(vid);
1058         if (unlikely(!dev))
1059                 return -1;
1060
1061         if (unlikely(!(dev->protocol_features &
1062             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1063                 return 0;
1064
1065         if (unlikely(vq_is_packed(dev)))
1066                 return -1;
1067
1068         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1069                 return -1;
1070
1071         vq = dev->virtqueue[vring_idx];
1072         if (unlikely(!vq))
1073                 return -1;
1074
1075         if (unlikely(!vq->inflight_split))
1076                 return -1;
1077
1078         if (unlikely(idx >= vq->size))
1079                 return -1;
1080
1081         rte_smp_mb();
1082
1083         vq->inflight_split->desc[idx].inflight = 0;
1084
1085         rte_smp_mb();
1086
1087         vq->inflight_split->used_idx = last_used_idx;
1088         return 0;
1089 }
1090
1091 int
1092 rte_vhost_clr_inflight_desc_packed(int vid, uint16_t vring_idx,
1093                                    uint16_t head)
1094 {
1095         struct rte_vhost_inflight_info_packed *inflight_info;
1096         struct virtio_net *dev;
1097         struct vhost_virtqueue *vq;
1098
1099         dev = get_device(vid);
1100         if (unlikely(!dev))
1101                 return -1;
1102
1103         if (unlikely(!(dev->protocol_features &
1104             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1105                 return 0;
1106
1107         if (unlikely(!vq_is_packed(dev)))
1108                 return -1;
1109
1110         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1111                 return -1;
1112
1113         vq = dev->virtqueue[vring_idx];
1114         if (unlikely(!vq))
1115                 return -1;
1116
1117         inflight_info = vq->inflight_packed;
1118         if (unlikely(!inflight_info))
1119                 return -1;
1120
1121         if (unlikely(head >= vq->size))
1122                 return -1;
1123
1124         rte_smp_mb();
1125
1126         inflight_info->desc[head].inflight = 0;
1127
1128         rte_smp_mb();
1129
1130         inflight_info->old_free_head = inflight_info->free_head;
1131         inflight_info->old_used_idx = inflight_info->used_idx;
1132         inflight_info->old_used_wrap_counter = inflight_info->used_wrap_counter;
1133
1134         return 0;
1135 }
1136
1137 int
1138 rte_vhost_set_last_inflight_io_split(int vid, uint16_t vring_idx,
1139                                      uint16_t idx)
1140 {
1141         struct virtio_net *dev;
1142         struct vhost_virtqueue *vq;
1143
1144         dev = get_device(vid);
1145         if (unlikely(!dev))
1146                 return -1;
1147
1148         if (unlikely(!(dev->protocol_features &
1149             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1150                 return 0;
1151
1152         if (unlikely(vq_is_packed(dev)))
1153                 return -1;
1154
1155         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1156                 return -1;
1157
1158         vq = dev->virtqueue[vring_idx];
1159         if (unlikely(!vq))
1160                 return -1;
1161
1162         if (unlikely(!vq->inflight_split))
1163                 return -1;
1164
1165         vq->inflight_split->last_inflight_io = idx;
1166         return 0;
1167 }
1168
1169 int
1170 rte_vhost_set_last_inflight_io_packed(int vid, uint16_t vring_idx,
1171                                       uint16_t head)
1172 {
1173         struct rte_vhost_inflight_info_packed *inflight_info;
1174         struct virtio_net *dev;
1175         struct vhost_virtqueue *vq;
1176         uint16_t last;
1177
1178         dev = get_device(vid);
1179         if (unlikely(!dev))
1180                 return -1;
1181
1182         if (unlikely(!(dev->protocol_features &
1183             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1184                 return 0;
1185
1186         if (unlikely(!vq_is_packed(dev)))
1187                 return -1;
1188
1189         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1190                 return -1;
1191
1192         vq = dev->virtqueue[vring_idx];
1193         if (unlikely(!vq))
1194                 return -1;
1195
1196         inflight_info = vq->inflight_packed;
1197         if (unlikely(!inflight_info))
1198                 return -1;
1199
1200         if (unlikely(head >= vq->size))
1201                 return -1;
1202
1203         last = inflight_info->desc[head].last;
1204         if (unlikely(last >= vq->size))
1205                 return -1;
1206
1207         inflight_info->desc[last].next = inflight_info->free_head;
1208         inflight_info->free_head = head;
1209         inflight_info->used_idx += inflight_info->desc[head].num;
1210         if (inflight_info->used_idx >= inflight_info->desc_num) {
1211                 inflight_info->used_idx -= inflight_info->desc_num;
1212                 inflight_info->used_wrap_counter =
1213                         !inflight_info->used_wrap_counter;
1214         }
1215
1216         return 0;
1217 }
1218
1219 int
1220 rte_vhost_vring_call(int vid, uint16_t vring_idx)
1221 {
1222         struct virtio_net *dev;
1223         struct vhost_virtqueue *vq;
1224
1225         dev = get_device(vid);
1226         if (!dev)
1227                 return -1;
1228
1229         if (vring_idx >= VHOST_MAX_VRING)
1230                 return -1;
1231
1232         vq = dev->virtqueue[vring_idx];
1233         if (!vq)
1234                 return -1;
1235
1236         if (vq_is_packed(dev))
1237                 vhost_vring_call_packed(dev, vq);
1238         else
1239                 vhost_vring_call_split(dev, vq);
1240
1241         return 0;
1242 }
1243
1244 uint16_t
1245 rte_vhost_avail_entries(int vid, uint16_t queue_id)
1246 {
1247         struct virtio_net *dev;
1248         struct vhost_virtqueue *vq;
1249         uint16_t ret = 0;
1250
1251         dev = get_device(vid);
1252         if (!dev)
1253                 return 0;
1254
1255         vq = dev->virtqueue[queue_id];
1256
1257         rte_spinlock_lock(&vq->access_lock);
1258
1259         if (unlikely(!vq->enabled || vq->avail == NULL))
1260                 goto out;
1261
1262         ret = *(volatile uint16_t *)&vq->avail->idx - vq->last_used_idx;
1263
1264 out:
1265         rte_spinlock_unlock(&vq->access_lock);
1266         return ret;
1267 }
1268
1269 static inline int
1270 vhost_enable_notify_split(struct virtio_net *dev,
1271                 struct vhost_virtqueue *vq, int enable)
1272 {
1273         if (vq->used == NULL)
1274                 return -1;
1275
1276         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
1277                 if (enable)
1278                         vq->used->flags &= ~VRING_USED_F_NO_NOTIFY;
1279                 else
1280                         vq->used->flags |= VRING_USED_F_NO_NOTIFY;
1281         } else {
1282                 if (enable)
1283                         vhost_avail_event(vq) = vq->last_avail_idx;
1284         }
1285         return 0;
1286 }
1287
1288 static inline int
1289 vhost_enable_notify_packed(struct virtio_net *dev,
1290                 struct vhost_virtqueue *vq, int enable)
1291 {
1292         uint16_t flags;
1293
1294         if (vq->device_event == NULL)
1295                 return -1;
1296
1297         if (!enable) {
1298                 vq->device_event->flags = VRING_EVENT_F_DISABLE;
1299                 return 0;
1300         }
1301
1302         flags = VRING_EVENT_F_ENABLE;
1303         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
1304                 flags = VRING_EVENT_F_DESC;
1305                 vq->device_event->off_wrap = vq->last_avail_idx |
1306                         vq->avail_wrap_counter << 15;
1307         }
1308
1309         rte_smp_wmb();
1310
1311         vq->device_event->flags = flags;
1312         return 0;
1313 }
1314
1315 int
1316 vhost_enable_guest_notification(struct virtio_net *dev,
1317                 struct vhost_virtqueue *vq, int enable)
1318 {
1319         /*
1320          * If the virtqueue is not ready yet, it will be applied
1321          * when it will become ready.
1322          */
1323         if (!vq->ready)
1324                 return 0;
1325
1326         if (vq_is_packed(dev))
1327                 return vhost_enable_notify_packed(dev, vq, enable);
1328         else
1329                 return vhost_enable_notify_split(dev, vq, enable);
1330 }
1331
1332 int
1333 rte_vhost_enable_guest_notification(int vid, uint16_t queue_id, int enable)
1334 {
1335         struct virtio_net *dev = get_device(vid);
1336         struct vhost_virtqueue *vq;
1337         int ret;
1338
1339         if (!dev)
1340                 return -1;
1341
1342         vq = dev->virtqueue[queue_id];
1343
1344         rte_spinlock_lock(&vq->access_lock);
1345
1346         vq->notif_enable = enable;
1347         ret = vhost_enable_guest_notification(dev, vq, enable);
1348
1349         rte_spinlock_unlock(&vq->access_lock);
1350
1351         return ret;
1352 }
1353
1354 void
1355 rte_vhost_log_write(int vid, uint64_t addr, uint64_t len)
1356 {
1357         struct virtio_net *dev = get_device(vid);
1358
1359         if (dev == NULL)
1360                 return;
1361
1362         vhost_log_write(dev, addr, len);
1363 }
1364
1365 void
1366 rte_vhost_log_used_vring(int vid, uint16_t vring_idx,
1367                          uint64_t offset, uint64_t len)
1368 {
1369         struct virtio_net *dev;
1370         struct vhost_virtqueue *vq;
1371
1372         dev = get_device(vid);
1373         if (dev == NULL)
1374                 return;
1375
1376         if (vring_idx >= VHOST_MAX_VRING)
1377                 return;
1378         vq = dev->virtqueue[vring_idx];
1379         if (!vq)
1380                 return;
1381
1382         vhost_log_used_vring(dev, vq, offset, len);
1383 }
1384
1385 uint32_t
1386 rte_vhost_rx_queue_count(int vid, uint16_t qid)
1387 {
1388         struct virtio_net *dev;
1389         struct vhost_virtqueue *vq;
1390         uint32_t ret = 0;
1391
1392         dev = get_device(vid);
1393         if (dev == NULL)
1394                 return 0;
1395
1396         if (unlikely(qid >= dev->nr_vring || (qid & 1) == 0)) {
1397                 VHOST_LOG_DATA(ERR, "(%d) %s: invalid virtqueue idx %d.\n",
1398                         dev->vid, __func__, qid);
1399                 return 0;
1400         }
1401
1402         vq = dev->virtqueue[qid];
1403         if (vq == NULL)
1404                 return 0;
1405
1406         rte_spinlock_lock(&vq->access_lock);
1407
1408         if (unlikely(vq->enabled == 0 || vq->avail == NULL))
1409                 goto out;
1410
1411         ret = *((volatile uint16_t *)&vq->avail->idx) - vq->last_avail_idx;
1412
1413 out:
1414         rte_spinlock_unlock(&vq->access_lock);
1415         return ret;
1416 }
1417
1418 struct rte_vdpa_device *
1419 rte_vhost_get_vdpa_device(int vid)
1420 {
1421         struct virtio_net *dev = get_device(vid);
1422
1423         if (dev == NULL)
1424                 return NULL;
1425
1426         return dev->vdpa_dev;
1427 }
1428
1429 int rte_vhost_get_log_base(int vid, uint64_t *log_base,
1430                 uint64_t *log_size)
1431 {
1432         struct virtio_net *dev = get_device(vid);
1433
1434         if (dev == NULL || log_base == NULL || log_size == NULL)
1435                 return -1;
1436
1437         *log_base = dev->log_base;
1438         *log_size = dev->log_size;
1439
1440         return 0;
1441 }
1442
1443 int rte_vhost_get_vring_base(int vid, uint16_t queue_id,
1444                 uint16_t *last_avail_idx, uint16_t *last_used_idx)
1445 {
1446         struct vhost_virtqueue *vq;
1447         struct virtio_net *dev = get_device(vid);
1448
1449         if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
1450                 return -1;
1451
1452         vq = dev->virtqueue[queue_id];
1453         if (!vq)
1454                 return -1;
1455
1456         if (vq_is_packed(dev)) {
1457                 *last_avail_idx = (vq->avail_wrap_counter << 15) |
1458                                   vq->last_avail_idx;
1459                 *last_used_idx = (vq->used_wrap_counter << 15) |
1460                                  vq->last_used_idx;
1461         } else {
1462                 *last_avail_idx = vq->last_avail_idx;
1463                 *last_used_idx = vq->last_used_idx;
1464         }
1465
1466         return 0;
1467 }
1468
1469 int rte_vhost_set_vring_base(int vid, uint16_t queue_id,
1470                 uint16_t last_avail_idx, uint16_t last_used_idx)
1471 {
1472         struct vhost_virtqueue *vq;
1473         struct virtio_net *dev = get_device(vid);
1474
1475         if (!dev)
1476                 return -1;
1477
1478         vq = dev->virtqueue[queue_id];
1479         if (!vq)
1480                 return -1;
1481
1482         if (vq_is_packed(dev)) {
1483                 vq->last_avail_idx = last_avail_idx & 0x7fff;
1484                 vq->avail_wrap_counter = !!(last_avail_idx & (1 << 15));
1485                 vq->last_used_idx = last_used_idx & 0x7fff;
1486                 vq->used_wrap_counter = !!(last_used_idx & (1 << 15));
1487         } else {
1488                 vq->last_avail_idx = last_avail_idx;
1489                 vq->last_used_idx = last_used_idx;
1490         }
1491
1492         return 0;
1493 }
1494
1495 int
1496 rte_vhost_get_vring_base_from_inflight(int vid,
1497                                        uint16_t queue_id,
1498                                        uint16_t *last_avail_idx,
1499                                        uint16_t *last_used_idx)
1500 {
1501         struct rte_vhost_inflight_info_packed *inflight_info;
1502         struct virtio_net *dev = get_device(vid);
1503
1504         if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
1505                 return -1;
1506
1507         if (!vq_is_packed(dev))
1508                 return -1;
1509
1510         inflight_info = dev->virtqueue[queue_id]->inflight_packed;
1511         if (!inflight_info)
1512                 return -1;
1513
1514         *last_avail_idx = (inflight_info->old_used_wrap_counter << 15) |
1515                           inflight_info->old_used_idx;
1516         *last_used_idx = *last_avail_idx;
1517
1518         return 0;
1519 }
1520
1521 int rte_vhost_extern_callback_register(int vid,
1522                 struct rte_vhost_user_extern_ops const * const ops, void *ctx)
1523 {
1524         struct virtio_net *dev = get_device(vid);
1525
1526         if (dev == NULL || ops == NULL)
1527                 return -1;
1528
1529         dev->extern_ops = *ops;
1530         dev->extern_data = ctx;
1531         return 0;
1532 }
1533
1534 int rte_vhost_async_channel_register(int vid, uint16_t queue_id,
1535                                         uint32_t features,
1536                                         struct rte_vhost_async_channel_ops *ops)
1537 {
1538         struct vhost_virtqueue *vq;
1539         struct virtio_net *dev = get_device(vid);
1540         struct rte_vhost_async_features f;
1541
1542         if (dev == NULL || ops == NULL)
1543                 return -1;
1544
1545         f.intval = features;
1546
1547         vq = dev->virtqueue[queue_id];
1548
1549         if (unlikely(vq == NULL || !dev->async_copy))
1550                 return -1;
1551
1552         /* packed queue is not supported */
1553         if (unlikely(vq_is_packed(dev) || !f.async_inorder)) {
1554                 VHOST_LOG_CONFIG(ERR,
1555                         "async copy is not supported on packed queue or non-inorder mode "
1556                         "(vid %d, qid: %d)\n", vid, queue_id);
1557                 return -1;
1558         }
1559
1560         if (unlikely(ops->check_completed_copies == NULL ||
1561                 ops->transfer_data == NULL))
1562                 return -1;
1563
1564         rte_spinlock_lock(&vq->access_lock);
1565
1566         if (unlikely(vq->async_registered)) {
1567                 VHOST_LOG_CONFIG(ERR,
1568                         "async register failed: channel already registered "
1569                         "(vid %d, qid: %d)\n", vid, queue_id);
1570                 goto reg_out;
1571         }
1572
1573         vq->async_pkts_pending = rte_malloc(NULL,
1574                         vq->size * sizeof(uintptr_t),
1575                         RTE_CACHE_LINE_SIZE);
1576         vq->async_pending_info = rte_malloc(NULL,
1577                         vq->size * sizeof(uint64_t),
1578                         RTE_CACHE_LINE_SIZE);
1579         if (!vq->async_pkts_pending || !vq->async_pending_info) {
1580                 if (vq->async_pkts_pending)
1581                         rte_free(vq->async_pkts_pending);
1582
1583                 if (vq->async_pending_info)
1584                         rte_free(vq->async_pending_info);
1585
1586                 VHOST_LOG_CONFIG(ERR,
1587                                 "async register failed: cannot allocate memory for vq data "
1588                                 "(vid %d, qid: %d)\n", vid, queue_id);
1589                 goto reg_out;
1590         }
1591
1592         vq->async_ops.check_completed_copies = ops->check_completed_copies;
1593         vq->async_ops.transfer_data = ops->transfer_data;
1594
1595         vq->async_inorder = f.async_inorder;
1596         vq->async_threshold = f.async_threshold;
1597
1598         vq->async_registered = true;
1599
1600 reg_out:
1601         rte_spinlock_unlock(&vq->access_lock);
1602
1603         return 0;
1604 }
1605
1606 int rte_vhost_async_channel_unregister(int vid, uint16_t queue_id)
1607 {
1608         struct vhost_virtqueue *vq;
1609         struct virtio_net *dev = get_device(vid);
1610         int ret = -1;
1611
1612         if (dev == NULL)
1613                 return ret;
1614
1615         vq = dev->virtqueue[queue_id];
1616
1617         if (vq == NULL)
1618                 return ret;
1619
1620         ret = 0;
1621         rte_spinlock_lock(&vq->access_lock);
1622
1623         if (!vq->async_registered)
1624                 goto out;
1625
1626         if (vq->async_pkts_inflight_n) {
1627                 VHOST_LOG_CONFIG(ERR, "Failed to unregister async channel. "
1628                         "async inflight packets must be completed before unregistration.\n");
1629                 ret = -1;
1630                 goto out;
1631         }
1632
1633         if (vq->async_pkts_pending) {
1634                 rte_free(vq->async_pkts_pending);
1635                 vq->async_pkts_pending = NULL;
1636         }
1637
1638         if (vq->async_pending_info) {
1639                 rte_free(vq->async_pending_info);
1640                 vq->async_pending_info = NULL;
1641         }
1642
1643         vq->async_ops.transfer_data = NULL;
1644         vq->async_ops.check_completed_copies = NULL;
1645         vq->async_registered = false;
1646
1647 out:
1648         rte_spinlock_unlock(&vq->access_lock);
1649
1650         return ret;
1651 }
1652
1653 RTE_LOG_REGISTER(vhost_config_log_level, lib.vhost.config, INFO);
1654 RTE_LOG_REGISTER(vhost_data_log_level, lib.vhost.data, WARNING);