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 static void
 328 vhost_free_async_mem(struct vhost_virtqueue *vq)
 329 {
 330         if (vq->async_pkts_pending)
 331                 rte_free(vq->async_pkts_pending);
 332         if (vq->async_pkts_info)
 333                 rte_free(vq->async_pkts_info);
 334         if (vq->it_pool)
 335                 rte_free(vq->it_pool);
 336         if (vq->vec_pool)
 337                 rte_free(vq->vec_pool);
 338
 339         vq->async_pkts_pending = NULL;
 340         vq->async_pkts_info = NULL;
 341         vq->it_pool = NULL;
 342         vq->vec_pool = NULL;
 343 }
 344
 345 void
 346 free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq)
 347 {
 348         if (vq_is_packed(dev))
 349                 rte_free(vq->shadow_used_packed);
 350         else {
 351                 rte_free(vq->shadow_used_split);
 352                 vhost_free_async_mem(vq);
 353         }
 354         rte_free(vq->batch_copy_elems);
 355         rte_mempool_free(vq->iotlb_pool);
 356         rte_free(vq);
 357 }
 358
 359 /*
 360  * Release virtqueues and device memory.
 361  */
 362 static void
 363 free_device(struct virtio_net *dev)
 364 {
 365         uint32_t i;
 366
 367         for (i = 0; i < dev->nr_vring; i++)
 368                 free_vq(dev, dev->virtqueue[i]);
 369
 370         rte_free(dev);
 371 }
 372
 373 static __rte_always_inline int
 374 log_translate(struct virtio_net *dev, struct vhost_virtqueue *vq)
 375 {
 376         if (likely(!(vq->ring_addrs.flags & (1 << VHOST_VRING_F_LOG))))
 377                 return 0;
 378
 379         vq->log_guest_addr = translate_log_addr(dev, vq,
 380                                                 vq->ring_addrs.log_guest_addr);
 381         if (vq->log_guest_addr == 0)
 382                 return -1;
 383
 384         return 0;
 385 }
 386
 387 /*
 388  * Converts vring log address to GPA
 389  * If IOMMU is enabled, the log address is IOVA
 390  * If IOMMU not enabled, the log address is already GPA
 391  *
 392  * Caller should have iotlb_lock read-locked
 393  */
 394 uint64_t
 395 translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
 396                 uint64_t log_addr)
 397 {
 398         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 399                 const uint64_t exp_size = sizeof(uint64_t);
 400                 uint64_t hva, gpa;
 401                 uint64_t size = exp_size;
 402
 403                 hva = vhost_iova_to_vva(dev, vq, log_addr,
 404                                         &size, VHOST_ACCESS_RW);
 405
 406                 if (size != exp_size)
 407                         return 0;
 408
 409                 gpa = hva_to_gpa(dev, hva, exp_size);
 410                 if (!gpa) {
 411                         VHOST_LOG_CONFIG(ERR,
 412                                 "VQ: Failed to find GPA for log_addr: 0x%"
 413                                 PRIx64 " hva: 0x%" PRIx64 "\n",
 414                                 log_addr, hva);
 415                         return 0;
 416                 }
 417                 return gpa;
 418
 419         } else
 420                 return log_addr;
 421 }
 422
 423 /* Caller should have iotlb_lock read-locked */
 424 static int
 425 vring_translate_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 426 {
 427         uint64_t req_size, size;
 428
 429         req_size = sizeof(struct vring_desc) * vq->size;
 430         size = req_size;
 431         vq->desc = (struct vring_desc *)(uintptr_t)vhost_iova_to_vva(dev, vq,
 432                                                 vq->ring_addrs.desc_user_addr,
 433                                                 &size, VHOST_ACCESS_RW);
 434         if (!vq->desc || size != req_size)
 435                 return -1;
 436
 437         req_size = sizeof(struct vring_avail);
 438         req_size += sizeof(uint16_t) * vq->size;
 439         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 440                 req_size += sizeof(uint16_t);
 441         size = req_size;
 442         vq->avail = (struct vring_avail *)(uintptr_t)vhost_iova_to_vva(dev, vq,
 443                                                 vq->ring_addrs.avail_user_addr,
 444                                                 &size, VHOST_ACCESS_RW);
 445         if (!vq->avail || size != req_size)
 446                 return -1;
 447
 448         req_size = sizeof(struct vring_used);
 449         req_size += sizeof(struct vring_used_elem) * vq->size;
 450         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 451                 req_size += sizeof(uint16_t);
 452         size = req_size;
 453         vq->used = (struct vring_used *)(uintptr_t)vhost_iova_to_vva(dev, vq,
 454                                                 vq->ring_addrs.used_user_addr,
 455                                                 &size, VHOST_ACCESS_RW);
 456         if (!vq->used || size != req_size)
 457                 return -1;
 458
 459         return 0;
 460 }
 461
 462 /* Caller should have iotlb_lock read-locked */
 463 static int
 464 vring_translate_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
 465 {
 466         uint64_t req_size, size;
 467
 468         req_size = sizeof(struct vring_packed_desc) * vq->size;
 469         size = req_size;
 470         vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
 471                 vhost_iova_to_vva(dev, vq, vq->ring_addrs.desc_user_addr,
 472                                 &size, VHOST_ACCESS_RW);
 473         if (!vq->desc_packed || size != req_size)
 474                 return -1;
 475
 476         req_size = sizeof(struct vring_packed_desc_event);
 477         size = req_size;
 478         vq->driver_event = (struct vring_packed_desc_event *)(uintptr_t)
 479                 vhost_iova_to_vva(dev, vq, vq->ring_addrs.avail_user_addr,
 480                                 &size, VHOST_ACCESS_RW);
 481         if (!vq->driver_event || size != req_size)
 482                 return -1;
 483
 484         req_size = sizeof(struct vring_packed_desc_event);
 485         size = req_size;
 486         vq->device_event = (struct vring_packed_desc_event *)(uintptr_t)
 487                 vhost_iova_to_vva(dev, vq, vq->ring_addrs.used_user_addr,
 488                                 &size, VHOST_ACCESS_RW);
 489         if (!vq->device_event || size != req_size)
 490                 return -1;
 491
 492         return 0;
 493 }
 494
 495 int
 496 vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq)
 497 {
 498
 499         if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 500                 return -1;
 501
 502         if (vq_is_packed(dev)) {
 503                 if (vring_translate_packed(dev, vq) < 0)
 504                         return -1;
 505         } else {
 506                 if (vring_translate_split(dev, vq) < 0)
 507                         return -1;
 508         }
 509
 510         if (log_translate(dev, vq) < 0)
 511                 return -1;
 512
 513         vq->access_ok = 1;
 514
 515         return 0;
 516 }
 517
 518 void
 519 vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq)
 520 {
 521         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 522                 vhost_user_iotlb_wr_lock(vq);
 523
 524         vq->access_ok = 0;
 525         vq->desc = NULL;
 526         vq->avail = NULL;
 527         vq->used = NULL;
 528         vq->log_guest_addr = 0;
 529
 530         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 531                 vhost_user_iotlb_wr_unlock(vq);
 532 }
 533
 534 static void
 535 init_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 536 {
 537         struct vhost_virtqueue *vq;
 538
 539         if (vring_idx >= VHOST_MAX_VRING) {
 540                 VHOST_LOG_CONFIG(ERR,
 541                                 "Failed not init vring, out of bound (%d)\n",
 542                                 vring_idx);
 543                 return;
 544         }
 545
 546         vq = dev->virtqueue[vring_idx];
 547
 548         memset(vq, 0, sizeof(struct vhost_virtqueue));
 549
 550         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 551         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
 552         vq->notif_enable = VIRTIO_UNINITIALIZED_NOTIF;
 553
 554         vhost_user_iotlb_init(dev, vring_idx);
 555         /* Backends are set to -1 indicating an inactive device. */
 556         vq->backend = -1;
 557 }
 558
 559 static void
 560 reset_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 561 {
 562         struct vhost_virtqueue *vq;
 563         int callfd;
 564
 565         if (vring_idx >= VHOST_MAX_VRING) {
 566                 VHOST_LOG_CONFIG(ERR,
 567                                 "Failed not init vring, out of bound (%d)\n",
 568                                 vring_idx);
 569                 return;
 570         }
 571
 572         vq = dev->virtqueue[vring_idx];
 573         callfd = vq->callfd;
 574         init_vring_queue(dev, vring_idx);
 575         vq->callfd = callfd;
 576 }
 577
 578 int
 579 alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 580 {
 581         struct vhost_virtqueue *vq;
 582         uint32_t i;
 583
 584         /* Also allocate holes, if any, up to requested vring index. */
 585         for (i = 0; i <= vring_idx; i++) {
 586                 if (dev->virtqueue[i])
 587                         continue;
 588
 589                 vq = rte_malloc(NULL, sizeof(struct vhost_virtqueue), 0);
 590                 if (vq == NULL) {
 591                         VHOST_LOG_CONFIG(ERR,
 592                                 "Failed to allocate memory for vring:%u.\n", i);
 593                         return -1;
 594                 }
 595
 596                 dev->virtqueue[i] = vq;
 597                 init_vring_queue(dev, vring_idx);
 598                 rte_spinlock_init(&vq->access_lock);
 599                 vq->avail_wrap_counter = 1;
 600                 vq->used_wrap_counter = 1;
 601                 vq->signalled_used_valid = false;
 602         }
 603
 604         dev->nr_vring = RTE_MAX(dev->nr_vring, vring_idx + 1);
 605
 606         return 0;
 607 }
 608
 609 /*
 610  * Reset some variables in device structure, while keeping few
 611  * others untouched, such as vid, ifname, nr_vring: they
 612  * should be same unless the device is removed.
 613  */
 614 void
 615 reset_device(struct virtio_net *dev)
 616 {
 617         uint32_t i;
 618
 619         dev->features = 0;
 620         dev->protocol_features = 0;
 621         dev->flags &= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 622
 623         for (i = 0; i < dev->nr_vring; i++)
 624                 reset_vring_queue(dev, i);
 625 }
 626
 627 /*
 628  * Invoked when there is a new vhost-user connection established (when
 629  * there is a new virtio device being attached).
 630  */
 631 int
 632 vhost_new_device(void)
 633 {
 634         struct virtio_net *dev;
 635         int i;
 636
 637         for (i = 0; i < MAX_VHOST_DEVICE; i++) {
 638                 if (vhost_devices[i] == NULL)
 639                         break;
 640         }
 641
 642         if (i == MAX_VHOST_DEVICE) {
 643                 VHOST_LOG_CONFIG(ERR,
 644                         "Failed to find a free slot for new device.\n");
 645                 return -1;
 646         }
 647
 648         dev = rte_zmalloc(NULL, sizeof(struct virtio_net), 0);
 649         if (dev == NULL) {
 650                 VHOST_LOG_CONFIG(ERR,
 651                         "Failed to allocate memory for new dev.\n");
 652                 return -1;
 653         }
 654
 655         vhost_devices[i] = dev;
 656         dev->vid = i;
 657         dev->flags = VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 658         dev->slave_req_fd = -1;
 659         dev->postcopy_ufd = -1;
 660         rte_spinlock_init(&dev->slave_req_lock);
 661
 662         return i;
 663 }
 664
 665 void
 666 vhost_destroy_device_notify(struct virtio_net *dev)
 667 {
 668         struct rte_vdpa_device *vdpa_dev;
 669
 670         if (dev->flags & VIRTIO_DEV_RUNNING) {
 671                 vdpa_dev = dev->vdpa_dev;
 672                 if (vdpa_dev)
 673                         vdpa_dev->ops->dev_close(dev->vid);
 674                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 675                 dev->notify_ops->destroy_device(dev->vid);
 676         }
 677 }
 678
 679 /*
 680  * Invoked when there is the vhost-user connection is broken (when
 681  * the virtio device is being detached).
 682  */
 683 void
 684 vhost_destroy_device(int vid)
 685 {
 686         struct virtio_net *dev = get_device(vid);
 687
 688         if (dev == NULL)
 689                 return;
 690
 691         vhost_destroy_device_notify(dev);
 692
 693         cleanup_device(dev, 1);
 694         free_device(dev);
 695
 696         vhost_devices[vid] = NULL;
 697 }
 698
 699 void
 700 vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *vdpa_dev)
 701 {
 702         struct virtio_net *dev = get_device(vid);
 703
 704         if (dev == NULL)
 705                 return;
 706
 707         dev->vdpa_dev = vdpa_dev;
 708 }
 709
 710 void
 711 vhost_set_ifname(int vid, const char *if_name, unsigned int if_len)
 712 {
 713         struct virtio_net *dev;
 714         unsigned int len;
 715
 716         dev = get_device(vid);
 717         if (dev == NULL)
 718                 return;
 719
 720         len = if_len > sizeof(dev->ifname) ?
 721                 sizeof(dev->ifname) : if_len;
 722
 723         strncpy(dev->ifname, if_name, len);
 724         dev->ifname[sizeof(dev->ifname) - 1] = '\0';
 725 }
 726
 727 void
 728 vhost_set_builtin_virtio_net(int vid, bool enable)
 729 {
 730         struct virtio_net *dev = get_device(vid);
 731
 732         if (dev == NULL)
 733                 return;
 734
 735         if (enable)
 736                 dev->flags |= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 737         else
 738                 dev->flags &= ~VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 739 }
 740
 741 void
 742 vhost_enable_extbuf(int vid)
 743 {
 744         struct virtio_net *dev = get_device(vid);
 745
 746         if (dev == NULL)
 747                 return;
 748
 749         dev->extbuf = 1;
 750 }
 751
 752 void
 753 vhost_enable_linearbuf(int vid)
 754 {
 755         struct virtio_net *dev = get_device(vid);
 756
 757         if (dev == NULL)
 758                 return;
 759
 760         dev->linearbuf = 1;
 761 }
 762
 763 int
 764 rte_vhost_get_mtu(int vid, uint16_t *mtu)
 765 {
 766         struct virtio_net *dev = get_device(vid);
 767
 768         if (dev == NULL || mtu == NULL)
 769                 return -ENODEV;
 770
 771         if (!(dev->flags & VIRTIO_DEV_READY))
 772                 return -EAGAIN;
 773
 774         if (!(dev->features & (1ULL << VIRTIO_NET_F_MTU)))
 775                 return -ENOTSUP;
 776
 777         *mtu = dev->mtu;
 778
 779         return 0;
 780 }
 781
 782 int
 783 rte_vhost_get_numa_node(int vid)
 784 {
 785 #ifdef RTE_LIBRTE_VHOST_NUMA
 786         struct virtio_net *dev = get_device(vid);
 787         int numa_node;
 788         int ret;
 789
 790         if (dev == NULL || numa_available() != 0)
 791                 return -1;
 792
 793         ret = get_mempolicy(&numa_node, NULL, 0, dev,
 794                             MPOL_F_NODE | MPOL_F_ADDR);
 795         if (ret < 0) {
 796                 VHOST_LOG_CONFIG(ERR,
 797                         "(%d) failed to query numa node: %s\n",
 798                         vid, rte_strerror(errno));
 799                 return -1;
 800         }
 801
 802         return numa_node;
 803 #else
 804         RTE_SET_USED(vid);
 805         return -1;
 806 #endif
 807 }
 808
 809 uint32_t
 810 rte_vhost_get_queue_num(int vid)
 811 {
 812         struct virtio_net *dev = get_device(vid);
 813
 814         if (dev == NULL)
 815                 return 0;
 816
 817         return dev->nr_vring / 2;
 818 }
 819
 820 uint16_t
 821 rte_vhost_get_vring_num(int vid)
 822 {
 823         struct virtio_net *dev = get_device(vid);
 824
 825         if (dev == NULL)
 826                 return 0;
 827
 828         return dev->nr_vring;
 829 }
 830
 831 int
 832 rte_vhost_get_ifname(int vid, char *buf, size_t len)
 833 {
 834         struct virtio_net *dev = get_device(vid);
 835
 836         if (dev == NULL || buf == NULL)
 837                 return -1;
 838
 839         len = RTE_MIN(len, sizeof(dev->ifname));
 840
 841         strncpy(buf, dev->ifname, len);
 842         buf[len - 1] = '\0';
 843
 844         return 0;
 845 }
 846
 847 int
 848 rte_vhost_get_negotiated_features(int vid, uint64_t *features)
 849 {
 850         struct virtio_net *dev;
 851
 852         dev = get_device(vid);
 853         if (dev == NULL || features == NULL)
 854                 return -1;
 855
 856         *features = dev->features;
 857         return 0;
 858 }
 859
 860 int
 861 rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem)
 862 {
 863         struct virtio_net *dev;
 864         struct rte_vhost_memory *m;
 865         size_t size;
 866
 867         dev = get_device(vid);
 868         if (dev == NULL || mem == NULL)
 869                 return -1;
 870
 871         size = dev->mem->nregions * sizeof(struct rte_vhost_mem_region);
 872         m = malloc(sizeof(struct rte_vhost_memory) + size);
 873         if (!m)
 874                 return -1;
 875
 876         m->nregions = dev->mem->nregions;
 877         memcpy(m->regions, dev->mem->regions, size);
 878         *mem = m;
 879
 880         return 0;
 881 }
 882
 883 int
 884 rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
 885                           struct rte_vhost_vring *vring)
 886 {
 887         struct virtio_net *dev;
 888         struct vhost_virtqueue *vq;
 889
 890         dev = get_device(vid);
 891         if (dev == NULL || vring == NULL)
 892                 return -1;
 893
 894         if (vring_idx >= VHOST_MAX_VRING)
 895                 return -1;
 896
 897         vq = dev->virtqueue[vring_idx];
 898         if (!vq)
 899                 return -1;
 900
 901         if (vq_is_packed(dev)) {
 902                 vring->desc_packed = vq->desc_packed;
 903                 vring->driver_event = vq->driver_event;
 904                 vring->device_event = vq->device_event;
 905         } else {
 906                 vring->desc = vq->desc;
 907                 vring->avail = vq->avail;
 908                 vring->used = vq->used;
 909         }
 910         vring->log_guest_addr  = vq->log_guest_addr;
 911
 912         vring->callfd  = vq->callfd;
 913         vring->kickfd  = vq->kickfd;
 914         vring->size    = vq->size;
 915
 916         return 0;
 917 }
 918
 919 int
 920 rte_vhost_get_vhost_ring_inflight(int vid, uint16_t vring_idx,
 921                                   struct rte_vhost_ring_inflight *vring)
 922 {
 923         struct virtio_net *dev;
 924         struct vhost_virtqueue *vq;
 925
 926         dev = get_device(vid);
 927         if (unlikely(!dev))
 928                 return -1;
 929
 930         if (vring_idx >= VHOST_MAX_VRING)
 931                 return -1;
 932
 933         vq = dev->virtqueue[vring_idx];
 934         if (unlikely(!vq))
 935                 return -1;
 936
 937         if (vq_is_packed(dev)) {
 938                 if (unlikely(!vq->inflight_packed))
 939                         return -1;
 940
 941                 vring->inflight_packed = vq->inflight_packed;
 942         } else {
 943                 if (unlikely(!vq->inflight_split))
 944                         return -1;
 945
 946                 vring->inflight_split = vq->inflight_split;
 947         }
 948
 949         vring->resubmit_inflight = vq->resubmit_inflight;
 950
 951         return 0;
 952 }
 953
 954 int
 955 rte_vhost_set_inflight_desc_split(int vid, uint16_t vring_idx,
 956                                   uint16_t idx)
 957 {
 958         struct vhost_virtqueue *vq;
 959         struct virtio_net *dev;
 960
 961         dev = get_device(vid);
 962         if (unlikely(!dev))
 963                 return -1;
 964
 965         if (unlikely(!(dev->protocol_features &
 966             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
 967                 return 0;
 968
 969         if (unlikely(vq_is_packed(dev)))
 970                 return -1;
 971
 972         if (unlikely(vring_idx >= VHOST_MAX_VRING))
 973                 return -1;
 974
 975         vq = dev->virtqueue[vring_idx];
 976         if (unlikely(!vq))
 977                 return -1;
 978
 979         if (unlikely(!vq->inflight_split))
 980                 return -1;
 981
 982         if (unlikely(idx >= vq->size))
 983                 return -1;
 984
 985         vq->inflight_split->desc[idx].counter = vq->global_counter++;
 986         vq->inflight_split->desc[idx].inflight = 1;
 987         return 0;
 988 }
 989
 990 int
 991 rte_vhost_set_inflight_desc_packed(int vid, uint16_t vring_idx,
 992                                    uint16_t head, uint16_t last,
 993                                    uint16_t *inflight_entry)
 994 {
 995         struct rte_vhost_inflight_info_packed *inflight_info;
 996         struct virtio_net *dev;
 997         struct vhost_virtqueue *vq;
 998         struct vring_packed_desc *desc;
 999         uint16_t old_free_head, free_head;
1000
1001         dev = get_device(vid);
1002         if (unlikely(!dev))
1003                 return -1;
1004
1005         if (unlikely(!(dev->protocol_features &
1006             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1007                 return 0;
1008
1009         if (unlikely(!vq_is_packed(dev)))
1010                 return -1;
1011
1012         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1013                 return -1;
1014
1015         vq = dev->virtqueue[vring_idx];
1016         if (unlikely(!vq))
1017                 return -1;
1018
1019         inflight_info = vq->inflight_packed;
1020         if (unlikely(!inflight_info))
1021                 return -1;
1022
1023         if (unlikely(head >= vq->size))
1024                 return -1;
1025
1026         desc = vq->desc_packed;
1027         old_free_head = inflight_info->old_free_head;
1028         if (unlikely(old_free_head >= vq->size))
1029                 return -1;
1030
1031         free_head = old_free_head;
1032
1033         /* init header descriptor */
1034         inflight_info->desc[old_free_head].num = 0;
1035         inflight_info->desc[old_free_head].counter = vq->global_counter++;
1036         inflight_info->desc[old_free_head].inflight = 1;
1037
1038         /* save desc entry in flight entry */
1039         while (head != ((last + 1) % vq->size)) {
1040                 inflight_info->desc[old_free_head].num++;
1041                 inflight_info->desc[free_head].addr = desc[head].addr;
1042                 inflight_info->desc[free_head].len = desc[head].len;
1043                 inflight_info->desc[free_head].flags = desc[head].flags;
1044                 inflight_info->desc[free_head].id = desc[head].id;
1045
1046                 inflight_info->desc[old_free_head].last = free_head;
1047                 free_head = inflight_info->desc[free_head].next;
1048                 inflight_info->free_head = free_head;
1049                 head = (head + 1) % vq->size;
1050         }
1051
1052         inflight_info->old_free_head = free_head;
1053         *inflight_entry = old_free_head;
1054
1055         return 0;
1056 }
1057
1058 int
1059 rte_vhost_clr_inflight_desc_split(int vid, uint16_t vring_idx,
1060                                   uint16_t last_used_idx, uint16_t idx)
1061 {
1062         struct virtio_net *dev;
1063         struct vhost_virtqueue *vq;
1064
1065         dev = get_device(vid);
1066         if (unlikely(!dev))
1067                 return -1;
1068
1069         if (unlikely(!(dev->protocol_features &
1070             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1071                 return 0;
1072
1073         if (unlikely(vq_is_packed(dev)))
1074                 return -1;
1075
1076         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1077                 return -1;
1078
1079         vq = dev->virtqueue[vring_idx];
1080         if (unlikely(!vq))
1081                 return -1;
1082
1083         if (unlikely(!vq->inflight_split))
1084                 return -1;
1085
1086         if (unlikely(idx >= vq->size))
1087                 return -1;
1088
1089         rte_smp_mb();
1090
1091         vq->inflight_split->desc[idx].inflight = 0;
1092
1093         rte_smp_mb();
1094
1095         vq->inflight_split->used_idx = last_used_idx;
1096         return 0;
1097 }
1098
1099 int
1100 rte_vhost_clr_inflight_desc_packed(int vid, uint16_t vring_idx,
1101                                    uint16_t head)
1102 {
1103         struct rte_vhost_inflight_info_packed *inflight_info;
1104         struct virtio_net *dev;
1105         struct vhost_virtqueue *vq;
1106
1107         dev = get_device(vid);
1108         if (unlikely(!dev))
1109                 return -1;
1110
1111         if (unlikely(!(dev->protocol_features &
1112             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1113                 return 0;
1114
1115         if (unlikely(!vq_is_packed(dev)))
1116                 return -1;
1117
1118         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1119                 return -1;
1120
1121         vq = dev->virtqueue[vring_idx];
1122         if (unlikely(!vq))
1123                 return -1;
1124
1125         inflight_info = vq->inflight_packed;
1126         if (unlikely(!inflight_info))
1127                 return -1;
1128
1129         if (unlikely(head >= vq->size))
1130                 return -1;
1131
1132         rte_smp_mb();
1133
1134         inflight_info->desc[head].inflight = 0;
1135
1136         rte_smp_mb();
1137
1138         inflight_info->old_free_head = inflight_info->free_head;
1139         inflight_info->old_used_idx = inflight_info->used_idx;
1140         inflight_info->old_used_wrap_counter = inflight_info->used_wrap_counter;
1141
1142         return 0;
1143 }
1144
1145 int
1146 rte_vhost_set_last_inflight_io_split(int vid, uint16_t vring_idx,
1147                                      uint16_t idx)
1148 {
1149         struct virtio_net *dev;
1150         struct vhost_virtqueue *vq;
1151
1152         dev = get_device(vid);
1153         if (unlikely(!dev))
1154                 return -1;
1155
1156         if (unlikely(!(dev->protocol_features &
1157             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1158                 return 0;
1159
1160         if (unlikely(vq_is_packed(dev)))
1161                 return -1;
1162
1163         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1164                 return -1;
1165
1166         vq = dev->virtqueue[vring_idx];
1167         if (unlikely(!vq))
1168                 return -1;
1169
1170         if (unlikely(!vq->inflight_split))
1171                 return -1;
1172
1173         vq->inflight_split->last_inflight_io = idx;
1174         return 0;
1175 }
1176
1177 int
1178 rte_vhost_set_last_inflight_io_packed(int vid, uint16_t vring_idx,
1179                                       uint16_t head)
1180 {
1181         struct rte_vhost_inflight_info_packed *inflight_info;
1182         struct virtio_net *dev;
1183         struct vhost_virtqueue *vq;
1184         uint16_t last;
1185
1186         dev = get_device(vid);
1187         if (unlikely(!dev))
1188                 return -1;
1189
1190         if (unlikely(!(dev->protocol_features &
1191             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1192                 return 0;
1193
1194         if (unlikely(!vq_is_packed(dev)))
1195                 return -1;
1196
1197         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1198                 return -1;
1199
1200         vq = dev->virtqueue[vring_idx];
1201         if (unlikely(!vq))
1202                 return -1;
1203
1204         inflight_info = vq->inflight_packed;
1205         if (unlikely(!inflight_info))
1206                 return -1;
1207
1208         if (unlikely(head >= vq->size))
1209                 return -1;
1210
1211         last = inflight_info->desc[head].last;
1212         if (unlikely(last >= vq->size))
1213                 return -1;
1214
1215         inflight_info->desc[last].next = inflight_info->free_head;
1216         inflight_info->free_head = head;
1217         inflight_info->used_idx += inflight_info->desc[head].num;
1218         if (inflight_info->used_idx >= inflight_info->desc_num) {
1219                 inflight_info->used_idx -= inflight_info->desc_num;
1220                 inflight_info->used_wrap_counter =
1221                         !inflight_info->used_wrap_counter;
1222         }
1223
1224         return 0;
1225 }
1226
1227 int
1228 rte_vhost_vring_call(int vid, uint16_t vring_idx)
1229 {
1230         struct virtio_net *dev;
1231         struct vhost_virtqueue *vq;
1232
1233         dev = get_device(vid);
1234         if (!dev)
1235                 return -1;
1236
1237         if (vring_idx >= VHOST_MAX_VRING)
1238                 return -1;
1239
1240         vq = dev->virtqueue[vring_idx];
1241         if (!vq)
1242                 return -1;
1243
1244         if (vq_is_packed(dev))
1245                 vhost_vring_call_packed(dev, vq);
1246         else
1247                 vhost_vring_call_split(dev, vq);
1248
1249         return 0;
1250 }
1251
1252 uint16_t
1253 rte_vhost_avail_entries(int vid, uint16_t queue_id)
1254 {
1255         struct virtio_net *dev;
1256         struct vhost_virtqueue *vq;
1257         uint16_t ret = 0;
1258
1259         dev = get_device(vid);
1260         if (!dev)
1261                 return 0;
1262
1263         vq = dev->virtqueue[queue_id];
1264
1265         rte_spinlock_lock(&vq->access_lock);
1266
1267         if (unlikely(!vq->enabled || vq->avail == NULL))
1268                 goto out;
1269
1270         ret = *(volatile uint16_t *)&vq->avail->idx - vq->last_used_idx;
1271
1272 out:
1273         rte_spinlock_unlock(&vq->access_lock);
1274         return ret;
1275 }
1276
1277 static inline int
1278 vhost_enable_notify_split(struct virtio_net *dev,
1279                 struct vhost_virtqueue *vq, int enable)
1280 {
1281         if (vq->used == NULL)
1282                 return -1;
1283
1284         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
1285                 if (enable)
1286                         vq->used->flags &= ~VRING_USED_F_NO_NOTIFY;
1287                 else
1288                         vq->used->flags |= VRING_USED_F_NO_NOTIFY;
1289         } else {
1290                 if (enable)
1291                         vhost_avail_event(vq) = vq->last_avail_idx;
1292         }
1293         return 0;
1294 }
1295
1296 static inline int
1297 vhost_enable_notify_packed(struct virtio_net *dev,
1298                 struct vhost_virtqueue *vq, int enable)
1299 {
1300         uint16_t flags;
1301
1302         if (vq->device_event == NULL)
1303                 return -1;
1304
1305         if (!enable) {
1306                 vq->device_event->flags = VRING_EVENT_F_DISABLE;
1307                 return 0;
1308         }
1309
1310         flags = VRING_EVENT_F_ENABLE;
1311         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
1312                 flags = VRING_EVENT_F_DESC;
1313                 vq->device_event->off_wrap = vq->last_avail_idx |
1314                         vq->avail_wrap_counter << 15;
1315         }
1316
1317         rte_smp_wmb();
1318
1319         vq->device_event->flags = flags;
1320         return 0;
1321 }
1322
1323 int
1324 vhost_enable_guest_notification(struct virtio_net *dev,
1325                 struct vhost_virtqueue *vq, int enable)
1326 {
1327         /*
1328          * If the virtqueue is not ready yet, it will be applied
1329          * when it will become ready.
1330          */
1331         if (!vq->ready)
1332                 return 0;
1333
1334         if (vq_is_packed(dev))
1335                 return vhost_enable_notify_packed(dev, vq, enable);
1336         else
1337                 return vhost_enable_notify_split(dev, vq, enable);
1338 }
1339
1340 int
1341 rte_vhost_enable_guest_notification(int vid, uint16_t queue_id, int enable)
1342 {
1343         struct virtio_net *dev = get_device(vid);
1344         struct vhost_virtqueue *vq;
1345         int ret;
1346
1347         if (!dev)
1348                 return -1;
1349
1350         vq = dev->virtqueue[queue_id];
1351
1352         rte_spinlock_lock(&vq->access_lock);
1353
1354         vq->notif_enable = enable;
1355         ret = vhost_enable_guest_notification(dev, vq, enable);
1356
1357         rte_spinlock_unlock(&vq->access_lock);
1358
1359         return ret;
1360 }
1361
1362 void
1363 rte_vhost_log_write(int vid, uint64_t addr, uint64_t len)
1364 {
1365         struct virtio_net *dev = get_device(vid);
1366
1367         if (dev == NULL)
1368                 return;
1369
1370         vhost_log_write(dev, addr, len);
1371 }
1372
1373 void
1374 rte_vhost_log_used_vring(int vid, uint16_t vring_idx,
1375                          uint64_t offset, uint64_t len)
1376 {
1377         struct virtio_net *dev;
1378         struct vhost_virtqueue *vq;
1379
1380         dev = get_device(vid);
1381         if (dev == NULL)
1382                 return;
1383
1384         if (vring_idx >= VHOST_MAX_VRING)
1385                 return;
1386         vq = dev->virtqueue[vring_idx];
1387         if (!vq)
1388                 return;
1389
1390         vhost_log_used_vring(dev, vq, offset, len);
1391 }
1392
1393 uint32_t
1394 rte_vhost_rx_queue_count(int vid, uint16_t qid)
1395 {
1396         struct virtio_net *dev;
1397         struct vhost_virtqueue *vq;
1398         uint32_t ret = 0;
1399
1400         dev = get_device(vid);
1401         if (dev == NULL)
1402                 return 0;
1403
1404         if (unlikely(qid >= dev->nr_vring || (qid & 1) == 0)) {
1405                 VHOST_LOG_DATA(ERR, "(%d) %s: invalid virtqueue idx %d.\n",
1406                         dev->vid, __func__, qid);
1407                 return 0;
1408         }
1409
1410         vq = dev->virtqueue[qid];
1411         if (vq == NULL)
1412                 return 0;
1413
1414         rte_spinlock_lock(&vq->access_lock);
1415
1416         if (unlikely(vq->enabled == 0 || vq->avail == NULL))
1417                 goto out;
1418
1419         ret = *((volatile uint16_t *)&vq->avail->idx) - vq->last_avail_idx;
1420
1421 out:
1422         rte_spinlock_unlock(&vq->access_lock);
1423         return ret;
1424 }
1425
1426 struct rte_vdpa_device *
1427 rte_vhost_get_vdpa_device(int vid)
1428 {
1429         struct virtio_net *dev = get_device(vid);
1430
1431         if (dev == NULL)
1432                 return NULL;
1433
1434         return dev->vdpa_dev;
1435 }
1436
1437 int rte_vhost_get_log_base(int vid, uint64_t *log_base,
1438                 uint64_t *log_size)
1439 {
1440         struct virtio_net *dev = get_device(vid);
1441
1442         if (dev == NULL || log_base == NULL || log_size == NULL)
1443                 return -1;
1444
1445         *log_base = dev->log_base;
1446         *log_size = dev->log_size;
1447
1448         return 0;
1449 }
1450
1451 int rte_vhost_get_vring_base(int vid, uint16_t queue_id,
1452                 uint16_t *last_avail_idx, uint16_t *last_used_idx)
1453 {
1454         struct vhost_virtqueue *vq;
1455         struct virtio_net *dev = get_device(vid);
1456
1457         if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
1458                 return -1;
1459
1460         vq = dev->virtqueue[queue_id];
1461         if (!vq)
1462                 return -1;
1463
1464         if (vq_is_packed(dev)) {
1465                 *last_avail_idx = (vq->avail_wrap_counter << 15) |
1466                                   vq->last_avail_idx;
1467                 *last_used_idx = (vq->used_wrap_counter << 15) |
1468                                  vq->last_used_idx;
1469         } else {
1470                 *last_avail_idx = vq->last_avail_idx;
1471                 *last_used_idx = vq->last_used_idx;
1472         }
1473
1474         return 0;
1475 }
1476
1477 int rte_vhost_set_vring_base(int vid, uint16_t queue_id,
1478                 uint16_t last_avail_idx, uint16_t last_used_idx)
1479 {
1480         struct vhost_virtqueue *vq;
1481         struct virtio_net *dev = get_device(vid);
1482
1483         if (!dev)
1484                 return -1;
1485
1486         vq = dev->virtqueue[queue_id];
1487         if (!vq)
1488                 return -1;
1489
1490         if (vq_is_packed(dev)) {
1491                 vq->last_avail_idx = last_avail_idx & 0x7fff;
1492                 vq->avail_wrap_counter = !!(last_avail_idx & (1 << 15));
1493                 vq->last_used_idx = last_used_idx & 0x7fff;
1494                 vq->used_wrap_counter = !!(last_used_idx & (1 << 15));
1495         } else {
1496                 vq->last_avail_idx = last_avail_idx;
1497                 vq->last_used_idx = last_used_idx;
1498         }
1499
1500         return 0;
1501 }
1502
1503 int
1504 rte_vhost_get_vring_base_from_inflight(int vid,
1505                                        uint16_t queue_id,
1506                                        uint16_t *last_avail_idx,
1507                                        uint16_t *last_used_idx)
1508 {
1509         struct rte_vhost_inflight_info_packed *inflight_info;
1510         struct virtio_net *dev = get_device(vid);
1511
1512         if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
1513                 return -1;
1514
1515         if (!vq_is_packed(dev))
1516                 return -1;
1517
1518         inflight_info = dev->virtqueue[queue_id]->inflight_packed;
1519         if (!inflight_info)
1520                 return -1;
1521
1522         *last_avail_idx = (inflight_info->old_used_wrap_counter << 15) |
1523                           inflight_info->old_used_idx;
1524         *last_used_idx = *last_avail_idx;
1525
1526         return 0;
1527 }
1528
1529 int rte_vhost_extern_callback_register(int vid,
1530                 struct rte_vhost_user_extern_ops const * const ops, void *ctx)
1531 {
1532         struct virtio_net *dev = get_device(vid);
1533
1534         if (dev == NULL || ops == NULL)
1535                 return -1;
1536
1537         dev->extern_ops = *ops;
1538         dev->extern_data = ctx;
1539         return 0;
1540 }
1541
1542 int rte_vhost_async_channel_register(int vid, uint16_t queue_id,
1543                                         uint32_t features,
1544                                         struct rte_vhost_async_channel_ops *ops)
1545 {
1546         struct vhost_virtqueue *vq;
1547         struct virtio_net *dev = get_device(vid);
1548         struct rte_vhost_async_features f;
1549         int node;
1550
1551         if (dev == NULL || ops == NULL)
1552                 return -1;
1553
1554         f.intval = features;
1555
1556         vq = dev->virtqueue[queue_id];
1557
1558         if (unlikely(vq == NULL || !dev->async_copy))
1559                 return -1;
1560
1561         /* packed queue is not supported */
1562         if (unlikely(vq_is_packed(dev) || !f.async_inorder)) {
1563                 VHOST_LOG_CONFIG(ERR,
1564                         "async copy is not supported on packed queue or non-inorder mode "
1565                         "(vid %d, qid: %d)\n", vid, queue_id);
1566                 return -1;
1567         }
1568
1569         if (unlikely(ops->check_completed_copies == NULL ||
1570                 ops->transfer_data == NULL))
1571                 return -1;
1572
1573         rte_spinlock_lock(&vq->access_lock);
1574
1575         if (unlikely(vq->async_registered)) {
1576                 VHOST_LOG_CONFIG(ERR,
1577                         "async register failed: channel already registered "
1578                         "(vid %d, qid: %d)\n", vid, queue_id);
1579                 goto reg_out;
1580         }
1581
1582 #ifdef RTE_LIBRTE_VHOST_NUMA
1583         if (get_mempolicy(&node, NULL, 0, vq, MPOL_F_NODE | MPOL_F_ADDR)) {
1584                 VHOST_LOG_CONFIG(ERR,
1585                         "unable to get numa information in async register. "
1586                         "allocating async buffer memory on the caller thread node\n");
1587                 node = SOCKET_ID_ANY;
1588         }
1589 #else
1590         node = SOCKET_ID_ANY;
1591 #endif
1592
1593         vq->async_pkts_pending = rte_malloc_socket(NULL,
1594                         vq->size * sizeof(uintptr_t),
1595                         RTE_CACHE_LINE_SIZE, node);
1596         vq->async_pkts_info = rte_malloc_socket(NULL,
1597                         vq->size * sizeof(struct async_inflight_info),
1598                         RTE_CACHE_LINE_SIZE, node);
1599         vq->it_pool = rte_malloc_socket(NULL,
1600                         VHOST_MAX_ASYNC_IT * sizeof(struct rte_vhost_iov_iter),
1601                         RTE_CACHE_LINE_SIZE, node);
1602         vq->vec_pool = rte_malloc_socket(NULL,
1603                         VHOST_MAX_ASYNC_VEC * sizeof(struct iovec),
1604                         RTE_CACHE_LINE_SIZE, node);
1605         if (!vq->async_pkts_pending || !vq->async_pkts_info ||
1606                 !vq->it_pool || !vq->vec_pool) {
1607                 vhost_free_async_mem(vq);
1608                 VHOST_LOG_CONFIG(ERR,
1609                                 "async register failed: cannot allocate memory for vq data "
1610                                 "(vid %d, qid: %d)\n", vid, queue_id);
1611                 goto reg_out;
1612         }
1613
1614         vq->async_ops.check_completed_copies = ops->check_completed_copies;
1615         vq->async_ops.transfer_data = ops->transfer_data;
1616
1617         vq->async_inorder = f.async_inorder;
1618         vq->async_threshold = f.async_threshold;
1619
1620         vq->async_registered = true;
1621
1622 reg_out:
1623         rte_spinlock_unlock(&vq->access_lock);
1624
1625         return 0;
1626 }
1627
1628 int rte_vhost_async_channel_unregister(int vid, uint16_t queue_id)
1629 {
1630         struct vhost_virtqueue *vq;
1631         struct virtio_net *dev = get_device(vid);
1632         int ret = -1;
1633
1634         if (dev == NULL)
1635                 return ret;
1636
1637         vq = dev->virtqueue[queue_id];
1638
1639         if (vq == NULL)
1640                 return ret;
1641
1642         ret = 0;
1643
1644         if (!vq->async_registered)
1645                 return ret;
1646
1647         if (!rte_spinlock_trylock(&vq->access_lock)) {
1648                 VHOST_LOG_CONFIG(ERR, "Failed to unregister async channel. "
1649                         "virt queue busy.\n");
1650                 return -1;
1651         }
1652
1653         if (vq->async_pkts_inflight_n) {
1654                 VHOST_LOG_CONFIG(ERR, "Failed to unregister async channel. "
1655                         "async inflight packets must be completed before unregistration.\n");
1656                 ret = -1;
1657                 goto out;
1658         }
1659
1660         vhost_free_async_mem(vq);
1661
1662         vq->async_ops.transfer_data = NULL;
1663         vq->async_ops.check_completed_copies = NULL;
1664         vq->async_registered = false;
1665
1666 out:
1667         rte_spinlock_unlock(&vq->access_lock);
1668
1669         return ret;
1670 }
1671
1672 RTE_LOG_REGISTER(vhost_config_log_level, lib.vhost.config, INFO);
1673 RTE_LOG_REGISTER(vhost_data_log_level, lib.vhost.data, WARNING);