lib/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 pthread_mutex_t vhost_dev_lock = PTHREAD_MUTEX_INITIALIZER;
  30
  31 /* Called with iotlb_lock read-locked */
  32 uint64_t
  33 __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
  34                     uint64_t iova, uint64_t *size, uint8_t perm)
  35 {
  36         uint64_t vva, tmp_size;
  37
  38         if (unlikely(!*size))
  39                 return 0;
  40
  41         tmp_size = *size;
  42
  43         vva = vhost_user_iotlb_cache_find(vq, iova, &tmp_size, perm);
  44         if (tmp_size == *size)
  45                 return vva;
  46
  47         iova += tmp_size;
  48
  49         if (!vhost_user_iotlb_pending_miss(vq, iova, perm)) {
  50                 /*
  51                  * iotlb_lock is read-locked for a full burst,
  52                  * but it only protects the iotlb cache.
  53                  * In case of IOTLB miss, we might block on the socket,
  54                  * which could cause a deadlock with QEMU if an IOTLB update
  55                  * is being handled. We can safely unlock here to avoid it.
  56                  */
  57                 vhost_user_iotlb_rd_unlock(vq);
  58
  59                 vhost_user_iotlb_pending_insert(vq, iova, perm);
  60                 if (vhost_user_iotlb_miss(dev, iova, perm)) {
  61                         VHOST_LOG_CONFIG(ERR,
  62                                 "IOTLB miss req failed for IOVA 0x%" PRIx64 "\n",
  63                                 iova);
  64                         vhost_user_iotlb_pending_remove(vq, iova, 1, perm);
  65                 }
  66
  67                 vhost_user_iotlb_rd_lock(vq);
  68         }
  69
  70         return 0;
  71 }
  72
  73 #define VHOST_LOG_PAGE  4096
  74
  75 /*
  76  * Atomically set a bit in memory.
  77  */
  78 static __rte_always_inline void
  79 vhost_set_bit(unsigned int nr, volatile uint8_t *addr)
  80 {
  81 #if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION < 70100)
  82         /*
  83          * __sync_ built-ins are deprecated, but __atomic_ ones
  84          * are sub-optimized in older GCC versions.
  85          */
  86         __sync_fetch_and_or_1(addr, (1U << nr));
  87 #else
  88         __atomic_fetch_or(addr, (1U << nr), __ATOMIC_RELAXED);
  89 #endif
  90 }
  91
  92 static __rte_always_inline void
  93 vhost_log_page(uint8_t *log_base, uint64_t page)
  94 {
  95         vhost_set_bit(page % 8, &log_base[page / 8]);
  96 }
  97
  98 void
  99 __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
 100 {
 101         uint64_t page;
 102
 103         if (unlikely(!dev->log_base || !len))
 104                 return;
 105
 106         if (unlikely(dev->log_size <= ((addr + len - 1) / VHOST_LOG_PAGE / 8)))
 107                 return;
 108
 109         /* To make sure guest memory updates are committed before logging */
 110         rte_atomic_thread_fence(__ATOMIC_RELEASE);
 111
 112         page = addr / VHOST_LOG_PAGE;
 113         while (page * VHOST_LOG_PAGE < addr + len) {
 114                 vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
 115                 page += 1;
 116         }
 117 }
 118
 119 void
 120 __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 121                              uint64_t iova, uint64_t len)
 122 {
 123         uint64_t hva, gpa, map_len;
 124         map_len = len;
 125
 126         hva = __vhost_iova_to_vva(dev, vq, iova, &map_len, VHOST_ACCESS_RW);
 127         if (map_len != len) {
 128                 VHOST_LOG_DATA(ERR,
 129                         "Failed to write log for IOVA 0x%" PRIx64 ". No IOTLB entry found\n",
 130                         iova);
 131                 return;
 132         }
 133
 134         gpa = hva_to_gpa(dev, hva, len);
 135         if (gpa)
 136                 __vhost_log_write(dev, gpa, len);
 137 }
 138
 139 void
 140 __vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
 141 {
 142         unsigned long *log_base;
 143         int i;
 144
 145         if (unlikely(!dev->log_base))
 146                 return;
 147
 148         /* No cache, nothing to sync */
 149         if (unlikely(!vq->log_cache))
 150                 return;
 151
 152         rte_atomic_thread_fence(__ATOMIC_RELEASE);
 153
 154         log_base = (unsigned long *)(uintptr_t)dev->log_base;
 155
 156         for (i = 0; i < vq->log_cache_nb_elem; i++) {
 157                 struct log_cache_entry *elem = vq->log_cache + i;
 158
 159 #if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION < 70100)
 160                 /*
 161                  * '__sync' builtins are deprecated, but '__atomic' ones
 162                  * are sub-optimized in older GCC versions.
 163                  */
 164                 __sync_fetch_and_or(log_base + elem->offset, elem->val);
 165 #else
 166                 __atomic_fetch_or(log_base + elem->offset, elem->val,
 167                                 __ATOMIC_RELAXED);
 168 #endif
 169         }
 170
 171         rte_atomic_thread_fence(__ATOMIC_RELEASE);
 172
 173         vq->log_cache_nb_elem = 0;
 174 }
 175
 176 static __rte_always_inline void
 177 vhost_log_cache_page(struct virtio_net *dev, struct vhost_virtqueue *vq,
 178                         uint64_t page)
 179 {
 180         uint32_t bit_nr = page % (sizeof(unsigned long) << 3);
 181         uint32_t offset = page / (sizeof(unsigned long) << 3);
 182         int i;
 183
 184         if (unlikely(!vq->log_cache)) {
 185                 /* No logging cache allocated, write dirty log map directly */
 186                 rte_atomic_thread_fence(__ATOMIC_RELEASE);
 187                 vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
 188
 189                 return;
 190         }
 191
 192         for (i = 0; i < vq->log_cache_nb_elem; i++) {
 193                 struct log_cache_entry *elem = vq->log_cache + i;
 194
 195                 if (elem->offset == offset) {
 196                         elem->val |= (1UL << bit_nr);
 197                         return;
 198                 }
 199         }
 200
 201         if (unlikely(i >= VHOST_LOG_CACHE_NR)) {
 202                 /*
 203                  * No more room for a new log cache entry,
 204                  * so write the dirty log map directly.
 205                  */
 206                 rte_atomic_thread_fence(__ATOMIC_RELEASE);
 207                 vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
 208
 209                 return;
 210         }
 211
 212         vq->log_cache[i].offset = offset;
 213         vq->log_cache[i].val = (1UL << bit_nr);
 214         vq->log_cache_nb_elem++;
 215 }
 216
 217 void
 218 __vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
 219                         uint64_t addr, uint64_t len)
 220 {
 221         uint64_t page;
 222
 223         if (unlikely(!dev->log_base || !len))
 224                 return;
 225
 226         if (unlikely(dev->log_size <= ((addr + len - 1) / VHOST_LOG_PAGE / 8)))
 227                 return;
 228
 229         page = addr / VHOST_LOG_PAGE;
 230         while (page * VHOST_LOG_PAGE < addr + len) {
 231                 vhost_log_cache_page(dev, vq, page);
 232                 page += 1;
 233         }
 234 }
 235
 236 void
 237 __vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 238                              uint64_t iova, uint64_t len)
 239 {
 240         uint64_t hva, gpa, map_len;
 241         map_len = len;
 242
 243         hva = __vhost_iova_to_vva(dev, vq, iova, &map_len, VHOST_ACCESS_RW);
 244         if (map_len != len) {
 245                 VHOST_LOG_DATA(ERR,
 246                         "Failed to write log for IOVA 0x%" PRIx64 ". No IOTLB entry found\n",
 247                         iova);
 248                 return;
 249         }
 250
 251         gpa = hva_to_gpa(dev, hva, len);
 252         if (gpa)
 253                 __vhost_log_cache_write(dev, vq, gpa, len);
 254 }
 255
 256 void *
 257 vhost_alloc_copy_ind_table(struct virtio_net *dev, struct vhost_virtqueue *vq,
 258                 uint64_t desc_addr, uint64_t desc_len)
 259 {
 260         void *idesc;
 261         uint64_t src, dst;
 262         uint64_t len, remain = desc_len;
 263
 264         idesc = rte_malloc(__func__, desc_len, 0);
 265         if (unlikely(!idesc))
 266                 return NULL;
 267
 268         dst = (uint64_t)(uintptr_t)idesc;
 269
 270         while (remain) {
 271                 len = remain;
 272                 src = vhost_iova_to_vva(dev, vq, desc_addr, &len,
 273                                 VHOST_ACCESS_RO);
 274                 if (unlikely(!src || !len)) {
 275                         rte_free(idesc);
 276                         return NULL;
 277                 }
 278
 279                 rte_memcpy((void *)(uintptr_t)dst, (void *)(uintptr_t)src, len);
 280
 281                 remain -= len;
 282                 dst += len;
 283                 desc_addr += len;
 284         }
 285
 286         return idesc;
 287 }
 288
 289 void
 290 cleanup_vq(struct vhost_virtqueue *vq, int destroy)
 291 {
 292         if ((vq->callfd >= 0) && (destroy != 0))
 293                 close(vq->callfd);
 294         if (vq->kickfd >= 0)
 295                 close(vq->kickfd);
 296 }
 297
 298 void
 299 cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq)
 300 {
 301         if (!(dev->protocol_features &
 302             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
 303                 return;
 304
 305         if (vq_is_packed(dev)) {
 306                 if (vq->inflight_packed)
 307                         vq->inflight_packed = NULL;
 308         } else {
 309                 if (vq->inflight_split)
 310                         vq->inflight_split = NULL;
 311         }
 312
 313         if (vq->resubmit_inflight) {
 314                 if (vq->resubmit_inflight->resubmit_list) {
 315                         free(vq->resubmit_inflight->resubmit_list);
 316                         vq->resubmit_inflight->resubmit_list = NULL;
 317                 }
 318                 free(vq->resubmit_inflight);
 319                 vq->resubmit_inflight = NULL;
 320         }
 321 }
 322
 323 /*
 324  * Unmap any memory, close any file descriptors and
 325  * free any memory owned by a device.
 326  */
 327 void
 328 cleanup_device(struct virtio_net *dev, int destroy)
 329 {
 330         uint32_t i;
 331
 332         vhost_backend_cleanup(dev);
 333
 334         for (i = 0; i < dev->nr_vring; i++) {
 335                 cleanup_vq(dev->virtqueue[i], destroy);
 336                 cleanup_vq_inflight(dev, dev->virtqueue[i]);
 337         }
 338 }
 339
 340 static void
 341 vhost_free_async_mem(struct vhost_virtqueue *vq)
 342 {
 343         rte_free(vq->async_pkts_info);
 344
 345         rte_free(vq->async_buffers_packed);
 346         vq->async_buffers_packed = NULL;
 347         rte_free(vq->async_descs_split);
 348         vq->async_descs_split = NULL;
 349
 350         rte_free(vq->it_pool);
 351         rte_free(vq->vec_pool);
 352
 353         vq->async_pkts_info = NULL;
 354         vq->it_pool = NULL;
 355         vq->vec_pool = NULL;
 356 }
 357
 358 void
 359 free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq)
 360 {
 361         if (vq_is_packed(dev))
 362                 rte_free(vq->shadow_used_packed);
 363         else
 364                 rte_free(vq->shadow_used_split);
 365
 366         vhost_free_async_mem(vq);
 367         rte_free(vq->batch_copy_elems);
 368         rte_mempool_free(vq->iotlb_pool);
 369         rte_free(vq->log_cache);
 370         rte_free(vq);
 371 }
 372
 373 /*
 374  * Release virtqueues and device memory.
 375  */
 376 static void
 377 free_device(struct virtio_net *dev)
 378 {
 379         uint32_t i;
 380
 381         for (i = 0; i < dev->nr_vring; i++)
 382                 free_vq(dev, dev->virtqueue[i]);
 383
 384         rte_free(dev);
 385 }
 386
 387 static __rte_always_inline int
 388 log_translate(struct virtio_net *dev, struct vhost_virtqueue *vq)
 389 {
 390         if (likely(!(vq->ring_addrs.flags & (1 << VHOST_VRING_F_LOG))))
 391                 return 0;
 392
 393         vq->log_guest_addr = translate_log_addr(dev, vq,
 394                                                 vq->ring_addrs.log_guest_addr);
 395         if (vq->log_guest_addr == 0)
 396                 return -1;
 397
 398         return 0;
 399 }
 400
 401 /*
 402  * Converts vring log address to GPA
 403  * If IOMMU is enabled, the log address is IOVA
 404  * If IOMMU not enabled, the log address is already GPA
 405  *
 406  * Caller should have iotlb_lock read-locked
 407  */
 408 uint64_t
 409 translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
 410                 uint64_t log_addr)
 411 {
 412         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 413                 const uint64_t exp_size = sizeof(uint64_t);
 414                 uint64_t hva, gpa;
 415                 uint64_t size = exp_size;
 416
 417                 hva = vhost_iova_to_vva(dev, vq, log_addr,
 418                                         &size, VHOST_ACCESS_RW);
 419
 420                 if (size != exp_size)
 421                         return 0;
 422
 423                 gpa = hva_to_gpa(dev, hva, exp_size);
 424                 if (!gpa) {
 425                         VHOST_LOG_CONFIG(ERR,
 426                                 "VQ: Failed to find GPA for log_addr: 0x%"
 427                                 PRIx64 " hva: 0x%" PRIx64 "\n",
 428                                 log_addr, hva);
 429                         return 0;
 430                 }
 431                 return gpa;
 432
 433         } else
 434                 return log_addr;
 435 }
 436
 437 /* Caller should have iotlb_lock read-locked */
 438 static int
 439 vring_translate_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 440 {
 441         uint64_t req_size, size;
 442
 443         req_size = sizeof(struct vring_desc) * vq->size;
 444         size = req_size;
 445         vq->desc = (struct vring_desc *)(uintptr_t)vhost_iova_to_vva(dev, vq,
 446                                                 vq->ring_addrs.desc_user_addr,
 447                                                 &size, VHOST_ACCESS_RW);
 448         if (!vq->desc || size != req_size)
 449                 return -1;
 450
 451         req_size = sizeof(struct vring_avail);
 452         req_size += sizeof(uint16_t) * vq->size;
 453         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 454                 req_size += sizeof(uint16_t);
 455         size = req_size;
 456         vq->avail = (struct vring_avail *)(uintptr_t)vhost_iova_to_vva(dev, vq,
 457                                                 vq->ring_addrs.avail_user_addr,
 458                                                 &size, VHOST_ACCESS_RW);
 459         if (!vq->avail || size != req_size)
 460                 return -1;
 461
 462         req_size = sizeof(struct vring_used);
 463         req_size += sizeof(struct vring_used_elem) * vq->size;
 464         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 465                 req_size += sizeof(uint16_t);
 466         size = req_size;
 467         vq->used = (struct vring_used *)(uintptr_t)vhost_iova_to_vva(dev, vq,
 468                                                 vq->ring_addrs.used_user_addr,
 469                                                 &size, VHOST_ACCESS_RW);
 470         if (!vq->used || size != req_size)
 471                 return -1;
 472
 473         return 0;
 474 }
 475
 476 /* Caller should have iotlb_lock read-locked */
 477 static int
 478 vring_translate_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
 479 {
 480         uint64_t req_size, size;
 481
 482         req_size = sizeof(struct vring_packed_desc) * vq->size;
 483         size = req_size;
 484         vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
 485                 vhost_iova_to_vva(dev, vq, vq->ring_addrs.desc_user_addr,
 486                                 &size, VHOST_ACCESS_RW);
 487         if (!vq->desc_packed || size != req_size)
 488                 return -1;
 489
 490         req_size = sizeof(struct vring_packed_desc_event);
 491         size = req_size;
 492         vq->driver_event = (struct vring_packed_desc_event *)(uintptr_t)
 493                 vhost_iova_to_vva(dev, vq, vq->ring_addrs.avail_user_addr,
 494                                 &size, VHOST_ACCESS_RW);
 495         if (!vq->driver_event || size != req_size)
 496                 return -1;
 497
 498         req_size = sizeof(struct vring_packed_desc_event);
 499         size = req_size;
 500         vq->device_event = (struct vring_packed_desc_event *)(uintptr_t)
 501                 vhost_iova_to_vva(dev, vq, vq->ring_addrs.used_user_addr,
 502                                 &size, VHOST_ACCESS_RW);
 503         if (!vq->device_event || size != req_size)
 504                 return -1;
 505
 506         return 0;
 507 }
 508
 509 int
 510 vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq)
 511 {
 512
 513         if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 514                 return -1;
 515
 516         if (vq_is_packed(dev)) {
 517                 if (vring_translate_packed(dev, vq) < 0)
 518                         return -1;
 519         } else {
 520                 if (vring_translate_split(dev, vq) < 0)
 521                         return -1;
 522         }
 523
 524         if (log_translate(dev, vq) < 0)
 525                 return -1;
 526
 527         vq->access_ok = true;
 528
 529         return 0;
 530 }
 531
 532 void
 533 vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq)
 534 {
 535         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 536                 vhost_user_iotlb_wr_lock(vq);
 537
 538         vq->access_ok = false;
 539         vq->desc = NULL;
 540         vq->avail = NULL;
 541         vq->used = NULL;
 542         vq->log_guest_addr = 0;
 543
 544         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 545                 vhost_user_iotlb_wr_unlock(vq);
 546 }
 547
 548 static void
 549 init_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 550 {
 551         struct vhost_virtqueue *vq;
 552
 553         if (vring_idx >= VHOST_MAX_VRING) {
 554                 VHOST_LOG_CONFIG(ERR,
 555                                 "Failed not init vring, out of bound (%d)\n",
 556                                 vring_idx);
 557                 return;
 558         }
 559
 560         vq = dev->virtqueue[vring_idx];
 561         if (!vq) {
 562                 VHOST_LOG_CONFIG(ERR, "Virtqueue not allocated (%d)\n",
 563                                 vring_idx);
 564                 return;
 565         }
 566
 567         memset(vq, 0, sizeof(struct vhost_virtqueue));
 568
 569         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 570         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
 571         vq->notif_enable = VIRTIO_UNINITIALIZED_NOTIF;
 572
 573         vhost_user_iotlb_init(dev, vring_idx);
 574 }
 575
 576 static void
 577 reset_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 578 {
 579         struct vhost_virtqueue *vq;
 580         int callfd;
 581
 582         if (vring_idx >= VHOST_MAX_VRING) {
 583                 VHOST_LOG_CONFIG(ERR,
 584                                 "Failed not init vring, out of bound (%d)\n",
 585                                 vring_idx);
 586                 return;
 587         }
 588
 589         vq = dev->virtqueue[vring_idx];
 590         if (!vq) {
 591                 VHOST_LOG_CONFIG(ERR, "Virtqueue not allocated (%d)\n",
 592                                 vring_idx);
 593                 return;
 594         }
 595
 596         callfd = vq->callfd;
 597         init_vring_queue(dev, vring_idx);
 598         vq->callfd = callfd;
 599 }
 600
 601 int
 602 alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 603 {
 604         struct vhost_virtqueue *vq;
 605         uint32_t i;
 606
 607         /* Also allocate holes, if any, up to requested vring index. */
 608         for (i = 0; i <= vring_idx; i++) {
 609                 if (dev->virtqueue[i])
 610                         continue;
 611
 612                 vq = rte_zmalloc(NULL, sizeof(struct vhost_virtqueue), 0);
 613                 if (vq == NULL) {
 614                         VHOST_LOG_CONFIG(ERR,
 615                                 "Failed to allocate memory for vring:%u.\n", i);
 616                         return -1;
 617                 }
 618
 619                 dev->virtqueue[i] = vq;
 620                 init_vring_queue(dev, i);
 621                 rte_spinlock_init(&vq->access_lock);
 622                 vq->avail_wrap_counter = 1;
 623                 vq->used_wrap_counter = 1;
 624                 vq->signalled_used_valid = false;
 625         }
 626
 627         dev->nr_vring = RTE_MAX(dev->nr_vring, vring_idx + 1);
 628
 629         return 0;
 630 }
 631
 632 /*
 633  * Reset some variables in device structure, while keeping few
 634  * others untouched, such as vid, ifname, nr_vring: they
 635  * should be same unless the device is removed.
 636  */
 637 void
 638 reset_device(struct virtio_net *dev)
 639 {
 640         uint32_t i;
 641
 642         dev->features = 0;
 643         dev->protocol_features = 0;
 644         dev->flags &= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 645
 646         for (i = 0; i < dev->nr_vring; i++)
 647                 reset_vring_queue(dev, i);
 648 }
 649
 650 /*
 651  * Invoked when there is a new vhost-user connection established (when
 652  * there is a new virtio device being attached).
 653  */
 654 int
 655 vhost_new_device(void)
 656 {
 657         struct virtio_net *dev;
 658         int i;
 659
 660         pthread_mutex_lock(&vhost_dev_lock);
 661         for (i = 0; i < MAX_VHOST_DEVICE; i++) {
 662                 if (vhost_devices[i] == NULL)
 663                         break;
 664         }
 665
 666         if (i == MAX_VHOST_DEVICE) {
 667                 VHOST_LOG_CONFIG(ERR,
 668                         "Failed to find a free slot for new device.\n");
 669                 pthread_mutex_unlock(&vhost_dev_lock);
 670                 return -1;
 671         }
 672
 673         dev = rte_zmalloc(NULL, sizeof(struct virtio_net), 0);
 674         if (dev == NULL) {
 675                 VHOST_LOG_CONFIG(ERR,
 676                         "Failed to allocate memory for new dev.\n");
 677                 pthread_mutex_unlock(&vhost_dev_lock);
 678                 return -1;
 679         }
 680
 681         vhost_devices[i] = dev;
 682         pthread_mutex_unlock(&vhost_dev_lock);
 683
 684         dev->vid = i;
 685         dev->flags = VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 686         dev->slave_req_fd = -1;
 687         dev->postcopy_ufd = -1;
 688         rte_spinlock_init(&dev->slave_req_lock);
 689
 690         return i;
 691 }
 692
 693 void
 694 vhost_destroy_device_notify(struct virtio_net *dev)
 695 {
 696         struct rte_vdpa_device *vdpa_dev;
 697
 698         if (dev->flags & VIRTIO_DEV_RUNNING) {
 699                 vdpa_dev = dev->vdpa_dev;
 700                 if (vdpa_dev)
 701                         vdpa_dev->ops->dev_close(dev->vid);
 702                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 703                 dev->notify_ops->destroy_device(dev->vid);
 704         }
 705 }
 706
 707 /*
 708  * Invoked when there is the vhost-user connection is broken (when
 709  * the virtio device is being detached).
 710  */
 711 void
 712 vhost_destroy_device(int vid)
 713 {
 714         struct virtio_net *dev = get_device(vid);
 715
 716         if (dev == NULL)
 717                 return;
 718
 719         vhost_destroy_device_notify(dev);
 720
 721         cleanup_device(dev, 1);
 722         free_device(dev);
 723
 724         vhost_devices[vid] = NULL;
 725 }
 726
 727 void
 728 vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *vdpa_dev)
 729 {
 730         struct virtio_net *dev = get_device(vid);
 731
 732         if (dev == NULL)
 733                 return;
 734
 735         dev->vdpa_dev = vdpa_dev;
 736 }
 737
 738 void
 739 vhost_set_ifname(int vid, const char *if_name, unsigned int if_len)
 740 {
 741         struct virtio_net *dev;
 742         unsigned int len;
 743
 744         dev = get_device(vid);
 745         if (dev == NULL)
 746                 return;
 747
 748         len = if_len > sizeof(dev->ifname) ?
 749                 sizeof(dev->ifname) : if_len;
 750
 751         strncpy(dev->ifname, if_name, len);
 752         dev->ifname[sizeof(dev->ifname) - 1] = '\0';
 753 }
 754
 755 void
 756 vhost_setup_virtio_net(int vid, bool enable, bool compliant_ol_flags)
 757 {
 758         struct virtio_net *dev = get_device(vid);
 759
 760         if (dev == NULL)
 761                 return;
 762
 763         if (enable)
 764                 dev->flags |= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 765         else
 766                 dev->flags &= ~VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 767         if (!compliant_ol_flags)
 768                 dev->flags |= VIRTIO_DEV_LEGACY_OL_FLAGS;
 769         else
 770                 dev->flags &= ~VIRTIO_DEV_LEGACY_OL_FLAGS;
 771 }
 772
 773 void
 774 vhost_enable_extbuf(int vid)
 775 {
 776         struct virtio_net *dev = get_device(vid);
 777
 778         if (dev == NULL)
 779                 return;
 780
 781         dev->extbuf = 1;
 782 }
 783
 784 void
 785 vhost_enable_linearbuf(int vid)
 786 {
 787         struct virtio_net *dev = get_device(vid);
 788
 789         if (dev == NULL)
 790                 return;
 791
 792         dev->linearbuf = 1;
 793 }
 794
 795 int
 796 rte_vhost_get_mtu(int vid, uint16_t *mtu)
 797 {
 798         struct virtio_net *dev = get_device(vid);
 799
 800         if (dev == NULL || mtu == NULL)
 801                 return -ENODEV;
 802
 803         if (!(dev->flags & VIRTIO_DEV_READY))
 804                 return -EAGAIN;
 805
 806         if (!(dev->features & (1ULL << VIRTIO_NET_F_MTU)))
 807                 return -ENOTSUP;
 808
 809         *mtu = dev->mtu;
 810
 811         return 0;
 812 }
 813
 814 int
 815 rte_vhost_get_numa_node(int vid)
 816 {
 817 #ifdef RTE_LIBRTE_VHOST_NUMA
 818         struct virtio_net *dev = get_device(vid);
 819         int numa_node;
 820         int ret;
 821
 822         if (dev == NULL || numa_available() != 0)
 823                 return -1;
 824
 825         ret = get_mempolicy(&numa_node, NULL, 0, dev,
 826                             MPOL_F_NODE | MPOL_F_ADDR);
 827         if (ret < 0) {
 828                 VHOST_LOG_CONFIG(ERR,
 829                         "(%d) failed to query numa node: %s\n",
 830                         vid, rte_strerror(errno));
 831                 return -1;
 832         }
 833
 834         return numa_node;
 835 #else
 836         RTE_SET_USED(vid);
 837         return -1;
 838 #endif
 839 }
 840
 841 uint32_t
 842 rte_vhost_get_queue_num(int vid)
 843 {
 844         struct virtio_net *dev = get_device(vid);
 845
 846         if (dev == NULL)
 847                 return 0;
 848
 849         return dev->nr_vring / 2;
 850 }
 851
 852 uint16_t
 853 rte_vhost_get_vring_num(int vid)
 854 {
 855         struct virtio_net *dev = get_device(vid);
 856
 857         if (dev == NULL)
 858                 return 0;
 859
 860         return dev->nr_vring;
 861 }
 862
 863 int
 864 rte_vhost_get_ifname(int vid, char *buf, size_t len)
 865 {
 866         struct virtio_net *dev = get_device(vid);
 867
 868         if (dev == NULL || buf == NULL)
 869                 return -1;
 870
 871         len = RTE_MIN(len, sizeof(dev->ifname));
 872
 873         strncpy(buf, dev->ifname, len);
 874         buf[len - 1] = '\0';
 875
 876         return 0;
 877 }
 878
 879 int
 880 rte_vhost_get_negotiated_features(int vid, uint64_t *features)
 881 {
 882         struct virtio_net *dev;
 883
 884         dev = get_device(vid);
 885         if (dev == NULL || features == NULL)
 886                 return -1;
 887
 888         *features = dev->features;
 889         return 0;
 890 }
 891
 892 int
 893 rte_vhost_get_negotiated_protocol_features(int vid,
 894                                            uint64_t *protocol_features)
 895 {
 896         struct virtio_net *dev;
 897
 898         dev = get_device(vid);
 899         if (dev == NULL || protocol_features == NULL)
 900                 return -1;
 901
 902         *protocol_features = dev->protocol_features;
 903         return 0;
 904 }
 905
 906 int
 907 rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem)
 908 {
 909         struct virtio_net *dev;
 910         struct rte_vhost_memory *m;
 911         size_t size;
 912
 913         dev = get_device(vid);
 914         if (dev == NULL || mem == NULL)
 915                 return -1;
 916
 917         size = dev->mem->nregions * sizeof(struct rte_vhost_mem_region);
 918         m = malloc(sizeof(struct rte_vhost_memory) + size);
 919         if (!m)
 920                 return -1;
 921
 922         m->nregions = dev->mem->nregions;
 923         memcpy(m->regions, dev->mem->regions, size);
 924         *mem = m;
 925
 926         return 0;
 927 }
 928
 929 int
 930 rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
 931                           struct rte_vhost_vring *vring)
 932 {
 933         struct virtio_net *dev;
 934         struct vhost_virtqueue *vq;
 935
 936         dev = get_device(vid);
 937         if (dev == NULL || vring == NULL)
 938                 return -1;
 939
 940         if (vring_idx >= VHOST_MAX_VRING)
 941                 return -1;
 942
 943         vq = dev->virtqueue[vring_idx];
 944         if (!vq)
 945                 return -1;
 946
 947         if (vq_is_packed(dev)) {
 948                 vring->desc_packed = vq->desc_packed;
 949                 vring->driver_event = vq->driver_event;
 950                 vring->device_event = vq->device_event;
 951         } else {
 952                 vring->desc = vq->desc;
 953                 vring->avail = vq->avail;
 954                 vring->used = vq->used;
 955         }
 956         vring->log_guest_addr  = vq->log_guest_addr;
 957
 958         vring->callfd  = vq->callfd;
 959         vring->kickfd  = vq->kickfd;
 960         vring->size    = vq->size;
 961
 962         return 0;
 963 }
 964
 965 int
 966 rte_vhost_get_vhost_ring_inflight(int vid, uint16_t vring_idx,
 967                                   struct rte_vhost_ring_inflight *vring)
 968 {
 969         struct virtio_net *dev;
 970         struct vhost_virtqueue *vq;
 971
 972         dev = get_device(vid);
 973         if (unlikely(!dev))
 974                 return -1;
 975
 976         if (vring_idx >= VHOST_MAX_VRING)
 977                 return -1;
 978
 979         vq = dev->virtqueue[vring_idx];
 980         if (unlikely(!vq))
 981                 return -1;
 982
 983         if (vq_is_packed(dev)) {
 984                 if (unlikely(!vq->inflight_packed))
 985                         return -1;
 986
 987                 vring->inflight_packed = vq->inflight_packed;
 988         } else {
 989                 if (unlikely(!vq->inflight_split))
 990                         return -1;
 991
 992                 vring->inflight_split = vq->inflight_split;
 993         }
 994
 995         vring->resubmit_inflight = vq->resubmit_inflight;
 996
 997         return 0;
 998 }
 999
1000 int
1001 rte_vhost_set_inflight_desc_split(int vid, uint16_t vring_idx,
1002                                   uint16_t idx)
1003 {
1004         struct vhost_virtqueue *vq;
1005         struct virtio_net *dev;
1006
1007         dev = get_device(vid);
1008         if (unlikely(!dev))
1009                 return -1;
1010
1011         if (unlikely(!(dev->protocol_features &
1012             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1013                 return 0;
1014
1015         if (unlikely(vq_is_packed(dev)))
1016                 return -1;
1017
1018         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1019                 return -1;
1020
1021         vq = dev->virtqueue[vring_idx];
1022         if (unlikely(!vq))
1023                 return -1;
1024
1025         if (unlikely(!vq->inflight_split))
1026                 return -1;
1027
1028         if (unlikely(idx >= vq->size))
1029                 return -1;
1030
1031         vq->inflight_split->desc[idx].counter = vq->global_counter++;
1032         vq->inflight_split->desc[idx].inflight = 1;
1033         return 0;
1034 }
1035
1036 int
1037 rte_vhost_set_inflight_desc_packed(int vid, uint16_t vring_idx,
1038                                    uint16_t head, uint16_t last,
1039                                    uint16_t *inflight_entry)
1040 {
1041         struct rte_vhost_inflight_info_packed *inflight_info;
1042         struct virtio_net *dev;
1043         struct vhost_virtqueue *vq;
1044         struct vring_packed_desc *desc;
1045         uint16_t old_free_head, free_head;
1046
1047         dev = get_device(vid);
1048         if (unlikely(!dev))
1049                 return -1;
1050
1051         if (unlikely(!(dev->protocol_features &
1052             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1053                 return 0;
1054
1055         if (unlikely(!vq_is_packed(dev)))
1056                 return -1;
1057
1058         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1059                 return -1;
1060
1061         vq = dev->virtqueue[vring_idx];
1062         if (unlikely(!vq))
1063                 return -1;
1064
1065         inflight_info = vq->inflight_packed;
1066         if (unlikely(!inflight_info))
1067                 return -1;
1068
1069         if (unlikely(head >= vq->size))
1070                 return -1;
1071
1072         desc = vq->desc_packed;
1073         old_free_head = inflight_info->old_free_head;
1074         if (unlikely(old_free_head >= vq->size))
1075                 return -1;
1076
1077         free_head = old_free_head;
1078
1079         /* init header descriptor */
1080         inflight_info->desc[old_free_head].num = 0;
1081         inflight_info->desc[old_free_head].counter = vq->global_counter++;
1082         inflight_info->desc[old_free_head].inflight = 1;
1083
1084         /* save desc entry in flight entry */
1085         while (head != ((last + 1) % vq->size)) {
1086                 inflight_info->desc[old_free_head].num++;
1087                 inflight_info->desc[free_head].addr = desc[head].addr;
1088                 inflight_info->desc[free_head].len = desc[head].len;
1089                 inflight_info->desc[free_head].flags = desc[head].flags;
1090                 inflight_info->desc[free_head].id = desc[head].id;
1091
1092                 inflight_info->desc[old_free_head].last = free_head;
1093                 free_head = inflight_info->desc[free_head].next;
1094                 inflight_info->free_head = free_head;
1095                 head = (head + 1) % vq->size;
1096         }
1097
1098         inflight_info->old_free_head = free_head;
1099         *inflight_entry = old_free_head;
1100
1101         return 0;
1102 }
1103
1104 int
1105 rte_vhost_clr_inflight_desc_split(int vid, uint16_t vring_idx,
1106                                   uint16_t last_used_idx, uint16_t idx)
1107 {
1108         struct virtio_net *dev;
1109         struct vhost_virtqueue *vq;
1110
1111         dev = get_device(vid);
1112         if (unlikely(!dev))
1113                 return -1;
1114
1115         if (unlikely(!(dev->protocol_features &
1116             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1117                 return 0;
1118
1119         if (unlikely(vq_is_packed(dev)))
1120                 return -1;
1121
1122         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1123                 return -1;
1124
1125         vq = dev->virtqueue[vring_idx];
1126         if (unlikely(!vq))
1127                 return -1;
1128
1129         if (unlikely(!vq->inflight_split))
1130                 return -1;
1131
1132         if (unlikely(idx >= vq->size))
1133                 return -1;
1134
1135         rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1136
1137         vq->inflight_split->desc[idx].inflight = 0;
1138
1139         rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1140
1141         vq->inflight_split->used_idx = last_used_idx;
1142         return 0;
1143 }
1144
1145 int
1146 rte_vhost_clr_inflight_desc_packed(int vid, uint16_t vring_idx,
1147                                    uint16_t head)
1148 {
1149         struct rte_vhost_inflight_info_packed *inflight_info;
1150         struct virtio_net *dev;
1151         struct vhost_virtqueue *vq;
1152
1153         dev = get_device(vid);
1154         if (unlikely(!dev))
1155                 return -1;
1156
1157         if (unlikely(!(dev->protocol_features &
1158             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1159                 return 0;
1160
1161         if (unlikely(!vq_is_packed(dev)))
1162                 return -1;
1163
1164         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1165                 return -1;
1166
1167         vq = dev->virtqueue[vring_idx];
1168         if (unlikely(!vq))
1169                 return -1;
1170
1171         inflight_info = vq->inflight_packed;
1172         if (unlikely(!inflight_info))
1173                 return -1;
1174
1175         if (unlikely(head >= vq->size))
1176                 return -1;
1177
1178         rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1179
1180         inflight_info->desc[head].inflight = 0;
1181
1182         rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1183
1184         inflight_info->old_free_head = inflight_info->free_head;
1185         inflight_info->old_used_idx = inflight_info->used_idx;
1186         inflight_info->old_used_wrap_counter = inflight_info->used_wrap_counter;
1187
1188         return 0;
1189 }
1190
1191 int
1192 rte_vhost_set_last_inflight_io_split(int vid, uint16_t vring_idx,
1193                                      uint16_t idx)
1194 {
1195         struct virtio_net *dev;
1196         struct vhost_virtqueue *vq;
1197
1198         dev = get_device(vid);
1199         if (unlikely(!dev))
1200                 return -1;
1201
1202         if (unlikely(!(dev->protocol_features &
1203             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1204                 return 0;
1205
1206         if (unlikely(vq_is_packed(dev)))
1207                 return -1;
1208
1209         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1210                 return -1;
1211
1212         vq = dev->virtqueue[vring_idx];
1213         if (unlikely(!vq))
1214                 return -1;
1215
1216         if (unlikely(!vq->inflight_split))
1217                 return -1;
1218
1219         vq->inflight_split->last_inflight_io = idx;
1220         return 0;
1221 }
1222
1223 int
1224 rte_vhost_set_last_inflight_io_packed(int vid, uint16_t vring_idx,
1225                                       uint16_t head)
1226 {
1227         struct rte_vhost_inflight_info_packed *inflight_info;
1228         struct virtio_net *dev;
1229         struct vhost_virtqueue *vq;
1230         uint16_t last;
1231
1232         dev = get_device(vid);
1233         if (unlikely(!dev))
1234                 return -1;
1235
1236         if (unlikely(!(dev->protocol_features &
1237             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1238                 return 0;
1239
1240         if (unlikely(!vq_is_packed(dev)))
1241                 return -1;
1242
1243         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1244                 return -1;
1245
1246         vq = dev->virtqueue[vring_idx];
1247         if (unlikely(!vq))
1248                 return -1;
1249
1250         inflight_info = vq->inflight_packed;
1251         if (unlikely(!inflight_info))
1252                 return -1;
1253
1254         if (unlikely(head >= vq->size))
1255                 return -1;
1256
1257         last = inflight_info->desc[head].last;
1258         if (unlikely(last >= vq->size))
1259                 return -1;
1260
1261         inflight_info->desc[last].next = inflight_info->free_head;
1262         inflight_info->free_head = head;
1263         inflight_info->used_idx += inflight_info->desc[head].num;
1264         if (inflight_info->used_idx >= inflight_info->desc_num) {
1265                 inflight_info->used_idx -= inflight_info->desc_num;
1266                 inflight_info->used_wrap_counter =
1267                         !inflight_info->used_wrap_counter;
1268         }
1269
1270         return 0;
1271 }
1272
1273 int
1274 rte_vhost_vring_call(int vid, uint16_t vring_idx)
1275 {
1276         struct virtio_net *dev;
1277         struct vhost_virtqueue *vq;
1278
1279         dev = get_device(vid);
1280         if (!dev)
1281                 return -1;
1282
1283         if (vring_idx >= VHOST_MAX_VRING)
1284                 return -1;
1285
1286         vq = dev->virtqueue[vring_idx];
1287         if (!vq)
1288                 return -1;
1289
1290         if (vq_is_packed(dev))
1291                 vhost_vring_call_packed(dev, vq);
1292         else
1293                 vhost_vring_call_split(dev, vq);
1294
1295         return 0;
1296 }
1297
1298 uint16_t
1299 rte_vhost_avail_entries(int vid, uint16_t queue_id)
1300 {
1301         struct virtio_net *dev;
1302         struct vhost_virtqueue *vq;
1303         uint16_t ret = 0;
1304
1305         dev = get_device(vid);
1306         if (!dev)
1307                 return 0;
1308
1309         if (queue_id >= VHOST_MAX_VRING)
1310                 return 0;
1311
1312         vq = dev->virtqueue[queue_id];
1313         if (!vq)
1314                 return 0;
1315
1316         rte_spinlock_lock(&vq->access_lock);
1317
1318         if (unlikely(!vq->enabled || vq->avail == NULL))
1319                 goto out;
1320
1321         ret = *(volatile uint16_t *)&vq->avail->idx - vq->last_used_idx;
1322
1323 out:
1324         rte_spinlock_unlock(&vq->access_lock);
1325         return ret;
1326 }
1327
1328 static inline int
1329 vhost_enable_notify_split(struct virtio_net *dev,
1330                 struct vhost_virtqueue *vq, int enable)
1331 {
1332         if (vq->used == NULL)
1333                 return -1;
1334
1335         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
1336                 if (enable)
1337                         vq->used->flags &= ~VRING_USED_F_NO_NOTIFY;
1338                 else
1339                         vq->used->flags |= VRING_USED_F_NO_NOTIFY;
1340         } else {
1341                 if (enable)
1342                         vhost_avail_event(vq) = vq->last_avail_idx;
1343         }
1344         return 0;
1345 }
1346
1347 static inline int
1348 vhost_enable_notify_packed(struct virtio_net *dev,
1349                 struct vhost_virtqueue *vq, int enable)
1350 {
1351         uint16_t flags;
1352
1353         if (vq->device_event == NULL)
1354                 return -1;
1355
1356         if (!enable) {
1357                 vq->device_event->flags = VRING_EVENT_F_DISABLE;
1358                 return 0;
1359         }
1360
1361         flags = VRING_EVENT_F_ENABLE;
1362         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
1363                 flags = VRING_EVENT_F_DESC;
1364                 vq->device_event->off_wrap = vq->last_avail_idx |
1365                         vq->avail_wrap_counter << 15;
1366         }
1367
1368         rte_atomic_thread_fence(__ATOMIC_RELEASE);
1369
1370         vq->device_event->flags = flags;
1371         return 0;
1372 }
1373
1374 int
1375 vhost_enable_guest_notification(struct virtio_net *dev,
1376                 struct vhost_virtqueue *vq, int enable)
1377 {
1378         /*
1379          * If the virtqueue is not ready yet, it will be applied
1380          * when it will become ready.
1381          */
1382         if (!vq->ready)
1383                 return 0;
1384
1385         if (vq_is_packed(dev))
1386                 return vhost_enable_notify_packed(dev, vq, enable);
1387         else
1388                 return vhost_enable_notify_split(dev, vq, enable);
1389 }
1390
1391 int
1392 rte_vhost_enable_guest_notification(int vid, uint16_t queue_id, int enable)
1393 {
1394         struct virtio_net *dev = get_device(vid);
1395         struct vhost_virtqueue *vq;
1396         int ret;
1397
1398         if (!dev)
1399                 return -1;
1400
1401         if (queue_id >= VHOST_MAX_VRING)
1402                 return -1;
1403
1404         vq = dev->virtqueue[queue_id];
1405         if (!vq)
1406                 return -1;
1407
1408         rte_spinlock_lock(&vq->access_lock);
1409
1410         vq->notif_enable = enable;
1411         ret = vhost_enable_guest_notification(dev, vq, enable);
1412
1413         rte_spinlock_unlock(&vq->access_lock);
1414
1415         return ret;
1416 }
1417
1418 void
1419 rte_vhost_log_write(int vid, uint64_t addr, uint64_t len)
1420 {
1421         struct virtio_net *dev = get_device(vid);
1422
1423         if (dev == NULL)
1424                 return;
1425
1426         vhost_log_write(dev, addr, len);
1427 }
1428
1429 void
1430 rte_vhost_log_used_vring(int vid, uint16_t vring_idx,
1431                          uint64_t offset, uint64_t len)
1432 {
1433         struct virtio_net *dev;
1434         struct vhost_virtqueue *vq;
1435
1436         dev = get_device(vid);
1437         if (dev == NULL)
1438                 return;
1439
1440         if (vring_idx >= VHOST_MAX_VRING)
1441                 return;
1442         vq = dev->virtqueue[vring_idx];
1443         if (!vq)
1444                 return;
1445
1446         vhost_log_used_vring(dev, vq, offset, len);
1447 }
1448
1449 uint32_t
1450 rte_vhost_rx_queue_count(int vid, uint16_t qid)
1451 {
1452         struct virtio_net *dev;
1453         struct vhost_virtqueue *vq;
1454         uint32_t ret = 0;
1455
1456         dev = get_device(vid);
1457         if (dev == NULL)
1458                 return 0;
1459
1460         if (unlikely(qid >= dev->nr_vring || (qid & 1) == 0)) {
1461                 VHOST_LOG_DATA(ERR, "(%d) %s: invalid virtqueue idx %d.\n",
1462                         dev->vid, __func__, qid);
1463                 return 0;
1464         }
1465
1466         vq = dev->virtqueue[qid];
1467         if (vq == NULL)
1468                 return 0;
1469
1470         rte_spinlock_lock(&vq->access_lock);
1471
1472         if (unlikely(!vq->enabled || vq->avail == NULL))
1473                 goto out;
1474
1475         ret = *((volatile uint16_t *)&vq->avail->idx) - vq->last_avail_idx;
1476
1477 out:
1478         rte_spinlock_unlock(&vq->access_lock);
1479         return ret;
1480 }
1481
1482 struct rte_vdpa_device *
1483 rte_vhost_get_vdpa_device(int vid)
1484 {
1485         struct virtio_net *dev = get_device(vid);
1486
1487         if (dev == NULL)
1488                 return NULL;
1489
1490         return dev->vdpa_dev;
1491 }
1492
1493 int rte_vhost_get_log_base(int vid, uint64_t *log_base,
1494                 uint64_t *log_size)
1495 {
1496         struct virtio_net *dev = get_device(vid);
1497
1498         if (dev == NULL || log_base == NULL || log_size == NULL)
1499                 return -1;
1500
1501         *log_base = dev->log_base;
1502         *log_size = dev->log_size;
1503
1504         return 0;
1505 }
1506
1507 int rte_vhost_get_vring_base(int vid, uint16_t queue_id,
1508                 uint16_t *last_avail_idx, uint16_t *last_used_idx)
1509 {
1510         struct vhost_virtqueue *vq;
1511         struct virtio_net *dev = get_device(vid);
1512
1513         if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
1514                 return -1;
1515
1516         if (queue_id >= VHOST_MAX_VRING)
1517                 return -1;
1518
1519         vq = dev->virtqueue[queue_id];
1520         if (!vq)
1521                 return -1;
1522
1523         if (vq_is_packed(dev)) {
1524                 *last_avail_idx = (vq->avail_wrap_counter << 15) |
1525                                   vq->last_avail_idx;
1526                 *last_used_idx = (vq->used_wrap_counter << 15) |
1527                                  vq->last_used_idx;
1528         } else {
1529                 *last_avail_idx = vq->last_avail_idx;
1530                 *last_used_idx = vq->last_used_idx;
1531         }
1532
1533         return 0;
1534 }
1535
1536 int rte_vhost_set_vring_base(int vid, uint16_t queue_id,
1537                 uint16_t last_avail_idx, uint16_t last_used_idx)
1538 {
1539         struct vhost_virtqueue *vq;
1540         struct virtio_net *dev = get_device(vid);
1541
1542         if (!dev)
1543                 return -1;
1544
1545         if (queue_id >= VHOST_MAX_VRING)
1546                 return -1;
1547
1548         vq = dev->virtqueue[queue_id];
1549         if (!vq)
1550                 return -1;
1551
1552         if (vq_is_packed(dev)) {
1553                 vq->last_avail_idx = last_avail_idx & 0x7fff;
1554                 vq->avail_wrap_counter = !!(last_avail_idx & (1 << 15));
1555                 vq->last_used_idx = last_used_idx & 0x7fff;
1556                 vq->used_wrap_counter = !!(last_used_idx & (1 << 15));
1557         } else {
1558                 vq->last_avail_idx = last_avail_idx;
1559                 vq->last_used_idx = last_used_idx;
1560         }
1561
1562         return 0;
1563 }
1564
1565 int
1566 rte_vhost_get_vring_base_from_inflight(int vid,
1567                                        uint16_t queue_id,
1568                                        uint16_t *last_avail_idx,
1569                                        uint16_t *last_used_idx)
1570 {
1571         struct rte_vhost_inflight_info_packed *inflight_info;
1572         struct vhost_virtqueue *vq;
1573         struct virtio_net *dev = get_device(vid);
1574
1575         if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
1576                 return -1;
1577
1578         if (queue_id >= VHOST_MAX_VRING)
1579                 return -1;
1580
1581         vq = dev->virtqueue[queue_id];
1582         if (!vq)
1583                 return -1;
1584
1585         if (!vq_is_packed(dev))
1586                 return -1;
1587
1588         inflight_info = vq->inflight_packed;
1589         if (!inflight_info)
1590                 return -1;
1591
1592         *last_avail_idx = (inflight_info->old_used_wrap_counter << 15) |
1593                           inflight_info->old_used_idx;
1594         *last_used_idx = *last_avail_idx;
1595
1596         return 0;
1597 }
1598
1599 int rte_vhost_extern_callback_register(int vid,
1600                 struct rte_vhost_user_extern_ops const * const ops, void *ctx)
1601 {
1602         struct virtio_net *dev = get_device(vid);
1603
1604         if (dev == NULL || ops == NULL)
1605                 return -1;
1606
1607         dev->extern_ops = *ops;
1608         dev->extern_data = ctx;
1609         return 0;
1610 }
1611
1612 int rte_vhost_async_channel_register(int vid, uint16_t queue_id,
1613                                         uint32_t features,
1614                                         struct rte_vhost_async_channel_ops *ops)
1615 {
1616         struct vhost_virtqueue *vq;
1617         struct virtio_net *dev = get_device(vid);
1618         struct rte_vhost_async_features f;
1619         int node;
1620
1621         if (dev == NULL || ops == NULL)
1622                 return -1;
1623
1624         f.intval = features;
1625
1626         if (queue_id >= VHOST_MAX_VRING)
1627                 return -1;
1628
1629         vq = dev->virtqueue[queue_id];
1630
1631         if (unlikely(vq == NULL || !dev->async_copy))
1632                 return -1;
1633
1634         if (unlikely(!f.async_inorder)) {
1635                 VHOST_LOG_CONFIG(ERR,
1636                         "async copy is not supported on non-inorder mode "
1637                         "(vid %d, qid: %d)\n", vid, queue_id);
1638                 return -1;
1639         }
1640
1641         if (unlikely(ops->check_completed_copies == NULL ||
1642                 ops->transfer_data == NULL))
1643                 return -1;
1644
1645         rte_spinlock_lock(&vq->access_lock);
1646
1647         if (unlikely(vq->async_registered)) {
1648                 VHOST_LOG_CONFIG(ERR,
1649                         "async register failed: channel already registered "
1650                         "(vid %d, qid: %d)\n", vid, queue_id);
1651                 goto reg_out;
1652         }
1653
1654 #ifdef RTE_LIBRTE_VHOST_NUMA
1655         if (get_mempolicy(&node, NULL, 0, vq, MPOL_F_NODE | MPOL_F_ADDR)) {
1656                 VHOST_LOG_CONFIG(ERR,
1657                         "unable to get numa information in async register. "
1658                         "allocating async buffer memory on the caller thread node\n");
1659                 node = SOCKET_ID_ANY;
1660         }
1661 #else
1662         node = SOCKET_ID_ANY;
1663 #endif
1664
1665         vq->async_pkts_info = rte_malloc_socket(NULL,
1666                         vq->size * sizeof(struct async_inflight_info),
1667                         RTE_CACHE_LINE_SIZE, node);
1668         if (!vq->async_pkts_info) {
1669                 vhost_free_async_mem(vq);
1670                 VHOST_LOG_CONFIG(ERR,
1671                         "async register failed: cannot allocate memory for async_pkts_info "
1672                         "(vid %d, qid: %d)\n", vid, queue_id);
1673                 goto reg_out;
1674         }
1675
1676         vq->it_pool = rte_malloc_socket(NULL,
1677                         VHOST_MAX_ASYNC_IT * sizeof(struct rte_vhost_iov_iter),
1678                         RTE_CACHE_LINE_SIZE, node);
1679         if (!vq->it_pool) {
1680                 vhost_free_async_mem(vq);
1681                 VHOST_LOG_CONFIG(ERR,
1682                         "async register failed: cannot allocate memory for it_pool "
1683                         "(vid %d, qid: %d)\n", vid, queue_id);
1684                 goto reg_out;
1685         }
1686
1687         vq->vec_pool = rte_malloc_socket(NULL,
1688                         VHOST_MAX_ASYNC_VEC * sizeof(struct iovec),
1689                         RTE_CACHE_LINE_SIZE, node);
1690         if (!vq->vec_pool) {
1691                 vhost_free_async_mem(vq);
1692                 VHOST_LOG_CONFIG(ERR,
1693                         "async register failed: cannot allocate memory for vec_pool "
1694                         "(vid %d, qid: %d)\n", vid, queue_id);
1695                 goto reg_out;
1696         }
1697
1698         if (vq_is_packed(dev)) {
1699                 vq->async_buffers_packed = rte_malloc_socket(NULL,
1700                         vq->size * sizeof(struct vring_used_elem_packed),
1701                         RTE_CACHE_LINE_SIZE, node);
1702                 if (!vq->async_buffers_packed) {
1703                         vhost_free_async_mem(vq);
1704                         VHOST_LOG_CONFIG(ERR,
1705                                 "async register failed: cannot allocate memory for async buffers "
1706                                 "(vid %d, qid: %d)\n", vid, queue_id);
1707                         goto reg_out;
1708                 }
1709         } else {
1710                 vq->async_descs_split = rte_malloc_socket(NULL,
1711                         vq->size * sizeof(struct vring_used_elem),
1712                         RTE_CACHE_LINE_SIZE, node);
1713                 if (!vq->async_descs_split) {
1714                         vhost_free_async_mem(vq);
1715                         VHOST_LOG_CONFIG(ERR,
1716                                 "async register failed: cannot allocate memory for async descs "
1717                                 "(vid %d, qid: %d)\n", vid, queue_id);
1718                         goto reg_out;
1719                 }
1720         }
1721
1722         vq->async_ops.check_completed_copies = ops->check_completed_copies;
1723         vq->async_ops.transfer_data = ops->transfer_data;
1724
1725         vq->async_inorder = f.async_inorder;
1726         vq->async_threshold = f.async_threshold;
1727
1728         vq->async_registered = true;
1729
1730 reg_out:
1731         rte_spinlock_unlock(&vq->access_lock);
1732
1733         return 0;
1734 }
1735
1736 int rte_vhost_async_channel_unregister(int vid, uint16_t queue_id)
1737 {
1738         struct vhost_virtqueue *vq;
1739         struct virtio_net *dev = get_device(vid);
1740         int ret = -1;
1741
1742         if (dev == NULL)
1743                 return ret;
1744
1745         if (queue_id >= VHOST_MAX_VRING)
1746                 return ret;
1747
1748         vq = dev->virtqueue[queue_id];
1749
1750         if (vq == NULL)
1751                 return ret;
1752
1753         ret = 0;
1754
1755         if (!vq->async_registered)
1756                 return ret;
1757
1758         if (!rte_spinlock_trylock(&vq->access_lock)) {
1759                 VHOST_LOG_CONFIG(ERR, "Failed to unregister async channel. "
1760                         "virt queue busy.\n");
1761                 return -1;
1762         }
1763
1764         if (vq->async_pkts_inflight_n) {
1765                 VHOST_LOG_CONFIG(ERR, "Failed to unregister async channel. "
1766                         "async inflight packets must be completed before unregistration.\n");
1767                 ret = -1;
1768                 goto out;
1769         }
1770
1771         vhost_free_async_mem(vq);
1772
1773         vq->async_ops.transfer_data = NULL;
1774         vq->async_ops.check_completed_copies = NULL;
1775         vq->async_registered = false;
1776
1777 out:
1778         rte_spinlock_unlock(&vq->access_lock);
1779
1780         return ret;
1781 }
1782
1783 RTE_LOG_REGISTER_SUFFIX(vhost_config_log_level, config, INFO);
1784 RTE_LOG_REGISTER_SUFFIX(vhost_data_log_level, data, WARNING);