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_atomic_thread_fence(__ATOMIC_RELEASE);
 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_atomic_thread_fence(__ATOMIC_RELEASE);
 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_atomic_thread_fence(__ATOMIC_RELEASE);
 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_atomic_thread_fence(__ATOMIC_RELEASE);
 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         if (!vq) {
 548                 VHOST_LOG_CONFIG(ERR, "Virtqueue not allocated (%d)\n",
 549                                 vring_idx);
 550                 return;
 551         }
 552
 553         memset(vq, 0, sizeof(struct vhost_virtqueue));
 554
 555         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 556         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
 557         vq->notif_enable = VIRTIO_UNINITIALIZED_NOTIF;
 558
 559         vhost_user_iotlb_init(dev, vring_idx);
 560         /* Backends are set to -1 indicating an inactive device. */
 561         vq->backend = -1;
 562 }
 563
 564 static void
 565 reset_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 566 {
 567         struct vhost_virtqueue *vq;
 568         int callfd;
 569
 570         if (vring_idx >= VHOST_MAX_VRING) {
 571                 VHOST_LOG_CONFIG(ERR,
 572                                 "Failed not init vring, out of bound (%d)\n",
 573                                 vring_idx);
 574                 return;
 575         }
 576
 577         vq = dev->virtqueue[vring_idx];
 578         if (!vq) {
 579                 VHOST_LOG_CONFIG(ERR, "Virtqueue not allocated (%d)\n",
 580                                 vring_idx);
 581                 return;
 582         }
 583
 584         callfd = vq->callfd;
 585         init_vring_queue(dev, vring_idx);
 586         vq->callfd = callfd;
 587 }
 588
 589 int
 590 alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 591 {
 592         struct vhost_virtqueue *vq;
 593         uint32_t i;
 594
 595         /* Also allocate holes, if any, up to requested vring index. */
 596         for (i = 0; i <= vring_idx; i++) {
 597                 if (dev->virtqueue[i])
 598                         continue;
 599
 600                 vq = rte_malloc(NULL, sizeof(struct vhost_virtqueue), 0);
 601                 if (vq == NULL) {
 602                         VHOST_LOG_CONFIG(ERR,
 603                                 "Failed to allocate memory for vring:%u.\n", i);
 604                         return -1;
 605                 }
 606
 607                 dev->virtqueue[i] = vq;
 608                 init_vring_queue(dev, i);
 609                 rte_spinlock_init(&vq->access_lock);
 610                 vq->avail_wrap_counter = 1;
 611                 vq->used_wrap_counter = 1;
 612                 vq->signalled_used_valid = false;
 613         }
 614
 615         dev->nr_vring = RTE_MAX(dev->nr_vring, vring_idx + 1);
 616
 617         return 0;
 618 }
 619
 620 /*
 621  * Reset some variables in device structure, while keeping few
 622  * others untouched, such as vid, ifname, nr_vring: they
 623  * should be same unless the device is removed.
 624  */
 625 void
 626 reset_device(struct virtio_net *dev)
 627 {
 628         uint32_t i;
 629
 630         dev->features = 0;
 631         dev->protocol_features = 0;
 632         dev->flags &= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 633
 634         for (i = 0; i < dev->nr_vring; i++)
 635                 reset_vring_queue(dev, i);
 636 }
 637
 638 /*
 639  * Invoked when there is a new vhost-user connection established (when
 640  * there is a new virtio device being attached).
 641  */
 642 int
 643 vhost_new_device(void)
 644 {
 645         struct virtio_net *dev;
 646         int i;
 647
 648         for (i = 0; i < MAX_VHOST_DEVICE; i++) {
 649                 if (vhost_devices[i] == NULL)
 650                         break;
 651         }
 652
 653         if (i == MAX_VHOST_DEVICE) {
 654                 VHOST_LOG_CONFIG(ERR,
 655                         "Failed to find a free slot for new device.\n");
 656                 return -1;
 657         }
 658
 659         dev = rte_zmalloc(NULL, sizeof(struct virtio_net), 0);
 660         if (dev == NULL) {
 661                 VHOST_LOG_CONFIG(ERR,
 662                         "Failed to allocate memory for new dev.\n");
 663                 return -1;
 664         }
 665
 666         vhost_devices[i] = dev;
 667         dev->vid = i;
 668         dev->flags = VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 669         dev->slave_req_fd = -1;
 670         dev->postcopy_ufd = -1;
 671         rte_spinlock_init(&dev->slave_req_lock);
 672
 673         return i;
 674 }
 675
 676 void
 677 vhost_destroy_device_notify(struct virtio_net *dev)
 678 {
 679         struct rte_vdpa_device *vdpa_dev;
 680
 681         if (dev->flags & VIRTIO_DEV_RUNNING) {
 682                 vdpa_dev = dev->vdpa_dev;
 683                 if (vdpa_dev)
 684                         vdpa_dev->ops->dev_close(dev->vid);
 685                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 686                 dev->notify_ops->destroy_device(dev->vid);
 687         }
 688 }
 689
 690 /*
 691  * Invoked when there is the vhost-user connection is broken (when
 692  * the virtio device is being detached).
 693  */
 694 void
 695 vhost_destroy_device(int vid)
 696 {
 697         struct virtio_net *dev = get_device(vid);
 698
 699         if (dev == NULL)
 700                 return;
 701
 702         vhost_destroy_device_notify(dev);
 703
 704         cleanup_device(dev, 1);
 705         free_device(dev);
 706
 707         vhost_devices[vid] = NULL;
 708 }
 709
 710 void
 711 vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *vdpa_dev)
 712 {
 713         struct virtio_net *dev = get_device(vid);
 714
 715         if (dev == NULL)
 716                 return;
 717
 718         dev->vdpa_dev = vdpa_dev;
 719 }
 720
 721 void
 722 vhost_set_ifname(int vid, const char *if_name, unsigned int if_len)
 723 {
 724         struct virtio_net *dev;
 725         unsigned int len;
 726
 727         dev = get_device(vid);
 728         if (dev == NULL)
 729                 return;
 730
 731         len = if_len > sizeof(dev->ifname) ?
 732                 sizeof(dev->ifname) : if_len;
 733
 734         strncpy(dev->ifname, if_name, len);
 735         dev->ifname[sizeof(dev->ifname) - 1] = '\0';
 736 }
 737
 738 void
 739 vhost_set_builtin_virtio_net(int vid, bool enable)
 740 {
 741         struct virtio_net *dev = get_device(vid);
 742
 743         if (dev == NULL)
 744                 return;
 745
 746         if (enable)
 747                 dev->flags |= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 748         else
 749                 dev->flags &= ~VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 750 }
 751
 752 void
 753 vhost_enable_extbuf(int vid)
 754 {
 755         struct virtio_net *dev = get_device(vid);
 756
 757         if (dev == NULL)
 758                 return;
 759
 760         dev->extbuf = 1;
 761 }
 762
 763 void
 764 vhost_enable_linearbuf(int vid)
 765 {
 766         struct virtio_net *dev = get_device(vid);
 767
 768         if (dev == NULL)
 769                 return;
 770
 771         dev->linearbuf = 1;
 772 }
 773
 774 int
 775 rte_vhost_get_mtu(int vid, uint16_t *mtu)
 776 {
 777         struct virtio_net *dev = get_device(vid);
 778
 779         if (dev == NULL || mtu == NULL)
 780                 return -ENODEV;
 781
 782         if (!(dev->flags & VIRTIO_DEV_READY))
 783                 return -EAGAIN;
 784
 785         if (!(dev->features & (1ULL << VIRTIO_NET_F_MTU)))
 786                 return -ENOTSUP;
 787
 788         *mtu = dev->mtu;
 789
 790         return 0;
 791 }
 792
 793 int
 794 rte_vhost_get_numa_node(int vid)
 795 {
 796 #ifdef RTE_LIBRTE_VHOST_NUMA
 797         struct virtio_net *dev = get_device(vid);
 798         int numa_node;
 799         int ret;
 800
 801         if (dev == NULL || numa_available() != 0)
 802                 return -1;
 803
 804         ret = get_mempolicy(&numa_node, NULL, 0, dev,
 805                             MPOL_F_NODE | MPOL_F_ADDR);
 806         if (ret < 0) {
 807                 VHOST_LOG_CONFIG(ERR,
 808                         "(%d) failed to query numa node: %s\n",
 809                         vid, rte_strerror(errno));
 810                 return -1;
 811         }
 812
 813         return numa_node;
 814 #else
 815         RTE_SET_USED(vid);
 816         return -1;
 817 #endif
 818 }
 819
 820 uint32_t
 821 rte_vhost_get_queue_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 / 2;
 829 }
 830
 831 uint16_t
 832 rte_vhost_get_vring_num(int vid)
 833 {
 834         struct virtio_net *dev = get_device(vid);
 835
 836         if (dev == NULL)
 837                 return 0;
 838
 839         return dev->nr_vring;
 840 }
 841
 842 int
 843 rte_vhost_get_ifname(int vid, char *buf, size_t len)
 844 {
 845         struct virtio_net *dev = get_device(vid);
 846
 847         if (dev == NULL || buf == NULL)
 848                 return -1;
 849
 850         len = RTE_MIN(len, sizeof(dev->ifname));
 851
 852         strncpy(buf, dev->ifname, len);
 853         buf[len - 1] = '\0';
 854
 855         return 0;
 856 }
 857
 858 int
 859 rte_vhost_get_negotiated_features(int vid, uint64_t *features)
 860 {
 861         struct virtio_net *dev;
 862
 863         dev = get_device(vid);
 864         if (dev == NULL || features == NULL)
 865                 return -1;
 866
 867         *features = dev->features;
 868         return 0;
 869 }
 870
 871 int
 872 rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem)
 873 {
 874         struct virtio_net *dev;
 875         struct rte_vhost_memory *m;
 876         size_t size;
 877
 878         dev = get_device(vid);
 879         if (dev == NULL || mem == NULL)
 880                 return -1;
 881
 882         size = dev->mem->nregions * sizeof(struct rte_vhost_mem_region);
 883         m = malloc(sizeof(struct rte_vhost_memory) + size);
 884         if (!m)
 885                 return -1;
 886
 887         m->nregions = dev->mem->nregions;
 888         memcpy(m->regions, dev->mem->regions, size);
 889         *mem = m;
 890
 891         return 0;
 892 }
 893
 894 int
 895 rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
 896                           struct rte_vhost_vring *vring)
 897 {
 898         struct virtio_net *dev;
 899         struct vhost_virtqueue *vq;
 900
 901         dev = get_device(vid);
 902         if (dev == NULL || vring == NULL)
 903                 return -1;
 904
 905         if (vring_idx >= VHOST_MAX_VRING)
 906                 return -1;
 907
 908         vq = dev->virtqueue[vring_idx];
 909         if (!vq)
 910                 return -1;
 911
 912         if (vq_is_packed(dev)) {
 913                 vring->desc_packed = vq->desc_packed;
 914                 vring->driver_event = vq->driver_event;
 915                 vring->device_event = vq->device_event;
 916         } else {
 917                 vring->desc = vq->desc;
 918                 vring->avail = vq->avail;
 919                 vring->used = vq->used;
 920         }
 921         vring->log_guest_addr  = vq->log_guest_addr;
 922
 923         vring->callfd  = vq->callfd;
 924         vring->kickfd  = vq->kickfd;
 925         vring->size    = vq->size;
 926
 927         return 0;
 928 }
 929
 930 int
 931 rte_vhost_get_vhost_ring_inflight(int vid, uint16_t vring_idx,
 932                                   struct rte_vhost_ring_inflight *vring)
 933 {
 934         struct virtio_net *dev;
 935         struct vhost_virtqueue *vq;
 936
 937         dev = get_device(vid);
 938         if (unlikely(!dev))
 939                 return -1;
 940
 941         if (vring_idx >= VHOST_MAX_VRING)
 942                 return -1;
 943
 944         vq = dev->virtqueue[vring_idx];
 945         if (unlikely(!vq))
 946                 return -1;
 947
 948         if (vq_is_packed(dev)) {
 949                 if (unlikely(!vq->inflight_packed))
 950                         return -1;
 951
 952                 vring->inflight_packed = vq->inflight_packed;
 953         } else {
 954                 if (unlikely(!vq->inflight_split))
 955                         return -1;
 956
 957                 vring->inflight_split = vq->inflight_split;
 958         }
 959
 960         vring->resubmit_inflight = vq->resubmit_inflight;
 961
 962         return 0;
 963 }
 964
 965 int
 966 rte_vhost_set_inflight_desc_split(int vid, uint16_t vring_idx,
 967                                   uint16_t idx)
 968 {
 969         struct vhost_virtqueue *vq;
 970         struct virtio_net *dev;
 971
 972         dev = get_device(vid);
 973         if (unlikely(!dev))
 974                 return -1;
 975
 976         if (unlikely(!(dev->protocol_features &
 977             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
 978                 return 0;
 979
 980         if (unlikely(vq_is_packed(dev)))
 981                 return -1;
 982
 983         if (unlikely(vring_idx >= VHOST_MAX_VRING))
 984                 return -1;
 985
 986         vq = dev->virtqueue[vring_idx];
 987         if (unlikely(!vq))
 988                 return -1;
 989
 990         if (unlikely(!vq->inflight_split))
 991                 return -1;
 992
 993         if (unlikely(idx >= vq->size))
 994                 return -1;
 995
 996         vq->inflight_split->desc[idx].counter = vq->global_counter++;
 997         vq->inflight_split->desc[idx].inflight = 1;
 998         return 0;
 999 }
1000
1001 int
1002 rte_vhost_set_inflight_desc_packed(int vid, uint16_t vring_idx,
1003                                    uint16_t head, uint16_t last,
1004                                    uint16_t *inflight_entry)
1005 {
1006         struct rte_vhost_inflight_info_packed *inflight_info;
1007         struct virtio_net *dev;
1008         struct vhost_virtqueue *vq;
1009         struct vring_packed_desc *desc;
1010         uint16_t old_free_head, free_head;
1011
1012         dev = get_device(vid);
1013         if (unlikely(!dev))
1014                 return -1;
1015
1016         if (unlikely(!(dev->protocol_features &
1017             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1018                 return 0;
1019
1020         if (unlikely(!vq_is_packed(dev)))
1021                 return -1;
1022
1023         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1024                 return -1;
1025
1026         vq = dev->virtqueue[vring_idx];
1027         if (unlikely(!vq))
1028                 return -1;
1029
1030         inflight_info = vq->inflight_packed;
1031         if (unlikely(!inflight_info))
1032                 return -1;
1033
1034         if (unlikely(head >= vq->size))
1035                 return -1;
1036
1037         desc = vq->desc_packed;
1038         old_free_head = inflight_info->old_free_head;
1039         if (unlikely(old_free_head >= vq->size))
1040                 return -1;
1041
1042         free_head = old_free_head;
1043
1044         /* init header descriptor */
1045         inflight_info->desc[old_free_head].num = 0;
1046         inflight_info->desc[old_free_head].counter = vq->global_counter++;
1047         inflight_info->desc[old_free_head].inflight = 1;
1048
1049         /* save desc entry in flight entry */
1050         while (head != ((last + 1) % vq->size)) {
1051                 inflight_info->desc[old_free_head].num++;
1052                 inflight_info->desc[free_head].addr = desc[head].addr;
1053                 inflight_info->desc[free_head].len = desc[head].len;
1054                 inflight_info->desc[free_head].flags = desc[head].flags;
1055                 inflight_info->desc[free_head].id = desc[head].id;
1056
1057                 inflight_info->desc[old_free_head].last = free_head;
1058                 free_head = inflight_info->desc[free_head].next;
1059                 inflight_info->free_head = free_head;
1060                 head = (head + 1) % vq->size;
1061         }
1062
1063         inflight_info->old_free_head = free_head;
1064         *inflight_entry = old_free_head;
1065
1066         return 0;
1067 }
1068
1069 int
1070 rte_vhost_clr_inflight_desc_split(int vid, uint16_t vring_idx,
1071                                   uint16_t last_used_idx, uint16_t idx)
1072 {
1073         struct virtio_net *dev;
1074         struct vhost_virtqueue *vq;
1075
1076         dev = get_device(vid);
1077         if (unlikely(!dev))
1078                 return -1;
1079
1080         if (unlikely(!(dev->protocol_features &
1081             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1082                 return 0;
1083
1084         if (unlikely(vq_is_packed(dev)))
1085                 return -1;
1086
1087         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1088                 return -1;
1089
1090         vq = dev->virtqueue[vring_idx];
1091         if (unlikely(!vq))
1092                 return -1;
1093
1094         if (unlikely(!vq->inflight_split))
1095                 return -1;
1096
1097         if (unlikely(idx >= vq->size))
1098                 return -1;
1099
1100         rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1101
1102         vq->inflight_split->desc[idx].inflight = 0;
1103
1104         rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1105
1106         vq->inflight_split->used_idx = last_used_idx;
1107         return 0;
1108 }
1109
1110 int
1111 rte_vhost_clr_inflight_desc_packed(int vid, uint16_t vring_idx,
1112                                    uint16_t head)
1113 {
1114         struct rte_vhost_inflight_info_packed *inflight_info;
1115         struct virtio_net *dev;
1116         struct vhost_virtqueue *vq;
1117
1118         dev = get_device(vid);
1119         if (unlikely(!dev))
1120                 return -1;
1121
1122         if (unlikely(!(dev->protocol_features &
1123             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1124                 return 0;
1125
1126         if (unlikely(!vq_is_packed(dev)))
1127                 return -1;
1128
1129         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1130                 return -1;
1131
1132         vq = dev->virtqueue[vring_idx];
1133         if (unlikely(!vq))
1134                 return -1;
1135
1136         inflight_info = vq->inflight_packed;
1137         if (unlikely(!inflight_info))
1138                 return -1;
1139
1140         if (unlikely(head >= vq->size))
1141                 return -1;
1142
1143         rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1144
1145         inflight_info->desc[head].inflight = 0;
1146
1147         rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1148
1149         inflight_info->old_free_head = inflight_info->free_head;
1150         inflight_info->old_used_idx = inflight_info->used_idx;
1151         inflight_info->old_used_wrap_counter = inflight_info->used_wrap_counter;
1152
1153         return 0;
1154 }
1155
1156 int
1157 rte_vhost_set_last_inflight_io_split(int vid, uint16_t vring_idx,
1158                                      uint16_t idx)
1159 {
1160         struct virtio_net *dev;
1161         struct vhost_virtqueue *vq;
1162
1163         dev = get_device(vid);
1164         if (unlikely(!dev))
1165                 return -1;
1166
1167         if (unlikely(!(dev->protocol_features &
1168             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1169                 return 0;
1170
1171         if (unlikely(vq_is_packed(dev)))
1172                 return -1;
1173
1174         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1175                 return -1;
1176
1177         vq = dev->virtqueue[vring_idx];
1178         if (unlikely(!vq))
1179                 return -1;
1180
1181         if (unlikely(!vq->inflight_split))
1182                 return -1;
1183
1184         vq->inflight_split->last_inflight_io = idx;
1185         return 0;
1186 }
1187
1188 int
1189 rte_vhost_set_last_inflight_io_packed(int vid, uint16_t vring_idx,
1190                                       uint16_t head)
1191 {
1192         struct rte_vhost_inflight_info_packed *inflight_info;
1193         struct virtio_net *dev;
1194         struct vhost_virtqueue *vq;
1195         uint16_t last;
1196
1197         dev = get_device(vid);
1198         if (unlikely(!dev))
1199                 return -1;
1200
1201         if (unlikely(!(dev->protocol_features &
1202             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1203                 return 0;
1204
1205         if (unlikely(!vq_is_packed(dev)))
1206                 return -1;
1207
1208         if (unlikely(vring_idx >= VHOST_MAX_VRING))
1209                 return -1;
1210
1211         vq = dev->virtqueue[vring_idx];
1212         if (unlikely(!vq))
1213                 return -1;
1214
1215         inflight_info = vq->inflight_packed;
1216         if (unlikely(!inflight_info))
1217                 return -1;
1218
1219         if (unlikely(head >= vq->size))
1220                 return -1;
1221
1222         last = inflight_info->desc[head].last;
1223         if (unlikely(last >= vq->size))
1224                 return -1;
1225
1226         inflight_info->desc[last].next = inflight_info->free_head;
1227         inflight_info->free_head = head;
1228         inflight_info->used_idx += inflight_info->desc[head].num;
1229         if (inflight_info->used_idx >= inflight_info->desc_num) {
1230                 inflight_info->used_idx -= inflight_info->desc_num;
1231                 inflight_info->used_wrap_counter =
1232                         !inflight_info->used_wrap_counter;
1233         }
1234
1235         return 0;
1236 }
1237
1238 int
1239 rte_vhost_vring_call(int vid, uint16_t vring_idx)
1240 {
1241         struct virtio_net *dev;
1242         struct vhost_virtqueue *vq;
1243
1244         dev = get_device(vid);
1245         if (!dev)
1246                 return -1;
1247
1248         if (vring_idx >= VHOST_MAX_VRING)
1249                 return -1;
1250
1251         vq = dev->virtqueue[vring_idx];
1252         if (!vq)
1253                 return -1;
1254
1255         if (vq_is_packed(dev))
1256                 vhost_vring_call_packed(dev, vq);
1257         else
1258                 vhost_vring_call_split(dev, vq);
1259
1260         return 0;
1261 }
1262
1263 uint16_t
1264 rte_vhost_avail_entries(int vid, uint16_t queue_id)
1265 {
1266         struct virtio_net *dev;
1267         struct vhost_virtqueue *vq;
1268         uint16_t ret = 0;
1269
1270         dev = get_device(vid);
1271         if (!dev)
1272                 return 0;
1273
1274         if (queue_id >= VHOST_MAX_VRING)
1275                 return 0;
1276
1277         vq = dev->virtqueue[queue_id];
1278         if (!vq)
1279                 return 0;
1280
1281         rte_spinlock_lock(&vq->access_lock);
1282
1283         if (unlikely(!vq->enabled || vq->avail == NULL))
1284                 goto out;
1285
1286         ret = *(volatile uint16_t *)&vq->avail->idx - vq->last_used_idx;
1287
1288 out:
1289         rte_spinlock_unlock(&vq->access_lock);
1290         return ret;
1291 }
1292
1293 static inline int
1294 vhost_enable_notify_split(struct virtio_net *dev,
1295                 struct vhost_virtqueue *vq, int enable)
1296 {
1297         if (vq->used == NULL)
1298                 return -1;
1299
1300         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
1301                 if (enable)
1302                         vq->used->flags &= ~VRING_USED_F_NO_NOTIFY;
1303                 else
1304                         vq->used->flags |= VRING_USED_F_NO_NOTIFY;
1305         } else {
1306                 if (enable)
1307                         vhost_avail_event(vq) = vq->last_avail_idx;
1308         }
1309         return 0;
1310 }
1311
1312 static inline int
1313 vhost_enable_notify_packed(struct virtio_net *dev,
1314                 struct vhost_virtqueue *vq, int enable)
1315 {
1316         uint16_t flags;
1317
1318         if (vq->device_event == NULL)
1319                 return -1;
1320
1321         if (!enable) {
1322                 vq->device_event->flags = VRING_EVENT_F_DISABLE;
1323                 return 0;
1324         }
1325
1326         flags = VRING_EVENT_F_ENABLE;
1327         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
1328                 flags = VRING_EVENT_F_DESC;
1329                 vq->device_event->off_wrap = vq->last_avail_idx |
1330                         vq->avail_wrap_counter << 15;
1331         }
1332
1333         rte_atomic_thread_fence(__ATOMIC_RELEASE);
1334
1335         vq->device_event->flags = flags;
1336         return 0;
1337 }
1338
1339 int
1340 vhost_enable_guest_notification(struct virtio_net *dev,
1341                 struct vhost_virtqueue *vq, int enable)
1342 {
1343         /*
1344          * If the virtqueue is not ready yet, it will be applied
1345          * when it will become ready.
1346          */
1347         if (!vq->ready)
1348                 return 0;
1349
1350         if (vq_is_packed(dev))
1351                 return vhost_enable_notify_packed(dev, vq, enable);
1352         else
1353                 return vhost_enable_notify_split(dev, vq, enable);
1354 }
1355
1356 int
1357 rte_vhost_enable_guest_notification(int vid, uint16_t queue_id, int enable)
1358 {
1359         struct virtio_net *dev = get_device(vid);
1360         struct vhost_virtqueue *vq;
1361         int ret;
1362
1363         if (!dev)
1364                 return -1;
1365
1366         if (queue_id >= VHOST_MAX_VRING)
1367                 return -1;
1368
1369         vq = dev->virtqueue[queue_id];
1370         if (!vq)
1371                 return -1;
1372
1373         rte_spinlock_lock(&vq->access_lock);
1374
1375         vq->notif_enable = enable;
1376         ret = vhost_enable_guest_notification(dev, vq, enable);
1377
1378         rte_spinlock_unlock(&vq->access_lock);
1379
1380         return ret;
1381 }
1382
1383 void
1384 rte_vhost_log_write(int vid, uint64_t addr, uint64_t len)
1385 {
1386         struct virtio_net *dev = get_device(vid);
1387
1388         if (dev == NULL)
1389                 return;
1390
1391         vhost_log_write(dev, addr, len);
1392 }
1393
1394 void
1395 rte_vhost_log_used_vring(int vid, uint16_t vring_idx,
1396                          uint64_t offset, uint64_t len)
1397 {
1398         struct virtio_net *dev;
1399         struct vhost_virtqueue *vq;
1400
1401         dev = get_device(vid);
1402         if (dev == NULL)
1403                 return;
1404
1405         if (vring_idx >= VHOST_MAX_VRING)
1406                 return;
1407         vq = dev->virtqueue[vring_idx];
1408         if (!vq)
1409                 return;
1410
1411         vhost_log_used_vring(dev, vq, offset, len);
1412 }
1413
1414 uint32_t
1415 rte_vhost_rx_queue_count(int vid, uint16_t qid)
1416 {
1417         struct virtio_net *dev;
1418         struct vhost_virtqueue *vq;
1419         uint32_t ret = 0;
1420
1421         dev = get_device(vid);
1422         if (dev == NULL)
1423                 return 0;
1424
1425         if (unlikely(qid >= dev->nr_vring || (qid & 1) == 0)) {
1426                 VHOST_LOG_DATA(ERR, "(%d) %s: invalid virtqueue idx %d.\n",
1427                         dev->vid, __func__, qid);
1428                 return 0;
1429         }
1430
1431         vq = dev->virtqueue[qid];
1432         if (vq == NULL)
1433                 return 0;
1434
1435         rte_spinlock_lock(&vq->access_lock);
1436
1437         if (unlikely(vq->enabled == 0 || vq->avail == NULL))
1438                 goto out;
1439
1440         ret = *((volatile uint16_t *)&vq->avail->idx) - vq->last_avail_idx;
1441
1442 out:
1443         rte_spinlock_unlock(&vq->access_lock);
1444         return ret;
1445 }
1446
1447 struct rte_vdpa_device *
1448 rte_vhost_get_vdpa_device(int vid)
1449 {
1450         struct virtio_net *dev = get_device(vid);
1451
1452         if (dev == NULL)
1453                 return NULL;
1454
1455         return dev->vdpa_dev;
1456 }
1457
1458 int rte_vhost_get_log_base(int vid, uint64_t *log_base,
1459                 uint64_t *log_size)
1460 {
1461         struct virtio_net *dev = get_device(vid);
1462
1463         if (dev == NULL || log_base == NULL || log_size == NULL)
1464                 return -1;
1465
1466         *log_base = dev->log_base;
1467         *log_size = dev->log_size;
1468
1469         return 0;
1470 }
1471
1472 int rte_vhost_get_vring_base(int vid, uint16_t queue_id,
1473                 uint16_t *last_avail_idx, uint16_t *last_used_idx)
1474 {
1475         struct vhost_virtqueue *vq;
1476         struct virtio_net *dev = get_device(vid);
1477
1478         if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
1479                 return -1;
1480
1481         if (queue_id >= VHOST_MAX_VRING)
1482                 return -1;
1483
1484         vq = dev->virtqueue[queue_id];
1485         if (!vq)
1486                 return -1;
1487
1488         if (vq_is_packed(dev)) {
1489                 *last_avail_idx = (vq->avail_wrap_counter << 15) |
1490                                   vq->last_avail_idx;
1491                 *last_used_idx = (vq->used_wrap_counter << 15) |
1492                                  vq->last_used_idx;
1493         } else {
1494                 *last_avail_idx = vq->last_avail_idx;
1495                 *last_used_idx = vq->last_used_idx;
1496         }
1497
1498         return 0;
1499 }
1500
1501 int rte_vhost_set_vring_base(int vid, uint16_t queue_id,
1502                 uint16_t last_avail_idx, uint16_t last_used_idx)
1503 {
1504         struct vhost_virtqueue *vq;
1505         struct virtio_net *dev = get_device(vid);
1506
1507         if (!dev)
1508                 return -1;
1509
1510         if (queue_id >= VHOST_MAX_VRING)
1511                 return -1;
1512
1513         vq = dev->virtqueue[queue_id];
1514         if (!vq)
1515                 return -1;
1516
1517         if (vq_is_packed(dev)) {
1518                 vq->last_avail_idx = last_avail_idx & 0x7fff;
1519                 vq->avail_wrap_counter = !!(last_avail_idx & (1 << 15));
1520                 vq->last_used_idx = last_used_idx & 0x7fff;
1521                 vq->used_wrap_counter = !!(last_used_idx & (1 << 15));
1522         } else {
1523                 vq->last_avail_idx = last_avail_idx;
1524                 vq->last_used_idx = last_used_idx;
1525         }
1526
1527         return 0;
1528 }
1529
1530 int
1531 rte_vhost_get_vring_base_from_inflight(int vid,
1532                                        uint16_t queue_id,
1533                                        uint16_t *last_avail_idx,
1534                                        uint16_t *last_used_idx)
1535 {
1536         struct rte_vhost_inflight_info_packed *inflight_info;
1537         struct vhost_virtqueue *vq;
1538         struct virtio_net *dev = get_device(vid);
1539
1540         if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
1541                 return -1;
1542
1543         if (queue_id >= VHOST_MAX_VRING)
1544                 return -1;
1545
1546         vq = dev->virtqueue[queue_id];
1547         if (!vq)
1548                 return -1;
1549
1550         if (!vq_is_packed(dev))
1551                 return -1;
1552
1553         inflight_info = vq->inflight_packed;
1554         if (!inflight_info)
1555                 return -1;
1556
1557         *last_avail_idx = (inflight_info->old_used_wrap_counter << 15) |
1558                           inflight_info->old_used_idx;
1559         *last_used_idx = *last_avail_idx;
1560
1561         return 0;
1562 }
1563
1564 int rte_vhost_extern_callback_register(int vid,
1565                 struct rte_vhost_user_extern_ops const * const ops, void *ctx)
1566 {
1567         struct virtio_net *dev = get_device(vid);
1568
1569         if (dev == NULL || ops == NULL)
1570                 return -1;
1571
1572         dev->extern_ops = *ops;
1573         dev->extern_data = ctx;
1574         return 0;
1575 }
1576
1577 int rte_vhost_async_channel_register(int vid, uint16_t queue_id,
1578                                         uint32_t features,
1579                                         struct rte_vhost_async_channel_ops *ops)
1580 {
1581         struct vhost_virtqueue *vq;
1582         struct virtio_net *dev = get_device(vid);
1583         struct rte_vhost_async_features f;
1584         int node;
1585
1586         if (dev == NULL || ops == NULL)
1587                 return -1;
1588
1589         f.intval = features;
1590
1591         if (queue_id >= VHOST_MAX_VRING)
1592                 return -1;
1593
1594         vq = dev->virtqueue[queue_id];
1595
1596         if (unlikely(vq == NULL || !dev->async_copy))
1597                 return -1;
1598
1599         /* packed queue is not supported */
1600         if (unlikely(vq_is_packed(dev) || !f.async_inorder)) {
1601                 VHOST_LOG_CONFIG(ERR,
1602                         "async copy is not supported on packed queue or non-inorder mode "
1603                         "(vid %d, qid: %d)\n", vid, queue_id);
1604                 return -1;
1605         }
1606
1607         if (unlikely(ops->check_completed_copies == NULL ||
1608                 ops->transfer_data == NULL))
1609                 return -1;
1610
1611         rte_spinlock_lock(&vq->access_lock);
1612
1613         if (unlikely(vq->async_registered)) {
1614                 VHOST_LOG_CONFIG(ERR,
1615                         "async register failed: channel already registered "
1616                         "(vid %d, qid: %d)\n", vid, queue_id);
1617                 goto reg_out;
1618         }
1619
1620 #ifdef RTE_LIBRTE_VHOST_NUMA
1621         if (get_mempolicy(&node, NULL, 0, vq, MPOL_F_NODE | MPOL_F_ADDR)) {
1622                 VHOST_LOG_CONFIG(ERR,
1623                         "unable to get numa information in async register. "
1624                         "allocating async buffer memory on the caller thread node\n");
1625                 node = SOCKET_ID_ANY;
1626         }
1627 #else
1628         node = SOCKET_ID_ANY;
1629 #endif
1630
1631         vq->async_pkts_pending = rte_malloc_socket(NULL,
1632                         vq->size * sizeof(uintptr_t),
1633                         RTE_CACHE_LINE_SIZE, node);
1634         vq->async_pkts_info = rte_malloc_socket(NULL,
1635                         vq->size * sizeof(struct async_inflight_info),
1636                         RTE_CACHE_LINE_SIZE, node);
1637         vq->it_pool = rte_malloc_socket(NULL,
1638                         VHOST_MAX_ASYNC_IT * sizeof(struct rte_vhost_iov_iter),
1639                         RTE_CACHE_LINE_SIZE, node);
1640         vq->vec_pool = rte_malloc_socket(NULL,
1641                         VHOST_MAX_ASYNC_VEC * sizeof(struct iovec),
1642                         RTE_CACHE_LINE_SIZE, node);
1643         if (!vq->async_pkts_pending || !vq->async_pkts_info ||
1644                 !vq->it_pool || !vq->vec_pool) {
1645                 vhost_free_async_mem(vq);
1646                 VHOST_LOG_CONFIG(ERR,
1647                                 "async register failed: cannot allocate memory for vq data "
1648                                 "(vid %d, qid: %d)\n", vid, queue_id);
1649                 goto reg_out;
1650         }
1651
1652         vq->async_ops.check_completed_copies = ops->check_completed_copies;
1653         vq->async_ops.transfer_data = ops->transfer_data;
1654
1655         vq->async_inorder = f.async_inorder;
1656         vq->async_threshold = f.async_threshold;
1657
1658         vq->async_registered = true;
1659
1660 reg_out:
1661         rte_spinlock_unlock(&vq->access_lock);
1662
1663         return 0;
1664 }
1665
1666 int rte_vhost_async_channel_unregister(int vid, uint16_t queue_id)
1667 {
1668         struct vhost_virtqueue *vq;
1669         struct virtio_net *dev = get_device(vid);
1670         int ret = -1;
1671
1672         if (dev == NULL)
1673                 return ret;
1674
1675         if (queue_id >= VHOST_MAX_VRING)
1676                 return ret;
1677
1678         vq = dev->virtqueue[queue_id];
1679
1680         if (vq == NULL)
1681                 return ret;
1682
1683         ret = 0;
1684
1685         if (!vq->async_registered)
1686                 return ret;
1687
1688         if (!rte_spinlock_trylock(&vq->access_lock)) {
1689                 VHOST_LOG_CONFIG(ERR, "Failed to unregister async channel. "
1690                         "virt queue busy.\n");
1691                 return -1;
1692         }
1693
1694         if (vq->async_pkts_inflight_n) {
1695                 VHOST_LOG_CONFIG(ERR, "Failed to unregister async channel. "
1696                         "async inflight packets must be completed before unregistration.\n");
1697                 ret = -1;
1698                 goto out;
1699         }
1700
1701         vhost_free_async_mem(vq);
1702
1703         vq->async_ops.transfer_data = NULL;
1704         vq->async_ops.check_completed_copies = NULL;
1705         vq->async_registered = false;
1706
1707 out:
1708         rte_spinlock_unlock(&vq->access_lock);
1709
1710         return ret;
1711 }
1712
1713 RTE_LOG_REGISTER(vhost_config_log_level, lib.vhost.config, INFO);
1714 RTE_LOG_REGISTER(vhost_data_log_level, lib.vhost.data, WARNING);