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                         RTE_LOG(ERR, VHOST_CONFIG,
  61                                 "IOTLB miss req failed for IOVA 0x%" PRIx64 "\n",
  62                                 iova);
  63                         vhost_user_iotlb_pending_remove(vq, iova, 1, perm);
  64                 }
  65
  66                 vhost_user_iotlb_rd_lock(vq);
  67         }
  68
  69         return 0;
  70 }
  71
  72 #define VHOST_LOG_PAGE  4096
  73
  74 /*
  75  * Atomically set a bit in memory.
  76  */
  77 static __rte_always_inline void
  78 vhost_set_bit(unsigned int nr, volatile uint8_t *addr)
  79 {
  80 #if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION < 70100)
  81         /*
  82          * __sync_ built-ins are deprecated, but __atomic_ ones
  83          * are sub-optimized in older GCC versions.
  84          */
  85         __sync_fetch_and_or_1(addr, (1U << nr));
  86 #else
  87         __atomic_fetch_or(addr, (1U << nr), __ATOMIC_RELAXED);
  88 #endif
  89 }
  90
  91 static __rte_always_inline void
  92 vhost_log_page(uint8_t *log_base, uint64_t page)
  93 {
  94         vhost_set_bit(page % 8, &log_base[page / 8]);
  95 }
  96
  97 void
  98 __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
  99 {
 100         uint64_t page;
 101
 102         if (unlikely(!dev->log_base || !len))
 103                 return;
 104
 105         if (unlikely(dev->log_size <= ((addr + len - 1) / VHOST_LOG_PAGE / 8)))
 106                 return;
 107
 108         /* To make sure guest memory updates are committed before logging */
 109         rte_smp_wmb();
 110
 111         page = addr / VHOST_LOG_PAGE;
 112         while (page * VHOST_LOG_PAGE < addr + len) {
 113                 vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
 114                 page += 1;
 115         }
 116 }
 117
 118 void
 119 __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 120                              uint64_t iova, uint64_t len)
 121 {
 122         uint64_t hva, gpa, map_len;
 123         map_len = len;
 124
 125         hva = __vhost_iova_to_vva(dev, vq, iova, &map_len, VHOST_ACCESS_RW);
 126         if (map_len != len) {
 127                 RTE_LOG(ERR, VHOST_CONFIG,
 128                         "Failed to write log for IOVA 0x%" PRIx64 ". No IOTLB entry found\n",
 129                         iova);
 130                 return;
 131         }
 132
 133         gpa = hva_to_gpa(dev, hva, len);
 134         if (gpa)
 135                 __vhost_log_write(dev, gpa, len);
 136 }
 137
 138 void
 139 __vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
 140 {
 141         unsigned long *log_base;
 142         int i;
 143
 144         if (unlikely(!dev->log_base))
 145                 return;
 146
 147         rte_smp_wmb();
 148
 149         log_base = (unsigned long *)(uintptr_t)dev->log_base;
 150
 151         for (i = 0; i < vq->log_cache_nb_elem; i++) {
 152                 struct log_cache_entry *elem = vq->log_cache + i;
 153
 154 #if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION < 70100)
 155                 /*
 156                  * '__sync' builtins are deprecated, but '__atomic' ones
 157                  * are sub-optimized in older GCC versions.
 158                  */
 159                 __sync_fetch_and_or(log_base + elem->offset, elem->val);
 160 #else
 161                 __atomic_fetch_or(log_base + elem->offset, elem->val,
 162                                 __ATOMIC_RELAXED);
 163 #endif
 164         }
 165
 166         rte_smp_wmb();
 167
 168         vq->log_cache_nb_elem = 0;
 169 }
 170
 171 static __rte_always_inline void
 172 vhost_log_cache_page(struct virtio_net *dev, struct vhost_virtqueue *vq,
 173                         uint64_t page)
 174 {
 175         uint32_t bit_nr = page % (sizeof(unsigned long) << 3);
 176         uint32_t offset = page / (sizeof(unsigned long) << 3);
 177         int i;
 178
 179         for (i = 0; i < vq->log_cache_nb_elem; i++) {
 180                 struct log_cache_entry *elem = vq->log_cache + i;
 181
 182                 if (elem->offset == offset) {
 183                         elem->val |= (1UL << bit_nr);
 184                         return;
 185                 }
 186         }
 187
 188         if (unlikely(i >= VHOST_LOG_CACHE_NR)) {
 189                 /*
 190                  * No more room for a new log cache entry,
 191                  * so write the dirty log map directly.
 192                  */
 193                 rte_smp_wmb();
 194                 vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
 195
 196                 return;
 197         }
 198
 199         vq->log_cache[i].offset = offset;
 200         vq->log_cache[i].val = (1UL << bit_nr);
 201         vq->log_cache_nb_elem++;
 202 }
 203
 204 void
 205 __vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
 206                         uint64_t addr, uint64_t len)
 207 {
 208         uint64_t page;
 209
 210         if (unlikely(!dev->log_base || !len))
 211                 return;
 212
 213         if (unlikely(dev->log_size <= ((addr + len - 1) / VHOST_LOG_PAGE / 8)))
 214                 return;
 215
 216         page = addr / VHOST_LOG_PAGE;
 217         while (page * VHOST_LOG_PAGE < addr + len) {
 218                 vhost_log_cache_page(dev, vq, page);
 219                 page += 1;
 220         }
 221 }
 222
 223 void
 224 __vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 225                              uint64_t iova, uint64_t len)
 226 {
 227         uint64_t hva, gpa, map_len;
 228         map_len = len;
 229
 230         hva = __vhost_iova_to_vva(dev, vq, iova, &map_len, VHOST_ACCESS_RW);
 231         if (map_len != len) {
 232                 RTE_LOG(ERR, VHOST_CONFIG,
 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 /*
 286  * Unmap any memory, close any file descriptors and
 287  * free any memory owned by a device.
 288  */
 289 void
 290 cleanup_device(struct virtio_net *dev, int destroy)
 291 {
 292         uint32_t i;
 293
 294         vhost_backend_cleanup(dev);
 295
 296         for (i = 0; i < dev->nr_vring; i++)
 297                 cleanup_vq(dev->virtqueue[i], destroy);
 298 }
 299
 300 void
 301 free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq)
 302 {
 303         if (vq_is_packed(dev))
 304                 rte_free(vq->shadow_used_packed);
 305         else
 306                 rte_free(vq->shadow_used_split);
 307         rte_free(vq->batch_copy_elems);
 308         rte_mempool_free(vq->iotlb_pool);
 309         rte_free(vq);
 310 }
 311
 312 /*
 313  * Release virtqueues and device memory.
 314  */
 315 static void
 316 free_device(struct virtio_net *dev)
 317 {
 318         uint32_t i;
 319
 320         for (i = 0; i < dev->nr_vring; i++)
 321                 free_vq(dev, dev->virtqueue[i]);
 322
 323         rte_free(dev);
 324 }
 325
 326 static int
 327 vring_translate_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 328 {
 329         uint64_t req_size, size;
 330
 331         req_size = sizeof(struct vring_desc) * vq->size;
 332         size = req_size;
 333         vq->desc = (struct vring_desc *)(uintptr_t)vhost_iova_to_vva(dev, vq,
 334                                                 vq->ring_addrs.desc_user_addr,
 335                                                 &size, VHOST_ACCESS_RW);
 336         if (!vq->desc || size != req_size)
 337                 return -1;
 338
 339         req_size = sizeof(struct vring_avail);
 340         req_size += sizeof(uint16_t) * vq->size;
 341         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 342                 req_size += sizeof(uint16_t);
 343         size = req_size;
 344         vq->avail = (struct vring_avail *)(uintptr_t)vhost_iova_to_vva(dev, vq,
 345                                                 vq->ring_addrs.avail_user_addr,
 346                                                 &size, VHOST_ACCESS_RW);
 347         if (!vq->avail || size != req_size)
 348                 return -1;
 349
 350         req_size = sizeof(struct vring_used);
 351         req_size += sizeof(struct vring_used_elem) * vq->size;
 352         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 353                 req_size += sizeof(uint16_t);
 354         size = req_size;
 355         vq->used = (struct vring_used *)(uintptr_t)vhost_iova_to_vva(dev, vq,
 356                                                 vq->ring_addrs.used_user_addr,
 357                                                 &size, VHOST_ACCESS_RW);
 358         if (!vq->used || size != req_size)
 359                 return -1;
 360
 361         return 0;
 362 }
 363
 364 static int
 365 vring_translate_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
 366 {
 367         uint64_t req_size, size;
 368
 369         req_size = sizeof(struct vring_packed_desc) * vq->size;
 370         size = req_size;
 371         vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
 372                 vhost_iova_to_vva(dev, vq, vq->ring_addrs.desc_user_addr,
 373                                 &size, VHOST_ACCESS_RW);
 374         if (!vq->desc_packed || size != req_size)
 375                 return -1;
 376
 377         req_size = sizeof(struct vring_packed_desc_event);
 378         size = req_size;
 379         vq->driver_event = (struct vring_packed_desc_event *)(uintptr_t)
 380                 vhost_iova_to_vva(dev, vq, vq->ring_addrs.avail_user_addr,
 381                                 &size, VHOST_ACCESS_RW);
 382         if (!vq->driver_event || size != req_size)
 383                 return -1;
 384
 385         req_size = sizeof(struct vring_packed_desc_event);
 386         size = req_size;
 387         vq->device_event = (struct vring_packed_desc_event *)(uintptr_t)
 388                 vhost_iova_to_vva(dev, vq, vq->ring_addrs.used_user_addr,
 389                                 &size, VHOST_ACCESS_RW);
 390         if (!vq->device_event || size != req_size)
 391                 return -1;
 392
 393         return 0;
 394 }
 395
 396 int
 397 vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq)
 398 {
 399
 400         if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 401                 return -1;
 402
 403         if (vq_is_packed(dev)) {
 404                 if (vring_translate_packed(dev, vq) < 0)
 405                         return -1;
 406         } else {
 407                 if (vring_translate_split(dev, vq) < 0)
 408                         return -1;
 409         }
 410         vq->access_ok = 1;
 411
 412         return 0;
 413 }
 414
 415 void
 416 vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq)
 417 {
 418         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 419                 vhost_user_iotlb_wr_lock(vq);
 420
 421         vq->access_ok = 0;
 422         vq->desc = NULL;
 423         vq->avail = NULL;
 424         vq->used = NULL;
 425         vq->log_guest_addr = 0;
 426
 427         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 428                 vhost_user_iotlb_wr_unlock(vq);
 429 }
 430
 431 static void
 432 init_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 433 {
 434         struct vhost_virtqueue *vq;
 435
 436         if (vring_idx >= VHOST_MAX_VRING) {
 437                 RTE_LOG(ERR, VHOST_CONFIG,
 438                                 "Failed not init vring, out of bound (%d)\n",
 439                                 vring_idx);
 440                 return;
 441         }
 442
 443         vq = dev->virtqueue[vring_idx];
 444
 445         memset(vq, 0, sizeof(struct vhost_virtqueue));
 446
 447         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 448         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
 449
 450         vhost_user_iotlb_init(dev, vring_idx);
 451         /* Backends are set to -1 indicating an inactive device. */
 452         vq->backend = -1;
 453
 454         TAILQ_INIT(&vq->zmbuf_list);
 455 }
 456
 457 static void
 458 reset_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 459 {
 460         struct vhost_virtqueue *vq;
 461         int callfd;
 462
 463         if (vring_idx >= VHOST_MAX_VRING) {
 464                 RTE_LOG(ERR, VHOST_CONFIG,
 465                                 "Failed not init vring, out of bound (%d)\n",
 466                                 vring_idx);
 467                 return;
 468         }
 469
 470         vq = dev->virtqueue[vring_idx];
 471         callfd = vq->callfd;
 472         init_vring_queue(dev, vring_idx);
 473         vq->callfd = callfd;
 474 }
 475
 476 int
 477 alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 478 {
 479         struct vhost_virtqueue *vq;
 480
 481         vq = rte_malloc(NULL, sizeof(struct vhost_virtqueue), 0);
 482         if (vq == NULL) {
 483                 RTE_LOG(ERR, VHOST_CONFIG,
 484                         "Failed to allocate memory for vring:%u.\n", vring_idx);
 485                 return -1;
 486         }
 487
 488         dev->virtqueue[vring_idx] = vq;
 489         init_vring_queue(dev, vring_idx);
 490         rte_spinlock_init(&vq->access_lock);
 491         vq->avail_wrap_counter = 1;
 492         vq->used_wrap_counter = 1;
 493         vq->signalled_used_valid = false;
 494
 495         dev->nr_vring += 1;
 496
 497         return 0;
 498 }
 499
 500 /*
 501  * Reset some variables in device structure, while keeping few
 502  * others untouched, such as vid, ifname, nr_vring: they
 503  * should be same unless the device is removed.
 504  */
 505 void
 506 reset_device(struct virtio_net *dev)
 507 {
 508         uint32_t i;
 509
 510         dev->features = 0;
 511         dev->protocol_features = 0;
 512         dev->flags &= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 513
 514         for (i = 0; i < dev->nr_vring; i++)
 515                 reset_vring_queue(dev, i);
 516 }
 517
 518 /*
 519  * Invoked when there is a new vhost-user connection established (when
 520  * there is a new virtio device being attached).
 521  */
 522 int
 523 vhost_new_device(void)
 524 {
 525         struct virtio_net *dev;
 526         int i;
 527
 528         for (i = 0; i < MAX_VHOST_DEVICE; i++) {
 529                 if (vhost_devices[i] == NULL)
 530                         break;
 531         }
 532
 533         if (i == MAX_VHOST_DEVICE) {
 534                 RTE_LOG(ERR, VHOST_CONFIG,
 535                         "Failed to find a free slot for new device.\n");
 536                 return -1;
 537         }
 538
 539         dev = rte_zmalloc(NULL, sizeof(struct virtio_net), 0);
 540         if (dev == NULL) {
 541                 RTE_LOG(ERR, VHOST_CONFIG,
 542                         "Failed to allocate memory for new dev.\n");
 543                 return -1;
 544         }
 545
 546         vhost_devices[i] = dev;
 547         dev->vid = i;
 548         dev->flags = VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 549         dev->slave_req_fd = -1;
 550         dev->vdpa_dev_id = -1;
 551         dev->postcopy_ufd = -1;
 552         rte_spinlock_init(&dev->slave_req_lock);
 553
 554         return i;
 555 }
 556
 557 void
 558 vhost_destroy_device_notify(struct virtio_net *dev)
 559 {
 560         struct rte_vdpa_device *vdpa_dev;
 561         int did;
 562
 563         if (dev->flags & VIRTIO_DEV_RUNNING) {
 564                 did = dev->vdpa_dev_id;
 565                 vdpa_dev = rte_vdpa_get_device(did);
 566                 if (vdpa_dev && vdpa_dev->ops->dev_close)
 567                         vdpa_dev->ops->dev_close(dev->vid);
 568                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 569                 dev->notify_ops->destroy_device(dev->vid);
 570         }
 571 }
 572
 573 /*
 574  * Invoked when there is the vhost-user connection is broken (when
 575  * the virtio device is being detached).
 576  */
 577 void
 578 vhost_destroy_device(int vid)
 579 {
 580         struct virtio_net *dev = get_device(vid);
 581
 582         if (dev == NULL)
 583                 return;
 584
 585         vhost_destroy_device_notify(dev);
 586
 587         cleanup_device(dev, 1);
 588         free_device(dev);
 589
 590         vhost_devices[vid] = NULL;
 591 }
 592
 593 void
 594 vhost_attach_vdpa_device(int vid, int did)
 595 {
 596         struct virtio_net *dev = get_device(vid);
 597
 598         if (dev == NULL)
 599                 return;
 600
 601         if (rte_vdpa_get_device(did) == NULL)
 602                 return;
 603
 604         dev->vdpa_dev_id = did;
 605 }
 606
 607 void
 608 vhost_set_ifname(int vid, const char *if_name, unsigned int if_len)
 609 {
 610         struct virtio_net *dev;
 611         unsigned int len;
 612
 613         dev = get_device(vid);
 614         if (dev == NULL)
 615                 return;
 616
 617         len = if_len > sizeof(dev->ifname) ?
 618                 sizeof(dev->ifname) : if_len;
 619
 620         strncpy(dev->ifname, if_name, len);
 621         dev->ifname[sizeof(dev->ifname) - 1] = '\0';
 622 }
 623
 624 void
 625 vhost_enable_dequeue_zero_copy(int vid)
 626 {
 627         struct virtio_net *dev = get_device(vid);
 628
 629         if (dev == NULL)
 630                 return;
 631
 632         dev->dequeue_zero_copy = 1;
 633 }
 634
 635 void
 636 vhost_set_builtin_virtio_net(int vid, bool enable)
 637 {
 638         struct virtio_net *dev = get_device(vid);
 639
 640         if (dev == NULL)
 641                 return;
 642
 643         if (enable)
 644                 dev->flags |= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 645         else
 646                 dev->flags &= ~VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 647 }
 648
 649 int
 650 rte_vhost_get_mtu(int vid, uint16_t *mtu)
 651 {
 652         struct virtio_net *dev = get_device(vid);
 653
 654         if (dev == NULL || mtu == NULL)
 655                 return -ENODEV;
 656
 657         if (!(dev->flags & VIRTIO_DEV_READY))
 658                 return -EAGAIN;
 659
 660         if (!(dev->features & (1ULL << VIRTIO_NET_F_MTU)))
 661                 return -ENOTSUP;
 662
 663         *mtu = dev->mtu;
 664
 665         return 0;
 666 }
 667
 668 int
 669 rte_vhost_get_numa_node(int vid)
 670 {
 671 #ifdef RTE_LIBRTE_VHOST_NUMA
 672         struct virtio_net *dev = get_device(vid);
 673         int numa_node;
 674         int ret;
 675
 676         if (dev == NULL || numa_available() != 0)
 677                 return -1;
 678
 679         ret = get_mempolicy(&numa_node, NULL, 0, dev,
 680                             MPOL_F_NODE | MPOL_F_ADDR);
 681         if (ret < 0) {
 682                 RTE_LOG(ERR, VHOST_CONFIG,
 683                         "(%d) failed to query numa node: %s\n",
 684                         vid, rte_strerror(errno));
 685                 return -1;
 686         }
 687
 688         return numa_node;
 689 #else
 690         RTE_SET_USED(vid);
 691         return -1;
 692 #endif
 693 }
 694
 695 uint32_t
 696 rte_vhost_get_queue_num(int vid)
 697 {
 698         struct virtio_net *dev = get_device(vid);
 699
 700         if (dev == NULL)
 701                 return 0;
 702
 703         return dev->nr_vring / 2;
 704 }
 705
 706 uint16_t
 707 rte_vhost_get_vring_num(int vid)
 708 {
 709         struct virtio_net *dev = get_device(vid);
 710
 711         if (dev == NULL)
 712                 return 0;
 713
 714         return dev->nr_vring;
 715 }
 716
 717 int
 718 rte_vhost_get_ifname(int vid, char *buf, size_t len)
 719 {
 720         struct virtio_net *dev = get_device(vid);
 721
 722         if (dev == NULL || buf == NULL)
 723                 return -1;
 724
 725         len = RTE_MIN(len, sizeof(dev->ifname));
 726
 727         strncpy(buf, dev->ifname, len);
 728         buf[len - 1] = '\0';
 729
 730         return 0;
 731 }
 732
 733 int
 734 rte_vhost_get_negotiated_features(int vid, uint64_t *features)
 735 {
 736         struct virtio_net *dev;
 737
 738         dev = get_device(vid);
 739         if (dev == NULL || features == NULL)
 740                 return -1;
 741
 742         *features = dev->features;
 743         return 0;
 744 }
 745
 746 int
 747 rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem)
 748 {
 749         struct virtio_net *dev;
 750         struct rte_vhost_memory *m;
 751         size_t size;
 752
 753         dev = get_device(vid);
 754         if (dev == NULL || mem == NULL)
 755                 return -1;
 756
 757         size = dev->mem->nregions * sizeof(struct rte_vhost_mem_region);
 758         m = malloc(sizeof(struct rte_vhost_memory) + size);
 759         if (!m)
 760                 return -1;
 761
 762         m->nregions = dev->mem->nregions;
 763         memcpy(m->regions, dev->mem->regions, size);
 764         *mem = m;
 765
 766         return 0;
 767 }
 768
 769 int
 770 rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
 771                           struct rte_vhost_vring *vring)
 772 {
 773         struct virtio_net *dev;
 774         struct vhost_virtqueue *vq;
 775
 776         dev = get_device(vid);
 777         if (dev == NULL || vring == NULL)
 778                 return -1;
 779
 780         if (vring_idx >= VHOST_MAX_VRING)
 781                 return -1;
 782
 783         vq = dev->virtqueue[vring_idx];
 784         if (!vq)
 785                 return -1;
 786
 787         vring->desc  = vq->desc;
 788         vring->avail = vq->avail;
 789         vring->used  = vq->used;
 790         vring->log_guest_addr  = vq->log_guest_addr;
 791
 792         vring->callfd  = vq->callfd;
 793         vring->kickfd  = vq->kickfd;
 794         vring->size    = vq->size;
 795
 796         return 0;
 797 }
 798
 799 int
 800 rte_vhost_vring_call(int vid, uint16_t vring_idx)
 801 {
 802         struct virtio_net *dev;
 803         struct vhost_virtqueue *vq;
 804
 805         dev = get_device(vid);
 806         if (!dev)
 807                 return -1;
 808
 809         if (vring_idx >= VHOST_MAX_VRING)
 810                 return -1;
 811
 812         vq = dev->virtqueue[vring_idx];
 813         if (!vq)
 814                 return -1;
 815
 816         if (vq_is_packed(dev))
 817                 vhost_vring_call_packed(dev, vq);
 818         else
 819                 vhost_vring_call_split(dev, vq);
 820
 821         return 0;
 822 }
 823
 824 uint16_t
 825 rte_vhost_avail_entries(int vid, uint16_t queue_id)
 826 {
 827         struct virtio_net *dev;
 828         struct vhost_virtqueue *vq;
 829         uint16_t ret = 0;
 830
 831         dev = get_device(vid);
 832         if (!dev)
 833                 return 0;
 834
 835         vq = dev->virtqueue[queue_id];
 836
 837         rte_spinlock_lock(&vq->access_lock);
 838
 839         if (unlikely(!vq->enabled || vq->avail == NULL))
 840                 goto out;
 841
 842         ret = *(volatile uint16_t *)&vq->avail->idx - vq->last_used_idx;
 843
 844 out:
 845         rte_spinlock_unlock(&vq->access_lock);
 846         return ret;
 847 }
 848
 849 static inline int
 850 vhost_enable_notify_split(struct virtio_net *dev,
 851                 struct vhost_virtqueue *vq, int enable)
 852 {
 853         if (vq->used == NULL)
 854                 return -1;
 855
 856         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
 857                 if (enable)
 858                         vq->used->flags &= ~VRING_USED_F_NO_NOTIFY;
 859                 else
 860                         vq->used->flags |= VRING_USED_F_NO_NOTIFY;
 861         } else {
 862                 if (enable)
 863                         vhost_avail_event(vq) = vq->last_avail_idx;
 864         }
 865         return 0;
 866 }
 867
 868 static inline int
 869 vhost_enable_notify_packed(struct virtio_net *dev,
 870                 struct vhost_virtqueue *vq, int enable)
 871 {
 872         uint16_t flags;
 873
 874         if (vq->device_event == NULL)
 875                 return -1;
 876
 877         if (!enable) {
 878                 vq->device_event->flags = VRING_EVENT_F_DISABLE;
 879                 return 0;
 880         }
 881
 882         flags = VRING_EVENT_F_ENABLE;
 883         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
 884                 flags = VRING_EVENT_F_DESC;
 885                 vq->device_event->off_wrap = vq->last_avail_idx |
 886                         vq->avail_wrap_counter << 15;
 887         }
 888
 889         rte_smp_wmb();
 890
 891         vq->device_event->flags = flags;
 892         return 0;
 893 }
 894
 895 int
 896 rte_vhost_enable_guest_notification(int vid, uint16_t queue_id, int enable)
 897 {
 898         struct virtio_net *dev = get_device(vid);
 899         struct vhost_virtqueue *vq;
 900         int ret;
 901
 902         if (!dev)
 903                 return -1;
 904
 905         vq = dev->virtqueue[queue_id];
 906
 907         rte_spinlock_lock(&vq->access_lock);
 908
 909         if (vq_is_packed(dev))
 910                 ret = vhost_enable_notify_packed(dev, vq, enable);
 911         else
 912                 ret = vhost_enable_notify_split(dev, vq, enable);
 913
 914         rte_spinlock_unlock(&vq->access_lock);
 915
 916         return ret;
 917 }
 918
 919 void
 920 rte_vhost_log_write(int vid, uint64_t addr, uint64_t len)
 921 {
 922         struct virtio_net *dev = get_device(vid);
 923
 924         if (dev == NULL)
 925                 return;
 926
 927         vhost_log_write(dev, addr, len);
 928 }
 929
 930 void
 931 rte_vhost_log_used_vring(int vid, uint16_t vring_idx,
 932                          uint64_t offset, uint64_t len)
 933 {
 934         struct virtio_net *dev;
 935         struct vhost_virtqueue *vq;
 936
 937         dev = get_device(vid);
 938         if (dev == NULL)
 939                 return;
 940
 941         if (vring_idx >= VHOST_MAX_VRING)
 942                 return;
 943         vq = dev->virtqueue[vring_idx];
 944         if (!vq)
 945                 return;
 946
 947         vhost_log_used_vring(dev, vq, offset, len);
 948 }
 949
 950 uint32_t
 951 rte_vhost_rx_queue_count(int vid, uint16_t qid)
 952 {
 953         struct virtio_net *dev;
 954         struct vhost_virtqueue *vq;
 955         uint32_t ret = 0;
 956
 957         dev = get_device(vid);
 958         if (dev == NULL)
 959                 return 0;
 960
 961         if (unlikely(qid >= dev->nr_vring || (qid & 1) == 0)) {
 962                 RTE_LOG(ERR, VHOST_DATA, "(%d) %s: invalid virtqueue idx %d.\n",
 963                         dev->vid, __func__, qid);
 964                 return 0;
 965         }
 966
 967         vq = dev->virtqueue[qid];
 968         if (vq == NULL)
 969                 return 0;
 970
 971         rte_spinlock_lock(&vq->access_lock);
 972
 973         if (unlikely(vq->enabled == 0 || vq->avail == NULL))
 974                 goto out;
 975
 976         ret = *((volatile uint16_t *)&vq->avail->idx) - vq->last_avail_idx;
 977
 978 out:
 979         rte_spinlock_unlock(&vq->access_lock);
 980         return ret;
 981 }
 982
 983 int rte_vhost_get_vdpa_device_id(int vid)
 984 {
 985         struct virtio_net *dev = get_device(vid);
 986
 987         if (dev == NULL)
 988                 return -1;
 989
 990         return dev->vdpa_dev_id;
 991 }
 992
 993 int rte_vhost_get_log_base(int vid, uint64_t *log_base,
 994                 uint64_t *log_size)
 995 {
 996         struct virtio_net *dev = get_device(vid);
 997
 998         if (dev == NULL || log_base == NULL || log_size == NULL)
 999                 return -1;
1000
1001         *log_base = dev->log_base;
1002         *log_size = dev->log_size;
1003
1004         return 0;
1005 }
1006
1007 int rte_vhost_get_vring_base(int vid, uint16_t queue_id,
1008                 uint16_t *last_avail_idx, uint16_t *last_used_idx)
1009 {
1010         struct virtio_net *dev = get_device(vid);
1011
1012         if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
1013                 return -1;
1014
1015         *last_avail_idx = dev->virtqueue[queue_id]->last_avail_idx;
1016         *last_used_idx = dev->virtqueue[queue_id]->last_used_idx;
1017
1018         return 0;
1019 }
1020
1021 int rte_vhost_set_vring_base(int vid, uint16_t queue_id,
1022                 uint16_t last_avail_idx, uint16_t last_used_idx)
1023 {
1024         struct virtio_net *dev = get_device(vid);
1025
1026         if (!dev)
1027                 return -1;
1028
1029         dev->virtqueue[queue_id]->last_avail_idx = last_avail_idx;
1030         dev->virtqueue[queue_id]->last_used_idx = last_used_idx;
1031
1032         return 0;
1033 }
1034
1035 int rte_vhost_extern_callback_register(int vid,
1036                 struct rte_vhost_user_extern_ops const * const ops, void *ctx)
1037 {
1038         struct virtio_net *dev = get_device(vid);
1039
1040         if (dev == NULL || ops == NULL)
1041                 return -1;
1042
1043         dev->extern_ops = *ops;
1044         dev->extern_data = ctx;
1045         return 0;
1046 }