lib/librte_vhost/vhost.h

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2018 Intel Corporation
   3  */
   4
   5 #ifndef _VHOST_NET_CDEV_H_
   6 #define _VHOST_NET_CDEV_H_
   7 #include <stdint.h>
   8 #include <stdio.h>
   9 #include <stdbool.h>
  10 #include <sys/types.h>
  11 #include <sys/queue.h>
  12 #include <unistd.h>
  13 #include <linux/vhost.h>
  14 #include <linux/virtio_net.h>
  15 #include <sys/socket.h>
  16 #include <linux/if.h>
  17
  18 #include <rte_log.h>
  19 #include <rte_ether.h>
  20 #include <rte_rwlock.h>
  21 #include <rte_malloc.h>
  22
  23 #include "rte_vhost.h"
  24 #include "rte_vdpa.h"
  25 #include "rte_vdpa_dev.h"
  26
  27 /* Used to indicate that the device is running on a data core */
  28 #define VIRTIO_DEV_RUNNING 1
  29 /* Used to indicate that the device is ready to operate */
  30 #define VIRTIO_DEV_READY 2
  31 /* Used to indicate that the built-in vhost net device backend is enabled */
  32 #define VIRTIO_DEV_BUILTIN_VIRTIO_NET 4
  33 /* Used to indicate that the device has its own data path and configured */
  34 #define VIRTIO_DEV_VDPA_CONFIGURED 8
  35
  36 /* Backend value set by guest. */
  37 #define VIRTIO_DEV_STOPPED -1
  38
  39 #define BUF_VECTOR_MAX 256
  40
  41 #define VHOST_LOG_CACHE_NR 32
  42
  43 #define PACKED_DESC_ENQUEUE_USED_FLAG(w)        \
  44         ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED | VRING_DESC_F_WRITE) : \
  45                 VRING_DESC_F_WRITE)
  46 #define PACKED_DESC_DEQUEUE_USED_FLAG(w)        \
  47         ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED) : 0x0)
  48 #define PACKED_DESC_SINGLE_DEQUEUE_FLAG (VRING_DESC_F_NEXT | \
  49                                          VRING_DESC_F_INDIRECT)
  50
  51 #define PACKED_BATCH_SIZE (RTE_CACHE_LINE_SIZE / \
  52                             sizeof(struct vring_packed_desc))
  53 #define PACKED_BATCH_MASK (PACKED_BATCH_SIZE - 1)
  54
  55 #ifdef VHOST_GCC_UNROLL_PRAGMA
  56 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("GCC unroll 4") \
  57         for (iter = val; iter < size; iter++)
  58 #endif
  59
  60 #ifdef VHOST_CLANG_UNROLL_PRAGMA
  61 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll 4") \
  62         for (iter = val; iter < size; iter++)
  63 #endif
  64
  65 #ifdef VHOST_ICC_UNROLL_PRAGMA
  66 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll (4)") \
  67         for (iter = val; iter < size; iter++)
  68 #endif
  69
  70 #ifndef vhost_for_each_try_unroll
  71 #define vhost_for_each_try_unroll(iter, val, num) \
  72         for (iter = val; iter < num; iter++)
  73 #endif
  74
  75 /**
  76  * Structure contains buffer address, length and descriptor index
  77  * from vring to do scatter RX.
  78  */
  79 struct buf_vector {
  80         uint64_t buf_iova;
  81         uint64_t buf_addr;
  82         uint32_t buf_len;
  83         uint32_t desc_idx;
  84 };
  85
  86 /*
  87  * A structure to hold some fields needed in zero copy code path,
  88  * mainly for associating an mbuf with the right desc_idx.
  89  */
  90 struct zcopy_mbuf {
  91         struct rte_mbuf *mbuf;
  92         uint32_t desc_idx;
  93         uint16_t desc_count;
  94         uint16_t in_use;
  95
  96         TAILQ_ENTRY(zcopy_mbuf) next;
  97 };
  98 TAILQ_HEAD(zcopy_mbuf_list, zcopy_mbuf);
  99
 100 /*
 101  * Structure contains the info for each batched memory copy.
 102  */
 103 struct batch_copy_elem {
 104         void *dst;
 105         void *src;
 106         uint32_t len;
 107         uint64_t log_addr;
 108 };
 109
 110 /*
 111  * Structure that contains the info for batched dirty logging.
 112  */
 113 struct log_cache_entry {
 114         uint32_t offset;
 115         unsigned long val;
 116 };
 117
 118 struct vring_used_elem_packed {
 119         uint16_t id;
 120         uint16_t flags;
 121         uint32_t len;
 122         uint32_t count;
 123 };
 124
 125 /**
 126  * Structure contains variables relevant to RX/TX virtqueues.
 127  */
 128 struct vhost_virtqueue {
 129         union {
 130                 struct vring_desc       *desc;
 131                 struct vring_packed_desc   *desc_packed;
 132         };
 133         union {
 134                 struct vring_avail      *avail;
 135                 struct vring_packed_desc_event *driver_event;
 136         };
 137         union {
 138                 struct vring_used       *used;
 139                 struct vring_packed_desc_event *device_event;
 140         };
 141         uint32_t                size;
 142
 143         uint16_t                last_avail_idx;
 144         uint16_t                last_used_idx;
 145         /* Last used index we notify to front end. */
 146         uint16_t                signalled_used;
 147         bool                    signalled_used_valid;
 148 #define VIRTIO_INVALID_EVENTFD          (-1)
 149 #define VIRTIO_UNINITIALIZED_EVENTFD    (-2)
 150
 151         /* Backend value to determine if device should started/stopped */
 152         int                     backend;
 153         int                     enabled;
 154         int                     access_ok;
 155         int                     ready;
 156         rte_spinlock_t          access_lock;
 157
 158         /* Used to notify the guest (trigger interrupt) */
 159         int                     callfd;
 160         /* Currently unused as polling mode is enabled */
 161         int                     kickfd;
 162
 163         /* Physical address of used ring, for logging */
 164         uint64_t                log_guest_addr;
 165
 166         /* inflight share memory info */
 167         union {
 168                 struct rte_vhost_inflight_info_split *inflight_split;
 169                 struct rte_vhost_inflight_info_packed *inflight_packed;
 170         };
 171         struct rte_vhost_resubmit_info *resubmit_inflight;
 172         uint64_t                global_counter;
 173
 174         uint16_t                nr_zmbuf;
 175         uint16_t                zmbuf_size;
 176         uint16_t                last_zmbuf_idx;
 177         struct zcopy_mbuf       *zmbufs;
 178         struct zcopy_mbuf_list  zmbuf_list;
 179
 180         union {
 181                 struct vring_used_elem  *shadow_used_split;
 182                 struct vring_used_elem_packed *shadow_used_packed;
 183         };
 184         uint16_t                shadow_used_idx;
 185         /* Record packed ring enqueue latest desc cache aligned index */
 186         uint16_t                shadow_aligned_idx;
 187         /* Record packed ring first dequeue desc index */
 188         uint16_t                shadow_last_used_idx;
 189         struct vhost_vring_addr ring_addrs;
 190
 191         struct batch_copy_elem  *batch_copy_elems;
 192         uint16_t                batch_copy_nb_elems;
 193         bool                    used_wrap_counter;
 194         bool                    avail_wrap_counter;
 195
 196         struct log_cache_entry log_cache[VHOST_LOG_CACHE_NR];
 197         uint16_t log_cache_nb_elem;
 198
 199         rte_rwlock_t    iotlb_lock;
 200         rte_rwlock_t    iotlb_pending_lock;
 201         struct rte_mempool *iotlb_pool;
 202         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
 203         int                             iotlb_cache_nr;
 204         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
 205 } __rte_cache_aligned;
 206
 207 #define VHOST_MAX_VRING                 0x100
 208 #define VHOST_MAX_QUEUE_PAIRS           0x80
 209
 210 /* Declare IOMMU related bits for older kernels */
 211 #ifndef VIRTIO_F_IOMMU_PLATFORM
 212
 213 #define VIRTIO_F_IOMMU_PLATFORM 33
 214
 215 struct vhost_iotlb_msg {
 216         __u64 iova;
 217         __u64 size;
 218         __u64 uaddr;
 219 #define VHOST_ACCESS_RO      0x1
 220 #define VHOST_ACCESS_WO      0x2
 221 #define VHOST_ACCESS_RW      0x3
 222         __u8 perm;
 223 #define VHOST_IOTLB_MISS           1
 224 #define VHOST_IOTLB_UPDATE         2
 225 #define VHOST_IOTLB_INVALIDATE     3
 226 #define VHOST_IOTLB_ACCESS_FAIL    4
 227         __u8 type;
 228 };
 229
 230 #define VHOST_IOTLB_MSG 0x1
 231
 232 struct vhost_msg {
 233         int type;
 234         union {
 235                 struct vhost_iotlb_msg iotlb;
 236                 __u8 padding[64];
 237         };
 238 };
 239 #endif
 240
 241 /*
 242  * Define virtio 1.0 for older kernels
 243  */
 244 #ifndef VIRTIO_F_VERSION_1
 245  #define VIRTIO_F_VERSION_1 32
 246 #endif
 247
 248 /* Declare packed ring related bits for older kernels */
 249 #ifndef VIRTIO_F_RING_PACKED
 250
 251 #define VIRTIO_F_RING_PACKED 34
 252
 253 struct vring_packed_desc {
 254         uint64_t addr;
 255         uint32_t len;
 256         uint16_t id;
 257         uint16_t flags;
 258 };
 259
 260 struct vring_packed_desc_event {
 261         uint16_t off_wrap;
 262         uint16_t flags;
 263 };
 264 #endif
 265
 266 /*
 267  * Declare below packed ring defines unconditionally
 268  * as Kernel header might use different names.
 269  */
 270 #define VRING_DESC_F_AVAIL      (1ULL << 7)
 271 #define VRING_DESC_F_USED       (1ULL << 15)
 272
 273 #define VRING_EVENT_F_ENABLE 0x0
 274 #define VRING_EVENT_F_DISABLE 0x1
 275 #define VRING_EVENT_F_DESC 0x2
 276
 277 /*
 278  * Available and used descs are in same order
 279  */
 280 #ifndef VIRTIO_F_IN_ORDER
 281 #define VIRTIO_F_IN_ORDER      35
 282 #endif
 283
 284 /* Features supported by this builtin vhost-user net driver. */
 285 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
 286                                 (1ULL << VIRTIO_F_ANY_LAYOUT) | \
 287                                 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
 288                                 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
 289                                 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
 290                                 (1ULL << VIRTIO_NET_F_MQ)      | \
 291                                 (1ULL << VIRTIO_F_VERSION_1)   | \
 292                                 (1ULL << VHOST_F_LOG_ALL)      | \
 293                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
 294                                 (1ULL << VIRTIO_NET_F_GSO) | \
 295                                 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
 296                                 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
 297                                 (1ULL << VIRTIO_NET_F_HOST_UFO) | \
 298                                 (1ULL << VIRTIO_NET_F_HOST_ECN) | \
 299                                 (1ULL << VIRTIO_NET_F_CSUM)    | \
 300                                 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
 301                                 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
 302                                 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
 303                                 (1ULL << VIRTIO_NET_F_GUEST_UFO) | \
 304                                 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
 305                                 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
 306                                 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \
 307                                 (1ULL << VIRTIO_NET_F_MTU)  | \
 308                                 (1ULL << VIRTIO_F_IN_ORDER) | \
 309                                 (1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
 310                                 (1ULL << VIRTIO_F_RING_PACKED))
 311
 312
 313 struct guest_page {
 314         uint64_t guest_phys_addr;
 315         uint64_t host_phys_addr;
 316         uint64_t size;
 317 };
 318
 319 struct inflight_mem_info {
 320         int             fd;
 321         void            *addr;
 322         uint64_t        size;
 323 };
 324
 325 /**
 326  * Device structure contains all configuration information relating
 327  * to the device.
 328  */
 329 struct virtio_net {
 330         /* Frontend (QEMU) memory and memory region information */
 331         struct rte_vhost_memory *mem;
 332         uint64_t                features;
 333         uint64_t                protocol_features;
 334         int                     vid;
 335         uint32_t                flags;
 336         uint16_t                vhost_hlen;
 337         /* to tell if we need broadcast rarp packet */
 338         int16_t                 broadcast_rarp;
 339         uint32_t                nr_vring;
 340         int                     dequeue_zero_copy;
 341         int                     extbuf;
 342         int                     linearbuf;
 343         struct vhost_virtqueue  *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
 344         struct inflight_mem_info *inflight_info;
 345 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
 346         char                    ifname[IF_NAME_SZ];
 347         uint64_t                log_size;
 348         uint64_t                log_base;
 349         uint64_t                log_addr;
 350         struct rte_ether_addr   mac;
 351         uint16_t                mtu;
 352
 353         struct vhost_device_ops const *notify_ops;
 354
 355         uint32_t                nr_guest_pages;
 356         uint32_t                max_guest_pages;
 357         struct guest_page       *guest_pages;
 358
 359         int                     slave_req_fd;
 360         rte_spinlock_t          slave_req_lock;
 361
 362         int                     postcopy_ufd;
 363         int                     postcopy_listening;
 364
 365         struct rte_vdpa_device *vdpa_dev;
 366
 367         /* context data for the external message handlers */
 368         void                    *extern_data;
 369         /* pre and post vhost user message handlers for the device */
 370         struct rte_vhost_user_extern_ops extern_ops;
 371 } __rte_cache_aligned;
 372
 373 static __rte_always_inline bool
 374 vq_is_packed(struct virtio_net *dev)
 375 {
 376         return dev->features & (1ull << VIRTIO_F_RING_PACKED);
 377 }
 378
 379 static inline bool
 380 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
 381 {
 382         uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
 383
 384         return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
 385                 wrap_counter != !!(flags & VRING_DESC_F_USED);
 386 }
 387
 388 static inline void
 389 vq_inc_last_used_packed(struct vhost_virtqueue *vq, uint16_t num)
 390 {
 391         vq->last_used_idx += num;
 392         if (vq->last_used_idx >= vq->size) {
 393                 vq->used_wrap_counter ^= 1;
 394                 vq->last_used_idx -= vq->size;
 395         }
 396 }
 397
 398 static inline void
 399 vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
 400 {
 401         vq->last_avail_idx += num;
 402         if (vq->last_avail_idx >= vq->size) {
 403                 vq->avail_wrap_counter ^= 1;
 404                 vq->last_avail_idx -= vq->size;
 405         }
 406 }
 407
 408 void __vhost_log_cache_write(struct virtio_net *dev,
 409                 struct vhost_virtqueue *vq,
 410                 uint64_t addr, uint64_t len);
 411 void __vhost_log_cache_write_iova(struct virtio_net *dev,
 412                 struct vhost_virtqueue *vq,
 413                 uint64_t iova, uint64_t len);
 414 void __vhost_log_cache_sync(struct virtio_net *dev,
 415                 struct vhost_virtqueue *vq);
 416 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
 417 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 418                             uint64_t iova, uint64_t len);
 419
 420 static __rte_always_inline void
 421 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
 422 {
 423         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 424                 __vhost_log_write(dev, addr, len);
 425 }
 426
 427 static __rte_always_inline void
 428 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
 429 {
 430         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 431                 __vhost_log_cache_sync(dev, vq);
 432 }
 433
 434 static __rte_always_inline void
 435 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
 436                         uint64_t addr, uint64_t len)
 437 {
 438         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 439                 __vhost_log_cache_write(dev, vq, addr, len);
 440 }
 441
 442 static __rte_always_inline void
 443 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 444                         uint64_t offset, uint64_t len)
 445 {
 446         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
 447                 if (unlikely(vq->log_guest_addr == 0))
 448                         return;
 449                 __vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset,
 450                                         len);
 451         }
 452 }
 453
 454 static __rte_always_inline void
 455 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 456                      uint64_t offset, uint64_t len)
 457 {
 458         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
 459                 if (unlikely(vq->log_guest_addr == 0))
 460                         return;
 461                 __vhost_log_write(dev, vq->log_guest_addr + offset, len);
 462         }
 463 }
 464
 465 static __rte_always_inline void
 466 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 467                            uint64_t iova, uint64_t len)
 468 {
 469         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 470                 return;
 471
 472         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 473                 __vhost_log_cache_write_iova(dev, vq, iova, len);
 474         else
 475                 __vhost_log_cache_write(dev, vq, iova, len);
 476 }
 477
 478 static __rte_always_inline void
 479 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 480                            uint64_t iova, uint64_t len)
 481 {
 482         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 483                 return;
 484
 485         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 486                 __vhost_log_write_iova(dev, vq, iova, len);
 487         else
 488                 __vhost_log_write(dev, iova, len);
 489 }
 490
 491 extern int vhost_config_log_level;
 492 extern int vhost_data_log_level;
 493
 494 #define VHOST_LOG_CONFIG(level, fmt, args...)                   \
 495         rte_log(RTE_LOG_ ## level, vhost_config_log_level,      \
 496                 "VHOST_CONFIG: " fmt, ##args)
 497
 498 #define VHOST_LOG_DATA(level, fmt, args...) \
 499         (void)((RTE_LOG_ ## level <= RTE_LOG_DP_LEVEL) ?        \
 500          rte_log(RTE_LOG_ ## level,  vhost_data_log_level,      \
 501                 "VHOST_DATA : " fmt, ##args) :                  \
 502          0)
 503
 504 #ifdef RTE_LIBRTE_VHOST_DEBUG
 505 #define VHOST_MAX_PRINT_BUFF 6072
 506 #define PRINT_PACKET(device, addr, size, header) do { \
 507         char *pkt_addr = (char *)(addr); \
 508         unsigned int index; \
 509         char packet[VHOST_MAX_PRINT_BUFF]; \
 510         \
 511         if ((header)) \
 512                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
 513         else \
 514                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
 515         for (index = 0; index < (size); index++) { \
 516                 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
 517                         "%02hhx ", pkt_addr[index]); \
 518         } \
 519         snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
 520         \
 521         VHOST_LOG_DATA(DEBUG, "%s", packet); \
 522 } while (0)
 523 #else
 524 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
 525 #endif
 526
 527 #define MAX_VHOST_DEVICE        1024
 528 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
 529
 530 #define VHOST_BINARY_SEARCH_THRESH 256
 531
 532 static __rte_always_inline int guest_page_addrcmp(const void *p1,
 533                                                 const void *p2)
 534 {
 535         const struct guest_page *page1 = (const struct guest_page *)p1;
 536         const struct guest_page *page2 = (const struct guest_page *)p2;
 537
 538         if (page1->guest_phys_addr > page2->guest_phys_addr)
 539                 return 1;
 540         if (page1->guest_phys_addr < page2->guest_phys_addr)
 541                 return -1;
 542
 543         return 0;
 544 }
 545
 546 /* Convert guest physical address to host physical address */
 547 static __rte_always_inline rte_iova_t
 548 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
 549 {
 550         uint32_t i;
 551         struct guest_page *page;
 552         struct guest_page key;
 553
 554         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
 555                 key.guest_phys_addr = gpa;
 556                 page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages,
 557                                sizeof(struct guest_page), guest_page_addrcmp);
 558                 if (page) {
 559                         if (gpa + size < page->guest_phys_addr + page->size)
 560                                 return gpa - page->guest_phys_addr +
 561                                         page->host_phys_addr;
 562                 }
 563         } else {
 564                 for (i = 0; i < dev->nr_guest_pages; i++) {
 565                         page = &dev->guest_pages[i];
 566
 567                         if (gpa >= page->guest_phys_addr &&
 568                             gpa + size < page->guest_phys_addr +
 569                             page->size)
 570                                 return gpa - page->guest_phys_addr +
 571                                        page->host_phys_addr;
 572                 }
 573         }
 574
 575         return 0;
 576 }
 577
 578 static __rte_always_inline uint64_t
 579 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
 580 {
 581         struct rte_vhost_mem_region *r;
 582         uint32_t i;
 583
 584         if (unlikely(!dev || !dev->mem))
 585                 return 0;
 586
 587         for (i = 0; i < dev->mem->nregions; i++) {
 588                 r = &dev->mem->regions[i];
 589
 590                 if (vva >= r->host_user_addr &&
 591                     vva + len <  r->host_user_addr + r->size) {
 592                         return r->guest_phys_addr + vva - r->host_user_addr;
 593                 }
 594         }
 595         return 0;
 596 }
 597
 598 static __rte_always_inline struct virtio_net *
 599 get_device(int vid)
 600 {
 601         struct virtio_net *dev = vhost_devices[vid];
 602
 603         if (unlikely(!dev)) {
 604                 VHOST_LOG_CONFIG(ERR,
 605                         "(%d) device not found.\n", vid);
 606         }
 607
 608         return dev;
 609 }
 610
 611 int vhost_new_device(void);
 612 void cleanup_device(struct virtio_net *dev, int destroy);
 613 void reset_device(struct virtio_net *dev);
 614 void vhost_destroy_device(int);
 615 void vhost_destroy_device_notify(struct virtio_net *dev);
 616
 617 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
 618 void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
 619 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
 620
 621 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
 622
 623 void vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *dev);
 624
 625 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
 626 void vhost_enable_dequeue_zero_copy(int vid);
 627 void vhost_set_builtin_virtio_net(int vid, bool enable);
 628 void vhost_enable_extbuf(int vid);
 629 void vhost_enable_linearbuf(int vid);
 630
 631 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
 632
 633 /*
 634  * Backend-specific cleanup.
 635  *
 636  * TODO: fix it; we have one backend now
 637  */
 638 void vhost_backend_cleanup(struct virtio_net *dev);
 639
 640 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 641                         uint64_t iova, uint64_t *len, uint8_t perm);
 642 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
 643                         struct vhost_virtqueue *vq,
 644                         uint64_t desc_addr, uint64_t desc_len);
 645 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 646 uint64_t translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
 647                 uint64_t log_addr);
 648 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 649
 650 static __rte_always_inline uint64_t
 651 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 652                         uint64_t iova, uint64_t *len, uint8_t perm)
 653 {
 654         if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 655                 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
 656
 657         return __vhost_iova_to_vva(dev, vq, iova, len, perm);
 658 }
 659
 660 #define vhost_avail_event(vr) \
 661         (*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
 662 #define vhost_used_event(vr) \
 663         (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
 664
 665 /*
 666  * The following is used with VIRTIO_RING_F_EVENT_IDX.
 667  * Assuming a given event_idx value from the other size, if we have
 668  * just incremented index from old to new_idx, should we trigger an
 669  * event?
 670  */
 671 static __rte_always_inline int
 672 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
 673 {
 674         return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
 675 }
 676
 677 static __rte_always_inline void
 678 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 679 {
 680         /* Flush used->idx update before we read avail->flags. */
 681         rte_smp_mb();
 682
 683         /* Don't kick guest if we don't reach index specified by guest. */
 684         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
 685                 uint16_t old = vq->signalled_used;
 686                 uint16_t new = vq->last_used_idx;
 687                 bool signalled_used_valid = vq->signalled_used_valid;
 688
 689                 vq->signalled_used = new;
 690                 vq->signalled_used_valid = true;
 691
 692                 VHOST_LOG_DATA(DEBUG, "%s: used_event_idx=%d, old=%d, new=%d\n",
 693                         __func__,
 694                         vhost_used_event(vq),
 695                         old, new);
 696
 697                 if ((vhost_need_event(vhost_used_event(vq), new, old) &&
 698                                         (vq->callfd >= 0)) ||
 699                                 unlikely(!signalled_used_valid)) {
 700                         eventfd_write(vq->callfd, (eventfd_t) 1);
 701                         if (dev->notify_ops->guest_notified)
 702                                 dev->notify_ops->guest_notified(dev->vid);
 703                 }
 704         } else {
 705                 /* Kick the guest if necessary. */
 706                 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
 707                                 && (vq->callfd >= 0)) {
 708                         eventfd_write(vq->callfd, (eventfd_t)1);
 709                         if (dev->notify_ops->guest_notified)
 710                                 dev->notify_ops->guest_notified(dev->vid);
 711                 }
 712         }
 713 }
 714
 715 static __rte_always_inline void
 716 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
 717 {
 718         uint16_t old, new, off, off_wrap;
 719         bool signalled_used_valid, kick = false;
 720
 721         /* Flush used desc update. */
 722         rte_smp_mb();
 723
 724         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
 725                 if (vq->driver_event->flags !=
 726                                 VRING_EVENT_F_DISABLE)
 727                         kick = true;
 728                 goto kick;
 729         }
 730
 731         old = vq->signalled_used;
 732         new = vq->last_used_idx;
 733         vq->signalled_used = new;
 734         signalled_used_valid = vq->signalled_used_valid;
 735         vq->signalled_used_valid = true;
 736
 737         if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
 738                 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
 739                         kick = true;
 740                 goto kick;
 741         }
 742
 743         if (unlikely(!signalled_used_valid)) {
 744                 kick = true;
 745                 goto kick;
 746         }
 747
 748         rte_smp_rmb();
 749
 750         off_wrap = vq->driver_event->off_wrap;
 751         off = off_wrap & ~(1 << 15);
 752
 753         if (new <= old)
 754                 old -= vq->size;
 755
 756         if (vq->used_wrap_counter != off_wrap >> 15)
 757                 off -= vq->size;
 758
 759         if (vhost_need_event(off, new, old))
 760                 kick = true;
 761 kick:
 762         if (kick) {
 763                 eventfd_write(vq->callfd, (eventfd_t)1);
 764                 if (dev->notify_ops->guest_notified)
 765                         dev->notify_ops->guest_notified(dev->vid);
 766         }
 767 }
 768
 769 static __rte_always_inline void
 770 free_ind_table(void *idesc)
 771 {
 772         rte_free(idesc);
 773 }
 774
 775 static __rte_always_inline void
 776 restore_mbuf(struct rte_mbuf *m)
 777 {
 778         uint32_t mbuf_size, priv_size;
 779
 780         while (m) {
 781                 priv_size = rte_pktmbuf_priv_size(m->pool);
 782                 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
 783                 /* start of buffer is after mbuf structure and priv data */
 784
 785                 m->buf_addr = (char *)m + mbuf_size;
 786                 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
 787                 m = m->next;
 788         }
 789 }
 790
 791 static __rte_always_inline bool
 792 mbuf_is_consumed(struct rte_mbuf *m)
 793 {
 794         while (m) {
 795                 if (rte_mbuf_refcnt_read(m) > 1)
 796                         return false;
 797                 m = m->next;
 798         }
 799
 800         return true;
 801 }
 802
 803 static __rte_always_inline void
 804 put_zmbuf(struct zcopy_mbuf *zmbuf)
 805 {
 806         zmbuf->in_use = 0;
 807 }
 808
 809 #endif /* _VHOST_NET_CDEV_H_ */