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 #include "rte_vhost_async.h"
  28
  29 /* Used to indicate that the device is running on a data core */
  30 #define VIRTIO_DEV_RUNNING 1
  31 /* Used to indicate that the device is ready to operate */
  32 #define VIRTIO_DEV_READY 2
  33 /* Used to indicate that the built-in vhost net device backend is enabled */
  34 #define VIRTIO_DEV_BUILTIN_VIRTIO_NET 4
  35 /* Used to indicate that the device has its own data path and configured */
  36 #define VIRTIO_DEV_VDPA_CONFIGURED 8
  37 /* Used to indicate that the feature negotiation failed */
  38 #define VIRTIO_DEV_FEATURES_FAILED 16
  39
  40 /* Backend value set by guest. */
  41 #define VIRTIO_DEV_STOPPED -1
  42
  43 #define BUF_VECTOR_MAX 256
  44
  45 #define VHOST_LOG_CACHE_NR 32
  46
  47 #define MAX_PKT_BURST 32
  48
  49 #define VHOST_MAX_ASYNC_IT (MAX_PKT_BURST * 2)
  50 #define VHOST_MAX_ASYNC_VEC (BUF_VECTOR_MAX * 2)
  51
  52 #define PACKED_DESC_ENQUEUE_USED_FLAG(w)        \
  53         ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED | VRING_DESC_F_WRITE) : \
  54                 VRING_DESC_F_WRITE)
  55 #define PACKED_DESC_DEQUEUE_USED_FLAG(w)        \
  56         ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED) : 0x0)
  57 #define PACKED_DESC_SINGLE_DEQUEUE_FLAG (VRING_DESC_F_NEXT | \
  58                                          VRING_DESC_F_INDIRECT)
  59
  60 #define PACKED_BATCH_SIZE (RTE_CACHE_LINE_SIZE / \
  61                             sizeof(struct vring_packed_desc))
  62 #define PACKED_BATCH_MASK (PACKED_BATCH_SIZE - 1)
  63
  64 #ifdef VHOST_GCC_UNROLL_PRAGMA
  65 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("GCC unroll 4") \
  66         for (iter = val; iter < size; iter++)
  67 #endif
  68
  69 #ifdef VHOST_CLANG_UNROLL_PRAGMA
  70 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll 4") \
  71         for (iter = val; iter < size; iter++)
  72 #endif
  73
  74 #ifdef VHOST_ICC_UNROLL_PRAGMA
  75 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll (4)") \
  76         for (iter = val; iter < size; iter++)
  77 #endif
  78
  79 #ifndef vhost_for_each_try_unroll
  80 #define vhost_for_each_try_unroll(iter, val, num) \
  81         for (iter = val; iter < num; iter++)
  82 #endif
  83
  84 /**
  85  * Structure contains buffer address, length and descriptor index
  86  * from vring to do scatter RX.
  87  */
  88 struct buf_vector {
  89         uint64_t buf_iova;
  90         uint64_t buf_addr;
  91         uint32_t buf_len;
  92         uint32_t desc_idx;
  93 };
  94
  95 /*
  96  * Structure contains the info for each batched memory copy.
  97  */
  98 struct batch_copy_elem {
  99         void *dst;
 100         void *src;
 101         uint32_t len;
 102         uint64_t log_addr;
 103 };
 104
 105 /*
 106  * Structure that contains the info for batched dirty logging.
 107  */
 108 struct log_cache_entry {
 109         uint32_t offset;
 110         unsigned long val;
 111 };
 112
 113 struct vring_used_elem_packed {
 114         uint16_t id;
 115         uint16_t flags;
 116         uint32_t len;
 117         uint32_t count;
 118 };
 119
 120 /**
 121  * Structure contains variables relevant to RX/TX virtqueues.
 122  */
 123 struct vhost_virtqueue {
 124         union {
 125                 struct vring_desc       *desc;
 126                 struct vring_packed_desc   *desc_packed;
 127         };
 128         union {
 129                 struct vring_avail      *avail;
 130                 struct vring_packed_desc_event *driver_event;
 131         };
 132         union {
 133                 struct vring_used       *used;
 134                 struct vring_packed_desc_event *device_event;
 135         };
 136         uint32_t                size;
 137
 138         uint16_t                last_avail_idx;
 139         uint16_t                last_used_idx;
 140         /* Last used index we notify to front end. */
 141         uint16_t                signalled_used;
 142         bool                    signalled_used_valid;
 143 #define VIRTIO_INVALID_EVENTFD          (-1)
 144 #define VIRTIO_UNINITIALIZED_EVENTFD    (-2)
 145
 146         /* Backend value to determine if device should started/stopped */
 147         int                     backend;
 148         int                     enabled;
 149         int                     access_ok;
 150         int                     ready;
 151         int                     notif_enable;
 152 #define VIRTIO_UNINITIALIZED_NOTIF      (-1)
 153
 154         rte_spinlock_t          access_lock;
 155
 156         /* Used to notify the guest (trigger interrupt) */
 157         int                     callfd;
 158         /* Currently unused as polling mode is enabled */
 159         int                     kickfd;
 160
 161         /* Physical address of used ring, for logging */
 162         uint64_t                log_guest_addr;
 163
 164         /* inflight share memory info */
 165         union {
 166                 struct rte_vhost_inflight_info_split *inflight_split;
 167                 struct rte_vhost_inflight_info_packed *inflight_packed;
 168         };
 169         struct rte_vhost_resubmit_info *resubmit_inflight;
 170         uint64_t                global_counter;
 171
 172         union {
 173                 struct vring_used_elem  *shadow_used_split;
 174                 struct vring_used_elem_packed *shadow_used_packed;
 175         };
 176         uint16_t                shadow_used_idx;
 177         /* Record packed ring enqueue latest desc cache aligned index */
 178         uint16_t                shadow_aligned_idx;
 179         /* Record packed ring first dequeue desc index */
 180         uint16_t                shadow_last_used_idx;
 181         struct vhost_vring_addr ring_addrs;
 182
 183         struct batch_copy_elem  *batch_copy_elems;
 184         uint16_t                batch_copy_nb_elems;
 185         bool                    used_wrap_counter;
 186         bool                    avail_wrap_counter;
 187
 188         struct log_cache_entry log_cache[VHOST_LOG_CACHE_NR];
 189         uint16_t log_cache_nb_elem;
 190
 191         rte_rwlock_t    iotlb_lock;
 192         rte_rwlock_t    iotlb_pending_lock;
 193         struct rte_mempool *iotlb_pool;
 194         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
 195         int                             iotlb_cache_nr;
 196         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
 197
 198         /* operation callbacks for async dma */
 199         struct rte_vhost_async_channel_ops      async_ops;
 200
 201         struct rte_vhost_iov_iter it_pool[VHOST_MAX_ASYNC_IT];
 202         struct iovec vec_pool[VHOST_MAX_ASYNC_VEC];
 203
 204         /* async data transfer status */
 205         uintptr_t       **async_pkts_pending;
 206         struct async_inflight_info *async_pkts_info;
 207         uint16_t        async_pkts_idx;
 208         uint16_t        async_pkts_inflight_n;
 209         uint16_t        async_last_pkts_n;
 210
 211         /* vq async features */
 212         bool            async_inorder;
 213         bool            async_registered;
 214         uint16_t        async_threshold;
 215 } __rte_cache_aligned;
 216
 217 /* Virtio device status as per Virtio specification */
 218 #define VIRTIO_DEVICE_STATUS_RESET              0x00
 219 #define VIRTIO_DEVICE_STATUS_ACK                0x01
 220 #define VIRTIO_DEVICE_STATUS_DRIVER             0x02
 221 #define VIRTIO_DEVICE_STATUS_DRIVER_OK          0x04
 222 #define VIRTIO_DEVICE_STATUS_FEATURES_OK        0x08
 223 #define VIRTIO_DEVICE_STATUS_DEV_NEED_RESET     0x40
 224 #define VIRTIO_DEVICE_STATUS_FAILED             0x80
 225
 226 #define VHOST_MAX_VRING                 0x100
 227 #define VHOST_MAX_QUEUE_PAIRS           0x80
 228
 229 /* Declare IOMMU related bits for older kernels */
 230 #ifndef VIRTIO_F_IOMMU_PLATFORM
 231
 232 #define VIRTIO_F_IOMMU_PLATFORM 33
 233
 234 struct vhost_iotlb_msg {
 235         __u64 iova;
 236         __u64 size;
 237         __u64 uaddr;
 238 #define VHOST_ACCESS_RO      0x1
 239 #define VHOST_ACCESS_WO      0x2
 240 #define VHOST_ACCESS_RW      0x3
 241         __u8 perm;
 242 #define VHOST_IOTLB_MISS           1
 243 #define VHOST_IOTLB_UPDATE         2
 244 #define VHOST_IOTLB_INVALIDATE     3
 245 #define VHOST_IOTLB_ACCESS_FAIL    4
 246         __u8 type;
 247 };
 248
 249 #define VHOST_IOTLB_MSG 0x1
 250
 251 struct vhost_msg {
 252         int type;
 253         union {
 254                 struct vhost_iotlb_msg iotlb;
 255                 __u8 padding[64];
 256         };
 257 };
 258 #endif
 259
 260 /*
 261  * Define virtio 1.0 for older kernels
 262  */
 263 #ifndef VIRTIO_F_VERSION_1
 264  #define VIRTIO_F_VERSION_1 32
 265 #endif
 266
 267 /* Declare packed ring related bits for older kernels */
 268 #ifndef VIRTIO_F_RING_PACKED
 269
 270 #define VIRTIO_F_RING_PACKED 34
 271
 272 struct vring_packed_desc {
 273         uint64_t addr;
 274         uint32_t len;
 275         uint16_t id;
 276         uint16_t flags;
 277 };
 278
 279 struct vring_packed_desc_event {
 280         uint16_t off_wrap;
 281         uint16_t flags;
 282 };
 283 #endif
 284
 285 /*
 286  * Declare below packed ring defines unconditionally
 287  * as Kernel header might use different names.
 288  */
 289 #define VRING_DESC_F_AVAIL      (1ULL << 7)
 290 #define VRING_DESC_F_USED       (1ULL << 15)
 291
 292 #define VRING_EVENT_F_ENABLE 0x0
 293 #define VRING_EVENT_F_DISABLE 0x1
 294 #define VRING_EVENT_F_DESC 0x2
 295
 296 /*
 297  * Available and used descs are in same order
 298  */
 299 #ifndef VIRTIO_F_IN_ORDER
 300 #define VIRTIO_F_IN_ORDER      35
 301 #endif
 302
 303 /* Features supported by this builtin vhost-user net driver. */
 304 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
 305                                 (1ULL << VIRTIO_F_ANY_LAYOUT) | \
 306                                 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
 307                                 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
 308                                 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
 309                                 (1ULL << VIRTIO_NET_F_MQ)      | \
 310                                 (1ULL << VIRTIO_F_VERSION_1)   | \
 311                                 (1ULL << VHOST_F_LOG_ALL)      | \
 312                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
 313                                 (1ULL << VIRTIO_NET_F_GSO) | \
 314                                 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
 315                                 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
 316                                 (1ULL << VIRTIO_NET_F_HOST_UFO) | \
 317                                 (1ULL << VIRTIO_NET_F_HOST_ECN) | \
 318                                 (1ULL << VIRTIO_NET_F_CSUM)    | \
 319                                 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
 320                                 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
 321                                 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
 322                                 (1ULL << VIRTIO_NET_F_GUEST_UFO) | \
 323                                 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
 324                                 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
 325                                 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \
 326                                 (1ULL << VIRTIO_NET_F_MTU)  | \
 327                                 (1ULL << VIRTIO_F_IN_ORDER) | \
 328                                 (1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
 329                                 (1ULL << VIRTIO_F_RING_PACKED))
 330
 331
 332 struct guest_page {
 333         uint64_t guest_phys_addr;
 334         uint64_t host_phys_addr;
 335         uint64_t size;
 336 };
 337
 338 struct inflight_mem_info {
 339         int             fd;
 340         void            *addr;
 341         uint64_t        size;
 342 };
 343
 344 /**
 345  * Device structure contains all configuration information relating
 346  * to the device.
 347  */
 348 struct virtio_net {
 349         /* Frontend (QEMU) memory and memory region information */
 350         struct rte_vhost_memory *mem;
 351         uint64_t                features;
 352         uint64_t                protocol_features;
 353         int                     vid;
 354         uint32_t                flags;
 355         uint16_t                vhost_hlen;
 356         /* to tell if we need broadcast rarp packet */
 357         int16_t                 broadcast_rarp;
 358         uint32_t                nr_vring;
 359         int                     async_copy;
 360         int                     extbuf;
 361         int                     linearbuf;
 362         struct vhost_virtqueue  *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
 363         struct inflight_mem_info *inflight_info;
 364 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
 365         char                    ifname[IF_NAME_SZ];
 366         uint64_t                log_size;
 367         uint64_t                log_base;
 368         uint64_t                log_addr;
 369         struct rte_ether_addr   mac;
 370         uint16_t                mtu;
 371         uint8_t                 status;
 372
 373         struct vhost_device_ops const *notify_ops;
 374
 375         uint32_t                nr_guest_pages;
 376         uint32_t                max_guest_pages;
 377         struct guest_page       *guest_pages;
 378
 379         int                     slave_req_fd;
 380         rte_spinlock_t          slave_req_lock;
 381
 382         int                     postcopy_ufd;
 383         int                     postcopy_listening;
 384
 385         struct rte_vdpa_device *vdpa_dev;
 386
 387         /* context data for the external message handlers */
 388         void                    *extern_data;
 389         /* pre and post vhost user message handlers for the device */
 390         struct rte_vhost_user_extern_ops extern_ops;
 391 } __rte_cache_aligned;
 392
 393 static __rte_always_inline bool
 394 vq_is_packed(struct virtio_net *dev)
 395 {
 396         return dev->features & (1ull << VIRTIO_F_RING_PACKED);
 397 }
 398
 399 static inline bool
 400 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
 401 {
 402         uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
 403
 404         return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
 405                 wrap_counter != !!(flags & VRING_DESC_F_USED);
 406 }
 407
 408 static inline void
 409 vq_inc_last_used_packed(struct vhost_virtqueue *vq, uint16_t num)
 410 {
 411         vq->last_used_idx += num;
 412         if (vq->last_used_idx >= vq->size) {
 413                 vq->used_wrap_counter ^= 1;
 414                 vq->last_used_idx -= vq->size;
 415         }
 416 }
 417
 418 static inline void
 419 vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
 420 {
 421         vq->last_avail_idx += num;
 422         if (vq->last_avail_idx >= vq->size) {
 423                 vq->avail_wrap_counter ^= 1;
 424                 vq->last_avail_idx -= vq->size;
 425         }
 426 }
 427
 428 void __vhost_log_cache_write(struct virtio_net *dev,
 429                 struct vhost_virtqueue *vq,
 430                 uint64_t addr, uint64_t len);
 431 void __vhost_log_cache_write_iova(struct virtio_net *dev,
 432                 struct vhost_virtqueue *vq,
 433                 uint64_t iova, uint64_t len);
 434 void __vhost_log_cache_sync(struct virtio_net *dev,
 435                 struct vhost_virtqueue *vq);
 436 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
 437 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 438                             uint64_t iova, uint64_t len);
 439
 440 static __rte_always_inline void
 441 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
 442 {
 443         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 444                 __vhost_log_write(dev, addr, len);
 445 }
 446
 447 static __rte_always_inline void
 448 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
 449 {
 450         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 451                 __vhost_log_cache_sync(dev, vq);
 452 }
 453
 454 static __rte_always_inline void
 455 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
 456                         uint64_t addr, uint64_t len)
 457 {
 458         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 459                 __vhost_log_cache_write(dev, vq, addr, len);
 460 }
 461
 462 static __rte_always_inline void
 463 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 464                         uint64_t offset, uint64_t len)
 465 {
 466         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
 467                 if (unlikely(vq->log_guest_addr == 0))
 468                         return;
 469                 __vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset,
 470                                         len);
 471         }
 472 }
 473
 474 static __rte_always_inline void
 475 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 476                      uint64_t offset, uint64_t len)
 477 {
 478         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
 479                 if (unlikely(vq->log_guest_addr == 0))
 480                         return;
 481                 __vhost_log_write(dev, vq->log_guest_addr + offset, len);
 482         }
 483 }
 484
 485 static __rte_always_inline void
 486 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 487                            uint64_t iova, uint64_t len)
 488 {
 489         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 490                 return;
 491
 492         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 493                 __vhost_log_cache_write_iova(dev, vq, iova, len);
 494         else
 495                 __vhost_log_cache_write(dev, vq, iova, len);
 496 }
 497
 498 static __rte_always_inline void
 499 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 500                            uint64_t iova, uint64_t len)
 501 {
 502         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 503                 return;
 504
 505         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 506                 __vhost_log_write_iova(dev, vq, iova, len);
 507         else
 508                 __vhost_log_write(dev, iova, len);
 509 }
 510
 511 extern int vhost_config_log_level;
 512 extern int vhost_data_log_level;
 513
 514 #define VHOST_LOG_CONFIG(level, fmt, args...)                   \
 515         rte_log(RTE_LOG_ ## level, vhost_config_log_level,      \
 516                 "VHOST_CONFIG: " fmt, ##args)
 517
 518 #define VHOST_LOG_DATA(level, fmt, args...) \
 519         (void)((RTE_LOG_ ## level <= RTE_LOG_DP_LEVEL) ?        \
 520          rte_log(RTE_LOG_ ## level,  vhost_data_log_level,      \
 521                 "VHOST_DATA : " fmt, ##args) :                  \
 522          0)
 523
 524 #ifdef RTE_LIBRTE_VHOST_DEBUG
 525 #define VHOST_MAX_PRINT_BUFF 6072
 526 #define PRINT_PACKET(device, addr, size, header) do { \
 527         char *pkt_addr = (char *)(addr); \
 528         unsigned int index; \
 529         char packet[VHOST_MAX_PRINT_BUFF]; \
 530         \
 531         if ((header)) \
 532                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
 533         else \
 534                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
 535         for (index = 0; index < (size); index++) { \
 536                 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
 537                         "%02hhx ", pkt_addr[index]); \
 538         } \
 539         snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
 540         \
 541         VHOST_LOG_DATA(DEBUG, "%s", packet); \
 542 } while (0)
 543 #else
 544 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
 545 #endif
 546
 547 #define MAX_VHOST_DEVICE        1024
 548 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
 549
 550 #define VHOST_BINARY_SEARCH_THRESH 256
 551
 552 static __rte_always_inline int guest_page_addrcmp(const void *p1,
 553                                                 const void *p2)
 554 {
 555         const struct guest_page *page1 = (const struct guest_page *)p1;
 556         const struct guest_page *page2 = (const struct guest_page *)p2;
 557
 558         if (page1->guest_phys_addr > page2->guest_phys_addr)
 559                 return 1;
 560         if (page1->guest_phys_addr < page2->guest_phys_addr)
 561                 return -1;
 562
 563         return 0;
 564 }
 565
 566 /* Convert guest physical address to host physical address */
 567 static __rte_always_inline rte_iova_t
 568 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
 569 {
 570         uint32_t i;
 571         struct guest_page *page;
 572         struct guest_page key;
 573
 574         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
 575                 key.guest_phys_addr = gpa;
 576                 page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages,
 577                                sizeof(struct guest_page), guest_page_addrcmp);
 578                 if (page) {
 579                         if (gpa + size < page->guest_phys_addr + page->size)
 580                                 return gpa - page->guest_phys_addr +
 581                                         page->host_phys_addr;
 582                 }
 583         } else {
 584                 for (i = 0; i < dev->nr_guest_pages; i++) {
 585                         page = &dev->guest_pages[i];
 586
 587                         if (gpa >= page->guest_phys_addr &&
 588                             gpa + size < page->guest_phys_addr +
 589                             page->size)
 590                                 return gpa - page->guest_phys_addr +
 591                                        page->host_phys_addr;
 592                 }
 593         }
 594
 595         return 0;
 596 }
 597
 598 static __rte_always_inline rte_iova_t
 599 gpa_to_first_hpa(struct virtio_net *dev, uint64_t gpa,
 600         uint64_t gpa_size, uint64_t *hpa_size)
 601 {
 602         uint32_t i;
 603         struct guest_page *page;
 604         struct guest_page key;
 605
 606         *hpa_size = gpa_size;
 607         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
 608                 key.guest_phys_addr = gpa & ~(dev->guest_pages[0].size - 1);
 609                 page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages,
 610                                sizeof(struct guest_page), guest_page_addrcmp);
 611                 if (page) {
 612                         if (gpa + gpa_size <=
 613                                         page->guest_phys_addr + page->size) {
 614                                 return gpa - page->guest_phys_addr +
 615                                         page->host_phys_addr;
 616                         } else if (gpa < page->guest_phys_addr +
 617                                                 page->size) {
 618                                 *hpa_size = page->guest_phys_addr +
 619                                         page->size - gpa;
 620                                 return gpa - page->guest_phys_addr +
 621                                         page->host_phys_addr;
 622                         }
 623                 }
 624         } else {
 625                 for (i = 0; i < dev->nr_guest_pages; i++) {
 626                         page = &dev->guest_pages[i];
 627
 628                         if (gpa >= page->guest_phys_addr) {
 629                                 if (gpa + gpa_size <=
 630                                         page->guest_phys_addr + page->size) {
 631                                         return gpa - page->guest_phys_addr +
 632                                                 page->host_phys_addr;
 633                                 } else if (gpa < page->guest_phys_addr +
 634                                                         page->size) {
 635                                         *hpa_size = page->guest_phys_addr +
 636                                                 page->size - gpa;
 637                                         return gpa - page->guest_phys_addr +
 638                                                 page->host_phys_addr;
 639                                 }
 640                         }
 641                 }
 642         }
 643
 644         *hpa_size = 0;
 645         return 0;
 646 }
 647
 648 static __rte_always_inline uint64_t
 649 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
 650 {
 651         struct rte_vhost_mem_region *r;
 652         uint32_t i;
 653
 654         if (unlikely(!dev || !dev->mem))
 655                 return 0;
 656
 657         for (i = 0; i < dev->mem->nregions; i++) {
 658                 r = &dev->mem->regions[i];
 659
 660                 if (vva >= r->host_user_addr &&
 661                     vva + len <  r->host_user_addr + r->size) {
 662                         return r->guest_phys_addr + vva - r->host_user_addr;
 663                 }
 664         }
 665         return 0;
 666 }
 667
 668 static __rte_always_inline struct virtio_net *
 669 get_device(int vid)
 670 {
 671         struct virtio_net *dev = vhost_devices[vid];
 672
 673         if (unlikely(!dev)) {
 674                 VHOST_LOG_CONFIG(ERR,
 675                         "(%d) device not found.\n", vid);
 676         }
 677
 678         return dev;
 679 }
 680
 681 int vhost_new_device(void);
 682 void cleanup_device(struct virtio_net *dev, int destroy);
 683 void reset_device(struct virtio_net *dev);
 684 void vhost_destroy_device(int);
 685 void vhost_destroy_device_notify(struct virtio_net *dev);
 686
 687 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
 688 void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
 689 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
 690
 691 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
 692
 693 void vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *dev);
 694
 695 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
 696 void vhost_set_builtin_virtio_net(int vid, bool enable);
 697 void vhost_enable_extbuf(int vid);
 698 void vhost_enable_linearbuf(int vid);
 699 int vhost_enable_guest_notification(struct virtio_net *dev,
 700                 struct vhost_virtqueue *vq, int enable);
 701
 702 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
 703
 704 /*
 705  * Backend-specific cleanup.
 706  *
 707  * TODO: fix it; we have one backend now
 708  */
 709 void vhost_backend_cleanup(struct virtio_net *dev);
 710
 711 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 712                         uint64_t iova, uint64_t *len, uint8_t perm);
 713 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
 714                         struct vhost_virtqueue *vq,
 715                         uint64_t desc_addr, uint64_t desc_len);
 716 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 717 uint64_t translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
 718                 uint64_t log_addr);
 719 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 720
 721 static __rte_always_inline uint64_t
 722 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 723                         uint64_t iova, uint64_t *len, uint8_t perm)
 724 {
 725         if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 726                 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
 727
 728         return __vhost_iova_to_vva(dev, vq, iova, len, perm);
 729 }
 730
 731 #define vhost_avail_event(vr) \
 732         (*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
 733 #define vhost_used_event(vr) \
 734         (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
 735
 736 /*
 737  * The following is used with VIRTIO_RING_F_EVENT_IDX.
 738  * Assuming a given event_idx value from the other size, if we have
 739  * just incremented index from old to new_idx, should we trigger an
 740  * event?
 741  */
 742 static __rte_always_inline int
 743 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
 744 {
 745         return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
 746 }
 747
 748 static __rte_always_inline void
 749 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 750 {
 751         /* Flush used->idx update before we read avail->flags. */
 752         rte_smp_mb();
 753
 754         /* Don't kick guest if we don't reach index specified by guest. */
 755         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
 756                 uint16_t old = vq->signalled_used;
 757                 uint16_t new = vq->async_pkts_inflight_n ?
 758                                         vq->used->idx:vq->last_used_idx;
 759                 bool signalled_used_valid = vq->signalled_used_valid;
 760
 761                 vq->signalled_used = new;
 762                 vq->signalled_used_valid = true;
 763
 764                 VHOST_LOG_DATA(DEBUG, "%s: used_event_idx=%d, old=%d, new=%d\n",
 765                         __func__,
 766                         vhost_used_event(vq),
 767                         old, new);
 768
 769                 if ((vhost_need_event(vhost_used_event(vq), new, old) &&
 770                                         (vq->callfd >= 0)) ||
 771                                 unlikely(!signalled_used_valid)) {
 772                         eventfd_write(vq->callfd, (eventfd_t) 1);
 773                         if (dev->notify_ops->guest_notified)
 774                                 dev->notify_ops->guest_notified(dev->vid);
 775                 }
 776         } else {
 777                 /* Kick the guest if necessary. */
 778                 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
 779                                 && (vq->callfd >= 0)) {
 780                         eventfd_write(vq->callfd, (eventfd_t)1);
 781                         if (dev->notify_ops->guest_notified)
 782                                 dev->notify_ops->guest_notified(dev->vid);
 783                 }
 784         }
 785 }
 786
 787 static __rte_always_inline void
 788 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
 789 {
 790         uint16_t old, new, off, off_wrap;
 791         bool signalled_used_valid, kick = false;
 792
 793         /* Flush used desc update. */
 794         rte_smp_mb();
 795
 796         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
 797                 if (vq->driver_event->flags !=
 798                                 VRING_EVENT_F_DISABLE)
 799                         kick = true;
 800                 goto kick;
 801         }
 802
 803         old = vq->signalled_used;
 804         new = vq->last_used_idx;
 805         vq->signalled_used = new;
 806         signalled_used_valid = vq->signalled_used_valid;
 807         vq->signalled_used_valid = true;
 808
 809         if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
 810                 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
 811                         kick = true;
 812                 goto kick;
 813         }
 814
 815         if (unlikely(!signalled_used_valid)) {
 816                 kick = true;
 817                 goto kick;
 818         }
 819
 820         rte_smp_rmb();
 821
 822         off_wrap = vq->driver_event->off_wrap;
 823         off = off_wrap & ~(1 << 15);
 824
 825         if (new <= old)
 826                 old -= vq->size;
 827
 828         if (vq->used_wrap_counter != off_wrap >> 15)
 829                 off -= vq->size;
 830
 831         if (vhost_need_event(off, new, old))
 832                 kick = true;
 833 kick:
 834         if (kick) {
 835                 eventfd_write(vq->callfd, (eventfd_t)1);
 836                 if (dev->notify_ops->guest_notified)
 837                         dev->notify_ops->guest_notified(dev->vid);
 838         }
 839 }
 840
 841 static __rte_always_inline void
 842 free_ind_table(void *idesc)
 843 {
 844         rte_free(idesc);
 845 }
 846
 847 static __rte_always_inline void
 848 restore_mbuf(struct rte_mbuf *m)
 849 {
 850         uint32_t mbuf_size, priv_size;
 851
 852         while (m) {
 853                 priv_size = rte_pktmbuf_priv_size(m->pool);
 854                 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
 855                 /* start of buffer is after mbuf structure and priv data */
 856
 857                 m->buf_addr = (char *)m + mbuf_size;
 858                 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
 859                 m = m->next;
 860         }
 861 }
 862
 863 static __rte_always_inline bool
 864 mbuf_is_consumed(struct rte_mbuf *m)
 865 {
 866         while (m) {
 867                 if (rte_mbuf_refcnt_read(m) > 1)
 868                         return false;
 869                 m = m->next;
 870         }
 871
 872         return true;
 873 }
 874
 875 #endif /* _VHOST_NET_CDEV_H_ */