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