lib/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 ((uint32_t)1 << 0)
  31 /* Used to indicate that the device is ready to operate */
  32 #define VIRTIO_DEV_READY ((uint32_t)1 << 1)
  33 /* Used to indicate that the built-in vhost net device backend is enabled */
  34 #define VIRTIO_DEV_BUILTIN_VIRTIO_NET ((uint32_t)1 << 2)
  35 /* Used to indicate that the device has its own data path and configured */
  36 #define VIRTIO_DEV_VDPA_CONFIGURED ((uint32_t)1 << 3)
  37 /* Used to indicate that the feature negotiation failed */
  38 #define VIRTIO_DEV_FEATURES_FAILED ((uint32_t)1 << 4)
  39 /* Used to indicate that the virtio_net tx code should fill TX ol_flags */
  40 #define VIRTIO_DEV_LEGACY_OL_FLAGS ((uint32_t)1 << 5)
  41
  42 /* Backend value set by guest. */
  43 #define VIRTIO_DEV_STOPPED -1
  44
  45 #define BUF_VECTOR_MAX 256
  46
  47 #define VHOST_LOG_CACHE_NR 32
  48
  49 #define MAX_PKT_BURST 32
  50
  51 #define VHOST_MAX_ASYNC_IT (MAX_PKT_BURST)
  52 #define VHOST_MAX_ASYNC_VEC (BUF_VECTOR_MAX * 2)
  53
  54 #define PACKED_DESC_ENQUEUE_USED_FLAG(w)        \
  55         ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED | VRING_DESC_F_WRITE) : \
  56                 VRING_DESC_F_WRITE)
  57 #define PACKED_DESC_DEQUEUE_USED_FLAG(w)        \
  58         ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED) : 0x0)
  59 #define PACKED_DESC_SINGLE_DEQUEUE_FLAG (VRING_DESC_F_NEXT | \
  60                                          VRING_DESC_F_INDIRECT)
  61
  62 #define PACKED_BATCH_SIZE (RTE_CACHE_LINE_SIZE / \
  63                             sizeof(struct vring_packed_desc))
  64 #define PACKED_BATCH_MASK (PACKED_BATCH_SIZE - 1)
  65
  66 #ifdef VHOST_GCC_UNROLL_PRAGMA
  67 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("GCC unroll 4") \
  68         for (iter = val; iter < size; iter++)
  69 #endif
  70
  71 #ifdef VHOST_CLANG_UNROLL_PRAGMA
  72 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll 4") \
  73         for (iter = val; iter < size; iter++)
  74 #endif
  75
  76 #ifdef VHOST_ICC_UNROLL_PRAGMA
  77 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll (4)") \
  78         for (iter = val; iter < size; iter++)
  79 #endif
  80
  81 #ifndef vhost_for_each_try_unroll
  82 #define vhost_for_each_try_unroll(iter, val, num) \
  83         for (iter = val; iter < num; iter++)
  84 #endif
  85
  86 /**
  87  * Structure contains buffer address, length and descriptor index
  88  * from vring to do scatter RX.
  89  */
  90 struct buf_vector {
  91         uint64_t buf_iova;
  92         uint64_t buf_addr;
  93         uint32_t buf_len;
  94         uint32_t desc_idx;
  95 };
  96
  97 /*
  98  * Structure contains the info for each batched memory copy.
  99  */
 100 struct batch_copy_elem {
 101         void *dst;
 102         void *src;
 103         uint32_t len;
 104         uint64_t log_addr;
 105 };
 106
 107 /*
 108  * Structure that contains the info for batched dirty logging.
 109  */
 110 struct log_cache_entry {
 111         uint32_t offset;
 112         unsigned long val;
 113 };
 114
 115 struct vring_used_elem_packed {
 116         uint16_t id;
 117         uint16_t flags;
 118         uint32_t len;
 119         uint32_t count;
 120 };
 121
 122 /**
 123  * inflight async packet information
 124  */
 125 struct async_inflight_info {
 126         struct rte_mbuf *mbuf;
 127         uint16_t descs; /* num of descs inflight */
 128         uint16_t nr_buffers; /* num of buffers inflight for packed ring */
 129 };
 130
 131 struct vhost_async {
 132         /* operation callbacks for DMA */
 133         struct rte_vhost_async_channel_ops ops;
 134
 135         struct rte_vhost_iov_iter iov_iter[VHOST_MAX_ASYNC_IT];
 136         struct rte_vhost_iovec iovec[VHOST_MAX_ASYNC_VEC];
 137
 138         /* data transfer status */
 139         struct async_inflight_info *pkts_info;
 140         uint16_t pkts_idx;
 141         uint16_t pkts_inflight_n;
 142         uint16_t last_pkts_n;
 143         union {
 144                 struct vring_used_elem  *descs_split;
 145                 struct vring_used_elem_packed *buffers_packed;
 146         };
 147         union {
 148                 uint16_t desc_idx_split;
 149                 uint16_t buffer_idx_packed;
 150         };
 151         union {
 152                 uint16_t last_desc_idx_split;
 153                 uint16_t last_buffer_idx_packed;
 154         };
 155 };
 156
 157 /**
 158  * Structure contains variables relevant to RX/TX virtqueues.
 159  */
 160 struct vhost_virtqueue {
 161         union {
 162                 struct vring_desc       *desc;
 163                 struct vring_packed_desc   *desc_packed;
 164         };
 165         union {
 166                 struct vring_avail      *avail;
 167                 struct vring_packed_desc_event *driver_event;
 168         };
 169         union {
 170                 struct vring_used       *used;
 171                 struct vring_packed_desc_event *device_event;
 172         };
 173         uint16_t                size;
 174
 175         uint16_t                last_avail_idx;
 176         uint16_t                last_used_idx;
 177         /* Last used index we notify to front end. */
 178         uint16_t                signalled_used;
 179         bool                    signalled_used_valid;
 180 #define VIRTIO_INVALID_EVENTFD          (-1)
 181 #define VIRTIO_UNINITIALIZED_EVENTFD    (-2)
 182
 183         bool                    enabled;
 184         bool                    access_ok;
 185         bool                    ready;
 186
 187         rte_spinlock_t          access_lock;
 188
 189
 190         union {
 191                 struct vring_used_elem  *shadow_used_split;
 192                 struct vring_used_elem_packed *shadow_used_packed;
 193         };
 194         uint16_t                shadow_used_idx;
 195         /* Record packed ring enqueue latest desc cache aligned index */
 196         uint16_t                shadow_aligned_idx;
 197         /* Record packed ring first dequeue desc index */
 198         uint16_t                shadow_last_used_idx;
 199
 200         uint16_t                batch_copy_nb_elems;
 201         struct batch_copy_elem  *batch_copy_elems;
 202         int                     numa_node;
 203         bool                    used_wrap_counter;
 204         bool                    avail_wrap_counter;
 205
 206         /* Physical address of used ring, for logging */
 207         uint16_t                log_cache_nb_elem;
 208         uint64_t                log_guest_addr;
 209         struct log_cache_entry  *log_cache;
 210
 211         rte_rwlock_t    iotlb_lock;
 212         rte_rwlock_t    iotlb_pending_lock;
 213         struct rte_mempool *iotlb_pool;
 214         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
 215         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
 216         int                             iotlb_cache_nr;
 217
 218         /* Used to notify the guest (trigger interrupt) */
 219         int                     callfd;
 220         /* Currently unused as polling mode is enabled */
 221         int                     kickfd;
 222
 223         /* inflight share memory info */
 224         union {
 225                 struct rte_vhost_inflight_info_split *inflight_split;
 226                 struct rte_vhost_inflight_info_packed *inflight_packed;
 227         };
 228         struct rte_vhost_resubmit_info *resubmit_inflight;
 229         uint64_t                global_counter;
 230
 231         struct vhost_async      *async;
 232
 233         int                     notif_enable;
 234 #define VIRTIO_UNINITIALIZED_NOTIF      (-1)
 235
 236         struct vhost_vring_addr ring_addrs;
 237 } __rte_cache_aligned;
 238
 239 /* Virtio device status as per Virtio specification */
 240 #define VIRTIO_DEVICE_STATUS_RESET              0x00
 241 #define VIRTIO_DEVICE_STATUS_ACK                0x01
 242 #define VIRTIO_DEVICE_STATUS_DRIVER             0x02
 243 #define VIRTIO_DEVICE_STATUS_DRIVER_OK          0x04
 244 #define VIRTIO_DEVICE_STATUS_FEATURES_OK        0x08
 245 #define VIRTIO_DEVICE_STATUS_DEV_NEED_RESET     0x40
 246 #define VIRTIO_DEVICE_STATUS_FAILED             0x80
 247
 248 #define VHOST_MAX_VRING                 0x100
 249 #define VHOST_MAX_QUEUE_PAIRS           0x80
 250
 251 /* Declare IOMMU related bits for older kernels */
 252 #ifndef VIRTIO_F_IOMMU_PLATFORM
 253
 254 #define VIRTIO_F_IOMMU_PLATFORM 33
 255
 256 struct vhost_iotlb_msg {
 257         __u64 iova;
 258         __u64 size;
 259         __u64 uaddr;
 260 #define VHOST_ACCESS_RO      0x1
 261 #define VHOST_ACCESS_WO      0x2
 262 #define VHOST_ACCESS_RW      0x3
 263         __u8 perm;
 264 #define VHOST_IOTLB_MISS           1
 265 #define VHOST_IOTLB_UPDATE         2
 266 #define VHOST_IOTLB_INVALIDATE     3
 267 #define VHOST_IOTLB_ACCESS_FAIL    4
 268         __u8 type;
 269 };
 270
 271 #define VHOST_IOTLB_MSG 0x1
 272
 273 struct vhost_msg {
 274         int type;
 275         union {
 276                 struct vhost_iotlb_msg iotlb;
 277                 __u8 padding[64];
 278         };
 279 };
 280 #endif
 281
 282 /*
 283  * Define virtio 1.0 for older kernels
 284  */
 285 #ifndef VIRTIO_F_VERSION_1
 286  #define VIRTIO_F_VERSION_1 32
 287 #endif
 288
 289 /* Declare packed ring related bits for older kernels */
 290 #ifndef VIRTIO_F_RING_PACKED
 291
 292 #define VIRTIO_F_RING_PACKED 34
 293
 294 struct vring_packed_desc {
 295         uint64_t addr;
 296         uint32_t len;
 297         uint16_t id;
 298         uint16_t flags;
 299 };
 300
 301 struct vring_packed_desc_event {
 302         uint16_t off_wrap;
 303         uint16_t flags;
 304 };
 305 #endif
 306
 307 /*
 308  * Declare below packed ring defines unconditionally
 309  * as Kernel header might use different names.
 310  */
 311 #define VRING_DESC_F_AVAIL      (1ULL << 7)
 312 #define VRING_DESC_F_USED       (1ULL << 15)
 313
 314 #define VRING_EVENT_F_ENABLE 0x0
 315 #define VRING_EVENT_F_DISABLE 0x1
 316 #define VRING_EVENT_F_DESC 0x2
 317
 318 /*
 319  * Available and used descs are in same order
 320  */
 321 #ifndef VIRTIO_F_IN_ORDER
 322 #define VIRTIO_F_IN_ORDER      35
 323 #endif
 324
 325 /* Features supported by this builtin vhost-user net driver. */
 326 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
 327                                 (1ULL << VIRTIO_F_ANY_LAYOUT) | \
 328                                 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
 329                                 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
 330                                 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
 331                                 (1ULL << VIRTIO_NET_F_MQ)      | \
 332                                 (1ULL << VIRTIO_F_VERSION_1)   | \
 333                                 (1ULL << VHOST_F_LOG_ALL)      | \
 334                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
 335                                 (1ULL << VIRTIO_NET_F_GSO) | \
 336                                 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
 337                                 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
 338                                 (1ULL << VIRTIO_NET_F_HOST_UFO) | \
 339                                 (1ULL << VIRTIO_NET_F_HOST_ECN) | \
 340                                 (1ULL << VIRTIO_NET_F_CSUM)    | \
 341                                 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
 342                                 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
 343                                 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
 344                                 (1ULL << VIRTIO_NET_F_GUEST_UFO) | \
 345                                 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
 346                                 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
 347                                 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \
 348                                 (1ULL << VIRTIO_NET_F_MTU)  | \
 349                                 (1ULL << VIRTIO_F_IN_ORDER) | \
 350                                 (1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
 351                                 (1ULL << VIRTIO_F_RING_PACKED))
 352
 353
 354 struct guest_page {
 355         uint64_t guest_phys_addr;
 356         uint64_t host_phys_addr;
 357         uint64_t size;
 358 };
 359
 360 struct inflight_mem_info {
 361         int             fd;
 362         void            *addr;
 363         uint64_t        size;
 364 };
 365
 366 /**
 367  * Device structure contains all configuration information relating
 368  * to the device.
 369  */
 370 struct virtio_net {
 371         /* Frontend (QEMU) memory and memory region information */
 372         struct rte_vhost_memory *mem;
 373         uint64_t                features;
 374         uint64_t                protocol_features;
 375         int                     vid;
 376         uint32_t                flags;
 377         uint16_t                vhost_hlen;
 378         /* to tell if we need broadcast rarp packet */
 379         int16_t                 broadcast_rarp;
 380         uint32_t                nr_vring;
 381         int                     async_copy;
 382
 383         int                     extbuf;
 384         int                     linearbuf;
 385         struct vhost_virtqueue  *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
 386         struct inflight_mem_info *inflight_info;
 387 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
 388         char                    ifname[IF_NAME_SZ];
 389         uint64_t                log_size;
 390         uint64_t                log_base;
 391         uint64_t                log_addr;
 392         struct rte_ether_addr   mac;
 393         uint16_t                mtu;
 394         uint8_t                 status;
 395
 396         struct vhost_device_ops const *notify_ops;
 397
 398         uint32_t                nr_guest_pages;
 399         uint32_t                max_guest_pages;
 400         struct guest_page       *guest_pages;
 401
 402         int                     slave_req_fd;
 403         rte_spinlock_t          slave_req_lock;
 404
 405         int                     postcopy_ufd;
 406         int                     postcopy_listening;
 407
 408         struct rte_vdpa_device *vdpa_dev;
 409
 410         /* context data for the external message handlers */
 411         void                    *extern_data;
 412         /* pre and post vhost user message handlers for the device */
 413         struct rte_vhost_user_extern_ops extern_ops;
 414 } __rte_cache_aligned;
 415
 416 static __rte_always_inline bool
 417 vq_is_packed(struct virtio_net *dev)
 418 {
 419         return dev->features & (1ull << VIRTIO_F_RING_PACKED);
 420 }
 421
 422 static inline bool
 423 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
 424 {
 425         uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
 426
 427         return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
 428                 wrap_counter != !!(flags & VRING_DESC_F_USED);
 429 }
 430
 431 static inline void
 432 vq_inc_last_used_packed(struct vhost_virtqueue *vq, uint16_t num)
 433 {
 434         vq->last_used_idx += num;
 435         if (vq->last_used_idx >= vq->size) {
 436                 vq->used_wrap_counter ^= 1;
 437                 vq->last_used_idx -= vq->size;
 438         }
 439 }
 440
 441 static inline void
 442 vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
 443 {
 444         vq->last_avail_idx += num;
 445         if (vq->last_avail_idx >= vq->size) {
 446                 vq->avail_wrap_counter ^= 1;
 447                 vq->last_avail_idx -= vq->size;
 448         }
 449 }
 450
 451 void __vhost_log_cache_write(struct virtio_net *dev,
 452                 struct vhost_virtqueue *vq,
 453                 uint64_t addr, uint64_t len);
 454 void __vhost_log_cache_write_iova(struct virtio_net *dev,
 455                 struct vhost_virtqueue *vq,
 456                 uint64_t iova, uint64_t len);
 457 void __vhost_log_cache_sync(struct virtio_net *dev,
 458                 struct vhost_virtqueue *vq);
 459 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
 460 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 461                             uint64_t iova, uint64_t len);
 462
 463 static __rte_always_inline void
 464 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
 465 {
 466         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 467                 __vhost_log_write(dev, addr, len);
 468 }
 469
 470 static __rte_always_inline void
 471 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
 472 {
 473         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 474                 __vhost_log_cache_sync(dev, vq);
 475 }
 476
 477 static __rte_always_inline void
 478 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
 479                         uint64_t addr, uint64_t len)
 480 {
 481         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 482                 __vhost_log_cache_write(dev, vq, addr, len);
 483 }
 484
 485 static __rte_always_inline void
 486 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 487                         uint64_t offset, uint64_t len)
 488 {
 489         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
 490                 if (unlikely(vq->log_guest_addr == 0))
 491                         return;
 492                 __vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset,
 493                                         len);
 494         }
 495 }
 496
 497 static __rte_always_inline void
 498 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 499                      uint64_t offset, uint64_t len)
 500 {
 501         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
 502                 if (unlikely(vq->log_guest_addr == 0))
 503                         return;
 504                 __vhost_log_write(dev, vq->log_guest_addr + offset, len);
 505         }
 506 }
 507
 508 static __rte_always_inline void
 509 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 510                            uint64_t iova, uint64_t len)
 511 {
 512         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 513                 return;
 514
 515         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 516                 __vhost_log_cache_write_iova(dev, vq, iova, len);
 517         else
 518                 __vhost_log_cache_write(dev, vq, iova, len);
 519 }
 520
 521 static __rte_always_inline void
 522 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 523                            uint64_t iova, uint64_t len)
 524 {
 525         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 526                 return;
 527
 528         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 529                 __vhost_log_write_iova(dev, vq, iova, len);
 530         else
 531                 __vhost_log_write(dev, iova, len);
 532 }
 533
 534 extern int vhost_config_log_level;
 535 extern int vhost_data_log_level;
 536
 537 #define VHOST_LOG_CONFIG(level, fmt, args...)                   \
 538         rte_log(RTE_LOG_ ## level, vhost_config_log_level,      \
 539                 "VHOST_CONFIG: " fmt, ##args)
 540
 541 #define VHOST_LOG_DATA(level, fmt, args...) \
 542         (void)((RTE_LOG_ ## level <= RTE_LOG_DP_LEVEL) ?        \
 543          rte_log(RTE_LOG_ ## level,  vhost_data_log_level,      \
 544                 "VHOST_DATA : " fmt, ##args) :                  \
 545          0)
 546
 547 #ifdef RTE_LIBRTE_VHOST_DEBUG
 548 #define VHOST_MAX_PRINT_BUFF 6072
 549 #define PRINT_PACKET(device, addr, size, header) do { \
 550         char *pkt_addr = (char *)(addr); \
 551         unsigned int index; \
 552         char packet[VHOST_MAX_PRINT_BUFF]; \
 553         \
 554         if ((header)) \
 555                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
 556         else \
 557                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
 558         for (index = 0; index < (size); index++) { \
 559                 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
 560                         "%02hhx ", pkt_addr[index]); \
 561         } \
 562         snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
 563         \
 564         VHOST_LOG_DATA(DEBUG, "%s", packet); \
 565 } while (0)
 566 #else
 567 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
 568 #endif
 569
 570 #define MAX_VHOST_DEVICE        1024
 571 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
 572
 573 #define VHOST_BINARY_SEARCH_THRESH 256
 574
 575 static __rte_always_inline int guest_page_addrcmp(const void *p1,
 576                                                 const void *p2)
 577 {
 578         const struct guest_page *page1 = (const struct guest_page *)p1;
 579         const struct guest_page *page2 = (const struct guest_page *)p2;
 580
 581         if (page1->guest_phys_addr > page2->guest_phys_addr)
 582                 return 1;
 583         if (page1->guest_phys_addr < page2->guest_phys_addr)
 584                 return -1;
 585
 586         return 0;
 587 }
 588
 589 static __rte_always_inline rte_iova_t
 590 gpa_to_first_hpa(struct virtio_net *dev, uint64_t gpa,
 591         uint64_t gpa_size, uint64_t *hpa_size)
 592 {
 593         uint32_t i;
 594         struct guest_page *page;
 595         struct guest_page key;
 596
 597         *hpa_size = gpa_size;
 598         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
 599                 key.guest_phys_addr = gpa & ~(dev->guest_pages[0].size - 1);
 600                 page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages,
 601                                sizeof(struct guest_page), guest_page_addrcmp);
 602                 if (page) {
 603                         if (gpa + gpa_size <=
 604                                         page->guest_phys_addr + page->size) {
 605                                 return gpa - page->guest_phys_addr +
 606                                         page->host_phys_addr;
 607                         } else if (gpa < page->guest_phys_addr +
 608                                                 page->size) {
 609                                 *hpa_size = page->guest_phys_addr +
 610                                         page->size - gpa;
 611                                 return gpa - page->guest_phys_addr +
 612                                         page->host_phys_addr;
 613                         }
 614                 }
 615         } else {
 616                 for (i = 0; i < dev->nr_guest_pages; i++) {
 617                         page = &dev->guest_pages[i];
 618
 619                         if (gpa >= page->guest_phys_addr) {
 620                                 if (gpa + gpa_size <=
 621                                         page->guest_phys_addr + page->size) {
 622                                         return gpa - page->guest_phys_addr +
 623                                                 page->host_phys_addr;
 624                                 } else if (gpa < page->guest_phys_addr +
 625                                                         page->size) {
 626                                         *hpa_size = page->guest_phys_addr +
 627                                                 page->size - gpa;
 628                                         return gpa - page->guest_phys_addr +
 629                                                 page->host_phys_addr;
 630                                 }
 631                         }
 632                 }
 633         }
 634
 635         *hpa_size = 0;
 636         return 0;
 637 }
 638
 639 /* Convert guest physical address to host physical address */
 640 static __rte_always_inline rte_iova_t
 641 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
 642 {
 643         rte_iova_t hpa;
 644         uint64_t hpa_size;
 645
 646         hpa = gpa_to_first_hpa(dev, gpa, size, &hpa_size);
 647         return hpa_size == size ? hpa : 0;
 648 }
 649
 650 static __rte_always_inline uint64_t
 651 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
 652 {
 653         struct rte_vhost_mem_region *r;
 654         uint32_t i;
 655
 656         if (unlikely(!dev || !dev->mem))
 657                 return 0;
 658
 659         for (i = 0; i < dev->mem->nregions; i++) {
 660                 r = &dev->mem->regions[i];
 661
 662                 if (vva >= r->host_user_addr &&
 663                     vva + len <  r->host_user_addr + r->size) {
 664                         return r->guest_phys_addr + vva - r->host_user_addr;
 665                 }
 666         }
 667         return 0;
 668 }
 669
 670 static __rte_always_inline struct virtio_net *
 671 get_device(int vid)
 672 {
 673         struct virtio_net *dev = vhost_devices[vid];
 674
 675         if (unlikely(!dev)) {
 676                 VHOST_LOG_CONFIG(ERR,
 677                         "(%d) device not found.\n", vid);
 678         }
 679
 680         return dev;
 681 }
 682
 683 int vhost_new_device(void);
 684 void cleanup_device(struct virtio_net *dev, int destroy);
 685 void reset_device(struct virtio_net *dev);
 686 void vhost_destroy_device(int);
 687 void vhost_destroy_device_notify(struct virtio_net *dev);
 688
 689 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
 690 void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
 691 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
 692
 693 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
 694
 695 void vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *dev);
 696
 697 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
 698 void vhost_setup_virtio_net(int vid, bool enable, bool legacy_ol_flags);
 699 void vhost_enable_extbuf(int vid);
 700 void vhost_enable_linearbuf(int vid);
 701 int vhost_enable_guest_notification(struct virtio_net *dev,
 702                 struct vhost_virtqueue *vq, int enable);
 703
 704 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
 705
 706 /*
 707  * Backend-specific cleanup.
 708  *
 709  * TODO: fix it; we have one backend now
 710  */
 711 void vhost_backend_cleanup(struct virtio_net *dev);
 712
 713 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 714                         uint64_t iova, uint64_t *len, uint8_t perm);
 715 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
 716                         struct vhost_virtqueue *vq,
 717                         uint64_t desc_addr, uint64_t desc_len);
 718 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 719 uint64_t translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
 720                 uint64_t log_addr);
 721 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 722
 723 static __rte_always_inline uint64_t
 724 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 725                         uint64_t iova, uint64_t *len, uint8_t perm)
 726 {
 727         if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 728                 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
 729
 730         return __vhost_iova_to_vva(dev, vq, iova, len, perm);
 731 }
 732
 733 #define vhost_avail_event(vr) \
 734         (*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
 735 #define vhost_used_event(vr) \
 736         (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
 737
 738 /*
 739  * The following is used with VIRTIO_RING_F_EVENT_IDX.
 740  * Assuming a given event_idx value from the other size, if we have
 741  * just incremented index from old to new_idx, should we trigger an
 742  * event?
 743  */
 744 static __rte_always_inline int
 745 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
 746 {
 747         return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
 748 }
 749
 750 static __rte_always_inline void
 751 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 752 {
 753         /* Flush used->idx update before we read avail->flags. */
 754         rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
 755
 756         /* Don't kick guest if we don't reach index specified by guest. */
 757         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
 758                 uint16_t old = vq->signalled_used;
 759                 uint16_t new = vq->last_used_idx;
 760                 bool signalled_used_valid = vq->signalled_used_valid;
 761
 762                 vq->signalled_used = new;
 763                 vq->signalled_used_valid = true;
 764
 765                 VHOST_LOG_DATA(DEBUG, "%s: used_event_idx=%d, old=%d, new=%d\n",
 766                         __func__,
 767                         vhost_used_event(vq),
 768                         old, new);
 769
 770                 if ((vhost_need_event(vhost_used_event(vq), new, old) &&
 771                                         (vq->callfd >= 0)) ||
 772                                 unlikely(!signalled_used_valid)) {
 773                         eventfd_write(vq->callfd, (eventfd_t) 1);
 774                         if (dev->notify_ops->guest_notified)
 775                                 dev->notify_ops->guest_notified(dev->vid);
 776                 }
 777         } else {
 778                 /* Kick the guest if necessary. */
 779                 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
 780                                 && (vq->callfd >= 0)) {
 781                         eventfd_write(vq->callfd, (eventfd_t)1);
 782                         if (dev->notify_ops->guest_notified)
 783                                 dev->notify_ops->guest_notified(dev->vid);
 784                 }
 785         }
 786 }
 787
 788 static __rte_always_inline void
 789 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
 790 {
 791         uint16_t old, new, off, off_wrap;
 792         bool signalled_used_valid, kick = false;
 793
 794         /* Flush used desc update. */
 795         rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
 796
 797         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
 798                 if (vq->driver_event->flags !=
 799                                 VRING_EVENT_F_DISABLE)
 800                         kick = true;
 801                 goto kick;
 802         }
 803
 804         old = vq->signalled_used;
 805         new = vq->last_used_idx;
 806         vq->signalled_used = new;
 807         signalled_used_valid = vq->signalled_used_valid;
 808         vq->signalled_used_valid = true;
 809
 810         if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
 811                 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
 812                         kick = true;
 813                 goto kick;
 814         }
 815
 816         if (unlikely(!signalled_used_valid)) {
 817                 kick = true;
 818                 goto kick;
 819         }
 820
 821         rte_atomic_thread_fence(__ATOMIC_ACQUIRE);
 822
 823         off_wrap = vq->driver_event->off_wrap;
 824         off = off_wrap & ~(1 << 15);
 825
 826         if (new <= old)
 827                 old -= vq->size;
 828
 829         if (vq->used_wrap_counter != off_wrap >> 15)
 830                 off -= vq->size;
 831
 832         if (vhost_need_event(off, new, old))
 833                 kick = true;
 834 kick:
 835         if (kick) {
 836                 eventfd_write(vq->callfd, (eventfd_t)1);
 837                 if (dev->notify_ops->guest_notified)
 838                         dev->notify_ops->guest_notified(dev->vid);
 839         }
 840 }
 841
 842 static __rte_always_inline void
 843 free_ind_table(void *idesc)
 844 {
 845         rte_free(idesc);
 846 }
 847
 848 static __rte_always_inline void
 849 restore_mbuf(struct rte_mbuf *m)
 850 {
 851         uint32_t mbuf_size, priv_size;
 852
 853         while (m) {
 854                 priv_size = rte_pktmbuf_priv_size(m->pool);
 855                 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
 856                 /* start of buffer is after mbuf structure and priv data */
 857
 858                 m->buf_addr = (char *)m + mbuf_size;
 859                 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
 860                 m = m->next;
 861         }
 862 }
 863
 864 static __rte_always_inline bool
 865 mbuf_is_consumed(struct rte_mbuf *m)
 866 {
 867         while (m) {
 868                 if (rte_mbuf_refcnt_read(m) > 1)
 869                         return false;
 870                 m = m->next;
 871         }
 872
 873         return true;
 874 }
 875
 876 #endif /* _VHOST_NET_CDEV_H_ */