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