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 ASYNC_MAX_POLL_SEG 255
  50
  51 #define VHOST_MAX_ASYNC_IT (MAX_PKT_BURST * 2)
  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  * 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         uint32_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         /* Backend value to determine if device should started/stopped */
 149         int                     backend;
 150         int                     enabled;
 151         int                     access_ok;
 152         int                     ready;
 153         int                     notif_enable;
 154 #define VIRTIO_UNINITIALIZED_NOTIF      (-1)
 155
 156         rte_spinlock_t          access_lock;
 157
 158         /* Used to notify the guest (trigger interrupt) */
 159         int                     callfd;
 160         /* Currently unused as polling mode is enabled */
 161         int                     kickfd;
 162
 163         /* Physical address of used ring, for logging */
 164         uint64_t                log_guest_addr;
 165
 166         /* inflight share memory info */
 167         union {
 168                 struct rte_vhost_inflight_info_split *inflight_split;
 169                 struct rte_vhost_inflight_info_packed *inflight_packed;
 170         };
 171         struct rte_vhost_resubmit_info *resubmit_inflight;
 172         uint64_t                global_counter;
 173
 174         union {
 175                 struct vring_used_elem  *shadow_used_split;
 176                 struct vring_used_elem_packed *shadow_used_packed;
 177         };
 178         uint16_t                shadow_used_idx;
 179         /* Record packed ring enqueue latest desc cache aligned index */
 180         uint16_t                shadow_aligned_idx;
 181         /* Record packed ring first dequeue desc index */
 182         uint16_t                shadow_last_used_idx;
 183         struct vhost_vring_addr ring_addrs;
 184
 185         struct batch_copy_elem  *batch_copy_elems;
 186         uint16_t                batch_copy_nb_elems;
 187         bool                    used_wrap_counter;
 188         bool                    avail_wrap_counter;
 189
 190         struct log_cache_entry log_cache[VHOST_LOG_CACHE_NR];
 191         uint16_t log_cache_nb_elem;
 192
 193         rte_rwlock_t    iotlb_lock;
 194         rte_rwlock_t    iotlb_pending_lock;
 195         struct rte_mempool *iotlb_pool;
 196         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
 197         int                             iotlb_cache_nr;
 198         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
 199
 200         /* operation callbacks for async dma */
 201         struct rte_vhost_async_channel_ops      async_ops;
 202
 203         struct rte_vhost_iov_iter it_pool[VHOST_MAX_ASYNC_IT];
 204         struct iovec vec_pool[VHOST_MAX_ASYNC_VEC];
 205
 206         /* async data transfer status */
 207         uintptr_t       **async_pkts_pending;
 208         #define         ASYNC_PENDING_INFO_N_MSK 0xFFFF
 209         #define         ASYNC_PENDING_INFO_N_SFT 16
 210         uint64_t        *async_pending_info;
 211         uint16_t        async_pkts_idx;
 212         uint16_t        async_pkts_inflight_n;
 213         uint16_t        async_last_seg_n;
 214
 215         /* vq async features */
 216         bool            async_inorder;
 217         bool            async_registered;
 218         uint16_t        async_threshold;
 219 } __rte_cache_aligned;
 220
 221 /* Virtio device status as per Virtio specification */
 222 #define VIRTIO_DEVICE_STATUS_RESET              0x00
 223 #define VIRTIO_DEVICE_STATUS_ACK                0x01
 224 #define VIRTIO_DEVICE_STATUS_DRIVER             0x02
 225 #define VIRTIO_DEVICE_STATUS_DRIVER_OK          0x04
 226 #define VIRTIO_DEVICE_STATUS_FEATURES_OK        0x08
 227 #define VIRTIO_DEVICE_STATUS_DEV_NEED_RESET     0x40
 228 #define VIRTIO_DEVICE_STATUS_FAILED             0x80
 229
 230 #define VHOST_MAX_VRING                 0x100
 231 #define VHOST_MAX_QUEUE_PAIRS           0x80
 232
 233 /* Declare IOMMU related bits for older kernels */
 234 #ifndef VIRTIO_F_IOMMU_PLATFORM
 235
 236 #define VIRTIO_F_IOMMU_PLATFORM 33
 237
 238 struct vhost_iotlb_msg {
 239         __u64 iova;
 240         __u64 size;
 241         __u64 uaddr;
 242 #define VHOST_ACCESS_RO      0x1
 243 #define VHOST_ACCESS_WO      0x2
 244 #define VHOST_ACCESS_RW      0x3
 245         __u8 perm;
 246 #define VHOST_IOTLB_MISS           1
 247 #define VHOST_IOTLB_UPDATE         2
 248 #define VHOST_IOTLB_INVALIDATE     3
 249 #define VHOST_IOTLB_ACCESS_FAIL    4
 250         __u8 type;
 251 };
 252
 253 #define VHOST_IOTLB_MSG 0x1
 254
 255 struct vhost_msg {
 256         int type;
 257         union {
 258                 struct vhost_iotlb_msg iotlb;
 259                 __u8 padding[64];
 260         };
 261 };
 262 #endif
 263
 264 /*
 265  * Define virtio 1.0 for older kernels
 266  */
 267 #ifndef VIRTIO_F_VERSION_1
 268  #define VIRTIO_F_VERSION_1 32
 269 #endif
 270
 271 /* Declare packed ring related bits for older kernels */
 272 #ifndef VIRTIO_F_RING_PACKED
 273
 274 #define VIRTIO_F_RING_PACKED 34
 275
 276 struct vring_packed_desc {
 277         uint64_t addr;
 278         uint32_t len;
 279         uint16_t id;
 280         uint16_t flags;
 281 };
 282
 283 struct vring_packed_desc_event {
 284         uint16_t off_wrap;
 285         uint16_t flags;
 286 };
 287 #endif
 288
 289 /*
 290  * Declare below packed ring defines unconditionally
 291  * as Kernel header might use different names.
 292  */
 293 #define VRING_DESC_F_AVAIL      (1ULL << 7)
 294 #define VRING_DESC_F_USED       (1ULL << 15)
 295
 296 #define VRING_EVENT_F_ENABLE 0x0
 297 #define VRING_EVENT_F_DISABLE 0x1
 298 #define VRING_EVENT_F_DESC 0x2
 299
 300 /*
 301  * Available and used descs are in same order
 302  */
 303 #ifndef VIRTIO_F_IN_ORDER
 304 #define VIRTIO_F_IN_ORDER      35
 305 #endif
 306
 307 /* Features supported by this builtin vhost-user net driver. */
 308 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
 309                                 (1ULL << VIRTIO_F_ANY_LAYOUT) | \
 310                                 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
 311                                 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
 312                                 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
 313                                 (1ULL << VIRTIO_NET_F_MQ)      | \
 314                                 (1ULL << VIRTIO_F_VERSION_1)   | \
 315                                 (1ULL << VHOST_F_LOG_ALL)      | \
 316                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
 317                                 (1ULL << VIRTIO_NET_F_GSO) | \
 318                                 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
 319                                 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
 320                                 (1ULL << VIRTIO_NET_F_HOST_UFO) | \
 321                                 (1ULL << VIRTIO_NET_F_HOST_ECN) | \
 322                                 (1ULL << VIRTIO_NET_F_CSUM)    | \
 323                                 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
 324                                 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
 325                                 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
 326                                 (1ULL << VIRTIO_NET_F_GUEST_UFO) | \
 327                                 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
 328                                 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
 329                                 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \
 330                                 (1ULL << VIRTIO_NET_F_MTU)  | \
 331                                 (1ULL << VIRTIO_F_IN_ORDER) | \
 332                                 (1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
 333                                 (1ULL << VIRTIO_F_RING_PACKED))
 334
 335
 336 struct guest_page {
 337         uint64_t guest_phys_addr;
 338         uint64_t host_phys_addr;
 339         uint64_t size;
 340 };
 341
 342 struct inflight_mem_info {
 343         int             fd;
 344         void            *addr;
 345         uint64_t        size;
 346 };
 347
 348 /**
 349  * Device structure contains all configuration information relating
 350  * to the device.
 351  */
 352 struct virtio_net {
 353         /* Frontend (QEMU) memory and memory region information */
 354         struct rte_vhost_memory *mem;
 355         uint64_t                features;
 356         uint64_t                protocol_features;
 357         int                     vid;
 358         uint32_t                flags;
 359         uint16_t                vhost_hlen;
 360         /* to tell if we need broadcast rarp packet */
 361         int16_t                 broadcast_rarp;
 362         uint32_t                nr_vring;
 363         int                     async_copy;
 364         int                     extbuf;
 365         int                     linearbuf;
 366         struct vhost_virtqueue  *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
 367         struct inflight_mem_info *inflight_info;
 368 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
 369         char                    ifname[IF_NAME_SZ];
 370         uint64_t                log_size;
 371         uint64_t                log_base;
 372         uint64_t                log_addr;
 373         struct rte_ether_addr   mac;
 374         uint16_t                mtu;
 375         uint8_t                 status;
 376
 377         struct vhost_device_ops const *notify_ops;
 378
 379         uint32_t                nr_guest_pages;
 380         uint32_t                max_guest_pages;
 381         struct guest_page       *guest_pages;
 382
 383         int                     slave_req_fd;
 384         rte_spinlock_t          slave_req_lock;
 385
 386         int                     postcopy_ufd;
 387         int                     postcopy_listening;
 388
 389         struct rte_vdpa_device *vdpa_dev;
 390
 391         /* context data for the external message handlers */
 392         void                    *extern_data;
 393         /* pre and post vhost user message handlers for the device */
 394         struct rte_vhost_user_extern_ops extern_ops;
 395 } __rte_cache_aligned;
 396
 397 static __rte_always_inline bool
 398 vq_is_packed(struct virtio_net *dev)
 399 {
 400         return dev->features & (1ull << VIRTIO_F_RING_PACKED);
 401 }
 402
 403 static inline bool
 404 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
 405 {
 406         uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
 407
 408         return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
 409                 wrap_counter != !!(flags & VRING_DESC_F_USED);
 410 }
 411
 412 static inline void
 413 vq_inc_last_used_packed(struct vhost_virtqueue *vq, uint16_t num)
 414 {
 415         vq->last_used_idx += num;
 416         if (vq->last_used_idx >= vq->size) {
 417                 vq->used_wrap_counter ^= 1;
 418                 vq->last_used_idx -= vq->size;
 419         }
 420 }
 421
 422 static inline void
 423 vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
 424 {
 425         vq->last_avail_idx += num;
 426         if (vq->last_avail_idx >= vq->size) {
 427                 vq->avail_wrap_counter ^= 1;
 428                 vq->last_avail_idx -= vq->size;
 429         }
 430 }
 431
 432 void __vhost_log_cache_write(struct virtio_net *dev,
 433                 struct vhost_virtqueue *vq,
 434                 uint64_t addr, uint64_t len);
 435 void __vhost_log_cache_write_iova(struct virtio_net *dev,
 436                 struct vhost_virtqueue *vq,
 437                 uint64_t iova, uint64_t len);
 438 void __vhost_log_cache_sync(struct virtio_net *dev,
 439                 struct vhost_virtqueue *vq);
 440 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
 441 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 442                             uint64_t iova, uint64_t len);
 443
 444 static __rte_always_inline void
 445 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
 446 {
 447         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 448                 __vhost_log_write(dev, addr, len);
 449 }
 450
 451 static __rte_always_inline void
 452 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
 453 {
 454         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 455                 __vhost_log_cache_sync(dev, vq);
 456 }
 457
 458 static __rte_always_inline void
 459 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
 460                         uint64_t addr, uint64_t len)
 461 {
 462         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 463                 __vhost_log_cache_write(dev, vq, addr, len);
 464 }
 465
 466 static __rte_always_inline void
 467 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 468                         uint64_t offset, uint64_t len)
 469 {
 470         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
 471                 if (unlikely(vq->log_guest_addr == 0))
 472                         return;
 473                 __vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset,
 474                                         len);
 475         }
 476 }
 477
 478 static __rte_always_inline void
 479 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 480                      uint64_t offset, uint64_t len)
 481 {
 482         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
 483                 if (unlikely(vq->log_guest_addr == 0))
 484                         return;
 485                 __vhost_log_write(dev, vq->log_guest_addr + offset, len);
 486         }
 487 }
 488
 489 static __rte_always_inline void
 490 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 491                            uint64_t iova, uint64_t len)
 492 {
 493         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 494                 return;
 495
 496         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 497                 __vhost_log_cache_write_iova(dev, vq, iova, len);
 498         else
 499                 __vhost_log_cache_write(dev, vq, iova, len);
 500 }
 501
 502 static __rte_always_inline void
 503 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 504                            uint64_t iova, uint64_t len)
 505 {
 506         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 507                 return;
 508
 509         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 510                 __vhost_log_write_iova(dev, vq, iova, len);
 511         else
 512                 __vhost_log_write(dev, iova, len);
 513 }
 514
 515 extern int vhost_config_log_level;
 516 extern int vhost_data_log_level;
 517
 518 #define VHOST_LOG_CONFIG(level, fmt, args...)                   \
 519         rte_log(RTE_LOG_ ## level, vhost_config_log_level,      \
 520                 "VHOST_CONFIG: " fmt, ##args)
 521
 522 #define VHOST_LOG_DATA(level, fmt, args...) \
 523         (void)((RTE_LOG_ ## level <= RTE_LOG_DP_LEVEL) ?        \
 524          rte_log(RTE_LOG_ ## level,  vhost_data_log_level,      \
 525                 "VHOST_DATA : " fmt, ##args) :                  \
 526          0)
 527
 528 #ifdef RTE_LIBRTE_VHOST_DEBUG
 529 #define VHOST_MAX_PRINT_BUFF 6072
 530 #define PRINT_PACKET(device, addr, size, header) do { \
 531         char *pkt_addr = (char *)(addr); \
 532         unsigned int index; \
 533         char packet[VHOST_MAX_PRINT_BUFF]; \
 534         \
 535         if ((header)) \
 536                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
 537         else \
 538                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
 539         for (index = 0; index < (size); index++) { \
 540                 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
 541                         "%02hhx ", pkt_addr[index]); \
 542         } \
 543         snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
 544         \
 545         VHOST_LOG_DATA(DEBUG, "%s", packet); \
 546 } while (0)
 547 #else
 548 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
 549 #endif
 550
 551 #define MAX_VHOST_DEVICE        1024
 552 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
 553
 554 #define VHOST_BINARY_SEARCH_THRESH 256
 555
 556 static __rte_always_inline int guest_page_addrcmp(const void *p1,
 557                                                 const void *p2)
 558 {
 559         const struct guest_page *page1 = (const struct guest_page *)p1;
 560         const struct guest_page *page2 = (const struct guest_page *)p2;
 561
 562         if (page1->guest_phys_addr > page2->guest_phys_addr)
 563                 return 1;
 564         if (page1->guest_phys_addr < page2->guest_phys_addr)
 565                 return -1;
 566
 567         return 0;
 568 }
 569
 570 /* Convert guest physical address to host physical address */
 571 static __rte_always_inline rte_iova_t
 572 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
 573 {
 574         uint32_t i;
 575         struct guest_page *page;
 576         struct guest_page key;
 577
 578         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
 579                 key.guest_phys_addr = gpa;
 580                 page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages,
 581                                sizeof(struct guest_page), guest_page_addrcmp);
 582                 if (page) {
 583                         if (gpa + size < page->guest_phys_addr + page->size)
 584                                 return gpa - page->guest_phys_addr +
 585                                         page->host_phys_addr;
 586                 }
 587         } else {
 588                 for (i = 0; i < dev->nr_guest_pages; i++) {
 589                         page = &dev->guest_pages[i];
 590
 591                         if (gpa >= page->guest_phys_addr &&
 592                             gpa + size < page->guest_phys_addr +
 593                             page->size)
 594                                 return gpa - page->guest_phys_addr +
 595                                        page->host_phys_addr;
 596                 }
 597         }
 598
 599         return 0;
 600 }
 601
 602 static __rte_always_inline rte_iova_t
 603 gpa_to_first_hpa(struct virtio_net *dev, uint64_t gpa,
 604         uint64_t gpa_size, uint64_t *hpa_size)
 605 {
 606         uint32_t i;
 607         struct guest_page *page;
 608         struct guest_page key;
 609
 610         *hpa_size = gpa_size;
 611         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
 612                 key.guest_phys_addr = gpa & ~(dev->guest_pages[0].size - 1);
 613                 page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages,
 614                                sizeof(struct guest_page), guest_page_addrcmp);
 615                 if (page) {
 616                         if (gpa + gpa_size <=
 617                                         page->guest_phys_addr + page->size) {
 618                                 return gpa - page->guest_phys_addr +
 619                                         page->host_phys_addr;
 620                         } else if (gpa < page->guest_phys_addr +
 621                                                 page->size) {
 622                                 *hpa_size = page->guest_phys_addr +
 623                                         page->size - gpa;
 624                                 return gpa - page->guest_phys_addr +
 625                                         page->host_phys_addr;
 626                         }
 627                 }
 628         } else {
 629                 for (i = 0; i < dev->nr_guest_pages; i++) {
 630                         page = &dev->guest_pages[i];
 631
 632                         if (gpa >= page->guest_phys_addr) {
 633                                 if (gpa + gpa_size <=
 634                                         page->guest_phys_addr + page->size) {
 635                                         return gpa - page->guest_phys_addr +
 636                                                 page->host_phys_addr;
 637                                 } else if (gpa < page->guest_phys_addr +
 638                                                         page->size) {
 639                                         *hpa_size = page->guest_phys_addr +
 640                                                 page->size - gpa;
 641                                         return gpa - page->guest_phys_addr +
 642                                                 page->host_phys_addr;
 643                                 }
 644                         }
 645                 }
 646         }
 647
 648         *hpa_size = 0;
 649         return 0;
 650 }
 651
 652 static __rte_always_inline uint64_t
 653 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
 654 {
 655         struct rte_vhost_mem_region *r;
 656         uint32_t i;
 657
 658         if (unlikely(!dev || !dev->mem))
 659                 return 0;
 660
 661         for (i = 0; i < dev->mem->nregions; i++) {
 662                 r = &dev->mem->regions[i];
 663
 664                 if (vva >= r->host_user_addr &&
 665                     vva + len <  r->host_user_addr + r->size) {
 666                         return r->guest_phys_addr + vva - r->host_user_addr;
 667                 }
 668         }
 669         return 0;
 670 }
 671
 672 static __rte_always_inline struct virtio_net *
 673 get_device(int vid)
 674 {
 675         struct virtio_net *dev = vhost_devices[vid];
 676
 677         if (unlikely(!dev)) {
 678                 VHOST_LOG_CONFIG(ERR,
 679                         "(%d) device not found.\n", vid);
 680         }
 681
 682         return dev;
 683 }
 684
 685 int vhost_new_device(void);
 686 void cleanup_device(struct virtio_net *dev, int destroy);
 687 void reset_device(struct virtio_net *dev);
 688 void vhost_destroy_device(int);
 689 void vhost_destroy_device_notify(struct virtio_net *dev);
 690
 691 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
 692 void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
 693 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
 694
 695 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
 696
 697 void vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *dev);
 698
 699 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
 700 void vhost_set_builtin_virtio_net(int vid, bool enable);
 701 void vhost_enable_extbuf(int vid);
 702 void vhost_enable_linearbuf(int vid);
 703 int vhost_enable_guest_notification(struct virtio_net *dev,
 704                 struct vhost_virtqueue *vq, int enable);
 705
 706 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
 707
 708 /*
 709  * Backend-specific cleanup.
 710  *
 711  * TODO: fix it; we have one backend now
 712  */
 713 void vhost_backend_cleanup(struct virtio_net *dev);
 714
 715 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 716                         uint64_t iova, uint64_t *len, uint8_t perm);
 717 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
 718                         struct vhost_virtqueue *vq,
 719                         uint64_t desc_addr, uint64_t desc_len);
 720 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 721 uint64_t translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
 722                 uint64_t log_addr);
 723 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 724
 725 static __rte_always_inline uint64_t
 726 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 727                         uint64_t iova, uint64_t *len, uint8_t perm)
 728 {
 729         if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 730                 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
 731
 732         return __vhost_iova_to_vva(dev, vq, iova, len, perm);
 733 }
 734
 735 #define vhost_avail_event(vr) \
 736         (*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
 737 #define vhost_used_event(vr) \
 738         (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
 739
 740 /*
 741  * The following is used with VIRTIO_RING_F_EVENT_IDX.
 742  * Assuming a given event_idx value from the other size, if we have
 743  * just incremented index from old to new_idx, should we trigger an
 744  * event?
 745  */
 746 static __rte_always_inline int
 747 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
 748 {
 749         return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
 750 }
 751
 752 static __rte_always_inline void
 753 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 754 {
 755         /* Flush used->idx update before we read avail->flags. */
 756         rte_smp_mb();
 757
 758         /* Don't kick guest if we don't reach index specified by guest. */
 759         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
 760                 uint16_t old = vq->signalled_used;
 761                 uint16_t new = vq->async_pkts_inflight_n ?
 762                                         vq->used->idx:vq->last_used_idx;
 763                 bool signalled_used_valid = vq->signalled_used_valid;
 764
 765                 vq->signalled_used = new;
 766                 vq->signalled_used_valid = true;
 767
 768                 VHOST_LOG_DATA(DEBUG, "%s: used_event_idx=%d, old=%d, new=%d\n",
 769                         __func__,
 770                         vhost_used_event(vq),
 771                         old, new);
 772
 773                 if ((vhost_need_event(vhost_used_event(vq), new, old) &&
 774                                         (vq->callfd >= 0)) ||
 775                                 unlikely(!signalled_used_valid)) {
 776                         eventfd_write(vq->callfd, (eventfd_t) 1);
 777                         if (dev->notify_ops->guest_notified)
 778                                 dev->notify_ops->guest_notified(dev->vid);
 779                 }
 780         } else {
 781                 /* Kick the guest if necessary. */
 782                 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
 783                                 && (vq->callfd >= 0)) {
 784                         eventfd_write(vq->callfd, (eventfd_t)1);
 785                         if (dev->notify_ops->guest_notified)
 786                                 dev->notify_ops->guest_notified(dev->vid);
 787                 }
 788         }
 789 }
 790
 791 static __rte_always_inline void
 792 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
 793 {
 794         uint16_t old, new, off, off_wrap;
 795         bool signalled_used_valid, kick = false;
 796
 797         /* Flush used desc update. */
 798         rte_smp_mb();
 799
 800         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
 801                 if (vq->driver_event->flags !=
 802                                 VRING_EVENT_F_DISABLE)
 803                         kick = true;
 804                 goto kick;
 805         }
 806
 807         old = vq->signalled_used;
 808         new = vq->last_used_idx;
 809         vq->signalled_used = new;
 810         signalled_used_valid = vq->signalled_used_valid;
 811         vq->signalled_used_valid = true;
 812
 813         if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
 814                 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
 815                         kick = true;
 816                 goto kick;
 817         }
 818
 819         if (unlikely(!signalled_used_valid)) {
 820                 kick = true;
 821                 goto kick;
 822         }
 823
 824         rte_smp_rmb();
 825
 826         off_wrap = vq->driver_event->off_wrap;
 827         off = off_wrap & ~(1 << 15);
 828
 829         if (new <= old)
 830                 old -= vq->size;
 831
 832         if (vq->used_wrap_counter != off_wrap >> 15)
 833                 off -= vq->size;
 834
 835         if (vhost_need_event(off, new, old))
 836                 kick = true;
 837 kick:
 838         if (kick) {
 839                 eventfd_write(vq->callfd, (eventfd_t)1);
 840                 if (dev->notify_ops->guest_notified)
 841                         dev->notify_ops->guest_notified(dev->vid);
 842         }
 843 }
 844
 845 static __rte_always_inline void
 846 free_ind_table(void *idesc)
 847 {
 848         rte_free(idesc);
 849 }
 850
 851 static __rte_always_inline void
 852 restore_mbuf(struct rte_mbuf *m)
 853 {
 854         uint32_t mbuf_size, priv_size;
 855
 856         while (m) {
 857                 priv_size = rte_pktmbuf_priv_size(m->pool);
 858                 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
 859                 /* start of buffer is after mbuf structure and priv data */
 860
 861                 m->buf_addr = (char *)m + mbuf_size;
 862                 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
 863                 m = m->next;
 864         }
 865 }
 866
 867 static __rte_always_inline bool
 868 mbuf_is_consumed(struct rte_mbuf *m)
 869 {
 870         while (m) {
 871                 if (rte_mbuf_refcnt_read(m) > 1)
 872                         return false;
 873                 m = m->next;
 874         }
 875
 876         return true;
 877 }
 878
 879 #endif /* _VHOST_NET_CDEV_H_ */