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