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  * A structure to hold some fields needed in zero copy code path,
  99  * mainly for associating an mbuf with the right desc_idx.
 100  */
 101 struct zcopy_mbuf {
 102         struct rte_mbuf *mbuf;
 103         uint32_t desc_idx;
 104         uint16_t desc_count;
 105         uint16_t in_use;
 106
 107         TAILQ_ENTRY(zcopy_mbuf) next;
 108 };
 109 TAILQ_HEAD(zcopy_mbuf_list, zcopy_mbuf);
 110
 111 /*
 112  * Structure contains the info for each batched memory copy.
 113  */
 114 struct batch_copy_elem {
 115         void *dst;
 116         void *src;
 117         uint32_t len;
 118         uint64_t log_addr;
 119 };
 120
 121 /*
 122  * Structure that contains the info for batched dirty logging.
 123  */
 124 struct log_cache_entry {
 125         uint32_t offset;
 126         unsigned long val;
 127 };
 128
 129 struct vring_used_elem_packed {
 130         uint16_t id;
 131         uint16_t flags;
 132         uint32_t len;
 133         uint32_t count;
 134 };
 135
 136 /**
 137  * Structure contains variables relevant to RX/TX virtqueues.
 138  */
 139 struct vhost_virtqueue {
 140         union {
 141                 struct vring_desc       *desc;
 142                 struct vring_packed_desc   *desc_packed;
 143         };
 144         union {
 145                 struct vring_avail      *avail;
 146                 struct vring_packed_desc_event *driver_event;
 147         };
 148         union {
 149                 struct vring_used       *used;
 150                 struct vring_packed_desc_event *device_event;
 151         };
 152         uint32_t                size;
 153
 154         uint16_t                last_avail_idx;
 155         uint16_t                last_used_idx;
 156         /* Last used index we notify to front end. */
 157         uint16_t                signalled_used;
 158         bool                    signalled_used_valid;
 159 #define VIRTIO_INVALID_EVENTFD          (-1)
 160 #define VIRTIO_UNINITIALIZED_EVENTFD    (-2)
 161
 162         /* Backend value to determine if device should started/stopped */
 163         int                     backend;
 164         int                     enabled;
 165         int                     access_ok;
 166         int                     ready;
 167         int                     notif_enable;
 168 #define VIRTIO_UNINITIALIZED_NOTIF      (-1)
 169
 170         rte_spinlock_t          access_lock;
 171
 172         /* Used to notify the guest (trigger interrupt) */
 173         int                     callfd;
 174         /* Currently unused as polling mode is enabled */
 175         int                     kickfd;
 176
 177         /* Physical address of used ring, for logging */
 178         uint64_t                log_guest_addr;
 179
 180         /* inflight share memory info */
 181         union {
 182                 struct rte_vhost_inflight_info_split *inflight_split;
 183                 struct rte_vhost_inflight_info_packed *inflight_packed;
 184         };
 185         struct rte_vhost_resubmit_info *resubmit_inflight;
 186         uint64_t                global_counter;
 187
 188         uint16_t                nr_zmbuf;
 189         uint16_t                zmbuf_size;
 190         uint16_t                last_zmbuf_idx;
 191         struct zcopy_mbuf       *zmbufs;
 192         struct zcopy_mbuf_list  zmbuf_list;
 193
 194         union {
 195                 struct vring_used_elem  *shadow_used_split;
 196                 struct vring_used_elem_packed *shadow_used_packed;
 197         };
 198         uint16_t                shadow_used_idx;
 199         /* Record packed ring enqueue latest desc cache aligned index */
 200         uint16_t                shadow_aligned_idx;
 201         /* Record packed ring first dequeue desc index */
 202         uint16_t                shadow_last_used_idx;
 203         struct vhost_vring_addr ring_addrs;
 204
 205         struct batch_copy_elem  *batch_copy_elems;
 206         uint16_t                batch_copy_nb_elems;
 207         bool                    used_wrap_counter;
 208         bool                    avail_wrap_counter;
 209
 210         struct log_cache_entry log_cache[VHOST_LOG_CACHE_NR];
 211         uint16_t log_cache_nb_elem;
 212
 213         rte_rwlock_t    iotlb_lock;
 214         rte_rwlock_t    iotlb_pending_lock;
 215         struct rte_mempool *iotlb_pool;
 216         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
 217         int                             iotlb_cache_nr;
 218         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
 219
 220         /* operation callbacks for async dma */
 221         struct rte_vhost_async_channel_ops      async_ops;
 222
 223         struct rte_vhost_iov_iter it_pool[VHOST_MAX_ASYNC_IT];
 224         struct iovec vec_pool[VHOST_MAX_ASYNC_VEC];
 225
 226         /* async data transfer status */
 227         uintptr_t       **async_pkts_pending;
 228         #define         ASYNC_PENDING_INFO_N_MSK 0xFFFF
 229         #define         ASYNC_PENDING_INFO_N_SFT 16
 230         uint64_t        *async_pending_info;
 231         uint16_t        async_pkts_idx;
 232         uint16_t        async_pkts_inflight_n;
 233         uint16_t        async_last_seg_n;
 234
 235         /* vq async features */
 236         bool            async_inorder;
 237         bool            async_registered;
 238         uint16_t        async_threshold;
 239 } __rte_cache_aligned;
 240
 241 /* Virtio device status as per Virtio specification */
 242 #define VIRTIO_DEVICE_STATUS_RESET              0x00
 243 #define VIRTIO_DEVICE_STATUS_ACK                0x01
 244 #define VIRTIO_DEVICE_STATUS_DRIVER             0x02
 245 #define VIRTIO_DEVICE_STATUS_DRIVER_OK          0x04
 246 #define VIRTIO_DEVICE_STATUS_FEATURES_OK        0x08
 247 #define VIRTIO_DEVICE_STATUS_DEV_NEED_RESET     0x40
 248 #define VIRTIO_DEVICE_STATUS_FAILED             0x80
 249
 250 #define VHOST_MAX_VRING                 0x100
 251 #define VHOST_MAX_QUEUE_PAIRS           0x80
 252
 253 /* Declare IOMMU related bits for older kernels */
 254 #ifndef VIRTIO_F_IOMMU_PLATFORM
 255
 256 #define VIRTIO_F_IOMMU_PLATFORM 33
 257
 258 struct vhost_iotlb_msg {
 259         __u64 iova;
 260         __u64 size;
 261         __u64 uaddr;
 262 #define VHOST_ACCESS_RO      0x1
 263 #define VHOST_ACCESS_WO      0x2
 264 #define VHOST_ACCESS_RW      0x3
 265         __u8 perm;
 266 #define VHOST_IOTLB_MISS           1
 267 #define VHOST_IOTLB_UPDATE         2
 268 #define VHOST_IOTLB_INVALIDATE     3
 269 #define VHOST_IOTLB_ACCESS_FAIL    4
 270         __u8 type;
 271 };
 272
 273 #define VHOST_IOTLB_MSG 0x1
 274
 275 struct vhost_msg {
 276         int type;
 277         union {
 278                 struct vhost_iotlb_msg iotlb;
 279                 __u8 padding[64];
 280         };
 281 };
 282 #endif
 283
 284 /*
 285  * Define virtio 1.0 for older kernels
 286  */
 287 #ifndef VIRTIO_F_VERSION_1
 288  #define VIRTIO_F_VERSION_1 32
 289 #endif
 290
 291 /* Declare packed ring related bits for older kernels */
 292 #ifndef VIRTIO_F_RING_PACKED
 293
 294 #define VIRTIO_F_RING_PACKED 34
 295
 296 struct vring_packed_desc {
 297         uint64_t addr;
 298         uint32_t len;
 299         uint16_t id;
 300         uint16_t flags;
 301 };
 302
 303 struct vring_packed_desc_event {
 304         uint16_t off_wrap;
 305         uint16_t flags;
 306 };
 307 #endif
 308
 309 /*
 310  * Declare below packed ring defines unconditionally
 311  * as Kernel header might use different names.
 312  */
 313 #define VRING_DESC_F_AVAIL      (1ULL << 7)
 314 #define VRING_DESC_F_USED       (1ULL << 15)
 315
 316 #define VRING_EVENT_F_ENABLE 0x0
 317 #define VRING_EVENT_F_DISABLE 0x1
 318 #define VRING_EVENT_F_DESC 0x2
 319
 320 /*
 321  * Available and used descs are in same order
 322  */
 323 #ifndef VIRTIO_F_IN_ORDER
 324 #define VIRTIO_F_IN_ORDER      35
 325 #endif
 326
 327 /* Features supported by this builtin vhost-user net driver. */
 328 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
 329                                 (1ULL << VIRTIO_F_ANY_LAYOUT) | \
 330                                 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
 331                                 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
 332                                 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
 333                                 (1ULL << VIRTIO_NET_F_MQ)      | \
 334                                 (1ULL << VIRTIO_F_VERSION_1)   | \
 335                                 (1ULL << VHOST_F_LOG_ALL)      | \
 336                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
 337                                 (1ULL << VIRTIO_NET_F_GSO) | \
 338                                 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
 339                                 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
 340                                 (1ULL << VIRTIO_NET_F_HOST_UFO) | \
 341                                 (1ULL << VIRTIO_NET_F_HOST_ECN) | \
 342                                 (1ULL << VIRTIO_NET_F_CSUM)    | \
 343                                 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
 344                                 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
 345                                 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
 346                                 (1ULL << VIRTIO_NET_F_GUEST_UFO) | \
 347                                 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
 348                                 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
 349                                 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \
 350                                 (1ULL << VIRTIO_NET_F_MTU)  | \
 351                                 (1ULL << VIRTIO_F_IN_ORDER) | \
 352                                 (1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
 353                                 (1ULL << VIRTIO_F_RING_PACKED))
 354
 355
 356 struct guest_page {
 357         uint64_t guest_phys_addr;
 358         uint64_t host_phys_addr;
 359         uint64_t size;
 360 };
 361
 362 struct inflight_mem_info {
 363         int             fd;
 364         void            *addr;
 365         uint64_t        size;
 366 };
 367
 368 /**
 369  * Device structure contains all configuration information relating
 370  * to the device.
 371  */
 372 struct virtio_net {
 373         /* Frontend (QEMU) memory and memory region information */
 374         struct rte_vhost_memory *mem;
 375         uint64_t                features;
 376         uint64_t                protocol_features;
 377         int                     vid;
 378         uint32_t                flags;
 379         uint16_t                vhost_hlen;
 380         /* to tell if we need broadcast rarp packet */
 381         int16_t                 broadcast_rarp;
 382         uint32_t                nr_vring;
 383         int                     dequeue_zero_copy;
 384         int                     async_copy;
 385         int                     extbuf;
 386         int                     linearbuf;
 387         struct vhost_virtqueue  *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
 388         struct inflight_mem_info *inflight_info;
 389 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
 390         char                    ifname[IF_NAME_SZ];
 391         uint64_t                log_size;
 392         uint64_t                log_base;
 393         uint64_t                log_addr;
 394         struct rte_ether_addr   mac;
 395         uint16_t                mtu;
 396         uint8_t                 status;
 397
 398         struct vhost_device_ops const *notify_ops;
 399
 400         uint32_t                nr_guest_pages;
 401         uint32_t                max_guest_pages;
 402         struct guest_page       *guest_pages;
 403
 404         int                     slave_req_fd;
 405         rte_spinlock_t          slave_req_lock;
 406
 407         int                     postcopy_ufd;
 408         int                     postcopy_listening;
 409
 410         struct rte_vdpa_device *vdpa_dev;
 411
 412         /* context data for the external message handlers */
 413         void                    *extern_data;
 414         /* pre and post vhost user message handlers for the device */
 415         struct rte_vhost_user_extern_ops extern_ops;
 416 } __rte_cache_aligned;
 417
 418 static __rte_always_inline bool
 419 vq_is_packed(struct virtio_net *dev)
 420 {
 421         return dev->features & (1ull << VIRTIO_F_RING_PACKED);
 422 }
 423
 424 static inline bool
 425 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
 426 {
 427         uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
 428
 429         return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
 430                 wrap_counter != !!(flags & VRING_DESC_F_USED);
 431 }
 432
 433 static inline void
 434 vq_inc_last_used_packed(struct vhost_virtqueue *vq, uint16_t num)
 435 {
 436         vq->last_used_idx += num;
 437         if (vq->last_used_idx >= vq->size) {
 438                 vq->used_wrap_counter ^= 1;
 439                 vq->last_used_idx -= vq->size;
 440         }
 441 }
 442
 443 static inline void
 444 vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
 445 {
 446         vq->last_avail_idx += num;
 447         if (vq->last_avail_idx >= vq->size) {
 448                 vq->avail_wrap_counter ^= 1;
 449                 vq->last_avail_idx -= vq->size;
 450         }
 451 }
 452
 453 void __vhost_log_cache_write(struct virtio_net *dev,
 454                 struct vhost_virtqueue *vq,
 455                 uint64_t addr, uint64_t len);
 456 void __vhost_log_cache_write_iova(struct virtio_net *dev,
 457                 struct vhost_virtqueue *vq,
 458                 uint64_t iova, uint64_t len);
 459 void __vhost_log_cache_sync(struct virtio_net *dev,
 460                 struct vhost_virtqueue *vq);
 461 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
 462 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 463                             uint64_t iova, uint64_t len);
 464
 465 static __rte_always_inline void
 466 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
 467 {
 468         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 469                 __vhost_log_write(dev, addr, len);
 470 }
 471
 472 static __rte_always_inline void
 473 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
 474 {
 475         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 476                 __vhost_log_cache_sync(dev, vq);
 477 }
 478
 479 static __rte_always_inline void
 480 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
 481                         uint64_t addr, uint64_t len)
 482 {
 483         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 484                 __vhost_log_cache_write(dev, vq, addr, len);
 485 }
 486
 487 static __rte_always_inline void
 488 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 489                         uint64_t offset, uint64_t len)
 490 {
 491         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
 492                 if (unlikely(vq->log_guest_addr == 0))
 493                         return;
 494                 __vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset,
 495                                         len);
 496         }
 497 }
 498
 499 static __rte_always_inline void
 500 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 501                      uint64_t offset, uint64_t len)
 502 {
 503         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
 504                 if (unlikely(vq->log_guest_addr == 0))
 505                         return;
 506                 __vhost_log_write(dev, vq->log_guest_addr + offset, len);
 507         }
 508 }
 509
 510 static __rte_always_inline void
 511 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 512                            uint64_t iova, uint64_t len)
 513 {
 514         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 515                 return;
 516
 517         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 518                 __vhost_log_cache_write_iova(dev, vq, iova, len);
 519         else
 520                 __vhost_log_cache_write(dev, vq, iova, len);
 521 }
 522
 523 static __rte_always_inline void
 524 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 525                            uint64_t iova, uint64_t len)
 526 {
 527         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 528                 return;
 529
 530         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 531                 __vhost_log_write_iova(dev, vq, iova, len);
 532         else
 533                 __vhost_log_write(dev, iova, len);
 534 }
 535
 536 extern int vhost_config_log_level;
 537 extern int vhost_data_log_level;
 538
 539 #define VHOST_LOG_CONFIG(level, fmt, args...)                   \
 540         rte_log(RTE_LOG_ ## level, vhost_config_log_level,      \
 541                 "VHOST_CONFIG: " fmt, ##args)
 542
 543 #define VHOST_LOG_DATA(level, fmt, args...) \
 544         (void)((RTE_LOG_ ## level <= RTE_LOG_DP_LEVEL) ?        \
 545          rte_log(RTE_LOG_ ## level,  vhost_data_log_level,      \
 546                 "VHOST_DATA : " fmt, ##args) :                  \
 547          0)
 548
 549 #ifdef RTE_LIBRTE_VHOST_DEBUG
 550 #define VHOST_MAX_PRINT_BUFF 6072
 551 #define PRINT_PACKET(device, addr, size, header) do { \
 552         char *pkt_addr = (char *)(addr); \
 553         unsigned int index; \
 554         char packet[VHOST_MAX_PRINT_BUFF]; \
 555         \
 556         if ((header)) \
 557                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
 558         else \
 559                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
 560         for (index = 0; index < (size); index++) { \
 561                 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
 562                         "%02hhx ", pkt_addr[index]); \
 563         } \
 564         snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
 565         \
 566         VHOST_LOG_DATA(DEBUG, "%s", packet); \
 567 } while (0)
 568 #else
 569 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
 570 #endif
 571
 572 #define MAX_VHOST_DEVICE        1024
 573 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
 574
 575 #define VHOST_BINARY_SEARCH_THRESH 256
 576
 577 static __rte_always_inline int guest_page_addrcmp(const void *p1,
 578                                                 const void *p2)
 579 {
 580         const struct guest_page *page1 = (const struct guest_page *)p1;
 581         const struct guest_page *page2 = (const struct guest_page *)p2;
 582
 583         if (page1->guest_phys_addr > page2->guest_phys_addr)
 584                 return 1;
 585         if (page1->guest_phys_addr < page2->guest_phys_addr)
 586                 return -1;
 587
 588         return 0;
 589 }
 590
 591 /* Convert guest physical address to host physical address */
 592 static __rte_always_inline rte_iova_t
 593 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
 594 {
 595         uint32_t i;
 596         struct guest_page *page;
 597         struct guest_page key;
 598
 599         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
 600                 key.guest_phys_addr = gpa;
 601                 page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages,
 602                                sizeof(struct guest_page), guest_page_addrcmp);
 603                 if (page) {
 604                         if (gpa + size < page->guest_phys_addr + page->size)
 605                                 return gpa - page->guest_phys_addr +
 606                                         page->host_phys_addr;
 607                 }
 608         } else {
 609                 for (i = 0; i < dev->nr_guest_pages; i++) {
 610                         page = &dev->guest_pages[i];
 611
 612                         if (gpa >= page->guest_phys_addr &&
 613                             gpa + size < page->guest_phys_addr +
 614                             page->size)
 615                                 return gpa - page->guest_phys_addr +
 616                                        page->host_phys_addr;
 617                 }
 618         }
 619
 620         return 0;
 621 }
 622
 623 static __rte_always_inline rte_iova_t
 624 gpa_to_first_hpa(struct virtio_net *dev, uint64_t gpa,
 625         uint64_t gpa_size, uint64_t *hpa_size)
 626 {
 627         uint32_t i;
 628         struct guest_page *page;
 629         struct guest_page key;
 630
 631         *hpa_size = gpa_size;
 632         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
 633                 key.guest_phys_addr = gpa & ~(dev->guest_pages[0].size - 1);
 634                 page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages,
 635                                sizeof(struct guest_page), guest_page_addrcmp);
 636                 if (page) {
 637                         if (gpa + gpa_size <=
 638                                         page->guest_phys_addr + page->size) {
 639                                 return gpa - page->guest_phys_addr +
 640                                         page->host_phys_addr;
 641                         } else if (gpa < page->guest_phys_addr +
 642                                                 page->size) {
 643                                 *hpa_size = page->guest_phys_addr +
 644                                         page->size - gpa;
 645                                 return gpa - page->guest_phys_addr +
 646                                         page->host_phys_addr;
 647                         }
 648                 }
 649         } else {
 650                 for (i = 0; i < dev->nr_guest_pages; i++) {
 651                         page = &dev->guest_pages[i];
 652
 653                         if (gpa >= page->guest_phys_addr) {
 654                                 if (gpa + gpa_size <=
 655                                         page->guest_phys_addr + page->size) {
 656                                         return gpa - page->guest_phys_addr +
 657                                                 page->host_phys_addr;
 658                                 } else if (gpa < page->guest_phys_addr +
 659                                                         page->size) {
 660                                         *hpa_size = page->guest_phys_addr +
 661                                                 page->size - gpa;
 662                                         return gpa - page->guest_phys_addr +
 663                                                 page->host_phys_addr;
 664                                 }
 665                         }
 666                 }
 667         }
 668
 669         *hpa_size = 0;
 670         return 0;
 671 }
 672
 673 static __rte_always_inline uint64_t
 674 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
 675 {
 676         struct rte_vhost_mem_region *r;
 677         uint32_t i;
 678
 679         if (unlikely(!dev || !dev->mem))
 680                 return 0;
 681
 682         for (i = 0; i < dev->mem->nregions; i++) {
 683                 r = &dev->mem->regions[i];
 684
 685                 if (vva >= r->host_user_addr &&
 686                     vva + len <  r->host_user_addr + r->size) {
 687                         return r->guest_phys_addr + vva - r->host_user_addr;
 688                 }
 689         }
 690         return 0;
 691 }
 692
 693 static __rte_always_inline struct virtio_net *
 694 get_device(int vid)
 695 {
 696         struct virtio_net *dev = vhost_devices[vid];
 697
 698         if (unlikely(!dev)) {
 699                 VHOST_LOG_CONFIG(ERR,
 700                         "(%d) device not found.\n", vid);
 701         }
 702
 703         return dev;
 704 }
 705
 706 int vhost_new_device(void);
 707 void cleanup_device(struct virtio_net *dev, int destroy);
 708 void reset_device(struct virtio_net *dev);
 709 void vhost_destroy_device(int);
 710 void vhost_destroy_device_notify(struct virtio_net *dev);
 711
 712 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
 713 void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
 714 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
 715
 716 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
 717
 718 void vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *dev);
 719
 720 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
 721 void vhost_enable_dequeue_zero_copy(int vid);
 722 void vhost_set_builtin_virtio_net(int vid, bool enable);
 723 void vhost_enable_extbuf(int vid);
 724 void vhost_enable_linearbuf(int vid);
 725 int vhost_enable_guest_notification(struct virtio_net *dev,
 726                 struct vhost_virtqueue *vq, int enable);
 727
 728 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
 729
 730 /*
 731  * Backend-specific cleanup.
 732  *
 733  * TODO: fix it; we have one backend now
 734  */
 735 void vhost_backend_cleanup(struct virtio_net *dev);
 736
 737 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 738                         uint64_t iova, uint64_t *len, uint8_t perm);
 739 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
 740                         struct vhost_virtqueue *vq,
 741                         uint64_t desc_addr, uint64_t desc_len);
 742 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 743 uint64_t translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
 744                 uint64_t log_addr);
 745 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 746
 747 static __rte_always_inline uint64_t
 748 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 749                         uint64_t iova, uint64_t *len, uint8_t perm)
 750 {
 751         if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 752                 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
 753
 754         return __vhost_iova_to_vva(dev, vq, iova, len, perm);
 755 }
 756
 757 #define vhost_avail_event(vr) \
 758         (*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
 759 #define vhost_used_event(vr) \
 760         (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
 761
 762 /*
 763  * The following is used with VIRTIO_RING_F_EVENT_IDX.
 764  * Assuming a given event_idx value from the other size, if we have
 765  * just incremented index from old to new_idx, should we trigger an
 766  * event?
 767  */
 768 static __rte_always_inline int
 769 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
 770 {
 771         return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
 772 }
 773
 774 static __rte_always_inline void
 775 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 776 {
 777         /* Flush used->idx update before we read avail->flags. */
 778         rte_smp_mb();
 779
 780         /* Don't kick guest if we don't reach index specified by guest. */
 781         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
 782                 uint16_t old = vq->signalled_used;
 783                 uint16_t new = vq->async_pkts_inflight_n ?
 784                                         vq->used->idx:vq->last_used_idx;
 785                 bool signalled_used_valid = vq->signalled_used_valid;
 786
 787                 vq->signalled_used = new;
 788                 vq->signalled_used_valid = true;
 789
 790                 VHOST_LOG_DATA(DEBUG, "%s: used_event_idx=%d, old=%d, new=%d\n",
 791                         __func__,
 792                         vhost_used_event(vq),
 793                         old, new);
 794
 795                 if ((vhost_need_event(vhost_used_event(vq), new, old) &&
 796                                         (vq->callfd >= 0)) ||
 797                                 unlikely(!signalled_used_valid)) {
 798                         eventfd_write(vq->callfd, (eventfd_t) 1);
 799                         if (dev->notify_ops->guest_notified)
 800                                 dev->notify_ops->guest_notified(dev->vid);
 801                 }
 802         } else {
 803                 /* Kick the guest if necessary. */
 804                 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
 805                                 && (vq->callfd >= 0)) {
 806                         eventfd_write(vq->callfd, (eventfd_t)1);
 807                         if (dev->notify_ops->guest_notified)
 808                                 dev->notify_ops->guest_notified(dev->vid);
 809                 }
 810         }
 811 }
 812
 813 static __rte_always_inline void
 814 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
 815 {
 816         uint16_t old, new, off, off_wrap;
 817         bool signalled_used_valid, kick = false;
 818
 819         /* Flush used desc update. */
 820         rte_smp_mb();
 821
 822         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
 823                 if (vq->driver_event->flags !=
 824                                 VRING_EVENT_F_DISABLE)
 825                         kick = true;
 826                 goto kick;
 827         }
 828
 829         old = vq->signalled_used;
 830         new = vq->last_used_idx;
 831         vq->signalled_used = new;
 832         signalled_used_valid = vq->signalled_used_valid;
 833         vq->signalled_used_valid = true;
 834
 835         if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
 836                 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
 837                         kick = true;
 838                 goto kick;
 839         }
 840
 841         if (unlikely(!signalled_used_valid)) {
 842                 kick = true;
 843                 goto kick;
 844         }
 845
 846         rte_smp_rmb();
 847
 848         off_wrap = vq->driver_event->off_wrap;
 849         off = off_wrap & ~(1 << 15);
 850
 851         if (new <= old)
 852                 old -= vq->size;
 853
 854         if (vq->used_wrap_counter != off_wrap >> 15)
 855                 off -= vq->size;
 856
 857         if (vhost_need_event(off, new, old))
 858                 kick = true;
 859 kick:
 860         if (kick) {
 861                 eventfd_write(vq->callfd, (eventfd_t)1);
 862                 if (dev->notify_ops->guest_notified)
 863                         dev->notify_ops->guest_notified(dev->vid);
 864         }
 865 }
 866
 867 static __rte_always_inline void
 868 free_ind_table(void *idesc)
 869 {
 870         rte_free(idesc);
 871 }
 872
 873 static __rte_always_inline void
 874 restore_mbuf(struct rte_mbuf *m)
 875 {
 876         uint32_t mbuf_size, priv_size;
 877
 878         while (m) {
 879                 priv_size = rte_pktmbuf_priv_size(m->pool);
 880                 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
 881                 /* start of buffer is after mbuf structure and priv data */
 882
 883                 m->buf_addr = (char *)m + mbuf_size;
 884                 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
 885                 m = m->next;
 886         }
 887 }
 888
 889 static __rte_always_inline bool
 890 mbuf_is_consumed(struct rte_mbuf *m)
 891 {
 892         while (m) {
 893                 if (rte_mbuf_refcnt_read(m) > 1)
 894                         return false;
 895                 m = m->next;
 896         }
 897
 898         return true;
 899 }
 900
 901 static __rte_always_inline void
 902 put_zmbuf(struct zcopy_mbuf *zmbuf)
 903 {
 904         zmbuf->in_use = 0;
 905 }
 906
 907 #endif /* _VHOST_NET_CDEV_H_ */