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
  38 /* Backend value set by guest. */
  39 #define VIRTIO_DEV_STOPPED -1
  40
  41 #define BUF_VECTOR_MAX 256
  42
  43 #define VHOST_LOG_CACHE_NR 32
  44
  45 #define MAX_PKT_BURST 32
  46
  47 #define VHOST_MAX_ASYNC_IT (MAX_PKT_BURST * 2)
  48 #define VHOST_MAX_ASYNC_VEC (BUF_VECTOR_MAX * 2)
  49
  50 #define PACKED_DESC_ENQUEUE_USED_FLAG(w)        \
  51         ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED | VRING_DESC_F_WRITE) : \
  52                 VRING_DESC_F_WRITE)
  53 #define PACKED_DESC_DEQUEUE_USED_FLAG(w)        \
  54         ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED) : 0x0)
  55 #define PACKED_DESC_SINGLE_DEQUEUE_FLAG (VRING_DESC_F_NEXT | \
  56                                          VRING_DESC_F_INDIRECT)
  57
  58 #define PACKED_BATCH_SIZE (RTE_CACHE_LINE_SIZE / \
  59                             sizeof(struct vring_packed_desc))
  60 #define PACKED_BATCH_MASK (PACKED_BATCH_SIZE - 1)
  61
  62 #ifdef VHOST_GCC_UNROLL_PRAGMA
  63 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("GCC unroll 4") \
  64         for (iter = val; iter < size; iter++)
  65 #endif
  66
  67 #ifdef VHOST_CLANG_UNROLL_PRAGMA
  68 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll 4") \
  69         for (iter = val; iter < size; iter++)
  70 #endif
  71
  72 #ifdef VHOST_ICC_UNROLL_PRAGMA
  73 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll (4)") \
  74         for (iter = val; iter < size; iter++)
  75 #endif
  76
  77 #ifndef vhost_for_each_try_unroll
  78 #define vhost_for_each_try_unroll(iter, val, num) \
  79         for (iter = val; iter < num; iter++)
  80 #endif
  81
  82 /**
  83  * Structure contains buffer address, length and descriptor index
  84  * from vring to do scatter RX.
  85  */
  86 struct buf_vector {
  87         uint64_t buf_iova;
  88         uint64_t buf_addr;
  89         uint32_t buf_len;
  90         uint32_t desc_idx;
  91 };
  92
  93 /*
  94  * A structure to hold some fields needed in zero copy code path,
  95  * mainly for associating an mbuf with the right desc_idx.
  96  */
  97 struct zcopy_mbuf {
  98         struct rte_mbuf *mbuf;
  99         uint32_t desc_idx;
 100         uint16_t desc_count;
 101         uint16_t in_use;
 102
 103         TAILQ_ENTRY(zcopy_mbuf) next;
 104 };
 105 TAILQ_HEAD(zcopy_mbuf_list, zcopy_mbuf);
 106
 107 /*
 108  * Structure contains the info for each batched memory copy.
 109  */
 110 struct batch_copy_elem {
 111         void *dst;
 112         void *src;
 113         uint32_t len;
 114         uint64_t log_addr;
 115 };
 116
 117 /*
 118  * Structure that contains the info for batched dirty logging.
 119  */
 120 struct log_cache_entry {
 121         uint32_t offset;
 122         unsigned long val;
 123 };
 124
 125 struct vring_used_elem_packed {
 126         uint16_t id;
 127         uint16_t flags;
 128         uint32_t len;
 129         uint32_t count;
 130 };
 131
 132 /**
 133  * Structure contains variables relevant to RX/TX virtqueues.
 134  */
 135 struct vhost_virtqueue {
 136         union {
 137                 struct vring_desc       *desc;
 138                 struct vring_packed_desc   *desc_packed;
 139         };
 140         union {
 141                 struct vring_avail      *avail;
 142                 struct vring_packed_desc_event *driver_event;
 143         };
 144         union {
 145                 struct vring_used       *used;
 146                 struct vring_packed_desc_event *device_event;
 147         };
 148         uint32_t                size;
 149
 150         uint16_t                last_avail_idx;
 151         uint16_t                last_used_idx;
 152         /* Last used index we notify to front end. */
 153         uint16_t                signalled_used;
 154         bool                    signalled_used_valid;
 155 #define VIRTIO_INVALID_EVENTFD          (-1)
 156 #define VIRTIO_UNINITIALIZED_EVENTFD    (-2)
 157
 158         /* Backend value to determine if device should started/stopped */
 159         int                     backend;
 160         int                     enabled;
 161         int                     access_ok;
 162         int                     ready;
 163         rte_spinlock_t          access_lock;
 164
 165         /* Used to notify the guest (trigger interrupt) */
 166         int                     callfd;
 167         /* Currently unused as polling mode is enabled */
 168         int                     kickfd;
 169
 170         /* Physical address of used ring, for logging */
 171         uint64_t                log_guest_addr;
 172
 173         /* inflight share memory info */
 174         union {
 175                 struct rte_vhost_inflight_info_split *inflight_split;
 176                 struct rte_vhost_inflight_info_packed *inflight_packed;
 177         };
 178         struct rte_vhost_resubmit_info *resubmit_inflight;
 179         uint64_t                global_counter;
 180
 181         uint16_t                nr_zmbuf;
 182         uint16_t                zmbuf_size;
 183         uint16_t                last_zmbuf_idx;
 184         struct zcopy_mbuf       *zmbufs;
 185         struct zcopy_mbuf_list  zmbuf_list;
 186
 187         union {
 188                 struct vring_used_elem  *shadow_used_split;
 189                 struct vring_used_elem_packed *shadow_used_packed;
 190         };
 191         uint16_t                shadow_used_idx;
 192         /* Record packed ring enqueue latest desc cache aligned index */
 193         uint16_t                shadow_aligned_idx;
 194         /* Record packed ring first dequeue desc index */
 195         uint16_t                shadow_last_used_idx;
 196         struct vhost_vring_addr ring_addrs;
 197
 198         struct batch_copy_elem  *batch_copy_elems;
 199         uint16_t                batch_copy_nb_elems;
 200         bool                    used_wrap_counter;
 201         bool                    avail_wrap_counter;
 202
 203         struct log_cache_entry log_cache[VHOST_LOG_CACHE_NR];
 204         uint16_t log_cache_nb_elem;
 205
 206         rte_rwlock_t    iotlb_lock;
 207         rte_rwlock_t    iotlb_pending_lock;
 208         struct rte_mempool *iotlb_pool;
 209         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
 210         int                             iotlb_cache_nr;
 211         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
 212
 213         /* operation callbacks for async dma */
 214         struct rte_vhost_async_channel_ops      async_ops;
 215
 216         struct rte_vhost_iov_iter it_pool[VHOST_MAX_ASYNC_IT];
 217         struct iovec vec_pool[VHOST_MAX_ASYNC_VEC];
 218
 219         /* async data transfer status */
 220         uintptr_t       **async_pkts_pending;
 221         #define         ASYNC_PENDING_INFO_N_MSK 0xFFFF
 222         #define         ASYNC_PENDING_INFO_N_SFT 16
 223         uint64_t        *async_pending_info;
 224         uint16_t        async_pkts_idx;
 225         uint16_t        async_pkts_inflight_n;
 226
 227         /* vq async features */
 228         bool            async_inorder;
 229         bool            async_registered;
 230         uint16_t        async_threshold;
 231 } __rte_cache_aligned;
 232
 233 #define VHOST_MAX_VRING                 0x100
 234 #define VHOST_MAX_QUEUE_PAIRS           0x80
 235
 236 /* Declare IOMMU related bits for older kernels */
 237 #ifndef VIRTIO_F_IOMMU_PLATFORM
 238
 239 #define VIRTIO_F_IOMMU_PLATFORM 33
 240
 241 struct vhost_iotlb_msg {
 242         __u64 iova;
 243         __u64 size;
 244         __u64 uaddr;
 245 #define VHOST_ACCESS_RO      0x1
 246 #define VHOST_ACCESS_WO      0x2
 247 #define VHOST_ACCESS_RW      0x3
 248         __u8 perm;
 249 #define VHOST_IOTLB_MISS           1
 250 #define VHOST_IOTLB_UPDATE         2
 251 #define VHOST_IOTLB_INVALIDATE     3
 252 #define VHOST_IOTLB_ACCESS_FAIL    4
 253         __u8 type;
 254 };
 255
 256 #define VHOST_IOTLB_MSG 0x1
 257
 258 struct vhost_msg {
 259         int type;
 260         union {
 261                 struct vhost_iotlb_msg iotlb;
 262                 __u8 padding[64];
 263         };
 264 };
 265 #endif
 266
 267 /*
 268  * Define virtio 1.0 for older kernels
 269  */
 270 #ifndef VIRTIO_F_VERSION_1
 271  #define VIRTIO_F_VERSION_1 32
 272 #endif
 273
 274 /* Declare packed ring related bits for older kernels */
 275 #ifndef VIRTIO_F_RING_PACKED
 276
 277 #define VIRTIO_F_RING_PACKED 34
 278
 279 struct vring_packed_desc {
 280         uint64_t addr;
 281         uint32_t len;
 282         uint16_t id;
 283         uint16_t flags;
 284 };
 285
 286 struct vring_packed_desc_event {
 287         uint16_t off_wrap;
 288         uint16_t flags;
 289 };
 290 #endif
 291
 292 /*
 293  * Declare below packed ring defines unconditionally
 294  * as Kernel header might use different names.
 295  */
 296 #define VRING_DESC_F_AVAIL      (1ULL << 7)
 297 #define VRING_DESC_F_USED       (1ULL << 15)
 298
 299 #define VRING_EVENT_F_ENABLE 0x0
 300 #define VRING_EVENT_F_DISABLE 0x1
 301 #define VRING_EVENT_F_DESC 0x2
 302
 303 /*
 304  * Available and used descs are in same order
 305  */
 306 #ifndef VIRTIO_F_IN_ORDER
 307 #define VIRTIO_F_IN_ORDER      35
 308 #endif
 309
 310 /* Features supported by this builtin vhost-user net driver. */
 311 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
 312                                 (1ULL << VIRTIO_F_ANY_LAYOUT) | \
 313                                 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
 314                                 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
 315                                 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
 316                                 (1ULL << VIRTIO_NET_F_MQ)      | \
 317                                 (1ULL << VIRTIO_F_VERSION_1)   | \
 318                                 (1ULL << VHOST_F_LOG_ALL)      | \
 319                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
 320                                 (1ULL << VIRTIO_NET_F_GSO) | \
 321                                 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
 322                                 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
 323                                 (1ULL << VIRTIO_NET_F_HOST_UFO) | \
 324                                 (1ULL << VIRTIO_NET_F_HOST_ECN) | \
 325                                 (1ULL << VIRTIO_NET_F_CSUM)    | \
 326                                 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
 327                                 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
 328                                 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
 329                                 (1ULL << VIRTIO_NET_F_GUEST_UFO) | \
 330                                 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
 331                                 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
 332                                 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \
 333                                 (1ULL << VIRTIO_NET_F_MTU)  | \
 334                                 (1ULL << VIRTIO_F_IN_ORDER) | \
 335                                 (1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
 336                                 (1ULL << VIRTIO_F_RING_PACKED))
 337
 338
 339 struct guest_page {
 340         uint64_t guest_phys_addr;
 341         uint64_t host_phys_addr;
 342         uint64_t size;
 343 };
 344
 345 struct inflight_mem_info {
 346         int             fd;
 347         void            *addr;
 348         uint64_t        size;
 349 };
 350
 351 /**
 352  * Device structure contains all configuration information relating
 353  * to the device.
 354  */
 355 struct virtio_net {
 356         /* Frontend (QEMU) memory and memory region information */
 357         struct rte_vhost_memory *mem;
 358         uint64_t                features;
 359         uint64_t                protocol_features;
 360         int                     vid;
 361         uint32_t                flags;
 362         uint16_t                vhost_hlen;
 363         /* to tell if we need broadcast rarp packet */
 364         int16_t                 broadcast_rarp;
 365         uint32_t                nr_vring;
 366         int                     dequeue_zero_copy;
 367         int                     async_copy;
 368         int                     extbuf;
 369         int                     linearbuf;
 370         struct vhost_virtqueue  *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
 371         struct inflight_mem_info *inflight_info;
 372 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
 373         char                    ifname[IF_NAME_SZ];
 374         uint64_t                log_size;
 375         uint64_t                log_base;
 376         uint64_t                log_addr;
 377         struct rte_ether_addr   mac;
 378         uint16_t                mtu;
 379
 380         struct vhost_device_ops const *notify_ops;
 381
 382         uint32_t                nr_guest_pages;
 383         uint32_t                max_guest_pages;
 384         struct guest_page       *guest_pages;
 385
 386         int                     slave_req_fd;
 387         rte_spinlock_t          slave_req_lock;
 388
 389         int                     postcopy_ufd;
 390         int                     postcopy_listening;
 391
 392         struct rte_vdpa_device *vdpa_dev;
 393
 394         /* context data for the external message handlers */
 395         void                    *extern_data;
 396         /* pre and post vhost user message handlers for the device */
 397         struct rte_vhost_user_extern_ops extern_ops;
 398 } __rte_cache_aligned;
 399
 400 static __rte_always_inline bool
 401 vq_is_packed(struct virtio_net *dev)
 402 {
 403         return dev->features & (1ull << VIRTIO_F_RING_PACKED);
 404 }
 405
 406 static inline bool
 407 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
 408 {
 409         uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
 410
 411         return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
 412                 wrap_counter != !!(flags & VRING_DESC_F_USED);
 413 }
 414
 415 static inline void
 416 vq_inc_last_used_packed(struct vhost_virtqueue *vq, uint16_t num)
 417 {
 418         vq->last_used_idx += num;
 419         if (vq->last_used_idx >= vq->size) {
 420                 vq->used_wrap_counter ^= 1;
 421                 vq->last_used_idx -= vq->size;
 422         }
 423 }
 424
 425 static inline void
 426 vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
 427 {
 428         vq->last_avail_idx += num;
 429         if (vq->last_avail_idx >= vq->size) {
 430                 vq->avail_wrap_counter ^= 1;
 431                 vq->last_avail_idx -= vq->size;
 432         }
 433 }
 434
 435 void __vhost_log_cache_write(struct virtio_net *dev,
 436                 struct vhost_virtqueue *vq,
 437                 uint64_t addr, uint64_t len);
 438 void __vhost_log_cache_write_iova(struct virtio_net *dev,
 439                 struct vhost_virtqueue *vq,
 440                 uint64_t iova, uint64_t len);
 441 void __vhost_log_cache_sync(struct virtio_net *dev,
 442                 struct vhost_virtqueue *vq);
 443 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
 444 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 445                             uint64_t iova, uint64_t len);
 446
 447 static __rte_always_inline void
 448 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
 449 {
 450         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 451                 __vhost_log_write(dev, addr, len);
 452 }
 453
 454 static __rte_always_inline void
 455 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
 456 {
 457         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 458                 __vhost_log_cache_sync(dev, vq);
 459 }
 460
 461 static __rte_always_inline void
 462 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
 463                         uint64_t addr, uint64_t len)
 464 {
 465         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 466                 __vhost_log_cache_write(dev, vq, addr, len);
 467 }
 468
 469 static __rte_always_inline void
 470 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 471                         uint64_t offset, uint64_t len)
 472 {
 473         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
 474                 if (unlikely(vq->log_guest_addr == 0))
 475                         return;
 476                 __vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset,
 477                                         len);
 478         }
 479 }
 480
 481 static __rte_always_inline void
 482 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 483                      uint64_t offset, uint64_t len)
 484 {
 485         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
 486                 if (unlikely(vq->log_guest_addr == 0))
 487                         return;
 488                 __vhost_log_write(dev, vq->log_guest_addr + offset, len);
 489         }
 490 }
 491
 492 static __rte_always_inline void
 493 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 494                            uint64_t iova, uint64_t len)
 495 {
 496         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 497                 return;
 498
 499         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 500                 __vhost_log_cache_write_iova(dev, vq, iova, len);
 501         else
 502                 __vhost_log_cache_write(dev, vq, iova, len);
 503 }
 504
 505 static __rte_always_inline void
 506 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 507                            uint64_t iova, uint64_t len)
 508 {
 509         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 510                 return;
 511
 512         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 513                 __vhost_log_write_iova(dev, vq, iova, len);
 514         else
 515                 __vhost_log_write(dev, iova, len);
 516 }
 517
 518 extern int vhost_config_log_level;
 519 extern int vhost_data_log_level;
 520
 521 #define VHOST_LOG_CONFIG(level, fmt, args...)                   \
 522         rte_log(RTE_LOG_ ## level, vhost_config_log_level,      \
 523                 "VHOST_CONFIG: " fmt, ##args)
 524
 525 #define VHOST_LOG_DATA(level, fmt, args...) \
 526         (void)((RTE_LOG_ ## level <= RTE_LOG_DP_LEVEL) ?        \
 527          rte_log(RTE_LOG_ ## level,  vhost_data_log_level,      \
 528                 "VHOST_DATA : " fmt, ##args) :                  \
 529          0)
 530
 531 #ifdef RTE_LIBRTE_VHOST_DEBUG
 532 #define VHOST_MAX_PRINT_BUFF 6072
 533 #define PRINT_PACKET(device, addr, size, header) do { \
 534         char *pkt_addr = (char *)(addr); \
 535         unsigned int index; \
 536         char packet[VHOST_MAX_PRINT_BUFF]; \
 537         \
 538         if ((header)) \
 539                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
 540         else \
 541                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
 542         for (index = 0; index < (size); index++) { \
 543                 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
 544                         "%02hhx ", pkt_addr[index]); \
 545         } \
 546         snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
 547         \
 548         VHOST_LOG_DATA(DEBUG, "%s", packet); \
 549 } while (0)
 550 #else
 551 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
 552 #endif
 553
 554 #define MAX_VHOST_DEVICE        1024
 555 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
 556
 557 #define VHOST_BINARY_SEARCH_THRESH 256
 558
 559 static __rte_always_inline int guest_page_addrcmp(const void *p1,
 560                                                 const void *p2)
 561 {
 562         const struct guest_page *page1 = (const struct guest_page *)p1;
 563         const struct guest_page *page2 = (const struct guest_page *)p2;
 564
 565         if (page1->guest_phys_addr > page2->guest_phys_addr)
 566                 return 1;
 567         if (page1->guest_phys_addr < page2->guest_phys_addr)
 568                 return -1;
 569
 570         return 0;
 571 }
 572
 573 /* Convert guest physical address to host physical address */
 574 static __rte_always_inline rte_iova_t
 575 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
 576 {
 577         uint32_t i;
 578         struct guest_page *page;
 579         struct guest_page key;
 580
 581         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
 582                 key.guest_phys_addr = gpa;
 583                 page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages,
 584                                sizeof(struct guest_page), guest_page_addrcmp);
 585                 if (page) {
 586                         if (gpa + size < page->guest_phys_addr + page->size)
 587                                 return gpa - page->guest_phys_addr +
 588                                         page->host_phys_addr;
 589                 }
 590         } else {
 591                 for (i = 0; i < dev->nr_guest_pages; i++) {
 592                         page = &dev->guest_pages[i];
 593
 594                         if (gpa >= page->guest_phys_addr &&
 595                             gpa + size < page->guest_phys_addr +
 596                             page->size)
 597                                 return gpa - page->guest_phys_addr +
 598                                        page->host_phys_addr;
 599                 }
 600         }
 601
 602         return 0;
 603 }
 604
 605 static __rte_always_inline uint64_t
 606 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
 607 {
 608         struct rte_vhost_mem_region *r;
 609         uint32_t i;
 610
 611         if (unlikely(!dev || !dev->mem))
 612                 return 0;
 613
 614         for (i = 0; i < dev->mem->nregions; i++) {
 615                 r = &dev->mem->regions[i];
 616
 617                 if (vva >= r->host_user_addr &&
 618                     vva + len <  r->host_user_addr + r->size) {
 619                         return r->guest_phys_addr + vva - r->host_user_addr;
 620                 }
 621         }
 622         return 0;
 623 }
 624
 625 static __rte_always_inline struct virtio_net *
 626 get_device(int vid)
 627 {
 628         struct virtio_net *dev = vhost_devices[vid];
 629
 630         if (unlikely(!dev)) {
 631                 VHOST_LOG_CONFIG(ERR,
 632                         "(%d) device not found.\n", vid);
 633         }
 634
 635         return dev;
 636 }
 637
 638 int vhost_new_device(void);
 639 void cleanup_device(struct virtio_net *dev, int destroy);
 640 void reset_device(struct virtio_net *dev);
 641 void vhost_destroy_device(int);
 642 void vhost_destroy_device_notify(struct virtio_net *dev);
 643
 644 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
 645 void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
 646 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
 647
 648 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
 649
 650 void vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *dev);
 651
 652 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
 653 void vhost_enable_dequeue_zero_copy(int vid);
 654 void vhost_set_builtin_virtio_net(int vid, bool enable);
 655 void vhost_enable_extbuf(int vid);
 656 void vhost_enable_linearbuf(int vid);
 657
 658 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
 659
 660 /*
 661  * Backend-specific cleanup.
 662  *
 663  * TODO: fix it; we have one backend now
 664  */
 665 void vhost_backend_cleanup(struct virtio_net *dev);
 666
 667 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 668                         uint64_t iova, uint64_t *len, uint8_t perm);
 669 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
 670                         struct vhost_virtqueue *vq,
 671                         uint64_t desc_addr, uint64_t desc_len);
 672 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 673 uint64_t translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
 674                 uint64_t log_addr);
 675 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 676
 677 static __rte_always_inline uint64_t
 678 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 679                         uint64_t iova, uint64_t *len, uint8_t perm)
 680 {
 681         if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 682                 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
 683
 684         return __vhost_iova_to_vva(dev, vq, iova, len, perm);
 685 }
 686
 687 #define vhost_avail_event(vr) \
 688         (*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
 689 #define vhost_used_event(vr) \
 690         (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
 691
 692 /*
 693  * The following is used with VIRTIO_RING_F_EVENT_IDX.
 694  * Assuming a given event_idx value from the other size, if we have
 695  * just incremented index from old to new_idx, should we trigger an
 696  * event?
 697  */
 698 static __rte_always_inline int
 699 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
 700 {
 701         return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
 702 }
 703
 704 static __rte_always_inline void
 705 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 706 {
 707         /* Flush used->idx update before we read avail->flags. */
 708         rte_smp_mb();
 709
 710         /* Don't kick guest if we don't reach index specified by guest. */
 711         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
 712                 uint16_t old = vq->signalled_used;
 713                 uint16_t new = vq->async_pkts_inflight_n ?
 714                                         vq->used->idx:vq->last_used_idx;
 715                 bool signalled_used_valid = vq->signalled_used_valid;
 716
 717                 vq->signalled_used = new;
 718                 vq->signalled_used_valid = true;
 719
 720                 VHOST_LOG_DATA(DEBUG, "%s: used_event_idx=%d, old=%d, new=%d\n",
 721                         __func__,
 722                         vhost_used_event(vq),
 723                         old, new);
 724
 725                 if ((vhost_need_event(vhost_used_event(vq), new, old) &&
 726                                         (vq->callfd >= 0)) ||
 727                                 unlikely(!signalled_used_valid)) {
 728                         eventfd_write(vq->callfd, (eventfd_t) 1);
 729                         if (dev->notify_ops->guest_notified)
 730                                 dev->notify_ops->guest_notified(dev->vid);
 731                 }
 732         } else {
 733                 /* Kick the guest if necessary. */
 734                 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
 735                                 && (vq->callfd >= 0)) {
 736                         eventfd_write(vq->callfd, (eventfd_t)1);
 737                         if (dev->notify_ops->guest_notified)
 738                                 dev->notify_ops->guest_notified(dev->vid);
 739                 }
 740         }
 741 }
 742
 743 static __rte_always_inline void
 744 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
 745 {
 746         uint16_t old, new, off, off_wrap;
 747         bool signalled_used_valid, kick = false;
 748
 749         /* Flush used desc update. */
 750         rte_smp_mb();
 751
 752         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
 753                 if (vq->driver_event->flags !=
 754                                 VRING_EVENT_F_DISABLE)
 755                         kick = true;
 756                 goto kick;
 757         }
 758
 759         old = vq->signalled_used;
 760         new = vq->last_used_idx;
 761         vq->signalled_used = new;
 762         signalled_used_valid = vq->signalled_used_valid;
 763         vq->signalled_used_valid = true;
 764
 765         if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
 766                 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
 767                         kick = true;
 768                 goto kick;
 769         }
 770
 771         if (unlikely(!signalled_used_valid)) {
 772                 kick = true;
 773                 goto kick;
 774         }
 775
 776         rte_smp_rmb();
 777
 778         off_wrap = vq->driver_event->off_wrap;
 779         off = off_wrap & ~(1 << 15);
 780
 781         if (new <= old)
 782                 old -= vq->size;
 783
 784         if (vq->used_wrap_counter != off_wrap >> 15)
 785                 off -= vq->size;
 786
 787         if (vhost_need_event(off, new, old))
 788                 kick = true;
 789 kick:
 790         if (kick) {
 791                 eventfd_write(vq->callfd, (eventfd_t)1);
 792                 if (dev->notify_ops->guest_notified)
 793                         dev->notify_ops->guest_notified(dev->vid);
 794         }
 795 }
 796
 797 static __rte_always_inline void
 798 free_ind_table(void *idesc)
 799 {
 800         rte_free(idesc);
 801 }
 802
 803 static __rte_always_inline void
 804 restore_mbuf(struct rte_mbuf *m)
 805 {
 806         uint32_t mbuf_size, priv_size;
 807
 808         while (m) {
 809                 priv_size = rte_pktmbuf_priv_size(m->pool);
 810                 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
 811                 /* start of buffer is after mbuf structure and priv data */
 812
 813                 m->buf_addr = (char *)m + mbuf_size;
 814                 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
 815                 m = m->next;
 816         }
 817 }
 818
 819 static __rte_always_inline bool
 820 mbuf_is_consumed(struct rte_mbuf *m)
 821 {
 822         while (m) {
 823                 if (rte_mbuf_refcnt_read(m) > 1)
 824                         return false;
 825                 m = m->next;
 826         }
 827
 828         return true;
 829 }
 830
 831 static __rte_always_inline void
 832 put_zmbuf(struct zcopy_mbuf *zmbuf)
 833 {
 834         zmbuf->in_use = 0;
 835 }
 836
 837 #endif /* _VHOST_NET_CDEV_H_ */