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