lib/vhost/vhost.h

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2018 Intel Corporation
   3  */
   4
   5 #ifndef _VHOST_NET_CDEV_H_
   6 #define _VHOST_NET_CDEV_H_
   7 #include <stdint.h>
   8 #include <stdio.h>
   9 #include <stdbool.h>
  10 #include <sys/types.h>
  11 #include <sys/queue.h>
  12 #include <unistd.h>
  13 #include <linux/vhost.h>
  14 #include <linux/virtio_net.h>
  15 #include <sys/socket.h>
  16 #include <linux/if.h>
  17
  18 #include <rte_log.h>
  19 #include <rte_ether.h>
  20 #include <rte_rwlock.h>
  21 #include <rte_malloc.h>
  22
  23 #include "rte_vhost.h"
  24 #include "rte_vdpa.h"
  25 #include "rte_vdpa_dev.h"
  26
  27 #include "rte_vhost_async.h"
  28
  29 /* Used to indicate that the device is running on a data core */
  30 #define VIRTIO_DEV_RUNNING ((uint32_t)1 << 0)
  31 /* Used to indicate that the device is ready to operate */
  32 #define VIRTIO_DEV_READY ((uint32_t)1 << 1)
  33 /* Used to indicate that the built-in vhost net device backend is enabled */
  34 #define VIRTIO_DEV_BUILTIN_VIRTIO_NET ((uint32_t)1 << 2)
  35 /* Used to indicate that the device has its own data path and configured */
  36 #define VIRTIO_DEV_VDPA_CONFIGURED ((uint32_t)1 << 3)
  37 /* Used to indicate that the feature negotiation failed */
  38 #define VIRTIO_DEV_FEATURES_FAILED ((uint32_t)1 << 4)
  39 /* Used to indicate that the virtio_net tx code should fill TX ol_flags */
  40 #define VIRTIO_DEV_LEGACY_OL_FLAGS ((uint32_t)1 << 5)
  41
  42 /* Backend value set by guest. */
  43 #define VIRTIO_DEV_STOPPED -1
  44
  45 #define BUF_VECTOR_MAX 256
  46
  47 #define VHOST_LOG_CACHE_NR 32
  48
  49 #define MAX_PKT_BURST 32
  50
  51 #define VHOST_MAX_ASYNC_IT (MAX_PKT_BURST * 2)
  52 #define VHOST_MAX_ASYNC_VEC (BUF_VECTOR_MAX * 4)
  53
  54 #define PACKED_DESC_ENQUEUE_USED_FLAG(w)        \
  55         ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED | VRING_DESC_F_WRITE) : \
  56                 VRING_DESC_F_WRITE)
  57 #define PACKED_DESC_DEQUEUE_USED_FLAG(w)        \
  58         ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED) : 0x0)
  59 #define PACKED_DESC_SINGLE_DEQUEUE_FLAG (VRING_DESC_F_NEXT | \
  60                                          VRING_DESC_F_INDIRECT)
  61
  62 #define PACKED_BATCH_SIZE (RTE_CACHE_LINE_SIZE / \
  63                             sizeof(struct vring_packed_desc))
  64 #define PACKED_BATCH_MASK (PACKED_BATCH_SIZE - 1)
  65
  66 #ifdef VHOST_GCC_UNROLL_PRAGMA
  67 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("GCC unroll 4") \
  68         for (iter = val; iter < size; iter++)
  69 #endif
  70
  71 #ifdef VHOST_CLANG_UNROLL_PRAGMA
  72 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll 4") \
  73         for (iter = val; iter < size; iter++)
  74 #endif
  75
  76 #ifdef VHOST_ICC_UNROLL_PRAGMA
  77 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll (4)") \
  78         for (iter = val; iter < size; iter++)
  79 #endif
  80
  81 #ifndef vhost_for_each_try_unroll
  82 #define vhost_for_each_try_unroll(iter, val, num) \
  83         for (iter = val; iter < num; iter++)
  84 #endif
  85
  86 /**
  87  * Structure contains buffer address, length and descriptor index
  88  * from vring to do scatter RX.
  89  */
  90 struct buf_vector {
  91         uint64_t buf_iova;
  92         uint64_t buf_addr;
  93         uint32_t buf_len;
  94         uint32_t desc_idx;
  95 };
  96
  97 /*
  98  * Structure contains the info for each batched memory copy.
  99  */
 100 struct batch_copy_elem {
 101         void *dst;
 102         void *src;
 103         uint32_t len;
 104         uint64_t log_addr;
 105 };
 106
 107 /*
 108  * Structure that contains the info for batched dirty logging.
 109  */
 110 struct log_cache_entry {
 111         uint32_t offset;
 112         unsigned long val;
 113 };
 114
 115 struct vring_used_elem_packed {
 116         uint16_t id;
 117         uint16_t flags;
 118         uint32_t len;
 119         uint32_t count;
 120 };
 121
 122 /**
 123  * Structure contains variables relevant to RX/TX virtqueues.
 124  */
 125 struct vhost_virtqueue {
 126         union {
 127                 struct vring_desc       *desc;
 128                 struct vring_packed_desc   *desc_packed;
 129         };
 130         union {
 131                 struct vring_avail      *avail;
 132                 struct vring_packed_desc_event *driver_event;
 133         };
 134         union {
 135                 struct vring_used       *used;
 136                 struct vring_packed_desc_event *device_event;
 137         };
 138         uint16_t                size;
 139
 140         uint16_t                last_avail_idx;
 141         uint16_t                last_used_idx;
 142         /* Last used index we notify to front end. */
 143         uint16_t                signalled_used;
 144         bool                    signalled_used_valid;
 145 #define VIRTIO_INVALID_EVENTFD          (-1)
 146 #define VIRTIO_UNINITIALIZED_EVENTFD    (-2)
 147
 148         bool                    enabled;
 149         bool                    access_ok;
 150         bool                    ready;
 151
 152         rte_spinlock_t          access_lock;
 153
 154
 155         union {
 156                 struct vring_used_elem  *shadow_used_split;
 157                 struct vring_used_elem_packed *shadow_used_packed;
 158         };
 159         uint16_t                shadow_used_idx;
 160         /* Record packed ring enqueue latest desc cache aligned index */
 161         uint16_t                shadow_aligned_idx;
 162         /* Record packed ring first dequeue desc index */
 163         uint16_t                shadow_last_used_idx;
 164
 165         uint16_t                batch_copy_nb_elems;
 166         struct batch_copy_elem  *batch_copy_elems;
 167         int                     numa_node;
 168         bool                    used_wrap_counter;
 169         bool                    avail_wrap_counter;
 170
 171         /* Physical address of used ring, for logging */
 172         uint16_t                log_cache_nb_elem;
 173         uint64_t                log_guest_addr;
 174         struct log_cache_entry  *log_cache;
 175
 176         rte_rwlock_t    iotlb_lock;
 177         rte_rwlock_t    iotlb_pending_lock;
 178         struct rte_mempool *iotlb_pool;
 179         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
 180         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
 181         int                             iotlb_cache_nr;
 182
 183         /* Used to notify the guest (trigger interrupt) */
 184         int                     callfd;
 185         /* Currently unused as polling mode is enabled */
 186         int                     kickfd;
 187
 188         /* inflight share memory info */
 189         union {
 190                 struct rte_vhost_inflight_info_split *inflight_split;
 191                 struct rte_vhost_inflight_info_packed *inflight_packed;
 192         };
 193         struct rte_vhost_resubmit_info *resubmit_inflight;
 194         uint64_t                global_counter;
 195
 196         /* operation callbacks for async dma */
 197         struct rte_vhost_async_channel_ops      async_ops;
 198
 199         struct rte_vhost_iov_iter *it_pool;
 200         struct iovec *vec_pool;
 201
 202         /* async data transfer status */
 203         struct async_inflight_info *async_pkts_info;
 204         uint16_t        async_pkts_idx;
 205         uint16_t        async_pkts_inflight_n;
 206         uint16_t        async_last_pkts_n;
 207         union {
 208                 struct vring_used_elem  *async_descs_split;
 209                 struct vring_used_elem_packed *async_buffers_packed;
 210         };
 211         union {
 212                 uint16_t async_desc_idx_split;
 213                 uint16_t async_buffer_idx_packed;
 214         };
 215         union {
 216                 uint16_t last_async_desc_idx_split;
 217                 uint16_t last_async_buffer_idx_packed;
 218         };
 219
 220         /* vq async features */
 221         bool            async_registered;
 222
 223         int                     notif_enable;
 224 #define VIRTIO_UNINITIALIZED_NOTIF      (-1)
 225
 226         struct vhost_vring_addr ring_addrs;
 227 } __rte_cache_aligned;
 228
 229 /* Virtio device status as per Virtio specification */
 230 #define VIRTIO_DEVICE_STATUS_RESET              0x00
 231 #define VIRTIO_DEVICE_STATUS_ACK                0x01
 232 #define VIRTIO_DEVICE_STATUS_DRIVER             0x02
 233 #define VIRTIO_DEVICE_STATUS_DRIVER_OK          0x04
 234 #define VIRTIO_DEVICE_STATUS_FEATURES_OK        0x08
 235 #define VIRTIO_DEVICE_STATUS_DEV_NEED_RESET     0x40
 236 #define VIRTIO_DEVICE_STATUS_FAILED             0x80
 237
 238 #define VHOST_MAX_VRING                 0x100
 239 #define VHOST_MAX_QUEUE_PAIRS           0x80
 240
 241 /* Declare IOMMU related bits for older kernels */
 242 #ifndef VIRTIO_F_IOMMU_PLATFORM
 243
 244 #define VIRTIO_F_IOMMU_PLATFORM 33
 245
 246 struct vhost_iotlb_msg {
 247         __u64 iova;
 248         __u64 size;
 249         __u64 uaddr;
 250 #define VHOST_ACCESS_RO      0x1
 251 #define VHOST_ACCESS_WO      0x2
 252 #define VHOST_ACCESS_RW      0x3
 253         __u8 perm;
 254 #define VHOST_IOTLB_MISS           1
 255 #define VHOST_IOTLB_UPDATE         2
 256 #define VHOST_IOTLB_INVALIDATE     3
 257 #define VHOST_IOTLB_ACCESS_FAIL    4
 258         __u8 type;
 259 };
 260
 261 #define VHOST_IOTLB_MSG 0x1
 262
 263 struct vhost_msg {
 264         int type;
 265         union {
 266                 struct vhost_iotlb_msg iotlb;
 267                 __u8 padding[64];
 268         };
 269 };
 270 #endif
 271
 272 /*
 273  * Define virtio 1.0 for older kernels
 274  */
 275 #ifndef VIRTIO_F_VERSION_1
 276  #define VIRTIO_F_VERSION_1 32
 277 #endif
 278
 279 /* Declare packed ring related bits for older kernels */
 280 #ifndef VIRTIO_F_RING_PACKED
 281
 282 #define VIRTIO_F_RING_PACKED 34
 283
 284 struct vring_packed_desc {
 285         uint64_t addr;
 286         uint32_t len;
 287         uint16_t id;
 288         uint16_t flags;
 289 };
 290
 291 struct vring_packed_desc_event {
 292         uint16_t off_wrap;
 293         uint16_t flags;
 294 };
 295 #endif
 296
 297 /*
 298  * Declare below packed ring defines unconditionally
 299  * as Kernel header might use different names.
 300  */
 301 #define VRING_DESC_F_AVAIL      (1ULL << 7)
 302 #define VRING_DESC_F_USED       (1ULL << 15)
 303
 304 #define VRING_EVENT_F_ENABLE 0x0
 305 #define VRING_EVENT_F_DISABLE 0x1
 306 #define VRING_EVENT_F_DESC 0x2
 307
 308 /*
 309  * Available and used descs are in same order
 310  */
 311 #ifndef VIRTIO_F_IN_ORDER
 312 #define VIRTIO_F_IN_ORDER      35
 313 #endif
 314
 315 /* Features supported by this builtin vhost-user net driver. */
 316 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
 317                                 (1ULL << VIRTIO_F_ANY_LAYOUT) | \
 318                                 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
 319                                 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
 320                                 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
 321                                 (1ULL << VIRTIO_NET_F_MQ)      | \
 322                                 (1ULL << VIRTIO_F_VERSION_1)   | \
 323                                 (1ULL << VHOST_F_LOG_ALL)      | \
 324                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
 325                                 (1ULL << VIRTIO_NET_F_GSO) | \
 326                                 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
 327                                 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
 328                                 (1ULL << VIRTIO_NET_F_HOST_UFO) | \
 329                                 (1ULL << VIRTIO_NET_F_HOST_ECN) | \
 330                                 (1ULL << VIRTIO_NET_F_CSUM)    | \
 331                                 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
 332                                 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
 333                                 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
 334                                 (1ULL << VIRTIO_NET_F_GUEST_UFO) | \
 335                                 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
 336                                 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
 337                                 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \
 338                                 (1ULL << VIRTIO_NET_F_MTU)  | \
 339                                 (1ULL << VIRTIO_F_IN_ORDER) | \
 340                                 (1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
 341                                 (1ULL << VIRTIO_F_RING_PACKED))
 342
 343
 344 struct guest_page {
 345         uint64_t guest_phys_addr;
 346         uint64_t host_phys_addr;
 347         uint64_t size;
 348 };
 349
 350 struct inflight_mem_info {
 351         int             fd;
 352         void            *addr;
 353         uint64_t        size;
 354 };
 355
 356 /**
 357  * Device structure contains all configuration information relating
 358  * to the device.
 359  */
 360 struct virtio_net {
 361         /* Frontend (QEMU) memory and memory region information */
 362         struct rte_vhost_memory *mem;
 363         uint64_t                features;
 364         uint64_t                protocol_features;
 365         int                     vid;
 366         uint32_t                flags;
 367         uint16_t                vhost_hlen;
 368         /* to tell if we need broadcast rarp packet */
 369         int16_t                 broadcast_rarp;
 370         uint32_t                nr_vring;
 371         int                     async_copy;
 372
 373         /* Record the dma map status for each region. */
 374         bool                    *async_map_status;
 375
 376         int                     extbuf;
 377         int                     linearbuf;
 378         struct vhost_virtqueue  *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
 379         struct inflight_mem_info *inflight_info;
 380 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
 381         char                    ifname[IF_NAME_SZ];
 382         uint64_t                log_size;
 383         uint64_t                log_base;
 384         uint64_t                log_addr;
 385         struct rte_ether_addr   mac;
 386         uint16_t                mtu;
 387         uint8_t                 status;
 388
 389         struct vhost_device_ops const *notify_ops;
 390
 391         uint32_t                nr_guest_pages;
 392         uint32_t                max_guest_pages;
 393         struct guest_page       *guest_pages;
 394
 395         int                     slave_req_fd;
 396         rte_spinlock_t          slave_req_lock;
 397
 398         int                     postcopy_ufd;
 399         int                     postcopy_listening;
 400
 401         struct rte_vdpa_device *vdpa_dev;
 402
 403         /* context data for the external message handlers */
 404         void                    *extern_data;
 405         /* pre and post vhost user message handlers for the device */
 406         struct rte_vhost_user_extern_ops extern_ops;
 407 } __rte_cache_aligned;
 408
 409 static __rte_always_inline bool
 410 vq_is_packed(struct virtio_net *dev)
 411 {
 412         return dev->features & (1ull << VIRTIO_F_RING_PACKED);
 413 }
 414
 415 static inline bool
 416 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
 417 {
 418         uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
 419
 420         return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
 421                 wrap_counter != !!(flags & VRING_DESC_F_USED);
 422 }
 423
 424 static inline void
 425 vq_inc_last_used_packed(struct vhost_virtqueue *vq, uint16_t num)
 426 {
 427         vq->last_used_idx += num;
 428         if (vq->last_used_idx >= vq->size) {
 429                 vq->used_wrap_counter ^= 1;
 430                 vq->last_used_idx -= vq->size;
 431         }
 432 }
 433
 434 static inline void
 435 vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
 436 {
 437         vq->last_avail_idx += num;
 438         if (vq->last_avail_idx >= vq->size) {
 439                 vq->avail_wrap_counter ^= 1;
 440                 vq->last_avail_idx -= vq->size;
 441         }
 442 }
 443
 444 void __vhost_log_cache_write(struct virtio_net *dev,
 445                 struct vhost_virtqueue *vq,
 446                 uint64_t addr, uint64_t len);
 447 void __vhost_log_cache_write_iova(struct virtio_net *dev,
 448                 struct vhost_virtqueue *vq,
 449                 uint64_t iova, uint64_t len);
 450 void __vhost_log_cache_sync(struct virtio_net *dev,
 451                 struct vhost_virtqueue *vq);
 452 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
 453 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 454                             uint64_t iova, uint64_t len);
 455
 456 static __rte_always_inline void
 457 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
 458 {
 459         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 460                 __vhost_log_write(dev, addr, len);
 461 }
 462
 463 static __rte_always_inline void
 464 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
 465 {
 466         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 467                 __vhost_log_cache_sync(dev, vq);
 468 }
 469
 470 static __rte_always_inline void
 471 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
 472                         uint64_t addr, uint64_t len)
 473 {
 474         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 475                 __vhost_log_cache_write(dev, vq, addr, len);
 476 }
 477
 478 static __rte_always_inline void
 479 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 480                         uint64_t offset, uint64_t len)
 481 {
 482         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
 483                 if (unlikely(vq->log_guest_addr == 0))
 484                         return;
 485                 __vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset,
 486                                         len);
 487         }
 488 }
 489
 490 static __rte_always_inline void
 491 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 492                      uint64_t offset, uint64_t len)
 493 {
 494         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
 495                 if (unlikely(vq->log_guest_addr == 0))
 496                         return;
 497                 __vhost_log_write(dev, vq->log_guest_addr + offset, len);
 498         }
 499 }
 500
 501 static __rte_always_inline void
 502 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 503                            uint64_t iova, uint64_t len)
 504 {
 505         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 506                 return;
 507
 508         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 509                 __vhost_log_cache_write_iova(dev, vq, iova, len);
 510         else
 511                 __vhost_log_cache_write(dev, vq, iova, len);
 512 }
 513
 514 static __rte_always_inline void
 515 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 516                            uint64_t iova, uint64_t len)
 517 {
 518         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 519                 return;
 520
 521         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 522                 __vhost_log_write_iova(dev, vq, iova, len);
 523         else
 524                 __vhost_log_write(dev, iova, len);
 525 }
 526
 527 extern int vhost_config_log_level;
 528 extern int vhost_data_log_level;
 529
 530 #define VHOST_LOG_CONFIG(level, fmt, args...)                   \
 531         rte_log(RTE_LOG_ ## level, vhost_config_log_level,      \
 532                 "VHOST_CONFIG: " fmt, ##args)
 533
 534 #define VHOST_LOG_DATA(level, fmt, args...) \
 535         (void)((RTE_LOG_ ## level <= RTE_LOG_DP_LEVEL) ?        \
 536          rte_log(RTE_LOG_ ## level,  vhost_data_log_level,      \
 537                 "VHOST_DATA : " fmt, ##args) :                  \
 538          0)
 539
 540 #ifdef RTE_LIBRTE_VHOST_DEBUG
 541 #define VHOST_MAX_PRINT_BUFF 6072
 542 #define PRINT_PACKET(device, addr, size, header) do { \
 543         char *pkt_addr = (char *)(addr); \
 544         unsigned int index; \
 545         char packet[VHOST_MAX_PRINT_BUFF]; \
 546         \
 547         if ((header)) \
 548                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
 549         else \
 550                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
 551         for (index = 0; index < (size); index++) { \
 552                 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
 553                         "%02hhx ", pkt_addr[index]); \
 554         } \
 555         snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
 556         \
 557         VHOST_LOG_DATA(DEBUG, "%s", packet); \
 558 } while (0)
 559 #else
 560 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
 561 #endif
 562
 563 #define MAX_VHOST_DEVICE        1024
 564 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
 565
 566 #define VHOST_BINARY_SEARCH_THRESH 256
 567
 568 static __rte_always_inline int guest_page_addrcmp(const void *p1,
 569                                                 const void *p2)
 570 {
 571         const struct guest_page *page1 = (const struct guest_page *)p1;
 572         const struct guest_page *page2 = (const struct guest_page *)p2;
 573
 574         if (page1->guest_phys_addr > page2->guest_phys_addr)
 575                 return 1;
 576         if (page1->guest_phys_addr < page2->guest_phys_addr)
 577                 return -1;
 578
 579         return 0;
 580 }
 581
 582 static __rte_always_inline rte_iova_t
 583 gpa_to_first_hpa(struct virtio_net *dev, uint64_t gpa,
 584         uint64_t gpa_size, uint64_t *hpa_size)
 585 {
 586         uint32_t i;
 587         struct guest_page *page;
 588         struct guest_page key;
 589
 590         *hpa_size = gpa_size;
 591         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
 592                 key.guest_phys_addr = gpa & ~(dev->guest_pages[0].size - 1);
 593                 page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages,
 594                                sizeof(struct guest_page), guest_page_addrcmp);
 595                 if (page) {
 596                         if (gpa + gpa_size <=
 597                                         page->guest_phys_addr + page->size) {
 598                                 return gpa - page->guest_phys_addr +
 599                                         page->host_phys_addr;
 600                         } else if (gpa < page->guest_phys_addr +
 601                                                 page->size) {
 602                                 *hpa_size = page->guest_phys_addr +
 603                                         page->size - gpa;
 604                                 return gpa - page->guest_phys_addr +
 605                                         page->host_phys_addr;
 606                         }
 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                                 if (gpa + gpa_size <=
 614                                         page->guest_phys_addr + page->size) {
 615                                         return gpa - page->guest_phys_addr +
 616                                                 page->host_phys_addr;
 617                                 } else if (gpa < page->guest_phys_addr +
 618                                                         page->size) {
 619                                         *hpa_size = page->guest_phys_addr +
 620                                                 page->size - gpa;
 621                                         return gpa - page->guest_phys_addr +
 622                                                 page->host_phys_addr;
 623                                 }
 624                         }
 625                 }
 626         }
 627
 628         *hpa_size = 0;
 629         return 0;
 630 }
 631
 632 /* Convert guest physical address to host physical address */
 633 static __rte_always_inline rte_iova_t
 634 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
 635 {
 636         rte_iova_t hpa;
 637         uint64_t hpa_size;
 638
 639         hpa = gpa_to_first_hpa(dev, gpa, size, &hpa_size);
 640         return hpa_size == size ? hpa : 0;
 641 }
 642
 643 static __rte_always_inline uint64_t
 644 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
 645 {
 646         struct rte_vhost_mem_region *r;
 647         uint32_t i;
 648
 649         if (unlikely(!dev || !dev->mem))
 650                 return 0;
 651
 652         for (i = 0; i < dev->mem->nregions; i++) {
 653                 r = &dev->mem->regions[i];
 654
 655                 if (vva >= r->host_user_addr &&
 656                     vva + len <  r->host_user_addr + r->size) {
 657                         return r->guest_phys_addr + vva - r->host_user_addr;
 658                 }
 659         }
 660         return 0;
 661 }
 662
 663 static __rte_always_inline struct virtio_net *
 664 get_device(int vid)
 665 {
 666         struct virtio_net *dev = vhost_devices[vid];
 667
 668         if (unlikely(!dev)) {
 669                 VHOST_LOG_CONFIG(ERR,
 670                         "(%d) device not found.\n", vid);
 671         }
 672
 673         return dev;
 674 }
 675
 676 int vhost_new_device(void);
 677 void cleanup_device(struct virtio_net *dev, int destroy);
 678 void reset_device(struct virtio_net *dev);
 679 void vhost_destroy_device(int);
 680 void vhost_destroy_device_notify(struct virtio_net *dev);
 681
 682 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
 683 void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
 684 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
 685
 686 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
 687
 688 void vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *dev);
 689
 690 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
 691 void vhost_setup_virtio_net(int vid, bool enable, bool legacy_ol_flags);
 692 void vhost_enable_extbuf(int vid);
 693 void vhost_enable_linearbuf(int vid);
 694 int vhost_enable_guest_notification(struct virtio_net *dev,
 695                 struct vhost_virtqueue *vq, int enable);
 696
 697 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
 698
 699 /*
 700  * Backend-specific cleanup.
 701  *
 702  * TODO: fix it; we have one backend now
 703  */
 704 void vhost_backend_cleanup(struct virtio_net *dev);
 705
 706 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 707                         uint64_t iova, uint64_t *len, uint8_t perm);
 708 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
 709                         struct vhost_virtqueue *vq,
 710                         uint64_t desc_addr, uint64_t desc_len);
 711 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 712 uint64_t translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
 713                 uint64_t log_addr);
 714 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 715
 716 static __rte_always_inline uint64_t
 717 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 718                         uint64_t iova, uint64_t *len, uint8_t perm)
 719 {
 720         if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 721                 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
 722
 723         return __vhost_iova_to_vva(dev, vq, iova, len, perm);
 724 }
 725
 726 #define vhost_avail_event(vr) \
 727         (*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
 728 #define vhost_used_event(vr) \
 729         (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
 730
 731 /*
 732  * The following is used with VIRTIO_RING_F_EVENT_IDX.
 733  * Assuming a given event_idx value from the other size, if we have
 734  * just incremented index from old to new_idx, should we trigger an
 735  * event?
 736  */
 737 static __rte_always_inline int
 738 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
 739 {
 740         return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
 741 }
 742
 743 static __rte_always_inline void
 744 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 745 {
 746         /* Flush used->idx update before we read avail->flags. */
 747         rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
 748
 749         /* Don't kick guest if we don't reach index specified by guest. */
 750         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
 751                 uint16_t old = vq->signalled_used;
 752                 uint16_t new = vq->last_used_idx;
 753                 bool signalled_used_valid = vq->signalled_used_valid;
 754
 755                 vq->signalled_used = new;
 756                 vq->signalled_used_valid = true;
 757
 758                 VHOST_LOG_DATA(DEBUG, "%s: used_event_idx=%d, old=%d, new=%d\n",
 759                         __func__,
 760                         vhost_used_event(vq),
 761                         old, new);
 762
 763                 if ((vhost_need_event(vhost_used_event(vq), new, old) &&
 764                                         (vq->callfd >= 0)) ||
 765                                 unlikely(!signalled_used_valid)) {
 766                         eventfd_write(vq->callfd, (eventfd_t) 1);
 767                         if (dev->notify_ops->guest_notified)
 768                                 dev->notify_ops->guest_notified(dev->vid);
 769                 }
 770         } else {
 771                 /* Kick the guest if necessary. */
 772                 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
 773                                 && (vq->callfd >= 0)) {
 774                         eventfd_write(vq->callfd, (eventfd_t)1);
 775                         if (dev->notify_ops->guest_notified)
 776                                 dev->notify_ops->guest_notified(dev->vid);
 777                 }
 778         }
 779 }
 780
 781 static __rte_always_inline void
 782 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
 783 {
 784         uint16_t old, new, off, off_wrap;
 785         bool signalled_used_valid, kick = false;
 786
 787         /* Flush used desc update. */
 788         rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
 789
 790         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
 791                 if (vq->driver_event->flags !=
 792                                 VRING_EVENT_F_DISABLE)
 793                         kick = true;
 794                 goto kick;
 795         }
 796
 797         old = vq->signalled_used;
 798         new = vq->last_used_idx;
 799         vq->signalled_used = new;
 800         signalled_used_valid = vq->signalled_used_valid;
 801         vq->signalled_used_valid = true;
 802
 803         if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
 804                 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
 805                         kick = true;
 806                 goto kick;
 807         }
 808
 809         if (unlikely(!signalled_used_valid)) {
 810                 kick = true;
 811                 goto kick;
 812         }
 813
 814         rte_atomic_thread_fence(__ATOMIC_ACQUIRE);
 815
 816         off_wrap = vq->driver_event->off_wrap;
 817         off = off_wrap & ~(1 << 15);
 818
 819         if (new <= old)
 820                 old -= vq->size;
 821
 822         if (vq->used_wrap_counter != off_wrap >> 15)
 823                 off -= vq->size;
 824
 825         if (vhost_need_event(off, new, old))
 826                 kick = true;
 827 kick:
 828         if (kick) {
 829                 eventfd_write(vq->callfd, (eventfd_t)1);
 830                 if (dev->notify_ops->guest_notified)
 831                         dev->notify_ops->guest_notified(dev->vid);
 832         }
 833 }
 834
 835 static __rte_always_inline void
 836 free_ind_table(void *idesc)
 837 {
 838         rte_free(idesc);
 839 }
 840
 841 static __rte_always_inline void
 842 restore_mbuf(struct rte_mbuf *m)
 843 {
 844         uint32_t mbuf_size, priv_size;
 845
 846         while (m) {
 847                 priv_size = rte_pktmbuf_priv_size(m->pool);
 848                 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
 849                 /* start of buffer is after mbuf structure and priv data */
 850
 851                 m->buf_addr = (char *)m + mbuf_size;
 852                 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
 853                 m = m->next;
 854         }
 855 }
 856
 857 static __rte_always_inline bool
 858 mbuf_is_consumed(struct rte_mbuf *m)
 859 {
 860         while (m) {
 861                 if (rte_mbuf_refcnt_read(m) > 1)
 862                         return false;
 863                 m = m->next;
 864         }
 865
 866         return true;
 867 }
 868
 869 #endif /* _VHOST_NET_CDEV_H_ */