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 VHOST_MAX_ASYNC_IT (MAX_PKT_BURST * 2)
  50 #define VHOST_MAX_ASYNC_VEC (BUF_VECTOR_MAX * 4)
  51
  52 #define PACKED_DESC_ENQUEUE_USED_FLAG(w)        \
  53         ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED | VRING_DESC_F_WRITE) : \
  54                 VRING_DESC_F_WRITE)
  55 #define PACKED_DESC_DEQUEUE_USED_FLAG(w)        \
  56         ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED) : 0x0)
  57 #define PACKED_DESC_SINGLE_DEQUEUE_FLAG (VRING_DESC_F_NEXT | \
  58                                          VRING_DESC_F_INDIRECT)
  59
  60 #define PACKED_BATCH_SIZE (RTE_CACHE_LINE_SIZE / \
  61                             sizeof(struct vring_packed_desc))
  62 #define PACKED_BATCH_MASK (PACKED_BATCH_SIZE - 1)
  63
  64 #ifdef VHOST_GCC_UNROLL_PRAGMA
  65 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("GCC unroll 4") \
  66         for (iter = val; iter < size; iter++)
  67 #endif
  68
  69 #ifdef VHOST_CLANG_UNROLL_PRAGMA
  70 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll 4") \
  71         for (iter = val; iter < size; iter++)
  72 #endif
  73
  74 #ifdef VHOST_ICC_UNROLL_PRAGMA
  75 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll (4)") \
  76         for (iter = val; iter < size; iter++)
  77 #endif
  78
  79 #ifndef vhost_for_each_try_unroll
  80 #define vhost_for_each_try_unroll(iter, val, num) \
  81         for (iter = val; iter < num; iter++)
  82 #endif
  83
  84 /**
  85  * Structure contains buffer address, length and descriptor index
  86  * from vring to do scatter RX.
  87  */
  88 struct buf_vector {
  89         uint64_t buf_iova;
  90         uint64_t buf_addr;
  91         uint32_t buf_len;
  92         uint32_t desc_idx;
  93 };
  94
  95 /*
  96  * Structure contains the info for each batched memory copy.
  97  */
  98 struct batch_copy_elem {
  99         void *dst;
 100         void *src;
 101         uint32_t len;
 102         uint64_t log_addr;
 103 };
 104
 105 /*
 106  * Structure that contains the info for batched dirty logging.
 107  */
 108 struct log_cache_entry {
 109         uint32_t offset;
 110         unsigned long val;
 111 };
 112
 113 struct vring_used_elem_packed {
 114         uint16_t id;
 115         uint16_t flags;
 116         uint32_t len;
 117         uint32_t count;
 118 };
 119
 120 /**
 121  * Structure contains variables relevant to RX/TX virtqueues.
 122  */
 123 struct vhost_virtqueue {
 124         union {
 125                 struct vring_desc       *desc;
 126                 struct vring_packed_desc   *desc_packed;
 127         };
 128         union {
 129                 struct vring_avail      *avail;
 130                 struct vring_packed_desc_event *driver_event;
 131         };
 132         union {
 133                 struct vring_used       *used;
 134                 struct vring_packed_desc_event *device_event;
 135         };
 136         uint16_t                size;
 137
 138         uint16_t                last_avail_idx;
 139         uint16_t                last_used_idx;
 140         /* Last used index we notify to front end. */
 141         uint16_t                signalled_used;
 142         bool                    signalled_used_valid;
 143 #define VIRTIO_INVALID_EVENTFD          (-1)
 144 #define VIRTIO_UNINITIALIZED_EVENTFD    (-2)
 145
 146         bool                    enabled;
 147         bool                    access_ok;
 148         bool                    ready;
 149
 150         rte_spinlock_t          access_lock;
 151
 152
 153         union {
 154                 struct vring_used_elem  *shadow_used_split;
 155                 struct vring_used_elem_packed *shadow_used_packed;
 156         };
 157         uint16_t                shadow_used_idx;
 158         /* Record packed ring enqueue latest desc cache aligned index */
 159         uint16_t                shadow_aligned_idx;
 160         /* Record packed ring first dequeue desc index */
 161         uint16_t                shadow_last_used_idx;
 162
 163         uint16_t                batch_copy_nb_elems;
 164         struct batch_copy_elem  *batch_copy_elems;
 165         bool                    used_wrap_counter;
 166         bool                    avail_wrap_counter;
 167
 168         /* Physical address of used ring, for logging */
 169         uint16_t                log_cache_nb_elem;
 170         uint64_t                log_guest_addr;
 171         struct log_cache_entry  *log_cache;
 172
 173         rte_rwlock_t    iotlb_lock;
 174         rte_rwlock_t    iotlb_pending_lock;
 175         struct rte_mempool *iotlb_pool;
 176         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
 177         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
 178         int                             iotlb_cache_nr;
 179
 180         /* Used to notify the guest (trigger interrupt) */
 181         int                     callfd;
 182         /* Currently unused as polling mode is enabled */
 183         int                     kickfd;
 184
 185         /* inflight share memory info */
 186         union {
 187                 struct rte_vhost_inflight_info_split *inflight_split;
 188                 struct rte_vhost_inflight_info_packed *inflight_packed;
 189         };
 190         struct rte_vhost_resubmit_info *resubmit_inflight;
 191         uint64_t                global_counter;
 192
 193         /* operation callbacks for async dma */
 194         struct rte_vhost_async_channel_ops      async_ops;
 195
 196         struct rte_vhost_iov_iter *it_pool;
 197         struct iovec *vec_pool;
 198
 199         /* async data transfer status */
 200         struct async_inflight_info *async_pkts_info;
 201         uint16_t        async_pkts_idx;
 202         uint16_t        async_pkts_inflight_n;
 203         uint16_t        async_last_pkts_n;
 204         struct vring_used_elem  *async_descs_split;
 205         uint16_t async_desc_idx;
 206         uint16_t last_async_desc_idx;
 207
 208         /* vq async features */
 209         bool            async_inorder;
 210         bool            async_registered;
 211         uint16_t        async_threshold;
 212
 213         int                     notif_enable;
 214 #define VIRTIO_UNINITIALIZED_NOTIF      (-1)
 215
 216         struct vhost_vring_addr ring_addrs;
 217 } __rte_cache_aligned;
 218
 219 /* Virtio device status as per Virtio specification */
 220 #define VIRTIO_DEVICE_STATUS_RESET              0x00
 221 #define VIRTIO_DEVICE_STATUS_ACK                0x01
 222 #define VIRTIO_DEVICE_STATUS_DRIVER             0x02
 223 #define VIRTIO_DEVICE_STATUS_DRIVER_OK          0x04
 224 #define VIRTIO_DEVICE_STATUS_FEATURES_OK        0x08
 225 #define VIRTIO_DEVICE_STATUS_DEV_NEED_RESET     0x40
 226 #define VIRTIO_DEVICE_STATUS_FAILED             0x80
 227
 228 #define VHOST_MAX_VRING                 0x100
 229 #define VHOST_MAX_QUEUE_PAIRS           0x80
 230
 231 /* Declare IOMMU related bits for older kernels */
 232 #ifndef VIRTIO_F_IOMMU_PLATFORM
 233
 234 #define VIRTIO_F_IOMMU_PLATFORM 33
 235
 236 struct vhost_iotlb_msg {
 237         __u64 iova;
 238         __u64 size;
 239         __u64 uaddr;
 240 #define VHOST_ACCESS_RO      0x1
 241 #define VHOST_ACCESS_WO      0x2
 242 #define VHOST_ACCESS_RW      0x3
 243         __u8 perm;
 244 #define VHOST_IOTLB_MISS           1
 245 #define VHOST_IOTLB_UPDATE         2
 246 #define VHOST_IOTLB_INVALIDATE     3
 247 #define VHOST_IOTLB_ACCESS_FAIL    4
 248         __u8 type;
 249 };
 250
 251 #define VHOST_IOTLB_MSG 0x1
 252
 253 struct vhost_msg {
 254         int type;
 255         union {
 256                 struct vhost_iotlb_msg iotlb;
 257                 __u8 padding[64];
 258         };
 259 };
 260 #endif
 261
 262 /*
 263  * Define virtio 1.0 for older kernels
 264  */
 265 #ifndef VIRTIO_F_VERSION_1
 266  #define VIRTIO_F_VERSION_1 32
 267 #endif
 268
 269 /* Declare packed ring related bits for older kernels */
 270 #ifndef VIRTIO_F_RING_PACKED
 271
 272 #define VIRTIO_F_RING_PACKED 34
 273
 274 struct vring_packed_desc {
 275         uint64_t addr;
 276         uint32_t len;
 277         uint16_t id;
 278         uint16_t flags;
 279 };
 280
 281 struct vring_packed_desc_event {
 282         uint16_t off_wrap;
 283         uint16_t flags;
 284 };
 285 #endif
 286
 287 /*
 288  * Declare below packed ring defines unconditionally
 289  * as Kernel header might use different names.
 290  */
 291 #define VRING_DESC_F_AVAIL      (1ULL << 7)
 292 #define VRING_DESC_F_USED       (1ULL << 15)
 293
 294 #define VRING_EVENT_F_ENABLE 0x0
 295 #define VRING_EVENT_F_DISABLE 0x1
 296 #define VRING_EVENT_F_DESC 0x2
 297
 298 /*
 299  * Available and used descs are in same order
 300  */
 301 #ifndef VIRTIO_F_IN_ORDER
 302 #define VIRTIO_F_IN_ORDER      35
 303 #endif
 304
 305 /* Features supported by this builtin vhost-user net driver. */
 306 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
 307                                 (1ULL << VIRTIO_F_ANY_LAYOUT) | \
 308                                 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
 309                                 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
 310                                 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
 311                                 (1ULL << VIRTIO_NET_F_MQ)      | \
 312                                 (1ULL << VIRTIO_F_VERSION_1)   | \
 313                                 (1ULL << VHOST_F_LOG_ALL)      | \
 314                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
 315                                 (1ULL << VIRTIO_NET_F_GSO) | \
 316                                 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
 317                                 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
 318                                 (1ULL << VIRTIO_NET_F_HOST_UFO) | \
 319                                 (1ULL << VIRTIO_NET_F_HOST_ECN) | \
 320                                 (1ULL << VIRTIO_NET_F_CSUM)    | \
 321                                 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
 322                                 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
 323                                 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
 324                                 (1ULL << VIRTIO_NET_F_GUEST_UFO) | \
 325                                 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
 326                                 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
 327                                 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \
 328                                 (1ULL << VIRTIO_NET_F_MTU)  | \
 329                                 (1ULL << VIRTIO_F_IN_ORDER) | \
 330                                 (1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
 331                                 (1ULL << VIRTIO_F_RING_PACKED))
 332
 333
 334 struct guest_page {
 335         uint64_t guest_phys_addr;
 336         uint64_t host_phys_addr;
 337         uint64_t size;
 338 };
 339
 340 struct inflight_mem_info {
 341         int             fd;
 342         void            *addr;
 343         uint64_t        size;
 344 };
 345
 346 /**
 347  * Device structure contains all configuration information relating
 348  * to the device.
 349  */
 350 struct virtio_net {
 351         /* Frontend (QEMU) memory and memory region information */
 352         struct rte_vhost_memory *mem;
 353         uint64_t                features;
 354         uint64_t                protocol_features;
 355         int                     vid;
 356         uint32_t                flags;
 357         uint16_t                vhost_hlen;
 358         /* to tell if we need broadcast rarp packet */
 359         int16_t                 broadcast_rarp;
 360         uint32_t                nr_vring;
 361         int                     async_copy;
 362         int                     extbuf;
 363         int                     linearbuf;
 364         struct vhost_virtqueue  *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
 365         struct inflight_mem_info *inflight_info;
 366 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
 367         char                    ifname[IF_NAME_SZ];
 368         uint64_t                log_size;
 369         uint64_t                log_base;
 370         uint64_t                log_addr;
 371         struct rte_ether_addr   mac;
 372         uint16_t                mtu;
 373         uint8_t                 status;
 374
 375         struct vhost_device_ops const *notify_ops;
 376
 377         uint32_t                nr_guest_pages;
 378         uint32_t                max_guest_pages;
 379         struct guest_page       *guest_pages;
 380
 381         int                     slave_req_fd;
 382         rte_spinlock_t          slave_req_lock;
 383
 384         int                     postcopy_ufd;
 385         int                     postcopy_listening;
 386
 387         struct rte_vdpa_device *vdpa_dev;
 388
 389         /* context data for the external message handlers */
 390         void                    *extern_data;
 391         /* pre and post vhost user message handlers for the device */
 392         struct rte_vhost_user_extern_ops extern_ops;
 393 } __rte_cache_aligned;
 394
 395 static __rte_always_inline bool
 396 vq_is_packed(struct virtio_net *dev)
 397 {
 398         return dev->features & (1ull << VIRTIO_F_RING_PACKED);
 399 }
 400
 401 static inline bool
 402 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
 403 {
 404         uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
 405
 406         return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
 407                 wrap_counter != !!(flags & VRING_DESC_F_USED);
 408 }
 409
 410 static inline void
 411 vq_inc_last_used_packed(struct vhost_virtqueue *vq, uint16_t num)
 412 {
 413         vq->last_used_idx += num;
 414         if (vq->last_used_idx >= vq->size) {
 415                 vq->used_wrap_counter ^= 1;
 416                 vq->last_used_idx -= vq->size;
 417         }
 418 }
 419
 420 static inline void
 421 vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
 422 {
 423         vq->last_avail_idx += num;
 424         if (vq->last_avail_idx >= vq->size) {
 425                 vq->avail_wrap_counter ^= 1;
 426                 vq->last_avail_idx -= vq->size;
 427         }
 428 }
 429
 430 void __vhost_log_cache_write(struct virtio_net *dev,
 431                 struct vhost_virtqueue *vq,
 432                 uint64_t addr, uint64_t len);
 433 void __vhost_log_cache_write_iova(struct virtio_net *dev,
 434                 struct vhost_virtqueue *vq,
 435                 uint64_t iova, uint64_t len);
 436 void __vhost_log_cache_sync(struct virtio_net *dev,
 437                 struct vhost_virtqueue *vq);
 438 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
 439 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 440                             uint64_t iova, uint64_t len);
 441
 442 static __rte_always_inline void
 443 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
 444 {
 445         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 446                 __vhost_log_write(dev, addr, len);
 447 }
 448
 449 static __rte_always_inline void
 450 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
 451 {
 452         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 453                 __vhost_log_cache_sync(dev, vq);
 454 }
 455
 456 static __rte_always_inline void
 457 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
 458                         uint64_t addr, uint64_t len)
 459 {
 460         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 461                 __vhost_log_cache_write(dev, vq, addr, len);
 462 }
 463
 464 static __rte_always_inline void
 465 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 466                         uint64_t offset, uint64_t len)
 467 {
 468         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
 469                 if (unlikely(vq->log_guest_addr == 0))
 470                         return;
 471                 __vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset,
 472                                         len);
 473         }
 474 }
 475
 476 static __rte_always_inline void
 477 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 478                      uint64_t offset, uint64_t len)
 479 {
 480         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
 481                 if (unlikely(vq->log_guest_addr == 0))
 482                         return;
 483                 __vhost_log_write(dev, vq->log_guest_addr + offset, len);
 484         }
 485 }
 486
 487 static __rte_always_inline void
 488 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 489                            uint64_t iova, uint64_t len)
 490 {
 491         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 492                 return;
 493
 494         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 495                 __vhost_log_cache_write_iova(dev, vq, iova, len);
 496         else
 497                 __vhost_log_cache_write(dev, vq, iova, len);
 498 }
 499
 500 static __rte_always_inline void
 501 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 502                            uint64_t iova, uint64_t len)
 503 {
 504         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 505                 return;
 506
 507         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 508                 __vhost_log_write_iova(dev, vq, iova, len);
 509         else
 510                 __vhost_log_write(dev, iova, len);
 511 }
 512
 513 extern int vhost_config_log_level;
 514 extern int vhost_data_log_level;
 515
 516 #define VHOST_LOG_CONFIG(level, fmt, args...)                   \
 517         rte_log(RTE_LOG_ ## level, vhost_config_log_level,      \
 518                 "VHOST_CONFIG: " fmt, ##args)
 519
 520 #define VHOST_LOG_DATA(level, fmt, args...) \
 521         (void)((RTE_LOG_ ## level <= RTE_LOG_DP_LEVEL) ?        \
 522          rte_log(RTE_LOG_ ## level,  vhost_data_log_level,      \
 523                 "VHOST_DATA : " fmt, ##args) :                  \
 524          0)
 525
 526 #ifdef RTE_LIBRTE_VHOST_DEBUG
 527 #define VHOST_MAX_PRINT_BUFF 6072
 528 #define PRINT_PACKET(device, addr, size, header) do { \
 529         char *pkt_addr = (char *)(addr); \
 530         unsigned int index; \
 531         char packet[VHOST_MAX_PRINT_BUFF]; \
 532         \
 533         if ((header)) \
 534                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
 535         else \
 536                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
 537         for (index = 0; index < (size); index++) { \
 538                 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
 539                         "%02hhx ", pkt_addr[index]); \
 540         } \
 541         snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
 542         \
 543         VHOST_LOG_DATA(DEBUG, "%s", packet); \
 544 } while (0)
 545 #else
 546 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
 547 #endif
 548
 549 #define MAX_VHOST_DEVICE        1024
 550 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
 551
 552 #define VHOST_BINARY_SEARCH_THRESH 256
 553
 554 static __rte_always_inline int guest_page_addrcmp(const void *p1,
 555                                                 const void *p2)
 556 {
 557         const struct guest_page *page1 = (const struct guest_page *)p1;
 558         const struct guest_page *page2 = (const struct guest_page *)p2;
 559
 560         if (page1->guest_phys_addr > page2->guest_phys_addr)
 561                 return 1;
 562         if (page1->guest_phys_addr < page2->guest_phys_addr)
 563                 return -1;
 564
 565         return 0;
 566 }
 567
 568 static __rte_always_inline rte_iova_t
 569 gpa_to_first_hpa(struct virtio_net *dev, uint64_t gpa,
 570         uint64_t gpa_size, uint64_t *hpa_size)
 571 {
 572         uint32_t i;
 573         struct guest_page *page;
 574         struct guest_page key;
 575
 576         *hpa_size = gpa_size;
 577         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
 578                 key.guest_phys_addr = gpa & ~(dev->guest_pages[0].size - 1);
 579                 page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages,
 580                                sizeof(struct guest_page), guest_page_addrcmp);
 581                 if (page) {
 582                         if (gpa + gpa_size <=
 583                                         page->guest_phys_addr + page->size) {
 584                                 return gpa - page->guest_phys_addr +
 585                                         page->host_phys_addr;
 586                         } else if (gpa < page->guest_phys_addr +
 587                                                 page->size) {
 588                                 *hpa_size = page->guest_phys_addr +
 589                                         page->size - gpa;
 590                                 return gpa - page->guest_phys_addr +
 591                                         page->host_phys_addr;
 592                         }
 593                 }
 594         } else {
 595                 for (i = 0; i < dev->nr_guest_pages; i++) {
 596                         page = &dev->guest_pages[i];
 597
 598                         if (gpa >= page->guest_phys_addr) {
 599                                 if (gpa + gpa_size <=
 600                                         page->guest_phys_addr + page->size) {
 601                                         return gpa - page->guest_phys_addr +
 602                                                 page->host_phys_addr;
 603                                 } else if (gpa < page->guest_phys_addr +
 604                                                         page->size) {
 605                                         *hpa_size = page->guest_phys_addr +
 606                                                 page->size - gpa;
 607                                         return gpa - page->guest_phys_addr +
 608                                                 page->host_phys_addr;
 609                                 }
 610                         }
 611                 }
 612         }
 613
 614         *hpa_size = 0;
 615         return 0;
 616 }
 617
 618 /* Convert guest physical address to host physical address */
 619 static __rte_always_inline rte_iova_t
 620 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
 621 {
 622         rte_iova_t hpa;
 623         uint64_t hpa_size;
 624
 625         hpa = gpa_to_first_hpa(dev, gpa, size, &hpa_size);
 626         return hpa_size == size ? hpa : 0;
 627 }
 628
 629 static __rte_always_inline uint64_t
 630 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
 631 {
 632         struct rte_vhost_mem_region *r;
 633         uint32_t i;
 634
 635         if (unlikely(!dev || !dev->mem))
 636                 return 0;
 637
 638         for (i = 0; i < dev->mem->nregions; i++) {
 639                 r = &dev->mem->regions[i];
 640
 641                 if (vva >= r->host_user_addr &&
 642                     vva + len <  r->host_user_addr + r->size) {
 643                         return r->guest_phys_addr + vva - r->host_user_addr;
 644                 }
 645         }
 646         return 0;
 647 }
 648
 649 static __rte_always_inline struct virtio_net *
 650 get_device(int vid)
 651 {
 652         struct virtio_net *dev = vhost_devices[vid];
 653
 654         if (unlikely(!dev)) {
 655                 VHOST_LOG_CONFIG(ERR,
 656                         "(%d) device not found.\n", vid);
 657         }
 658
 659         return dev;
 660 }
 661
 662 int vhost_new_device(void);
 663 void cleanup_device(struct virtio_net *dev, int destroy);
 664 void reset_device(struct virtio_net *dev);
 665 void vhost_destroy_device(int);
 666 void vhost_destroy_device_notify(struct virtio_net *dev);
 667
 668 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
 669 void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
 670 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
 671
 672 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
 673
 674 void vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *dev);
 675
 676 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
 677 void vhost_set_builtin_virtio_net(int vid, bool enable);
 678 void vhost_enable_extbuf(int vid);
 679 void vhost_enable_linearbuf(int vid);
 680 int vhost_enable_guest_notification(struct virtio_net *dev,
 681                 struct vhost_virtqueue *vq, int enable);
 682
 683 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
 684
 685 /*
 686  * Backend-specific cleanup.
 687  *
 688  * TODO: fix it; we have one backend now
 689  */
 690 void vhost_backend_cleanup(struct virtio_net *dev);
 691
 692 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 693                         uint64_t iova, uint64_t *len, uint8_t perm);
 694 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
 695                         struct vhost_virtqueue *vq,
 696                         uint64_t desc_addr, uint64_t desc_len);
 697 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 698 uint64_t translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
 699                 uint64_t log_addr);
 700 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 701
 702 static __rte_always_inline uint64_t
 703 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 704                         uint64_t iova, uint64_t *len, uint8_t perm)
 705 {
 706         if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 707                 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
 708
 709         return __vhost_iova_to_vva(dev, vq, iova, len, perm);
 710 }
 711
 712 #define vhost_avail_event(vr) \
 713         (*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
 714 #define vhost_used_event(vr) \
 715         (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
 716
 717 /*
 718  * The following is used with VIRTIO_RING_F_EVENT_IDX.
 719  * Assuming a given event_idx value from the other size, if we have
 720  * just incremented index from old to new_idx, should we trigger an
 721  * event?
 722  */
 723 static __rte_always_inline int
 724 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
 725 {
 726         return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
 727 }
 728
 729 static __rte_always_inline void
 730 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 731 {
 732         /* Flush used->idx update before we read avail->flags. */
 733         rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
 734
 735         /* Don't kick guest if we don't reach index specified by guest. */
 736         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
 737                 uint16_t old = vq->signalled_used;
 738                 uint16_t new = vq->last_used_idx;
 739                 bool signalled_used_valid = vq->signalled_used_valid;
 740
 741                 vq->signalled_used = new;
 742                 vq->signalled_used_valid = true;
 743
 744                 VHOST_LOG_DATA(DEBUG, "%s: used_event_idx=%d, old=%d, new=%d\n",
 745                         __func__,
 746                         vhost_used_event(vq),
 747                         old, new);
 748
 749                 if ((vhost_need_event(vhost_used_event(vq), new, old) &&
 750                                         (vq->callfd >= 0)) ||
 751                                 unlikely(!signalled_used_valid)) {
 752                         eventfd_write(vq->callfd, (eventfd_t) 1);
 753                         if (dev->notify_ops->guest_notified)
 754                                 dev->notify_ops->guest_notified(dev->vid);
 755                 }
 756         } else {
 757                 /* Kick the guest if necessary. */
 758                 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
 759                                 && (vq->callfd >= 0)) {
 760                         eventfd_write(vq->callfd, (eventfd_t)1);
 761                         if (dev->notify_ops->guest_notified)
 762                                 dev->notify_ops->guest_notified(dev->vid);
 763                 }
 764         }
 765 }
 766
 767 static __rte_always_inline void
 768 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
 769 {
 770         uint16_t old, new, off, off_wrap;
 771         bool signalled_used_valid, kick = false;
 772
 773         /* Flush used desc update. */
 774         rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
 775
 776         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
 777                 if (vq->driver_event->flags !=
 778                                 VRING_EVENT_F_DISABLE)
 779                         kick = true;
 780                 goto kick;
 781         }
 782
 783         old = vq->signalled_used;
 784         new = vq->last_used_idx;
 785         vq->signalled_used = new;
 786         signalled_used_valid = vq->signalled_used_valid;
 787         vq->signalled_used_valid = true;
 788
 789         if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
 790                 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
 791                         kick = true;
 792                 goto kick;
 793         }
 794
 795         if (unlikely(!signalled_used_valid)) {
 796                 kick = true;
 797                 goto kick;
 798         }
 799
 800         rte_atomic_thread_fence(__ATOMIC_ACQUIRE);
 801
 802         off_wrap = vq->driver_event->off_wrap;
 803         off = off_wrap & ~(1 << 15);
 804
 805         if (new <= old)
 806                 old -= vq->size;
 807
 808         if (vq->used_wrap_counter != off_wrap >> 15)
 809                 off -= vq->size;
 810
 811         if (vhost_need_event(off, new, old))
 812                 kick = true;
 813 kick:
 814         if (kick) {
 815                 eventfd_write(vq->callfd, (eventfd_t)1);
 816                 if (dev->notify_ops->guest_notified)
 817                         dev->notify_ops->guest_notified(dev->vid);
 818         }
 819 }
 820
 821 static __rte_always_inline void
 822 free_ind_table(void *idesc)
 823 {
 824         rte_free(idesc);
 825 }
 826
 827 static __rte_always_inline void
 828 restore_mbuf(struct rte_mbuf *m)
 829 {
 830         uint32_t mbuf_size, priv_size;
 831
 832         while (m) {
 833                 priv_size = rte_pktmbuf_priv_size(m->pool);
 834                 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
 835                 /* start of buffer is after mbuf structure and priv data */
 836
 837                 m->buf_addr = (char *)m + mbuf_size;
 838                 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
 839                 m = m->next;
 840         }
 841 }
 842
 843 static __rte_always_inline bool
 844 mbuf_is_consumed(struct rte_mbuf *m)
 845 {
 846         while (m) {
 847                 if (rte_mbuf_refcnt_read(m) > 1)
 848                         return false;
 849                 m = m->next;
 850         }
 851
 852         return true;
 853 }
 854
 855 #endif /* _VHOST_NET_CDEV_H_ */