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
  26 /* Used to indicate that the device is running on a data core */
  27 #define VIRTIO_DEV_RUNNING 1
  28 /* Used to indicate that the device is ready to operate */
  29 #define VIRTIO_DEV_READY 2
  30 /* Used to indicate that the built-in vhost net device backend is enabled */
  31 #define VIRTIO_DEV_BUILTIN_VIRTIO_NET 4
  32 /* Used to indicate that the device has its own data path and configured */
  33 #define VIRTIO_DEV_VDPA_CONFIGURED 8
  34
  35 /* Backend value set by guest. */
  36 #define VIRTIO_DEV_STOPPED -1
  37
  38 #define BUF_VECTOR_MAX 256
  39
  40 #define VHOST_LOG_CACHE_NR 32
  41
  42 #define PACKED_DESC_ENQUEUE_USED_FLAG(w)        \
  43         ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED | VRING_DESC_F_WRITE) : \
  44                 VRING_DESC_F_WRITE)
  45 #define PACKED_DESC_DEQUEUE_USED_FLAG(w)        \
  46         ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED) : 0x0)
  47 #define PACKED_DESC_SINGLE_DEQUEUE_FLAG (VRING_DESC_F_NEXT | \
  48                                          VRING_DESC_F_INDIRECT)
  49
  50 #define PACKED_BATCH_SIZE (RTE_CACHE_LINE_SIZE / \
  51                             sizeof(struct vring_packed_desc))
  52 #define PACKED_BATCH_MASK (PACKED_BATCH_SIZE - 1)
  53
  54 #ifdef VHOST_GCC_UNROLL_PRAGMA
  55 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("GCC unroll 4") \
  56         for (iter = val; iter < size; iter++)
  57 #endif
  58
  59 #ifdef VHOST_CLANG_UNROLL_PRAGMA
  60 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll 4") \
  61         for (iter = val; iter < size; iter++)
  62 #endif
  63
  64 #ifdef VHOST_ICC_UNROLL_PRAGMA
  65 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll (4)") \
  66         for (iter = val; iter < size; iter++)
  67 #endif
  68
  69 #ifndef vhost_for_each_try_unroll
  70 #define vhost_for_each_try_unroll(iter, val, num) \
  71         for (iter = val; iter < num; iter++)
  72 #endif
  73
  74 /**
  75  * Structure contains buffer address, length and descriptor index
  76  * from vring to do scatter RX.
  77  */
  78 struct buf_vector {
  79         uint64_t buf_iova;
  80         uint64_t buf_addr;
  81         uint32_t buf_len;
  82         uint32_t desc_idx;
  83 };
  84
  85 /*
  86  * A structure to hold some fields needed in zero copy code path,
  87  * mainly for associating an mbuf with the right desc_idx.
  88  */
  89 struct zcopy_mbuf {
  90         struct rte_mbuf *mbuf;
  91         uint32_t desc_idx;
  92         uint16_t desc_count;
  93         uint16_t in_use;
  94
  95         TAILQ_ENTRY(zcopy_mbuf) next;
  96 };
  97 TAILQ_HEAD(zcopy_mbuf_list, zcopy_mbuf);
  98
  99 /*
 100  * Structure contains the info for each batched memory copy.
 101  */
 102 struct batch_copy_elem {
 103         void *dst;
 104         void *src;
 105         uint32_t len;
 106         uint64_t log_addr;
 107 };
 108
 109 /*
 110  * Structure that contains the info for batched dirty logging.
 111  */
 112 struct log_cache_entry {
 113         uint32_t offset;
 114         unsigned long val;
 115 };
 116
 117 struct vring_used_elem_packed {
 118         uint16_t id;
 119         uint16_t flags;
 120         uint32_t len;
 121         uint32_t count;
 122 };
 123
 124 /**
 125  * Structure contains variables relevant to RX/TX virtqueues.
 126  */
 127 struct vhost_virtqueue {
 128         union {
 129                 struct vring_desc       *desc;
 130                 struct vring_packed_desc   *desc_packed;
 131         };
 132         union {
 133                 struct vring_avail      *avail;
 134                 struct vring_packed_desc_event *driver_event;
 135         };
 136         union {
 137                 struct vring_used       *used;
 138                 struct vring_packed_desc_event *device_event;
 139         };
 140         uint32_t                size;
 141
 142         uint16_t                last_avail_idx;
 143         uint16_t                last_used_idx;
 144         /* Last used index we notify to front end. */
 145         uint16_t                signalled_used;
 146         bool                    signalled_used_valid;
 147 #define VIRTIO_INVALID_EVENTFD          (-1)
 148 #define VIRTIO_UNINITIALIZED_EVENTFD    (-2)
 149
 150         /* Backend value to determine if device should started/stopped */
 151         int                     backend;
 152         int                     enabled;
 153         int                     access_ok;
 154         rte_spinlock_t          access_lock;
 155
 156         /* Used to notify the guest (trigger interrupt) */
 157         int                     callfd;
 158         /* Currently unused as polling mode is enabled */
 159         int                     kickfd;
 160
 161         /* Physical address of used ring, for logging */
 162         uint64_t                log_guest_addr;
 163
 164         /* inflight share memory info */
 165         union {
 166                 struct rte_vhost_inflight_info_split *inflight_split;
 167                 struct rte_vhost_inflight_info_packed *inflight_packed;
 168         };
 169         struct rte_vhost_resubmit_info *resubmit_inflight;
 170         uint64_t                global_counter;
 171
 172         uint16_t                nr_zmbuf;
 173         uint16_t                zmbuf_size;
 174         uint16_t                last_zmbuf_idx;
 175         struct zcopy_mbuf       *zmbufs;
 176         struct zcopy_mbuf_list  zmbuf_list;
 177
 178         union {
 179                 struct vring_used_elem  *shadow_used_split;
 180                 struct vring_used_elem_packed *shadow_used_packed;
 181         };
 182         uint16_t                shadow_used_idx;
 183         /* Record packed ring enqueue latest desc cache aligned index */
 184         uint16_t                shadow_aligned_idx;
 185         /* Record packed ring first dequeue desc index */
 186         uint16_t                shadow_last_used_idx;
 187         struct vhost_vring_addr ring_addrs;
 188
 189         struct batch_copy_elem  *batch_copy_elems;
 190         uint16_t                batch_copy_nb_elems;
 191         bool                    used_wrap_counter;
 192         bool                    avail_wrap_counter;
 193
 194         struct log_cache_entry log_cache[VHOST_LOG_CACHE_NR];
 195         uint16_t log_cache_nb_elem;
 196
 197         rte_rwlock_t    iotlb_lock;
 198         rte_rwlock_t    iotlb_pending_lock;
 199         struct rte_mempool *iotlb_pool;
 200         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
 201         int                             iotlb_cache_nr;
 202         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
 203 } __rte_cache_aligned;
 204
 205 /* Old kernels have no such macros defined */
 206 #ifndef VIRTIO_NET_F_GUEST_ANNOUNCE
 207  #define VIRTIO_NET_F_GUEST_ANNOUNCE 21
 208 #endif
 209
 210 #ifndef VIRTIO_NET_F_MQ
 211  #define VIRTIO_NET_F_MQ                22
 212 #endif
 213
 214 #define VHOST_MAX_VRING                 0x100
 215 #define VHOST_MAX_QUEUE_PAIRS           0x80
 216
 217 #ifndef VIRTIO_NET_F_MTU
 218  #define VIRTIO_NET_F_MTU 3
 219 #endif
 220
 221 #ifndef VIRTIO_F_ANY_LAYOUT
 222  #define VIRTIO_F_ANY_LAYOUT            27
 223 #endif
 224
 225 /* Declare IOMMU related bits for older kernels */
 226 #ifndef VIRTIO_F_IOMMU_PLATFORM
 227
 228 #define VIRTIO_F_IOMMU_PLATFORM 33
 229
 230 struct vhost_iotlb_msg {
 231         __u64 iova;
 232         __u64 size;
 233         __u64 uaddr;
 234 #define VHOST_ACCESS_RO      0x1
 235 #define VHOST_ACCESS_WO      0x2
 236 #define VHOST_ACCESS_RW      0x3
 237         __u8 perm;
 238 #define VHOST_IOTLB_MISS           1
 239 #define VHOST_IOTLB_UPDATE         2
 240 #define VHOST_IOTLB_INVALIDATE     3
 241 #define VHOST_IOTLB_ACCESS_FAIL    4
 242         __u8 type;
 243 };
 244
 245 #define VHOST_IOTLB_MSG 0x1
 246
 247 struct vhost_msg {
 248         int type;
 249         union {
 250                 struct vhost_iotlb_msg iotlb;
 251                 __u8 padding[64];
 252         };
 253 };
 254 #endif
 255
 256 /*
 257  * Define virtio 1.0 for older kernels
 258  */
 259 #ifndef VIRTIO_F_VERSION_1
 260  #define VIRTIO_F_VERSION_1 32
 261 #endif
 262
 263 /* Declare packed ring related bits for older kernels */
 264 #ifndef VIRTIO_F_RING_PACKED
 265
 266 #define VIRTIO_F_RING_PACKED 34
 267
 268 struct vring_packed_desc {
 269         uint64_t addr;
 270         uint32_t len;
 271         uint16_t id;
 272         uint16_t flags;
 273 };
 274
 275 struct vring_packed_desc_event {
 276         uint16_t off_wrap;
 277         uint16_t flags;
 278 };
 279 #endif
 280
 281 /*
 282  * Declare below packed ring defines unconditionally
 283  * as Kernel header might use different names.
 284  */
 285 #define VRING_DESC_F_AVAIL      (1ULL << 7)
 286 #define VRING_DESC_F_USED       (1ULL << 15)
 287
 288 #define VRING_EVENT_F_ENABLE 0x0
 289 #define VRING_EVENT_F_DISABLE 0x1
 290 #define VRING_EVENT_F_DESC 0x2
 291
 292 /*
 293  * Available and used descs are in same order
 294  */
 295 #ifndef VIRTIO_F_IN_ORDER
 296 #define VIRTIO_F_IN_ORDER      35
 297 #endif
 298
 299 /* Features supported by this builtin vhost-user net driver. */
 300 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
 301                                 (1ULL << VIRTIO_F_ANY_LAYOUT) | \
 302                                 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
 303                                 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
 304                                 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
 305                                 (1ULL << VIRTIO_NET_F_MQ)      | \
 306                                 (1ULL << VIRTIO_F_VERSION_1)   | \
 307                                 (1ULL << VHOST_F_LOG_ALL)      | \
 308                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
 309                                 (1ULL << VIRTIO_NET_F_GSO) | \
 310                                 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
 311                                 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
 312                                 (1ULL << VIRTIO_NET_F_HOST_UFO) | \
 313                                 (1ULL << VIRTIO_NET_F_HOST_ECN) | \
 314                                 (1ULL << VIRTIO_NET_F_CSUM)    | \
 315                                 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
 316                                 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
 317                                 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
 318                                 (1ULL << VIRTIO_NET_F_GUEST_UFO) | \
 319                                 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
 320                                 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
 321                                 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \
 322                                 (1ULL << VIRTIO_NET_F_MTU)  | \
 323                                 (1ULL << VIRTIO_F_IN_ORDER) | \
 324                                 (1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
 325                                 (1ULL << VIRTIO_F_RING_PACKED))
 326
 327
 328 struct guest_page {
 329         uint64_t guest_phys_addr;
 330         uint64_t host_phys_addr;
 331         uint64_t size;
 332 };
 333
 334 struct inflight_mem_info {
 335         int             fd;
 336         void            *addr;
 337         uint64_t        size;
 338 };
 339
 340 /**
 341  * Device structure contains all configuration information relating
 342  * to the device.
 343  */
 344 struct virtio_net {
 345         /* Frontend (QEMU) memory and memory region information */
 346         struct rte_vhost_memory *mem;
 347         uint64_t                features;
 348         uint64_t                protocol_features;
 349         int                     vid;
 350         uint32_t                flags;
 351         uint16_t                vhost_hlen;
 352         /* to tell if we need broadcast rarp packet */
 353         int16_t                 broadcast_rarp;
 354         uint32_t                nr_vring;
 355         int                     dequeue_zero_copy;
 356         int                     extbuf;
 357         int                     linearbuf;
 358         struct vhost_virtqueue  *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
 359         struct inflight_mem_info *inflight_info;
 360 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
 361         char                    ifname[IF_NAME_SZ];
 362         uint64_t                log_size;
 363         uint64_t                log_base;
 364         uint64_t                log_addr;
 365         struct rte_ether_addr   mac;
 366         uint16_t                mtu;
 367
 368         struct vhost_device_ops const *notify_ops;
 369
 370         uint32_t                nr_guest_pages;
 371         uint32_t                max_guest_pages;
 372         struct guest_page       *guest_pages;
 373
 374         int                     slave_req_fd;
 375         rte_spinlock_t          slave_req_lock;
 376
 377         int                     postcopy_ufd;
 378         int                     postcopy_listening;
 379
 380         /*
 381          * Device id to identify a specific backend device.
 382          * It's set to -1 for the default software implementation.
 383          */
 384         int                     vdpa_dev_id;
 385
 386         /* context data for the external message handlers */
 387         void                    *extern_data;
 388         /* pre and post vhost user message handlers for the device */
 389         struct rte_vhost_user_extern_ops extern_ops;
 390 } __rte_cache_aligned;
 391
 392 static __rte_always_inline bool
 393 vq_is_packed(struct virtio_net *dev)
 394 {
 395         return dev->features & (1ull << VIRTIO_F_RING_PACKED);
 396 }
 397
 398 static inline bool
 399 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
 400 {
 401         uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
 402
 403         return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
 404                 wrap_counter != !!(flags & VRING_DESC_F_USED);
 405 }
 406
 407 static inline void
 408 vq_inc_last_used_packed(struct vhost_virtqueue *vq, uint16_t num)
 409 {
 410         vq->last_used_idx += num;
 411         if (vq->last_used_idx >= vq->size) {
 412                 vq->used_wrap_counter ^= 1;
 413                 vq->last_used_idx -= vq->size;
 414         }
 415 }
 416
 417 static inline void
 418 vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
 419 {
 420         vq->last_avail_idx += num;
 421         if (vq->last_avail_idx >= vq->size) {
 422                 vq->avail_wrap_counter ^= 1;
 423                 vq->last_avail_idx -= vq->size;
 424         }
 425 }
 426
 427 void __vhost_log_cache_write(struct virtio_net *dev,
 428                 struct vhost_virtqueue *vq,
 429                 uint64_t addr, uint64_t len);
 430 void __vhost_log_cache_write_iova(struct virtio_net *dev,
 431                 struct vhost_virtqueue *vq,
 432                 uint64_t iova, uint64_t len);
 433 void __vhost_log_cache_sync(struct virtio_net *dev,
 434                 struct vhost_virtqueue *vq);
 435 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
 436 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 437                             uint64_t iova, uint64_t len);
 438
 439 static __rte_always_inline void
 440 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
 441 {
 442         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 443                 __vhost_log_write(dev, addr, len);
 444 }
 445
 446 static __rte_always_inline void
 447 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
 448 {
 449         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 450                 __vhost_log_cache_sync(dev, vq);
 451 }
 452
 453 static __rte_always_inline void
 454 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
 455                         uint64_t addr, uint64_t len)
 456 {
 457         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 458                 __vhost_log_cache_write(dev, vq, addr, len);
 459 }
 460
 461 static __rte_always_inline void
 462 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 463                         uint64_t offset, uint64_t len)
 464 {
 465         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
 466                 if (unlikely(vq->log_guest_addr == 0))
 467                         return;
 468                 __vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset,
 469                                         len);
 470         }
 471 }
 472
 473 static __rte_always_inline void
 474 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 475                      uint64_t offset, uint64_t len)
 476 {
 477         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
 478                 if (unlikely(vq->log_guest_addr == 0))
 479                         return;
 480                 __vhost_log_write(dev, vq->log_guest_addr + offset, len);
 481         }
 482 }
 483
 484 static __rte_always_inline void
 485 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 486                            uint64_t iova, uint64_t len)
 487 {
 488         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 489                 return;
 490
 491         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 492                 __vhost_log_cache_write_iova(dev, vq, iova, len);
 493         else
 494                 __vhost_log_cache_write(dev, vq, iova, len);
 495 }
 496
 497 static __rte_always_inline void
 498 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 499                            uint64_t iova, uint64_t len)
 500 {
 501         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 502                 return;
 503
 504         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 505                 __vhost_log_write_iova(dev, vq, iova, len);
 506         else
 507                 __vhost_log_write(dev, iova, len);
 508 }
 509
 510 extern int vhost_config_log_level;
 511 extern int vhost_data_log_level;
 512
 513 #define VHOST_LOG_CONFIG(level, fmt, args...)                   \
 514         rte_log(RTE_LOG_ ## level, vhost_config_log_level,      \
 515                 "VHOST_CONFIG: " fmt, ##args)
 516
 517 #define VHOST_LOG_DATA(level, fmt, args...) \
 518         (void)((RTE_LOG_ ## level <= RTE_LOG_DP_LEVEL) ?        \
 519          rte_log(RTE_LOG_ ## level,  vhost_data_log_level,      \
 520                 "VHOST_DATA : " fmt, ##args) :                  \
 521          0)
 522
 523 #ifdef RTE_LIBRTE_VHOST_DEBUG
 524 #define VHOST_MAX_PRINT_BUFF 6072
 525 #define PRINT_PACKET(device, addr, size, header) do { \
 526         char *pkt_addr = (char *)(addr); \
 527         unsigned int index; \
 528         char packet[VHOST_MAX_PRINT_BUFF]; \
 529         \
 530         if ((header)) \
 531                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
 532         else \
 533                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
 534         for (index = 0; index < (size); index++) { \
 535                 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
 536                         "%02hhx ", pkt_addr[index]); \
 537         } \
 538         snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
 539         \
 540         VHOST_LOG_DATA(DEBUG, "%s", packet); \
 541 } while (0)
 542 #else
 543 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
 544 #endif
 545
 546 #define MAX_VHOST_DEVICE        1024
 547 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
 548
 549 #define VHOST_BINARY_SEARCH_THRESH 256
 550
 551 static __rte_always_inline int guest_page_addrcmp(const void *p1,
 552                                                 const void *p2)
 553 {
 554         const struct guest_page *page1 = (const struct guest_page *)p1;
 555         const struct guest_page *page2 = (const struct guest_page *)p2;
 556
 557         if (page1->guest_phys_addr > page2->guest_phys_addr)
 558                 return 1;
 559         if (page1->guest_phys_addr < page2->guest_phys_addr)
 560                 return -1;
 561
 562         return 0;
 563 }
 564
 565 /* Convert guest physical address to host physical address */
 566 static __rte_always_inline rte_iova_t
 567 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
 568 {
 569         uint32_t i;
 570         struct guest_page *page;
 571         struct guest_page key;
 572
 573         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
 574                 key.guest_phys_addr = gpa;
 575                 page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages,
 576                                sizeof(struct guest_page), guest_page_addrcmp);
 577                 if (page) {
 578                         if (gpa + size < page->guest_phys_addr + page->size)
 579                                 return gpa - page->guest_phys_addr +
 580                                         page->host_phys_addr;
 581                 }
 582         } else {
 583                 for (i = 0; i < dev->nr_guest_pages; i++) {
 584                         page = &dev->guest_pages[i];
 585
 586                         if (gpa >= page->guest_phys_addr &&
 587                             gpa + size < page->guest_phys_addr +
 588                             page->size)
 589                                 return gpa - page->guest_phys_addr +
 590                                        page->host_phys_addr;
 591                 }
 592         }
 593
 594         return 0;
 595 }
 596
 597 static __rte_always_inline uint64_t
 598 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
 599 {
 600         struct rte_vhost_mem_region *r;
 601         uint32_t i;
 602
 603         if (unlikely(!dev || !dev->mem))
 604                 return 0;
 605
 606         for (i = 0; i < dev->mem->nregions; i++) {
 607                 r = &dev->mem->regions[i];
 608
 609                 if (vva >= r->host_user_addr &&
 610                     vva + len <  r->host_user_addr + r->size) {
 611                         return r->guest_phys_addr + vva - r->host_user_addr;
 612                 }
 613         }
 614         return 0;
 615 }
 616
 617 static __rte_always_inline struct virtio_net *
 618 get_device(int vid)
 619 {
 620         struct virtio_net *dev = vhost_devices[vid];
 621
 622         if (unlikely(!dev)) {
 623                 VHOST_LOG_CONFIG(ERR,
 624                         "(%d) device not found.\n", vid);
 625         }
 626
 627         return dev;
 628 }
 629
 630 int vhost_new_device(void);
 631 void cleanup_device(struct virtio_net *dev, int destroy);
 632 void reset_device(struct virtio_net *dev);
 633 void vhost_destroy_device(int);
 634 void vhost_destroy_device_notify(struct virtio_net *dev);
 635
 636 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
 637 void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
 638 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
 639
 640 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
 641
 642 void vhost_attach_vdpa_device(int vid, int did);
 643
 644 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
 645 void vhost_enable_dequeue_zero_copy(int vid);
 646 void vhost_set_builtin_virtio_net(int vid, bool enable);
 647 void vhost_enable_extbuf(int vid);
 648 void vhost_enable_linearbuf(int vid);
 649
 650 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
 651
 652 /*
 653  * Backend-specific cleanup.
 654  *
 655  * TODO: fix it; we have one backend now
 656  */
 657 void vhost_backend_cleanup(struct virtio_net *dev);
 658
 659 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 660                         uint64_t iova, uint64_t *len, uint8_t perm);
 661 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
 662                         struct vhost_virtqueue *vq,
 663                         uint64_t desc_addr, uint64_t desc_len);
 664 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 665 uint64_t translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
 666                 uint64_t log_addr);
 667 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 668
 669 static __rte_always_inline uint64_t
 670 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 671                         uint64_t iova, uint64_t *len, uint8_t perm)
 672 {
 673         if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 674                 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
 675
 676         return __vhost_iova_to_vva(dev, vq, iova, len, perm);
 677 }
 678
 679 #define vhost_avail_event(vr) \
 680         (*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
 681 #define vhost_used_event(vr) \
 682         (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
 683
 684 /*
 685  * The following is used with VIRTIO_RING_F_EVENT_IDX.
 686  * Assuming a given event_idx value from the other size, if we have
 687  * just incremented index from old to new_idx, should we trigger an
 688  * event?
 689  */
 690 static __rte_always_inline int
 691 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
 692 {
 693         return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
 694 }
 695
 696 static __rte_always_inline void
 697 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 698 {
 699         /* Flush used->idx update before we read avail->flags. */
 700         rte_smp_mb();
 701
 702         /* Don't kick guest if we don't reach index specified by guest. */
 703         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
 704                 uint16_t old = vq->signalled_used;
 705                 uint16_t new = vq->last_used_idx;
 706                 bool signalled_used_valid = vq->signalled_used_valid;
 707
 708                 vq->signalled_used = new;
 709                 vq->signalled_used_valid = true;
 710
 711                 VHOST_LOG_DATA(DEBUG, "%s: used_event_idx=%d, old=%d, new=%d\n",
 712                         __func__,
 713                         vhost_used_event(vq),
 714                         old, new);
 715
 716                 if ((vhost_need_event(vhost_used_event(vq), new, old) &&
 717                                         (vq->callfd >= 0)) ||
 718                                 unlikely(!signalled_used_valid)) {
 719                         eventfd_write(vq->callfd, (eventfd_t) 1);
 720                         if (dev->notify_ops->guest_notified)
 721                                 dev->notify_ops->guest_notified(dev->vid);
 722                 }
 723         } else {
 724                 /* Kick the guest if necessary. */
 725                 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
 726                                 && (vq->callfd >= 0)) {
 727                         eventfd_write(vq->callfd, (eventfd_t)1);
 728                         if (dev->notify_ops->guest_notified)
 729                                 dev->notify_ops->guest_notified(dev->vid);
 730                 }
 731         }
 732 }
 733
 734 static __rte_always_inline void
 735 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
 736 {
 737         uint16_t old, new, off, off_wrap;
 738         bool signalled_used_valid, kick = false;
 739
 740         /* Flush used desc update. */
 741         rte_smp_mb();
 742
 743         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
 744                 if (vq->driver_event->flags !=
 745                                 VRING_EVENT_F_DISABLE)
 746                         kick = true;
 747                 goto kick;
 748         }
 749
 750         old = vq->signalled_used;
 751         new = vq->last_used_idx;
 752         vq->signalled_used = new;
 753         signalled_used_valid = vq->signalled_used_valid;
 754         vq->signalled_used_valid = true;
 755
 756         if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
 757                 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
 758                         kick = true;
 759                 goto kick;
 760         }
 761
 762         if (unlikely(!signalled_used_valid)) {
 763                 kick = true;
 764                 goto kick;
 765         }
 766
 767         rte_smp_rmb();
 768
 769         off_wrap = vq->driver_event->off_wrap;
 770         off = off_wrap & ~(1 << 15);
 771
 772         if (new <= old)
 773                 old -= vq->size;
 774
 775         if (vq->used_wrap_counter != off_wrap >> 15)
 776                 off -= vq->size;
 777
 778         if (vhost_need_event(off, new, old))
 779                 kick = true;
 780 kick:
 781         if (kick) {
 782                 eventfd_write(vq->callfd, (eventfd_t)1);
 783                 if (dev->notify_ops->guest_notified)
 784                         dev->notify_ops->guest_notified(dev->vid);
 785         }
 786 }
 787
 788 static __rte_always_inline void
 789 free_ind_table(void *idesc)
 790 {
 791         rte_free(idesc);
 792 }
 793
 794 static __rte_always_inline void
 795 restore_mbuf(struct rte_mbuf *m)
 796 {
 797         uint32_t mbuf_size, priv_size;
 798
 799         while (m) {
 800                 priv_size = rte_pktmbuf_priv_size(m->pool);
 801                 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
 802                 /* start of buffer is after mbuf structure and priv data */
 803
 804                 m->buf_addr = (char *)m + mbuf_size;
 805                 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
 806                 m = m->next;
 807         }
 808 }
 809
 810 static __rte_always_inline bool
 811 mbuf_is_consumed(struct rte_mbuf *m)
 812 {
 813         while (m) {
 814                 if (rte_mbuf_refcnt_read(m) > 1)
 815                         return false;
 816                 m = m->next;
 817         }
 818
 819         return true;
 820 }
 821
 822 static __rte_always_inline void
 823 put_zmbuf(struct zcopy_mbuf *zmbuf)
 824 {
 825         zmbuf->in_use = 0;
 826 }
 827
 828 #endif /* _VHOST_NET_CDEV_H_ */