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 /**
  43  * Structure contains buffer address, length and descriptor index
  44  * from vring to do scatter RX.
  45  */
  46 struct buf_vector {
  47         uint64_t buf_iova;
  48         uint64_t buf_addr;
  49         uint32_t buf_len;
  50         uint32_t desc_idx;
  51 };
  52
  53 /*
  54  * A structure to hold some fields needed in zero copy code path,
  55  * mainly for associating an mbuf with the right desc_idx.
  56  */
  57 struct zcopy_mbuf {
  58         struct rte_mbuf *mbuf;
  59         uint32_t desc_idx;
  60         uint16_t desc_count;
  61         uint16_t in_use;
  62
  63         TAILQ_ENTRY(zcopy_mbuf) next;
  64 };
  65 TAILQ_HEAD(zcopy_mbuf_list, zcopy_mbuf);
  66
  67 /*
  68  * Structure contains the info for each batched memory copy.
  69  */
  70 struct batch_copy_elem {
  71         void *dst;
  72         void *src;
  73         uint32_t len;
  74         uint64_t log_addr;
  75 };
  76
  77 /*
  78  * Structure that contains the info for batched dirty logging.
  79  */
  80 struct log_cache_entry {
  81         uint32_t offset;
  82         unsigned long val;
  83 };
  84
  85 struct vring_used_elem_packed {
  86         uint16_t id;
  87         uint32_t len;
  88         uint32_t count;
  89 };
  90
  91 /**
  92  * Structure contains variables relevant to RX/TX virtqueues.
  93  */
  94 struct vhost_virtqueue {
  95         union {
  96                 struct vring_desc       *desc;
  97                 struct vring_packed_desc   *desc_packed;
  98         };
  99         union {
 100                 struct vring_avail      *avail;
 101                 struct vring_packed_desc_event *driver_event;
 102         };
 103         union {
 104                 struct vring_used       *used;
 105                 struct vring_packed_desc_event *device_event;
 106         };
 107         uint32_t                size;
 108
 109         uint16_t                last_avail_idx;
 110         uint16_t                last_used_idx;
 111         /* Last used index we notify to front end. */
 112         uint16_t                signalled_used;
 113         bool                    signalled_used_valid;
 114 #define VIRTIO_INVALID_EVENTFD          (-1)
 115 #define VIRTIO_UNINITIALIZED_EVENTFD    (-2)
 116
 117         /* Backend value to determine if device should started/stopped */
 118         int                     backend;
 119         int                     enabled;
 120         int                     access_ok;
 121         rte_spinlock_t          access_lock;
 122
 123         /* Used to notify the guest (trigger interrupt) */
 124         int                     callfd;
 125         /* Currently unused as polling mode is enabled */
 126         int                     kickfd;
 127
 128         /* Physical address of used ring, for logging */
 129         uint64_t                log_guest_addr;
 130
 131         /* inflight share memory info */
 132         union {
 133                 struct rte_vhost_inflight_info_split *inflight_split;
 134                 struct rte_vhost_inflight_info_packed *inflight_packed;
 135         };
 136         struct rte_vhost_resubmit_info *resubmit_inflight;
 137         uint64_t                global_counter;
 138
 139         uint16_t                nr_zmbuf;
 140         uint16_t                zmbuf_size;
 141         uint16_t                last_zmbuf_idx;
 142         struct zcopy_mbuf       *zmbufs;
 143         struct zcopy_mbuf_list  zmbuf_list;
 144
 145         union {
 146                 struct vring_used_elem  *shadow_used_split;
 147                 struct vring_used_elem_packed *shadow_used_packed;
 148         };
 149         uint16_t                shadow_used_idx;
 150         struct vhost_vring_addr ring_addrs;
 151
 152         struct batch_copy_elem  *batch_copy_elems;
 153         uint16_t                batch_copy_nb_elems;
 154         bool                    used_wrap_counter;
 155         bool                    avail_wrap_counter;
 156
 157         struct log_cache_entry log_cache[VHOST_LOG_CACHE_NR];
 158         uint16_t log_cache_nb_elem;
 159
 160         rte_rwlock_t    iotlb_lock;
 161         rte_rwlock_t    iotlb_pending_lock;
 162         struct rte_mempool *iotlb_pool;
 163         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
 164         int                             iotlb_cache_nr;
 165         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
 166 } __rte_cache_aligned;
 167
 168 /* Old kernels have no such macros defined */
 169 #ifndef VIRTIO_NET_F_GUEST_ANNOUNCE
 170  #define VIRTIO_NET_F_GUEST_ANNOUNCE 21
 171 #endif
 172
 173 #ifndef VIRTIO_NET_F_MQ
 174  #define VIRTIO_NET_F_MQ                22
 175 #endif
 176
 177 #define VHOST_MAX_VRING                 0x100
 178 #define VHOST_MAX_QUEUE_PAIRS           0x80
 179
 180 #ifndef VIRTIO_NET_F_MTU
 181  #define VIRTIO_NET_F_MTU 3
 182 #endif
 183
 184 #ifndef VIRTIO_F_ANY_LAYOUT
 185  #define VIRTIO_F_ANY_LAYOUT            27
 186 #endif
 187
 188 /* Declare IOMMU related bits for older kernels */
 189 #ifndef VIRTIO_F_IOMMU_PLATFORM
 190
 191 #define VIRTIO_F_IOMMU_PLATFORM 33
 192
 193 struct vhost_iotlb_msg {
 194         __u64 iova;
 195         __u64 size;
 196         __u64 uaddr;
 197 #define VHOST_ACCESS_RO      0x1
 198 #define VHOST_ACCESS_WO      0x2
 199 #define VHOST_ACCESS_RW      0x3
 200         __u8 perm;
 201 #define VHOST_IOTLB_MISS           1
 202 #define VHOST_IOTLB_UPDATE         2
 203 #define VHOST_IOTLB_INVALIDATE     3
 204 #define VHOST_IOTLB_ACCESS_FAIL    4
 205         __u8 type;
 206 };
 207
 208 #define VHOST_IOTLB_MSG 0x1
 209
 210 struct vhost_msg {
 211         int type;
 212         union {
 213                 struct vhost_iotlb_msg iotlb;
 214                 __u8 padding[64];
 215         };
 216 };
 217 #endif
 218
 219 /*
 220  * Define virtio 1.0 for older kernels
 221  */
 222 #ifndef VIRTIO_F_VERSION_1
 223  #define VIRTIO_F_VERSION_1 32
 224 #endif
 225
 226 /* Declare packed ring related bits for older kernels */
 227 #ifndef VIRTIO_F_RING_PACKED
 228
 229 #define VIRTIO_F_RING_PACKED 34
 230
 231 struct vring_packed_desc {
 232         uint64_t addr;
 233         uint32_t len;
 234         uint16_t id;
 235         uint16_t flags;
 236 };
 237
 238 struct vring_packed_desc_event {
 239         uint16_t off_wrap;
 240         uint16_t flags;
 241 };
 242 #endif
 243
 244 /*
 245  * Declare below packed ring defines unconditionally
 246  * as Kernel header might use different names.
 247  */
 248 #define VRING_DESC_F_AVAIL      (1ULL << 7)
 249 #define VRING_DESC_F_USED       (1ULL << 15)
 250
 251 #define VRING_EVENT_F_ENABLE 0x0
 252 #define VRING_EVENT_F_DISABLE 0x1
 253 #define VRING_EVENT_F_DESC 0x2
 254
 255 /*
 256  * Available and used descs are in same order
 257  */
 258 #ifndef VIRTIO_F_IN_ORDER
 259 #define VIRTIO_F_IN_ORDER      35
 260 #endif
 261
 262 /* Features supported by this builtin vhost-user net driver. */
 263 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
 264                                 (1ULL << VIRTIO_F_ANY_LAYOUT) | \
 265                                 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
 266                                 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
 267                                 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
 268                                 (1ULL << VIRTIO_NET_F_MQ)      | \
 269                                 (1ULL << VIRTIO_F_VERSION_1)   | \
 270                                 (1ULL << VHOST_F_LOG_ALL)      | \
 271                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
 272                                 (1ULL << VIRTIO_NET_F_GSO) | \
 273                                 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
 274                                 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
 275                                 (1ULL << VIRTIO_NET_F_HOST_UFO) | \
 276                                 (1ULL << VIRTIO_NET_F_HOST_ECN) | \
 277                                 (1ULL << VIRTIO_NET_F_CSUM)    | \
 278                                 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
 279                                 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
 280                                 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
 281                                 (1ULL << VIRTIO_NET_F_GUEST_UFO) | \
 282                                 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
 283                                 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
 284                                 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \
 285                                 (1ULL << VIRTIO_NET_F_MTU)  | \
 286                                 (1ULL << VIRTIO_F_IN_ORDER) | \
 287                                 (1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
 288                                 (1ULL << VIRTIO_F_RING_PACKED))
 289
 290
 291 struct guest_page {
 292         uint64_t guest_phys_addr;
 293         uint64_t host_phys_addr;
 294         uint64_t size;
 295 };
 296
 297 struct inflight_mem_info {
 298         int             fd;
 299         void            *addr;
 300         uint64_t        size;
 301 };
 302
 303 /**
 304  * Device structure contains all configuration information relating
 305  * to the device.
 306  */
 307 struct virtio_net {
 308         /* Frontend (QEMU) memory and memory region information */
 309         struct rte_vhost_memory *mem;
 310         uint64_t                features;
 311         uint64_t                protocol_features;
 312         int                     vid;
 313         uint32_t                flags;
 314         uint16_t                vhost_hlen;
 315         /* to tell if we need broadcast rarp packet */
 316         rte_atomic16_t          broadcast_rarp;
 317         uint32_t                nr_vring;
 318         int                     dequeue_zero_copy;
 319         int                     extbuf;
 320         int                     linearbuf;
 321         struct vhost_virtqueue  *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
 322         struct inflight_mem_info *inflight_info;
 323 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
 324         char                    ifname[IF_NAME_SZ];
 325         uint64_t                log_size;
 326         uint64_t                log_base;
 327         uint64_t                log_addr;
 328         struct rte_ether_addr   mac;
 329         uint16_t                mtu;
 330
 331         struct vhost_device_ops const *notify_ops;
 332
 333         uint32_t                nr_guest_pages;
 334         uint32_t                max_guest_pages;
 335         struct guest_page       *guest_pages;
 336
 337         int                     slave_req_fd;
 338         rte_spinlock_t          slave_req_lock;
 339
 340         int                     postcopy_ufd;
 341         int                     postcopy_listening;
 342
 343         /*
 344          * Device id to identify a specific backend device.
 345          * It's set to -1 for the default software implementation.
 346          */
 347         int                     vdpa_dev_id;
 348
 349         /* context data for the external message handlers */
 350         void                    *extern_data;
 351         /* pre and post vhost user message handlers for the device */
 352         struct rte_vhost_user_extern_ops extern_ops;
 353 } __rte_cache_aligned;
 354
 355 static __rte_always_inline bool
 356 vq_is_packed(struct virtio_net *dev)
 357 {
 358         return dev->features & (1ull << VIRTIO_F_RING_PACKED);
 359 }
 360
 361 static inline bool
 362 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
 363 {
 364         uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
 365
 366         return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
 367                 wrap_counter != !!(flags & VRING_DESC_F_USED);
 368 }
 369
 370 static inline void
 371 vq_inc_last_used_packed(struct vhost_virtqueue *vq, uint16_t num)
 372 {
 373         vq->last_used_idx += num;
 374         if (vq->last_used_idx >= vq->size) {
 375                 vq->used_wrap_counter ^= 1;
 376                 vq->last_used_idx -= vq->size;
 377         }
 378 }
 379
 380 static inline void
 381 vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
 382 {
 383         vq->last_avail_idx += num;
 384         if (vq->last_avail_idx >= vq->size) {
 385                 vq->avail_wrap_counter ^= 1;
 386                 vq->last_avail_idx -= vq->size;
 387         }
 388 }
 389
 390 void __vhost_log_cache_write(struct virtio_net *dev,
 391                 struct vhost_virtqueue *vq,
 392                 uint64_t addr, uint64_t len);
 393 void __vhost_log_cache_write_iova(struct virtio_net *dev,
 394                 struct vhost_virtqueue *vq,
 395                 uint64_t iova, uint64_t len);
 396 void __vhost_log_cache_sync(struct virtio_net *dev,
 397                 struct vhost_virtqueue *vq);
 398 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
 399 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 400                             uint64_t iova, uint64_t len);
 401
 402 static __rte_always_inline void
 403 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
 404 {
 405         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 406                 __vhost_log_write(dev, addr, len);
 407 }
 408
 409 static __rte_always_inline void
 410 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
 411 {
 412         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 413                 __vhost_log_cache_sync(dev, vq);
 414 }
 415
 416 static __rte_always_inline void
 417 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
 418                         uint64_t addr, uint64_t len)
 419 {
 420         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 421                 __vhost_log_cache_write(dev, vq, addr, len);
 422 }
 423
 424 static __rte_always_inline void
 425 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 426                         uint64_t offset, uint64_t len)
 427 {
 428         vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset, len);
 429 }
 430
 431 static __rte_always_inline void
 432 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 433                      uint64_t offset, uint64_t len)
 434 {
 435         vhost_log_write(dev, vq->log_guest_addr + offset, len);
 436 }
 437
 438 static __rte_always_inline void
 439 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 440                            uint64_t iova, uint64_t len)
 441 {
 442         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 443                 return;
 444
 445         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 446                 __vhost_log_cache_write_iova(dev, vq, iova, len);
 447         else
 448                 __vhost_log_cache_write(dev, vq, iova, len);
 449 }
 450
 451 static __rte_always_inline void
 452 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 453                            uint64_t iova, uint64_t len)
 454 {
 455         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 456                 return;
 457
 458         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 459                 __vhost_log_write_iova(dev, vq, iova, len);
 460         else
 461                 __vhost_log_write(dev, iova, len);
 462 }
 463
 464 /* Macros for printing using RTE_LOG */
 465 #define RTE_LOGTYPE_VHOST_CONFIG RTE_LOGTYPE_USER1
 466 #define RTE_LOGTYPE_VHOST_DATA   RTE_LOGTYPE_USER1
 467
 468 #ifdef RTE_LIBRTE_VHOST_DEBUG
 469 #define VHOST_MAX_PRINT_BUFF 6072
 470 #define VHOST_LOG_DEBUG(log_type, fmt, args...) \
 471         RTE_LOG(DEBUG, log_type, fmt, ##args)
 472 #define PRINT_PACKET(device, addr, size, header) do { \
 473         char *pkt_addr = (char *)(addr); \
 474         unsigned int index; \
 475         char packet[VHOST_MAX_PRINT_BUFF]; \
 476         \
 477         if ((header)) \
 478                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
 479         else \
 480                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
 481         for (index = 0; index < (size); index++) { \
 482                 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
 483                         "%02hhx ", pkt_addr[index]); \
 484         } \
 485         snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
 486         \
 487         VHOST_LOG_DEBUG(VHOST_DATA, "%s", packet); \
 488 } while (0)
 489 #else
 490 #define VHOST_LOG_DEBUG(log_type, fmt, args...) do {} while (0)
 491 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
 492 #endif
 493
 494 extern uint64_t VHOST_FEATURES;
 495 #define MAX_VHOST_DEVICE        1024
 496 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
 497
 498 /* Convert guest physical address to host physical address */
 499 static __rte_always_inline rte_iova_t
 500 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
 501 {
 502         uint32_t i;
 503         struct guest_page *page;
 504
 505         for (i = 0; i < dev->nr_guest_pages; i++) {
 506                 page = &dev->guest_pages[i];
 507
 508                 if (gpa >= page->guest_phys_addr &&
 509                     gpa + size < page->guest_phys_addr + page->size) {
 510                         return gpa - page->guest_phys_addr +
 511                                page->host_phys_addr;
 512                 }
 513         }
 514
 515         return 0;
 516 }
 517
 518 static __rte_always_inline uint64_t
 519 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
 520 {
 521         struct rte_vhost_mem_region *r;
 522         uint32_t i;
 523
 524         if (unlikely(!dev || !dev->mem))
 525                 return 0;
 526
 527         for (i = 0; i < dev->mem->nregions; i++) {
 528                 r = &dev->mem->regions[i];
 529
 530                 if (vva >= r->host_user_addr &&
 531                     vva + len <  r->host_user_addr + r->size) {
 532                         return r->guest_phys_addr + vva - r->host_user_addr;
 533                 }
 534         }
 535         return 0;
 536 }
 537
 538 static __rte_always_inline struct virtio_net *
 539 get_device(int vid)
 540 {
 541         struct virtio_net *dev = vhost_devices[vid];
 542
 543         if (unlikely(!dev)) {
 544                 RTE_LOG(ERR, VHOST_CONFIG,
 545                         "(%d) device not found.\n", vid);
 546         }
 547
 548         return dev;
 549 }
 550
 551 int vhost_new_device(void);
 552 void cleanup_device(struct virtio_net *dev, int destroy);
 553 void reset_device(struct virtio_net *dev);
 554 void vhost_destroy_device(int);
 555 void vhost_destroy_device_notify(struct virtio_net *dev);
 556
 557 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
 558 void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
 559 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
 560
 561 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
 562
 563 void vhost_attach_vdpa_device(int vid, int did);
 564
 565 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
 566 void vhost_enable_dequeue_zero_copy(int vid);
 567 void vhost_set_builtin_virtio_net(int vid, bool enable);
 568 void vhost_enable_extbuf(int vid);
 569 void vhost_enable_linearbuf(int vid);
 570
 571 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
 572
 573 /*
 574  * Backend-specific cleanup.
 575  *
 576  * TODO: fix it; we have one backend now
 577  */
 578 void vhost_backend_cleanup(struct virtio_net *dev);
 579
 580 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 581                         uint64_t iova, uint64_t *len, uint8_t perm);
 582 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
 583                         struct vhost_virtqueue *vq,
 584                         uint64_t desc_addr, uint64_t desc_len);
 585 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 586 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 587
 588 static __rte_always_inline uint64_t
 589 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 590                         uint64_t iova, uint64_t *len, uint8_t perm)
 591 {
 592         if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 593                 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
 594
 595         return __vhost_iova_to_vva(dev, vq, iova, len, perm);
 596 }
 597
 598 #define vhost_avail_event(vr) \
 599         (*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
 600 #define vhost_used_event(vr) \
 601         (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
 602
 603 /*
 604  * The following is used with VIRTIO_RING_F_EVENT_IDX.
 605  * Assuming a given event_idx value from the other size, if we have
 606  * just incremented index from old to new_idx, should we trigger an
 607  * event?
 608  */
 609 static __rte_always_inline int
 610 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
 611 {
 612         return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
 613 }
 614
 615 static __rte_always_inline void
 616 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 617 {
 618         /* Flush used->idx update before we read avail->flags. */
 619         rte_smp_mb();
 620
 621         /* Don't kick guest if we don't reach index specified by guest. */
 622         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
 623                 uint16_t old = vq->signalled_used;
 624                 uint16_t new = vq->last_used_idx;
 625                 bool signalled_used_valid = vq->signalled_used_valid;
 626
 627                 vq->signalled_used = new;
 628                 vq->signalled_used_valid = true;
 629
 630                 VHOST_LOG_DEBUG(VHOST_DATA, "%s: used_event_idx=%d, old=%d, new=%d\n",
 631                         __func__,
 632                         vhost_used_event(vq),
 633                         old, new);
 634
 635                 if ((vhost_need_event(vhost_used_event(vq), new, old) &&
 636                                         (vq->callfd >= 0)) ||
 637                                 unlikely(!signalled_used_valid)) {
 638                         eventfd_write(vq->callfd, (eventfd_t) 1);
 639                         if (dev->notify_ops->guest_notified)
 640                                 dev->notify_ops->guest_notified(dev->vid);
 641                 }
 642         } else {
 643                 /* Kick the guest if necessary. */
 644                 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
 645                                 && (vq->callfd >= 0)) {
 646                         eventfd_write(vq->callfd, (eventfd_t)1);
 647                         if (dev->notify_ops->guest_notified)
 648                                 dev->notify_ops->guest_notified(dev->vid);
 649                 }
 650         }
 651 }
 652
 653 static __rte_always_inline void
 654 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
 655 {
 656         uint16_t old, new, off, off_wrap;
 657         bool signalled_used_valid, kick = false;
 658
 659         /* Flush used desc update. */
 660         rte_smp_mb();
 661
 662         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
 663                 if (vq->driver_event->flags !=
 664                                 VRING_EVENT_F_DISABLE)
 665                         kick = true;
 666                 goto kick;
 667         }
 668
 669         old = vq->signalled_used;
 670         new = vq->last_used_idx;
 671         vq->signalled_used = new;
 672         signalled_used_valid = vq->signalled_used_valid;
 673         vq->signalled_used_valid = true;
 674
 675         if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
 676                 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
 677                         kick = true;
 678                 goto kick;
 679         }
 680
 681         if (unlikely(!signalled_used_valid)) {
 682                 kick = true;
 683                 goto kick;
 684         }
 685
 686         rte_smp_rmb();
 687
 688         off_wrap = vq->driver_event->off_wrap;
 689         off = off_wrap & ~(1 << 15);
 690
 691         if (new <= old)
 692                 old -= vq->size;
 693
 694         if (vq->used_wrap_counter != off_wrap >> 15)
 695                 off -= vq->size;
 696
 697         if (vhost_need_event(off, new, old))
 698                 kick = true;
 699 kick:
 700         if (kick) {
 701                 eventfd_write(vq->callfd, (eventfd_t)1);
 702                 if (dev->notify_ops->guest_notified)
 703                         dev->notify_ops->guest_notified(dev->vid);
 704         }
 705 }
 706
 707 static __rte_always_inline void
 708 free_ind_table(void *idesc)
 709 {
 710         rte_free(idesc);
 711 }
 712
 713 static __rte_always_inline void
 714 restore_mbuf(struct rte_mbuf *m)
 715 {
 716         uint32_t mbuf_size, priv_size;
 717
 718         while (m) {
 719                 priv_size = rte_pktmbuf_priv_size(m->pool);
 720                 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
 721                 /* start of buffer is after mbuf structure and priv data */
 722
 723                 m->buf_addr = (char *)m + mbuf_size;
 724                 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
 725                 m = m->next;
 726         }
 727 }
 728
 729 static __rte_always_inline bool
 730 mbuf_is_consumed(struct rte_mbuf *m)
 731 {
 732         while (m) {
 733                 if (rte_mbuf_refcnt_read(m) > 1)
 734                         return false;
 735                 m = m->next;
 736         }
 737
 738         return true;
 739 }
 740
 741 static __rte_always_inline void
 742 put_zmbuf(struct zcopy_mbuf *zmbuf)
 743 {
 744         zmbuf->in_use = 0;
 745 }
 746
 747 #endif /* _VHOST_NET_CDEV_H_ */