ceph/src/seastar/dpdk/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         uint16_t                nr_zmbuf;
 132         uint16_t                zmbuf_size;
 133         uint16_t                last_zmbuf_idx;
 134         struct zcopy_mbuf       *zmbufs;
 135         struct zcopy_mbuf_list  zmbuf_list;
 136
 137         union {
 138                 struct vring_used_elem  *shadow_used_split;
 139                 struct vring_used_elem_packed *shadow_used_packed;
 140         };
 141         uint16_t                shadow_used_idx;
 142         struct vhost_vring_addr ring_addrs;
 143
 144         struct batch_copy_elem  *batch_copy_elems;
 145         uint16_t                batch_copy_nb_elems;
 146         bool                    used_wrap_counter;
 147         bool                    avail_wrap_counter;
 148
 149         struct log_cache_entry log_cache[VHOST_LOG_CACHE_NR];
 150         uint16_t log_cache_nb_elem;
 151
 152         rte_rwlock_t    iotlb_lock;
 153         rte_rwlock_t    iotlb_pending_lock;
 154         struct rte_mempool *iotlb_pool;
 155         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
 156         int                             iotlb_cache_nr;
 157         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
 158 } __rte_cache_aligned;
 159
 160 /* Old kernels have no such macros defined */
 161 #ifndef VIRTIO_NET_F_GUEST_ANNOUNCE
 162  #define VIRTIO_NET_F_GUEST_ANNOUNCE 21
 163 #endif
 164
 165 #ifndef VIRTIO_NET_F_MQ
 166  #define VIRTIO_NET_F_MQ                22
 167 #endif
 168
 169 #define VHOST_MAX_VRING                 0x100
 170 #define VHOST_MAX_QUEUE_PAIRS           0x80
 171
 172 #ifndef VIRTIO_NET_F_MTU
 173  #define VIRTIO_NET_F_MTU 3
 174 #endif
 175
 176 #ifndef VIRTIO_F_ANY_LAYOUT
 177  #define VIRTIO_F_ANY_LAYOUT            27
 178 #endif
 179
 180 /* Declare IOMMU related bits for older kernels */
 181 #ifndef VIRTIO_F_IOMMU_PLATFORM
 182
 183 #define VIRTIO_F_IOMMU_PLATFORM 33
 184
 185 struct vhost_iotlb_msg {
 186         __u64 iova;
 187         __u64 size;
 188         __u64 uaddr;
 189 #define VHOST_ACCESS_RO      0x1
 190 #define VHOST_ACCESS_WO      0x2
 191 #define VHOST_ACCESS_RW      0x3
 192         __u8 perm;
 193 #define VHOST_IOTLB_MISS           1
 194 #define VHOST_IOTLB_UPDATE         2
 195 #define VHOST_IOTLB_INVALIDATE     3
 196 #define VHOST_IOTLB_ACCESS_FAIL    4
 197         __u8 type;
 198 };
 199
 200 #define VHOST_IOTLB_MSG 0x1
 201
 202 struct vhost_msg {
 203         int type;
 204         union {
 205                 struct vhost_iotlb_msg iotlb;
 206                 __u8 padding[64];
 207         };
 208 };
 209 #endif
 210
 211 /*
 212  * Define virtio 1.0 for older kernels
 213  */
 214 #ifndef VIRTIO_F_VERSION_1
 215  #define VIRTIO_F_VERSION_1 32
 216 #endif
 217
 218 /* Declare packed ring related bits for older kernels */
 219 #ifndef VIRTIO_F_RING_PACKED
 220
 221 #define VIRTIO_F_RING_PACKED 34
 222
 223 struct vring_packed_desc {
 224         uint64_t addr;
 225         uint32_t len;
 226         uint16_t id;
 227         uint16_t flags;
 228 };
 229
 230 struct vring_packed_desc_event {
 231         uint16_t off_wrap;
 232         uint16_t flags;
 233 };
 234 #endif
 235
 236 /*
 237  * Declare below packed ring defines unconditionally
 238  * as Kernel header might use different names.
 239  */
 240 #define VRING_DESC_F_AVAIL      (1ULL << 7)
 241 #define VRING_DESC_F_USED       (1ULL << 15)
 242
 243 #define VRING_EVENT_F_ENABLE 0x0
 244 #define VRING_EVENT_F_DISABLE 0x1
 245 #define VRING_EVENT_F_DESC 0x2
 246
 247 /*
 248  * Available and used descs are in same order
 249  */
 250 #ifndef VIRTIO_F_IN_ORDER
 251 #define VIRTIO_F_IN_ORDER      35
 252 #endif
 253
 254 /* Features supported by this builtin vhost-user net driver. */
 255 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
 256                                 (1ULL << VIRTIO_F_ANY_LAYOUT) | \
 257                                 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
 258                                 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
 259                                 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
 260                                 (1ULL << VIRTIO_NET_F_MQ)      | \
 261                                 (1ULL << VIRTIO_F_VERSION_1)   | \
 262                                 (1ULL << VHOST_F_LOG_ALL)      | \
 263                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
 264                                 (1ULL << VIRTIO_NET_F_GSO) | \
 265                                 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
 266                                 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
 267                                 (1ULL << VIRTIO_NET_F_HOST_UFO) | \
 268                                 (1ULL << VIRTIO_NET_F_HOST_ECN) | \
 269                                 (1ULL << VIRTIO_NET_F_CSUM)    | \
 270                                 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
 271                                 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
 272                                 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
 273                                 (1ULL << VIRTIO_NET_F_GUEST_UFO) | \
 274                                 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
 275                                 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
 276                                 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \
 277                                 (1ULL << VIRTIO_NET_F_MTU)  | \
 278                                 (1ULL << VIRTIO_F_IN_ORDER) | \
 279                                 (1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
 280                                 (1ULL << VIRTIO_F_RING_PACKED))
 281
 282
 283 struct guest_page {
 284         uint64_t guest_phys_addr;
 285         uint64_t host_phys_addr;
 286         uint64_t size;
 287 };
 288
 289 /**
 290  * Device structure contains all configuration information relating
 291  * to the device.
 292  */
 293 struct virtio_net {
 294         /* Frontend (QEMU) memory and memory region information */
 295         struct rte_vhost_memory *mem;
 296         uint64_t                features;
 297         uint64_t                protocol_features;
 298         int                     vid;
 299         uint32_t                flags;
 300         uint16_t                vhost_hlen;
 301         /* to tell if we need broadcast rarp packet */
 302         rte_atomic16_t          broadcast_rarp;
 303         uint32_t                nr_vring;
 304         int                     dequeue_zero_copy;
 305         struct vhost_virtqueue  *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
 306 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
 307         char                    ifname[IF_NAME_SZ];
 308         uint64_t                log_size;
 309         uint64_t                log_base;
 310         uint64_t                log_addr;
 311         struct ether_addr       mac;
 312         uint16_t                mtu;
 313
 314         struct vhost_device_ops const *notify_ops;
 315
 316         uint32_t                nr_guest_pages;
 317         uint32_t                max_guest_pages;
 318         struct guest_page       *guest_pages;
 319
 320         int                     slave_req_fd;
 321         rte_spinlock_t          slave_req_lock;
 322
 323         int                     postcopy_ufd;
 324         int                     postcopy_listening;
 325
 326         /*
 327          * Device id to identify a specific backend device.
 328          * It's set to -1 for the default software implementation.
 329          */
 330         int                     vdpa_dev_id;
 331
 332         /* context data for the external message handlers */
 333         void                    *extern_data;
 334         /* pre and post vhost user message handlers for the device */
 335         struct rte_vhost_user_extern_ops extern_ops;
 336 } __rte_cache_aligned;
 337
 338 static __rte_always_inline bool
 339 vq_is_packed(struct virtio_net *dev)
 340 {
 341         return dev->features & (1ull << VIRTIO_F_RING_PACKED);
 342 }
 343
 344 static inline bool
 345 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
 346 {
 347         uint16_t flags = *((volatile uint16_t *) &desc->flags);
 348
 349         return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
 350                 wrap_counter != !!(flags & VRING_DESC_F_USED);
 351 }
 352
 353 #define VHOST_LOG_PAGE  4096
 354
 355 /*
 356  * Atomically set a bit in memory.
 357  */
 358 static __rte_always_inline void
 359 vhost_set_bit(unsigned int nr, volatile uint8_t *addr)
 360 {
 361 #if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION < 70100)
 362         /*
 363          * __sync_ built-ins are deprecated, but __atomic_ ones
 364          * are sub-optimized in older GCC versions.
 365          */
 366         __sync_fetch_and_or_1(addr, (1U << nr));
 367 #else
 368         __atomic_fetch_or(addr, (1U << nr), __ATOMIC_RELAXED);
 369 #endif
 370 }
 371
 372 static __rte_always_inline void
 373 vhost_log_page(uint8_t *log_base, uint64_t page)
 374 {
 375         vhost_set_bit(page % 8, &log_base[page / 8]);
 376 }
 377
 378 static __rte_always_inline void
 379 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
 380 {
 381         uint64_t page;
 382
 383         if (likely(((dev->features & (1ULL << VHOST_F_LOG_ALL)) == 0) ||
 384                    !dev->log_base || !len))
 385                 return;
 386
 387         if (unlikely(dev->log_size <= ((addr + len - 1) / VHOST_LOG_PAGE / 8)))
 388                 return;
 389
 390         /* To make sure guest memory updates are committed before logging */
 391         rte_smp_wmb();
 392
 393         page = addr / VHOST_LOG_PAGE;
 394         while (page * VHOST_LOG_PAGE < addr + len) {
 395                 vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
 396                 page += 1;
 397         }
 398 }
 399
 400 static __rte_always_inline void
 401 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
 402 {
 403         unsigned long *log_base;
 404         int i;
 405
 406         if (likely(((dev->features & (1ULL << VHOST_F_LOG_ALL)) == 0) ||
 407                    !dev->log_base))
 408                 return;
 409
 410         rte_smp_wmb();
 411
 412         log_base = (unsigned long *)(uintptr_t)dev->log_base;
 413
 414         for (i = 0; i < vq->log_cache_nb_elem; i++) {
 415                 struct log_cache_entry *elem = vq->log_cache + i;
 416
 417 #if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION < 70100)
 418                 /*
 419                  * '__sync' builtins are deprecated, but '__atomic' ones
 420                  * are sub-optimized in older GCC versions.
 421                  */
 422                 __sync_fetch_and_or(log_base + elem->offset, elem->val);
 423 #else
 424                 __atomic_fetch_or(log_base + elem->offset, elem->val,
 425                                 __ATOMIC_RELAXED);
 426 #endif
 427         }
 428
 429         rte_smp_wmb();
 430
 431         vq->log_cache_nb_elem = 0;
 432 }
 433
 434 static __rte_always_inline void
 435 vhost_log_cache_page(struct virtio_net *dev, struct vhost_virtqueue *vq,
 436                         uint64_t page)
 437 {
 438         uint32_t bit_nr = page % (sizeof(unsigned long) << 3);
 439         uint32_t offset = page / (sizeof(unsigned long) << 3);
 440         int i;
 441
 442         for (i = 0; i < vq->log_cache_nb_elem; i++) {
 443                 struct log_cache_entry *elem = vq->log_cache + i;
 444
 445                 if (elem->offset == offset) {
 446                         elem->val |= (1UL << bit_nr);
 447                         return;
 448                 }
 449         }
 450
 451         if (unlikely(i >= VHOST_LOG_CACHE_NR)) {
 452                 /*
 453                  * No more room for a new log cache entry,
 454                  * so write the dirty log map directly.
 455                  */
 456                 rte_smp_wmb();
 457                 vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
 458
 459                 return;
 460         }
 461
 462         vq->log_cache[i].offset = offset;
 463         vq->log_cache[i].val = (1UL << bit_nr);
 464         vq->log_cache_nb_elem++;
 465 }
 466
 467 static __rte_always_inline void
 468 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
 469                         uint64_t addr, uint64_t len)
 470 {
 471         uint64_t page;
 472
 473         if (likely(((dev->features & (1ULL << VHOST_F_LOG_ALL)) == 0) ||
 474                    !dev->log_base || !len))
 475                 return;
 476
 477         if (unlikely(dev->log_size <= ((addr + len - 1) / VHOST_LOG_PAGE / 8)))
 478                 return;
 479
 480         page = addr / VHOST_LOG_PAGE;
 481         while (page * VHOST_LOG_PAGE < addr + len) {
 482                 vhost_log_cache_page(dev, vq, page);
 483                 page += 1;
 484         }
 485 }
 486
 487 static __rte_always_inline void
 488 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 489                         uint64_t offset, uint64_t len)
 490 {
 491         vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset, len);
 492 }
 493
 494 static __rte_always_inline void
 495 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 496                      uint64_t offset, uint64_t len)
 497 {
 498         vhost_log_write(dev, vq->log_guest_addr + offset, len);
 499 }
 500
 501 /* Macros for printing using RTE_LOG */
 502 #define RTE_LOGTYPE_VHOST_CONFIG RTE_LOGTYPE_USER1
 503 #define RTE_LOGTYPE_VHOST_DATA   RTE_LOGTYPE_USER1
 504
 505 #ifdef RTE_LIBRTE_VHOST_DEBUG
 506 #define VHOST_MAX_PRINT_BUFF 6072
 507 #define VHOST_LOG_DEBUG(log_type, fmt, args...) \
 508         RTE_LOG(DEBUG, log_type, fmt, ##args)
 509 #define PRINT_PACKET(device, addr, size, header) do { \
 510         char *pkt_addr = (char *)(addr); \
 511         unsigned int index; \
 512         char packet[VHOST_MAX_PRINT_BUFF]; \
 513         \
 514         if ((header)) \
 515                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
 516         else \
 517                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
 518         for (index = 0; index < (size); index++) { \
 519                 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
 520                         "%02hhx ", pkt_addr[index]); \
 521         } \
 522         snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
 523         \
 524         VHOST_LOG_DEBUG(VHOST_DATA, "%s", packet); \
 525 } while (0)
 526 #else
 527 #define VHOST_LOG_DEBUG(log_type, fmt, args...) do {} while (0)
 528 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
 529 #endif
 530
 531 extern uint64_t VHOST_FEATURES;
 532 #define MAX_VHOST_DEVICE        1024
 533 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
 534
 535 /* Convert guest physical address to host physical address */
 536 static __rte_always_inline rte_iova_t
 537 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
 538 {
 539         uint32_t i;
 540         struct guest_page *page;
 541
 542         for (i = 0; i < dev->nr_guest_pages; i++) {
 543                 page = &dev->guest_pages[i];
 544
 545                 if (gpa >= page->guest_phys_addr &&
 546                     gpa + size < page->guest_phys_addr + page->size) {
 547                         return gpa - page->guest_phys_addr +
 548                                page->host_phys_addr;
 549                 }
 550         }
 551
 552         return 0;
 553 }
 554
 555 static __rte_always_inline struct virtio_net *
 556 get_device(int vid)
 557 {
 558         struct virtio_net *dev = vhost_devices[vid];
 559
 560         if (unlikely(!dev)) {
 561                 RTE_LOG(ERR, VHOST_CONFIG,
 562                         "(%d) device not found.\n", vid);
 563         }
 564
 565         return dev;
 566 }
 567
 568 int vhost_new_device(void);
 569 void cleanup_device(struct virtio_net *dev, int destroy);
 570 void reset_device(struct virtio_net *dev);
 571 void vhost_destroy_device(int);
 572 void vhost_destroy_device_notify(struct virtio_net *dev);
 573
 574 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
 575 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
 576
 577 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
 578
 579 void vhost_attach_vdpa_device(int vid, int did);
 580
 581 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
 582 void vhost_enable_dequeue_zero_copy(int vid);
 583 void vhost_set_builtin_virtio_net(int vid, bool enable);
 584
 585 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
 586
 587 /*
 588  * Backend-specific cleanup.
 589  *
 590  * TODO: fix it; we have one backend now
 591  */
 592 void vhost_backend_cleanup(struct virtio_net *dev);
 593
 594 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 595                         uint64_t iova, uint64_t *len, uint8_t perm);
 596 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 597 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 598
 599 static __rte_always_inline uint64_t
 600 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 601                         uint64_t iova, uint64_t *len, uint8_t perm)
 602 {
 603         if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 604                 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
 605
 606         return __vhost_iova_to_vva(dev, vq, iova, len, perm);
 607 }
 608
 609 #define vhost_avail_event(vr) \
 610         (*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
 611 #define vhost_used_event(vr) \
 612         (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
 613
 614 /*
 615  * The following is used with VIRTIO_RING_F_EVENT_IDX.
 616  * Assuming a given event_idx value from the other size, if we have
 617  * just incremented index from old to new_idx, should we trigger an
 618  * event?
 619  */
 620 static __rte_always_inline int
 621 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
 622 {
 623         return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
 624 }
 625
 626 static __rte_always_inline void
 627 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 628 {
 629         /* Flush used->idx update before we read avail->flags. */
 630         rte_smp_mb();
 631
 632         /* Don't kick guest if we don't reach index specified by guest. */
 633         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
 634                 uint16_t old = vq->signalled_used;
 635                 uint16_t new = vq->last_used_idx;
 636                 bool signalled_used_valid = vq->signalled_used_valid;
 637
 638                 vq->signalled_used = new;
 639                 vq->signalled_used_valid = true;
 640
 641                 VHOST_LOG_DEBUG(VHOST_DATA, "%s: used_event_idx=%d, old=%d, new=%d\n",
 642                         __func__,
 643                         vhost_used_event(vq),
 644                         old, new);
 645
 646                 if ((vhost_need_event(vhost_used_event(vq), new, old) &&
 647                                         (vq->callfd >= 0)) ||
 648                                 unlikely(!signalled_used_valid))
 649                         eventfd_write(vq->callfd, (eventfd_t) 1);
 650         } else {
 651                 /* Kick the guest if necessary. */
 652                 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
 653                                 && (vq->callfd >= 0))
 654                         eventfd_write(vq->callfd, (eventfd_t)1);
 655         }
 656 }
 657
 658 static __rte_always_inline void
 659 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
 660 {
 661         uint16_t old, new, off, off_wrap;
 662         bool signalled_used_valid, kick = false;
 663
 664         /* Flush used desc update. */
 665         rte_smp_mb();
 666
 667         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
 668                 if (vq->driver_event->flags !=
 669                                 VRING_EVENT_F_DISABLE)
 670                         kick = true;
 671                 goto kick;
 672         }
 673
 674         old = vq->signalled_used;
 675         new = vq->last_used_idx;
 676         vq->signalled_used = new;
 677         signalled_used_valid = vq->signalled_used_valid;
 678         vq->signalled_used_valid = true;
 679
 680         if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
 681                 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
 682                         kick = true;
 683                 goto kick;
 684         }
 685
 686         if (unlikely(!signalled_used_valid)) {
 687                 kick = true;
 688                 goto kick;
 689         }
 690
 691         rte_smp_rmb();
 692
 693         off_wrap = vq->driver_event->off_wrap;
 694         off = off_wrap & ~(1 << 15);
 695
 696         if (new <= old)
 697                 old -= vq->size;
 698
 699         if (vq->used_wrap_counter != off_wrap >> 15)
 700                 off -= vq->size;
 701
 702         if (vhost_need_event(off, new, old))
 703                 kick = true;
 704 kick:
 705         if (kick)
 706                 eventfd_write(vq->callfd, (eventfd_t)1);
 707 }
 708
 709 static __rte_always_inline void *
 710 alloc_copy_ind_table(struct virtio_net *dev, struct vhost_virtqueue *vq,
 711                 uint64_t desc_addr, uint64_t desc_len)
 712 {
 713         void *idesc;
 714         uint64_t src, dst;
 715         uint64_t len, remain = desc_len;
 716
 717         idesc = rte_malloc(__func__, desc_len, 0);
 718         if (unlikely(!idesc))
 719                 return 0;
 720
 721         dst = (uint64_t)(uintptr_t)idesc;
 722
 723         while (remain) {
 724                 len = remain;
 725                 src = vhost_iova_to_vva(dev, vq, desc_addr, &len,
 726                                 VHOST_ACCESS_RO);
 727                 if (unlikely(!src || !len)) {
 728                         rte_free(idesc);
 729                         return 0;
 730                 }
 731
 732                 rte_memcpy((void *)(uintptr_t)dst, (void *)(uintptr_t)src, len);
 733
 734                 remain -= len;
 735                 dst += len;
 736                 desc_addr += len;
 737         }
 738
 739         return idesc;
 740 }
 741
 742 static __rte_always_inline void
 743 free_ind_table(void *idesc)
 744 {
 745         rte_free(idesc);
 746 }
 747
 748 static __rte_always_inline void
 749 restore_mbuf(struct rte_mbuf *m)
 750 {
 751         uint32_t mbuf_size, priv_size;
 752
 753         while (m) {
 754                 priv_size = rte_pktmbuf_priv_size(m->pool);
 755                 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
 756                 /* start of buffer is after mbuf structure and priv data */
 757
 758                 m->buf_addr = (char *)m + mbuf_size;
 759                 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
 760                 m = m->next;
 761         }
 762 }
 763
 764 static __rte_always_inline bool
 765 mbuf_is_consumed(struct rte_mbuf *m)
 766 {
 767         while (m) {
 768                 if (rte_mbuf_refcnt_read(m) > 1)
 769                         return false;
 770                 m = m->next;
 771         }
 772
 773         return true;
 774 }
 775
 776 static __rte_always_inline void
 777 put_zmbuf(struct zcopy_mbuf *zmbuf)
 778 {
 779         zmbuf->in_use = 0;
 780 }
 781
 782 #endif /* _VHOST_NET_CDEV_H_ */