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_SINGLE_DEQUEUE_FLAG (VRING_DESC_F_NEXT | \
  43                                          VRING_DESC_F_INDIRECT)
  44
  45 #define PACKED_BATCH_SIZE (RTE_CACHE_LINE_SIZE / \
  46                             sizeof(struct vring_packed_desc))
  47 #define PACKED_BATCH_MASK (PACKED_BATCH_SIZE - 1)
  48
  49 #ifdef VHOST_GCC_UNROLL_PRAGMA
  50 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("GCC unroll 4") \
  51         for (iter = val; iter < size; iter++)
  52 #endif
  53
  54 #ifdef VHOST_CLANG_UNROLL_PRAGMA
  55 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll 4") \
  56         for (iter = val; iter < size; iter++)
  57 #endif
  58
  59 #ifdef VHOST_ICC_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 #ifndef vhost_for_each_try_unroll
  65 #define vhost_for_each_try_unroll(iter, val, num) \
  66         for (iter = val; iter < num; iter++)
  67 #endif
  68
  69 /**
  70  * Structure contains buffer address, length and descriptor index
  71  * from vring to do scatter RX.
  72  */
  73 struct buf_vector {
  74         uint64_t buf_iova;
  75         uint64_t buf_addr;
  76         uint32_t buf_len;
  77         uint32_t desc_idx;
  78 };
  79
  80 /*
  81  * A structure to hold some fields needed in zero copy code path,
  82  * mainly for associating an mbuf with the right desc_idx.
  83  */
  84 struct zcopy_mbuf {
  85         struct rte_mbuf *mbuf;
  86         uint32_t desc_idx;
  87         uint16_t desc_count;
  88         uint16_t in_use;
  89
  90         TAILQ_ENTRY(zcopy_mbuf) next;
  91 };
  92 TAILQ_HEAD(zcopy_mbuf_list, zcopy_mbuf);
  93
  94 /*
  95  * Structure contains the info for each batched memory copy.
  96  */
  97 struct batch_copy_elem {
  98         void *dst;
  99         void *src;
 100         uint32_t len;
 101         uint64_t log_addr;
 102 };
 103
 104 /*
 105  * Structure that contains the info for batched dirty logging.
 106  */
 107 struct log_cache_entry {
 108         uint32_t offset;
 109         unsigned long val;
 110 };
 111
 112 struct vring_used_elem_packed {
 113         uint16_t id;
 114         uint32_t len;
 115         uint32_t count;
 116 };
 117
 118 /**
 119  * Structure contains variables relevant to RX/TX virtqueues.
 120  */
 121 struct vhost_virtqueue {
 122         union {
 123                 struct vring_desc       *desc;
 124                 struct vring_packed_desc   *desc_packed;
 125         };
 126         union {
 127                 struct vring_avail      *avail;
 128                 struct vring_packed_desc_event *driver_event;
 129         };
 130         union {
 131                 struct vring_used       *used;
 132                 struct vring_packed_desc_event *device_event;
 133         };
 134         uint32_t                size;
 135
 136         uint16_t                last_avail_idx;
 137         uint16_t                last_used_idx;
 138         /* Last used index we notify to front end. */
 139         uint16_t                signalled_used;
 140         bool                    signalled_used_valid;
 141 #define VIRTIO_INVALID_EVENTFD          (-1)
 142 #define VIRTIO_UNINITIALIZED_EVENTFD    (-2)
 143
 144         /* Backend value to determine if device should started/stopped */
 145         int                     backend;
 146         int                     enabled;
 147         int                     access_ok;
 148         rte_spinlock_t          access_lock;
 149
 150         /* Used to notify the guest (trigger interrupt) */
 151         int                     callfd;
 152         /* Currently unused as polling mode is enabled */
 153         int                     kickfd;
 154
 155         /* Physical address of used ring, for logging */
 156         uint64_t                log_guest_addr;
 157
 158         /* inflight share memory info */
 159         union {
 160                 struct rte_vhost_inflight_info_split *inflight_split;
 161                 struct rte_vhost_inflight_info_packed *inflight_packed;
 162         };
 163         struct rte_vhost_resubmit_info *resubmit_inflight;
 164         uint64_t                global_counter;
 165
 166         uint16_t                nr_zmbuf;
 167         uint16_t                zmbuf_size;
 168         uint16_t                last_zmbuf_idx;
 169         struct zcopy_mbuf       *zmbufs;
 170         struct zcopy_mbuf_list  zmbuf_list;
 171
 172         union {
 173                 struct vring_used_elem  *shadow_used_split;
 174                 struct vring_used_elem_packed *shadow_used_packed;
 175         };
 176         uint16_t                shadow_used_idx;
 177         /* Record packed ring enqueue latest desc cache aligned index */
 178         uint16_t                shadow_aligned_idx;
 179         struct vhost_vring_addr ring_addrs;
 180
 181         struct batch_copy_elem  *batch_copy_elems;
 182         uint16_t                batch_copy_nb_elems;
 183         bool                    used_wrap_counter;
 184         bool                    avail_wrap_counter;
 185
 186         struct log_cache_entry log_cache[VHOST_LOG_CACHE_NR];
 187         uint16_t log_cache_nb_elem;
 188
 189         rte_rwlock_t    iotlb_lock;
 190         rte_rwlock_t    iotlb_pending_lock;
 191         struct rte_mempool *iotlb_pool;
 192         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
 193         int                             iotlb_cache_nr;
 194         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
 195 } __rte_cache_aligned;
 196
 197 /* Old kernels have no such macros defined */
 198 #ifndef VIRTIO_NET_F_GUEST_ANNOUNCE
 199  #define VIRTIO_NET_F_GUEST_ANNOUNCE 21
 200 #endif
 201
 202 #ifndef VIRTIO_NET_F_MQ
 203  #define VIRTIO_NET_F_MQ                22
 204 #endif
 205
 206 #define VHOST_MAX_VRING                 0x100
 207 #define VHOST_MAX_QUEUE_PAIRS           0x80
 208
 209 #ifndef VIRTIO_NET_F_MTU
 210  #define VIRTIO_NET_F_MTU 3
 211 #endif
 212
 213 #ifndef VIRTIO_F_ANY_LAYOUT
 214  #define VIRTIO_F_ANY_LAYOUT            27
 215 #endif
 216
 217 /* Declare IOMMU related bits for older kernels */
 218 #ifndef VIRTIO_F_IOMMU_PLATFORM
 219
 220 #define VIRTIO_F_IOMMU_PLATFORM 33
 221
 222 struct vhost_iotlb_msg {
 223         __u64 iova;
 224         __u64 size;
 225         __u64 uaddr;
 226 #define VHOST_ACCESS_RO      0x1
 227 #define VHOST_ACCESS_WO      0x2
 228 #define VHOST_ACCESS_RW      0x3
 229         __u8 perm;
 230 #define VHOST_IOTLB_MISS           1
 231 #define VHOST_IOTLB_UPDATE         2
 232 #define VHOST_IOTLB_INVALIDATE     3
 233 #define VHOST_IOTLB_ACCESS_FAIL    4
 234         __u8 type;
 235 };
 236
 237 #define VHOST_IOTLB_MSG 0x1
 238
 239 struct vhost_msg {
 240         int type;
 241         union {
 242                 struct vhost_iotlb_msg iotlb;
 243                 __u8 padding[64];
 244         };
 245 };
 246 #endif
 247
 248 /*
 249  * Define virtio 1.0 for older kernels
 250  */
 251 #ifndef VIRTIO_F_VERSION_1
 252  #define VIRTIO_F_VERSION_1 32
 253 #endif
 254
 255 /* Declare packed ring related bits for older kernels */
 256 #ifndef VIRTIO_F_RING_PACKED
 257
 258 #define VIRTIO_F_RING_PACKED 34
 259
 260 struct vring_packed_desc {
 261         uint64_t addr;
 262         uint32_t len;
 263         uint16_t id;
 264         uint16_t flags;
 265 };
 266
 267 struct vring_packed_desc_event {
 268         uint16_t off_wrap;
 269         uint16_t flags;
 270 };
 271 #endif
 272
 273 /*
 274  * Declare below packed ring defines unconditionally
 275  * as Kernel header might use different names.
 276  */
 277 #define VRING_DESC_F_AVAIL      (1ULL << 7)
 278 #define VRING_DESC_F_USED       (1ULL << 15)
 279
 280 #define VRING_EVENT_F_ENABLE 0x0
 281 #define VRING_EVENT_F_DISABLE 0x1
 282 #define VRING_EVENT_F_DESC 0x2
 283
 284 /*
 285  * Available and used descs are in same order
 286  */
 287 #ifndef VIRTIO_F_IN_ORDER
 288 #define VIRTIO_F_IN_ORDER      35
 289 #endif
 290
 291 /* Features supported by this builtin vhost-user net driver. */
 292 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
 293                                 (1ULL << VIRTIO_F_ANY_LAYOUT) | \
 294                                 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
 295                                 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
 296                                 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
 297                                 (1ULL << VIRTIO_NET_F_MQ)      | \
 298                                 (1ULL << VIRTIO_F_VERSION_1)   | \
 299                                 (1ULL << VHOST_F_LOG_ALL)      | \
 300                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
 301                                 (1ULL << VIRTIO_NET_F_GSO) | \
 302                                 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
 303                                 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
 304                                 (1ULL << VIRTIO_NET_F_HOST_UFO) | \
 305                                 (1ULL << VIRTIO_NET_F_HOST_ECN) | \
 306                                 (1ULL << VIRTIO_NET_F_CSUM)    | \
 307                                 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
 308                                 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
 309                                 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
 310                                 (1ULL << VIRTIO_NET_F_GUEST_UFO) | \
 311                                 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
 312                                 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
 313                                 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \
 314                                 (1ULL << VIRTIO_NET_F_MTU)  | \
 315                                 (1ULL << VIRTIO_F_IN_ORDER) | \
 316                                 (1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
 317                                 (1ULL << VIRTIO_F_RING_PACKED))
 318
 319
 320 struct guest_page {
 321         uint64_t guest_phys_addr;
 322         uint64_t host_phys_addr;
 323         uint64_t size;
 324 };
 325
 326 struct inflight_mem_info {
 327         int             fd;
 328         void            *addr;
 329         uint64_t        size;
 330 };
 331
 332 /**
 333  * Device structure contains all configuration information relating
 334  * to the device.
 335  */
 336 struct virtio_net {
 337         /* Frontend (QEMU) memory and memory region information */
 338         struct rte_vhost_memory *mem;
 339         uint64_t                features;
 340         uint64_t                protocol_features;
 341         int                     vid;
 342         uint32_t                flags;
 343         uint16_t                vhost_hlen;
 344         /* to tell if we need broadcast rarp packet */
 345         rte_atomic16_t          broadcast_rarp;
 346         uint32_t                nr_vring;
 347         int                     dequeue_zero_copy;
 348         int                     extbuf;
 349         int                     linearbuf;
 350         struct vhost_virtqueue  *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
 351         struct inflight_mem_info *inflight_info;
 352 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
 353         char                    ifname[IF_NAME_SZ];
 354         uint64_t                log_size;
 355         uint64_t                log_base;
 356         uint64_t                log_addr;
 357         struct rte_ether_addr   mac;
 358         uint16_t                mtu;
 359
 360         struct vhost_device_ops const *notify_ops;
 361
 362         uint32_t                nr_guest_pages;
 363         uint32_t                max_guest_pages;
 364         struct guest_page       *guest_pages;
 365
 366         int                     slave_req_fd;
 367         rte_spinlock_t          slave_req_lock;
 368
 369         int                     postcopy_ufd;
 370         int                     postcopy_listening;
 371
 372         /*
 373          * Device id to identify a specific backend device.
 374          * It's set to -1 for the default software implementation.
 375          */
 376         int                     vdpa_dev_id;
 377
 378         /* context data for the external message handlers */
 379         void                    *extern_data;
 380         /* pre and post vhost user message handlers for the device */
 381         struct rte_vhost_user_extern_ops extern_ops;
 382 } __rte_cache_aligned;
 383
 384 static __rte_always_inline bool
 385 vq_is_packed(struct virtio_net *dev)
 386 {
 387         return dev->features & (1ull << VIRTIO_F_RING_PACKED);
 388 }
 389
 390 static inline bool
 391 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
 392 {
 393         uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
 394
 395         return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
 396                 wrap_counter != !!(flags & VRING_DESC_F_USED);
 397 }
 398
 399 static inline void
 400 vq_inc_last_used_packed(struct vhost_virtqueue *vq, uint16_t num)
 401 {
 402         vq->last_used_idx += num;
 403         if (vq->last_used_idx >= vq->size) {
 404                 vq->used_wrap_counter ^= 1;
 405                 vq->last_used_idx -= vq->size;
 406         }
 407 }
 408
 409 static inline void
 410 vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
 411 {
 412         vq->last_avail_idx += num;
 413         if (vq->last_avail_idx >= vq->size) {
 414                 vq->avail_wrap_counter ^= 1;
 415                 vq->last_avail_idx -= vq->size;
 416         }
 417 }
 418
 419 void __vhost_log_cache_write(struct virtio_net *dev,
 420                 struct vhost_virtqueue *vq,
 421                 uint64_t addr, uint64_t len);
 422 void __vhost_log_cache_write_iova(struct virtio_net *dev,
 423                 struct vhost_virtqueue *vq,
 424                 uint64_t iova, uint64_t len);
 425 void __vhost_log_cache_sync(struct virtio_net *dev,
 426                 struct vhost_virtqueue *vq);
 427 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
 428 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 429                             uint64_t iova, uint64_t len);
 430
 431 static __rte_always_inline void
 432 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
 433 {
 434         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 435                 __vhost_log_write(dev, addr, len);
 436 }
 437
 438 static __rte_always_inline void
 439 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
 440 {
 441         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 442                 __vhost_log_cache_sync(dev, vq);
 443 }
 444
 445 static __rte_always_inline void
 446 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
 447                         uint64_t addr, uint64_t len)
 448 {
 449         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 450                 __vhost_log_cache_write(dev, vq, addr, len);
 451 }
 452
 453 static __rte_always_inline void
 454 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 455                         uint64_t offset, uint64_t len)
 456 {
 457         vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset, len);
 458 }
 459
 460 static __rte_always_inline void
 461 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 462                      uint64_t offset, uint64_t len)
 463 {
 464         vhost_log_write(dev, vq->log_guest_addr + offset, len);
 465 }
 466
 467 static __rte_always_inline void
 468 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 469                            uint64_t iova, uint64_t len)
 470 {
 471         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 472                 return;
 473
 474         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 475                 __vhost_log_cache_write_iova(dev, vq, iova, len);
 476         else
 477                 __vhost_log_cache_write(dev, vq, iova, len);
 478 }
 479
 480 static __rte_always_inline void
 481 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 482                            uint64_t iova, uint64_t len)
 483 {
 484         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 485                 return;
 486
 487         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 488                 __vhost_log_write_iova(dev, vq, iova, len);
 489         else
 490                 __vhost_log_write(dev, iova, len);
 491 }
 492
 493 /* Macros for printing using RTE_LOG */
 494 #define RTE_LOGTYPE_VHOST_CONFIG RTE_LOGTYPE_USER1
 495 #define RTE_LOGTYPE_VHOST_DATA   RTE_LOGTYPE_USER1
 496
 497 #ifdef RTE_LIBRTE_VHOST_DEBUG
 498 #define VHOST_MAX_PRINT_BUFF 6072
 499 #define VHOST_LOG_DEBUG(log_type, fmt, args...) \
 500         RTE_LOG(DEBUG, log_type, fmt, ##args)
 501 #define PRINT_PACKET(device, addr, size, header) do { \
 502         char *pkt_addr = (char *)(addr); \
 503         unsigned int index; \
 504         char packet[VHOST_MAX_PRINT_BUFF]; \
 505         \
 506         if ((header)) \
 507                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
 508         else \
 509                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
 510         for (index = 0; index < (size); index++) { \
 511                 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
 512                         "%02hhx ", pkt_addr[index]); \
 513         } \
 514         snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
 515         \
 516         VHOST_LOG_DEBUG(VHOST_DATA, "%s", packet); \
 517 } while (0)
 518 #else
 519 #define VHOST_LOG_DEBUG(log_type, fmt, args...) do {} while (0)
 520 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
 521 #endif
 522
 523 extern uint64_t VHOST_FEATURES;
 524 #define MAX_VHOST_DEVICE        1024
 525 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
 526
 527 /* Convert guest physical address to host physical address */
 528 static __rte_always_inline rte_iova_t
 529 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
 530 {
 531         uint32_t i;
 532         struct guest_page *page;
 533
 534         for (i = 0; i < dev->nr_guest_pages; i++) {
 535                 page = &dev->guest_pages[i];
 536
 537                 if (gpa >= page->guest_phys_addr &&
 538                     gpa + size < page->guest_phys_addr + page->size) {
 539                         return gpa - page->guest_phys_addr +
 540                                page->host_phys_addr;
 541                 }
 542         }
 543
 544         return 0;
 545 }
 546
 547 static __rte_always_inline uint64_t
 548 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
 549 {
 550         struct rte_vhost_mem_region *r;
 551         uint32_t i;
 552
 553         if (unlikely(!dev || !dev->mem))
 554                 return 0;
 555
 556         for (i = 0; i < dev->mem->nregions; i++) {
 557                 r = &dev->mem->regions[i];
 558
 559                 if (vva >= r->host_user_addr &&
 560                     vva + len <  r->host_user_addr + r->size) {
 561                         return r->guest_phys_addr + vva - r->host_user_addr;
 562                 }
 563         }
 564         return 0;
 565 }
 566
 567 static __rte_always_inline struct virtio_net *
 568 get_device(int vid)
 569 {
 570         struct virtio_net *dev = vhost_devices[vid];
 571
 572         if (unlikely(!dev)) {
 573                 RTE_LOG(ERR, VHOST_CONFIG,
 574                         "(%d) device not found.\n", vid);
 575         }
 576
 577         return dev;
 578 }
 579
 580 int vhost_new_device(void);
 581 void cleanup_device(struct virtio_net *dev, int destroy);
 582 void reset_device(struct virtio_net *dev);
 583 void vhost_destroy_device(int);
 584 void vhost_destroy_device_notify(struct virtio_net *dev);
 585
 586 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
 587 void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
 588 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
 589
 590 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
 591
 592 void vhost_attach_vdpa_device(int vid, int did);
 593
 594 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
 595 void vhost_enable_dequeue_zero_copy(int vid);
 596 void vhost_set_builtin_virtio_net(int vid, bool enable);
 597 void vhost_enable_extbuf(int vid);
 598 void vhost_enable_linearbuf(int vid);
 599
 600 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
 601
 602 /*
 603  * Backend-specific cleanup.
 604  *
 605  * TODO: fix it; we have one backend now
 606  */
 607 void vhost_backend_cleanup(struct virtio_net *dev);
 608
 609 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 610                         uint64_t iova, uint64_t *len, uint8_t perm);
 611 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
 612                         struct vhost_virtqueue *vq,
 613                         uint64_t desc_addr, uint64_t desc_len);
 614 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 615 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 616
 617 static __rte_always_inline uint64_t
 618 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 619                         uint64_t iova, uint64_t *len, uint8_t perm)
 620 {
 621         if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 622                 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
 623
 624         return __vhost_iova_to_vva(dev, vq, iova, len, perm);
 625 }
 626
 627 #define vhost_avail_event(vr) \
 628         (*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
 629 #define vhost_used_event(vr) \
 630         (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
 631
 632 /*
 633  * The following is used with VIRTIO_RING_F_EVENT_IDX.
 634  * Assuming a given event_idx value from the other size, if we have
 635  * just incremented index from old to new_idx, should we trigger an
 636  * event?
 637  */
 638 static __rte_always_inline int
 639 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
 640 {
 641         return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
 642 }
 643
 644 static __rte_always_inline void
 645 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 646 {
 647         /* Flush used->idx update before we read avail->flags. */
 648         rte_smp_mb();
 649
 650         /* Don't kick guest if we don't reach index specified by guest. */
 651         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
 652                 uint16_t old = vq->signalled_used;
 653                 uint16_t new = vq->last_used_idx;
 654                 bool signalled_used_valid = vq->signalled_used_valid;
 655
 656                 vq->signalled_used = new;
 657                 vq->signalled_used_valid = true;
 658
 659                 VHOST_LOG_DEBUG(VHOST_DATA, "%s: used_event_idx=%d, old=%d, new=%d\n",
 660                         __func__,
 661                         vhost_used_event(vq),
 662                         old, new);
 663
 664                 if ((vhost_need_event(vhost_used_event(vq), new, old) &&
 665                                         (vq->callfd >= 0)) ||
 666                                 unlikely(!signalled_used_valid)) {
 667                         eventfd_write(vq->callfd, (eventfd_t) 1);
 668                         if (dev->notify_ops->guest_notified)
 669                                 dev->notify_ops->guest_notified(dev->vid);
 670                 }
 671         } else {
 672                 /* Kick the guest if necessary. */
 673                 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
 674                                 && (vq->callfd >= 0)) {
 675                         eventfd_write(vq->callfd, (eventfd_t)1);
 676                         if (dev->notify_ops->guest_notified)
 677                                 dev->notify_ops->guest_notified(dev->vid);
 678                 }
 679         }
 680 }
 681
 682 static __rte_always_inline void
 683 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
 684 {
 685         uint16_t old, new, off, off_wrap;
 686         bool signalled_used_valid, kick = false;
 687
 688         /* Flush used desc update. */
 689         rte_smp_mb();
 690
 691         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
 692                 if (vq->driver_event->flags !=
 693                                 VRING_EVENT_F_DISABLE)
 694                         kick = true;
 695                 goto kick;
 696         }
 697
 698         old = vq->signalled_used;
 699         new = vq->last_used_idx;
 700         vq->signalled_used = new;
 701         signalled_used_valid = vq->signalled_used_valid;
 702         vq->signalled_used_valid = true;
 703
 704         if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
 705                 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
 706                         kick = true;
 707                 goto kick;
 708         }
 709
 710         if (unlikely(!signalled_used_valid)) {
 711                 kick = true;
 712                 goto kick;
 713         }
 714
 715         rte_smp_rmb();
 716
 717         off_wrap = vq->driver_event->off_wrap;
 718         off = off_wrap & ~(1 << 15);
 719
 720         if (new <= old)
 721                 old -= vq->size;
 722
 723         if (vq->used_wrap_counter != off_wrap >> 15)
 724                 off -= vq->size;
 725
 726         if (vhost_need_event(off, new, old))
 727                 kick = true;
 728 kick:
 729         if (kick) {
 730                 eventfd_write(vq->callfd, (eventfd_t)1);
 731                 if (dev->notify_ops->guest_notified)
 732                         dev->notify_ops->guest_notified(dev->vid);
 733         }
 734 }
 735
 736 static __rte_always_inline void
 737 free_ind_table(void *idesc)
 738 {
 739         rte_free(idesc);
 740 }
 741
 742 static __rte_always_inline void
 743 restore_mbuf(struct rte_mbuf *m)
 744 {
 745         uint32_t mbuf_size, priv_size;
 746
 747         while (m) {
 748                 priv_size = rte_pktmbuf_priv_size(m->pool);
 749                 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
 750                 /* start of buffer is after mbuf structure and priv data */
 751
 752                 m->buf_addr = (char *)m + mbuf_size;
 753                 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
 754                 m = m->next;
 755         }
 756 }
 757
 758 static __rte_always_inline bool
 759 mbuf_is_consumed(struct rte_mbuf *m)
 760 {
 761         while (m) {
 762                 if (rte_mbuf_refcnt_read(m) > 1)
 763                         return false;
 764                 m = m->next;
 765         }
 766
 767         return true;
 768 }
 769
 770 static __rte_always_inline void
 771 put_zmbuf(struct zcopy_mbuf *zmbuf)
 772 {
 773         zmbuf->in_use = 0;
 774 }
 775
 776 #endif /* _VHOST_NET_CDEV_H_ */