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_SINGLE_DEQUEUE_FLAG (VRING_DESC_F_NEXT | \
  46                                          VRING_DESC_F_INDIRECT)
  47
  48 #define PACKED_BATCH_SIZE (RTE_CACHE_LINE_SIZE / \
  49                             sizeof(struct vring_packed_desc))
  50 #define PACKED_BATCH_MASK (PACKED_BATCH_SIZE - 1)
  51
  52 #ifdef VHOST_GCC_UNROLL_PRAGMA
  53 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("GCC unroll 4") \
  54         for (iter = val; iter < size; iter++)
  55 #endif
  56
  57 #ifdef VHOST_CLANG_UNROLL_PRAGMA
  58 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll 4") \
  59         for (iter = val; iter < size; iter++)
  60 #endif
  61
  62 #ifdef VHOST_ICC_UNROLL_PRAGMA
  63 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll (4)") \
  64         for (iter = val; iter < size; iter++)
  65 #endif
  66
  67 #ifndef vhost_for_each_try_unroll
  68 #define vhost_for_each_try_unroll(iter, val, num) \
  69         for (iter = val; iter < num; iter++)
  70 #endif
  71
  72 /**
  73  * Structure contains buffer address, length and descriptor index
  74  * from vring to do scatter RX.
  75  */
  76 struct buf_vector {
  77         uint64_t buf_iova;
  78         uint64_t buf_addr;
  79         uint32_t buf_len;
  80         uint32_t desc_idx;
  81 };
  82
  83 /*
  84  * A structure to hold some fields needed in zero copy code path,
  85  * mainly for associating an mbuf with the right desc_idx.
  86  */
  87 struct zcopy_mbuf {
  88         struct rte_mbuf *mbuf;
  89         uint32_t desc_idx;
  90         uint16_t desc_count;
  91         uint16_t in_use;
  92
  93         TAILQ_ENTRY(zcopy_mbuf) next;
  94 };
  95 TAILQ_HEAD(zcopy_mbuf_list, zcopy_mbuf);
  96
  97 /*
  98  * Structure contains the info for each batched memory copy.
  99  */
 100 struct batch_copy_elem {
 101         void *dst;
 102         void *src;
 103         uint32_t len;
 104         uint64_t log_addr;
 105 };
 106
 107 /*
 108  * Structure that contains the info for batched dirty logging.
 109  */
 110 struct log_cache_entry {
 111         uint32_t offset;
 112         unsigned long val;
 113 };
 114
 115 struct vring_used_elem_packed {
 116         uint16_t id;
 117         uint32_t len;
 118         uint32_t count;
 119 };
 120
 121 /**
 122  * Structure contains variables relevant to RX/TX virtqueues.
 123  */
 124 struct vhost_virtqueue {
 125         union {
 126                 struct vring_desc       *desc;
 127                 struct vring_packed_desc   *desc_packed;
 128         };
 129         union {
 130                 struct vring_avail      *avail;
 131                 struct vring_packed_desc_event *driver_event;
 132         };
 133         union {
 134                 struct vring_used       *used;
 135                 struct vring_packed_desc_event *device_event;
 136         };
 137         uint32_t                size;
 138
 139         uint16_t                last_avail_idx;
 140         uint16_t                last_used_idx;
 141         /* Last used index we notify to front end. */
 142         uint16_t                signalled_used;
 143         bool                    signalled_used_valid;
 144 #define VIRTIO_INVALID_EVENTFD          (-1)
 145 #define VIRTIO_UNINITIALIZED_EVENTFD    (-2)
 146
 147         /* Backend value to determine if device should started/stopped */
 148         int                     backend;
 149         int                     enabled;
 150         int                     access_ok;
 151         rte_spinlock_t          access_lock;
 152
 153         /* Used to notify the guest (trigger interrupt) */
 154         int                     callfd;
 155         /* Currently unused as polling mode is enabled */
 156         int                     kickfd;
 157
 158         /* Physical address of used ring, for logging */
 159         uint64_t                log_guest_addr;
 160
 161         /* inflight share memory info */
 162         union {
 163                 struct rte_vhost_inflight_info_split *inflight_split;
 164                 struct rte_vhost_inflight_info_packed *inflight_packed;
 165         };
 166         struct rte_vhost_resubmit_info *resubmit_inflight;
 167         uint64_t                global_counter;
 168
 169         uint16_t                nr_zmbuf;
 170         uint16_t                zmbuf_size;
 171         uint16_t                last_zmbuf_idx;
 172         struct zcopy_mbuf       *zmbufs;
 173         struct zcopy_mbuf_list  zmbuf_list;
 174
 175         union {
 176                 struct vring_used_elem  *shadow_used_split;
 177                 struct vring_used_elem_packed *shadow_used_packed;
 178         };
 179         uint16_t                shadow_used_idx;
 180         /* Record packed ring enqueue latest desc cache aligned index */
 181         uint16_t                shadow_aligned_idx;
 182         struct vhost_vring_addr ring_addrs;
 183
 184         struct batch_copy_elem  *batch_copy_elems;
 185         uint16_t                batch_copy_nb_elems;
 186         bool                    used_wrap_counter;
 187         bool                    avail_wrap_counter;
 188
 189         struct log_cache_entry log_cache[VHOST_LOG_CACHE_NR];
 190         uint16_t log_cache_nb_elem;
 191
 192         rte_rwlock_t    iotlb_lock;
 193         rte_rwlock_t    iotlb_pending_lock;
 194         struct rte_mempool *iotlb_pool;
 195         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
 196         int                             iotlb_cache_nr;
 197         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
 198 } __rte_cache_aligned;
 199
 200 /* Old kernels have no such macros defined */
 201 #ifndef VIRTIO_NET_F_GUEST_ANNOUNCE
 202  #define VIRTIO_NET_F_GUEST_ANNOUNCE 21
 203 #endif
 204
 205 #ifndef VIRTIO_NET_F_MQ
 206  #define VIRTIO_NET_F_MQ                22
 207 #endif
 208
 209 #define VHOST_MAX_VRING                 0x100
 210 #define VHOST_MAX_QUEUE_PAIRS           0x80
 211
 212 #ifndef VIRTIO_NET_F_MTU
 213  #define VIRTIO_NET_F_MTU 3
 214 #endif
 215
 216 #ifndef VIRTIO_F_ANY_LAYOUT
 217  #define VIRTIO_F_ANY_LAYOUT            27
 218 #endif
 219
 220 /* Declare IOMMU related bits for older kernels */
 221 #ifndef VIRTIO_F_IOMMU_PLATFORM
 222
 223 #define VIRTIO_F_IOMMU_PLATFORM 33
 224
 225 struct vhost_iotlb_msg {
 226         __u64 iova;
 227         __u64 size;
 228         __u64 uaddr;
 229 #define VHOST_ACCESS_RO      0x1
 230 #define VHOST_ACCESS_WO      0x2
 231 #define VHOST_ACCESS_RW      0x3
 232         __u8 perm;
 233 #define VHOST_IOTLB_MISS           1
 234 #define VHOST_IOTLB_UPDATE         2
 235 #define VHOST_IOTLB_INVALIDATE     3
 236 #define VHOST_IOTLB_ACCESS_FAIL    4
 237         __u8 type;
 238 };
 239
 240 #define VHOST_IOTLB_MSG 0x1
 241
 242 struct vhost_msg {
 243         int type;
 244         union {
 245                 struct vhost_iotlb_msg iotlb;
 246                 __u8 padding[64];
 247         };
 248 };
 249 #endif
 250
 251 /*
 252  * Define virtio 1.0 for older kernels
 253  */
 254 #ifndef VIRTIO_F_VERSION_1
 255  #define VIRTIO_F_VERSION_1 32
 256 #endif
 257
 258 /* Declare packed ring related bits for older kernels */
 259 #ifndef VIRTIO_F_RING_PACKED
 260
 261 #define VIRTIO_F_RING_PACKED 34
 262
 263 struct vring_packed_desc {
 264         uint64_t addr;
 265         uint32_t len;
 266         uint16_t id;
 267         uint16_t flags;
 268 };
 269
 270 struct vring_packed_desc_event {
 271         uint16_t off_wrap;
 272         uint16_t flags;
 273 };
 274 #endif
 275
 276 /*
 277  * Declare below packed ring defines unconditionally
 278  * as Kernel header might use different names.
 279  */
 280 #define VRING_DESC_F_AVAIL      (1ULL << 7)
 281 #define VRING_DESC_F_USED       (1ULL << 15)
 282
 283 #define VRING_EVENT_F_ENABLE 0x0
 284 #define VRING_EVENT_F_DISABLE 0x1
 285 #define VRING_EVENT_F_DESC 0x2
 286
 287 /*
 288  * Available and used descs are in same order
 289  */
 290 #ifndef VIRTIO_F_IN_ORDER
 291 #define VIRTIO_F_IN_ORDER      35
 292 #endif
 293
 294 /* Features supported by this builtin vhost-user net driver. */
 295 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
 296                                 (1ULL << VIRTIO_F_ANY_LAYOUT) | \
 297                                 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
 298                                 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
 299                                 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
 300                                 (1ULL << VIRTIO_NET_F_MQ)      | \
 301                                 (1ULL << VIRTIO_F_VERSION_1)   | \
 302                                 (1ULL << VHOST_F_LOG_ALL)      | \
 303                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
 304                                 (1ULL << VIRTIO_NET_F_GSO) | \
 305                                 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
 306                                 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
 307                                 (1ULL << VIRTIO_NET_F_HOST_UFO) | \
 308                                 (1ULL << VIRTIO_NET_F_HOST_ECN) | \
 309                                 (1ULL << VIRTIO_NET_F_CSUM)    | \
 310                                 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
 311                                 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
 312                                 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
 313                                 (1ULL << VIRTIO_NET_F_GUEST_UFO) | \
 314                                 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
 315                                 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
 316                                 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \
 317                                 (1ULL << VIRTIO_NET_F_MTU)  | \
 318                                 (1ULL << VIRTIO_F_IN_ORDER) | \
 319                                 (1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
 320                                 (1ULL << VIRTIO_F_RING_PACKED))
 321
 322
 323 struct guest_page {
 324         uint64_t guest_phys_addr;
 325         uint64_t host_phys_addr;
 326         uint64_t size;
 327 };
 328
 329 struct inflight_mem_info {
 330         int             fd;
 331         void            *addr;
 332         uint64_t        size;
 333 };
 334
 335 /**
 336  * Device structure contains all configuration information relating
 337  * to the device.
 338  */
 339 struct virtio_net {
 340         /* Frontend (QEMU) memory and memory region information */
 341         struct rte_vhost_memory *mem;
 342         uint64_t                features;
 343         uint64_t                protocol_features;
 344         int                     vid;
 345         uint32_t                flags;
 346         uint16_t                vhost_hlen;
 347         /* to tell if we need broadcast rarp packet */
 348         rte_atomic16_t          broadcast_rarp;
 349         uint32_t                nr_vring;
 350         int                     dequeue_zero_copy;
 351         int                     extbuf;
 352         int                     linearbuf;
 353         struct vhost_virtqueue  *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
 354         struct inflight_mem_info *inflight_info;
 355 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
 356         char                    ifname[IF_NAME_SZ];
 357         uint64_t                log_size;
 358         uint64_t                log_base;
 359         uint64_t                log_addr;
 360         struct rte_ether_addr   mac;
 361         uint16_t                mtu;
 362
 363         struct vhost_device_ops const *notify_ops;
 364
 365         uint32_t                nr_guest_pages;
 366         uint32_t                max_guest_pages;
 367         struct guest_page       *guest_pages;
 368
 369         int                     slave_req_fd;
 370         rte_spinlock_t          slave_req_lock;
 371
 372         int                     postcopy_ufd;
 373         int                     postcopy_listening;
 374
 375         /*
 376          * Device id to identify a specific backend device.
 377          * It's set to -1 for the default software implementation.
 378          */
 379         int                     vdpa_dev_id;
 380
 381         /* context data for the external message handlers */
 382         void                    *extern_data;
 383         /* pre and post vhost user message handlers for the device */
 384         struct rte_vhost_user_extern_ops extern_ops;
 385 } __rte_cache_aligned;
 386
 387 static __rte_always_inline bool
 388 vq_is_packed(struct virtio_net *dev)
 389 {
 390         return dev->features & (1ull << VIRTIO_F_RING_PACKED);
 391 }
 392
 393 static inline bool
 394 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
 395 {
 396         uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
 397
 398         return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
 399                 wrap_counter != !!(flags & VRING_DESC_F_USED);
 400 }
 401
 402 static inline void
 403 vq_inc_last_used_packed(struct vhost_virtqueue *vq, uint16_t num)
 404 {
 405         vq->last_used_idx += num;
 406         if (vq->last_used_idx >= vq->size) {
 407                 vq->used_wrap_counter ^= 1;
 408                 vq->last_used_idx -= vq->size;
 409         }
 410 }
 411
 412 static inline void
 413 vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
 414 {
 415         vq->last_avail_idx += num;
 416         if (vq->last_avail_idx >= vq->size) {
 417                 vq->avail_wrap_counter ^= 1;
 418                 vq->last_avail_idx -= vq->size;
 419         }
 420 }
 421
 422 void __vhost_log_cache_write(struct virtio_net *dev,
 423                 struct vhost_virtqueue *vq,
 424                 uint64_t addr, uint64_t len);
 425 void __vhost_log_cache_write_iova(struct virtio_net *dev,
 426                 struct vhost_virtqueue *vq,
 427                 uint64_t iova, uint64_t len);
 428 void __vhost_log_cache_sync(struct virtio_net *dev,
 429                 struct vhost_virtqueue *vq);
 430 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
 431 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 432                             uint64_t iova, uint64_t len);
 433
 434 static __rte_always_inline void
 435 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
 436 {
 437         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 438                 __vhost_log_write(dev, addr, len);
 439 }
 440
 441 static __rte_always_inline void
 442 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
 443 {
 444         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 445                 __vhost_log_cache_sync(dev, vq);
 446 }
 447
 448 static __rte_always_inline void
 449 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
 450                         uint64_t addr, uint64_t len)
 451 {
 452         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 453                 __vhost_log_cache_write(dev, vq, addr, len);
 454 }
 455
 456 static __rte_always_inline void
 457 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 458                         uint64_t offset, uint64_t len)
 459 {
 460         vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset, len);
 461 }
 462
 463 static __rte_always_inline void
 464 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 465                      uint64_t offset, uint64_t len)
 466 {
 467         vhost_log_write(dev, vq->log_guest_addr + offset, len);
 468 }
 469
 470 static __rte_always_inline void
 471 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 472                            uint64_t iova, uint64_t len)
 473 {
 474         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 475                 return;
 476
 477         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 478                 __vhost_log_cache_write_iova(dev, vq, iova, len);
 479         else
 480                 __vhost_log_cache_write(dev, vq, iova, len);
 481 }
 482
 483 static __rte_always_inline void
 484 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 485                            uint64_t iova, uint64_t len)
 486 {
 487         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 488                 return;
 489
 490         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 491                 __vhost_log_write_iova(dev, vq, iova, len);
 492         else
 493                 __vhost_log_write(dev, iova, len);
 494 }
 495
 496 /* Macros for printing using RTE_LOG */
 497 #define RTE_LOGTYPE_VHOST_CONFIG RTE_LOGTYPE_USER1
 498 #define RTE_LOGTYPE_VHOST_DATA   RTE_LOGTYPE_USER1
 499
 500 #ifdef RTE_LIBRTE_VHOST_DEBUG
 501 #define VHOST_MAX_PRINT_BUFF 6072
 502 #define VHOST_LOG_DEBUG(log_type, fmt, args...) \
 503         RTE_LOG(DEBUG, log_type, fmt, ##args)
 504 #define PRINT_PACKET(device, addr, size, header) do { \
 505         char *pkt_addr = (char *)(addr); \
 506         unsigned int index; \
 507         char packet[VHOST_MAX_PRINT_BUFF]; \
 508         \
 509         if ((header)) \
 510                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
 511         else \
 512                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
 513         for (index = 0; index < (size); index++) { \
 514                 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
 515                         "%02hhx ", pkt_addr[index]); \
 516         } \
 517         snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
 518         \
 519         VHOST_LOG_DEBUG(VHOST_DATA, "%s", packet); \
 520 } while (0)
 521 #else
 522 #define VHOST_LOG_DEBUG(log_type, fmt, args...) do {} while (0)
 523 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
 524 #endif
 525
 526 extern uint64_t VHOST_FEATURES;
 527 #define MAX_VHOST_DEVICE        1024
 528 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
 529
 530 /* Convert guest physical address to host physical address */
 531 static __rte_always_inline rte_iova_t
 532 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
 533 {
 534         uint32_t i;
 535         struct guest_page *page;
 536
 537         for (i = 0; i < dev->nr_guest_pages; i++) {
 538                 page = &dev->guest_pages[i];
 539
 540                 if (gpa >= page->guest_phys_addr &&
 541                     gpa + size < page->guest_phys_addr + page->size) {
 542                         return gpa - page->guest_phys_addr +
 543                                page->host_phys_addr;
 544                 }
 545         }
 546
 547         return 0;
 548 }
 549
 550 static __rte_always_inline uint64_t
 551 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
 552 {
 553         struct rte_vhost_mem_region *r;
 554         uint32_t i;
 555
 556         if (unlikely(!dev || !dev->mem))
 557                 return 0;
 558
 559         for (i = 0; i < dev->mem->nregions; i++) {
 560                 r = &dev->mem->regions[i];
 561
 562                 if (vva >= r->host_user_addr &&
 563                     vva + len <  r->host_user_addr + r->size) {
 564                         return r->guest_phys_addr + vva - r->host_user_addr;
 565                 }
 566         }
 567         return 0;
 568 }
 569
 570 static __rte_always_inline struct virtio_net *
 571 get_device(int vid)
 572 {
 573         struct virtio_net *dev = vhost_devices[vid];
 574
 575         if (unlikely(!dev)) {
 576                 RTE_LOG(ERR, VHOST_CONFIG,
 577                         "(%d) device not found.\n", vid);
 578         }
 579
 580         return dev;
 581 }
 582
 583 int vhost_new_device(void);
 584 void cleanup_device(struct virtio_net *dev, int destroy);
 585 void reset_device(struct virtio_net *dev);
 586 void vhost_destroy_device(int);
 587 void vhost_destroy_device_notify(struct virtio_net *dev);
 588
 589 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
 590 void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
 591 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
 592
 593 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
 594
 595 void vhost_attach_vdpa_device(int vid, int did);
 596
 597 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
 598 void vhost_enable_dequeue_zero_copy(int vid);
 599 void vhost_set_builtin_virtio_net(int vid, bool enable);
 600 void vhost_enable_extbuf(int vid);
 601 void vhost_enable_linearbuf(int vid);
 602
 603 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
 604
 605 /*
 606  * Backend-specific cleanup.
 607  *
 608  * TODO: fix it; we have one backend now
 609  */
 610 void vhost_backend_cleanup(struct virtio_net *dev);
 611
 612 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 613                         uint64_t iova, uint64_t *len, uint8_t perm);
 614 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
 615                         struct vhost_virtqueue *vq,
 616                         uint64_t desc_addr, uint64_t desc_len);
 617 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 618 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 619
 620 static __rte_always_inline uint64_t
 621 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 622                         uint64_t iova, uint64_t *len, uint8_t perm)
 623 {
 624         if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 625                 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
 626
 627         return __vhost_iova_to_vva(dev, vq, iova, len, perm);
 628 }
 629
 630 #define vhost_avail_event(vr) \
 631         (*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
 632 #define vhost_used_event(vr) \
 633         (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
 634
 635 /*
 636  * The following is used with VIRTIO_RING_F_EVENT_IDX.
 637  * Assuming a given event_idx value from the other size, if we have
 638  * just incremented index from old to new_idx, should we trigger an
 639  * event?
 640  */
 641 static __rte_always_inline int
 642 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
 643 {
 644         return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
 645 }
 646
 647 static __rte_always_inline void
 648 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 649 {
 650         /* Flush used->idx update before we read avail->flags. */
 651         rte_smp_mb();
 652
 653         /* Don't kick guest if we don't reach index specified by guest. */
 654         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
 655                 uint16_t old = vq->signalled_used;
 656                 uint16_t new = vq->last_used_idx;
 657                 bool signalled_used_valid = vq->signalled_used_valid;
 658
 659                 vq->signalled_used = new;
 660                 vq->signalled_used_valid = true;
 661
 662                 VHOST_LOG_DEBUG(VHOST_DATA, "%s: used_event_idx=%d, old=%d, new=%d\n",
 663                         __func__,
 664                         vhost_used_event(vq),
 665                         old, new);
 666
 667                 if ((vhost_need_event(vhost_used_event(vq), new, old) &&
 668                                         (vq->callfd >= 0)) ||
 669                                 unlikely(!signalled_used_valid)) {
 670                         eventfd_write(vq->callfd, (eventfd_t) 1);
 671                         if (dev->notify_ops->guest_notified)
 672                                 dev->notify_ops->guest_notified(dev->vid);
 673                 }
 674         } else {
 675                 /* Kick the guest if necessary. */
 676                 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
 677                                 && (vq->callfd >= 0)) {
 678                         eventfd_write(vq->callfd, (eventfd_t)1);
 679                         if (dev->notify_ops->guest_notified)
 680                                 dev->notify_ops->guest_notified(dev->vid);
 681                 }
 682         }
 683 }
 684
 685 static __rte_always_inline void
 686 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
 687 {
 688         uint16_t old, new, off, off_wrap;
 689         bool signalled_used_valid, kick = false;
 690
 691         /* Flush used desc update. */
 692         rte_smp_mb();
 693
 694         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
 695                 if (vq->driver_event->flags !=
 696                                 VRING_EVENT_F_DISABLE)
 697                         kick = true;
 698                 goto kick;
 699         }
 700
 701         old = vq->signalled_used;
 702         new = vq->last_used_idx;
 703         vq->signalled_used = new;
 704         signalled_used_valid = vq->signalled_used_valid;
 705         vq->signalled_used_valid = true;
 706
 707         if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
 708                 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
 709                         kick = true;
 710                 goto kick;
 711         }
 712
 713         if (unlikely(!signalled_used_valid)) {
 714                 kick = true;
 715                 goto kick;
 716         }
 717
 718         rte_smp_rmb();
 719
 720         off_wrap = vq->driver_event->off_wrap;
 721         off = off_wrap & ~(1 << 15);
 722
 723         if (new <= old)
 724                 old -= vq->size;
 725
 726         if (vq->used_wrap_counter != off_wrap >> 15)
 727                 off -= vq->size;
 728
 729         if (vhost_need_event(off, new, old))
 730                 kick = true;
 731 kick:
 732         if (kick) {
 733                 eventfd_write(vq->callfd, (eventfd_t)1);
 734                 if (dev->notify_ops->guest_notified)
 735                         dev->notify_ops->guest_notified(dev->vid);
 736         }
 737 }
 738
 739 static __rte_always_inline void
 740 free_ind_table(void *idesc)
 741 {
 742         rte_free(idesc);
 743 }
 744
 745 static __rte_always_inline void
 746 restore_mbuf(struct rte_mbuf *m)
 747 {
 748         uint32_t mbuf_size, priv_size;
 749
 750         while (m) {
 751                 priv_size = rte_pktmbuf_priv_size(m->pool);
 752                 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
 753                 /* start of buffer is after mbuf structure and priv data */
 754
 755                 m->buf_addr = (char *)m + mbuf_size;
 756                 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
 757                 m = m->next;
 758         }
 759 }
 760
 761 static __rte_always_inline bool
 762 mbuf_is_consumed(struct rte_mbuf *m)
 763 {
 764         while (m) {
 765                 if (rte_mbuf_refcnt_read(m) > 1)
 766                         return false;
 767                 m = m->next;
 768         }
 769
 770         return true;
 771 }
 772
 773 static __rte_always_inline void
 774 put_zmbuf(struct zcopy_mbuf *zmbuf)
 775 {
 776         zmbuf->in_use = 0;
 777 }
 778
 779 #endif /* _VHOST_NET_CDEV_H_ */