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