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