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 #include "rte_vhost_async.h"
  28
  29 /* Used to indicate that the device is running on a data core */
  30 #define VIRTIO_DEV_RUNNING 1
  31 /* Used to indicate that the device is ready to operate */
  32 #define VIRTIO_DEV_READY 2
  33 /* Used to indicate that the built-in vhost net device backend is enabled */
  34 #define VIRTIO_DEV_BUILTIN_VIRTIO_NET 4
  35 /* Used to indicate that the device has its own data path and configured */
  36 #define VIRTIO_DEV_VDPA_CONFIGURED 8
  37
  38 /* Backend value set by guest. */
  39 #define VIRTIO_DEV_STOPPED -1
  40
  41 #define BUF_VECTOR_MAX 256
  42
  43 #define VHOST_LOG_CACHE_NR 32
  44
  45 #define MAX_PKT_BURST 32
  46
  47 #define VHOST_MAX_ASYNC_IT (MAX_PKT_BURST * 2)
  48 #define VHOST_MAX_ASYNC_VEC (BUF_VECTOR_MAX * 2)
  49
  50 #define PACKED_DESC_ENQUEUE_USED_FLAG(w)        \
  51         ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED | VRING_DESC_F_WRITE) : \
  52                 VRING_DESC_F_WRITE)
  53 #define PACKED_DESC_DEQUEUE_USED_FLAG(w)        \
  54         ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED) : 0x0)
  55 #define PACKED_DESC_SINGLE_DEQUEUE_FLAG (VRING_DESC_F_NEXT | \
  56                                          VRING_DESC_F_INDIRECT)
  57
  58 #define PACKED_BATCH_SIZE (RTE_CACHE_LINE_SIZE / \
  59                             sizeof(struct vring_packed_desc))
  60 #define PACKED_BATCH_MASK (PACKED_BATCH_SIZE - 1)
  61
  62 #ifdef VHOST_GCC_UNROLL_PRAGMA
  63 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("GCC unroll 4") \
  64         for (iter = val; iter < size; iter++)
  65 #endif
  66
  67 #ifdef VHOST_CLANG_UNROLL_PRAGMA
  68 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll 4") \
  69         for (iter = val; iter < size; iter++)
  70 #endif
  71
  72 #ifdef VHOST_ICC_UNROLL_PRAGMA
  73 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll (4)") \
  74         for (iter = val; iter < size; iter++)
  75 #endif
  76
  77 #ifndef vhost_for_each_try_unroll
  78 #define vhost_for_each_try_unroll(iter, val, num) \
  79         for (iter = val; iter < num; iter++)
  80 #endif
  81
  82 /**
  83  * Structure contains buffer address, length and descriptor index
  84  * from vring to do scatter RX.
  85  */
  86 struct buf_vector {
  87         uint64_t buf_iova;
  88         uint64_t buf_addr;
  89         uint32_t buf_len;
  90         uint32_t desc_idx;
  91 };
  92
  93 /*
  94  * A structure to hold some fields needed in zero copy code path,
  95  * mainly for associating an mbuf with the right desc_idx.
  96  */
  97 struct zcopy_mbuf {
  98         struct rte_mbuf *mbuf;
  99         uint32_t desc_idx;
 100         uint16_t desc_count;
 101         uint16_t in_use;
 102
 103         TAILQ_ENTRY(zcopy_mbuf) next;
 104 };
 105 TAILQ_HEAD(zcopy_mbuf_list, zcopy_mbuf);
 106
 107 /*
 108  * Structure contains the info for each batched memory copy.
 109  */
 110 struct batch_copy_elem {
 111         void *dst;
 112         void *src;
 113         uint32_t len;
 114         uint64_t log_addr;
 115 };
 116
 117 /*
 118  * Structure that contains the info for batched dirty logging.
 119  */
 120 struct log_cache_entry {
 121         uint32_t offset;
 122         unsigned long val;
 123 };
 124
 125 struct vring_used_elem_packed {
 126         uint16_t id;
 127         uint16_t flags;
 128         uint32_t len;
 129         uint32_t count;
 130 };
 131
 132 /**
 133  * Structure contains variables relevant to RX/TX virtqueues.
 134  */
 135 struct vhost_virtqueue {
 136         union {
 137                 struct vring_desc       *desc;
 138                 struct vring_packed_desc   *desc_packed;
 139         };
 140         union {
 141                 struct vring_avail      *avail;
 142                 struct vring_packed_desc_event *driver_event;
 143         };
 144         union {
 145                 struct vring_used       *used;
 146                 struct vring_packed_desc_event *device_event;
 147         };
 148         uint32_t                size;
 149
 150         uint16_t                last_avail_idx;
 151         uint16_t                last_used_idx;
 152         /* Last used index we notify to front end. */
 153         uint16_t                signalled_used;
 154         bool                    signalled_used_valid;
 155 #define VIRTIO_INVALID_EVENTFD          (-1)
 156 #define VIRTIO_UNINITIALIZED_EVENTFD    (-2)
 157
 158         /* Backend value to determine if device should started/stopped */
 159         int                     backend;
 160         int                     enabled;
 161         int                     access_ok;
 162         int                     ready;
 163         rte_spinlock_t          access_lock;
 164
 165         /* Used to notify the guest (trigger interrupt) */
 166         int                     callfd;
 167         /* Currently unused as polling mode is enabled */
 168         int                     kickfd;
 169
 170         /* Physical address of used ring, for logging */
 171         uint64_t                log_guest_addr;
 172
 173         /* inflight share memory info */
 174         union {
 175                 struct rte_vhost_inflight_info_split *inflight_split;
 176                 struct rte_vhost_inflight_info_packed *inflight_packed;
 177         };
 178         struct rte_vhost_resubmit_info *resubmit_inflight;
 179         uint64_t                global_counter;
 180
 181         uint16_t                nr_zmbuf;
 182         uint16_t                zmbuf_size;
 183         uint16_t                last_zmbuf_idx;
 184         struct zcopy_mbuf       *zmbufs;
 185         struct zcopy_mbuf_list  zmbuf_list;
 186
 187         union {
 188                 struct vring_used_elem  *shadow_used_split;
 189                 struct vring_used_elem_packed *shadow_used_packed;
 190         };
 191         uint16_t                shadow_used_idx;
 192         /* Record packed ring enqueue latest desc cache aligned index */
 193         uint16_t                shadow_aligned_idx;
 194         /* Record packed ring first dequeue desc index */
 195         uint16_t                shadow_last_used_idx;
 196         struct vhost_vring_addr ring_addrs;
 197
 198         struct batch_copy_elem  *batch_copy_elems;
 199         uint16_t                batch_copy_nb_elems;
 200         bool                    used_wrap_counter;
 201         bool                    avail_wrap_counter;
 202
 203         struct log_cache_entry log_cache[VHOST_LOG_CACHE_NR];
 204         uint16_t log_cache_nb_elem;
 205
 206         rte_rwlock_t    iotlb_lock;
 207         rte_rwlock_t    iotlb_pending_lock;
 208         struct rte_mempool *iotlb_pool;
 209         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
 210         int                             iotlb_cache_nr;
 211         TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
 212
 213         /* operation callbacks for async dma */
 214         struct rte_vhost_async_channel_ops      async_ops;
 215
 216         struct rte_vhost_iov_iter it_pool[VHOST_MAX_ASYNC_IT];
 217         struct iovec vec_pool[VHOST_MAX_ASYNC_VEC];
 218
 219         /* async data transfer status */
 220         uintptr_t       **async_pkts_pending;
 221         #define         ASYNC_PENDING_INFO_N_MSK 0xFFFF
 222         #define         ASYNC_PENDING_INFO_N_SFT 16
 223         uint64_t        *async_pending_info;
 224         uint16_t        async_pkts_idx;
 225         uint16_t        async_pkts_inflight_n;
 226
 227         /* vq async features */
 228         bool            async_inorder;
 229         bool            async_registered;
 230         uint16_t        async_threshold;
 231 } __rte_cache_aligned;
 232
 233 /* Virtio device status as per Virtio specification */
 234 #define VIRTIO_DEVICE_STATUS_ACK                0x01
 235 #define VIRTIO_DEVICE_STATUS_DRIVER             0x02
 236 #define VIRTIO_DEVICE_STATUS_DRIVER_OK          0x04
 237 #define VIRTIO_DEVICE_STATUS_FEATURES_OK        0x08
 238 #define VIRTIO_DEVICE_STATUS_DEV_NEED_RESET     0x40
 239 #define VIRTIO_DEVICE_STATUS_FAILED             0x80
 240
 241 #define VHOST_MAX_VRING                 0x100
 242 #define VHOST_MAX_QUEUE_PAIRS           0x80
 243
 244 /* Declare IOMMU related bits for older kernels */
 245 #ifndef VIRTIO_F_IOMMU_PLATFORM
 246
 247 #define VIRTIO_F_IOMMU_PLATFORM 33
 248
 249 struct vhost_iotlb_msg {
 250         __u64 iova;
 251         __u64 size;
 252         __u64 uaddr;
 253 #define VHOST_ACCESS_RO      0x1
 254 #define VHOST_ACCESS_WO      0x2
 255 #define VHOST_ACCESS_RW      0x3
 256         __u8 perm;
 257 #define VHOST_IOTLB_MISS           1
 258 #define VHOST_IOTLB_UPDATE         2
 259 #define VHOST_IOTLB_INVALIDATE     3
 260 #define VHOST_IOTLB_ACCESS_FAIL    4
 261         __u8 type;
 262 };
 263
 264 #define VHOST_IOTLB_MSG 0x1
 265
 266 struct vhost_msg {
 267         int type;
 268         union {
 269                 struct vhost_iotlb_msg iotlb;
 270                 __u8 padding[64];
 271         };
 272 };
 273 #endif
 274
 275 /*
 276  * Define virtio 1.0 for older kernels
 277  */
 278 #ifndef VIRTIO_F_VERSION_1
 279  #define VIRTIO_F_VERSION_1 32
 280 #endif
 281
 282 /* Declare packed ring related bits for older kernels */
 283 #ifndef VIRTIO_F_RING_PACKED
 284
 285 #define VIRTIO_F_RING_PACKED 34
 286
 287 struct vring_packed_desc {
 288         uint64_t addr;
 289         uint32_t len;
 290         uint16_t id;
 291         uint16_t flags;
 292 };
 293
 294 struct vring_packed_desc_event {
 295         uint16_t off_wrap;
 296         uint16_t flags;
 297 };
 298 #endif
 299
 300 /*
 301  * Declare below packed ring defines unconditionally
 302  * as Kernel header might use different names.
 303  */
 304 #define VRING_DESC_F_AVAIL      (1ULL << 7)
 305 #define VRING_DESC_F_USED       (1ULL << 15)
 306
 307 #define VRING_EVENT_F_ENABLE 0x0
 308 #define VRING_EVENT_F_DISABLE 0x1
 309 #define VRING_EVENT_F_DESC 0x2
 310
 311 /*
 312  * Available and used descs are in same order
 313  */
 314 #ifndef VIRTIO_F_IN_ORDER
 315 #define VIRTIO_F_IN_ORDER      35
 316 #endif
 317
 318 /* Features supported by this builtin vhost-user net driver. */
 319 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
 320                                 (1ULL << VIRTIO_F_ANY_LAYOUT) | \
 321                                 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
 322                                 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
 323                                 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
 324                                 (1ULL << VIRTIO_NET_F_MQ)      | \
 325                                 (1ULL << VIRTIO_F_VERSION_1)   | \
 326                                 (1ULL << VHOST_F_LOG_ALL)      | \
 327                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
 328                                 (1ULL << VIRTIO_NET_F_GSO) | \
 329                                 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
 330                                 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
 331                                 (1ULL << VIRTIO_NET_F_HOST_UFO) | \
 332                                 (1ULL << VIRTIO_NET_F_HOST_ECN) | \
 333                                 (1ULL << VIRTIO_NET_F_CSUM)    | \
 334                                 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
 335                                 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
 336                                 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
 337                                 (1ULL << VIRTIO_NET_F_GUEST_UFO) | \
 338                                 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
 339                                 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
 340                                 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \
 341                                 (1ULL << VIRTIO_NET_F_MTU)  | \
 342                                 (1ULL << VIRTIO_F_IN_ORDER) | \
 343                                 (1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
 344                                 (1ULL << VIRTIO_F_RING_PACKED))
 345
 346
 347 struct guest_page {
 348         uint64_t guest_phys_addr;
 349         uint64_t host_phys_addr;
 350         uint64_t size;
 351 };
 352
 353 struct inflight_mem_info {
 354         int             fd;
 355         void            *addr;
 356         uint64_t        size;
 357 };
 358
 359 /**
 360  * Device structure contains all configuration information relating
 361  * to the device.
 362  */
 363 struct virtio_net {
 364         /* Frontend (QEMU) memory and memory region information */
 365         struct rte_vhost_memory *mem;
 366         uint64_t                features;
 367         uint64_t                protocol_features;
 368         int                     vid;
 369         uint32_t                flags;
 370         uint16_t                vhost_hlen;
 371         /* to tell if we need broadcast rarp packet */
 372         int16_t                 broadcast_rarp;
 373         uint32_t                nr_vring;
 374         int                     dequeue_zero_copy;
 375         int                     async_copy;
 376         int                     extbuf;
 377         int                     linearbuf;
 378         struct vhost_virtqueue  *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
 379         struct inflight_mem_info *inflight_info;
 380 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
 381         char                    ifname[IF_NAME_SZ];
 382         uint64_t                log_size;
 383         uint64_t                log_base;
 384         uint64_t                log_addr;
 385         struct rte_ether_addr   mac;
 386         uint16_t                mtu;
 387         uint8_t                 status;
 388
 389         struct vhost_device_ops const *notify_ops;
 390
 391         uint32_t                nr_guest_pages;
 392         uint32_t                max_guest_pages;
 393         struct guest_page       *guest_pages;
 394
 395         int                     slave_req_fd;
 396         rte_spinlock_t          slave_req_lock;
 397
 398         int                     postcopy_ufd;
 399         int                     postcopy_listening;
 400
 401         struct rte_vdpa_device *vdpa_dev;
 402
 403         /* context data for the external message handlers */
 404         void                    *extern_data;
 405         /* pre and post vhost user message handlers for the device */
 406         struct rte_vhost_user_extern_ops extern_ops;
 407 } __rte_cache_aligned;
 408
 409 static __rte_always_inline bool
 410 vq_is_packed(struct virtio_net *dev)
 411 {
 412         return dev->features & (1ull << VIRTIO_F_RING_PACKED);
 413 }
 414
 415 static inline bool
 416 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
 417 {
 418         uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
 419
 420         return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
 421                 wrap_counter != !!(flags & VRING_DESC_F_USED);
 422 }
 423
 424 static inline void
 425 vq_inc_last_used_packed(struct vhost_virtqueue *vq, uint16_t num)
 426 {
 427         vq->last_used_idx += num;
 428         if (vq->last_used_idx >= vq->size) {
 429                 vq->used_wrap_counter ^= 1;
 430                 vq->last_used_idx -= vq->size;
 431         }
 432 }
 433
 434 static inline void
 435 vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
 436 {
 437         vq->last_avail_idx += num;
 438         if (vq->last_avail_idx >= vq->size) {
 439                 vq->avail_wrap_counter ^= 1;
 440                 vq->last_avail_idx -= vq->size;
 441         }
 442 }
 443
 444 void __vhost_log_cache_write(struct virtio_net *dev,
 445                 struct vhost_virtqueue *vq,
 446                 uint64_t addr, uint64_t len);
 447 void __vhost_log_cache_write_iova(struct virtio_net *dev,
 448                 struct vhost_virtqueue *vq,
 449                 uint64_t iova, uint64_t len);
 450 void __vhost_log_cache_sync(struct virtio_net *dev,
 451                 struct vhost_virtqueue *vq);
 452 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
 453 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 454                             uint64_t iova, uint64_t len);
 455
 456 static __rte_always_inline void
 457 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
 458 {
 459         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 460                 __vhost_log_write(dev, addr, len);
 461 }
 462
 463 static __rte_always_inline void
 464 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
 465 {
 466         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 467                 __vhost_log_cache_sync(dev, vq);
 468 }
 469
 470 static __rte_always_inline void
 471 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
 472                         uint64_t addr, uint64_t len)
 473 {
 474         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
 475                 __vhost_log_cache_write(dev, vq, addr, len);
 476 }
 477
 478 static __rte_always_inline void
 479 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 480                         uint64_t offset, uint64_t len)
 481 {
 482         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
 483                 if (unlikely(vq->log_guest_addr == 0))
 484                         return;
 485                 __vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset,
 486                                         len);
 487         }
 488 }
 489
 490 static __rte_always_inline void
 491 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 492                      uint64_t offset, uint64_t len)
 493 {
 494         if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
 495                 if (unlikely(vq->log_guest_addr == 0))
 496                         return;
 497                 __vhost_log_write(dev, vq->log_guest_addr + offset, len);
 498         }
 499 }
 500
 501 static __rte_always_inline void
 502 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 503                            uint64_t iova, uint64_t len)
 504 {
 505         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 506                 return;
 507
 508         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 509                 __vhost_log_cache_write_iova(dev, vq, iova, len);
 510         else
 511                 __vhost_log_cache_write(dev, vq, iova, len);
 512 }
 513
 514 static __rte_always_inline void
 515 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
 516                            uint64_t iova, uint64_t len)
 517 {
 518         if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
 519                 return;
 520
 521         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 522                 __vhost_log_write_iova(dev, vq, iova, len);
 523         else
 524                 __vhost_log_write(dev, iova, len);
 525 }
 526
 527 extern int vhost_config_log_level;
 528 extern int vhost_data_log_level;
 529
 530 #define VHOST_LOG_CONFIG(level, fmt, args...)                   \
 531         rte_log(RTE_LOG_ ## level, vhost_config_log_level,      \
 532                 "VHOST_CONFIG: " fmt, ##args)
 533
 534 #define VHOST_LOG_DATA(level, fmt, args...) \
 535         (void)((RTE_LOG_ ## level <= RTE_LOG_DP_LEVEL) ?        \
 536          rte_log(RTE_LOG_ ## level,  vhost_data_log_level,      \
 537                 "VHOST_DATA : " fmt, ##args) :                  \
 538          0)
 539
 540 #ifdef RTE_LIBRTE_VHOST_DEBUG
 541 #define VHOST_MAX_PRINT_BUFF 6072
 542 #define PRINT_PACKET(device, addr, size, header) do { \
 543         char *pkt_addr = (char *)(addr); \
 544         unsigned int index; \
 545         char packet[VHOST_MAX_PRINT_BUFF]; \
 546         \
 547         if ((header)) \
 548                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
 549         else \
 550                 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
 551         for (index = 0; index < (size); index++) { \
 552                 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
 553                         "%02hhx ", pkt_addr[index]); \
 554         } \
 555         snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
 556         \
 557         VHOST_LOG_DATA(DEBUG, "%s", packet); \
 558 } while (0)
 559 #else
 560 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
 561 #endif
 562
 563 #define MAX_VHOST_DEVICE        1024
 564 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
 565
 566 #define VHOST_BINARY_SEARCH_THRESH 256
 567
 568 static __rte_always_inline int guest_page_addrcmp(const void *p1,
 569                                                 const void *p2)
 570 {
 571         const struct guest_page *page1 = (const struct guest_page *)p1;
 572         const struct guest_page *page2 = (const struct guest_page *)p2;
 573
 574         if (page1->guest_phys_addr > page2->guest_phys_addr)
 575                 return 1;
 576         if (page1->guest_phys_addr < page2->guest_phys_addr)
 577                 return -1;
 578
 579         return 0;
 580 }
 581
 582 /* Convert guest physical address to host physical address */
 583 static __rte_always_inline rte_iova_t
 584 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
 585 {
 586         uint32_t i;
 587         struct guest_page *page;
 588         struct guest_page key;
 589
 590         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
 591                 key.guest_phys_addr = gpa;
 592                 page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages,
 593                                sizeof(struct guest_page), guest_page_addrcmp);
 594                 if (page) {
 595                         if (gpa + size < page->guest_phys_addr + page->size)
 596                                 return gpa - page->guest_phys_addr +
 597                                         page->host_phys_addr;
 598                 }
 599         } else {
 600                 for (i = 0; i < dev->nr_guest_pages; i++) {
 601                         page = &dev->guest_pages[i];
 602
 603                         if (gpa >= page->guest_phys_addr &&
 604                             gpa + size < page->guest_phys_addr +
 605                             page->size)
 606                                 return gpa - page->guest_phys_addr +
 607                                        page->host_phys_addr;
 608                 }
 609         }
 610
 611         return 0;
 612 }
 613
 614 static __rte_always_inline uint64_t
 615 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
 616 {
 617         struct rte_vhost_mem_region *r;
 618         uint32_t i;
 619
 620         if (unlikely(!dev || !dev->mem))
 621                 return 0;
 622
 623         for (i = 0; i < dev->mem->nregions; i++) {
 624                 r = &dev->mem->regions[i];
 625
 626                 if (vva >= r->host_user_addr &&
 627                     vva + len <  r->host_user_addr + r->size) {
 628                         return r->guest_phys_addr + vva - r->host_user_addr;
 629                 }
 630         }
 631         return 0;
 632 }
 633
 634 static __rte_always_inline struct virtio_net *
 635 get_device(int vid)
 636 {
 637         struct virtio_net *dev = vhost_devices[vid];
 638
 639         if (unlikely(!dev)) {
 640                 VHOST_LOG_CONFIG(ERR,
 641                         "(%d) device not found.\n", vid);
 642         }
 643
 644         return dev;
 645 }
 646
 647 int vhost_new_device(void);
 648 void cleanup_device(struct virtio_net *dev, int destroy);
 649 void reset_device(struct virtio_net *dev);
 650 void vhost_destroy_device(int);
 651 void vhost_destroy_device_notify(struct virtio_net *dev);
 652
 653 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
 654 void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
 655 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
 656
 657 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
 658
 659 void vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *dev);
 660
 661 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
 662 void vhost_enable_dequeue_zero_copy(int vid);
 663 void vhost_set_builtin_virtio_net(int vid, bool enable);
 664 void vhost_enable_extbuf(int vid);
 665 void vhost_enable_linearbuf(int vid);
 666
 667 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
 668
 669 /*
 670  * Backend-specific cleanup.
 671  *
 672  * TODO: fix it; we have one backend now
 673  */
 674 void vhost_backend_cleanup(struct virtio_net *dev);
 675
 676 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 677                         uint64_t iova, uint64_t *len, uint8_t perm);
 678 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
 679                         struct vhost_virtqueue *vq,
 680                         uint64_t desc_addr, uint64_t desc_len);
 681 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 682 uint64_t translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
 683                 uint64_t log_addr);
 684 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
 685
 686 static __rte_always_inline uint64_t
 687 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 688                         uint64_t iova, uint64_t *len, uint8_t perm)
 689 {
 690         if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 691                 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
 692
 693         return __vhost_iova_to_vva(dev, vq, iova, len, perm);
 694 }
 695
 696 #define vhost_avail_event(vr) \
 697         (*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
 698 #define vhost_used_event(vr) \
 699         (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
 700
 701 /*
 702  * The following is used with VIRTIO_RING_F_EVENT_IDX.
 703  * Assuming a given event_idx value from the other size, if we have
 704  * just incremented index from old to new_idx, should we trigger an
 705  * event?
 706  */
 707 static __rte_always_inline int
 708 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
 709 {
 710         return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
 711 }
 712
 713 static __rte_always_inline void
 714 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 715 {
 716         /* Flush used->idx update before we read avail->flags. */
 717         rte_smp_mb();
 718
 719         /* Don't kick guest if we don't reach index specified by guest. */
 720         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
 721                 uint16_t old = vq->signalled_used;
 722                 uint16_t new = vq->async_pkts_inflight_n ?
 723                                         vq->used->idx:vq->last_used_idx;
 724                 bool signalled_used_valid = vq->signalled_used_valid;
 725
 726                 vq->signalled_used = new;
 727                 vq->signalled_used_valid = true;
 728
 729                 VHOST_LOG_DATA(DEBUG, "%s: used_event_idx=%d, old=%d, new=%d\n",
 730                         __func__,
 731                         vhost_used_event(vq),
 732                         old, new);
 733
 734                 if ((vhost_need_event(vhost_used_event(vq), new, old) &&
 735                                         (vq->callfd >= 0)) ||
 736                                 unlikely(!signalled_used_valid)) {
 737                         eventfd_write(vq->callfd, (eventfd_t) 1);
 738                         if (dev->notify_ops->guest_notified)
 739                                 dev->notify_ops->guest_notified(dev->vid);
 740                 }
 741         } else {
 742                 /* Kick the guest if necessary. */
 743                 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
 744                                 && (vq->callfd >= 0)) {
 745                         eventfd_write(vq->callfd, (eventfd_t)1);
 746                         if (dev->notify_ops->guest_notified)
 747                                 dev->notify_ops->guest_notified(dev->vid);
 748                 }
 749         }
 750 }
 751
 752 static __rte_always_inline void
 753 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
 754 {
 755         uint16_t old, new, off, off_wrap;
 756         bool signalled_used_valid, kick = false;
 757
 758         /* Flush used desc update. */
 759         rte_smp_mb();
 760
 761         if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
 762                 if (vq->driver_event->flags !=
 763                                 VRING_EVENT_F_DISABLE)
 764                         kick = true;
 765                 goto kick;
 766         }
 767
 768         old = vq->signalled_used;
 769         new = vq->last_used_idx;
 770         vq->signalled_used = new;
 771         signalled_used_valid = vq->signalled_used_valid;
 772         vq->signalled_used_valid = true;
 773
 774         if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
 775                 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
 776                         kick = true;
 777                 goto kick;
 778         }
 779
 780         if (unlikely(!signalled_used_valid)) {
 781                 kick = true;
 782                 goto kick;
 783         }
 784
 785         rte_smp_rmb();
 786
 787         off_wrap = vq->driver_event->off_wrap;
 788         off = off_wrap & ~(1 << 15);
 789
 790         if (new <= old)
 791                 old -= vq->size;
 792
 793         if (vq->used_wrap_counter != off_wrap >> 15)
 794                 off -= vq->size;
 795
 796         if (vhost_need_event(off, new, old))
 797                 kick = true;
 798 kick:
 799         if (kick) {
 800                 eventfd_write(vq->callfd, (eventfd_t)1);
 801                 if (dev->notify_ops->guest_notified)
 802                         dev->notify_ops->guest_notified(dev->vid);
 803         }
 804 }
 805
 806 static __rte_always_inline void
 807 free_ind_table(void *idesc)
 808 {
 809         rte_free(idesc);
 810 }
 811
 812 static __rte_always_inline void
 813 restore_mbuf(struct rte_mbuf *m)
 814 {
 815         uint32_t mbuf_size, priv_size;
 816
 817         while (m) {
 818                 priv_size = rte_pktmbuf_priv_size(m->pool);
 819                 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
 820                 /* start of buffer is after mbuf structure and priv data */
 821
 822                 m->buf_addr = (char *)m + mbuf_size;
 823                 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
 824                 m = m->next;
 825         }
 826 }
 827
 828 static __rte_always_inline bool
 829 mbuf_is_consumed(struct rte_mbuf *m)
 830 {
 831         while (m) {
 832                 if (rte_mbuf_refcnt_read(m) > 1)
 833                         return false;
 834                 m = m->next;
 835         }
 836
 837         return true;
 838 }
 839
 840 static __rte_always_inline void
 841 put_zmbuf(struct zcopy_mbuf *zmbuf)
 842 {
 843         zmbuf->in_use = 0;
 844 }
 845
 846 #endif /* _VHOST_NET_CDEV_H_ */