lib/vhost/virtio_net.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2016 Intel Corporation
   3  */
   4
   5 #include <stdint.h>
   6 #include <stdbool.h>
   7 #include <linux/virtio_net.h>
   8
   9 #include <rte_mbuf.h>
  10 #include <rte_memcpy.h>
  11 #include <rte_net.h>
  12 #include <rte_ether.h>
  13 #include <rte_ip.h>
  14 #include <rte_vhost.h>
  15 #include <rte_tcp.h>
  16 #include <rte_udp.h>
  17 #include <rte_sctp.h>
  18 #include <rte_arp.h>
  19 #include <rte_spinlock.h>
  20 #include <rte_malloc.h>
  21 #include <rte_vhost_async.h>
  22
  23 #include "iotlb.h"
  24 #include "vhost.h"
  25
  26 #define MAX_BATCH_LEN 256
  27
  28 static  __rte_always_inline bool
  29 rxvq_is_mergeable(struct virtio_net *dev)
  30 {
  31         return dev->features & (1ULL << VIRTIO_NET_F_MRG_RXBUF);
  32 }
  33
  34 static  __rte_always_inline bool
  35 virtio_net_is_inorder(struct virtio_net *dev)
  36 {
  37         return dev->features & (1ULL << VIRTIO_F_IN_ORDER);
  38 }
  39
  40 static bool
  41 is_valid_virt_queue_idx(uint32_t idx, int is_tx, uint32_t nr_vring)
  42 {
  43         return (is_tx ^ (idx & 1)) == 0 && idx < nr_vring;
  44 }
  45
  46 static inline void
  47 do_data_copy_enqueue(struct virtio_net *dev, struct vhost_virtqueue *vq)
  48 {
  49         struct batch_copy_elem *elem = vq->batch_copy_elems;
  50         uint16_t count = vq->batch_copy_nb_elems;
  51         int i;
  52
  53         for (i = 0; i < count; i++) {
  54                 rte_memcpy(elem[i].dst, elem[i].src, elem[i].len);
  55                 vhost_log_cache_write_iova(dev, vq, elem[i].log_addr,
  56                                            elem[i].len);
  57                 PRINT_PACKET(dev, (uintptr_t)elem[i].dst, elem[i].len, 0);
  58         }
  59
  60         vq->batch_copy_nb_elems = 0;
  61 }
  62
  63 static inline void
  64 do_data_copy_dequeue(struct vhost_virtqueue *vq)
  65 {
  66         struct batch_copy_elem *elem = vq->batch_copy_elems;
  67         uint16_t count = vq->batch_copy_nb_elems;
  68         int i;
  69
  70         for (i = 0; i < count; i++)
  71                 rte_memcpy(elem[i].dst, elem[i].src, elem[i].len);
  72
  73         vq->batch_copy_nb_elems = 0;
  74 }
  75
  76 static __rte_always_inline void
  77 do_flush_shadow_used_ring_split(struct virtio_net *dev,
  78                         struct vhost_virtqueue *vq,
  79                         uint16_t to, uint16_t from, uint16_t size)
  80 {
  81         rte_memcpy(&vq->used->ring[to],
  82                         &vq->shadow_used_split[from],
  83                         size * sizeof(struct vring_used_elem));
  84         vhost_log_cache_used_vring(dev, vq,
  85                         offsetof(struct vring_used, ring[to]),
  86                         size * sizeof(struct vring_used_elem));
  87 }
  88
  89 static __rte_always_inline void
  90 flush_shadow_used_ring_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
  91 {
  92         uint16_t used_idx = vq->last_used_idx & (vq->size - 1);
  93
  94         if (used_idx + vq->shadow_used_idx <= vq->size) {
  95                 do_flush_shadow_used_ring_split(dev, vq, used_idx, 0,
  96                                           vq->shadow_used_idx);
  97         } else {
  98                 uint16_t size;
  99
 100                 /* update used ring interval [used_idx, vq->size] */
 101                 size = vq->size - used_idx;
 102                 do_flush_shadow_used_ring_split(dev, vq, used_idx, 0, size);
 103
 104                 /* update the left half used ring interval [0, left_size] */
 105                 do_flush_shadow_used_ring_split(dev, vq, 0, size,
 106                                           vq->shadow_used_idx - size);
 107         }
 108         vq->last_used_idx += vq->shadow_used_idx;
 109
 110         vhost_log_cache_sync(dev, vq);
 111
 112         __atomic_add_fetch(&vq->used->idx, vq->shadow_used_idx,
 113                            __ATOMIC_RELEASE);
 114         vq->shadow_used_idx = 0;
 115         vhost_log_used_vring(dev, vq, offsetof(struct vring_used, idx),
 116                 sizeof(vq->used->idx));
 117 }
 118
 119 static __rte_always_inline void
 120 update_shadow_used_ring_split(struct vhost_virtqueue *vq,
 121                          uint16_t desc_idx, uint32_t len)
 122 {
 123         uint16_t i = vq->shadow_used_idx++;
 124
 125         vq->shadow_used_split[i].id  = desc_idx;
 126         vq->shadow_used_split[i].len = len;
 127 }
 128
 129 static __rte_always_inline void
 130 vhost_flush_enqueue_shadow_packed(struct virtio_net *dev,
 131                                   struct vhost_virtqueue *vq)
 132 {
 133         int i;
 134         uint16_t used_idx = vq->last_used_idx;
 135         uint16_t head_idx = vq->last_used_idx;
 136         uint16_t head_flags = 0;
 137
 138         /* Split loop in two to save memory barriers */
 139         for (i = 0; i < vq->shadow_used_idx; i++) {
 140                 vq->desc_packed[used_idx].id = vq->shadow_used_packed[i].id;
 141                 vq->desc_packed[used_idx].len = vq->shadow_used_packed[i].len;
 142
 143                 used_idx += vq->shadow_used_packed[i].count;
 144                 if (used_idx >= vq->size)
 145                         used_idx -= vq->size;
 146         }
 147
 148         /* The ordering for storing desc flags needs to be enforced. */
 149         rte_atomic_thread_fence(__ATOMIC_RELEASE);
 150
 151         for (i = 0; i < vq->shadow_used_idx; i++) {
 152                 uint16_t flags;
 153
 154                 if (vq->shadow_used_packed[i].len)
 155                         flags = VRING_DESC_F_WRITE;
 156                 else
 157                         flags = 0;
 158
 159                 if (vq->used_wrap_counter) {
 160                         flags |= VRING_DESC_F_USED;
 161                         flags |= VRING_DESC_F_AVAIL;
 162                 } else {
 163                         flags &= ~VRING_DESC_F_USED;
 164                         flags &= ~VRING_DESC_F_AVAIL;
 165                 }
 166
 167                 if (i > 0) {
 168                         vq->desc_packed[vq->last_used_idx].flags = flags;
 169
 170                         vhost_log_cache_used_vring(dev, vq,
 171                                         vq->last_used_idx *
 172                                         sizeof(struct vring_packed_desc),
 173                                         sizeof(struct vring_packed_desc));
 174                 } else {
 175                         head_idx = vq->last_used_idx;
 176                         head_flags = flags;
 177                 }
 178
 179                 vq_inc_last_used_packed(vq, vq->shadow_used_packed[i].count);
 180         }
 181
 182         vq->desc_packed[head_idx].flags = head_flags;
 183
 184         vhost_log_cache_used_vring(dev, vq,
 185                                 head_idx *
 186                                 sizeof(struct vring_packed_desc),
 187                                 sizeof(struct vring_packed_desc));
 188
 189         vq->shadow_used_idx = 0;
 190         vhost_log_cache_sync(dev, vq);
 191 }
 192
 193 static __rte_always_inline void
 194 vhost_flush_dequeue_shadow_packed(struct virtio_net *dev,
 195                                   struct vhost_virtqueue *vq)
 196 {
 197         struct vring_used_elem_packed *used_elem = &vq->shadow_used_packed[0];
 198
 199         vq->desc_packed[vq->shadow_last_used_idx].id = used_elem->id;
 200         /* desc flags is the synchronization point for virtio packed vring */
 201         __atomic_store_n(&vq->desc_packed[vq->shadow_last_used_idx].flags,
 202                          used_elem->flags, __ATOMIC_RELEASE);
 203
 204         vhost_log_cache_used_vring(dev, vq, vq->shadow_last_used_idx *
 205                                    sizeof(struct vring_packed_desc),
 206                                    sizeof(struct vring_packed_desc));
 207         vq->shadow_used_idx = 0;
 208         vhost_log_cache_sync(dev, vq);
 209 }
 210
 211 static __rte_always_inline void
 212 vhost_flush_enqueue_batch_packed(struct virtio_net *dev,
 213                                  struct vhost_virtqueue *vq,
 214                                  uint64_t *lens,
 215                                  uint16_t *ids)
 216 {
 217         uint16_t i;
 218         uint16_t flags;
 219         uint16_t last_used_idx;
 220         struct vring_packed_desc *desc_base;
 221
 222         last_used_idx = vq->last_used_idx;
 223         desc_base = &vq->desc_packed[last_used_idx];
 224
 225         flags = PACKED_DESC_ENQUEUE_USED_FLAG(vq->used_wrap_counter);
 226
 227         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
 228                 desc_base[i].id = ids[i];
 229                 desc_base[i].len = lens[i];
 230         }
 231
 232         rte_atomic_thread_fence(__ATOMIC_RELEASE);
 233
 234         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
 235                 desc_base[i].flags = flags;
 236         }
 237
 238         vhost_log_cache_used_vring(dev, vq, last_used_idx *
 239                                    sizeof(struct vring_packed_desc),
 240                                    sizeof(struct vring_packed_desc) *
 241                                    PACKED_BATCH_SIZE);
 242         vhost_log_cache_sync(dev, vq);
 243
 244         vq_inc_last_used_packed(vq, PACKED_BATCH_SIZE);
 245 }
 246
 247 static __rte_always_inline void
 248 vhost_shadow_dequeue_batch_packed_inorder(struct vhost_virtqueue *vq,
 249                                           uint16_t id)
 250 {
 251         vq->shadow_used_packed[0].id = id;
 252
 253         if (!vq->shadow_used_idx) {
 254                 vq->shadow_last_used_idx = vq->last_used_idx;
 255                 vq->shadow_used_packed[0].flags =
 256                         PACKED_DESC_DEQUEUE_USED_FLAG(vq->used_wrap_counter);
 257                 vq->shadow_used_packed[0].len = 0;
 258                 vq->shadow_used_packed[0].count = 1;
 259                 vq->shadow_used_idx++;
 260         }
 261
 262         vq_inc_last_used_packed(vq, PACKED_BATCH_SIZE);
 263 }
 264
 265 static __rte_always_inline void
 266 vhost_shadow_dequeue_batch_packed(struct virtio_net *dev,
 267                                   struct vhost_virtqueue *vq,
 268                                   uint16_t *ids)
 269 {
 270         uint16_t flags;
 271         uint16_t i;
 272         uint16_t begin;
 273
 274         flags = PACKED_DESC_DEQUEUE_USED_FLAG(vq->used_wrap_counter);
 275
 276         if (!vq->shadow_used_idx) {
 277                 vq->shadow_last_used_idx = vq->last_used_idx;
 278                 vq->shadow_used_packed[0].id  = ids[0];
 279                 vq->shadow_used_packed[0].len = 0;
 280                 vq->shadow_used_packed[0].count = 1;
 281                 vq->shadow_used_packed[0].flags = flags;
 282                 vq->shadow_used_idx++;
 283                 begin = 1;
 284         } else
 285                 begin = 0;
 286
 287         vhost_for_each_try_unroll(i, begin, PACKED_BATCH_SIZE) {
 288                 vq->desc_packed[vq->last_used_idx + i].id = ids[i];
 289                 vq->desc_packed[vq->last_used_idx + i].len = 0;
 290         }
 291
 292         rte_atomic_thread_fence(__ATOMIC_RELEASE);
 293         vhost_for_each_try_unroll(i, begin, PACKED_BATCH_SIZE)
 294                 vq->desc_packed[vq->last_used_idx + i].flags = flags;
 295
 296         vhost_log_cache_used_vring(dev, vq, vq->last_used_idx *
 297                                    sizeof(struct vring_packed_desc),
 298                                    sizeof(struct vring_packed_desc) *
 299                                    PACKED_BATCH_SIZE);
 300         vhost_log_cache_sync(dev, vq);
 301
 302         vq_inc_last_used_packed(vq, PACKED_BATCH_SIZE);
 303 }
 304
 305 static __rte_always_inline void
 306 vhost_shadow_dequeue_single_packed(struct vhost_virtqueue *vq,
 307                                    uint16_t buf_id,
 308                                    uint16_t count)
 309 {
 310         uint16_t flags;
 311
 312         flags = vq->desc_packed[vq->last_used_idx].flags;
 313         if (vq->used_wrap_counter) {
 314                 flags |= VRING_DESC_F_USED;
 315                 flags |= VRING_DESC_F_AVAIL;
 316         } else {
 317                 flags &= ~VRING_DESC_F_USED;
 318                 flags &= ~VRING_DESC_F_AVAIL;
 319         }
 320
 321         if (!vq->shadow_used_idx) {
 322                 vq->shadow_last_used_idx = vq->last_used_idx;
 323
 324                 vq->shadow_used_packed[0].id  = buf_id;
 325                 vq->shadow_used_packed[0].len = 0;
 326                 vq->shadow_used_packed[0].flags = flags;
 327                 vq->shadow_used_idx++;
 328         } else {
 329                 vq->desc_packed[vq->last_used_idx].id = buf_id;
 330                 vq->desc_packed[vq->last_used_idx].len = 0;
 331                 vq->desc_packed[vq->last_used_idx].flags = flags;
 332         }
 333
 334         vq_inc_last_used_packed(vq, count);
 335 }
 336
 337 static __rte_always_inline void
 338 vhost_shadow_dequeue_single_packed_inorder(struct vhost_virtqueue *vq,
 339                                            uint16_t buf_id,
 340                                            uint16_t count)
 341 {
 342         uint16_t flags;
 343
 344         vq->shadow_used_packed[0].id = buf_id;
 345
 346         flags = vq->desc_packed[vq->last_used_idx].flags;
 347         if (vq->used_wrap_counter) {
 348                 flags |= VRING_DESC_F_USED;
 349                 flags |= VRING_DESC_F_AVAIL;
 350         } else {
 351                 flags &= ~VRING_DESC_F_USED;
 352                 flags &= ~VRING_DESC_F_AVAIL;
 353         }
 354
 355         if (!vq->shadow_used_idx) {
 356                 vq->shadow_last_used_idx = vq->last_used_idx;
 357                 vq->shadow_used_packed[0].len = 0;
 358                 vq->shadow_used_packed[0].flags = flags;
 359                 vq->shadow_used_idx++;
 360         }
 361
 362         vq_inc_last_used_packed(vq, count);
 363 }
 364
 365 static __rte_always_inline void
 366 vhost_shadow_enqueue_packed(struct vhost_virtqueue *vq,
 367                                    uint32_t *len,
 368                                    uint16_t *id,
 369                                    uint16_t *count,
 370                                    uint16_t num_buffers)
 371 {
 372         uint16_t i;
 373
 374         for (i = 0; i < num_buffers; i++) {
 375                 /* enqueue shadow flush action aligned with batch num */
 376                 if (!vq->shadow_used_idx)
 377                         vq->shadow_aligned_idx = vq->last_used_idx &
 378                                 PACKED_BATCH_MASK;
 379                 vq->shadow_used_packed[vq->shadow_used_idx].id  = id[i];
 380                 vq->shadow_used_packed[vq->shadow_used_idx].len = len[i];
 381                 vq->shadow_used_packed[vq->shadow_used_idx].count = count[i];
 382                 vq->shadow_aligned_idx += count[i];
 383                 vq->shadow_used_idx++;
 384         }
 385 }
 386
 387 static __rte_always_inline void
 388 vhost_shadow_enqueue_single_packed(struct virtio_net *dev,
 389                                    struct vhost_virtqueue *vq,
 390                                    uint32_t *len,
 391                                    uint16_t *id,
 392                                    uint16_t *count,
 393                                    uint16_t num_buffers)
 394 {
 395         vhost_shadow_enqueue_packed(vq, len, id, count, num_buffers);
 396
 397         if (vq->shadow_aligned_idx >= PACKED_BATCH_SIZE) {
 398                 do_data_copy_enqueue(dev, vq);
 399                 vhost_flush_enqueue_shadow_packed(dev, vq);
 400         }
 401 }
 402
 403 /* avoid write operation when necessary, to lessen cache issues */
 404 #define ASSIGN_UNLESS_EQUAL(var, val) do {      \
 405         if ((var) != (val))                     \
 406                 (var) = (val);                  \
 407 } while (0)
 408
 409 static __rte_always_inline void
 410 virtio_enqueue_offload(struct rte_mbuf *m_buf, struct virtio_net_hdr *net_hdr)
 411 {
 412         uint64_t csum_l4 = m_buf->ol_flags & RTE_MBUF_F_TX_L4_MASK;
 413
 414         if (m_buf->ol_flags & RTE_MBUF_F_TX_TCP_SEG)
 415                 csum_l4 |= RTE_MBUF_F_TX_TCP_CKSUM;
 416
 417         if (csum_l4) {
 418                 net_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 419                 net_hdr->csum_start = m_buf->l2_len + m_buf->l3_len;
 420
 421                 switch (csum_l4) {
 422                 case RTE_MBUF_F_TX_TCP_CKSUM:
 423                         net_hdr->csum_offset = (offsetof(struct rte_tcp_hdr,
 424                                                 cksum));
 425                         break;
 426                 case RTE_MBUF_F_TX_UDP_CKSUM:
 427                         net_hdr->csum_offset = (offsetof(struct rte_udp_hdr,
 428                                                 dgram_cksum));
 429                         break;
 430                 case RTE_MBUF_F_TX_SCTP_CKSUM:
 431                         net_hdr->csum_offset = (offsetof(struct rte_sctp_hdr,
 432                                                 cksum));
 433                         break;
 434                 }
 435         } else {
 436                 ASSIGN_UNLESS_EQUAL(net_hdr->csum_start, 0);
 437                 ASSIGN_UNLESS_EQUAL(net_hdr->csum_offset, 0);
 438                 ASSIGN_UNLESS_EQUAL(net_hdr->flags, 0);
 439         }
 440
 441         /* IP cksum verification cannot be bypassed, then calculate here */
 442         if (m_buf->ol_flags & RTE_MBUF_F_TX_IP_CKSUM) {
 443                 struct rte_ipv4_hdr *ipv4_hdr;
 444
 445                 ipv4_hdr = rte_pktmbuf_mtod_offset(m_buf, struct rte_ipv4_hdr *,
 446                                                    m_buf->l2_len);
 447                 ipv4_hdr->hdr_checksum = 0;
 448                 ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
 449         }
 450
 451         if (m_buf->ol_flags & RTE_MBUF_F_TX_TCP_SEG) {
 452                 if (m_buf->ol_flags & RTE_MBUF_F_TX_IPV4)
 453                         net_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 454                 else
 455                         net_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 456                 net_hdr->gso_size = m_buf->tso_segsz;
 457                 net_hdr->hdr_len = m_buf->l2_len + m_buf->l3_len
 458                                         + m_buf->l4_len;
 459         } else if (m_buf->ol_flags & RTE_MBUF_F_TX_UDP_SEG) {
 460                 net_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
 461                 net_hdr->gso_size = m_buf->tso_segsz;
 462                 net_hdr->hdr_len = m_buf->l2_len + m_buf->l3_len +
 463                         m_buf->l4_len;
 464         } else {
 465                 ASSIGN_UNLESS_EQUAL(net_hdr->gso_type, 0);
 466                 ASSIGN_UNLESS_EQUAL(net_hdr->gso_size, 0);
 467                 ASSIGN_UNLESS_EQUAL(net_hdr->hdr_len, 0);
 468         }
 469 }
 470
 471 static __rte_always_inline int
 472 map_one_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
 473                 struct buf_vector *buf_vec, uint16_t *vec_idx,
 474                 uint64_t desc_iova, uint64_t desc_len, uint8_t perm)
 475 {
 476         uint16_t vec_id = *vec_idx;
 477
 478         while (desc_len) {
 479                 uint64_t desc_addr;
 480                 uint64_t desc_chunck_len = desc_len;
 481
 482                 if (unlikely(vec_id >= BUF_VECTOR_MAX))
 483                         return -1;
 484
 485                 desc_addr = vhost_iova_to_vva(dev, vq,
 486                                 desc_iova,
 487                                 &desc_chunck_len,
 488                                 perm);
 489                 if (unlikely(!desc_addr))
 490                         return -1;
 491
 492                 rte_prefetch0((void *)(uintptr_t)desc_addr);
 493
 494                 buf_vec[vec_id].buf_iova = desc_iova;
 495                 buf_vec[vec_id].buf_addr = desc_addr;
 496                 buf_vec[vec_id].buf_len  = desc_chunck_len;
 497
 498                 desc_len -= desc_chunck_len;
 499                 desc_iova += desc_chunck_len;
 500                 vec_id++;
 501         }
 502         *vec_idx = vec_id;
 503
 504         return 0;
 505 }
 506
 507 static __rte_always_inline int
 508 fill_vec_buf_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 509                          uint32_t avail_idx, uint16_t *vec_idx,
 510                          struct buf_vector *buf_vec, uint16_t *desc_chain_head,
 511                          uint32_t *desc_chain_len, uint8_t perm)
 512 {
 513         uint16_t idx = vq->avail->ring[avail_idx & (vq->size - 1)];
 514         uint16_t vec_id = *vec_idx;
 515         uint32_t len    = 0;
 516         uint64_t dlen;
 517         uint32_t nr_descs = vq->size;
 518         uint32_t cnt    = 0;
 519         struct vring_desc *descs = vq->desc;
 520         struct vring_desc *idesc = NULL;
 521
 522         if (unlikely(idx >= vq->size))
 523                 return -1;
 524
 525         *desc_chain_head = idx;
 526
 527         if (vq->desc[idx].flags & VRING_DESC_F_INDIRECT) {
 528                 dlen = vq->desc[idx].len;
 529                 nr_descs = dlen / sizeof(struct vring_desc);
 530                 if (unlikely(nr_descs > vq->size))
 531                         return -1;
 532
 533                 descs = (struct vring_desc *)(uintptr_t)
 534                         vhost_iova_to_vva(dev, vq, vq->desc[idx].addr,
 535                                                 &dlen,
 536                                                 VHOST_ACCESS_RO);
 537                 if (unlikely(!descs))
 538                         return -1;
 539
 540                 if (unlikely(dlen < vq->desc[idx].len)) {
 541                         /*
 542                          * The indirect desc table is not contiguous
 543                          * in process VA space, we have to copy it.
 544                          */
 545                         idesc = vhost_alloc_copy_ind_table(dev, vq,
 546                                         vq->desc[idx].addr, vq->desc[idx].len);
 547                         if (unlikely(!idesc))
 548                                 return -1;
 549
 550                         descs = idesc;
 551                 }
 552
 553                 idx = 0;
 554         }
 555
 556         while (1) {
 557                 if (unlikely(idx >= nr_descs || cnt++ >= nr_descs)) {
 558                         free_ind_table(idesc);
 559                         return -1;
 560                 }
 561
 562                 dlen = descs[idx].len;
 563                 len += dlen;
 564
 565                 if (unlikely(map_one_desc(dev, vq, buf_vec, &vec_id,
 566                                                 descs[idx].addr, dlen,
 567                                                 perm))) {
 568                         free_ind_table(idesc);
 569                         return -1;
 570                 }
 571
 572                 if ((descs[idx].flags & VRING_DESC_F_NEXT) == 0)
 573                         break;
 574
 575                 idx = descs[idx].next;
 576         }
 577
 578         *desc_chain_len = len;
 579         *vec_idx = vec_id;
 580
 581         if (unlikely(!!idesc))
 582                 free_ind_table(idesc);
 583
 584         return 0;
 585 }
 586
 587 /*
 588  * Returns -1 on fail, 0 on success
 589  */
 590 static inline int
 591 reserve_avail_buf_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 592                                 uint32_t size, struct buf_vector *buf_vec,
 593                                 uint16_t *num_buffers, uint16_t avail_head,
 594                                 uint16_t *nr_vec)
 595 {
 596         uint16_t cur_idx;
 597         uint16_t vec_idx = 0;
 598         uint16_t max_tries, tries = 0;
 599
 600         uint16_t head_idx = 0;
 601         uint32_t len = 0;
 602
 603         *num_buffers = 0;
 604         cur_idx  = vq->last_avail_idx;
 605
 606         if (rxvq_is_mergeable(dev))
 607                 max_tries = vq->size - 1;
 608         else
 609                 max_tries = 1;
 610
 611         while (size > 0) {
 612                 if (unlikely(cur_idx == avail_head))
 613                         return -1;
 614                 /*
 615                  * if we tried all available ring items, and still
 616                  * can't get enough buf, it means something abnormal
 617                  * happened.
 618                  */
 619                 if (unlikely(++tries > max_tries))
 620                         return -1;
 621
 622                 if (unlikely(fill_vec_buf_split(dev, vq, cur_idx,
 623                                                 &vec_idx, buf_vec,
 624                                                 &head_idx, &len,
 625                                                 VHOST_ACCESS_RW) < 0))
 626                         return -1;
 627                 len = RTE_MIN(len, size);
 628                 update_shadow_used_ring_split(vq, head_idx, len);
 629                 size -= len;
 630
 631                 cur_idx++;
 632                 *num_buffers += 1;
 633         }
 634
 635         *nr_vec = vec_idx;
 636
 637         return 0;
 638 }
 639
 640 static __rte_always_inline int
 641 fill_vec_buf_packed_indirect(struct virtio_net *dev,
 642                         struct vhost_virtqueue *vq,
 643                         struct vring_packed_desc *desc, uint16_t *vec_idx,
 644                         struct buf_vector *buf_vec, uint32_t *len, uint8_t perm)
 645 {
 646         uint16_t i;
 647         uint32_t nr_descs;
 648         uint16_t vec_id = *vec_idx;
 649         uint64_t dlen;
 650         struct vring_packed_desc *descs, *idescs = NULL;
 651
 652         dlen = desc->len;
 653         descs = (struct vring_packed_desc *)(uintptr_t)
 654                 vhost_iova_to_vva(dev, vq, desc->addr, &dlen, VHOST_ACCESS_RO);
 655         if (unlikely(!descs))
 656                 return -1;
 657
 658         if (unlikely(dlen < desc->len)) {
 659                 /*
 660                  * The indirect desc table is not contiguous
 661                  * in process VA space, we have to copy it.
 662                  */
 663                 idescs = vhost_alloc_copy_ind_table(dev,
 664                                 vq, desc->addr, desc->len);
 665                 if (unlikely(!idescs))
 666                         return -1;
 667
 668                 descs = idescs;
 669         }
 670
 671         nr_descs =  desc->len / sizeof(struct vring_packed_desc);
 672         if (unlikely(nr_descs >= vq->size)) {
 673                 free_ind_table(idescs);
 674                 return -1;
 675         }
 676
 677         for (i = 0; i < nr_descs; i++) {
 678                 if (unlikely(vec_id >= BUF_VECTOR_MAX)) {
 679                         free_ind_table(idescs);
 680                         return -1;
 681                 }
 682
 683                 dlen = descs[i].len;
 684                 *len += dlen;
 685                 if (unlikely(map_one_desc(dev, vq, buf_vec, &vec_id,
 686                                                 descs[i].addr, dlen,
 687                                                 perm)))
 688                         return -1;
 689         }
 690         *vec_idx = vec_id;
 691
 692         if (unlikely(!!idescs))
 693                 free_ind_table(idescs);
 694
 695         return 0;
 696 }
 697
 698 static __rte_always_inline int
 699 fill_vec_buf_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
 700                                 uint16_t avail_idx, uint16_t *desc_count,
 701                                 struct buf_vector *buf_vec, uint16_t *vec_idx,
 702                                 uint16_t *buf_id, uint32_t *len, uint8_t perm)
 703 {
 704         bool wrap_counter = vq->avail_wrap_counter;
 705         struct vring_packed_desc *descs = vq->desc_packed;
 706         uint16_t vec_id = *vec_idx;
 707         uint64_t dlen;
 708
 709         if (avail_idx < vq->last_avail_idx)
 710                 wrap_counter ^= 1;
 711
 712         /*
 713          * Perform a load-acquire barrier in desc_is_avail to
 714          * enforce the ordering between desc flags and desc
 715          * content.
 716          */
 717         if (unlikely(!desc_is_avail(&descs[avail_idx], wrap_counter)))
 718                 return -1;
 719
 720         *desc_count = 0;
 721         *len = 0;
 722
 723         while (1) {
 724                 if (unlikely(vec_id >= BUF_VECTOR_MAX))
 725                         return -1;
 726
 727                 if (unlikely(*desc_count >= vq->size))
 728                         return -1;
 729
 730                 *desc_count += 1;
 731                 *buf_id = descs[avail_idx].id;
 732
 733                 if (descs[avail_idx].flags & VRING_DESC_F_INDIRECT) {
 734                         if (unlikely(fill_vec_buf_packed_indirect(dev, vq,
 735                                                         &descs[avail_idx],
 736                                                         &vec_id, buf_vec,
 737                                                         len, perm) < 0))
 738                                 return -1;
 739                 } else {
 740                         dlen = descs[avail_idx].len;
 741                         *len += dlen;
 742
 743                         if (unlikely(map_one_desc(dev, vq, buf_vec, &vec_id,
 744                                                         descs[avail_idx].addr,
 745                                                         dlen,
 746                                                         perm)))
 747                                 return -1;
 748                 }
 749
 750                 if ((descs[avail_idx].flags & VRING_DESC_F_NEXT) == 0)
 751                         break;
 752
 753                 if (++avail_idx >= vq->size) {
 754                         avail_idx -= vq->size;
 755                         wrap_counter ^= 1;
 756                 }
 757         }
 758
 759         *vec_idx = vec_id;
 760
 761         return 0;
 762 }
 763
 764 static __rte_noinline void
 765 copy_vnet_hdr_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
 766                 struct buf_vector *buf_vec,
 767                 struct virtio_net_hdr_mrg_rxbuf *hdr)
 768 {
 769         uint64_t len;
 770         uint64_t remain = dev->vhost_hlen;
 771         uint64_t src = (uint64_t)(uintptr_t)hdr, dst;
 772         uint64_t iova = buf_vec->buf_iova;
 773
 774         while (remain) {
 775                 len = RTE_MIN(remain,
 776                                 buf_vec->buf_len);
 777                 dst = buf_vec->buf_addr;
 778                 rte_memcpy((void *)(uintptr_t)dst,
 779                                 (void *)(uintptr_t)src,
 780                                 len);
 781
 782                 PRINT_PACKET(dev, (uintptr_t)dst,
 783                                 (uint32_t)len, 0);
 784                 vhost_log_cache_write_iova(dev, vq,
 785                                 iova, len);
 786
 787                 remain -= len;
 788                 iova += len;
 789                 src += len;
 790                 buf_vec++;
 791         }
 792 }
 793
 794 static __rte_always_inline int
 795 async_iter_initialize(struct vhost_async *async)
 796 {
 797         struct rte_vhost_iov_iter *iter;
 798
 799         if (unlikely(async->iovec_idx >= VHOST_MAX_ASYNC_VEC)) {
 800                 VHOST_LOG_DATA(ERR, "no more async iovec available\n");
 801                 return -1;
 802         }
 803
 804         iter = async->iov_iter + async->iter_idx;
 805         iter->iov = async->iovec + async->iovec_idx;
 806         iter->nr_segs = 0;
 807
 808         return 0;
 809 }
 810
 811 static __rte_always_inline int
 812 async_iter_add_iovec(struct vhost_async *async, void *src, void *dst, size_t len)
 813 {
 814         struct rte_vhost_iov_iter *iter;
 815         struct rte_vhost_iovec *iovec;
 816
 817         if (unlikely(async->iovec_idx >= VHOST_MAX_ASYNC_VEC)) {
 818                 static bool vhost_max_async_vec_log;
 819
 820                 if (!vhost_max_async_vec_log) {
 821                         VHOST_LOG_DATA(ERR, "no more async iovec available\n");
 822                         vhost_max_async_vec_log = true;
 823                 }
 824
 825                 return -1;
 826         }
 827
 828         iter = async->iov_iter + async->iter_idx;
 829         iovec = async->iovec + async->iovec_idx;
 830
 831         iovec->src_addr = src;
 832         iovec->dst_addr = dst;
 833         iovec->len = len;
 834
 835         iter->nr_segs++;
 836         async->iovec_idx++;
 837
 838         return 0;
 839 }
 840
 841 static __rte_always_inline void
 842 async_iter_finalize(struct vhost_async *async)
 843 {
 844         async->iter_idx++;
 845 }
 846
 847 static __rte_always_inline void
 848 async_iter_cancel(struct vhost_async *async)
 849 {
 850         struct rte_vhost_iov_iter *iter;
 851
 852         iter = async->iov_iter + async->iter_idx;
 853         async->iovec_idx -= iter->nr_segs;
 854         iter->nr_segs = 0;
 855         iter->iov = NULL;
 856 }
 857
 858 static __rte_always_inline void
 859 async_iter_reset(struct vhost_async *async)
 860 {
 861         async->iter_idx = 0;
 862         async->iovec_idx = 0;
 863 }
 864
 865 static __rte_always_inline int
 866 async_mbuf_to_desc_seg(struct virtio_net *dev, struct vhost_virtqueue *vq,
 867                 struct rte_mbuf *m, uint32_t mbuf_offset,
 868                 uint64_t buf_iova, uint32_t cpy_len)
 869 {
 870         struct vhost_async *async = vq->async;
 871         uint64_t mapped_len;
 872         uint32_t buf_offset = 0;
 873         void *hpa;
 874
 875         while (cpy_len) {
 876                 hpa = (void *)(uintptr_t)gpa_to_first_hpa(dev,
 877                                 buf_iova + buf_offset, cpy_len, &mapped_len);
 878                 if (unlikely(!hpa)) {
 879                         VHOST_LOG_DATA(ERR, "(%d) %s: failed to get hpa.\n", dev->vid, __func__);
 880                         return -1;
 881                 }
 882
 883                 if (unlikely(async_iter_add_iovec(async,
 884                                                 (void *)(uintptr_t)rte_pktmbuf_iova_offset(m,
 885                                                         mbuf_offset),
 886                                                 hpa, (size_t)mapped_len)))
 887                         return -1;
 888
 889                 cpy_len -= (uint32_t)mapped_len;
 890                 mbuf_offset += (uint32_t)mapped_len;
 891                 buf_offset += (uint32_t)mapped_len;
 892         }
 893
 894         return 0;
 895 }
 896
 897 static __rte_always_inline void
 898 sync_mbuf_to_desc_seg(struct virtio_net *dev, struct vhost_virtqueue *vq,
 899                 struct rte_mbuf *m, uint32_t mbuf_offset,
 900                 uint64_t buf_addr, uint64_t buf_iova, uint32_t cpy_len)
 901 {
 902         struct batch_copy_elem *batch_copy = vq->batch_copy_elems;
 903
 904         if (likely(cpy_len > MAX_BATCH_LEN || vq->batch_copy_nb_elems >= vq->size)) {
 905                 rte_memcpy((void *)((uintptr_t)(buf_addr)),
 906                                 rte_pktmbuf_mtod_offset(m, void *, mbuf_offset),
 907                                 cpy_len);
 908                 vhost_log_cache_write_iova(dev, vq, buf_iova, cpy_len);
 909                 PRINT_PACKET(dev, (uintptr_t)(buf_addr), cpy_len, 0);
 910         } else {
 911                 batch_copy[vq->batch_copy_nb_elems].dst =
 912                         (void *)((uintptr_t)(buf_addr));
 913                 batch_copy[vq->batch_copy_nb_elems].src =
 914                         rte_pktmbuf_mtod_offset(m, void *, mbuf_offset);
 915                 batch_copy[vq->batch_copy_nb_elems].log_addr = buf_iova;
 916                 batch_copy[vq->batch_copy_nb_elems].len = cpy_len;
 917                 vq->batch_copy_nb_elems++;
 918         }
 919 }
 920
 921 static __rte_always_inline int
 922 mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
 923                 struct rte_mbuf *m, struct buf_vector *buf_vec,
 924                 uint16_t nr_vec, uint16_t num_buffers, bool is_async)
 925 {
 926         uint32_t vec_idx = 0;
 927         uint32_t mbuf_offset, mbuf_avail;
 928         uint32_t buf_offset, buf_avail;
 929         uint64_t buf_addr, buf_iova, buf_len;
 930         uint32_t cpy_len;
 931         uint64_t hdr_addr;
 932         struct rte_mbuf *hdr_mbuf;
 933         struct virtio_net_hdr_mrg_rxbuf tmp_hdr, *hdr = NULL;
 934         struct vhost_async *async = vq->async;
 935
 936         if (unlikely(m == NULL))
 937                 return -1;
 938
 939         buf_addr = buf_vec[vec_idx].buf_addr;
 940         buf_iova = buf_vec[vec_idx].buf_iova;
 941         buf_len = buf_vec[vec_idx].buf_len;
 942
 943         if (unlikely(buf_len < dev->vhost_hlen && nr_vec <= 1))
 944                 return -1;
 945
 946         hdr_mbuf = m;
 947         hdr_addr = buf_addr;
 948         if (unlikely(buf_len < dev->vhost_hlen)) {
 949                 memset(&tmp_hdr, 0, sizeof(struct virtio_net_hdr_mrg_rxbuf));
 950                 hdr = &tmp_hdr;
 951         } else
 952                 hdr = (struct virtio_net_hdr_mrg_rxbuf *)(uintptr_t)hdr_addr;
 953
 954         VHOST_LOG_DATA(DEBUG, "(%d) RX: num merge buffers %d\n",
 955                 dev->vid, num_buffers);
 956
 957         if (unlikely(buf_len < dev->vhost_hlen)) {
 958                 buf_offset = dev->vhost_hlen - buf_len;
 959                 vec_idx++;
 960                 buf_addr = buf_vec[vec_idx].buf_addr;
 961                 buf_iova = buf_vec[vec_idx].buf_iova;
 962                 buf_len = buf_vec[vec_idx].buf_len;
 963                 buf_avail = buf_len - buf_offset;
 964         } else {
 965                 buf_offset = dev->vhost_hlen;
 966                 buf_avail = buf_len - dev->vhost_hlen;
 967         }
 968
 969         mbuf_avail  = rte_pktmbuf_data_len(m);
 970         mbuf_offset = 0;
 971
 972         if (is_async) {
 973                 if (async_iter_initialize(async))
 974                         return -1;
 975         }
 976
 977         while (mbuf_avail != 0 || m->next != NULL) {
 978                 /* done with current buf, get the next one */
 979                 if (buf_avail == 0) {
 980                         vec_idx++;
 981                         if (unlikely(vec_idx >= nr_vec))
 982                                 goto error;
 983
 984                         buf_addr = buf_vec[vec_idx].buf_addr;
 985                         buf_iova = buf_vec[vec_idx].buf_iova;
 986                         buf_len = buf_vec[vec_idx].buf_len;
 987
 988                         buf_offset = 0;
 989                         buf_avail  = buf_len;
 990                 }
 991
 992                 /* done with current mbuf, get the next one */
 993                 if (mbuf_avail == 0) {
 994                         m = m->next;
 995
 996                         mbuf_offset = 0;
 997                         mbuf_avail  = rte_pktmbuf_data_len(m);
 998                 }
 999
1000                 if (hdr_addr) {
1001                         virtio_enqueue_offload(hdr_mbuf, &hdr->hdr);
1002                         if (rxvq_is_mergeable(dev))
1003                                 ASSIGN_UNLESS_EQUAL(hdr->num_buffers,
1004                                                 num_buffers);
1005
1006                         if (unlikely(hdr == &tmp_hdr)) {
1007                                 copy_vnet_hdr_to_desc(dev, vq, buf_vec, hdr);
1008                         } else {
1009                                 PRINT_PACKET(dev, (uintptr_t)hdr_addr,
1010                                                 dev->vhost_hlen, 0);
1011                                 vhost_log_cache_write_iova(dev, vq,
1012                                                 buf_vec[0].buf_iova,
1013                                                 dev->vhost_hlen);
1014                         }
1015
1016                         hdr_addr = 0;
1017                 }
1018
1019                 cpy_len = RTE_MIN(buf_avail, mbuf_avail);
1020
1021                 if (is_async) {
1022                         if (async_mbuf_to_desc_seg(dev, vq, m, mbuf_offset,
1023                                                 buf_iova + buf_offset, cpy_len) < 0)
1024                                 goto error;
1025                 } else {
1026                         sync_mbuf_to_desc_seg(dev, vq, m, mbuf_offset,
1027                                         buf_addr + buf_offset,
1028                                         buf_iova + buf_offset, cpy_len);
1029                 }
1030
1031                 mbuf_avail  -= cpy_len;
1032                 mbuf_offset += cpy_len;
1033                 buf_avail  -= cpy_len;
1034                 buf_offset += cpy_len;
1035         }
1036
1037         if (is_async)
1038                 async_iter_finalize(async);
1039
1040         return 0;
1041 error:
1042         if (is_async)
1043                 async_iter_cancel(async);
1044
1045         return -1;
1046 }
1047
1048 static __rte_always_inline int
1049 vhost_enqueue_single_packed(struct virtio_net *dev,
1050                             struct vhost_virtqueue *vq,
1051                             struct rte_mbuf *pkt,
1052                             struct buf_vector *buf_vec,
1053                             uint16_t *nr_descs)
1054 {
1055         uint16_t nr_vec = 0;
1056         uint16_t avail_idx = vq->last_avail_idx;
1057         uint16_t max_tries, tries = 0;
1058         uint16_t buf_id = 0;
1059         uint32_t len = 0;
1060         uint16_t desc_count;
1061         uint32_t size = pkt->pkt_len + sizeof(struct virtio_net_hdr_mrg_rxbuf);
1062         uint16_t num_buffers = 0;
1063         uint32_t buffer_len[vq->size];
1064         uint16_t buffer_buf_id[vq->size];
1065         uint16_t buffer_desc_count[vq->size];
1066
1067         if (rxvq_is_mergeable(dev))
1068                 max_tries = vq->size - 1;
1069         else
1070                 max_tries = 1;
1071
1072         while (size > 0) {
1073                 /*
1074                  * if we tried all available ring items, and still
1075                  * can't get enough buf, it means something abnormal
1076                  * happened.
1077                  */
1078                 if (unlikely(++tries > max_tries))
1079                         return -1;
1080
1081                 if (unlikely(fill_vec_buf_packed(dev, vq,
1082                                                 avail_idx, &desc_count,
1083                                                 buf_vec, &nr_vec,
1084                                                 &buf_id, &len,
1085                                                 VHOST_ACCESS_RW) < 0))
1086                         return -1;
1087
1088                 len = RTE_MIN(len, size);
1089                 size -= len;
1090
1091                 buffer_len[num_buffers] = len;
1092                 buffer_buf_id[num_buffers] = buf_id;
1093                 buffer_desc_count[num_buffers] = desc_count;
1094                 num_buffers += 1;
1095
1096                 *nr_descs += desc_count;
1097                 avail_idx += desc_count;
1098                 if (avail_idx >= vq->size)
1099                         avail_idx -= vq->size;
1100         }
1101
1102         if (mbuf_to_desc(dev, vq, pkt, buf_vec, nr_vec, num_buffers, false) < 0)
1103                 return -1;
1104
1105         vhost_shadow_enqueue_single_packed(dev, vq, buffer_len, buffer_buf_id,
1106                                            buffer_desc_count, num_buffers);
1107
1108         return 0;
1109 }
1110
1111 static __rte_noinline uint32_t
1112 virtio_dev_rx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
1113         struct rte_mbuf **pkts, uint32_t count)
1114 {
1115         uint32_t pkt_idx = 0;
1116         uint16_t num_buffers;
1117         struct buf_vector buf_vec[BUF_VECTOR_MAX];
1118         uint16_t avail_head;
1119
1120         /*
1121          * The ordering between avail index and
1122          * desc reads needs to be enforced.
1123          */
1124         avail_head = __atomic_load_n(&vq->avail->idx, __ATOMIC_ACQUIRE);
1125
1126         rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
1127
1128         for (pkt_idx = 0; pkt_idx < count; pkt_idx++) {
1129                 uint32_t pkt_len = pkts[pkt_idx]->pkt_len + dev->vhost_hlen;
1130                 uint16_t nr_vec = 0;
1131
1132                 if (unlikely(reserve_avail_buf_split(dev, vq,
1133                                                 pkt_len, buf_vec, &num_buffers,
1134                                                 avail_head, &nr_vec) < 0)) {
1135                         VHOST_LOG_DATA(DEBUG,
1136                                 "(%d) failed to get enough desc from vring\n",
1137                                 dev->vid);
1138                         vq->shadow_used_idx -= num_buffers;
1139                         break;
1140                 }
1141
1142                 VHOST_LOG_DATA(DEBUG, "(%d) current index %d | end index %d\n",
1143                         dev->vid, vq->last_avail_idx,
1144                         vq->last_avail_idx + num_buffers);
1145
1146                 if (mbuf_to_desc(dev, vq, pkts[pkt_idx], buf_vec, nr_vec,
1147                                         num_buffers, false) < 0) {
1148                         vq->shadow_used_idx -= num_buffers;
1149                         break;
1150                 }
1151
1152                 vq->last_avail_idx += num_buffers;
1153         }
1154
1155         do_data_copy_enqueue(dev, vq);
1156
1157         if (likely(vq->shadow_used_idx)) {
1158                 flush_shadow_used_ring_split(dev, vq);
1159                 vhost_vring_call_split(dev, vq);
1160         }
1161
1162         return pkt_idx;
1163 }
1164
1165 static __rte_always_inline int
1166 virtio_dev_rx_sync_batch_check(struct virtio_net *dev,
1167                            struct vhost_virtqueue *vq,
1168                            struct rte_mbuf **pkts,
1169                            uint64_t *desc_addrs,
1170                            uint64_t *lens)
1171 {
1172         bool wrap_counter = vq->avail_wrap_counter;
1173         struct vring_packed_desc *descs = vq->desc_packed;
1174         uint16_t avail_idx = vq->last_avail_idx;
1175         uint32_t buf_offset = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1176         uint16_t i;
1177
1178         if (unlikely(avail_idx & PACKED_BATCH_MASK))
1179                 return -1;
1180
1181         if (unlikely((avail_idx + PACKED_BATCH_SIZE) > vq->size))
1182                 return -1;
1183
1184         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
1185                 if (unlikely(pkts[i]->next != NULL))
1186                         return -1;
1187                 if (unlikely(!desc_is_avail(&descs[avail_idx + i],
1188                                             wrap_counter)))
1189                         return -1;
1190         }
1191
1192         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
1193                 lens[i] = descs[avail_idx + i].len;
1194
1195         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
1196                 if (unlikely(pkts[i]->pkt_len > (lens[i] - buf_offset)))
1197                         return -1;
1198         }
1199
1200         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
1201                 desc_addrs[i] = vhost_iova_to_vva(dev, vq,
1202                                                   descs[avail_idx + i].addr,
1203                                                   &lens[i],
1204                                                   VHOST_ACCESS_RW);
1205
1206         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
1207                 if (unlikely(!desc_addrs[i]))
1208                         return -1;
1209                 if (unlikely(lens[i] != descs[avail_idx + i].len))
1210                         return -1;
1211         }
1212
1213         return 0;
1214 }
1215
1216 static __rte_always_inline void
1217 virtio_dev_rx_batch_packed_copy(struct virtio_net *dev,
1218                            struct vhost_virtqueue *vq,
1219                            struct rte_mbuf **pkts,
1220                            uint64_t *desc_addrs,
1221                            uint64_t *lens)
1222 {
1223         uint32_t buf_offset = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1224         struct virtio_net_hdr_mrg_rxbuf *hdrs[PACKED_BATCH_SIZE];
1225         struct vring_packed_desc *descs = vq->desc_packed;
1226         uint16_t avail_idx = vq->last_avail_idx;
1227         uint16_t ids[PACKED_BATCH_SIZE];
1228         uint16_t i;
1229
1230         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
1231                 rte_prefetch0((void *)(uintptr_t)desc_addrs[i]);
1232                 hdrs[i] = (struct virtio_net_hdr_mrg_rxbuf *)
1233                                         (uintptr_t)desc_addrs[i];
1234                 lens[i] = pkts[i]->pkt_len +
1235                         sizeof(struct virtio_net_hdr_mrg_rxbuf);
1236         }
1237
1238         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
1239                 virtio_enqueue_offload(pkts[i], &hdrs[i]->hdr);
1240
1241         vq_inc_last_avail_packed(vq, PACKED_BATCH_SIZE);
1242
1243         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
1244                 rte_memcpy((void *)(uintptr_t)(desc_addrs[i] + buf_offset),
1245                            rte_pktmbuf_mtod_offset(pkts[i], void *, 0),
1246                            pkts[i]->pkt_len);
1247         }
1248
1249         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
1250                 vhost_log_cache_write_iova(dev, vq, descs[avail_idx + i].addr,
1251                                            lens[i]);
1252
1253         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
1254                 ids[i] = descs[avail_idx + i].id;
1255
1256         vhost_flush_enqueue_batch_packed(dev, vq, lens, ids);
1257 }
1258
1259 static __rte_always_inline int
1260 virtio_dev_rx_sync_batch_packed(struct virtio_net *dev,
1261                            struct vhost_virtqueue *vq,
1262                            struct rte_mbuf **pkts)
1263 {
1264         uint64_t desc_addrs[PACKED_BATCH_SIZE];
1265         uint64_t lens[PACKED_BATCH_SIZE];
1266
1267         if (virtio_dev_rx_sync_batch_check(dev, vq, pkts, desc_addrs, lens) == -1)
1268                 return -1;
1269
1270         if (vq->shadow_used_idx) {
1271                 do_data_copy_enqueue(dev, vq);
1272                 vhost_flush_enqueue_shadow_packed(dev, vq);
1273         }
1274
1275         virtio_dev_rx_batch_packed_copy(dev, vq, pkts, desc_addrs, lens);
1276
1277         return 0;
1278 }
1279
1280 static __rte_always_inline int16_t
1281 virtio_dev_rx_single_packed(struct virtio_net *dev,
1282                             struct vhost_virtqueue *vq,
1283                             struct rte_mbuf *pkt)
1284 {
1285         struct buf_vector buf_vec[BUF_VECTOR_MAX];
1286         uint16_t nr_descs = 0;
1287
1288         if (unlikely(vhost_enqueue_single_packed(dev, vq, pkt, buf_vec,
1289                                                  &nr_descs) < 0)) {
1290                 VHOST_LOG_DATA(DEBUG,
1291                                 "(%d) failed to get enough desc from vring\n",
1292                                 dev->vid);
1293                 return -1;
1294         }
1295
1296         VHOST_LOG_DATA(DEBUG, "(%d) current index %d | end index %d\n",
1297                         dev->vid, vq->last_avail_idx,
1298                         vq->last_avail_idx + nr_descs);
1299
1300         vq_inc_last_avail_packed(vq, nr_descs);
1301
1302         return 0;
1303 }
1304
1305 static __rte_noinline uint32_t
1306 virtio_dev_rx_packed(struct virtio_net *dev,
1307                      struct vhost_virtqueue *__rte_restrict vq,
1308                      struct rte_mbuf **__rte_restrict pkts,
1309                      uint32_t count)
1310 {
1311         uint32_t pkt_idx = 0;
1312
1313         do {
1314                 rte_prefetch0(&vq->desc_packed[vq->last_avail_idx]);
1315
1316                 if (count - pkt_idx >= PACKED_BATCH_SIZE) {
1317                         if (!virtio_dev_rx_sync_batch_packed(dev, vq,
1318                                                         &pkts[pkt_idx])) {
1319                                 pkt_idx += PACKED_BATCH_SIZE;
1320                                 continue;
1321                         }
1322                 }
1323
1324                 if (virtio_dev_rx_single_packed(dev, vq, pkts[pkt_idx]))
1325                         break;
1326                 pkt_idx++;
1327
1328         } while (pkt_idx < count);
1329
1330         if (vq->shadow_used_idx) {
1331                 do_data_copy_enqueue(dev, vq);
1332                 vhost_flush_enqueue_shadow_packed(dev, vq);
1333         }
1334
1335         if (pkt_idx)
1336                 vhost_vring_call_packed(dev, vq);
1337
1338         return pkt_idx;
1339 }
1340
1341 static __rte_always_inline uint32_t
1342 virtio_dev_rx(struct virtio_net *dev, uint16_t queue_id,
1343         struct rte_mbuf **pkts, uint32_t count)
1344 {
1345         struct vhost_virtqueue *vq;
1346         uint32_t nb_tx = 0;
1347
1348         VHOST_LOG_DATA(DEBUG, "(%d) %s\n", dev->vid, __func__);
1349         if (unlikely(!is_valid_virt_queue_idx(queue_id, 0, dev->nr_vring))) {
1350                 VHOST_LOG_DATA(ERR, "(%d) %s: invalid virtqueue idx %d.\n",
1351                         dev->vid, __func__, queue_id);
1352                 return 0;
1353         }
1354
1355         vq = dev->virtqueue[queue_id];
1356
1357         rte_spinlock_lock(&vq->access_lock);
1358
1359         if (unlikely(!vq->enabled))
1360                 goto out_access_unlock;
1361
1362         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1363                 vhost_user_iotlb_rd_lock(vq);
1364
1365         if (unlikely(!vq->access_ok))
1366                 if (unlikely(vring_translate(dev, vq) < 0))
1367                         goto out;
1368
1369         count = RTE_MIN((uint32_t)MAX_PKT_BURST, count);
1370         if (count == 0)
1371                 goto out;
1372
1373         if (vq_is_packed(dev))
1374                 nb_tx = virtio_dev_rx_packed(dev, vq, pkts, count);
1375         else
1376                 nb_tx = virtio_dev_rx_split(dev, vq, pkts, count);
1377
1378 out:
1379         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1380                 vhost_user_iotlb_rd_unlock(vq);
1381
1382 out_access_unlock:
1383         rte_spinlock_unlock(&vq->access_lock);
1384
1385         return nb_tx;
1386 }
1387
1388 uint16_t
1389 rte_vhost_enqueue_burst(int vid, uint16_t queue_id,
1390         struct rte_mbuf **__rte_restrict pkts, uint16_t count)
1391 {
1392         struct virtio_net *dev = get_device(vid);
1393
1394         if (!dev)
1395                 return 0;
1396
1397         if (unlikely(!(dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET))) {
1398                 VHOST_LOG_DATA(ERR,
1399                         "(%d) %s: built-in vhost net backend is disabled.\n",
1400                         dev->vid, __func__);
1401                 return 0;
1402         }
1403
1404         return virtio_dev_rx(dev, queue_id, pkts, count);
1405 }
1406
1407 static __rte_always_inline uint16_t
1408 async_get_first_inflight_pkt_idx(struct vhost_virtqueue *vq)
1409 {
1410         struct vhost_async *async = vq->async;
1411
1412         if (async->pkts_idx >= async->pkts_inflight_n)
1413                 return async->pkts_idx - async->pkts_inflight_n;
1414         else
1415                 return vq->size - async->pkts_inflight_n + async->pkts_idx;
1416 }
1417
1418 static __rte_always_inline void
1419 store_dma_desc_info_split(struct vring_used_elem *s_ring, struct vring_used_elem *d_ring,
1420                 uint16_t ring_size, uint16_t s_idx, uint16_t d_idx, uint16_t count)
1421 {
1422         size_t elem_size = sizeof(struct vring_used_elem);
1423
1424         if (d_idx + count <= ring_size) {
1425                 rte_memcpy(d_ring + d_idx, s_ring + s_idx, count * elem_size);
1426         } else {
1427                 uint16_t size = ring_size - d_idx;
1428
1429                 rte_memcpy(d_ring + d_idx, s_ring + s_idx, size * elem_size);
1430                 rte_memcpy(d_ring, s_ring + s_idx + size, (count - size) * elem_size);
1431         }
1432 }
1433
1434 static __rte_always_inline void
1435 store_dma_desc_info_packed(struct vring_used_elem_packed *s_ring,
1436                 struct vring_used_elem_packed *d_ring,
1437                 uint16_t ring_size, uint16_t s_idx, uint16_t d_idx, uint16_t count)
1438 {
1439         size_t elem_size = sizeof(struct vring_used_elem_packed);
1440
1441         if (d_idx + count <= ring_size) {
1442                 rte_memcpy(d_ring + d_idx, s_ring + s_idx, count * elem_size);
1443         } else {
1444                 uint16_t size = ring_size - d_idx;
1445
1446                 rte_memcpy(d_ring + d_idx, s_ring + s_idx, size * elem_size);
1447                 rte_memcpy(d_ring, s_ring + s_idx + size, (count - size) * elem_size);
1448         }
1449 }
1450
1451 static __rte_noinline uint32_t
1452 virtio_dev_rx_async_submit_split(struct virtio_net *dev,
1453         struct vhost_virtqueue *vq, uint16_t queue_id,
1454         struct rte_mbuf **pkts, uint32_t count)
1455 {
1456         struct buf_vector buf_vec[BUF_VECTOR_MAX];
1457         uint32_t pkt_idx = 0;
1458         uint16_t num_buffers;
1459         uint16_t avail_head;
1460
1461         struct vhost_async *async = vq->async;
1462         struct async_inflight_info *pkts_info = async->pkts_info;
1463         uint32_t pkt_err = 0;
1464         int32_t n_xfer;
1465         uint16_t slot_idx = 0;
1466
1467         /*
1468          * The ordering between avail index and desc reads need to be enforced.
1469          */
1470         avail_head = __atomic_load_n(&vq->avail->idx, __ATOMIC_ACQUIRE);
1471
1472         rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
1473
1474         async_iter_reset(async);
1475
1476         for (pkt_idx = 0; pkt_idx < count; pkt_idx++) {
1477                 uint32_t pkt_len = pkts[pkt_idx]->pkt_len + dev->vhost_hlen;
1478                 uint16_t nr_vec = 0;
1479
1480                 if (unlikely(reserve_avail_buf_split(dev, vq, pkt_len, buf_vec,
1481                                                 &num_buffers, avail_head, &nr_vec) < 0)) {
1482                         VHOST_LOG_DATA(DEBUG, "(%d) failed to get enough desc from vring\n",
1483                                         dev->vid);
1484                         vq->shadow_used_idx -= num_buffers;
1485                         break;
1486                 }
1487
1488                 VHOST_LOG_DATA(DEBUG, "(%d) current index %d | end index %d\n",
1489                         dev->vid, vq->last_avail_idx, vq->last_avail_idx + num_buffers);
1490
1491                 if (mbuf_to_desc(dev, vq, pkts[pkt_idx], buf_vec, nr_vec, num_buffers, true) < 0) {
1492                         vq->shadow_used_idx -= num_buffers;
1493                         break;
1494                 }
1495
1496                 slot_idx = (async->pkts_idx + pkt_idx) & (vq->size - 1);
1497                 pkts_info[slot_idx].descs = num_buffers;
1498                 pkts_info[slot_idx].mbuf = pkts[pkt_idx];
1499
1500                 vq->last_avail_idx += num_buffers;
1501         }
1502
1503         if (unlikely(pkt_idx == 0))
1504                 return 0;
1505
1506         n_xfer = async->ops.transfer_data(dev->vid, queue_id, async->iov_iter, 0, pkt_idx);
1507         if (unlikely(n_xfer < 0)) {
1508                 VHOST_LOG_DATA(ERR, "(%d) %s: failed to transfer data for queue id %d.\n",
1509                                 dev->vid, __func__, queue_id);
1510                 n_xfer = 0;
1511         }
1512
1513         pkt_err = pkt_idx - n_xfer;
1514         if (unlikely(pkt_err)) {
1515                 uint16_t num_descs = 0;
1516
1517                 /* update number of completed packets */
1518                 pkt_idx = n_xfer;
1519
1520                 /* calculate the sum of descriptors to revert */
1521                 while (pkt_err-- > 0) {
1522                         num_descs += pkts_info[slot_idx & (vq->size - 1)].descs;
1523                         slot_idx--;
1524                 }
1525
1526                 /* recover shadow used ring and available ring */
1527                 vq->shadow_used_idx -= num_descs;
1528                 vq->last_avail_idx -= num_descs;
1529         }
1530
1531         /* keep used descriptors */
1532         if (likely(vq->shadow_used_idx)) {
1533                 uint16_t to = async->desc_idx_split & (vq->size - 1);
1534
1535                 store_dma_desc_info_split(vq->shadow_used_split,
1536                                 async->descs_split, vq->size, 0, to,
1537                                 vq->shadow_used_idx);
1538
1539                 async->desc_idx_split += vq->shadow_used_idx;
1540
1541                 async->pkts_idx += pkt_idx;
1542                 if (async->pkts_idx >= vq->size)
1543                         async->pkts_idx -= vq->size;
1544
1545                 async->pkts_inflight_n += pkt_idx;
1546                 vq->shadow_used_idx = 0;
1547         }
1548
1549         return pkt_idx;
1550 }
1551
1552 static __rte_always_inline void
1553 vhost_update_used_packed(struct vhost_virtqueue *vq,
1554                         struct vring_used_elem_packed *shadow_ring,
1555                         uint16_t count)
1556 {
1557         int i;
1558         uint16_t used_idx = vq->last_used_idx;
1559         uint16_t head_idx = vq->last_used_idx;
1560         uint16_t head_flags = 0;
1561
1562         if (count == 0)
1563                 return;
1564
1565         /* Split loop in two to save memory barriers */
1566         for (i = 0; i < count; i++) {
1567                 vq->desc_packed[used_idx].id = shadow_ring[i].id;
1568                 vq->desc_packed[used_idx].len = shadow_ring[i].len;
1569
1570                 used_idx += shadow_ring[i].count;
1571                 if (used_idx >= vq->size)
1572                         used_idx -= vq->size;
1573         }
1574
1575         /* The ordering for storing desc flags needs to be enforced. */
1576         rte_atomic_thread_fence(__ATOMIC_RELEASE);
1577
1578         for (i = 0; i < count; i++) {
1579                 uint16_t flags;
1580
1581                 if (vq->shadow_used_packed[i].len)
1582                         flags = VRING_DESC_F_WRITE;
1583                 else
1584                         flags = 0;
1585
1586                 if (vq->used_wrap_counter) {
1587                         flags |= VRING_DESC_F_USED;
1588                         flags |= VRING_DESC_F_AVAIL;
1589                 } else {
1590                         flags &= ~VRING_DESC_F_USED;
1591                         flags &= ~VRING_DESC_F_AVAIL;
1592                 }
1593
1594                 if (i > 0) {
1595                         vq->desc_packed[vq->last_used_idx].flags = flags;
1596                 } else {
1597                         head_idx = vq->last_used_idx;
1598                         head_flags = flags;
1599                 }
1600
1601                 vq_inc_last_used_packed(vq, shadow_ring[i].count);
1602         }
1603
1604         vq->desc_packed[head_idx].flags = head_flags;
1605 }
1606
1607 static __rte_always_inline int
1608 vhost_enqueue_async_packed(struct virtio_net *dev,
1609                             struct vhost_virtqueue *vq,
1610                             struct rte_mbuf *pkt,
1611                             struct buf_vector *buf_vec,
1612                             uint16_t *nr_descs,
1613                             uint16_t *nr_buffers)
1614 {
1615         uint16_t nr_vec = 0;
1616         uint16_t avail_idx = vq->last_avail_idx;
1617         uint16_t max_tries, tries = 0;
1618         uint16_t buf_id = 0;
1619         uint32_t len = 0;
1620         uint16_t desc_count = 0;
1621         uint32_t size = pkt->pkt_len + sizeof(struct virtio_net_hdr_mrg_rxbuf);
1622         uint32_t buffer_len[vq->size];
1623         uint16_t buffer_buf_id[vq->size];
1624         uint16_t buffer_desc_count[vq->size];
1625
1626         if (rxvq_is_mergeable(dev))
1627                 max_tries = vq->size - 1;
1628         else
1629                 max_tries = 1;
1630
1631         while (size > 0) {
1632                 /*
1633                  * if we tried all available ring items, and still
1634                  * can't get enough buf, it means something abnormal
1635                  * happened.
1636                  */
1637                 if (unlikely(++tries > max_tries))
1638                         return -1;
1639
1640                 if (unlikely(fill_vec_buf_packed(dev, vq,
1641                                                 avail_idx, &desc_count,
1642                                                 buf_vec, &nr_vec,
1643                                                 &buf_id, &len,
1644                                                 VHOST_ACCESS_RW) < 0))
1645                         return -1;
1646
1647                 len = RTE_MIN(len, size);
1648                 size -= len;
1649
1650                 buffer_len[*nr_buffers] = len;
1651                 buffer_buf_id[*nr_buffers] = buf_id;
1652                 buffer_desc_count[*nr_buffers] = desc_count;
1653                 *nr_buffers += 1;
1654                 *nr_descs += desc_count;
1655                 avail_idx += desc_count;
1656                 if (avail_idx >= vq->size)
1657                         avail_idx -= vq->size;
1658         }
1659
1660         if (unlikely(mbuf_to_desc(dev, vq, pkt, buf_vec, nr_vec, *nr_buffers, true) < 0))
1661                 return -1;
1662
1663         vhost_shadow_enqueue_packed(vq, buffer_len, buffer_buf_id, buffer_desc_count, *nr_buffers);
1664
1665         return 0;
1666 }
1667
1668 static __rte_always_inline int16_t
1669 virtio_dev_rx_async_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
1670                             struct rte_mbuf *pkt, uint16_t *nr_descs, uint16_t *nr_buffers)
1671 {
1672         struct buf_vector buf_vec[BUF_VECTOR_MAX];
1673
1674         if (unlikely(vhost_enqueue_async_packed(dev, vq, pkt, buf_vec,
1675                                         nr_descs, nr_buffers) < 0)) {
1676                 VHOST_LOG_DATA(DEBUG, "(%d) failed to get enough desc from vring\n", dev->vid);
1677                 return -1;
1678         }
1679
1680         VHOST_LOG_DATA(DEBUG, "(%d) current index %d | end index %d\n",
1681                         dev->vid, vq->last_avail_idx, vq->last_avail_idx + *nr_descs);
1682
1683         return 0;
1684 }
1685
1686 static __rte_always_inline void
1687 dma_error_handler_packed(struct vhost_virtqueue *vq, uint16_t slot_idx,
1688                         uint32_t nr_err, uint32_t *pkt_idx)
1689 {
1690         uint16_t descs_err = 0;
1691         uint16_t buffers_err = 0;
1692         struct async_inflight_info *pkts_info = vq->async->pkts_info;
1693
1694         *pkt_idx -= nr_err;
1695         /* calculate the sum of buffers and descs of DMA-error packets. */
1696         while (nr_err-- > 0) {
1697                 descs_err += pkts_info[slot_idx % vq->size].descs;
1698                 buffers_err += pkts_info[slot_idx % vq->size].nr_buffers;
1699                 slot_idx--;
1700         }
1701
1702         if (vq->last_avail_idx >= descs_err) {
1703                 vq->last_avail_idx -= descs_err;
1704         } else {
1705                 vq->last_avail_idx = vq->last_avail_idx + vq->size - descs_err;
1706                 vq->avail_wrap_counter ^= 1;
1707         }
1708
1709         vq->shadow_used_idx -= buffers_err;
1710 }
1711
1712 static __rte_noinline uint32_t
1713 virtio_dev_rx_async_submit_packed(struct virtio_net *dev,
1714         struct vhost_virtqueue *vq, uint16_t queue_id,
1715         struct rte_mbuf **pkts, uint32_t count)
1716 {
1717         uint32_t pkt_idx = 0;
1718         uint32_t remained = count;
1719         int32_t n_xfer;
1720         uint16_t num_buffers;
1721         uint16_t num_descs;
1722
1723         struct vhost_async *async = vq->async;
1724         struct async_inflight_info *pkts_info = async->pkts_info;
1725         uint32_t pkt_err = 0;
1726         uint16_t slot_idx = 0;
1727
1728         do {
1729                 rte_prefetch0(&vq->desc_packed[vq->last_avail_idx]);
1730
1731                 num_buffers = 0;
1732                 num_descs = 0;
1733                 if (unlikely(virtio_dev_rx_async_packed(dev, vq, pkts[pkt_idx],
1734                                                 &num_descs, &num_buffers) < 0))
1735                         break;
1736
1737                 slot_idx = (async->pkts_idx + pkt_idx) % vq->size;
1738
1739                 pkts_info[slot_idx].descs = num_descs;
1740                 pkts_info[slot_idx].nr_buffers = num_buffers;
1741                 pkts_info[slot_idx].mbuf = pkts[pkt_idx];
1742
1743                 pkt_idx++;
1744                 remained--;
1745                 vq_inc_last_avail_packed(vq, num_descs);
1746         } while (pkt_idx < count);
1747
1748         if (unlikely(pkt_idx == 0))
1749                 return 0;
1750
1751         n_xfer = async->ops.transfer_data(dev->vid, queue_id, async->iov_iter, 0, pkt_idx);
1752         if (unlikely(n_xfer < 0)) {
1753                 VHOST_LOG_DATA(ERR, "(%d) %s: failed to transfer data for queue id %d.\n",
1754                                 dev->vid, __func__, queue_id);
1755                 n_xfer = 0;
1756         }
1757
1758         pkt_err = pkt_idx - n_xfer;
1759
1760         async_iter_reset(async);
1761
1762         if (unlikely(pkt_err))
1763                 dma_error_handler_packed(vq, slot_idx, pkt_err, &pkt_idx);
1764
1765         if (likely(vq->shadow_used_idx)) {
1766                 /* keep used descriptors. */
1767                 store_dma_desc_info_packed(vq->shadow_used_packed, async->buffers_packed,
1768                                         vq->size, 0, async->buffer_idx_packed,
1769                                         vq->shadow_used_idx);
1770
1771                 async->buffer_idx_packed += vq->shadow_used_idx;
1772                 if (async->buffer_idx_packed >= vq->size)
1773                         async->buffer_idx_packed -= vq->size;
1774
1775                 async->pkts_idx += pkt_idx;
1776                 if (async->pkts_idx >= vq->size)
1777                         async->pkts_idx -= vq->size;
1778
1779                 vq->shadow_used_idx = 0;
1780                 async->pkts_inflight_n += pkt_idx;
1781         }
1782
1783         return pkt_idx;
1784 }
1785
1786 static __rte_always_inline void
1787 write_back_completed_descs_split(struct vhost_virtqueue *vq, uint16_t n_descs)
1788 {
1789         struct vhost_async *async = vq->async;
1790         uint16_t nr_left = n_descs;
1791         uint16_t nr_copy;
1792         uint16_t to, from;
1793
1794         do {
1795                 from = async->last_desc_idx_split & (vq->size - 1);
1796                 nr_copy = nr_left + from <= vq->size ? nr_left : vq->size - from;
1797                 to = vq->last_used_idx & (vq->size - 1);
1798
1799                 if (to + nr_copy <= vq->size) {
1800                         rte_memcpy(&vq->used->ring[to], &async->descs_split[from],
1801                                         nr_copy * sizeof(struct vring_used_elem));
1802                 } else {
1803                         uint16_t size = vq->size - to;
1804
1805                         rte_memcpy(&vq->used->ring[to], &async->descs_split[from],
1806                                         size * sizeof(struct vring_used_elem));
1807                         rte_memcpy(&vq->used->ring[0], &async->descs_split[from + size],
1808                                         (nr_copy - size) * sizeof(struct vring_used_elem));
1809                 }
1810
1811                 async->last_desc_idx_split += nr_copy;
1812                 vq->last_used_idx += nr_copy;
1813                 nr_left -= nr_copy;
1814         } while (nr_left > 0);
1815 }
1816
1817 static __rte_always_inline void
1818 write_back_completed_descs_packed(struct vhost_virtqueue *vq,
1819                                 uint16_t n_buffers)
1820 {
1821         struct vhost_async *async = vq->async;
1822         uint16_t nr_left = n_buffers;
1823         uint16_t from, to;
1824
1825         do {
1826                 from = async->last_buffer_idx_packed;
1827                 to = (from + nr_left) % vq->size;
1828                 if (to > from) {
1829                         vhost_update_used_packed(vq, async->buffers_packed + from, to - from);
1830                         async->last_buffer_idx_packed += nr_left;
1831                         nr_left = 0;
1832                 } else {
1833                         vhost_update_used_packed(vq, async->buffers_packed + from,
1834                                 vq->size - from);
1835                         async->last_buffer_idx_packed = 0;
1836                         nr_left -= vq->size - from;
1837                 }
1838         } while (nr_left > 0);
1839 }
1840
1841 static __rte_always_inline uint16_t
1842 vhost_poll_enqueue_completed(struct virtio_net *dev, uint16_t queue_id,
1843                 struct rte_mbuf **pkts, uint16_t count)
1844 {
1845         struct vhost_virtqueue *vq = dev->virtqueue[queue_id];
1846         struct vhost_async *async = vq->async;
1847         struct async_inflight_info *pkts_info = async->pkts_info;
1848         int32_t n_cpl;
1849         uint16_t n_descs = 0, n_buffers = 0;
1850         uint16_t start_idx, from, i;
1851
1852         n_cpl = async->ops.check_completed_copies(dev->vid, queue_id, 0, count);
1853         if (unlikely(n_cpl < 0)) {
1854                 VHOST_LOG_DATA(ERR, "(%d) %s: failed to check completed copies for queue id %d.\n",
1855                                 dev->vid, __func__, queue_id);
1856                 return 0;
1857         }
1858
1859         if (n_cpl == 0)
1860                 return 0;
1861
1862         start_idx = async_get_first_inflight_pkt_idx(vq);
1863
1864         for (i = 0; i < n_cpl; i++) {
1865                 from = (start_idx + i) % vq->size;
1866                 /* Only used with packed ring */
1867                 n_buffers += pkts_info[from].nr_buffers;
1868                 /* Only used with split ring */
1869                 n_descs += pkts_info[from].descs;
1870                 pkts[i] = pkts_info[from].mbuf;
1871         }
1872
1873         async->pkts_inflight_n -= n_cpl;
1874
1875         if (likely(vq->enabled && vq->access_ok)) {
1876                 if (vq_is_packed(dev)) {
1877                         write_back_completed_descs_packed(vq, n_buffers);
1878                         vhost_vring_call_packed(dev, vq);
1879                 } else {
1880                         write_back_completed_descs_split(vq, n_descs);
1881                         __atomic_add_fetch(&vq->used->idx, n_descs, __ATOMIC_RELEASE);
1882                         vhost_vring_call_split(dev, vq);
1883                 }
1884         } else {
1885                 if (vq_is_packed(dev)) {
1886                         async->last_buffer_idx_packed += n_buffers;
1887                         if (async->last_buffer_idx_packed >= vq->size)
1888                                 async->last_buffer_idx_packed -= vq->size;
1889                 } else {
1890                         async->last_desc_idx_split += n_descs;
1891                 }
1892         }
1893
1894         return n_cpl;
1895 }
1896
1897 uint16_t
1898 rte_vhost_poll_enqueue_completed(int vid, uint16_t queue_id,
1899                 struct rte_mbuf **pkts, uint16_t count)
1900 {
1901         struct virtio_net *dev = get_device(vid);
1902         struct vhost_virtqueue *vq;
1903         uint16_t n_pkts_cpl = 0;
1904
1905         if (unlikely(!dev))
1906                 return 0;
1907
1908         VHOST_LOG_DATA(DEBUG, "(%d) %s\n", dev->vid, __func__);
1909         if (unlikely(!is_valid_virt_queue_idx(queue_id, 0, dev->nr_vring))) {
1910                 VHOST_LOG_DATA(ERR, "(%d) %s: invalid virtqueue idx %d.\n",
1911                         dev->vid, __func__, queue_id);
1912                 return 0;
1913         }
1914
1915         vq = dev->virtqueue[queue_id];
1916
1917         if (unlikely(!vq->async)) {
1918                 VHOST_LOG_DATA(ERR, "(%d) %s: async not registered for queue id %d.\n",
1919                         dev->vid, __func__, queue_id);
1920                 return 0;
1921         }
1922
1923         rte_spinlock_lock(&vq->access_lock);
1924
1925         n_pkts_cpl = vhost_poll_enqueue_completed(dev, queue_id, pkts, count);
1926
1927         rte_spinlock_unlock(&vq->access_lock);
1928
1929         return n_pkts_cpl;
1930 }
1931
1932 uint16_t
1933 rte_vhost_clear_queue_thread_unsafe(int vid, uint16_t queue_id,
1934                 struct rte_mbuf **pkts, uint16_t count)
1935 {
1936         struct virtio_net *dev = get_device(vid);
1937         struct vhost_virtqueue *vq;
1938         uint16_t n_pkts_cpl = 0;
1939
1940         if (!dev)
1941                 return 0;
1942
1943         VHOST_LOG_DATA(DEBUG, "(%d) %s\n", dev->vid, __func__);
1944         if (unlikely(!is_valid_virt_queue_idx(queue_id, 0, dev->nr_vring))) {
1945                 VHOST_LOG_DATA(ERR, "(%d) %s: invalid virtqueue idx %d.\n",
1946                         dev->vid, __func__, queue_id);
1947                 return 0;
1948         }
1949
1950         vq = dev->virtqueue[queue_id];
1951
1952         if (unlikely(!vq->async)) {
1953                 VHOST_LOG_DATA(ERR, "(%d) %s: async not registered for queue id %d.\n",
1954                         dev->vid, __func__, queue_id);
1955                 return 0;
1956         }
1957
1958         n_pkts_cpl = vhost_poll_enqueue_completed(dev, queue_id, pkts, count);
1959
1960         return n_pkts_cpl;
1961 }
1962
1963 static __rte_always_inline uint32_t
1964 virtio_dev_rx_async_submit(struct virtio_net *dev, uint16_t queue_id,
1965         struct rte_mbuf **pkts, uint32_t count)
1966 {
1967         struct vhost_virtqueue *vq;
1968         uint32_t nb_tx = 0;
1969
1970         VHOST_LOG_DATA(DEBUG, "(%d) %s\n", dev->vid, __func__);
1971         if (unlikely(!is_valid_virt_queue_idx(queue_id, 0, dev->nr_vring))) {
1972                 VHOST_LOG_DATA(ERR, "(%d) %s: invalid virtqueue idx %d.\n",
1973                         dev->vid, __func__, queue_id);
1974                 return 0;
1975         }
1976
1977         vq = dev->virtqueue[queue_id];
1978
1979         rte_spinlock_lock(&vq->access_lock);
1980
1981         if (unlikely(!vq->enabled || !vq->async))
1982                 goto out_access_unlock;
1983
1984         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1985                 vhost_user_iotlb_rd_lock(vq);
1986
1987         if (unlikely(!vq->access_ok))
1988                 if (unlikely(vring_translate(dev, vq) < 0))
1989                         goto out;
1990
1991         count = RTE_MIN((uint32_t)MAX_PKT_BURST, count);
1992         if (count == 0)
1993                 goto out;
1994
1995         if (vq_is_packed(dev))
1996                 nb_tx = virtio_dev_rx_async_submit_packed(dev, vq, queue_id,
1997                                 pkts, count);
1998         else
1999                 nb_tx = virtio_dev_rx_async_submit_split(dev, vq, queue_id,
2000                                 pkts, count);
2001
2002 out:
2003         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
2004                 vhost_user_iotlb_rd_unlock(vq);
2005
2006 out_access_unlock:
2007         rte_spinlock_unlock(&vq->access_lock);
2008
2009         return nb_tx;
2010 }
2011
2012 uint16_t
2013 rte_vhost_submit_enqueue_burst(int vid, uint16_t queue_id,
2014                 struct rte_mbuf **pkts, uint16_t count)
2015 {
2016         struct virtio_net *dev = get_device(vid);
2017
2018         if (!dev)
2019                 return 0;
2020
2021         if (unlikely(!(dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET))) {
2022                 VHOST_LOG_DATA(ERR,
2023                         "(%d) %s: built-in vhost net backend is disabled.\n",
2024                         dev->vid, __func__);
2025                 return 0;
2026         }
2027
2028         return virtio_dev_rx_async_submit(dev, queue_id, pkts, count);
2029 }
2030
2031 static inline bool
2032 virtio_net_with_host_offload(struct virtio_net *dev)
2033 {
2034         if (dev->features &
2035                         ((1ULL << VIRTIO_NET_F_CSUM) |
2036                          (1ULL << VIRTIO_NET_F_HOST_ECN) |
2037                          (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2038                          (1ULL << VIRTIO_NET_F_HOST_TSO6) |
2039                          (1ULL << VIRTIO_NET_F_HOST_UFO)))
2040                 return true;
2041
2042         return false;
2043 }
2044
2045 static int
2046 parse_headers(struct rte_mbuf *m, uint8_t *l4_proto)
2047 {
2048         struct rte_ipv4_hdr *ipv4_hdr;
2049         struct rte_ipv6_hdr *ipv6_hdr;
2050         struct rte_ether_hdr *eth_hdr;
2051         uint16_t ethertype;
2052         uint16_t data_len = rte_pktmbuf_data_len(m);
2053
2054         if (data_len < sizeof(struct rte_ether_hdr))
2055                 return -EINVAL;
2056
2057         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
2058
2059         m->l2_len = sizeof(struct rte_ether_hdr);
2060         ethertype = rte_be_to_cpu_16(eth_hdr->ether_type);
2061
2062         if (ethertype == RTE_ETHER_TYPE_VLAN) {
2063                 if (data_len < sizeof(struct rte_ether_hdr) +
2064                                 sizeof(struct rte_vlan_hdr))
2065                         goto error;
2066
2067                 struct rte_vlan_hdr *vlan_hdr =
2068                         (struct rte_vlan_hdr *)(eth_hdr + 1);
2069
2070                 m->l2_len += sizeof(struct rte_vlan_hdr);
2071                 ethertype = rte_be_to_cpu_16(vlan_hdr->eth_proto);
2072         }
2073
2074         switch (ethertype) {
2075         case RTE_ETHER_TYPE_IPV4:
2076                 if (data_len < m->l2_len + sizeof(struct rte_ipv4_hdr))
2077                         goto error;
2078                 ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
2079                                 m->l2_len);
2080                 m->l3_len = rte_ipv4_hdr_len(ipv4_hdr);
2081                 if (data_len < m->l2_len + m->l3_len)
2082                         goto error;
2083                 m->ol_flags |= RTE_MBUF_F_TX_IPV4;
2084                 *l4_proto = ipv4_hdr->next_proto_id;
2085                 break;
2086         case RTE_ETHER_TYPE_IPV6:
2087                 if (data_len < m->l2_len + sizeof(struct rte_ipv6_hdr))
2088                         goto error;
2089                 ipv6_hdr = rte_pktmbuf_mtod_offset(m, struct rte_ipv6_hdr *,
2090                                 m->l2_len);
2091                 m->l3_len = sizeof(struct rte_ipv6_hdr);
2092                 m->ol_flags |= RTE_MBUF_F_TX_IPV6;
2093                 *l4_proto = ipv6_hdr->proto;
2094                 break;
2095         default:
2096                 /* a valid L3 header is needed for further L4 parsing */
2097                 goto error;
2098         }
2099
2100         /* both CSUM and GSO need a valid L4 header */
2101         switch (*l4_proto) {
2102         case IPPROTO_TCP:
2103                 if (data_len < m->l2_len + m->l3_len +
2104                                 sizeof(struct rte_tcp_hdr))
2105                         goto error;
2106                 break;
2107         case IPPROTO_UDP:
2108                 if (data_len < m->l2_len + m->l3_len +
2109                                 sizeof(struct rte_udp_hdr))
2110                         goto error;
2111                 break;
2112         case IPPROTO_SCTP:
2113                 if (data_len < m->l2_len + m->l3_len +
2114                                 sizeof(struct rte_sctp_hdr))
2115                         goto error;
2116                 break;
2117         default:
2118                 goto error;
2119         }
2120
2121         return 0;
2122
2123 error:
2124         m->l2_len = 0;
2125         m->l3_len = 0;
2126         m->ol_flags = 0;
2127         return -EINVAL;
2128 }
2129
2130 static __rte_always_inline void
2131 vhost_dequeue_offload_legacy(struct virtio_net_hdr *hdr, struct rte_mbuf *m)
2132 {
2133         uint8_t l4_proto = 0;
2134         struct rte_tcp_hdr *tcp_hdr = NULL;
2135         uint16_t tcp_len;
2136         uint16_t data_len = rte_pktmbuf_data_len(m);
2137
2138         if (parse_headers(m, &l4_proto) < 0)
2139                 return;
2140
2141         if (hdr->flags == VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2142                 if (hdr->csum_start == (m->l2_len + m->l3_len)) {
2143                         switch (hdr->csum_offset) {
2144                         case (offsetof(struct rte_tcp_hdr, cksum)):
2145                                 if (l4_proto != IPPROTO_TCP)
2146                                         goto error;
2147                                 m->ol_flags |= RTE_MBUF_F_TX_TCP_CKSUM;
2148                                 break;
2149                         case (offsetof(struct rte_udp_hdr, dgram_cksum)):
2150                                 if (l4_proto != IPPROTO_UDP)
2151                                         goto error;
2152                                 m->ol_flags |= RTE_MBUF_F_TX_UDP_CKSUM;
2153                                 break;
2154                         case (offsetof(struct rte_sctp_hdr, cksum)):
2155                                 if (l4_proto != IPPROTO_SCTP)
2156                                         goto error;
2157                                 m->ol_flags |= RTE_MBUF_F_TX_SCTP_CKSUM;
2158                                 break;
2159                         default:
2160                                 goto error;
2161                         }
2162                 } else {
2163                         goto error;
2164                 }
2165         }
2166
2167         if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2168                 switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2169                 case VIRTIO_NET_HDR_GSO_TCPV4:
2170                 case VIRTIO_NET_HDR_GSO_TCPV6:
2171                         if (l4_proto != IPPROTO_TCP)
2172                                 goto error;
2173                         tcp_hdr = rte_pktmbuf_mtod_offset(m,
2174                                         struct rte_tcp_hdr *,
2175                                         m->l2_len + m->l3_len);
2176                         tcp_len = (tcp_hdr->data_off & 0xf0) >> 2;
2177                         if (data_len < m->l2_len + m->l3_len + tcp_len)
2178                                 goto error;
2179                         m->ol_flags |= RTE_MBUF_F_TX_TCP_SEG;
2180                         m->tso_segsz = hdr->gso_size;
2181                         m->l4_len = tcp_len;
2182                         break;
2183                 case VIRTIO_NET_HDR_GSO_UDP:
2184                         if (l4_proto != IPPROTO_UDP)
2185                                 goto error;
2186                         m->ol_flags |= RTE_MBUF_F_TX_UDP_SEG;
2187                         m->tso_segsz = hdr->gso_size;
2188                         m->l4_len = sizeof(struct rte_udp_hdr);
2189                         break;
2190                 default:
2191                         VHOST_LOG_DATA(WARNING,
2192                                 "unsupported gso type %u.\n", hdr->gso_type);
2193                         goto error;
2194                 }
2195         }
2196         return;
2197
2198 error:
2199         m->l2_len = 0;
2200         m->l3_len = 0;
2201         m->ol_flags = 0;
2202 }
2203
2204 static __rte_always_inline void
2205 vhost_dequeue_offload(struct virtio_net_hdr *hdr, struct rte_mbuf *m,
2206         bool legacy_ol_flags)
2207 {
2208         struct rte_net_hdr_lens hdr_lens;
2209         int l4_supported = 0;
2210         uint32_t ptype;
2211
2212         if (hdr->flags == 0 && hdr->gso_type == VIRTIO_NET_HDR_GSO_NONE)
2213                 return;
2214
2215         if (legacy_ol_flags) {
2216                 vhost_dequeue_offload_legacy(hdr, m);
2217                 return;
2218         }
2219
2220         m->ol_flags |= RTE_MBUF_F_RX_IP_CKSUM_UNKNOWN;
2221
2222         ptype = rte_net_get_ptype(m, &hdr_lens, RTE_PTYPE_ALL_MASK);
2223         m->packet_type = ptype;
2224         if ((ptype & RTE_PTYPE_L4_MASK) == RTE_PTYPE_L4_TCP ||
2225             (ptype & RTE_PTYPE_L4_MASK) == RTE_PTYPE_L4_UDP ||
2226             (ptype & RTE_PTYPE_L4_MASK) == RTE_PTYPE_L4_SCTP)
2227                 l4_supported = 1;
2228
2229         /* According to Virtio 1.1 spec, the device only needs to look at
2230          * VIRTIO_NET_HDR_F_NEEDS_CSUM in the packet transmission path.
2231          * This differs from the processing incoming packets path where the
2232          * driver could rely on VIRTIO_NET_HDR_F_DATA_VALID flag set by the
2233          * device.
2234          *
2235          * 5.1.6.2.1 Driver Requirements: Packet Transmission
2236          * The driver MUST NOT set the VIRTIO_NET_HDR_F_DATA_VALID and
2237          * VIRTIO_NET_HDR_F_RSC_INFO bits in flags.
2238          *
2239          * 5.1.6.2.2 Device Requirements: Packet Transmission
2240          * The device MUST ignore flag bits that it does not recognize.
2241          */
2242         if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2243                 uint32_t hdrlen;
2244
2245                 hdrlen = hdr_lens.l2_len + hdr_lens.l3_len + hdr_lens.l4_len;
2246                 if (hdr->csum_start <= hdrlen && l4_supported != 0) {
2247                         m->ol_flags |= RTE_MBUF_F_RX_L4_CKSUM_NONE;
2248                 } else {
2249                         /* Unknown proto or tunnel, do sw cksum. We can assume
2250                          * the cksum field is in the first segment since the
2251                          * buffers we provided to the host are large enough.
2252                          * In case of SCTP, this will be wrong since it's a CRC
2253                          * but there's nothing we can do.
2254                          */
2255                         uint16_t csum = 0, off;
2256
2257                         if (rte_raw_cksum_mbuf(m, hdr->csum_start,
2258                                         rte_pktmbuf_pkt_len(m) - hdr->csum_start, &csum) < 0)
2259                                 return;
2260                         if (likely(csum != 0xffff))
2261                                 csum = ~csum;
2262                         off = hdr->csum_offset + hdr->csum_start;
2263                         if (rte_pktmbuf_data_len(m) >= off + 1)
2264                                 *rte_pktmbuf_mtod_offset(m, uint16_t *, off) = csum;
2265                 }
2266         }
2267
2268         if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2269                 if (hdr->gso_size == 0)
2270                         return;
2271
2272                 switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2273                 case VIRTIO_NET_HDR_GSO_TCPV4:
2274                 case VIRTIO_NET_HDR_GSO_TCPV6:
2275                         if ((ptype & RTE_PTYPE_L4_MASK) != RTE_PTYPE_L4_TCP)
2276                                 break;
2277                         m->ol_flags |= RTE_MBUF_F_RX_LRO | RTE_MBUF_F_RX_L4_CKSUM_NONE;
2278                         m->tso_segsz = hdr->gso_size;
2279                         break;
2280                 case VIRTIO_NET_HDR_GSO_UDP:
2281                         if ((ptype & RTE_PTYPE_L4_MASK) != RTE_PTYPE_L4_UDP)
2282                                 break;
2283                         m->ol_flags |= RTE_MBUF_F_RX_LRO | RTE_MBUF_F_RX_L4_CKSUM_NONE;
2284                         m->tso_segsz = hdr->gso_size;
2285                         break;
2286                 default:
2287                         break;
2288                 }
2289         }
2290 }
2291
2292 static __rte_noinline void
2293 copy_vnet_hdr_from_desc(struct virtio_net_hdr *hdr,
2294                 struct buf_vector *buf_vec)
2295 {
2296         uint64_t len;
2297         uint64_t remain = sizeof(struct virtio_net_hdr);
2298         uint64_t src;
2299         uint64_t dst = (uint64_t)(uintptr_t)hdr;
2300
2301         while (remain) {
2302                 len = RTE_MIN(remain, buf_vec->buf_len);
2303                 src = buf_vec->buf_addr;
2304                 rte_memcpy((void *)(uintptr_t)dst,
2305                                 (void *)(uintptr_t)src, len);
2306
2307                 remain -= len;
2308                 dst += len;
2309                 buf_vec++;
2310         }
2311 }
2312
2313 static __rte_always_inline int
2314 copy_desc_to_mbuf(struct virtio_net *dev, struct vhost_virtqueue *vq,
2315                   struct buf_vector *buf_vec, uint16_t nr_vec,
2316                   struct rte_mbuf *m, struct rte_mempool *mbuf_pool,
2317                   bool legacy_ol_flags)
2318 {
2319         uint32_t buf_avail, buf_offset;
2320         uint64_t buf_addr, buf_len;
2321         uint32_t mbuf_avail, mbuf_offset;
2322         uint32_t cpy_len;
2323         struct rte_mbuf *cur = m, *prev = m;
2324         struct virtio_net_hdr tmp_hdr;
2325         struct virtio_net_hdr *hdr = NULL;
2326         /* A counter to avoid desc dead loop chain */
2327         uint16_t vec_idx = 0;
2328         struct batch_copy_elem *batch_copy = vq->batch_copy_elems;
2329         int error = 0;
2330
2331         buf_addr = buf_vec[vec_idx].buf_addr;
2332         buf_len = buf_vec[vec_idx].buf_len;
2333
2334         if (unlikely(buf_len < dev->vhost_hlen && nr_vec <= 1)) {
2335                 error = -1;
2336                 goto out;
2337         }
2338
2339         if (virtio_net_with_host_offload(dev)) {
2340                 if (unlikely(buf_len < sizeof(struct virtio_net_hdr))) {
2341                         /*
2342                          * No luck, the virtio-net header doesn't fit
2343                          * in a contiguous virtual area.
2344                          */
2345                         copy_vnet_hdr_from_desc(&tmp_hdr, buf_vec);
2346                         hdr = &tmp_hdr;
2347                 } else {
2348                         hdr = (struct virtio_net_hdr *)((uintptr_t)buf_addr);
2349                 }
2350         }
2351
2352         /*
2353          * A virtio driver normally uses at least 2 desc buffers
2354          * for Tx: the first for storing the header, and others
2355          * for storing the data.
2356          */
2357         if (unlikely(buf_len < dev->vhost_hlen)) {
2358                 buf_offset = dev->vhost_hlen - buf_len;
2359                 vec_idx++;
2360                 buf_addr = buf_vec[vec_idx].buf_addr;
2361                 buf_len = buf_vec[vec_idx].buf_len;
2362                 buf_avail  = buf_len - buf_offset;
2363         } else if (buf_len == dev->vhost_hlen) {
2364                 if (unlikely(++vec_idx >= nr_vec))
2365                         goto out;
2366                 buf_addr = buf_vec[vec_idx].buf_addr;
2367                 buf_len = buf_vec[vec_idx].buf_len;
2368
2369                 buf_offset = 0;
2370                 buf_avail = buf_len;
2371         } else {
2372                 buf_offset = dev->vhost_hlen;
2373                 buf_avail = buf_vec[vec_idx].buf_len - dev->vhost_hlen;
2374         }
2375
2376         PRINT_PACKET(dev,
2377                         (uintptr_t)(buf_addr + buf_offset),
2378                         (uint32_t)buf_avail, 0);
2379
2380         mbuf_offset = 0;
2381         mbuf_avail  = m->buf_len - RTE_PKTMBUF_HEADROOM;
2382         while (1) {
2383                 cpy_len = RTE_MIN(buf_avail, mbuf_avail);
2384
2385                 if (likely(cpy_len > MAX_BATCH_LEN ||
2386                                         vq->batch_copy_nb_elems >= vq->size ||
2387                                         (hdr && cur == m))) {
2388                         rte_memcpy(rte_pktmbuf_mtod_offset(cur, void *,
2389                                                 mbuf_offset),
2390                                         (void *)((uintptr_t)(buf_addr +
2391                                                         buf_offset)), cpy_len);
2392                 } else {
2393                         batch_copy[vq->batch_copy_nb_elems].dst =
2394                                 rte_pktmbuf_mtod_offset(cur, void *,
2395                                                 mbuf_offset);
2396                         batch_copy[vq->batch_copy_nb_elems].src =
2397                                 (void *)((uintptr_t)(buf_addr + buf_offset));
2398                         batch_copy[vq->batch_copy_nb_elems].len = cpy_len;
2399                         vq->batch_copy_nb_elems++;
2400                 }
2401
2402                 mbuf_avail  -= cpy_len;
2403                 mbuf_offset += cpy_len;
2404                 buf_avail -= cpy_len;
2405                 buf_offset += cpy_len;
2406
2407                 /* This buf reaches to its end, get the next one */
2408                 if (buf_avail == 0) {
2409                         if (++vec_idx >= nr_vec)
2410                                 break;
2411
2412                         buf_addr = buf_vec[vec_idx].buf_addr;
2413                         buf_len = buf_vec[vec_idx].buf_len;
2414
2415                         buf_offset = 0;
2416                         buf_avail  = buf_len;
2417
2418                         PRINT_PACKET(dev, (uintptr_t)buf_addr,
2419                                         (uint32_t)buf_avail, 0);
2420                 }
2421
2422                 /*
2423                  * This mbuf reaches to its end, get a new one
2424                  * to hold more data.
2425                  */
2426                 if (mbuf_avail == 0) {
2427                         cur = rte_pktmbuf_alloc(mbuf_pool);
2428                         if (unlikely(cur == NULL)) {
2429                                 VHOST_LOG_DATA(ERR, "Failed to "
2430                                         "allocate memory for mbuf.\n");
2431                                 error = -1;
2432                                 goto out;
2433                         }
2434
2435                         prev->next = cur;
2436                         prev->data_len = mbuf_offset;
2437                         m->nb_segs += 1;
2438                         m->pkt_len += mbuf_offset;
2439                         prev = cur;
2440
2441                         mbuf_offset = 0;
2442                         mbuf_avail  = cur->buf_len - RTE_PKTMBUF_HEADROOM;
2443                 }
2444         }
2445
2446         prev->data_len = mbuf_offset;
2447         m->pkt_len    += mbuf_offset;
2448
2449         if (hdr)
2450                 vhost_dequeue_offload(hdr, m, legacy_ol_flags);
2451
2452 out:
2453
2454         return error;
2455 }
2456
2457 static void
2458 virtio_dev_extbuf_free(void *addr __rte_unused, void *opaque)
2459 {
2460         rte_free(opaque);
2461 }
2462
2463 static int
2464 virtio_dev_extbuf_alloc(struct rte_mbuf *pkt, uint32_t size)
2465 {
2466         struct rte_mbuf_ext_shared_info *shinfo = NULL;
2467         uint32_t total_len = RTE_PKTMBUF_HEADROOM + size;
2468         uint16_t buf_len;
2469         rte_iova_t iova;
2470         void *buf;
2471
2472         total_len += sizeof(*shinfo) + sizeof(uintptr_t);
2473         total_len = RTE_ALIGN_CEIL(total_len, sizeof(uintptr_t));
2474
2475         if (unlikely(total_len > UINT16_MAX))
2476                 return -ENOSPC;
2477
2478         buf_len = total_len;
2479         buf = rte_malloc(NULL, buf_len, RTE_CACHE_LINE_SIZE);
2480         if (unlikely(buf == NULL))
2481                 return -ENOMEM;
2482
2483         /* Initialize shinfo */
2484         shinfo = rte_pktmbuf_ext_shinfo_init_helper(buf, &buf_len,
2485                                                 virtio_dev_extbuf_free, buf);
2486         if (unlikely(shinfo == NULL)) {
2487                 rte_free(buf);
2488                 VHOST_LOG_DATA(ERR, "Failed to init shinfo\n");
2489                 return -1;
2490         }
2491
2492         iova = rte_malloc_virt2iova(buf);
2493         rte_pktmbuf_attach_extbuf(pkt, buf, iova, buf_len, shinfo);
2494         rte_pktmbuf_reset_headroom(pkt);
2495
2496         return 0;
2497 }
2498
2499 /*
2500  * Prepare a host supported pktmbuf.
2501  */
2502 static __rte_always_inline int
2503 virtio_dev_pktmbuf_prep(struct virtio_net *dev, struct rte_mbuf *pkt,
2504                          uint32_t data_len)
2505 {
2506         if (rte_pktmbuf_tailroom(pkt) >= data_len)
2507                 return 0;
2508
2509         /* attach an external buffer if supported */
2510         if (dev->extbuf && !virtio_dev_extbuf_alloc(pkt, data_len))
2511                 return 0;
2512
2513         /* check if chained buffers are allowed */
2514         if (!dev->linearbuf)
2515                 return 0;
2516
2517         return -1;
2518 }
2519
2520 __rte_always_inline
2521 static uint16_t
2522 virtio_dev_tx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
2523         struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t count,
2524         bool legacy_ol_flags)
2525 {
2526         uint16_t i;
2527         uint16_t free_entries;
2528         uint16_t dropped = 0;
2529         static bool allocerr_warned;
2530
2531         /*
2532          * The ordering between avail index and
2533          * desc reads needs to be enforced.
2534          */
2535         free_entries = __atomic_load_n(&vq->avail->idx, __ATOMIC_ACQUIRE) -
2536                         vq->last_avail_idx;
2537         if (free_entries == 0)
2538                 return 0;
2539
2540         rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
2541
2542         VHOST_LOG_DATA(DEBUG, "(%d) %s\n", dev->vid, __func__);
2543
2544         count = RTE_MIN(count, MAX_PKT_BURST);
2545         count = RTE_MIN(count, free_entries);
2546         VHOST_LOG_DATA(DEBUG, "(%d) about to dequeue %u buffers\n",
2547                         dev->vid, count);
2548
2549         if (rte_pktmbuf_alloc_bulk(mbuf_pool, pkts, count))
2550                 return 0;
2551
2552         for (i = 0; i < count; i++) {
2553                 struct buf_vector buf_vec[BUF_VECTOR_MAX];
2554                 uint16_t head_idx;
2555                 uint32_t buf_len;
2556                 uint16_t nr_vec = 0;
2557                 int err;
2558
2559                 if (unlikely(fill_vec_buf_split(dev, vq,
2560                                                 vq->last_avail_idx + i,
2561                                                 &nr_vec, buf_vec,
2562                                                 &head_idx, &buf_len,
2563                                                 VHOST_ACCESS_RO) < 0))
2564                         break;
2565
2566                 update_shadow_used_ring_split(vq, head_idx, 0);
2567
2568                 err = virtio_dev_pktmbuf_prep(dev, pkts[i], buf_len);
2569                 if (unlikely(err)) {
2570                         /*
2571                          * mbuf allocation fails for jumbo packets when external
2572                          * buffer allocation is not allowed and linear buffer
2573                          * is required. Drop this packet.
2574                          */
2575                         if (!allocerr_warned) {
2576                                 VHOST_LOG_DATA(ERR,
2577                                         "Failed mbuf alloc of size %d from %s on %s.\n",
2578                                         buf_len, mbuf_pool->name, dev->ifname);
2579                                 allocerr_warned = true;
2580                         }
2581                         dropped += 1;
2582                         i++;
2583                         break;
2584                 }
2585
2586                 err = copy_desc_to_mbuf(dev, vq, buf_vec, nr_vec, pkts[i],
2587                                 mbuf_pool, legacy_ol_flags);
2588                 if (unlikely(err)) {
2589                         if (!allocerr_warned) {
2590                                 VHOST_LOG_DATA(ERR,
2591                                         "Failed to copy desc to mbuf on %s.\n",
2592                                         dev->ifname);
2593                                 allocerr_warned = true;
2594                         }
2595                         dropped += 1;
2596                         i++;
2597                         break;
2598                 }
2599         }
2600
2601         if (dropped)
2602                 rte_pktmbuf_free_bulk(&pkts[i - 1], count - i + 1);
2603
2604         vq->last_avail_idx += i;
2605
2606         do_data_copy_dequeue(vq);
2607         if (unlikely(i < count))
2608                 vq->shadow_used_idx = i;
2609         if (likely(vq->shadow_used_idx)) {
2610                 flush_shadow_used_ring_split(dev, vq);
2611                 vhost_vring_call_split(dev, vq);
2612         }
2613
2614         return (i - dropped);
2615 }
2616
2617 __rte_noinline
2618 static uint16_t
2619 virtio_dev_tx_split_legacy(struct virtio_net *dev,
2620         struct vhost_virtqueue *vq, struct rte_mempool *mbuf_pool,
2621         struct rte_mbuf **pkts, uint16_t count)
2622 {
2623         return virtio_dev_tx_split(dev, vq, mbuf_pool, pkts, count, true);
2624 }
2625
2626 __rte_noinline
2627 static uint16_t
2628 virtio_dev_tx_split_compliant(struct virtio_net *dev,
2629         struct vhost_virtqueue *vq, struct rte_mempool *mbuf_pool,
2630         struct rte_mbuf **pkts, uint16_t count)
2631 {
2632         return virtio_dev_tx_split(dev, vq, mbuf_pool, pkts, count, false);
2633 }
2634
2635 static __rte_always_inline int
2636 vhost_reserve_avail_batch_packed(struct virtio_net *dev,
2637                                  struct vhost_virtqueue *vq,
2638                                  struct rte_mbuf **pkts,
2639                                  uint16_t avail_idx,
2640                                  uintptr_t *desc_addrs,
2641                                  uint16_t *ids)
2642 {
2643         bool wrap = vq->avail_wrap_counter;
2644         struct vring_packed_desc *descs = vq->desc_packed;
2645         uint64_t lens[PACKED_BATCH_SIZE];
2646         uint64_t buf_lens[PACKED_BATCH_SIZE];
2647         uint32_t buf_offset = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2648         uint16_t flags, i;
2649
2650         if (unlikely(avail_idx & PACKED_BATCH_MASK))
2651                 return -1;
2652         if (unlikely((avail_idx + PACKED_BATCH_SIZE) > vq->size))
2653                 return -1;
2654
2655         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
2656                 flags = descs[avail_idx + i].flags;
2657                 if (unlikely((wrap != !!(flags & VRING_DESC_F_AVAIL)) ||
2658                              (wrap == !!(flags & VRING_DESC_F_USED))  ||
2659                              (flags & PACKED_DESC_SINGLE_DEQUEUE_FLAG)))
2660                         return -1;
2661         }
2662
2663         rte_atomic_thread_fence(__ATOMIC_ACQUIRE);
2664
2665         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
2666                 lens[i] = descs[avail_idx + i].len;
2667
2668         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
2669                 desc_addrs[i] = vhost_iova_to_vva(dev, vq,
2670                                                   descs[avail_idx + i].addr,
2671                                                   &lens[i], VHOST_ACCESS_RW);
2672         }
2673
2674         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
2675                 if (unlikely(!desc_addrs[i]))
2676                         return -1;
2677                 if (unlikely((lens[i] != descs[avail_idx + i].len)))
2678                         return -1;
2679         }
2680
2681         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
2682                 if (virtio_dev_pktmbuf_prep(dev, pkts[i], lens[i]))
2683                         goto err;
2684         }
2685
2686         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
2687                 buf_lens[i] = pkts[i]->buf_len - pkts[i]->data_off;
2688
2689         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
2690                 if (unlikely(buf_lens[i] < (lens[i] - buf_offset)))
2691                         goto err;
2692         }
2693
2694         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
2695                 pkts[i]->pkt_len = lens[i] - buf_offset;
2696                 pkts[i]->data_len = pkts[i]->pkt_len;
2697                 ids[i] = descs[avail_idx + i].id;
2698         }
2699
2700         return 0;
2701
2702 err:
2703         return -1;
2704 }
2705
2706 static __rte_always_inline int
2707 virtio_dev_tx_batch_packed(struct virtio_net *dev,
2708                            struct vhost_virtqueue *vq,
2709                            struct rte_mbuf **pkts,
2710                            bool legacy_ol_flags)
2711 {
2712         uint16_t avail_idx = vq->last_avail_idx;
2713         uint32_t buf_offset = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2714         struct virtio_net_hdr *hdr;
2715         uintptr_t desc_addrs[PACKED_BATCH_SIZE];
2716         uint16_t ids[PACKED_BATCH_SIZE];
2717         uint16_t i;
2718
2719         if (vhost_reserve_avail_batch_packed(dev, vq, pkts, avail_idx,
2720                                              desc_addrs, ids))
2721                 return -1;
2722
2723         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
2724                 rte_prefetch0((void *)(uintptr_t)desc_addrs[i]);
2725
2726         vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
2727                 rte_memcpy(rte_pktmbuf_mtod_offset(pkts[i], void *, 0),
2728                            (void *)(uintptr_t)(desc_addrs[i] + buf_offset),
2729                            pkts[i]->pkt_len);
2730
2731         if (virtio_net_with_host_offload(dev)) {
2732                 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
2733                         hdr = (struct virtio_net_hdr *)(desc_addrs[i]);
2734                         vhost_dequeue_offload(hdr, pkts[i], legacy_ol_flags);
2735                 }
2736         }
2737
2738         if (virtio_net_is_inorder(dev))
2739                 vhost_shadow_dequeue_batch_packed_inorder(vq,
2740                         ids[PACKED_BATCH_SIZE - 1]);
2741         else
2742                 vhost_shadow_dequeue_batch_packed(dev, vq, ids);
2743
2744         vq_inc_last_avail_packed(vq, PACKED_BATCH_SIZE);
2745
2746         return 0;
2747 }
2748
2749 static __rte_always_inline int
2750 vhost_dequeue_single_packed(struct virtio_net *dev,
2751                             struct vhost_virtqueue *vq,
2752                             struct rte_mempool *mbuf_pool,
2753                             struct rte_mbuf *pkts,
2754                             uint16_t *buf_id,
2755                             uint16_t *desc_count,
2756                             bool legacy_ol_flags)
2757 {
2758         struct buf_vector buf_vec[BUF_VECTOR_MAX];
2759         uint32_t buf_len;
2760         uint16_t nr_vec = 0;
2761         int err;
2762         static bool allocerr_warned;
2763
2764         if (unlikely(fill_vec_buf_packed(dev, vq,
2765                                          vq->last_avail_idx, desc_count,
2766                                          buf_vec, &nr_vec,
2767                                          buf_id, &buf_len,
2768                                          VHOST_ACCESS_RO) < 0))
2769                 return -1;
2770
2771         if (unlikely(virtio_dev_pktmbuf_prep(dev, pkts, buf_len))) {
2772                 if (!allocerr_warned) {
2773                         VHOST_LOG_DATA(ERR,
2774                                 "Failed mbuf alloc of size %d from %s on %s.\n",
2775                                 buf_len, mbuf_pool->name, dev->ifname);
2776                         allocerr_warned = true;
2777                 }
2778                 return -1;
2779         }
2780
2781         err = copy_desc_to_mbuf(dev, vq, buf_vec, nr_vec, pkts,
2782                                 mbuf_pool, legacy_ol_flags);
2783         if (unlikely(err)) {
2784                 if (!allocerr_warned) {
2785                         VHOST_LOG_DATA(ERR,
2786                                 "Failed to copy desc to mbuf on %s.\n",
2787                                 dev->ifname);
2788                         allocerr_warned = true;
2789                 }
2790                 return -1;
2791         }
2792
2793         return 0;
2794 }
2795
2796 static __rte_always_inline int
2797 virtio_dev_tx_single_packed(struct virtio_net *dev,
2798                             struct vhost_virtqueue *vq,
2799                             struct rte_mempool *mbuf_pool,
2800                             struct rte_mbuf *pkts,
2801                             bool legacy_ol_flags)
2802 {
2803
2804         uint16_t buf_id, desc_count = 0;
2805         int ret;
2806
2807         ret = vhost_dequeue_single_packed(dev, vq, mbuf_pool, pkts, &buf_id,
2808                                         &desc_count, legacy_ol_flags);
2809
2810         if (likely(desc_count > 0)) {
2811                 if (virtio_net_is_inorder(dev))
2812                         vhost_shadow_dequeue_single_packed_inorder(vq, buf_id,
2813                                                                    desc_count);
2814                 else
2815                         vhost_shadow_dequeue_single_packed(vq, buf_id,
2816                                         desc_count);
2817
2818                 vq_inc_last_avail_packed(vq, desc_count);
2819         }
2820
2821         return ret;
2822 }
2823
2824 __rte_always_inline
2825 static uint16_t
2826 virtio_dev_tx_packed(struct virtio_net *dev,
2827                      struct vhost_virtqueue *__rte_restrict vq,
2828                      struct rte_mempool *mbuf_pool,
2829                      struct rte_mbuf **__rte_restrict pkts,
2830                      uint32_t count,
2831                      bool legacy_ol_flags)
2832 {
2833         uint32_t pkt_idx = 0;
2834
2835         if (rte_pktmbuf_alloc_bulk(mbuf_pool, pkts, count))
2836                 return 0;
2837
2838         do {
2839                 rte_prefetch0(&vq->desc_packed[vq->last_avail_idx]);
2840
2841                 if (count - pkt_idx >= PACKED_BATCH_SIZE) {
2842                         if (!virtio_dev_tx_batch_packed(dev, vq,
2843                                                         &pkts[pkt_idx],
2844                                                         legacy_ol_flags)) {
2845                                 pkt_idx += PACKED_BATCH_SIZE;
2846                                 continue;
2847                         }
2848                 }
2849
2850                 if (virtio_dev_tx_single_packed(dev, vq, mbuf_pool,
2851                                                 pkts[pkt_idx],
2852                                                 legacy_ol_flags))
2853                         break;
2854                 pkt_idx++;
2855         } while (pkt_idx < count);
2856
2857         if (pkt_idx != count)
2858                 rte_pktmbuf_free_bulk(&pkts[pkt_idx], count - pkt_idx);
2859
2860         if (vq->shadow_used_idx) {
2861                 do_data_copy_dequeue(vq);
2862
2863                 vhost_flush_dequeue_shadow_packed(dev, vq);
2864                 vhost_vring_call_packed(dev, vq);
2865         }
2866
2867         return pkt_idx;
2868 }
2869
2870 __rte_noinline
2871 static uint16_t
2872 virtio_dev_tx_packed_legacy(struct virtio_net *dev,
2873         struct vhost_virtqueue *__rte_restrict vq, struct rte_mempool *mbuf_pool,
2874         struct rte_mbuf **__rte_restrict pkts, uint32_t count)
2875 {
2876         return virtio_dev_tx_packed(dev, vq, mbuf_pool, pkts, count, true);
2877 }
2878
2879 __rte_noinline
2880 static uint16_t
2881 virtio_dev_tx_packed_compliant(struct virtio_net *dev,
2882         struct vhost_virtqueue *__rte_restrict vq, struct rte_mempool *mbuf_pool,
2883         struct rte_mbuf **__rte_restrict pkts, uint32_t count)
2884 {
2885         return virtio_dev_tx_packed(dev, vq, mbuf_pool, pkts, count, false);
2886 }
2887
2888 uint16_t
2889 rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
2890         struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t count)
2891 {
2892         struct virtio_net *dev;
2893         struct rte_mbuf *rarp_mbuf = NULL;
2894         struct vhost_virtqueue *vq;
2895         int16_t success = 1;
2896
2897         dev = get_device(vid);
2898         if (!dev)
2899                 return 0;
2900
2901         if (unlikely(!(dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET))) {
2902                 VHOST_LOG_DATA(ERR,
2903                         "(%d) %s: built-in vhost net backend is disabled.\n",
2904                         dev->vid, __func__);
2905                 return 0;
2906         }
2907
2908         if (unlikely(!is_valid_virt_queue_idx(queue_id, 1, dev->nr_vring))) {
2909                 VHOST_LOG_DATA(ERR,
2910                         "(%d) %s: invalid virtqueue idx %d.\n",
2911                         dev->vid, __func__, queue_id);
2912                 return 0;
2913         }
2914
2915         vq = dev->virtqueue[queue_id];
2916
2917         if (unlikely(rte_spinlock_trylock(&vq->access_lock) == 0))
2918                 return 0;
2919
2920         if (unlikely(!vq->enabled)) {
2921                 count = 0;
2922                 goto out_access_unlock;
2923         }
2924
2925         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
2926                 vhost_user_iotlb_rd_lock(vq);
2927
2928         if (unlikely(!vq->access_ok))
2929                 if (unlikely(vring_translate(dev, vq) < 0)) {
2930                         count = 0;
2931                         goto out;
2932                 }
2933
2934         /*
2935          * Construct a RARP broadcast packet, and inject it to the "pkts"
2936          * array, to looks like that guest actually send such packet.
2937          *
2938          * Check user_send_rarp() for more information.
2939          *
2940          * broadcast_rarp shares a cacheline in the virtio_net structure
2941          * with some fields that are accessed during enqueue and
2942          * __atomic_compare_exchange_n causes a write if performed compare
2943          * and exchange. This could result in false sharing between enqueue
2944          * and dequeue.
2945          *
2946          * Prevent unnecessary false sharing by reading broadcast_rarp first
2947          * and only performing compare and exchange if the read indicates it
2948          * is likely to be set.
2949          */
2950         if (unlikely(__atomic_load_n(&dev->broadcast_rarp, __ATOMIC_ACQUIRE) &&
2951                         __atomic_compare_exchange_n(&dev->broadcast_rarp,
2952                         &success, 0, 0, __ATOMIC_RELEASE, __ATOMIC_RELAXED))) {
2953
2954                 rarp_mbuf = rte_net_make_rarp_packet(mbuf_pool, &dev->mac);
2955                 if (rarp_mbuf == NULL) {
2956                         VHOST_LOG_DATA(ERR, "Failed to make RARP packet.\n");
2957                         count = 0;
2958                         goto out;
2959                 }
2960                 /*
2961                  * Inject it to the head of "pkts" array, so that switch's mac
2962                  * learning table will get updated first.
2963                  */
2964                 pkts[0] = rarp_mbuf;
2965                 pkts++;
2966                 count -= 1;
2967         }
2968
2969         if (vq_is_packed(dev)) {
2970                 if (dev->flags & VIRTIO_DEV_LEGACY_OL_FLAGS)
2971                         count = virtio_dev_tx_packed_legacy(dev, vq, mbuf_pool, pkts, count);
2972                 else
2973                         count = virtio_dev_tx_packed_compliant(dev, vq, mbuf_pool, pkts, count);
2974         } else {
2975                 if (dev->flags & VIRTIO_DEV_LEGACY_OL_FLAGS)
2976                         count = virtio_dev_tx_split_legacy(dev, vq, mbuf_pool, pkts, count);
2977                 else
2978                         count = virtio_dev_tx_split_compliant(dev, vq, mbuf_pool, pkts, count);
2979         }
2980
2981 out:
2982         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
2983                 vhost_user_iotlb_rd_unlock(vq);
2984
2985 out_access_unlock:
2986         rte_spinlock_unlock(&vq->access_lock);
2987
2988         if (unlikely(rarp_mbuf != NULL))
2989                 count += 1;
2990
2991         return count;
2992 }