aboutsummaryrefslogtreecommitdiffstats
path: root/src
diff options
context:
space:
mode:
authorDamjan Marion <damarion@cisco.com>2023-11-06 00:07:15 +0000
committerOle Tr�an <otroan@employees.org>2023-11-06 10:15:12 +0000
commit7444fd2a0d177796cacb7c040323cdc327afa97a (patch)
treedc7050e8171b3628c0368821d5d1c1e85b8c6bd8 /src
parent60529a8ef81d08c7aa0f7eaea402435b3c6d1eb6 (diff)
pci: remove unnecessary ioctl() call and improve logging in vfio_set_irqs
Type: improvement Change-Id: Ic8e2785bf375882defe5a1d299948d522cdd4895 Signed-off-by: Damjan Marion <damarion@cisco.com>
Diffstat (limited to 'src')
-rw-r--r--src/vlib/linux/pci.c33
-rw-r--r--src/vlib/linux/vfio.c48
-rw-r--r--src/vlib/linux/vfio.h2
3 files changed, 49 insertions, 34 deletions
diff --git a/src/vlib/linux/pci.c b/src/vlib/linux/pci.c
index 62db8204182..0435d569bec 100644
--- a/src/vlib/linux/pci.c
+++ b/src/vlib/linux/pci.c
@@ -648,33 +648,14 @@ vfio_set_irqs (vlib_main_t * vm, linux_pci_device_t * p, u32 index, u32 start,
{
int data_len = efds ? count * sizeof (int) : 0;
u8 buf[sizeof (struct vfio_irq_set) + data_len];
- struct vfio_irq_info ii = { 0 };
struct vfio_irq_set *irq_set = (struct vfio_irq_set *) buf;
-
- ii.argsz = sizeof (struct vfio_irq_info);
- ii.index = index;
-
- if (ioctl (p->fd, VFIO_DEVICE_GET_IRQ_INFO, &ii) < 0)
- return clib_error_return_unix (0, "ioctl(VFIO_DEVICE_GET_IRQ_INFO) "
- "'%U'", format_vlib_pci_addr, &p->addr);
-
- log_debug (p, "%s index:%u count:%u flags: %s%s%s%s(0x%x)", __func__,
- ii.index, ii.count,
- ii.flags & VFIO_IRQ_INFO_EVENTFD ? "eventfd " : "",
- ii.flags & VFIO_IRQ_INFO_MASKABLE ? "maskable " : "",
- ii.flags & VFIO_IRQ_INFO_AUTOMASKED ? "automasked " : "",
- ii.flags & VFIO_IRQ_INFO_NORESIZE ? "noresize " : "", ii.flags);
-
- if (ii.count < start + count)
- return clib_error_return_unix (0, "vfio_set_irq: unexistng interrupt on "
- "'%U'", format_vlib_pci_addr, &p->addr);
-
-
if (efds)
{
+ int *data = (int *) irq_set->data;
flags |= VFIO_IRQ_SET_DATA_EVENTFD;
- clib_memcpy_fast (&irq_set->data, efds, data_len);
+ for (u32 i = 0; i < count; i++)
+ data[i] = efds[i];
}
else
flags |= VFIO_IRQ_SET_DATA_NONE;
@@ -689,11 +670,11 @@ vfio_set_irqs (vlib_main_t * vm, linux_pci_device_t * p, u32 index, u32 start,
irq_set->flags = flags;
if (ioctl (p->fd, VFIO_DEVICE_SET_IRQS, irq_set) < 0)
- return clib_error_return_unix (0, "%U:ioctl(VFIO_DEVICE_SET_IRQS) "
- "[index = %u, start = %u, count = %u, "
- "flags = 0x%x]",
+ return clib_error_return_unix (0, "%U:ioctl(VFIO_DEVICE_SET_IRQS)\n%U",
format_vlib_pci_addr, &p->addr,
- index, start, count, flags);
+ format_vfio_irq_set, irq_set);
+
+ log_debug (p, "%s:\n%U", __func__, format_vfio_irq_set, irq_set);
return 0;
}
diff --git a/src/vlib/linux/vfio.c b/src/vlib/linux/vfio.c
index ee04081d767..1462cc6c7ca 100644
--- a/src/vlib/linux/vfio.c
+++ b/src/vlib/linux/vfio.c
@@ -301,10 +301,44 @@ format_vfio_region_info (u8 * s, va_list * args)
return s;
}
-/*
- * fd.io coding-style-patch-verification: ON
- *
- * Local Variables:
- * eval: (c-set-style "gnu")
- * End:
- */
+u8 *
+format_vfio_irq_set (u8 *s, va_list *args)
+{
+ struct vfio_irq_set *is = va_arg (*args, struct vfio_irq_set *);
+ u32 indent = format_get_indent (s);
+
+ s = format (s, "index:%u start:%u count:%u flags: 0x%x", is->index,
+ is->start, is->count, is->flags);
+
+ s = format (s, " (data:");
+ if (is->flags & VFIO_IRQ_SET_DATA_NONE)
+ s = format (s, " none");
+ if (is->flags & VFIO_IRQ_SET_DATA_BOOL)
+ s = format (s, " bool");
+ if (is->flags & VFIO_IRQ_SET_DATA_EVENTFD)
+ s = format (s, " eventfd");
+
+ s = format (s, ", action:");
+ if (is->flags & VFIO_IRQ_SET_ACTION_MASK)
+ s = format (s, " mask");
+ if (is->flags & VFIO_IRQ_SET_ACTION_UNMASK)
+ s = format (s, " unmask");
+ if (is->flags & VFIO_IRQ_SET_ACTION_TRIGGER)
+ s = format (s, " trigger");
+ vec_add1 (s, ')');
+
+ if (is->flags & VFIO_IRQ_SET_DATA_EVENTFD)
+ {
+ s = format (s, "\n%U eventfd data:", format_white_space, indent);
+ for (u32 i = 0; i < is->count; i++)
+ s = format (s, " %d", ((int *) (is->data))[i]);
+ }
+ if (is->flags & VFIO_IRQ_SET_DATA_BOOL)
+ {
+ s = format (s, "\n%U bool data:", format_white_space, indent);
+ for (u32 i = 0; i < is->count; i++)
+ s = format (s, " %u", is->data);
+ }
+
+ return s;
+}
diff --git a/src/vlib/linux/vfio.h b/src/vlib/linux/vfio.h
index fe4f0f75346..c2bb2e9b667 100644
--- a/src/vlib/linux/vfio.h
+++ b/src/vlib/linux/vfio.h
@@ -50,7 +50,7 @@ clib_error_t *linux_vfio_group_get_device_fd (vlib_pci_addr_t * addr,
int *fd, int *is_noiommu);
format_function_t format_vfio_region_info;
-
+format_function_t format_vfio_irq_set;
#endif /* included_vlib_linux_vfio_h */
href='#n377'>377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 
/*
 *------------------------------------------------------------------
 * vhost-user-output
 *
 * Copyright (c) 2014-2018 Cisco and/or its affiliates.
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at:
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 *------------------------------------------------------------------
 */

#include <stddef.h>
#include <fcntl.h>		/* for open */
#include <sys/ioctl.h>
#include <sys/socket.h>
#include <sys/un.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <sys/uio.h>		/* for iovec */
#include <netinet/in.h>
#include <sys/vfs.h>

#include <linux/if_arp.h>
#include <linux/if_tun.h>

#include <vlib/vlib.h>
#include <vlib/unix/unix.h>

#include <vnet/ethernet/ethernet.h>
#include <vnet/devices/devices.h>
#include <vnet/feature/feature.h>
#include <vnet/ip/ip_psh_cksum.h>

#include <vnet/devices/virtio/vhost_user.h>
#include <vnet/devices/virtio/vhost_user_inline.h>

#include <vnet/gso/hdr_offset_parser.h>
/*
 * On the transmit side, we keep processing the buffers from vlib in the while
 * loop and prepare the copy order to be executed later. However, the static
 * array which we keep the copy order is limited to VHOST_USER_COPY_ARRAY_N
 * entries. In order to not corrupt memory, we have to do the copy when the
 * static array reaches the copy threshold. We subtract 40 in case the code
 * goes into the inner loop for a maximum of 64k frames which may require
 * more array entries. We subtract 200 because our default buffer size is
 * 2048 and the default desc len is likely 1536. While it takes less than 40
 * vlib buffers for the jumbo frame, it may take twice as much descriptors
 * for the same jumbo frame. Use 200 for the extra head room.
 */
#define VHOST_USER_TX_COPY_THRESHOLD (VHOST_USER_COPY_ARRAY_N - 200)

extern vnet_device_class_t vhost_user_device_class;

#define foreach_vhost_user_tx_func_error      \
  _(NONE, "no error")  \
  _(NOT_READY, "vhost vring not ready")  \
  _(DOWN, "vhost interface is down")  \
  _(PKT_DROP_NOBUF, "tx packet drops (no available descriptors)")  \
  _(PKT_DROP_NOMRG, "tx packet drops (cannot merge descriptors)")  \
  _(MMAP_FAIL, "mmap failure") \
  _(INDIRECT_OVERFLOW, "indirect descriptor table overflow")

typedef enum
{
#define _(f,s) VHOST_USER_TX_FUNC_ERROR_##f,
  foreach_vhost_user_tx_func_error
#undef _
    VHOST_USER_TX_FUNC_N_ERROR,
} vhost_user_tx_func_error_t;

static __clib_unused char *vhost_user_tx_func_error_strings[] = {
#define _(n,s) s,
  foreach_vhost_user_tx_func_error
#undef _
};

static __clib_unused u8 *
format_vhost_user_interface_name (u8 * s, va_list * args)
{
  u32 i = va_arg (*args, u32);
  u32 show_dev_instance = ~0;
  vhost_user_main_t *vum = &vhost_user_main;

  if (i < vec_len (vum->show_dev_instance_by_real_dev_instance))
    show_dev_instance = vum->show_dev_instance_by_real_dev_instance[i];

  if (show_dev_instance != ~0)
    i = show_dev_instance;

  s = format (s, "VirtualEthernet0/0/%d", i);
  return s;
}

static __clib_unused int
vhost_user_name_renumber (vnet_hw_interface_t * hi, u32 new_dev_instance)
{
  // FIXME: check if the new dev instance is already used
  vhost_user_main_t *vum = &vhost_user_main;
  vhost_user_intf_t *vui = pool_elt_at_index (vum->vhost_user_interfaces,
					      hi->dev_instance);

  vec_validate_init_empty (vum->show_dev_instance_by_real_dev_instance,
			   hi->dev_instance, ~0);

  vum->show_dev_instance_by_real_dev_instance[hi->dev_instance] =
    new_dev_instance;

  vu_log_debug (vui, "renumbered vhost-user interface dev_instance %d to %d",
		hi->dev_instance, new_dev_instance);

  return 0;
}

static_always_inline void
vhost_user_tx_trace (vhost_trace_t * t,
		     vhost_user_intf_t * vui, u16 qid,
		     vlib_buffer_t * b, vhost_user_vring_t * rxvq)
{
  vhost_user_main_t *vum = &vhost_user_main;
  u32 last_avail_idx = rxvq->last_avail_idx;
  u32 desc_current = rxvq->avail->ring[last_avail_idx & rxvq->qsz_mask];
  vnet_virtio_vring_desc_t *hdr_desc = 0;
  u32 hint = 0;

  clib_memset (t, 0, sizeof (*t));
  t->device_index = vui - vum->vhost_user_interfaces;
  t->qid = qid;

  hdr_desc = &rxvq->desc[desc_current];
  if (rxvq->desc[desc_current].flags & VRING_DESC_F_INDIRECT)
    {
      t->virtio_ring_flags |= 1 << VIRTIO_TRACE_F_INDIRECT;
      /* Header is the first here */
      hdr_desc = map_guest_mem (vui, rxvq->desc[desc_current].addr, &hint);
    }
  if (rxvq->desc[desc_current].flags & VRING_DESC_F_NEXT)
    {
      t->virtio_ring_flags |= 1 << VIRTIO_TRACE_F_SIMPLE_CHAINED;
    }
  if (!(rxvq->desc[desc_current].flags & VRING_DESC_F_NEXT) &&
      !(rxvq->desc[desc_current].flags & VRING_DESC_F_INDIRECT))
    {
      t->virtio_ring_flags |= 1 << VIRTIO_TRACE_F_SINGLE_DESC;
    }

  t->first_desc_len = hdr_desc ? hdr_desc->len : 0;
}

static_always_inline u32
vhost_user_tx_copy (vhost_user_intf_t * vui, vhost_copy_t * cpy,
		    u16 copy_len, u32 * map_hint)
{
  void *dst0, *dst1, *dst2, *dst3;
  if (PREDICT_TRUE (copy_len >= 4))
    {
      if (PREDICT_FALSE (!(dst2 = map_guest_mem (vui, cpy[0].dst, map_hint))))
	return 1;
      if (PREDICT_FALSE (!(dst3 = map_guest_mem (vui, cpy[1].dst, map_hint))))
	return 1;
      while (PREDICT_TRUE (copy_len >= 4))
	{
	  dst0 = dst2;
	  dst1 = dst3;

	  if (PREDICT_FALSE
	      (!(dst2 = map_guest_mem (vui, cpy[2].dst, map_hint))))
	    return 1;
	  if (PREDICT_FALSE
	      (!(dst3 = map_guest_mem (vui, cpy[3].dst, map_hint))))
	    return 1;

	  clib_prefetch_load ((void *) cpy[2].src);
	  clib_prefetch_load ((void *) cpy[3].src);

	  clib_memcpy_fast (dst0, (void *) cpy[0].src, cpy[0].len);
	  clib_memcpy_fast (dst1, (void *) cpy[1].src, cpy[1].len);

	  vhost_user_log_dirty_pages_2 (vui, cpy[0].dst, cpy[0].len, 1);
	  vhost_user_log_dirty_pages_2 (vui, cpy[1].dst, cpy[1].len, 1);
	  copy_len -= 2;
	  cpy += 2;
	}
    }
  while (copy_len)
    {
      if (PREDICT_FALSE (!(dst0 = map_guest_mem (vui, cpy->dst, map_hint))))
	return 1;
      clib_memcpy_fast (dst0, (void *) cpy->src, cpy->len);
      vhost_user_log_dirty_pages_2 (vui, cpy->dst, cpy->len, 1);
      copy_len -= 1;
      cpy += 1;
    }
  return 0;
}

static_always_inline void
vhost_user_handle_tx_offload (vhost_user_intf_t *vui, vlib_buffer_t *b,
			      vnet_virtio_net_hdr_t *hdr)
{
  generic_header_offset_t gho = { 0 };
  int is_ip4 = b->flags & VNET_BUFFER_F_IS_IP4;
  int is_ip6 = b->flags & VNET_BUFFER_F_IS_IP6;
  vnet_buffer_oflags_t oflags = vnet_buffer (b)->oflags;
  u16 psh_cksum = 0;
  ip4_header_t *ip4 = 0;
  ip6_header_t *ip6 = 0;

  ASSERT (!(is_ip4 && is_ip6));
  vnet_generic_header_offset_parser (b, &gho, 1 /* l2 */ , is_ip4, is_ip6);
  if (oflags & VNET_BUFFER_OFFLOAD_F_IP_CKSUM)
    {
      ip4 =
	(ip4_header_t *) (vlib_buffer_get_current (b) + gho.l3_hdr_offset);
      ip4->checksum = ip4_header_checksum (ip4);
      psh_cksum = ip4_pseudo_header_cksum (ip4);
    }
  else
    {
      ip6 = (ip6_header_t *) (vlib_buffer_get_current (b) + gho.l3_hdr_offset);
      psh_cksum = ip6_pseudo_header_cksum (ip6);
    }

  /* checksum offload */
  if (oflags & VNET_BUFFER_OFFLOAD_F_UDP_CKSUM)
    {
      udp_header_t *udp =
	(udp_header_t *) (vlib_buffer_get_current (b) + gho.l4_hdr_offset);
      udp->checksum = psh_cksum;
      hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
      hdr->csum_start = gho.l4_hdr_offset;
      hdr->csum_offset = offsetof (udp_header_t, checksum);
    }
  else if (oflags & VNET_BUFFER_OFFLOAD_F_TCP_CKSUM)
    {
      tcp_header_t *tcp =
	(tcp_header_t *) (vlib_buffer_get_current (b) + gho.l4_hdr_offset);
      tcp->checksum = psh_cksum;
      hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
      hdr->csum_start = gho.l4_hdr_offset;
      hdr->csum_offset = offsetof (tcp_header_t, checksum);
    }

  /* GSO offload */
  if (b->flags & VNET_BUFFER_F_GSO)
    {
      if (oflags & VNET_BUFFER_OFFLOAD_F_TCP_CKSUM)
	{
	  if (is_ip4 &&
	      (vui->features & VIRTIO_FEATURE (VIRTIO_NET_F_GUEST_TSO4)))
	    {
	      hdr->gso_size = vnet_buffer2 (b)->gso_size;
	      hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
	    }
	  else if (is_ip6 &&
		   (vui->features & VIRTIO_FEATURE (VIRTIO_NET_F_GUEST_TSO6)))
	    {
	      hdr->gso_size = vnet_buffer2 (b)->gso_size;
	      hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
	    }
	}
      else if ((vui->features & VIRTIO_FEATURE (VIRTIO_NET_F_GUEST_UFO)) &&
	       (oflags & VNET_BUFFER_OFFLOAD_F_UDP_CKSUM))
	{
	  hdr->gso_size = vnet_buffer2 (b)->gso_size;
	  hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
	}
    }
}

static_always_inline void
vhost_user_mark_desc_available (vlib_main_t * vm, vhost_user_intf_t * vui,
				vhost_user_vring_t * rxvq,
				u16 * n_descs_processed, u8 chained,
				vlib_frame_t * frame, u32 n_left)
{
  u16 desc_idx, flags;
  vnet_virtio_vring_packed_desc_t *desc_table = rxvq->packed_desc;
  u16 last_used_idx = rxvq->last_used_idx;

  if (PREDICT_FALSE (*n_descs_processed == 0))
    return;

  if (rxvq->used_wrap_counter)
    flags = desc_table[last_used_idx & rxvq->qsz_mask].flags |
      (VRING_DESC_F_AVAIL | VRING_DESC_F_USED);
  else
    flags = desc_table[last_used_idx & rxvq->qsz_mask].flags &
      ~(VRING_DESC_F_AVAIL | VRING_DESC_F_USED);

  vhost_user_advance_last_used_idx (rxvq);

  for (desc_idx = 1; desc_idx < *n_descs_processed; desc_idx++)
    {
      if (rxvq->used_wrap_counter)
	desc_table[rxvq->last_used_idx & rxvq->qsz_mask].flags |=
	  (VRING_DESC_F_AVAIL | VRING_DESC_F_USED);
      else
	desc_table[rxvq->last_used_idx & rxvq->qsz_mask].flags &=
	  ~(VRING_DESC_F_AVAIL | VRING_DESC_F_USED);
      vhost_user_advance_last_used_idx (rxvq);
    }

  desc_table[last_used_idx & rxvq->qsz_mask].flags = flags;

  *n_descs_processed = 0;

  if (chained)
    {
      vnet_virtio_vring_packed_desc_t *desc_table = rxvq->packed_desc;

      while (desc_table[rxvq->last_used_idx & rxvq->qsz_mask].flags &
	     VRING_DESC_F_NEXT)
	vhost_user_advance_last_used_idx (rxvq);

      /* Advance past the current chained table entries */
      vhost_user_advance_last_used_idx (rxvq);
    }

  /* interrupt (call) handling */
  if ((rxvq->callfd_idx != ~0) &&
      (rxvq->avail_event->flags != VRING_EVENT_F_DISABLE))
    {
      vhost_user_main_t *vum = &vhost_user_main;

      rxvq->n_since_last_int += frame->n_vectors - n_left;
      if (rxvq->n_since_last_int > vum->coalesce_frames)
	vhost_user_send_call (vm, vui, rxvq);
    }
}

static_always_inline void
vhost_user_tx_trace_packed (vhost_trace_t * t, vhost_user_intf_t * vui,
			    u16 qid, vlib_buffer_t * b,
			    vhost_user_vring_t * rxvq)
{
  vhost_user_main_t *vum = &vhost_user_main;
  u32 last_avail_idx = rxvq->last_avail_idx;
  u32 desc_current = last_avail_idx & rxvq->qsz_mask;
  vnet_virtio_vring_packed_desc_t *hdr_desc = 0;
  u32 hint = 0;

  clib_memset (t, 0, sizeof (*t));
  t->device_index = vui - vum->vhost_user_interfaces;
  t->qid = qid;

  hdr_desc = &rxvq->packed_desc[desc_current];
  if (rxvq->packed_desc[desc_current].flags & VRING_DESC_F_INDIRECT)
    {
      t->virtio_ring_flags |= 1 << VIRTIO_TRACE_F_INDIRECT;
      /* Header is the first here */
      hdr_desc = map_guest_mem (vui, rxvq->packed_desc[desc_current].addr,
				&hint);
    }
  if (rxvq->packed_desc[desc_current].flags & VRING_DESC_F_NEXT)
    {
      t->virtio_ring_flags |= 1 << VIRTIO_TRACE_F_SIMPLE_CHAINED;
    }
  if (!(rxvq->packed_desc[desc_current].flags & VRING_DESC_F_NEXT) &&
      !(rxvq->packed_desc[desc_current].flags & VRING_DESC_F_INDIRECT))
    {
      t->virtio_ring_flags |= 1 << VIRTIO_TRACE_F_SINGLE_DESC;
    }

  t->first_desc_len = hdr_desc ? hdr_desc->len : 0;
}

static_always_inline uword
vhost_user_device_class_packed (vlib_main_t *vm, vlib_node_runtime_t *node,
				vlib_frame_t *frame, vhost_user_intf_t *vui,
				vhost_user_vring_t *rxvq)
{
  u32 *buffers = vlib_frame_vector_args (frame);
  u32 n_left = frame->n_vectors;
  vhost_user_main_t *vum = &vhost_user_main;
  u32 qid = rxvq->qid;
  u8 error;
  u32 thread_index = vm->thread_index;
  vhost_cpu_t *cpu = &vum->cpus[thread_index];
  u32 map_hint = 0;
  u8 retry = 8;
  u16 copy_len;
  u16 tx_headers_len;
  vnet_virtio_vring_packed_desc_t *desc_table;
  u32 or_flags;
  u16 desc_head, desc_index, desc_len;
  u16 n_descs_processed;
  u8 indirect, chained;

retry:
  error = VHOST_USER_TX_FUNC_ERROR_NONE;
  tx_headers_len = 0;
  copy_len = 0;
  n_descs_processed = 0;

  while (n_left > 0)
    {
      vlib_buffer_t *b0, *current_b0;
      uword buffer_map_addr;
      u32 buffer_len;
      u16 bytes_left;
      u32 total_desc_len = 0;
      u16 n_entries = 0;

      indirect = 0;
      chained = 0;
      if (PREDICT_TRUE (n_left > 1))
	vlib_prefetch_buffer_with_index (vm, buffers[1], LOAD);

      b0 = vlib_get_buffer (vm, buffers[0]);
      if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
	{
	  cpu->current_trace = vlib_add_trace (vm, node, b0,
					       sizeof (*cpu->current_trace));
	  vhost_user_tx_trace_packed (cpu->current_trace, vui, qid / 2, b0,
				      rxvq);
	}

      desc_table = rxvq->packed_desc;
      desc_head = desc_index = rxvq->last_avail_idx & rxvq->qsz_mask;
      if (PREDICT_FALSE (!vhost_user_packed_desc_available (rxvq, desc_head)))
	{
	  error = VHOST_USER_TX_FUNC_ERROR_PKT_DROP_NOBUF;
	  goto done;
	}
      /*
       * Go deeper in case of indirect descriptor.
       * To test it, turn off mrg_rxbuf.
       */
      if (desc_table[desc_head].flags & VRING_DESC_F_INDIRECT)
	{
	  indirect = 1;
	  if (PREDICT_FALSE (desc_table[desc_head].len <
			     sizeof (vnet_virtio_vring_packed_desc_t)))
	    {
	      error = VHOST_USER_TX_FUNC_ERROR_INDIRECT_OVERFLOW;
	      goto done;
	    }
	  n_entries = desc_table[desc_head].len >> 4;
	  desc_table = map_guest_mem (vui, desc_table[desc_index].addr,
				      &map_hint);
	  if (PREDICT_FALSE (desc_table == 0))
	    {
	      error = VHOST_USER_TX_FUNC_ERROR_MMAP_FAIL;
	      goto done;
	    }
	  desc_index = 0;
	}
      else if (rxvq->packed_desc[desc_head].flags & VRING_DESC_F_NEXT)
	chained = 1;

      desc_len = vui->virtio_net_hdr_sz;
      buffer_map_addr = desc_table[desc_index].addr;
      buffer_len = desc_table[desc_index].len;

      /* Get a header from the header array */
      vnet_virtio_net_hdr_mrg_rxbuf_t *hdr = &cpu->tx_headers[tx_headers_len];
      tx_headers_len++;
      hdr->hdr.flags = 0;
      hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
      hdr->num_buffers = 1;

      or_flags = (b0->flags & VNET_BUFFER_F_OFFLOAD);

      /* Guest supports csum offload and buffer requires checksum offload? */
      if (or_flags &&
	  (vui->features & VIRTIO_FEATURE (VIRTIO_NET_F_GUEST_CSUM)))
	vhost_user_handle_tx_offload (vui, b0, &hdr->hdr);

      /* Prepare a copy order executed later for the header */
      ASSERT (copy_len < VHOST_USER_COPY_ARRAY_N);
      vhost_copy_t *cpy = &cpu->copy[copy_len];
      copy_len++;
      cpy->len = vui->virtio_net_hdr_sz;
      cpy->dst = buffer_map_addr;
      cpy->src = (uword) hdr;

      buffer_map_addr += vui->virtio_net_hdr_sz;
      buffer_len -= vui->virtio_net_hdr_sz;
      bytes_left = b0->current_length;
      current_b0 = b0;
      while (1)
	{
	  if (buffer_len == 0)
	    {
	      /* Get new output */
	      if (chained)
		{
		  /*
		   * Next one is chained
		   * Test it with both indirect and mrg_rxbuf off
		   */
		  if (PREDICT_FALSE (!(desc_table[desc_index].flags &
				       VRING_DESC_F_NEXT)))
		    {
		      /*
		       * Last descriptor in chain.
		       * Dequeue queued descriptors for this packet
		       */
		      vhost_user_dequeue_chained_descs (rxvq,
							&n_descs_processed);
		      error = VHOST_USER_TX_FUNC_ERROR_PKT_DROP_NOBUF;
		      goto done;
		    }
		  vhost_user_advance_last_avail_idx (rxvq);
		  desc_index = rxvq->last_avail_idx & rxvq->qsz_mask;
		  n_descs_processed++;
		  buffer_map_addr = desc_table[desc_index].addr;
		  buffer_len = desc_table[desc_index].len;
		  total_desc_len += desc_len;
		  desc_len = 0;
		}
	      else if (indirect)
		{
		  /*
		   * Indirect table
		   * Test it with mrg_rxnuf off
		   */
		  if (PREDICT_TRUE (n_entries > 0))
		    n_entries--;
		  else
		    {
		      /* Dequeue queued descriptors for this packet */
		      vhost_user_dequeue_chained_descs (rxvq,
							&n_descs_processed);
		      error = VHOST_USER_TX_FUNC_ERROR_INDIRECT_OVERFLOW;
		      goto done;
		    }
		  total_desc_len += desc_len;
		  desc_index = (desc_index + 1) & rxvq->qsz_mask;
		  buffer_map_addr = desc_table[desc_index].addr;
		  buffer_len = desc_table[desc_index].len;
		  desc_len = 0;
		}
	      else if (vui->virtio_net_hdr_sz == 12)
		{
		  /*
		   * MRG is available
		   * This is the default setting for the guest VM
		   */
		  vnet_virtio_net_hdr_mrg_rxbuf_t *hdr =
		    &cpu->tx_headers[tx_headers_len - 1];

		  desc_table[desc_index].len = desc_len;
		  vhost_user_advance_last_avail_idx (rxvq);
		  desc_head = desc_index =
		    rxvq->last_avail_idx & rxvq->qsz_mask;
		  hdr->num_buffers++;
		  n_descs_processed++;
		  desc_len = 0;

		  if (PREDICT_FALSE (!vhost_user_packed_desc_available
				     (rxvq, desc_index)))
		    {
		      /* Dequeue queued descriptors for this packet */
		      vhost_user_dequeue_descs (rxvq, hdr,
						&n_descs_processed);
		      error = VHOST_USER_TX_FUNC_ERROR_PKT_DROP_NOBUF;
		      goto done;
		    }

		  buffer_map_addr = desc_table[desc_index].addr;
		  buffer_len = desc_table[desc_index].len;
		}
	      else
		{
		  error = VHOST_USER_TX_FUNC_ERROR_PKT_DROP_NOMRG;
		  goto done;
		}
	    }

	  ASSERT (copy_len < VHOST_USER_COPY_ARRAY_N);
	  vhost_copy_t *cpy = &cpu->copy[copy_len];
	  copy_len++;
	  cpy->len = bytes_left;
	  cpy->len = (cpy->len > buffer_len) ? buffer_len : cpy->len;
	  cpy->dst = buffer_map_addr;
	  cpy->src = (uword) vlib_buffer_get_current (current_b0) +
	    current_b0->current_length - bytes_left;

	  bytes_left -= cpy->len;
	  buffer_len -= cpy->len;
	  buffer_map_addr += cpy->len;
	  desc_len += cpy->len;

	  clib_prefetch_load (&rxvq->packed_desc);

	  /* Check if vlib buffer has more data. If not, get more or break */
	  if (PREDICT_TRUE (!bytes_left))
	    {
	      if (PREDICT_FALSE
		  (current_b0->flags & VLIB_BUFFER_NEXT_PRESENT))
		{
		  current_b0 = vlib_get_buffer (vm, current_b0->next_buffer);
		  bytes_left = current_b0->current_length;
		}
	      else
		{
		  /* End of packet */
		  break;
		}
	    }
	}

      /* Move from available to used ring */
      total_desc_len += desc_len;
      rxvq->packed_desc[desc_head].len = total_desc_len;

      vhost_user_advance_last_avail_table_idx (vui, rxvq, chained);
      n_descs_processed++;

      if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
	cpu->current_trace->hdr = cpu->tx_headers[tx_headers_len - 1];

      n_left--;

      /*
       * Do the copy periodically to prevent
       * cpu->copy array overflow and corrupt memory
       */
      if (PREDICT_FALSE (copy_len >= VHOST_USER_TX_COPY_THRESHOLD) || chained)
	{
	  if (PREDICT_FALSE (vhost_user_tx_copy (vui, cpu->copy, copy_len,
						 &map_hint)))
	    vlib_error_count (vm, node->node_index,
			      VHOST_USER_TX_FUNC_ERROR_MMAP_FAIL, 1);
	  copy_len = 0;

	  /* give buffers back to driver */
	  vhost_user_mark_desc_available (vm, vui, rxvq, &n_descs_processed,
					  chained, frame, n_left);
	}

      buffers++;
    }

done:
  if (PREDICT_TRUE (copy_len))
    {
      if (PREDICT_FALSE (vhost_user_tx_copy (vui, cpu->copy, copy_len,
					     &map_hint)))
	vlib_error_count (vm, node->node_index,
			  VHOST_USER_TX_FUNC_ERROR_MMAP_FAIL, 1);

      vhost_user_mark_desc_available (vm, vui, rxvq, &n_descs_processed,
				      chained, frame, n_left);
    }

  /*
   * When n_left is set, error is always set to something too.
   * In case error is due to lack of remaining buffers, we go back up and
   * retry.
   * The idea is that it is better to waste some time on packets
   * that have been processed already than dropping them and get
   * more fresh packets with a good likelyhood that they will be dropped too.
   * This technique also gives more time to VM driver to pick-up packets.
   * In case the traffic flows from physical to virtual interfaces, this
   * technique will end-up leveraging the physical NIC buffer in order to
   * absorb the VM's CPU jitter.
   */
  if (n_left && (error == VHOST_USER_TX_FUNC_ERROR_PKT_DROP_NOBUF) && retry)
    {
      retry--;
      goto retry;
    }

  clib_spinlock_unlock (&rxvq->vring_lock);

  if (PREDICT_FALSE (n_left && error != VHOST_USER_TX_FUNC_ERROR_NONE))
    {
      vlib_error_count (vm, node->node_index, error, n_left);
      vlib_increment_simple_counter
	(vnet_main.interface_main.sw_if_counters +
	 VNET_INTERFACE_COUNTER_DROP, thread_index, vui->sw_if_index, n_left);
    }

  vlib_buffer_free (vm, vlib_frame_vector_args (frame), frame->n_vectors);
  return frame->n_vectors;
}

VNET_DEVICE_CLASS_TX_FN (vhost_user_device_class) (vlib_main_t * vm,
						   vlib_node_runtime_t *
						   node, vlib_frame_t * frame)
{
  u32 *buffers = vlib_frame_vector_args (frame);
  u32 n_left = frame->n_vectors;
  vhost_user_main_t *vum = &vhost_user_main;
  vnet_interface_output_runtime_t *rd = (void *) node->runtime_data;
  vhost_user_intf_t *vui =
    pool_elt_at_index (vum->vhost_user_interfaces, rd->dev_instance);
  u32 qid;
  vhost_user_vring_t *rxvq;
  u8 error;
  u32 thread_index = vm->thread_index;
  vhost_cpu_t *cpu = &vum->cpus[thread_index];
  u32 map_hint = 0;
  u8 retry = 8;
  u16 copy_len;
  u16 tx_headers_len;
  u32 or_flags;
  vnet_hw_if_tx_frame_t *tf = vlib_frame_scalar_args (frame);

  if (PREDICT_FALSE (!vui->admin_up))
    {
      error = VHOST_USER_TX_FUNC_ERROR_DOWN;
      goto done3;
    }

  if (PREDICT_FALSE (!vui->is_ready))
    {
      error = VHOST_USER_TX_FUNC_ERROR_NOT_READY;
      goto done3;
    }

  qid = VHOST_VRING_IDX_RX (tf->queue_id);
  rxvq = &vui->vrings[qid];
  ASSERT (tf->queue_id == rxvq->qid);

  if (PREDICT_FALSE (rxvq->avail == 0))
    {
      error = VHOST_USER_TX_FUNC_ERROR_MMAP_FAIL;
      goto done3;
    }
  if (tf->shared_queue)
    clib_spinlock_lock (&rxvq->vring_lock);

  if (vhost_user_is_packed_ring_supported (vui))
    return (vhost_user_device_class_packed (vm, node, frame, vui, rxvq));

retry:
  error = VHOST_USER_TX_FUNC_ERROR_NONE;
  tx_headers_len = 0;
  copy_len = 0;
  while (n_left > 0)
    {
      vlib_buffer_t *b0, *current_b0;
      u16 desc_head, desc_index, desc_len;
      vnet_virtio_vring_desc_t *desc_table;
      uword buffer_map_addr;
      u32 buffer_len;
      u16 bytes_left;

      if (PREDICT_TRUE (n_left > 1))
	vlib_prefetch_buffer_with_index (vm, buffers[1], LOAD);

      b0 = vlib_get_buffer (vm, buffers[0]);

      if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
	{
	  cpu->current_trace = vlib_add_trace (vm, node, b0,
					       sizeof (*cpu->current_trace));
	  vhost_user_tx_trace (cpu->current_trace, vui, qid / 2, b0, rxvq);
	}

      if (PREDICT_FALSE (rxvq->last_avail_idx == rxvq->avail->idx))
	{
	  error = VHOST_USER_TX_FUNC_ERROR_PKT_DROP_NOBUF;
	  goto done;
	}

      desc_table = rxvq->desc;
      desc_head = desc_index =
	rxvq->avail->ring[rxvq->last_avail_idx & rxvq->qsz_mask];

      /* Go deeper in case of indirect descriptor
       * I don't know of any driver providing indirect for RX. */
      if (PREDICT_FALSE (rxvq->desc[desc_head].flags & VRING_DESC_F_INDIRECT))
	{
	  if (PREDICT_FALSE (rxvq->desc[desc_head].len <
			     sizeof (vnet_virtio_vring_desc_t)))
	    {
	      error = VHOST_USER_TX_FUNC_ERROR_INDIRECT_OVERFLOW;
	      goto done;
	    }
	  if (PREDICT_FALSE
	      (!(desc_table =
		 map_guest_mem (vui, rxvq->desc[desc_index].addr,
				&map_hint))))
	    {
	      error = VHOST_USER_TX_FUNC_ERROR_MMAP_FAIL;
	      goto done;
	    }
	  desc_index = 0;
	}

      desc_len = vui->virtio_net_hdr_sz;
      buffer_map_addr = desc_table[desc_index].addr;
      buffer_len = desc_table[desc_index].len;

      {
	// Get a header from the header array
	vnet_virtio_net_hdr_mrg_rxbuf_t *hdr =
	  &cpu->tx_headers[tx_headers_len];
	tx_headers_len++;
	hdr->hdr.flags = 0;
	hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
	hdr->num_buffers = 1;	//This is local, no need to check

	or_flags = (b0->flags & VNET_BUFFER_F_OFFLOAD);

	/* Guest supports csum offload and buffer requires checksum offload? */
	if (or_flags
	    && (vui->features & VIRTIO_FEATURE (VIRTIO_NET_F_GUEST_CSUM)))
	  vhost_user_handle_tx_offload (vui, b0, &hdr->hdr);

	// Prepare a copy order executed later for the header
	ASSERT (copy_len < VHOST_USER_COPY_ARRAY_N);
	vhost_copy_t *cpy = &cpu->copy[copy_len];
	copy_len++;
	cpy->len = vui->virtio_net_hdr_sz;
	cpy->dst = buffer_map_addr;
	cpy->src = (uword) hdr;
      }

      buffer_map_addr += vui->virtio_net_hdr_sz;
      buffer_len -= vui->virtio_net_hdr_sz;
      bytes_left = b0->current_length;
      current_b0 = b0;
      while (1)
	{
	  if (buffer_len == 0)
	    {			//Get new output
	      if (desc_table[desc_index].flags & VRING_DESC_F_NEXT)
		{
		  //Next one is chained
		  desc_index = desc_table[desc_index].next;
		  buffer_map_addr = desc_table[desc_index].addr;
		  buffer_len = desc_table[desc_index].len;
		}
	      else if (vui->virtio_net_hdr_sz == 12)	//MRG is available
		{
		  vnet_virtio_net_hdr_mrg_rxbuf_t *hdr =
		    &cpu->tx_headers[tx_headers_len - 1];

		  //Move from available to used buffer
		  rxvq->used->ring[rxvq->last_used_idx & rxvq->qsz_mask].id =
		    desc_head;
		  rxvq->used->ring[rxvq->last_used_idx & rxvq->qsz_mask].len =
		    desc_len;
		  vhost_user_log_dirty_ring (vui, rxvq,
					     ring[rxvq->last_used_idx &
						  rxvq->qsz_mask]);

		  rxvq->last_avail_idx++;
		  rxvq->last_used_idx++;
		  hdr->num_buffers++;
		  desc_len = 0;

		  if (PREDICT_FALSE
		      (rxvq->last_avail_idx == rxvq->avail->idx))
		    {
		      //Dequeue queued descriptors for this packet
		      rxvq->last_used_idx -= hdr->num_buffers - 1;
		      rxvq->last_avail_idx -= hdr->num_buffers - 1;
		      error = VHOST_USER_TX_FUNC_ERROR_PKT_DROP_NOBUF;
		      goto done;
		    }

		  desc_table = rxvq->desc;
		  desc_head = desc_index =
		    rxvq->avail->ring[rxvq->last_avail_idx & rxvq->qsz_mask];
		  if (PREDICT_FALSE
		      (rxvq->desc[desc_head].flags & VRING_DESC_F_INDIRECT))
		    {
		      //It is seriously unlikely that a driver will put indirect descriptor
		      //after non-indirect descriptor.
		      if (PREDICT_FALSE (rxvq->desc[desc_head].len <
					 sizeof (vnet_virtio_vring_desc_t)))
			{
			  error = VHOST_USER_TX_FUNC_ERROR_INDIRECT_OVERFLOW;
			  goto done;
			}
		      if (PREDICT_FALSE
			  (!(desc_table =
			     map_guest_mem (vui,
					    rxvq->desc[desc_index].addr,
					    &map_hint))))
			{
			  error = VHOST_USER_TX_FUNC_ERROR_MMAP_FAIL;
			  goto done;
			}
		      desc_index = 0;
		    }
		  buffer_map_addr = desc_table[desc_index].addr;
		  buffer_len = desc_table[desc_index].len;
		}
	      else
		{
		  error = VHOST_USER_TX_FUNC_ERROR_PKT_DROP_NOMRG;
		  goto done;
		}
	    }

	  {
	    ASSERT (copy_len < VHOST_USER_COPY_ARRAY_N);
	    vhost_copy_t *cpy = &cpu->copy[copy_len];
	    copy_len++;
	    cpy->len = bytes_left;
	    cpy->len = (cpy->len > buffer_len) ? buffer_len : cpy->len;
	    cpy->dst = buffer_map_addr;
	    cpy->src = (uword) vlib_buffer_get_current (current_b0) +
	      current_b0->current_length - bytes_left;

	    bytes_left -= cpy->len;
	    buffer_len -= cpy->len;
	    buffer_map_addr += cpy->len;
	    desc_len += cpy->len;

	    clib_prefetch_load (&rxvq->desc);
	  }

	  // Check if vlib buffer has more data. If not, get more or break.
	  if (PREDICT_TRUE (!bytes_left))
	    {
	      if (PREDICT_FALSE
		  (current_b0->flags & VLIB_BUFFER_NEXT_PRESENT))
		{
		  current_b0 = vlib_get_buffer (vm, current_b0->next_buffer);
		  bytes_left = current_b0->current_length;
		}
	      else
		{
		  //End of packet
		  break;
		}
	    }
	}

      //Move from available to used ring
      rxvq->used->ring[rxvq->last_used_idx & rxvq->qsz_mask].id = desc_head;
      rxvq->used->ring[rxvq->last_used_idx & rxvq->qsz_mask].len = desc_len;
      vhost_user_log_dirty_ring (vui, rxvq,
				 ring[rxvq->last_used_idx & rxvq->qsz_mask]);
      rxvq->last_avail_idx++;
      rxvq->last_used_idx++;

      if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
	{
	  cpu->current_trace->hdr = cpu->tx_headers[tx_headers_len - 1];
	}

      n_left--;			//At the end for error counting when 'goto done' is invoked

      /*
       * Do the copy periodically to prevent
       * cpu->copy array overflow and corrupt memory
       */
      if (PREDICT_FALSE (copy_len >= VHOST_USER_TX_COPY_THRESHOLD))
	{
	  if (PREDICT_FALSE (vhost_user_tx_copy (vui, cpu->copy, copy_len,
						 &map_hint)))
	    {
	      vlib_error_count (vm, node->node_index,
				VHOST_USER_TX_FUNC_ERROR_MMAP_FAIL, 1);
	    }
	  copy_len = 0;

	  /* give buffers back to driver */
	  CLIB_MEMORY_BARRIER ();
	  rxvq->used->idx = rxvq->last_used_idx;
	  vhost_user_log_dirty_ring (vui, rxvq, idx);
	}
      buffers++;
    }

done:
  //Do the memory copies
  if (PREDICT_FALSE (vhost_user_tx_copy (vui, cpu->copy, copy_len,
					 &map_hint)))
    {
      vlib_error_count (vm, node->node_index,
			VHOST_USER_TX_FUNC_ERROR_MMAP_FAIL, 1);
    }

  CLIB_MEMORY_BARRIER ();
  rxvq->used->idx = rxvq->last_used_idx;
  vhost_user_log_dirty_ring (vui, rxvq, idx);

  /*
   * When n_left is set, error is always set to something too.
   * In case error is due to lack of remaining buffers, we go back up and
   * retry.
   * The idea is that it is better to waste some time on packets
   * that have been processed already than dropping them and get
   * more fresh packets with a good likelihood that they will be dropped too.
   * This technique also gives more time to VM driver to pick-up packets.
   * In case the traffic flows from physical to virtual interfaces, this
   * technique will end-up leveraging the physical NIC buffer in order to
   * absorb the VM's CPU jitter.
   */
  if (n_left && (error == VHOST_USER_TX_FUNC_ERROR_PKT_DROP_NOBUF) && retry)
    {
      retry--;
      goto retry;
    }

  /* interrupt (call) handling */
  if ((rxvq->callfd_idx != ~0) &&
      !(rxvq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT))
    {
      rxvq->n_since_last_int += frame->n_vectors - n_left;

      if (rxvq->n_since_last_int > vum->coalesce_frames)
	vhost_user_send_call (vm, vui, rxvq);
    }

  clib_spinlock_unlock (&rxvq->vring_lock);

done3:
  if (PREDICT_FALSE (n_left && error != VHOST_USER_TX_FUNC_ERROR_NONE))
    {
      vlib_error_count (vm, node->node_index, error, n_left);
      vlib_increment_simple_counter
	(vnet_main.interface_main.sw_if_counters
	 + VNET_INTERFACE_COUNTER_DROP,
	 thread_index, vui->sw_if_index, n_left);
    }

  vlib_buffer_free (vm, vlib_frame_vector_args (frame), frame->n_vectors);
  return frame->n_vectors;
}

static __clib_unused clib_error_t *
vhost_user_interface_rx_mode_change (vnet_main_t * vnm, u32 hw_if_index,
				     u32 qid, vnet_hw_if_rx_mode mode)
{
  vlib_main_t *vm = vnm->vlib_main;
  vnet_hw_interface_t *hif = vnet_get_hw_interface (vnm, hw_if_index);
  vhost_user_main_t *vum = &vhost_user_main;
  vhost_user_intf_t *vui =
    pool_elt_at_index (vum->vhost_user_interfaces, hif->dev_instance);
  vhost_user_vring_t *txvq = &vui->vrings[VHOST_VRING_IDX_TX (qid)];
  vhost_cpu_t *cpu;

  if (mode == txvq->mode)
    return 0;

  if ((mode != VNET_HW_IF_RX_MODE_POLLING) &&
      (mode != VNET_HW_IF_RX_MODE_ADAPTIVE) &&
      (mode != VNET_HW_IF_RX_MODE_INTERRUPT))
    {
      vu_log_err (vui, "unhandled mode %d changed for if %d queue %d", mode,
		  hw_if_index, qid);
      return clib_error_return (0, "unsupported");
    }

  if (txvq->thread_index == ~0)
    return clib_error_return (0, "Queue initialization is not finished yet");

  cpu = vec_elt_at_index (vum->cpus, txvq->thread_index);
  if ((mode == VNET_HW_IF_RX_MODE_INTERRUPT) ||
      (mode == VNET_HW_IF_RX_MODE_ADAPTIVE))
    {
      if (txvq->kickfd_idx == ~0)
	{
	  // We cannot support interrupt mode if the driver opts out
	  return clib_error_return (0, "Driver does not support interrupt");
	}
      if (txvq->mode == VNET_HW_IF_RX_MODE_POLLING)
	{
	  ASSERT (cpu->polling_q_count != 0);
	  if (cpu->polling_q_count)
	    cpu->polling_q_count--;
	  vum->ifq_count++;
	  // Start the timer if this is the first encounter on interrupt
	  // interface/queue
	  if ((vum->ifq_count == 1) &&
	      ((vum->coalesce_time > 0.0) || (vum->coalesce_frames > 0)))
	    vlib_process_signal_event (vm,
				       vhost_user_send_interrupt_node.index,
				       VHOST_USER_EVENT_START_TIMER, 0);
	}
    }
  else if (mode == VNET_HW_IF_RX_MODE_POLLING)
    {
      if (((txvq->mode == VNET_HW_IF_RX_MODE_INTERRUPT) ||
	   (txvq->mode == VNET_HW_IF_RX_MODE_ADAPTIVE)) && vum->ifq_count)
	{
	  cpu->polling_q_count++;
	  vum->ifq_count--;
	  // Stop the timer if there is no more interrupt interface/queue
	  if (vum->ifq_count == 0)
	    vlib_process_signal_event (vm,
				       vhost_user_send_interrupt_node.index,
				       VHOST_USER_EVENT_STOP_TIMER, 0);
	}
    }

  txvq->mode = mode;
  vhost_user_set_operation_mode (vui, txvq);

  return 0;
}

static __clib_unused clib_error_t *
vhost_user_interface_admin_up_down (vnet_main_t * vnm, u32 hw_if_index,
				    u32 flags)
{
  vnet_hw_interface_t *hif = vnet_get_hw_interface (vnm, hw_if_index);
  vhost_user_main_t *vum = &vhost_user_main;
  vhost_user_intf_t *vui =
    pool_elt_at_index (vum->vhost_user_interfaces, hif->dev_instance);
  u8 link_old, link_new;

  link_old = vui_is_link_up (vui);

  vui->admin_up = (flags & VNET_SW_INTERFACE_FLAG_ADMIN_UP) != 0;

  link_new = vui_is_link_up (vui);

  if (link_old != link_new)
    vnet_hw_interface_set_flags (vnm, vui->hw_if_index, link_new ?
				 VNET_HW_INTERFACE_FLAG_LINK_UP : 0);

  return /* no error */ 0;
}

/* *INDENT-OFF* */
VNET_DEVICE_CLASS (vhost_user_device_class) = {
  .name = "vhost-user",
  .tx_function_n_errors = VHOST_USER_TX_FUNC_N_ERROR,
  .tx_function_error_strings = vhost_user_tx_func_error_strings,
  .format_device_name = format_vhost_user_interface_name,
  .name_renumber = vhost_user_name_renumber,
  .admin_up_down_function = vhost_user_interface_admin_up_down,
  .rx_mode_change_function = vhost_user_interface_rx_mode_change,
  .format_tx_trace = format_vhost_trace,
};

/* *INDENT-ON* */

/*
 * fd.io coding-style-patch-verification: ON
 *
 * Local Variables:
 * eval: (c-set-style "gnu")
 * End:
 */