aboutsummaryrefslogtreecommitdiffstats
path: root/tests/vpp/perf/__init__.robot
diff options
context:
space:
mode:
authorPeter Mikus <pmikus@cisco.com>2017-09-22 14:59:40 +0200
committerPeter Mikus <pmikus@cisco.com>2017-10-12 08:39:16 +0000
commit9a261ea61549fc6a5c23369d2e236b002dc35038 (patch)
tree8004f366007d3ee0210bbf099b71491bb412fe88 /tests/vpp/perf/__init__.robot
parentf2573eccd38609fbc3d44f1fb9c706d08e50d49c (diff)
CSIT-748 vnf-agent integration
CSIT-773 Implement RF keywords for k8s - Implementation of Test Suite Setup for Ligato vnf-agent testing - Implementation of KubernetesUtil for controlling kubectl - Yaml templates for L2XC topology with 1cswitch and 1cnf - Yaml templates for L2BD topology with 1cswitch and 1cnf - ligato bootstrap script for creating vnf-agent image Change-Id: Iebefde0eb984a27a0afcdf29fe549ca4edf8a61e Signed-off-by: Peter Mikus <pmikus@cisco.com>
Diffstat (limited to 'tests/vpp/perf/__init__.robot')
-rw-r--r--tests/vpp/perf/__init__.robot1
1 files changed, 1 insertions, 0 deletions
diff --git a/tests/vpp/perf/__init__.robot b/tests/vpp/perf/__init__.robot
index 61dd83aaaf..18cb722e0f 100644
--- a/tests/vpp/perf/__init__.robot
+++ b/tests/vpp/perf/__init__.robot
@@ -19,6 +19,7 @@
| Suite Setup | Run Keywords | Setup performance global Variables
| ... | AND | Setup Framework | ${nodes}
| ... | AND | Setup All DUTs | ${nodes}
+| ... | AND | Show Vpp Version On All Duts | ${nodes}
| ... | AND | Get CPU Layout from all nodes | ${nodes}
| ... | AND | Update All Interface Data On All Nodes
| ... | ${nodes} | skip_tg=${True} | numa_node=${True}
/a> 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
/*
 *------------------------------------------------------------------
 * Copyright (c) 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 <unistd.h>
#include <fcntl.h>
#include <net/if.h>
#include <linux/if_link.h>
#include <linux/if_ether.h>

#include <vppinfra/linux/sysfs.h>
#include <vlib/vlib.h>
#include <vlib/unix/unix.h>
#include <vlib/pci/pci.h>
#include <vnet/ethernet/ethernet.h>

#include <rdma/rdma.h>

/* Default RSS hash key (from DPDK MLX driver) */
static u8 rdma_rss_hash_key[] = {
  0x2c, 0xc6, 0x81, 0xd1,
  0x5b, 0xdb, 0xf4, 0xf7,
  0xfc, 0xa2, 0x83, 0x19,
  0xdb, 0x1a, 0x3e, 0x94,
  0x6b, 0x9e, 0x38, 0xd9,
  0x2c, 0x9c, 0x03, 0xd1,
  0xad, 0x99, 0x44, 0xa7,
  0xd9, 0x56, 0x3d, 0x59,
  0x06, 0x3c, 0x25, 0xf3,
  0xfc, 0x1f, 0xdc, 0x2a,
};

rdma_main_t rdma_main;

#define rdma_log__(lvl, dev, f, ...) \
  do { \
      vlib_log((lvl), rdma_main.log_class, "%s: " f, \
               &(dev)->name, ##__VA_ARGS__); \
  } while (0)

#define rdma_log(lvl, dev, f, ...) \
   rdma_log__((lvl), (dev), "%s (%d): " f, strerror(errno), errno, ##__VA_ARGS__)

static struct ibv_flow *
rdma_rxq_init_flow (const rdma_device_t * rd, struct ibv_qp *qp,
		    const mac_address_t * mac, const mac_address_t * mask,
		    u32 flags)
{
  struct ibv_flow *flow;
  struct raw_eth_flow_attr
  {
    struct ibv_flow_attr attr;
    struct ibv_flow_spec_eth spec_eth;
  } __attribute__ ((packed)) fa;

  memset (&fa, 0, sizeof (fa));
  fa.attr.num_of_specs = 1;
  fa.attr.port = 1;
  fa.attr.flags = flags;
  fa.spec_eth.type = IBV_FLOW_SPEC_ETH;
  fa.spec_eth.size = sizeof (struct ibv_flow_spec_eth);

  memcpy (fa.spec_eth.val.dst_mac, mac, sizeof (fa.spec_eth.val.dst_mac));
  memcpy (fa.spec_eth.mask.dst_mac, mask, sizeof (fa.spec_eth.mask.dst_mac));

  flow = ibv_create_flow (qp, &fa.attr);
  if (!flow)
    rdma_log (VLIB_LOG_LEVEL_ERR, rd, "ibv_create_flow() failed");
  return flow;
}

static u32
rdma_rxq_destroy_flow (const rdma_device_t * rd, struct ibv_flow **flow)
{
  if (!*flow)
    return 0;

  if (ibv_destroy_flow (*flow))
    {
      rdma_log (VLIB_LOG_LEVEL_ERR, rd, "ibv_destroy_flow() failed");
      return ~0;
    }

  *flow = 0;
  return 0;
}

static u32
rdma_dev_set_promisc (rdma_device_t * rd)
{
  const mac_address_t all = {.bytes = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0} };
  int err;

  err = rdma_rxq_destroy_flow (rd, &rd->flow_mcast);
  if (err)
    return ~0;

  err = rdma_rxq_destroy_flow (rd, &rd->flow_ucast);
  if (err)
    return ~0;

  rd->flow_ucast = rdma_rxq_init_flow (rd, rd->rx_qp, &all, &all, 0);
  if (!rd->flow_ucast)
    return ~0;

  rd->flags |= RDMA_DEVICE_F_PROMISC;
  return 0;
}

static u32
rdma_dev_set_ucast (rdma_device_t * rd)
{
  const mac_address_t ucast = {.bytes = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff}
  };
  const mac_address_t mcast = {.bytes = {0x1, 0x0, 0x0, 0x0, 0x0, 0x0} };
  int err;

  err = rdma_rxq_destroy_flow (rd, &rd->flow_mcast);
  if (err)
    return ~0;

  err = rdma_rxq_destroy_flow (rd, &rd->flow_ucast);
  if (err)
    return ~0;

  /* receive only packets with src = our MAC */
  rd->flow_ucast = rdma_rxq_init_flow (rd, rd->rx_qp, &rd->hwaddr, &ucast, 0);
  if (!rd->flow_ucast)
    return ~0;

  /* receive multicast packets */
  rd->flow_mcast = rdma_rxq_init_flow (rd, rd->rx_qp, &mcast, &mcast,
				       IBV_FLOW_ATTR_FLAGS_DONT_TRAP
				       /* let others receive mcast packet too (eg. Linux) */
    );
  if (!rd->flow_mcast)
    return ~0;

  rd->flags &= ~RDMA_DEVICE_F_PROMISC;
  return 0;
}

static clib_error_t *
rdma_mac_change (vnet_hw_interface_t * hw, const u8 * old, const u8 * new)
{
  rdma_main_t *rm = &rdma_main;
  rdma_device_t *rd = vec_elt_at_index (rm->devices, hw->dev_instance);
  mac_address_from_bytes (&rd->hwaddr, new);
  if (!(rd->flags & RDMA_DEVICE_F_PROMISC) && rdma_dev_set_ucast (rd))
    {
      mac_address_from_bytes (&rd->hwaddr, old);
      return clib_error_return_unix (0, "MAC update failed");
    }
  return 0;
}

static u32
rdma_dev_change_mtu (rdma_device_t * rd)
{
  rdma_log__ (VLIB_LOG_LEVEL_ERR, rd, "MTU change not supported");
  return ~0;
}

static u32
rdma_flag_change (vnet_main_t * vnm, vnet_hw_interface_t * hw, u32 flags)
{
  rdma_main_t *rm = &rdma_main;
  rdma_device_t *rd = vec_elt_at_index (rm->devices, hw->dev_instance);

  switch (flags)
    {
    case 0:
      return rdma_dev_set_ucast (rd);
    case ETHERNET_INTERFACE_FLAG_ACCEPT_ALL:
      return rdma_dev_set_promisc (rd);
    case ETHERNET_INTERFACE_FLAG_MTU:
      return rdma_dev_change_mtu (rd);
    }

  rdma_log__ (VLIB_LOG_LEVEL_ERR, rd, "unknown flag %x requested", flags);
  return ~0;
}

static void
rdma_update_state (vnet_main_t * vnm, rdma_device_t * rd, int port)
{
  struct ibv_port_attr attr;
  u32 width = 0;
  u32 speed = 0;

  if (ibv_query_port (rd->ctx, port, &attr))
    {
      vnet_hw_interface_set_link_speed (vnm, rd->hw_if_index, 0);
      vnet_hw_interface_set_flags (vnm, rd->hw_if_index, 0);
      return;
    }

  /* update state */
  switch (attr.state)
    {
    case IBV_PORT_ACTIVE:	/* fallthrough */
    case IBV_PORT_ACTIVE_DEFER:
      rd->flags |= RDMA_DEVICE_F_LINK_UP;
      vnet_hw_interface_set_flags (vnm, rd->hw_if_index,
				   VNET_HW_INTERFACE_FLAG_LINK_UP);
      break;
    default:
      rd->flags &= ~RDMA_DEVICE_F_LINK_UP;
      vnet_hw_interface_set_flags (vnm, rd->hw_if_index, 0);
      break;
    }

  /* update speed */
  switch (attr.active_width)
    {
    case 1:
      width = 1;
      break;
    case 2:
      width = 4;
      break;
    case 4:
      width = 8;
      break;
    case 8:
      width = 12;
      break;
    }
  switch (attr.active_speed)
    {
    case 1:
      speed = 2500000;
      break;
    case 2:
      speed = 5000000;
      break;
    case 4:			/* fallthrough */
    case 8:
      speed = 10000000;
      break;
    case 16:
      speed = 14000000;
      break;
    case 32:
      speed = 25000000;
      break;
    }
  vnet_hw_interface_set_link_speed (vnm, rd->hw_if_index, width * speed);
}

static clib_error_t *
rdma_async_event_error_ready (clib_file_t * f)
{
  rdma_main_t *rm = &rdma_main;
  rdma_device_t *rd = vec_elt_at_index (rm->devices, f->private_data);
  return clib_error_return (0, "RDMA: %s: async event error", rd->name);
}

static clib_error_t *
rdma_async_event_read_ready (clib_file_t * f)
{
  vnet_main_t *vnm = vnet_get_main ();
  rdma_main_t *rm = &rdma_main;
  rdma_device_t *rd = vec_elt_at_index (rm->devices, f->private_data);
  int ret;
  struct ibv_async_event event;
  ret = ibv_get_async_event (rd->ctx, &event);
  if (ret < 0)
    return clib_error_return_unix (0, "ibv_get_async_event() failed");

  switch (event.event_type)
    {
    case IBV_EVENT_PORT_ACTIVE:
      rdma_update_state (vnm, rd, event.element.port_num);
      break;
    case IBV_EVENT_PORT_ERR:
      rdma_update_state (vnm, rd, event.element.port_num);
      break;
    case IBV_EVENT_DEVICE_FATAL:
      rd->flags &= ~RDMA_DEVICE_F_LINK_UP;
      vnet_hw_interface_set_flags (vnm, rd->hw_if_index, 0);
      vlib_log_emerg (rm->log_class, "%s: fatal error", rd->name);
      break;
    default:
      rdma_log__ (VLIB_LOG_LEVEL_ERR, rd, "unhandeld RDMA async event %i",
		  event.event_type);
      break;
    }

  ibv_ack_async_event (&event);
  return 0;
}

static clib_error_t *
rdma_async_event_init (rdma_device_t * rd)
{
  clib_file_t t = { 0 };
  int ret;

  /* make RDMA async event fd non-blocking */
  ret = fcntl (rd->ctx->async_fd, F_GETFL);
  if (ret < 0)
    return clib_error_return_unix (0, "fcntl(F_GETFL) failed");

  ret = fcntl (rd->ctx->async_fd, F_SETFL, ret | O_NONBLOCK);
  if (ret < 0)
    return clib_error_return_unix (0, "fcntl(F_SETFL, O_NONBLOCK) failed");

  /* register RDMA async event fd */
  t.read_function = rdma_async_event_read_ready;
  t.file_descriptor = rd->ctx->async_fd;
  t.error_function = rdma_async_event_error_ready;
  t.private_data = rd->dev_instance;
  t.description = format (0, "%v async event", rd->name);

  rd->async_event_clib_file_index = clib_file_add (&file_main, &t);
  return 0;
}

static void
rdma_async_event_cleanup (rdma_device_t * rd)
{
  clib_file_del_by_index (&file_main, rd->async_event_clib_file_index);
}

static clib_error_t *
rdma_register_interface (vnet_main_t * vnm, rdma_device_t * rd)
{
  return ethernet_register_interface (vnm, rdma_device_class.index,
				      rd->dev_instance, rd->hwaddr.bytes,
				      &rd->hw_if_index, rdma_flag_change);
}

static void
rdma_unregister_interface (vnet_main_t * vnm, rdma_device_t * rd)
{
  vnet_hw_interface_set_flags (vnm, rd->hw_if_index, 0);
  vnet_hw_interface_unassign_rx_thread (vnm, rd->hw_if_index, 0);
  ethernet_delete_interface (vnm, rd->hw_if_index);
}

static void
rdma_dev_cleanup (rdma_device_t * rd)
{
  rdma_main_t *rm = &rdma_main;
  rdma_rxq_t *rxq;
  rdma_txq_t *txq;

#define _(fn, arg) if (arg) \
  { \
    int rv; \
    if ((rv = fn (arg))) \
       rdma_log (VLIB_LOG_LEVEL_DEBUG, rd, #fn "() failed (rv = %d)", rv); \
  }

  _(ibv_destroy_flow, rd->flow_mcast);
  _(ibv_destroy_flow, rd->flow_ucast);
  _(ibv_dereg_mr, rd->mr);
  vec_foreach (txq, rd->txqs)
  {
    _(ibv_destroy_qp, txq->qp);
    _(ibv_destroy_cq, txq->cq);
  }
  vec_foreach (rxq, rd->rxqs)
  {
    _(ibv_destroy_wq, rxq->wq);
    _(ibv_destroy_cq, rxq->cq);
  }
  _(ibv_destroy_rwq_ind_table, rd->rx_rwq_ind_tbl);
  _(ibv_destroy_qp, rd->rx_qp);
  _(ibv_dealloc_pd, rd->pd);
  _(ibv_close_device, rd->ctx);
#undef _

  clib_error_free (rd->error);

  vec_free (rd->rxqs);
  vec_free (rd->txqs);
  vec_free (rd->name);
  vlib_pci_free_device_info (rd->pci);
  pool_put (rm->devices, rd);
}

static clib_error_t *
rdma_rxq_init (vlib_main_t * vm, rdma_device_t * rd, u16 qid, u32 n_desc)
{
  rdma_rxq_t *rxq;
  struct ibv_wq_init_attr wqia;
  struct ibv_wq_attr wqa;

  vec_validate_aligned (rd->rxqs, qid, CLIB_CACHE_LINE_BYTES);
  rxq = vec_elt_at_index (rd->rxqs, qid);
  rxq->size = n_desc;
  vec_validate_aligned (rxq->bufs, n_desc - 1, CLIB_CACHE_LINE_BYTES);

  if ((rxq->cq = ibv_create_cq (rd->ctx, n_desc, NULL, NULL, 0)) == 0)
    return clib_error_return_unix (0, "Create CQ Failed");

  memset (&wqia, 0, sizeof (wqia));
  wqia.wq_type = IBV_WQT_RQ;
  wqia.max_wr = n_desc;
  wqia.max_sge = 1;
  wqia.pd = rd->pd;
  wqia.cq = rxq->cq;
  if ((rxq->wq = ibv_create_wq (rd->ctx, &wqia)) == 0)
    return clib_error_return_unix (0, "Create WQ Failed");

  memset (&wqa, 0, sizeof (wqa));
  wqa.attr_mask = IBV_WQ_ATTR_STATE;
  wqa.wq_state = IBV_WQS_RDY;
  if (ibv_modify_wq (rxq->wq, &wqa) != 0)
    return clib_error_return_unix (0, "Modify WQ (RDY) Failed");

  return 0;
}

static clib_error_t *
rdma_rxq_finalize (vlib_main_t * vm, rdma_device_t * rd)
{
  struct ibv_rwq_ind_table_init_attr rwqia;
  struct ibv_qp_init_attr_ex qpia;
  struct ibv_wq **ind_tbl;
  u32 i;

  ASSERT (is_pow2 (vec_len (rd->rxqs))
	  && "rxq number should be a power of 2");

  ind_tbl = vec_new (struct ibv_wq *, vec_len (rd->rxqs));
  vec_foreach_index (i, rd->rxqs)
    ind_tbl[i] = vec_elt_at_index (rd->rxqs, i)->wq;
  memset (&rwqia, 0, sizeof (rwqia));
  rwqia.log_ind_tbl_size = min_log2 (vec_len (ind_tbl));
  rwqia.ind_tbl = ind_tbl;
  if ((rd->rx_rwq_ind_tbl = ibv_create_rwq_ind_table (rd->ctx, &rwqia)) == 0)
    return clib_error_return_unix (0, "RWQ indirection table create failed");
  vec_free (ind_tbl);

  memset (&qpia, 0, sizeof (qpia));
  qpia.qp_type = IBV_QPT_RAW_PACKET;
  qpia.comp_mask =
    IBV_QP_INIT_ATTR_PD | IBV_QP_INIT_ATTR_IND_TABLE |
    IBV_QP_INIT_ATTR_RX_HASH;
  qpia.pd = rd->pd;
  qpia.rwq_ind_tbl = rd->rx_rwq_ind_tbl;
  STATIC_ASSERT_SIZEOF (rdma_rss_hash_key, 40);
  qpia.rx_hash_conf.rx_hash_key_len = sizeof (rdma_rss_hash_key);
  qpia.rx_hash_conf.rx_hash_key = rdma_rss_hash_key;
  qpia.rx_hash_conf.rx_hash_function = IBV_RX_HASH_FUNC_TOEPLITZ;
  qpia.rx_hash_conf.rx_hash_fields_mask =
    IBV_RX_HASH_SRC_IPV4 | IBV_RX_HASH_DST_IPV4;
  if ((rd->rx_qp = ibv_create_qp_ex (rd->ctx, &qpia)) == 0)
    return clib_error_return_unix (0, "Queue Pair create failed");

  if (rdma_dev_set_ucast (rd))
    return clib_error_return_unix (0, "Set unicast mode failed");

  return 0;
}

static clib_error_t *
rdma_txq_init (vlib_main_t * vm, rdma_device_t * rd, u16 qid, u32 n_desc)
{
  rdma_txq_t *txq;
  struct ibv_qp_init_attr qpia;
  struct ibv_qp_attr qpa;
  int qp_flags;

  vec_validate_aligned (rd->txqs, qid, CLIB_CACHE_LINE_BYTES);
  txq = vec_elt_at_index (rd->txqs, qid);
  txq->size = n_desc;
  vec_validate_aligned (txq->bufs, n_desc - 1, CLIB_CACHE_LINE_BYTES);

  if ((txq->cq = ibv_create_cq (rd->ctx, n_desc, NULL, NULL, 0)) == 0)
    return clib_error_return_unix (0, "Create CQ Failed");

  memset (&qpia, 0, sizeof (qpia));
  qpia.send_cq = txq->cq;
  qpia.recv_cq = txq->cq;
  qpia.cap.max_send_wr = n_desc;
  qpia.cap.max_send_sge = 1;
  qpia.qp_type = IBV_QPT_RAW_PACKET;

  if ((txq->qp = ibv_create_qp (rd->pd, &qpia)) == 0)
    return clib_error_return_unix (0, "Queue Pair create failed");

  memset (&qpa, 0, sizeof (qpa));
  qp_flags = IBV_QP_STATE | IBV_QP_PORT;
  qpa.qp_state = IBV_QPS_INIT;
  qpa.port_num = 1;
  if (ibv_modify_qp (txq->qp, &qpa, qp_flags) != 0)
    return clib_error_return_unix (0, "Modify QP (init) Failed");

  memset (&qpa, 0, sizeof (qpa));
  qp_flags = IBV_QP_STATE;
  qpa.qp_state = IBV_QPS_RTR;
  if (ibv_modify_qp (txq->qp, &qpa, qp_flags) != 0)
    return clib_error_return_unix (0, "Modify QP (receive) Failed");

  memset (&qpa, 0, sizeof (qpa));
  qp_flags = IBV_QP_STATE;
  qpa.qp_state = IBV_QPS_RTS;
  if (ibv_modify_qp (txq->qp, &qpa, qp_flags) != 0)
    return clib_error_return_unix (0, "Modify QP (send) Failed");
  return 0;
}

static clib_error_t *
rdma_dev_init (vlib_main_t * vm, rdma_device_t * rd, u32 rxq_size,
	       u32 txq_size, u32 rxq_num)
{
  clib_error_t *err;
  vlib_buffer_main_t *bm = vm->buffer_main;
  vlib_thread_main_t *tm = vlib_get_thread_main ();
  u32 i;

  if (rd->ctx == 0)
    return clib_error_return_unix (0, "Device Open Failed");

  if ((rd->pd = ibv_alloc_pd (rd->ctx)) == 0)
    return clib_error_return_unix (0, "PD Alloc Failed");

  ethernet_mac_address_generate (rd->hwaddr.bytes);

  /*
   * /!\ WARNING /!\ creation order is important
   * We *must* create TX queues *before* RX queues, otherwise we will receive
   * the broacast packets we sent
   */
  for (i = 0; i < tm->n_vlib_mains; i++)
    if ((err = rdma_txq_init (vm, rd, i, txq_size)))
      return err;

  for (i = 0; i < rxq_num; i++)
    if ((err = rdma_rxq_init (vm, rd, i, rxq_size)))
      return err;
  if ((err = rdma_rxq_finalize (vm, rd)))
    return err;

  if ((rd->mr = ibv_reg_mr (rd->pd, (void *) bm->buffer_mem_start,
			    bm->buffer_mem_size,
			    IBV_ACCESS_LOCAL_WRITE)) == 0)
    return clib_error_return_unix (0, "Register MR Failed");
  rd->lkey = rd->mr->lkey;	/* avoid indirection in datapath */

  return 0;
}

static uword
sysfs_path_to_pci_addr (char *path, vlib_pci_addr_t * addr)
{
  uword rv;
  unformat_input_t in;
  u8 *s;

  s = clib_sysfs_link_to_name (path);
  if (!s)
    return 0;

  unformat_init_string (&in, (char *) s, strlen ((char *) s));
  rv = unformat (&in, "%U", unformat_vlib_pci_addr, addr);
  unformat_free (&in);
  vec_free (s);
  return rv;
}

void
rdma_create_if (vlib_main_t * vm, rdma_create_if_args_t * args)
{
  vnet_main_t *vnm = vnet_get_main ();
  rdma_main_t *rm = &rdma_main;
  rdma_device_t *rd;
  vlib_pci_addr_t pci_addr;
  struct ibv_device **dev_list;
  int n_devs;
  u8 *s;
  u16 qid;
  int i;

  args->rxq_size = args->rxq_size ? args->rxq_size : 2 * VLIB_FRAME_SIZE;
  args->txq_size = args->txq_size ? args->txq_size : 2 * VLIB_FRAME_SIZE;
  args->rxq_num = args->rxq_num ? args->rxq_num : 1;

  if (!is_pow2 (args->rxq_num))
    {
      args->rv = VNET_API_ERROR_INVALID_VALUE;
      args->error =
	clib_error_return (0, "rx queue number must be a power of two");
      goto err0;
    }

  if (args->rxq_size < VLIB_FRAME_SIZE || args->txq_size < VLIB_FRAME_SIZE ||
      !is_pow2 (args->rxq_size) || !is_pow2 (args->txq_size))
    {
      args->rv = VNET_API_ERROR_INVALID_VALUE;
      args->error =
	clib_error_return (0, "queue size must be a power of two >= %i",
			   VLIB_FRAME_SIZE);
      goto err0;
    }

  dev_list = ibv_get_device_list (&n_devs);
  if (n_devs == 0)
    {
      args->error =
	clib_error_return_unix (0,
				"no RDMA devices available. Is the ib_uverbs module loaded?");
      goto err0;
    }

  /* get PCI address */
  s = format (0, "/sys/class/net/%s/device%c", args->ifname, 0);
  if (sysfs_path_to_pci_addr ((char *) s, &pci_addr) == 0)
    {
      args->error =
	clib_error_return (0, "cannot find PCI address for device ");
      goto err1;
    }

  pool_get_zero (rm->devices, rd);
  rd->dev_instance = rd - rm->devices;
  rd->per_interface_next_index = VNET_DEVICE_INPUT_NEXT_ETHERNET_INPUT;
  rd->linux_ifname = format (0, "%s", args->ifname);

  if (!args->name || 0 == args->name[0])
    rd->name = format (0, "%s/%d", args->ifname, rd->dev_instance);
  else
    rd->name = format (0, "%s", args->name);

  rd->pci = vlib_pci_get_device_info (vm, &pci_addr, &args->error);
  if (!rd->pci)
    goto err2;

  /* if we failed to parse NUMA node, default to 0 */
  if (-1 == rd->pci->numa_node)
    rd->pci->numa_node = 0;

  rd->pool = vlib_buffer_pool_get_default_for_numa (vm, rd->pci->numa_node);

  if (strncmp ((char *) rd->pci->driver_name, "mlx5_core", 9))
    {
      args->error =
	clib_error_return (0,
			   "invalid interface (only mlx5 supported for now)");
      goto err2;
    }

  for (i = 0; i < n_devs; i++)
    {
      vlib_pci_addr_t addr;

      vec_reset_length (s);
      s = format (s, "%s/device%c", dev_list[i]->dev_path, 0);

      if (sysfs_path_to_pci_addr ((char *) s, &addr) == 0)
	continue;

      if (addr.as_u32 != rd->pci->addr.as_u32)
	continue;

      if ((rd->ctx = ibv_open_device (dev_list[i])))
	break;
    }

  if ((args->error =
       rdma_dev_init (vm, rd, args->rxq_size, args->txq_size, args->rxq_num)))
    goto err2;

  if ((args->error = rdma_register_interface (vnm, rd)))
    goto err2;

  if ((args->error = rdma_async_event_init (rd)))
    goto err3;

  rdma_update_state (vnm, rd, 1);

  vnet_sw_interface_t *sw = vnet_get_hw_sw_interface (vnm, rd->hw_if_index);
  args->sw_if_index = rd->sw_if_index = sw->sw_if_index;
  /*
   * FIXME: add support for interrupt mode
   * vnet_hw_interface_t *hw = vnet_get_hw_interface (vnm, rd->hw_if_index);
   * hw->flags |= VNET_HW_INTERFACE_FLAG_SUPPORTS_INT_MODE;
   */
  vnet_hw_interface_set_input_node (vnm, rd->hw_if_index,
				    rdma_input_node.index);
  vec_foreach_index (qid, rd->rxqs)
    vnet_hw_interface_assign_rx_thread (vnm, rd->hw_if_index, qid, ~0);

  vec_free (s);
  return;

err3:
  rdma_unregister_interface (vnm, rd);
err2:
  rdma_dev_cleanup (rd);
err1:
  ibv_free_device_list (dev_list);
  vec_free (s);
  args->rv = VNET_API_ERROR_INVALID_INTERFACE;
err0:
  vlib_log_err (rm->log_class, "%U", format_clib_error, args->error);
}

void
rdma_delete_if (vlib_main_t * vm, rdma_device_t * rd)
{
  rdma_async_event_cleanup (rd);
  rdma_unregister_interface (vnet_get_main (), rd);
  rdma_dev_cleanup (rd);
}

static clib_error_t *
rdma_interface_admin_up_down (vnet_main_t * vnm, u32 hw_if_index, u32 flags)
{
  vnet_hw_interface_t *hi = vnet_get_hw_interface (vnm, hw_if_index);
  rdma_main_t *rm = &rdma_main;
  rdma_device_t *rd = vec_elt_at_index (rm->devices, hi->dev_instance);
  uword is_up = (flags & VNET_SW_INTERFACE_FLAG_ADMIN_UP) != 0;

  if (rd->flags & RDMA_DEVICE_F_ERROR)
    return clib_error_return (0, "device is in error state");

  if (is_up)
    {
      vnet_hw_interface_set_flags (vnm, rd->hw_if_index,
				   VNET_HW_INTERFACE_FLAG_LINK_UP);
      rd->flags |= RDMA_DEVICE_F_ADMIN_UP;
    }
  else
    {
      vnet_hw_interface_set_flags (vnm, rd->hw_if_index, 0);
      rd->flags &= ~RDMA_DEVICE_F_ADMIN_UP;
    }
  return 0;
}

static void
rdma_set_interface_next_node (vnet_main_t * vnm, u32 hw_if_index,
			      u32 node_index)
{
  rdma_main_t *rm = &rdma_main;
  vnet_hw_interface_t *hw = vnet_get_hw_interface (vnm, hw_if_index);
  rdma_device_t *rd = pool_elt_at_index (rm->devices, hw->dev_instance);
  rd->per_interface_next_index =
    ~0 ==
    node_index ? VNET_DEVICE_INPUT_NEXT_ETHERNET_INPUT :
    vlib_node_add_next (vlib_get_main (), rdma_input_node.index, node_index);
}

static char *rdma_tx_func_error_strings[] = {
#define _(n,s) s,
  foreach_rdma_tx_func_error
#undef _
};

/* *INDENT-OFF* */
VNET_DEVICE_CLASS (rdma_device_class) =
{
  .name = "RDMA interface",
  .format_device = format_rdma_device,
  .format_device_name = format_rdma_device_name,
  .admin_up_down_function = rdma_interface_admin_up_down,
  .rx_redirect_to_node = rdma_set_interface_next_node,
  .tx_function_n_errors = RDMA_TX_N_ERROR,
  .tx_function_error_strings = rdma_tx_func_error_strings,
  .mac_addr_change_function = rdma_mac_change,
};
/* *INDENT-ON* */

clib_error_t *
rdma_init (vlib_main_t * vm)
{
  rdma_main_t *rm = &rdma_main;

  rm->log_class = vlib_log_register_class ("rdma", 0);

  return 0;
}

/* *INDENT-OFF* */
VLIB_INIT_FUNCTION (rdma_init) =
{
  .runs_after = VLIB_INITS ("pci_bus_init"),
};
/* *INDENT-OFF* */

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