FD.io VPP  v19.01.3-6-g70449b9b9
Vector Packet Processing
init.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2015 Cisco and/or its affiliates.
3  * Licensed under the Apache License, Version 2.0 (the "License");
4  * you may not use this file except in compliance with the License.
5  * You may obtain a copy of the License at:
6  *
7  * http://www.apache.org/licenses/LICENSE-2.0
8  *
9  * Unless required by applicable law or agreed to in writing, software
10  * distributed under the License is distributed on an "AS IS" BASIS,
11  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12  * See the License for the specific language governing permissions and
13  * limitations under the License.
14  */
15 #include <vnet/vnet.h>
16 #include <vppinfra/vec.h>
17 #include <vppinfra/error.h>
18 #include <vppinfra/format.h>
19 #include <vppinfra/bitmap.h>
20 #include <vppinfra/linux/sysfs.h>
21 #include <vlib/unix/unix.h>
22 #include <vlib/log.h>
23 
24 #include <vnet/ethernet/ethernet.h>
25 #include <dpdk/device/dpdk.h>
26 #include <vlib/pci/pci.h>
27 #include <vlib/vmbus/vmbus.h>
28 
29 #include <rte_ring.h>
30 
31 #include <stdio.h>
32 #include <stdlib.h>
33 #include <unistd.h>
34 #include <sys/stat.h>
35 #include <sys/mount.h>
36 #include <string.h>
37 #include <fcntl.h>
38 
39 #include <dpdk/device/dpdk_priv.h>
40 
41 #define ETHER_MAX_LEN 1518 /**< Maximum frame len, including CRC. */
42 
45 
46 #define LINK_STATE_ELOGS 0
47 
48 /* Port configuration, mildly modified Intel app values */
49 
50 static dpdk_port_type_t
51 port_type_from_speed_capa (struct rte_eth_dev_info *dev_info)
52 {
53 
54  if (dev_info->speed_capa & ETH_LINK_SPEED_100G)
56  else if (dev_info->speed_capa & ETH_LINK_SPEED_56G)
58  else if (dev_info->speed_capa & ETH_LINK_SPEED_50G)
60  else if (dev_info->speed_capa & ETH_LINK_SPEED_40G)
62  else if (dev_info->speed_capa & ETH_LINK_SPEED_25G)
64  else if (dev_info->speed_capa & ETH_LINK_SPEED_20G)
66  else if (dev_info->speed_capa & ETH_LINK_SPEED_10G)
68  else if (dev_info->speed_capa & ETH_LINK_SPEED_5G)
70  else if (dev_info->speed_capa & ETH_LINK_SPEED_2_5G)
72  else if (dev_info->speed_capa & ETH_LINK_SPEED_1G)
74 
76 }
77 
78 static dpdk_port_type_t
80 {
81  switch (link_speed)
82  {
83  case ETH_SPEED_NUM_1G:
85  case ETH_SPEED_NUM_2_5G:
87  case ETH_SPEED_NUM_5G:
89  case ETH_SPEED_NUM_10G:
91  case ETH_SPEED_NUM_20G:
93  case ETH_SPEED_NUM_25G:
95  case ETH_SPEED_NUM_40G:
97  case ETH_SPEED_NUM_50G:
99  case ETH_SPEED_NUM_56G:
101  case ETH_SPEED_NUM_100G:
103  default:
105  }
106 }
107 
108 static u32
110 {
111  dpdk_main_t *dm = &dpdk_main;
113  u32 old = 0;
114 
116  {
117  old = (xd->flags & DPDK_DEVICE_FLAG_PROMISC) != 0;
118 
120  xd->flags |= DPDK_DEVICE_FLAG_PROMISC;
121  else
122  xd->flags &= ~DPDK_DEVICE_FLAG_PROMISC;
123 
124  if (xd->flags & DPDK_DEVICE_FLAG_ADMIN_UP)
125  {
126  if (xd->flags & DPDK_DEVICE_FLAG_PROMISC)
127  rte_eth_promiscuous_enable (xd->port_id);
128  else
129  rte_eth_promiscuous_disable (xd->port_id);
130  }
131  }
132  else if (ETHERNET_INTERFACE_FLAG_CONFIG_MTU (flags))
133  {
134  xd->port_conf.rxmode.max_rx_pkt_len = hi->max_packet_bytes;
135  dpdk_device_setup (xd);
136  }
137  return old;
138 }
139 
140 static void
142 {
143  int q;
144  vec_validate (xd->lockp, xd->tx_q_used - 1);
145  for (q = 0; q < xd->tx_q_used; q++)
146  {
149  clib_memset ((void *) xd->lockp[q], 0, CLIB_CACHE_LINE_BYTES);
150  }
151 }
152 
153 static struct rte_mempool_ops *
154 get_ops_by_name (char *ops_name)
155 {
156  u32 i;
157 
158  for (i = 0; i < rte_mempool_ops_table.num_ops; i++)
159  {
160  if (!strcmp (ops_name, rte_mempool_ops_table.ops[i].name))
161  return &rte_mempool_ops_table.ops[i];
162  }
163 
164  return 0;
165 }
166 
167 static int
168 dpdk_ring_alloc (struct rte_mempool *mp)
169 {
170  u32 rg_flags = 0, count;
171  i32 ret;
172  char rg_name[RTE_RING_NAMESIZE];
173  struct rte_ring *r;
174 
175  ret = snprintf (rg_name, sizeof (rg_name), RTE_MEMPOOL_MZ_FORMAT, mp->name);
176  if (ret < 0 || ret >= (i32) sizeof (rg_name))
177  return -ENAMETOOLONG;
178 
179  /* ring flags */
180  if (mp->flags & MEMPOOL_F_SP_PUT)
181  rg_flags |= RING_F_SP_ENQ;
182  if (mp->flags & MEMPOOL_F_SC_GET)
183  rg_flags |= RING_F_SC_DEQ;
184 
185  count = rte_align32pow2 (mp->size + 1);
186  /*
187  * Allocate the ring that will be used to store objects.
188  * Ring functions will return appropriate errors if we are
189  * running as a secondary process etc., so no checks made
190  * in this function for that condition.
191  */
192  /* XXX can we get memory from the right socket? */
193  r = clib_mem_alloc_aligned (rte_ring_get_memsize (count),
195 
196  /* XXX rte_ring_lookup will not work */
197 
198  ret = rte_ring_init (r, rg_name, count, rg_flags);
199  if (ret)
200  return ret;
201 
202  mp->pool_data = r;
203 
204  return 0;
205 }
206 
207 static int
209 {
210 #if RTE_VERSION < RTE_VERSION_NUM(18, 11, 0, 0)
211  return ! !(xd->port_conf.rxmode.offloads & DEV_RX_OFFLOAD_CRC_STRIP);
212 #else
213  return !(xd->port_conf.rxmode.offloads & DEV_RX_OFFLOAD_KEEP_CRC);
214 #endif
215 }
216 
217 static clib_error_t *
219 {
220  u32 nports;
221  u32 mtu, max_rx_frame;
222  u32 nb_desc = 0;
223  int i;
224  clib_error_t *error;
230  dpdk_device_t *xd;
231  vlib_pci_addr_t last_pci_addr;
232  u32 last_pci_addr_port = 0;
234  uword *p_hqos;
235 
236  u32 next_hqos_cpu = 0;
237  u8 af_packet_instance_num = 0;
238  u8 bond_ether_instance_num = 0;
239  last_pci_addr.as_u32 = ~0;
240 
241  dm->hqos_cpu_first_index = 0;
242  dm->hqos_cpu_count = 0;
243 
244  /* find out which cpus will be used for I/O TX */
245  p_hqos = hash_get_mem (tm->thread_registrations_by_name, "hqos-threads");
246  tr_hqos = p_hqos ? (vlib_thread_registration_t *) p_hqos[0] : 0;
247 
248  if (tr_hqos && tr_hqos->count > 0)
249  {
250  dm->hqos_cpu_first_index = tr_hqos->first_index;
251  dm->hqos_cpu_count = tr_hqos->count;
252  }
253 
256 
257  nports = rte_eth_dev_count_avail ();
258 
259  if (nports < 1)
260  {
261  dpdk_log_notice ("DPDK drivers found no ports...");
262  }
263 
264  if (CLIB_DEBUG > 0)
265  dpdk_log_notice ("DPDK drivers found %d ports...", nports);
266 
267  if (dm->conf->enable_tcp_udp_checksum)
268  dm->buffer_flags_template &= ~(VNET_BUFFER_F_L4_CHECKSUM_CORRECT
269  | VNET_BUFFER_F_L4_CHECKSUM_COMPUTED);
270 
271  /* vlib_buffer_t template */
274  for (i = 0; i < tm->n_vlib_mains; i++)
275  {
278  fl = vlib_buffer_get_free_list (vm,
282  vnet_buffer (&ptd->buffer_template)->sw_if_index[VLIB_TX] = (u32) ~ 0;
283  }
284 
285  /* *INDENT-OFF* */
286  RTE_ETH_FOREACH_DEV(i)
287  {
288  u8 addr[6];
289  u8 vlan_strip = 0;
290  struct rte_eth_dev_info dev_info;
291  struct rte_pci_device *pci_dev;
292  struct rte_eth_link l;
293  dpdk_device_config_t *devconf = 0;
294  vlib_pci_addr_t pci_addr;
295  uword *p = 0;
296 
297  if (!rte_eth_dev_is_valid_port(i))
298  continue;
299 
300  rte_eth_link_get_nowait (i, &l);
301  rte_eth_dev_info_get (i, &dev_info);
302 
303  if (dev_info.device == 0)
304  {
305  clib_warning ("DPDK bug: missing device info. Skipping %s device",
306  dev_info.driver_name);
307  continue;
308  }
309 
310  pci_dev = dpdk_get_pci_device (&dev_info);
311 
312  if (pci_dev) /* bonded interface has no pci info */
313  {
314  pci_addr.domain = pci_dev->addr.domain;
315  pci_addr.bus = pci_dev->addr.bus;
316  pci_addr.slot = pci_dev->addr.devid;
317  pci_addr.function = pci_dev->addr.function;
319  pci_addr.as_u32);
320  }
321 
322 
323  /* Create vnet interface */
327  xd->cpu_socket = (i8) rte_eth_dev_socket_id (i);
328 
329  if (p)
330  {
331  devconf = pool_elt_at_index (dm->conf->dev_confs, p[0]);
332  xd->name = devconf->name;
333  }
334  else
335  devconf = &dm->conf->default_devconf;
336 
337  /* Handle interface naming for devices with multiple ports sharing same PCI ID */
338  if (pci_dev)
339  {
340  struct rte_eth_dev_info di = { 0 };
341  struct rte_pci_device *next_pci_dev;
342  rte_eth_dev_info_get (i + 1, &di);
343  next_pci_dev = di.device ? RTE_DEV_TO_PCI (di.device) : 0;
344  if (pci_dev && next_pci_dev &&
345  pci_addr.as_u32 != last_pci_addr.as_u32 &&
346  memcmp (&pci_dev->addr, &next_pci_dev->addr,
347  sizeof (struct rte_pci_addr)) == 0)
348  {
349  xd->interface_name_suffix = format (0, "0");
350  last_pci_addr.as_u32 = pci_addr.as_u32;
351  last_pci_addr_port = i;
352  }
353  else if (pci_addr.as_u32 == last_pci_addr.as_u32)
354  {
356  format (0, "%u", i - last_pci_addr_port);
357  }
358  else
359  {
360  last_pci_addr.as_u32 = ~0;
361  }
362  }
363  else
364  last_pci_addr.as_u32 = ~0;
365 
366  clib_memcpy (&xd->tx_conf, &dev_info.default_txconf,
367  sizeof (struct rte_eth_txconf));
368 
369  if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_IPV4_CKSUM)
370  {
371  xd->port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_IPV4_CKSUM;
372  xd->flags |= DPDK_DEVICE_FLAG_RX_IP4_CKSUM;
373  }
374 
375  if (dm->conf->no_multi_seg)
376  {
377  xd->port_conf.txmode.offloads &= ~DEV_TX_OFFLOAD_MULTI_SEGS;
378  xd->port_conf.rxmode.offloads &= ~DEV_RX_OFFLOAD_JUMBO_FRAME;
379  xd->port_conf.rxmode.offloads &= ~DEV_RX_OFFLOAD_SCATTER;
380  }
381  else
382  {
383  xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_MULTI_SEGS;
384  xd->port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
385  xd->port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_SCATTER;
386  xd->flags |= DPDK_DEVICE_FLAG_MAYBE_MULTISEG;
387  }
388 
389  xd->tx_q_used = clib_min (dev_info.max_tx_queues, tm->n_vlib_mains);
390 
391  if (devconf->num_tx_queues > 0
392  && devconf->num_tx_queues < xd->tx_q_used)
393  xd->tx_q_used = clib_min (xd->tx_q_used, devconf->num_tx_queues);
394 
395  if (devconf->num_rx_queues > 1
396  && dev_info.max_rx_queues >= devconf->num_rx_queues)
397  {
398  xd->rx_q_used = devconf->num_rx_queues;
399  xd->port_conf.rxmode.mq_mode = ETH_MQ_RX_RSS;
400  if (devconf->rss_fn == 0)
401  xd->port_conf.rx_adv_conf.rss_conf.rss_hf =
402  ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP;
403  else
404  {
405  u64 unsupported_bits;
406  xd->port_conf.rx_adv_conf.rss_conf.rss_hf = devconf->rss_fn;
407  unsupported_bits = xd->port_conf.rx_adv_conf.rss_conf.rss_hf;
408  unsupported_bits &= ~dev_info.flow_type_rss_offloads;
409  if (unsupported_bits)
410  dpdk_log_warn ("Unsupported RSS hash functions: %U",
411  format_dpdk_rss_hf_name, unsupported_bits);
412  }
413  xd->port_conf.rx_adv_conf.rss_conf.rss_hf &=
414  dev_info.flow_type_rss_offloads;
415  }
416  else
417  xd->rx_q_used = 1;
418 
419  xd->flags |= DPDK_DEVICE_FLAG_PMD;
420 
421  /* workaround for drivers not setting driver_name */
422  if ((!dev_info.driver_name) && (pci_dev))
423  dev_info.driver_name = pci_dev->driver->driver.name;
424 
425  ASSERT (dev_info.driver_name);
426 
427  if (!xd->pmd)
428  {
429 
430 
431 #define _(s,f) else if (dev_info.driver_name && \
432  !strcmp(dev_info.driver_name, s)) \
433  xd->pmd = VNET_DPDK_PMD_##f;
434  if (0)
435  ;
437 #undef _
438  else
440 
444 
445  switch (xd->pmd)
446  {
447  /* Drivers with valid speed_capa set */
448  case VNET_DPDK_PMD_E1000EM:
449  case VNET_DPDK_PMD_IGB:
450  case VNET_DPDK_PMD_IXGBE:
451  case VNET_DPDK_PMD_I40E:
452  xd->port_type = port_type_from_speed_capa (&dev_info);
453  xd->supported_flow_actions = VNET_FLOW_ACTION_MARK |
454  VNET_FLOW_ACTION_REDIRECT_TO_NODE |
455  VNET_FLOW_ACTION_BUFFER_ADVANCE |
456  VNET_FLOW_ACTION_COUNT | VNET_FLOW_ACTION_DROP;
457 
458  if (dm->conf->no_tx_checksum_offload == 0)
459  {
460  xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_TCP_CKSUM;
461  xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_UDP_CKSUM;
462  xd->flags |=
463  DPDK_DEVICE_FLAG_TX_OFFLOAD |
464  DPDK_DEVICE_FLAG_INTEL_PHDR_CKSUM;
465  }
466 
467 
468  break;
469  case VNET_DPDK_PMD_CXGBE:
470  case VNET_DPDK_PMD_MLX4:
471  case VNET_DPDK_PMD_MLX5:
472  case VNET_DPDK_PMD_QEDE:
473  xd->port_type = port_type_from_speed_capa (&dev_info);
474  break;
475 
476  /* SR-IOV VFs */
477  case VNET_DPDK_PMD_IGBVF:
478  case VNET_DPDK_PMD_IXGBEVF:
479  case VNET_DPDK_PMD_I40EVF:
481 #if RTE_VERSION < RTE_VERSION_NUM(18, 11, 0, 0)
482  xd->port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_CRC_STRIP;
483 #endif
484  break;
485 
486  case VNET_DPDK_PMD_THUNDERX:
488 #if RTE_VERSION < RTE_VERSION_NUM(18, 11, 0, 0)
489  xd->port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_CRC_STRIP;
490 #endif
491 
492  if (dm->conf->no_tx_checksum_offload == 0)
493  {
494  xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_TCP_CKSUM;
495  xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_UDP_CKSUM;
496  xd->flags |= DPDK_DEVICE_FLAG_TX_OFFLOAD;
497  }
498  break;
499 
500  case VNET_DPDK_PMD_ENA:
502  xd->port_conf.rxmode.offloads &= ~DEV_RX_OFFLOAD_SCATTER;
503  break;
504 
505  case VNET_DPDK_PMD_DPAA2:
507  break;
508 
509  /* Cisco VIC */
510  case VNET_DPDK_PMD_ENIC:
511  if (l.link_speed == 40000)
513  else
515  break;
516 
517  /* Intel Red Rock Canyon */
518  case VNET_DPDK_PMD_FM10K:
520 #if RTE_VERSION < RTE_VERSION_NUM(18, 11, 0, 0)
521  xd->port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_CRC_STRIP;
522 #endif
523  break;
524 
525  /* virtio */
526  case VNET_DPDK_PMD_VIRTIO:
530  break;
531 
532  /* vmxnet3 */
533  case VNET_DPDK_PMD_VMXNET3:
535  xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_MULTI_SEGS;
536  break;
537 
538  case VNET_DPDK_PMD_AF_PACKET:
540  xd->af_packet_instance_num = af_packet_instance_num++;
541  break;
542 
543  case VNET_DPDK_PMD_BOND:
545  xd->bond_instance_num = bond_ether_instance_num++;
546  break;
547 
548  case VNET_DPDK_PMD_VIRTIO_USER:
550  break;
551 
552  case VNET_DPDK_PMD_VHOST_ETHER:
554  break;
555 
556  case VNET_DPDK_PMD_LIOVF_ETHER:
558  break;
559 
560  case VNET_DPDK_PMD_FAILSAFE:
562  xd->port_conf.intr_conf.lsc = 1;
563  break;
564 
565  case VNET_DPDK_PMD_NETVSC:
566  xd->port_type = port_type_from_link_speed (l.link_speed);
567  break;
568 
569  default:
571  }
572 
573  if (devconf->num_rx_desc)
574  xd->nb_rx_desc = devconf->num_rx_desc;
575 
576  if (devconf->num_tx_desc)
577  xd->nb_tx_desc = devconf->num_tx_desc;
578  }
579 
580  if (xd->pmd == VNET_DPDK_PMD_AF_PACKET)
581  {
582  f64 now = vlib_time_now (vm);
583  u32 rnd;
584  rnd = (u32) (now * 1e6);
585  rnd = random_u32 (&rnd);
586  clib_memcpy (addr + 2, &rnd, sizeof (rnd));
587  addr[0] = 2;
588  addr[1] = 0xfe;
589  }
590  else
591  rte_eth_macaddr_get (i, (struct ether_addr *) addr);
592 
593  if (xd->tx_q_used < tm->n_vlib_mains)
595 
596  xd->port_id = i;
597  xd->device_index = xd - dm->devices;
598  xd->per_interface_next_index = ~0;
599 
600  /* assign interface to input thread */
601  int q;
602 
603  if (devconf->hqos_enabled)
604  {
605  xd->flags |= DPDK_DEVICE_FLAG_HQOS;
606 
607  int cpu;
608  if (devconf->hqos.hqos_thread_valid)
609  {
610  if (devconf->hqos.hqos_thread >= dm->hqos_cpu_count)
611  return clib_error_return (0, "invalid HQoS thread index");
612 
613  cpu = dm->hqos_cpu_first_index + devconf->hqos.hqos_thread;
614  }
615  else
616  {
617  if (dm->hqos_cpu_count == 0)
618  return clib_error_return (0, "no HQoS threads available");
619 
620  cpu = dm->hqos_cpu_first_index + next_hqos_cpu;
621 
622  next_hqos_cpu++;
623  if (next_hqos_cpu == dm->hqos_cpu_count)
624  next_hqos_cpu = 0;
625 
626  devconf->hqos.hqos_thread_valid = 1;
627  devconf->hqos.hqos_thread = cpu;
628  }
629 
631  vec_add2 (dm->devices_by_hqos_cpu[cpu], dq, 1);
632  dq->device = xd->device_index;
633  dq->queue_id = 0;
634  }
635 
636  /* count the number of descriptors used for this device */
637  nb_desc += xd->nb_rx_desc + xd->nb_tx_desc * xd->tx_q_used;
638 
640  (dm->vnet_main, dpdk_device_class.index, xd->device_index,
641  /* ethernet address */ addr,
643  if (error)
644  return error;
645 
646  /*
647  * Ensure default mtu is not > the mtu read from the hardware.
648  * Otherwise rte_eth_dev_configure() will fail and the port will
649  * not be available.
650  * Calculate max_frame_size and mtu supported by NIC
651  */
652  if (ETHERNET_MAX_PACKET_BYTES > dev_info.max_rx_pktlen)
653  {
654  /*
655  * This device does not support the platforms's max frame
656  * size. Use it's advertised mru instead.
657  */
658  max_rx_frame = dev_info.max_rx_pktlen;
659  mtu = dev_info.max_rx_pktlen - sizeof (ethernet_header_t);
660  }
661  else
662  {
663  /* VPP treats MTU and max_rx_pktlen both equal to
664  * ETHERNET_MAX_PACKET_BYTES, if dev_info.max_rx_pktlen >=
665  * ETHERNET_MAX_PACKET_BYTES + sizeof(ethernet_header_t)
666  */
667  if (dev_info.max_rx_pktlen >= (ETHERNET_MAX_PACKET_BYTES +
668  sizeof (ethernet_header_t)))
669  {
671  max_rx_frame = ETHERNET_MAX_PACKET_BYTES;
672 
673  /*
674  * Some platforms do not account for Ethernet FCS (4 bytes) in
675  * MTU calculations. To interop with them increase mru but only
676  * if the device's settings can support it.
677  */
678  if (dpdk_port_crc_strip_enabled (xd) &&
679  (dev_info.max_rx_pktlen >= (ETHERNET_MAX_PACKET_BYTES +
680  sizeof (ethernet_header_t) +
681  4)))
682  {
683  max_rx_frame += 4;
684  }
685  }
686  else
687  {
688  max_rx_frame = ETHERNET_MAX_PACKET_BYTES;
690 
691  if (dpdk_port_crc_strip_enabled (xd) &&
692  (dev_info.max_rx_pktlen >= (ETHERNET_MAX_PACKET_BYTES + 4)))
693  {
694  max_rx_frame += 4;
695  }
696  }
697  }
698 
699  if (xd->pmd == VNET_DPDK_PMD_FAILSAFE)
700  {
701  /* failsafe device numerables are reported with active device only,
702  * need to query the mtu for current device setup to overwrite
703  * reported value.
704  */
705  uint16_t dev_mtu;
706  if (!rte_eth_dev_get_mtu (i, &dev_mtu))
707  {
708  mtu = dev_mtu;
709  max_rx_frame = mtu + sizeof (ethernet_header_t);
710 
712  {
713  max_rx_frame += 4;
714  }
715  }
716  }
717 
718  /*Set port rxmode config */
719  xd->port_conf.rxmode.max_rx_pkt_len = max_rx_frame;
720 
722  xd->sw_if_index = sw->sw_if_index;
724  dpdk_input_node.index);
725 
726  if (devconf->workers)
727  {
728  int i;
729  q = 0;
730  clib_bitmap_foreach (i, devconf->workers, ({
731  vnet_hw_interface_assign_rx_thread (dm->vnet_main, xd->hw_if_index, q++,
732  vdm->first_worker_thread_index + i);
733  }));
734  }
735  else
736  for (q = 0; q < xd->rx_q_used; q++)
737  {
739  ~1);
740  }
741 
742  /*Get vnet hardware interface */
744 
745  /*Override default max_packet_bytes and max_supported_bytes set in
746  * ethernet_register_interface() above*/
747  if (hi)
748  {
749  hi->max_packet_bytes = mtu;
750  hi->max_supported_packet_bytes = max_rx_frame;
751  }
752 
753  if (dm->conf->no_tx_checksum_offload == 0)
754  if (xd->flags & DPDK_DEVICE_FLAG_TX_OFFLOAD && hi != NULL)
756 
757  dpdk_device_setup (xd);
758 
759  if (vec_len (xd->errors))
760  dpdk_log_err ("setup failed for device %U. Errors:\n %U",
763 
764  if (devconf->hqos_enabled)
765  {
766  clib_error_t *rv;
767  rv = dpdk_port_setup_hqos (xd, &devconf->hqos);
768  if (rv)
769  return rv;
770  }
771 
772  /*
773  * For cisco VIC vNIC, set default to VLAN strip enabled, unless
774  * specified otherwise in the startup config.
775  * For other NICs default to VLAN strip disabled, unless specified
776  * otherwise in the startup config.
777  */
778  if (xd->pmd == VNET_DPDK_PMD_ENIC)
779  {
780  if (devconf->vlan_strip_offload != DPDK_DEVICE_VLAN_STRIP_OFF)
781  vlan_strip = 1; /* remove vlan tag from VIC port by default */
782  else
783  dpdk_log_warn ("VLAN strip disabled for interface\n");
784  }
785  else if (devconf->vlan_strip_offload == DPDK_DEVICE_VLAN_STRIP_ON)
786  vlan_strip = 1;
787 
788  if (vlan_strip)
789  {
790  int vlan_off;
791  vlan_off = rte_eth_dev_get_vlan_offload (xd->port_id);
792  vlan_off |= ETH_VLAN_STRIP_OFFLOAD;
793  if (vlan_off)
794  xd->port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_STRIP;
795  else
796  xd->port_conf.rxmode.offloads &= ~DEV_RX_OFFLOAD_VLAN_STRIP;
797  if (rte_eth_dev_set_vlan_offload (xd->port_id, vlan_off) == 0)
798  dpdk_log_info ("VLAN strip enabled for interface\n");
799  else
800  dpdk_log_warn ("VLAN strip cannot be supported by interface\n");
801  }
802 
803  if (hi)
804  hi->max_packet_bytes = xd->port_conf.rxmode.max_rx_pkt_len
805  - sizeof (ethernet_header_t);
806  else
807  clib_warning ("hi NULL");
808 
809  if (dm->conf->no_multi_seg)
810  mtu = mtu > ETHER_MAX_LEN ? ETHER_MAX_LEN : mtu;
811 
812  rte_eth_dev_set_mtu (xd->port_id, mtu);
813  }
814  /* *INDENT-ON* */
815 
816  if (nb_desc > dm->conf->num_mbufs)
817  dpdk_log_err ("%d mbufs allocated but total rx/tx ring size is %d\n",
818  dm->conf->num_mbufs, nb_desc);
819 
820  return 0;
821 }
822 
823 static void
825 {
827  clib_error_t *error;
828  u8 *pci_addr = 0;
829  int num_whitelisted = vec_len (conf->dev_confs);
830  vlib_pci_device_info_t *d = 0;
831  vlib_pci_addr_t *addr = 0, *addrs;
832  int i;
833 
834  addrs = vlib_pci_get_all_dev_addrs ();
835  /* *INDENT-OFF* */
836  vec_foreach (addr, addrs)
837  {
838  dpdk_device_config_t * devconf = 0;
839  vec_reset_length (pci_addr);
840  pci_addr = format (pci_addr, "%U%c", format_vlib_pci_addr, addr, 0);
841  if (d)
842  {
844  d = 0;
845  }
846  d = vlib_pci_get_device_info (vm, addr, &error);
847  if (error)
848  {
849  clib_error_report (error);
850  continue;
851  }
852 
854  continue;
855 
856  if (num_whitelisted)
857  {
858  uword * p = hash_get (conf->device_config_index_by_pci_addr, addr->as_u32);
859 
860  if (!p)
861  {
862  skipped:
863  continue;
864  }
865 
866  devconf = pool_elt_at_index (conf->dev_confs, p[0]);
867  }
868 
869  /* Enforce Device blacklist by vendor and device */
870  for (i = 0; i < vec_len (conf->blacklist_by_pci_vendor_and_device); i++)
871  {
872  u16 vendor, device;
873  vendor = (u16)(conf->blacklist_by_pci_vendor_and_device[i] >> 16);
874  device = (u16)(conf->blacklist_by_pci_vendor_and_device[i] & 0xFFFF);
875  if (d->vendor_id == vendor && d->device_id == device)
876  {
877  /*
878  * Expected case: device isn't whitelisted,
879  * so blacklist it...
880  */
881  if (devconf == 0)
882  {
883  /* Device is blacklisted */
884  pool_get (conf->dev_confs, devconf);
885  hash_set (conf->device_config_index_by_pci_addr, addr->as_u32,
886  devconf - conf->dev_confs);
887  devconf->pci_addr.as_u32 = addr->as_u32;
888  devconf->is_blacklisted = 1;
889  goto skipped;
890  }
891  else /* explicitly whitelisted, ignore the device blacklist */
892  break;
893  }
894  }
895 
896  /* virtio */
897  if (d->vendor_id == 0x1af4 &&
900  ;
901  /* vmxnet3 */
902  else if (d->vendor_id == 0x15ad && d->device_id == 0x07b0)
903  {
904  /*
905  * For vmxnet3 PCI, unless it is explicitly specified in the whitelist,
906  * the default is to put it in the blacklist.
907  */
908  if (devconf == 0)
909  {
910  pool_get (conf->dev_confs, devconf);
911  hash_set (conf->device_config_index_by_pci_addr, addr->as_u32,
912  devconf - conf->dev_confs);
913  devconf->pci_addr.as_u32 = addr->as_u32;
914  devconf->is_blacklisted = 1;
915  }
916  }
917  /* all Intel network devices */
918  else if (d->vendor_id == 0x8086 && d->device_class == PCI_CLASS_NETWORK_ETHERNET)
919  ;
920  /* all Intel QAT devices VFs */
921  else if (d->vendor_id == 0x8086 && d->device_class == PCI_CLASS_PROCESSOR_CO &&
922  (d->device_id == 0x0443 || d->device_id == 0x37c9 || d->device_id == 0x19e3))
923  ;
924  /* Cisco VIC */
925  else if (d->vendor_id == 0x1137 && d->device_id == 0x0043)
926  ;
927  /* Chelsio T4/T5 */
928  else if (d->vendor_id == 0x1425 && (d->device_id & 0xe000) == 0x4000)
929  ;
930  /* Amazen Elastic Network Adapter */
931  else if (d->vendor_id == 0x1d0f && d->device_id >= 0xec20 && d->device_id <= 0xec21)
932  ;
933  /* Cavium Network Adapter */
934  else if (d->vendor_id == 0x177d && d->device_id == 0x9712)
935  ;
936  /* Cavium FastlinQ QL41000 Series */
937  else if (d->vendor_id == 0x1077 && d->device_id >= 0x8070 && d->device_id <= 0x8090)
938  ;
939  /* Mellanox mlx4 */
940  else if (d->vendor_id == 0x15b3 && d->device_id >= 0x1003 && d->device_id <= 0x1004)
941  {
942  continue;
943  }
944  /* Mellanox mlx5 */
945  else if (d->vendor_id == 0x15b3 && d->device_id >= 0x1013 && d->device_id <= 0x101a)
946  {
947  continue;
948  }
949  else
950  {
951  dpdk_log_warn ("Unsupported PCI device 0x%04x:0x%04x found "
952  "at PCI address %s\n", (u16) d->vendor_id, (u16) d->device_id,
953  pci_addr);
954  continue;
955  }
956 
957  error = vlib_pci_bind_to_uio (vm, addr, (char *) conf->uio_driver_name);
958 
959  if (error)
960  {
961  if (devconf == 0)
962  {
963  pool_get (conf->dev_confs, devconf);
964  hash_set (conf->device_config_index_by_pci_addr, addr->as_u32,
965  devconf - conf->dev_confs);
966  devconf->pci_addr.as_u32 = addr->as_u32;
967  }
968  devconf->is_blacklisted = 1;
969  clib_error_report (error);
970  }
971  }
972  /* *INDENT-ON* */
973  vec_free (pci_addr);
975 }
976 
977 static void
979 {
980  clib_error_t *error;
981  vlib_vmbus_addr_t *addrs, *addr = 0;
982 
983  addrs = vlib_vmbus_get_all_dev_addrs ();
984 
985  /* *INDENT-OFF* */
986  vec_foreach (addr, addrs)
987  {
988  error = vlib_vmbus_bind_to_uio (addr);
989 
990  if (error)
991  {
992  clib_error_report (error);
993  }
994  }
995  /* *INDENT-ON* */
996 }
997 
998 static clib_error_t *
999 dpdk_device_config (dpdk_config_main_t * conf, vlib_pci_addr_t pci_addr,
1000  unformat_input_t * input, u8 is_default)
1001 {
1002  clib_error_t *error = 0;
1003  uword *p;
1004  dpdk_device_config_t *devconf;
1005  unformat_input_t sub_input;
1006 
1007  if (is_default)
1008  {
1009  devconf = &conf->default_devconf;
1010  }
1011  else
1012  {
1013  p = hash_get (conf->device_config_index_by_pci_addr, pci_addr.as_u32);
1014 
1015  if (!p)
1016  {
1017  pool_get (conf->dev_confs, devconf);
1018  hash_set (conf->device_config_index_by_pci_addr, pci_addr.as_u32,
1019  devconf - conf->dev_confs);
1020  }
1021  else
1022  return clib_error_return (0,
1023  "duplicate configuration for PCI address %U",
1024  format_vlib_pci_addr, &pci_addr);
1025  }
1026 
1027  devconf->pci_addr.as_u32 = pci_addr.as_u32;
1028  devconf->hqos_enabled = 0;
1030 
1031  if (!input)
1032  return 0;
1033 
1034  unformat_skip_white_space (input);
1035  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
1036  {
1037  if (unformat (input, "num-rx-queues %u", &devconf->num_rx_queues))
1038  ;
1039  else if (unformat (input, "num-tx-queues %u", &devconf->num_tx_queues))
1040  ;
1041  else if (unformat (input, "num-rx-desc %u", &devconf->num_rx_desc))
1042  ;
1043  else if (unformat (input, "num-tx-desc %u", &devconf->num_tx_desc))
1044  ;
1045  else if (unformat (input, "name %s", &devconf->name))
1046  ;
1047  else if (unformat (input, "workers %U", unformat_bitmap_list,
1048  &devconf->workers))
1049  ;
1050  else
1051  if (unformat
1052  (input, "rss %U", unformat_vlib_cli_sub_input, &sub_input))
1053  {
1054  error = unformat_rss_fn (&sub_input, &devconf->rss_fn);
1055  if (error)
1056  break;
1057  }
1058  else if (unformat (input, "vlan-strip-offload off"))
1060  else if (unformat (input, "vlan-strip-offload on"))
1062  else
1063  if (unformat
1064  (input, "hqos %U", unformat_vlib_cli_sub_input, &sub_input))
1065  {
1066  devconf->hqos_enabled = 1;
1067  error = unformat_hqos (&sub_input, &devconf->hqos);
1068  if (error)
1069  break;
1070  }
1071  else if (unformat (input, "hqos"))
1072  {
1073  devconf->hqos_enabled = 1;
1074  }
1075  else
1076  {
1077  error = clib_error_return (0, "unknown input `%U'",
1078  format_unformat_error, input);
1079  break;
1080  }
1081  }
1082 
1083  if (error)
1084  return error;
1085 
1086  if (devconf->workers && devconf->num_rx_queues == 0)
1087  devconf->num_rx_queues = clib_bitmap_count_set_bits (devconf->workers);
1088  else if (devconf->workers &&
1089  clib_bitmap_count_set_bits (devconf->workers) !=
1090  devconf->num_rx_queues)
1091  error =
1092  clib_error_return (0,
1093  "%U: number of worker threads must be "
1094  "equal to number of rx queues", format_vlib_pci_addr,
1095  &pci_addr);
1096 
1097  return error;
1098 }
1099 
1100 static clib_error_t *
1102 {
1103  unformat_input_t input;
1104  u8 *line, *s = 0;
1105  int n, n_try;
1106 
1107  n = n_try = 4096;
1108  while (n == n_try)
1109  {
1110  uword len = vec_len (s);
1111  vec_resize (s, len + n_try);
1112 
1113  n = read (uf->file_descriptor, s + len, n_try);
1114  if (n < 0 && errno != EAGAIN)
1115  return clib_error_return_unix (0, "read");
1116  _vec_len (s) = len + (n < 0 ? 0 : n);
1117  }
1118 
1119  unformat_init_vector (&input, s);
1120 
1121  while (unformat_user (&input, unformat_line, &line))
1122  {
1123  dpdk_log_notice ("%v", line);
1124  vec_free (line);
1125  }
1126 
1127  unformat_free (&input);
1128  return 0;
1129 }
1130 
1131 static clib_error_t *
1133 {
1134  clib_error_t *error = 0;
1137  dpdk_device_config_t *devconf;
1138  vlib_pci_addr_t pci_addr;
1139  unformat_input_t sub_input;
1140  uword default_hugepage_sz, x;
1141  u8 *s, *tmp = 0;
1142  u32 log_level;
1143  int ret, i;
1144  int num_whitelisted = 0;
1145  u8 no_pci = 0;
1146  u8 no_vmbus = 0;
1147  u8 file_prefix = 0;
1148  u8 *socket_mem = 0;
1149  u8 *huge_dir_path = 0;
1150  u32 vendor, device;
1151 
1152  huge_dir_path =
1153  format (0, "%s/hugepages%c", vlib_unix_get_runtime_dir (), 0);
1154 
1155  conf->device_config_index_by_pci_addr = hash_create (0, sizeof (uword));
1156  log_level = RTE_LOG_NOTICE;
1157 
1158  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
1159  {
1160  /* Prime the pump */
1161  if (unformat (input, "no-hugetlb"))
1162  {
1163  vec_add1 (conf->eal_init_args, (u8 *) "--no-huge");
1164  }
1165 
1166  else if (unformat (input, "enable-tcp-udp-checksum"))
1167  conf->enable_tcp_udp_checksum = 1;
1168 
1169  else if (unformat (input, "no-tx-checksum-offload"))
1170  conf->no_tx_checksum_offload = 1;
1171 
1172  else if (unformat (input, "decimal-interface-names"))
1174 
1175  else if (unformat (input, "log-level %U", unformat_dpdk_log_level, &x))
1176  log_level = x;
1177 
1178  else if (unformat (input, "no-multi-seg"))
1179  conf->no_multi_seg = 1;
1180 
1181  else if (unformat (input, "dev default %U", unformat_vlib_cli_sub_input,
1182  &sub_input))
1183  {
1184  error =
1185  dpdk_device_config (conf, (vlib_pci_addr_t) (u32) ~ 1, &sub_input,
1186  1);
1187 
1188  if (error)
1189  return error;
1190  }
1191  else
1192  if (unformat
1193  (input, "dev %U %U", unformat_vlib_pci_addr, &pci_addr,
1194  unformat_vlib_cli_sub_input, &sub_input))
1195  {
1196  error = dpdk_device_config (conf, pci_addr, &sub_input, 0);
1197 
1198  if (error)
1199  return error;
1200 
1201  num_whitelisted++;
1202  }
1203  else if (unformat (input, "dev %U", unformat_vlib_pci_addr, &pci_addr))
1204  {
1205  error = dpdk_device_config (conf, pci_addr, 0, 0);
1206 
1207  if (error)
1208  return error;
1209 
1210  num_whitelisted++;
1211  }
1212  else if (unformat (input, "num-mem-channels %d", &conf->nchannels))
1213  conf->nchannels_set_manually = 0;
1214  else if (unformat (input, "num-mbufs %d", &conf->num_mbufs))
1215  ;
1216  else if (unformat (input, "uio-driver %s", &conf->uio_driver_name))
1217  ;
1218  else if (unformat (input, "socket-mem %s", &socket_mem))
1219  ;
1220  else if (unformat (input, "no-pci"))
1221  {
1222  no_pci = 1;
1223  tmp = format (0, "--no-pci%c", 0);
1224  vec_add1 (conf->eal_init_args, tmp);
1225  }
1226  else if (unformat (input, "blacklist %x:%x", &vendor, &device))
1227  {
1228  u32 blacklist_entry;
1229  if (vendor > 0xFFFF)
1230  return clib_error_return (0, "blacklist PCI vendor out of range");
1231  if (device > 0xFFFF)
1232  return clib_error_return (0, "blacklist PCI device out of range");
1233  blacklist_entry = (vendor << 16) | (device & 0xffff);
1235  blacklist_entry);
1236  }
1237  else if (unformat (input, "no-vmbus"))
1238  {
1239  no_vmbus = 1;
1240  tmp = format (0, "--no-vmbus%c", 0);
1241  vec_add1 (conf->eal_init_args, tmp);
1242  }
1243 
1244 #define _(a) \
1245  else if (unformat(input, #a)) \
1246  { \
1247  tmp = format (0, "--%s%c", #a, 0); \
1248  vec_add1 (conf->eal_init_args, tmp); \
1249  }
1251 #undef _
1252 #define _(a) \
1253  else if (unformat(input, #a " %s", &s)) \
1254  { \
1255  if (!strncmp(#a, "file-prefix", 11)) \
1256  file_prefix = 1; \
1257  tmp = format (0, "--%s%c", #a, 0); \
1258  vec_add1 (conf->eal_init_args, tmp); \
1259  vec_add1 (s, 0); \
1260  if (!strncmp(#a, "vdev", 4)) \
1261  if (strstr((char*)s, "af_packet")) \
1262  clib_warning ("af_packet obsoleted. Use CLI 'create host-interface'."); \
1263  vec_add1 (conf->eal_init_args, s); \
1264  }
1266 #undef _
1267 #define _(a,b) \
1268  else if (unformat(input, #a " %s", &s)) \
1269  { \
1270  tmp = format (0, "-%s%c", #b, 0); \
1271  vec_add1 (conf->eal_init_args, tmp); \
1272  vec_add1 (s, 0); \
1273  vec_add1 (conf->eal_init_args, s); \
1274  }
1276 #undef _
1277 #define _(a,b) \
1278  else if (unformat(input, #a " %s", &s)) \
1279  { \
1280  tmp = format (0, "-%s%c", #b, 0); \
1281  vec_add1 (conf->eal_init_args, tmp); \
1282  vec_add1 (s, 0); \
1283  vec_add1 (conf->eal_init_args, s); \
1284  conf->a##_set_manually = 1; \
1285  }
1287 #undef _
1288  else if (unformat (input, "default"))
1289  ;
1290 
1291  else if (unformat_skip_white_space (input))
1292  ;
1293  else
1294  {
1295  error = clib_error_return (0, "unknown input `%U'",
1296  format_unformat_error, input);
1297  goto done;
1298  }
1299  }
1300 
1301  if (!conf->uio_driver_name)
1302  conf->uio_driver_name = format (0, "auto%c", 0);
1303 
1304  default_hugepage_sz = clib_mem_get_default_hugepage_size ();
1305 
1306  /* *INDENT-OFF* */
1308  {
1309  clib_error_t *e;
1310  uword n_pages;
1311  /* preallocate at least 16MB of hugepages per socket,
1312  if more is needed it is up to consumer to preallocate more */
1313  n_pages = round_pow2 ((uword) 16 << 20, default_hugepage_sz);
1314  n_pages /= default_hugepage_sz;
1315 
1316  if ((e = clib_sysfs_prealloc_hugepages(x, 0, n_pages)))
1317  clib_error_report (e);
1318  }));
1319  /* *INDENT-ON* */
1320 
1321  if (!file_prefix)
1322  {
1323  tmp = format (0, "--file-prefix%c", 0);
1324  vec_add1 (conf->eal_init_args, tmp);
1325  tmp = format (0, "vpp%c", 0);
1326  vec_add1 (conf->eal_init_args, tmp);
1327  }
1328 
1329  if (error)
1330  return error;
1331 
1332  /* I'll bet that -c and -n must be the first and second args... */
1333  if (!conf->coremask_set_manually)
1334  {
1336  uword *coremask = 0;
1337  int i;
1338 
1339  /* main thread core */
1340  coremask = clib_bitmap_set (coremask, tm->main_lcore, 1);
1341 
1342  for (i = 0; i < vec_len (tm->registrations); i++)
1343  {
1344  tr = tm->registrations[i];
1345  coremask = clib_bitmap_or (coremask, tr->coremask);
1346  }
1347 
1348  vec_insert (conf->eal_init_args, 2, 1);
1349  conf->eal_init_args[1] = (u8 *) "-c";
1350  tmp = format (0, "%U%c", format_bitmap_hex, coremask, 0);
1351  conf->eal_init_args[2] = tmp;
1352  clib_bitmap_free (coremask);
1353  }
1354 
1355  if (!conf->nchannels_set_manually)
1356  {
1357  vec_insert (conf->eal_init_args, 2, 3);
1358  conf->eal_init_args[3] = (u8 *) "-n";
1359  tmp = format (0, "%d", conf->nchannels);
1360  conf->eal_init_args[4] = tmp;
1361  }
1362 
1363  if (no_pci == 0 && geteuid () == 0)
1364  dpdk_bind_devices_to_uio (conf);
1365 
1366  if (no_vmbus == 0 && geteuid () == 0)
1368 
1369 #define _(x) \
1370  if (devconf->x == 0 && conf->default_devconf.x > 0) \
1371  devconf->x = conf->default_devconf.x ;
1372 
1373  /* *INDENT-OFF* */
1374  pool_foreach (devconf, conf->dev_confs, ({
1375 
1376  /* default per-device config items */
1377  foreach_dpdk_device_config_item
1378 
1379  /* add DPDK EAL whitelist/blacklist entry */
1380  if (num_whitelisted > 0 && devconf->is_blacklisted == 0)
1381  {
1382  tmp = format (0, "-w%c", 0);
1383  vec_add1 (conf->eal_init_args, tmp);
1384  tmp = format (0, "%U%c", format_vlib_pci_addr, &devconf->pci_addr, 0);
1385  vec_add1 (conf->eal_init_args, tmp);
1386  }
1387  else if (num_whitelisted == 0 && devconf->is_blacklisted != 0)
1388  {
1389  tmp = format (0, "-b%c", 0);
1390  vec_add1 (conf->eal_init_args, tmp);
1391  tmp = format (0, "%U%c", format_vlib_pci_addr, &devconf->pci_addr, 0);
1392  vec_add1 (conf->eal_init_args, tmp);
1393  }
1394  }));
1395  /* *INDENT-ON* */
1396 
1397 #undef _
1398 
1399  /* set master-lcore */
1400  tmp = format (0, "--master-lcore%c", 0);
1401  vec_add1 (conf->eal_init_args, tmp);
1402  tmp = format (0, "%u%c", tm->main_lcore, 0);
1403  vec_add1 (conf->eal_init_args, tmp);
1404 
1405 
1406  if (socket_mem)
1407  clib_warning ("socket-mem argument is deprecated");
1408 
1409  /* NULL terminate the "argv" vector, in case of stupidity */
1410  vec_add1 (conf->eal_init_args, 0);
1411  _vec_len (conf->eal_init_args) -= 1;
1412 
1413  /* Set up DPDK eal and packet mbuf pool early. */
1414 
1415  rte_log_set_global_level (log_level);
1416  int log_fds[2] = { 0 };
1417  if (pipe (log_fds) == 0)
1418  {
1419  if (fcntl (log_fds[1], F_SETFL, O_NONBLOCK) == 0)
1420  {
1421  FILE *f = fdopen (log_fds[1], "a");
1422  if (f && rte_openlog_stream (f) == 0)
1423  {
1424  clib_file_t t = { 0 };
1426  t.file_descriptor = log_fds[0];
1427  t.description = format (0, "DPDK logging pipe");
1428  clib_file_add (&file_main, &t);
1429  }
1430  }
1431  else
1432  {
1433  close (log_fds[0]);
1434  close (log_fds[1]);
1435  }
1436  }
1437 
1438  vm = vlib_get_main ();
1439 
1440  /* make copy of args as rte_eal_init tends to mess up with arg array */
1441  for (i = 1; i < vec_len (conf->eal_init_args); i++)
1442  conf->eal_init_args_str = format (conf->eal_init_args_str, "%s ",
1443  conf->eal_init_args[i]);
1444 
1445  dpdk_log_warn ("EAL init args: %s", conf->eal_init_args_str);
1446  ret = rte_eal_init (vec_len (conf->eal_init_args),
1447  (char **) conf->eal_init_args);
1448 
1449  /* lazy umount hugepages */
1450  umount2 ((char *) huge_dir_path, MNT_DETACH);
1451  rmdir ((char *) huge_dir_path);
1452  vec_free (huge_dir_path);
1453 
1454  if (ret < 0)
1455  return clib_error_return (0, "rte_eal_init returned %d", ret);
1456 
1457  /* set custom ring memory allocator */
1458  {
1459  struct rte_mempool_ops *ops = NULL;
1460 
1461  ops = get_ops_by_name ("ring_sp_sc");
1462  ops->alloc = dpdk_ring_alloc;
1463 
1464  ops = get_ops_by_name ("ring_mp_sc");
1465  ops->alloc = dpdk_ring_alloc;
1466 
1467  ops = get_ops_by_name ("ring_sp_mc");
1468  ops->alloc = dpdk_ring_alloc;
1469 
1470  ops = get_ops_by_name ("ring_mp_mc");
1471  ops->alloc = dpdk_ring_alloc;
1472  }
1473 
1474  /* main thread 1st */
1475  error = dpdk_buffer_pool_create (vm, conf->num_mbufs, rte_socket_id ());
1476  if (error)
1477  return error;
1478 
1479  for (i = 0; i < RTE_MAX_LCORE; i++)
1480  {
1481  error = dpdk_buffer_pool_create (vm, conf->num_mbufs,
1482  rte_lcore_to_socket_id (i));
1483  if (error)
1484  return error;
1485  }
1486 
1487 done:
1488  return error;
1489 }
1490 
1492 
1493 void
1495 {
1496  vnet_main_t *vnm = vnet_get_main ();
1497  struct rte_eth_link prev_link = xd->link;
1498  u32 hw_flags = 0;
1499  u8 hw_flags_chg = 0;
1500 
1501  /* only update link state for PMD interfaces */
1502  if ((xd->flags & DPDK_DEVICE_FLAG_PMD) == 0)
1503  return;
1504 
1505  xd->time_last_link_update = now ? now : xd->time_last_link_update;
1506  clib_memset (&xd->link, 0, sizeof (xd->link));
1507  rte_eth_link_get_nowait (xd->port_id, &xd->link);
1508 
1509  if (LINK_STATE_ELOGS)
1510  {
1512  ELOG_TYPE_DECLARE (e) =
1513  {
1514  .format =
1515  "update-link-state: sw_if_index %d, admin_up %d,"
1516  "old link_state %d new link_state %d",.format_args = "i4i1i1i1",};
1517 
1518  struct
1519  {
1520  u32 sw_if_index;
1521  u8 admin_up;
1522  u8 old_link_state;
1523  u8 new_link_state;
1524  } *ed;
1525  ed = ELOG_DATA (&vm->elog_main, e);
1526  ed->sw_if_index = xd->sw_if_index;
1527  ed->admin_up = (xd->flags & DPDK_DEVICE_FLAG_ADMIN_UP) != 0;
1528  ed->old_link_state = (u8)
1530  ed->new_link_state = (u8) xd->link.link_status;
1531  }
1532 
1533  if ((xd->flags & (DPDK_DEVICE_FLAG_ADMIN_UP | DPDK_DEVICE_FLAG_BOND_SLAVE))
1534  && ((xd->link.link_status != 0) ^
1536  {
1537  hw_flags_chg = 1;
1538  hw_flags |= (xd->link.link_status ? VNET_HW_INTERFACE_FLAG_LINK_UP : 0);
1539  }
1540 
1541  if (hw_flags_chg || (xd->link.link_duplex != prev_link.link_duplex))
1542  {
1543  hw_flags_chg = 1;
1544  switch (xd->link.link_duplex)
1545  {
1546  case ETH_LINK_HALF_DUPLEX:
1548  break;
1549  case ETH_LINK_FULL_DUPLEX:
1551  break;
1552  default:
1553  break;
1554  }
1555  }
1556  if (xd->link.link_speed != prev_link.link_speed)
1558  xd->link.link_speed * 1000);
1559 
1560  if (hw_flags_chg)
1561  {
1562  if (LINK_STATE_ELOGS)
1563  {
1565 
1566  ELOG_TYPE_DECLARE (e) =
1567  {
1568  .format =
1569  "update-link-state: sw_if_index %d, new flags %d",.format_args
1570  = "i4i4",};
1571 
1572  struct
1573  {
1574  u32 sw_if_index;
1575  u32 flags;
1576  } *ed;
1577  ed = ELOG_DATA (&vm->elog_main, e);
1578  ed->sw_if_index = xd->sw_if_index;
1579  ed->flags = hw_flags;
1580  }
1581  vnet_hw_interface_set_flags (vnm, xd->hw_if_index, hw_flags);
1582  }
1583 }
1584 
1585 static uword
1587 {
1588  clib_error_t *error;
1589  vnet_main_t *vnm = vnet_get_main ();
1590  dpdk_main_t *dm = &dpdk_main;
1592  dpdk_device_t *xd;
1594  int i;
1595  int j;
1596 
1597  error = dpdk_lib_init (dm);
1598 
1599  if (error)
1600  clib_error_report (error);
1601 
1602  tm->worker_thread_release = 1;
1603 
1604  f64 now = vlib_time_now (vm);
1605  vec_foreach (xd, dm->devices)
1606  {
1607  dpdk_update_link_state (xd, now);
1608  }
1609 
1610  {
1611  /*
1612  * Extra set up for bond interfaces:
1613  * 1. Setup MACs for bond interfaces and their slave links which was set
1614  * in dpdk_device_setup() but needs to be done again here to take
1615  * effect.
1616  * 2. Set up info and register slave link state change callback handling.
1617  * 3. Set up info for bond interface related CLI support.
1618  */
1619  int nports = rte_eth_dev_count_avail ();
1620  if (nports > 0)
1621  {
1622  /* *INDENT-OFF* */
1623  RTE_ETH_FOREACH_DEV(i)
1624  {
1625  xd = NULL;
1626  for (j = 0; j < nports; j++)
1627  {
1628  if (dm->devices[j].port_id == i)
1629  {
1630  xd = &dm->devices[j];
1631  }
1632  }
1633  if (xd != NULL && xd->pmd == VNET_DPDK_PMD_BOND)
1634  {
1635  u8 addr[6];
1636  dpdk_portid_t slink[16];
1637  int nlink = rte_eth_bond_slaves_get (i, slink, 16);
1638  if (nlink > 0)
1639  {
1640  vnet_hw_interface_t *bhi;
1641  ethernet_interface_t *bei;
1642  int rv;
1643 
1644  /* Get MAC of 1st slave link */
1645  rte_eth_macaddr_get
1646  (slink[0], (struct ether_addr *) addr);
1647 
1648  /* Set MAC of bounded interface to that of 1st slave link */
1649  dpdk_log_info ("Set MAC for bond port %d BondEthernet%d",
1650  i, xd->bond_instance_num);
1651  rv = rte_eth_bond_mac_address_set
1652  (i, (struct ether_addr *) addr);
1653  if (rv)
1654  dpdk_log_warn ("Set MAC addr failure rv=%d", rv);
1655 
1656  /* Populate MAC of bonded interface in VPP hw tables */
1657  bhi = vnet_get_hw_interface
1658  (vnm, dm->devices[i].hw_if_index);
1659  bei = pool_elt_at_index
1660  (em->interfaces, bhi->hw_instance);
1661  clib_memcpy (bhi->hw_address, addr, 6);
1662  clib_memcpy (bei->address, addr, 6);
1663 
1664  /* Init l3 packet size allowed on bonded interface */
1666  while (nlink >= 1)
1667  { /* for all slave links */
1668  int slave = slink[--nlink];
1669  dpdk_device_t *sdev = &dm->devices[slave];
1670  vnet_hw_interface_t *shi;
1671  vnet_sw_interface_t *ssi;
1672  ethernet_interface_t *sei;
1673  /* Add MAC to all slave links except the first one */
1674  if (nlink)
1675  {
1676  dpdk_log_info ("Add MAC for slave port %d",
1677  slave);
1678  rv = rte_eth_dev_mac_addr_add
1679  (slave, (struct ether_addr *) addr, 0);
1680  if (rv)
1681  dpdk_log_warn ("Add MAC addr failure rv=%d",
1682  rv);
1683  }
1684  /* Setup slave link state change callback handling */
1685  rte_eth_dev_callback_register
1686  (slave, RTE_ETH_EVENT_INTR_LSC,
1688  dpdk_device_t *sxd = &dm->devices[slave];
1689  sxd->flags |= DPDK_DEVICE_FLAG_BOND_SLAVE;
1690  sxd->bond_port = i;
1691  /* Set slaves bitmap for bonded interface */
1692  bhi->bond_info = clib_bitmap_set
1693  (bhi->bond_info, sdev->hw_if_index, 1);
1694  /* Set MACs and slave link flags on slave interface */
1695  shi = vnet_get_hw_interface (vnm, sdev->hw_if_index);
1696  ssi = vnet_get_sw_interface (vnm, sdev->sw_if_index);
1697  sei = pool_elt_at_index
1698  (em->interfaces, shi->hw_instance);
1701  clib_memcpy (shi->hw_address, addr, 6);
1702  clib_memcpy (sei->address, addr, 6);
1703  /* Set l3 packet size allowed as the lowest of slave */
1704  if (bhi->max_packet_bytes > shi->max_packet_bytes)
1705  bhi->max_packet_bytes = shi->max_packet_bytes;
1706  }
1707  }
1708  }
1709  }
1710  /* *INDENT-ON* */
1711  }
1712  }
1713 
1714  while (1)
1715  {
1716  /*
1717  * check each time through the loop in case intervals are changed
1718  */
1719  f64 min_wait = dm->link_state_poll_interval < dm->stat_poll_interval ?
1721 
1722  vlib_process_wait_for_event_or_clock (vm, min_wait);
1723 
1724  if (dm->admin_up_down_in_progress)
1725  /* skip the poll if an admin up down is in progress (on any interface) */
1726  continue;
1727 
1728  vec_foreach (xd, dm->devices)
1729  {
1730  f64 now = vlib_time_now (vm);
1731  if ((now - xd->time_last_stats_update) >= dm->stat_poll_interval)
1732  dpdk_update_counters (xd, now);
1733  if ((now - xd->time_last_link_update) >= dm->link_state_poll_interval)
1734  dpdk_update_link_state (xd, now);
1735 
1736  }
1737  }
1738 
1739  return 0;
1740 }
1741 
1742 /* *INDENT-OFF* */
1744  .function = dpdk_process,
1745  .type = VLIB_NODE_TYPE_PROCESS,
1746  .name = "dpdk-process",
1747  .process_log2_n_stack_bytes = 17,
1748 };
1749 /* *INDENT-ON* */
1750 
1751 static clib_error_t *
1753 {
1754  dpdk_main_t *dm = &dpdk_main;
1755  clib_error_t *error = 0;
1756 
1757  /* verify that structs are cacheline aligned */
1758  STATIC_ASSERT (offsetof (dpdk_device_t, cacheline0) == 0,
1759  "Cache line marker must be 1st element in dpdk_device_t");
1760  STATIC_ASSERT (offsetof (dpdk_device_t, cacheline1) ==
1762  "Data in cache line 0 is bigger than cache line size");
1763  STATIC_ASSERT (offsetof (frame_queue_trace_t, cacheline0) == 0,
1764  "Cache line marker must be 1st element in frame_queue_trace_t");
1765  STATIC_ASSERT (RTE_CACHE_LINE_SIZE == 1 << CLIB_LOG2_CACHE_LINE_BYTES,
1766  "DPDK RTE CACHE LINE SIZE does not match with 1<<CLIB_LOG2_CACHE_LINE_BYTES");
1767 
1768  dm->vlib_main = vm;
1769  dm->vnet_main = vnet_get_main ();
1770  dm->conf = &dpdk_config_main;
1771 
1772  dm->conf->nchannels = 4;
1773  dm->conf->num_mbufs = dm->conf->num_mbufs ? dm->conf->num_mbufs : NB_MBUF;
1774  vec_add1 (dm->conf->eal_init_args, (u8 *) "vnet");
1775  vec_add1 (dm->conf->eal_init_args, (u8 *) "--in-memory");
1776 
1777  /* Default vlib_buffer_t flags, DISABLES tcp/udp checksumming... */
1778  dm->buffer_flags_template = (VLIB_BUFFER_TOTAL_LENGTH_VALID |
1779  VLIB_BUFFER_EXT_HDR_VALID |
1780  VNET_BUFFER_F_L4_CHECKSUM_COMPUTED |
1781  VNET_BUFFER_F_L4_CHECKSUM_CORRECT);
1782 
1785 
1786  /* init CLI */
1787  if ((error = vlib_call_init_function (vm, dpdk_cli_init)))
1788  return error;
1789 
1790  dm->log_default = vlib_log_register_class ("dpdk", 0);
1791 
1792  return error;
1793 }
1794 
1796 
1797 
1798 /*
1799  * fd.io coding-style-patch-verification: ON
1800  *
1801  * Local Variables:
1802  * eval: (c-set-style "gnu")
1803  * End:
1804  */
vlib_log_class_t vlib_log_register_class(char *class, char *subclass)
Definition: log.c:227
#define vec_validate(V, I)
Make sure vector is long enough for given index (no header, unspecified alignment) ...
Definition: vec.h:439
static void dpdk_bind_devices_to_uio(dpdk_config_main_t *conf)
Definition: init.c:824
f64 time_last_link_update
Definition: dpdk.h:257
vmrglw vmrglh hi
format_function_t format_vlib_pci_addr
Definition: pci.h:321
#define hash_set(h, key, value)
Definition: hash.h:255
#define VIRTIO_PCI_MODERN_DEVICEID_NET
Definition: pci_config.h:169
u32 flags
Definition: vhost_user.h:115
#define clib_min(x, y)
Definition: clib.h:295
ethernet_main_t ethernet_main
Definition: init.c:45
static f64 vlib_process_wait_for_event_or_clock(vlib_main_t *vm, f64 dt)
Suspend a cooperative multi-tasking thread Waits for an event, or for the indicated number of seconds...
Definition: node_funcs.h:673
u8 interface_name_format_decimal
Definition: dpdk.h:373
clib_error_t * vlib_vmbus_bind_to_uio(vlib_vmbus_addr_t *addr)
Definition: vmbus.c:203
vnet_main_t * vnet_get_main(void)
Definition: misc.c:47
#define NB_MBUF
Definition: dpdk.h:59
Optimized string handling code, including c11-compliant "safe C library" variants.
vnet_device_class_t dpdk_device_class
unsigned long u64
Definition: types.h:89
u32 sw_if_index
Definition: dpdk.h:208
#define DPDK_DEVICE_VLAN_STRIP_OFF
Definition: dpdk.h:340
#define NULL
Definition: clib.h:58
clib_memset(h->entries, 0, sizeof(h->entries[0]) *entries)
static f64 vlib_time_now(vlib_main_t *vm)
Definition: main.h:232
#define vec_add2_aligned(V, P, N, A)
Add N elements to end of vector V, return pointer to new elements in P.
Definition: vec.h:576
static uword * clib_bitmap_or(uword *ai, uword *bi)
Logical operator across two bitmaps.
static u32 dpdk_flag_change(vnet_main_t *vnm, vnet_hw_interface_t *hi, u32 flags)
Definition: init.c:109
struct rte_pci_device * dpdk_get_pci_device(const struct rte_eth_dev_info *info)
Definition: common.c:350
void dpdk_update_link_state(dpdk_device_t *xd, f64 now)
Definition: init.c:1494
u16 flags
Definition: dpdk.h:216
static vnet_hw_interface_t * vnet_get_hw_interface(vnet_main_t *vnm, u32 hw_if_index)
#define LINK_STATE_ELOGS
Definition: init.c:46
#define CLIB_LOG2_CACHE_LINE_BYTES
Definition: cache.h:50
static struct rte_mempool_ops * get_ops_by_name(char *ops_name)
Definition: init.c:154
u32 file_descriptor
Definition: file.h:54
#define vec_add1(V, E)
Add 1 element to end of vector (unspecified alignment).
Definition: vec.h:525
dpdk_device_and_queue_t ** devices_by_hqos_cpu
Definition: dpdk.h:412
#define DPDK_NB_RX_DESC_VIRTIO
Definition: dpdk_priv.h:21
clib_error_t * errors
Definition: dpdk.h:271
#define vec_add2(V, P, N)
Add N elements to end of vector V, return pointer to new elements in P.
Definition: vec.h:564
int i
u32 per_interface_next_index
Definition: dpdk.h:211
u8 enable_tcp_udp_checksum
Definition: dpdk.h:359
static uword * clib_bitmap_set(uword *ai, uword i, uword value)
Sets the ith bit of a bitmap to new_value Removes trailing zeros from the bitmap. ...
Definition: bitmap.h:167
uword unformat_user(unformat_input_t *input, unformat_function_t *func,...)
Definition: unformat.c:983
#define DPDK_DEVICE_VLAN_STRIP_ON
Definition: dpdk.h:341
static vnet_sw_interface_t * vnet_get_sw_interface(vnet_main_t *vnm, u32 sw_if_index)
u8 * format(u8 *s, const char *fmt,...)
Definition: format.c:419
#define DPDK_NB_TX_DESC_DEFAULT
Definition: dpdk_priv.h:20
u32 supported_flow_actions
Definition: dpdk.h:237
#define foreach_eal_double_hyphen_predicate_arg
Definition: dpdk_priv.h:32
unformat_function_t unformat_vlib_pci_addr
Definition: pci.h:320
int dpdk_port_state_callback(dpdk_portid_t port_id, enum rte_eth_event_type type, void *param, void *ret_param)
Definition: common.c:340
u16 bond_instance_num
Definition: dpdk.h:250
#define vec_validate_aligned(V, I, A)
Make sure vector is long enough for given index (no header, specified alignment)
Definition: vec.h:450
dpdk_device_config_hqos_t hqos
Definition: dpdk.h:348
vlib_pci_addr_t * vlib_pci_get_all_dev_addrs()
Definition: pci.c:1409
#define pool_get(P, E)
Allocate an object E from a pool P (unspecified alignment).
Definition: pool.h:236
vhost_vring_addr_t addr
Definition: vhost_user.h:121
u32 * blacklist_by_pci_vendor_and_device
Definition: dpdk.h:381
unsigned char u8
Definition: types.h:56
dpdk_config_main_t dpdk_config_main
Definition: init.c:44
#define vec_reset_length(v)
Reset vector length to zero NULL-pointer tolerant.
clib_file_function_t * read_function
Definition: file.h:67
#define ETHER_MAX_LEN
Maximum frame len, including CRC.
Definition: init.c:41
double f64
Definition: types.h:142
static vnet_sw_interface_t * vnet_get_hw_sw_interface(vnet_main_t *vnm, u32 hw_if_index)
#define clib_memcpy(d, s, n)
Definition: string.h:180
foreach_dpdk_device_config_item clib_bitmap_t * workers
Definition: dpdk.h:346
#define dpdk_log_warn(...)
Definition: dpdk.h:506
#define pool_foreach(VAR, POOL, BODY)
Iterate through pool.
Definition: pool.h:490
static clib_error_t * dpdk_log_read_ready(clib_file_t *uf)
Definition: init.c:1101
#define VLIB_INIT_FUNCTION(x)
Definition: init.h:163
dpdk_portid_t port_id
Definition: dpdk.h:205
vlib_node_registration_t dpdk_input_node
(constructor) VLIB_REGISTER_NODE (dpdk_input_node)
Definition: node.c:519
u32 sw_if_index
Definition: vxlan_gbp.api:37
dpdk_device_config_t default_devconf
Definition: dpdk.h:376
f64 stat_poll_interval
Definition: dpdk.h:439
static char * vlib_unix_get_runtime_dir(void)
Definition: unix.h:139
static dpdk_port_type_t port_type_from_speed_capa(struct rte_eth_dev_info *dev_info)
Definition: init.c:51
vnet_hw_interface_flags_t flags
Definition: interface.h:516
#define vec_elt_at_index(v, i)
Get vector value at index i checking that i is in bounds.
#define clib_error_return(e, args...)
Definition: error.h:99
u16 rx_q_used
Definition: dpdk.h:229
clib_file_main_t file_main
Definition: main.c:63
#define vec_resize(V, N)
Resize a vector (no header, unspecified alignment) Add N elements to end of given vector V...
Definition: vec.h:242
unsigned int u32
Definition: types.h:88
void dpdk_device_setup(dpdk_device_t *xd)
Definition: common.c:40
#define vlib_call_init_function(vm, x)
Definition: init.h:260
clib_error_t * unformat_rss_fn(unformat_input_t *input, uword *rss_fn)
Definition: format.c:902
#define DPDK_NB_TX_DESC_VIRTIO
Definition: dpdk_priv.h:22
struct rte_eth_conf port_conf
Definition: dpdk.h:233
static clib_error_t * dpdk_init(vlib_main_t *vm)
Definition: init.c:1752
#define fl(x, y)
u32 max_supported_packet_bytes
Definition: interface.h:555
f64 time_last_stats_update
Definition: dpdk.h:264
struct rte_eth_txconf tx_conf
Definition: dpdk.h:234
u8 * description
Definition: file.h:70
#define hash_get(h, key)
Definition: hash.h:249
#define clib_bitmap_foreach(i, ai, body)
Macro to iterate across set bits in a bitmap.
Definition: bitmap.h:361
#define pool_elt_at_index(p, i)
Returns pointer to element at given index.
Definition: pool.h:511
#define vec_insert(V, N, M)
Insert N vector elements starting at element M, initialize new elements to zero (no header...
Definition: vec.h:689
vlib_pci_addr_t pci_addr
Definition: dpdk.h:335
clib_error_t * dpdk_buffer_pool_create(vlib_main_t *vm, unsigned num_mbufs, unsigned socket_id)
Definition: buffer.c:444
static clib_error_t * dpdk_config(vlib_main_t *vm, unformat_input_t *input)
Definition: init.c:1132
#define foreach_eal_double_hyphen_arg
Definition: dpdk_priv.h:48
dpdk_portid_t bond_port
Definition: dpdk.h:254
#define ETHERNET_INTERFACE_FLAG_CONFIG_MTU(flags)
Definition: ethernet.h:180
u8 ** eal_init_args
Definition: dpdk.h:355
dpdk_per_thread_data_t * per_thread_data
Definition: dpdk.h:413
#define foreach_eal_single_hyphen_mandatory_arg
Definition: dpdk_priv.h:38
static dpdk_port_type_t port_type_from_link_speed(u32 link_speed)
Definition: init.c:79
struct _unformat_input_t unformat_input_t
unsigned short u16
Definition: types.h:57
#define clib_error_return_unix(e, args...)
Definition: error.h:102
#define foreach_dpdk_pmd
Definition: dpdk.h:65
vlib_buffer_t buffer_template
Definition: dpdk.h:397
#define VLIB_BUFFER_DEFAULT_FREE_LIST_INDEX
Definition: buffer.h:442
#define ELOG_DATA(em, f)
Definition: elog.h:484
#define VIRTIO_PCI_LEGACY_DEVICEID_NET
Definition: pci_config.h:168
dpdk_port_type_t port_type
Definition: dpdk.h:265
#define VLIB_CONFIG_FUNCTION(x, n,...)
Definition: init.h:172
vnet_sw_interface_flags_t flags
Definition: interface.h:706
signed char i8
Definition: types.h:45
u16 tx_q_used
Definition: dpdk.h:228
u16 nb_rx_desc
Definition: dpdk.h:230
uint16_t dpdk_portid_t
Definition: dpdk.h:124
#define dpdk_log_info(...)
Definition: dpdk.h:510
u32 hw_if_index
Definition: dpdk.h:207
u8 len
Definition: ip_types.api:49
void unformat_init_vector(unformat_input_t *input, u8 *vector_string)
Definition: unformat.c:1031
#define VNET_HW_INTERFACE_BOND_INFO_SLAVE
Definition: interface.h:575
u16 af_packet_instance_num
Definition: dpdk.h:249
#define foreach_eal_single_hyphen_arg
Definition: dpdk_priv.h:42
#define VLIB_REGISTER_NODE(x,...)
Definition: node.h:169
#define DPDK_NB_RX_DESC_DEFAULT
Definition: dpdk_priv.h:19
#define UNFORMAT_END_OF_INPUT
Definition: format.h:144
static clib_error_t * dpdk_device_config(dpdk_config_main_t *conf, vlib_pci_addr_t pci_addr, unformat_input_t *input, u8 is_default)
Definition: init.c:999
dpdk_device_t * devices
Definition: dpdk.h:411
vlib_main_t * vm
Definition: buffer.c:301
static void dpdk_update_counters(dpdk_device_t *xd, f64 now)
Definition: dpdk_priv.h:88
u8 nchannels_set_manually
Definition: dpdk.h:364
#define vec_free(V)
Free vector&#39;s memory (no header).
Definition: vec.h:341
#define dpdk_log_err(...)
Definition: dpdk.h:504
volatile u32 ** lockp
Definition: dpdk.h:199
dpdk_device_config_t * dev_confs
Definition: dpdk.h:377
#define clib_warning(format, args...)
Definition: error.h:59
dpdk_pmd_t pmd
Definition: dpdk.h:213
format_function_t format_dpdk_device_errors
Definition: dpdk.h:517
elog_main_t elog_main
Definition: main.h:157
#define ETHERNET_INTERFACE_FLAG_ACCEPT_ALL
Definition: ethernet.h:174
u8 coremask_set_manually
Definition: dpdk.h:363
#define ELOG_TYPE_DECLARE(f)
Definition: elog.h:442
static void dpdk_device_lock_init(dpdk_device_t *xd)
Definition: init.c:141
format_function_t format_dpdk_rss_hf_name
Definition: dpdk.h:523
static void dpdk_bind_vmbus_devices_to_uio(dpdk_config_main_t *conf)
Definition: init.c:978
u8 * interface_name_suffix
Definition: dpdk.h:222
signed int i32
Definition: types.h:77
#define hash_create(elts, value_bytes)
Definition: hash.h:696
#define ASSERT(truth)
void dpdk_device_config_hqos_default(dpdk_device_config_hqos_t *hqos)
Definition: hqos.c:205
format_function_t format_dpdk_device_name
Definition: dpdk.h:515
int hqos_cpu_count
Definition: dpdk.h:435
void vnet_hw_interface_assign_rx_thread(vnet_main_t *vnm, u32 hw_if_index, u16 queue_id, uword thread_index)
Definition: devices.c:139
vlib_pci_device_info_t * vlib_pci_get_device_info(vlib_main_t *vm, vlib_pci_addr_t *addr, clib_error_t **error)
Definition: pci.c:184
static uword clib_file_add(clib_file_main_t *um, clib_file_t *template)
Definition: file.h:96
vlib_log_class_t log_default
Definition: dpdk.h:453
static uword dpdk_process(vlib_main_t *vm, vlib_node_runtime_t *rt, vlib_frame_t *f)
Definition: init.c:1586
Bitmaps built as vectors of machine words.
clib_error_t * ethernet_register_interface(vnet_main_t *vnm, u32 dev_class_index, u32 dev_instance, u8 *address, u32 *hw_if_index_return, ethernet_flag_change_function_t flag_change)
Definition: interface.c:277
#define clib_error_report(e)
Definition: error.h:113
#define clib_bitmap_free(v)
Free a bitmap.
Definition: bitmap.h:92
uword clib_mem_get_default_hugepage_size(void)
Definition: mem.c:57
#define DPDK_LINK_POLL_INTERVAL
Definition: dpdk.h:277
size_t count
Definition: vapi.c:47
dpdk_main_t dpdk_main
Definition: init.c:43
uword * thread_registrations_by_name
Definition: threads.h:288
clib_error_t * dpdk_cli_init(vlib_main_t *vm)
Definition: cli.c:2074
dpdk_portid_t device_index
Definition: dpdk.h:202
struct rte_eth_link link
Definition: dpdk.h:256
u8 * name
Definition: dpdk.h:221
static vlib_main_t * vlib_get_main(void)
Definition: global_funcs.h:23
clib_error_t * dpdk_port_setup_hqos(dpdk_device_t *xd, dpdk_device_config_hqos_t *hqos)
Definition: hqos.c:247
dpdk_port_type_t
Definition: dpdk.h:101
static int dpdk_port_crc_strip_enabled(dpdk_device_t *xd)
Definition: init.c:208
static uword clib_bitmap_count_set_bits(uword *ai)
Return the number of set bits in a bitmap.
Definition: bitmap.h:462
Definition: defs.h:47
#define DPDK_STATS_POLL_INTERVAL
Definition: dpdk.h:274
vlib_vmbus_addr_t * vlib_vmbus_get_all_dev_addrs()
Definition: vmbus.c:368
unformat_function_t unformat_line
Definition: format.h:279
static vlib_node_registration_t dpdk_process_node
(constructor) VLIB_REGISTER_NODE (dpdk_process_node)
Definition: init.c:1743
#define vec_len(v)
Number of elements in vector (rvalue-only, NULL tolerant)
uword unformat_vlib_cli_sub_input(unformat_input_t *i, va_list *args)
Definition: cli.c:158
u8 admin_up_down_in_progress
Definition: dpdk.h:431
#define STATIC_ASSERT(truth,...)
clib_error_t * vnet_hw_interface_set_flags(vnet_main_t *vnm, u32 hw_if_index, vnet_hw_interface_flags_t flags)
Definition: interface.c:504
u8 no_tx_checksum_offload
Definition: dpdk.h:360
u64 uword
Definition: types.h:112
static void unformat_free(unformat_input_t *i)
Definition: format.h:162
static clib_error_t * dpdk_lib_init(dpdk_main_t *dm)
Definition: init.c:218
#define hash_get_mem(h, key)
Definition: hash.h:269
u32 buffer_flags_template
Definition: dpdk.h:416
static void * clib_mem_alloc_aligned(uword size, uword align)
Definition: mem.h:140
static void vlib_buffer_init_for_free_list(vlib_buffer_t *dst, vlib_buffer_free_list_t *fl)
#define vnet_buffer(b)
Definition: buffer.h:368
static u32 random_u32(u32 *seed)
32-bit random number generator
Definition: random.h:69
u8 * format_unformat_error(u8 *s, va_list *va)
Definition: unformat.c:91
int hqos_cpu_first_index
Definition: dpdk.h:434
static vlib_thread_main_t * vlib_get_thread_main()
Definition: global_funcs.h:32
#define ETHERNET_MAX_PACKET_BYTES
Definition: ethernet.h:166
#define vec_foreach(var, vec)
Vector iterator.
i8 cpu_socket
Definition: dpdk.h:214
#define ETHERNET_INTERFACE_FLAG_CONFIG_PROMISC(flags)
Definition: ethernet.h:175
uword * cpu_socket_bitmap
Definition: threads.h:323
Definition: file.h:51
static int dpdk_ring_alloc(struct rte_mempool *mp)
Definition: init.c:168
static vlib_buffer_free_list_t * vlib_buffer_get_free_list(vlib_main_t *vm, vlib_buffer_free_list_index_t free_list_index)
Definition: buffer_funcs.h:657
clib_error_t * vlib_pci_bind_to_uio(vlib_main_t *vm, vlib_pci_addr_t *addr, char *uio_drv_name)
Definition: pci.c:368
u8 * uio_driver_name
Definition: dpdk.h:357
vlib_thread_registration_t ** registrations
Definition: threads.h:286
static void vlib_pci_free_device_info(vlib_pci_device_info_t *di)
Definition: pci.h:111
#define CLIB_CACHE_LINE_BYTES
Definition: cache.h:59
unformat_function_t unformat_dpdk_log_level
Definition: dpdk.h:526
ethernet_interface_t * interfaces
Definition: ethernet.h:305
u32 flags
buffer flags: VLIB_BUFFER_FREE_LIST_INDEX_MASK: bits used to store free list index, VLIB_BUFFER_IS_TRACED: trace this buffer.
Definition: buffer.h:117
vnet_main_t * vnet_main
Definition: dpdk.h:443
u16 nb_tx_desc
Definition: dpdk.h:218
clib_error_t * unformat_hqos(unformat_input_t *input, dpdk_device_config_hqos_t *hqos)
Definition: format.c:939
u16 device_class
Definition: pci.h:72
uword * device_config_index_by_pci_addr
Definition: dpdk.h:378
uword unformat_skip_white_space(unformat_input_t *input)
Definition: unformat.c:815
static uword vnet_hw_interface_is_link_up(vnet_main_t *vnm, u32 hw_if_index)
volatile u32 worker_thread_release
Definition: threads.h:329
#define dpdk_log_notice(...)
Definition: dpdk.h:508
static void vnet_hw_interface_set_link_speed(vnet_main_t *vnm, u32 hw_if_index, u32 link_speed)
static void vnet_hw_interface_set_input_node(vnet_main_t *vnm, u32 hw_if_index, u32 node_index)
Definition: devices.h:79
vnet_device_main_t vnet_device_main
Definition: devices.c:22
uword unformat(unformat_input_t *i, const char *fmt,...)
Definition: unformat.c:972
f64 link_state_poll_interval
Definition: dpdk.h:438
CLIB vectors are ubiquitous dynamically resized arrays with by user defined "headers".
static uword unformat_check_input(unformat_input_t *i)
Definition: format.h:170
dpdk_config_main_t * conf
Definition: dpdk.h:444
vlib_main_t * vlib_main
Definition: dpdk.h:442