init.c revision 7fd40216
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/buffer.h>
26#include <dpdk/device/dpdk.h>
27#include <vlib/pci/pci.h>
28#include <vlib/vmbus/vmbus.h>
29
30#include <rte_ring.h>
31
32#include <stdio.h>
33#include <stdlib.h>
34#include <unistd.h>
35#include <sys/stat.h>
36#include <sys/mount.h>
37#include <string.h>
38#include <fcntl.h>
39#include <dirent.h>
40
41#include <dpdk/device/dpdk_priv.h>
42
43#define ETHER_MAX_LEN   1518  /**< Maximum frame len, including CRC. */
44
45dpdk_main_t dpdk_main;
46dpdk_config_main_t dpdk_config_main;
47
48#define LINK_STATE_ELOGS	0
49
50/* Port configuration, mildly modified Intel app values */
51
52static dpdk_port_type_t
53port_type_from_speed_capa (struct rte_eth_dev_info *dev_info)
54{
55
56  if (dev_info->speed_capa & ETH_LINK_SPEED_100G)
57    return VNET_DPDK_PORT_TYPE_ETH_100G;
58  else if (dev_info->speed_capa & ETH_LINK_SPEED_56G)
59    return VNET_DPDK_PORT_TYPE_ETH_56G;
60  else if (dev_info->speed_capa & ETH_LINK_SPEED_50G)
61    return VNET_DPDK_PORT_TYPE_ETH_50G;
62  else if (dev_info->speed_capa & ETH_LINK_SPEED_40G)
63    return VNET_DPDK_PORT_TYPE_ETH_40G;
64  else if (dev_info->speed_capa & ETH_LINK_SPEED_25G)
65    return VNET_DPDK_PORT_TYPE_ETH_25G;
66  else if (dev_info->speed_capa & ETH_LINK_SPEED_20G)
67    return VNET_DPDK_PORT_TYPE_ETH_20G;
68  else if (dev_info->speed_capa & ETH_LINK_SPEED_10G)
69    return VNET_DPDK_PORT_TYPE_ETH_10G;
70  else if (dev_info->speed_capa & ETH_LINK_SPEED_5G)
71    return VNET_DPDK_PORT_TYPE_ETH_5G;
72  else if (dev_info->speed_capa & ETH_LINK_SPEED_2_5G)
73    return VNET_DPDK_PORT_TYPE_ETH_2_5G;
74  else if (dev_info->speed_capa & ETH_LINK_SPEED_1G)
75    return VNET_DPDK_PORT_TYPE_ETH_1G;
76
77  return VNET_DPDK_PORT_TYPE_UNKNOWN;
78}
79
80static dpdk_port_type_t
81port_type_from_link_speed (u32 link_speed)
82{
83  switch (link_speed)
84    {
85    case ETH_SPEED_NUM_1G:
86      return VNET_DPDK_PORT_TYPE_ETH_1G;
87    case ETH_SPEED_NUM_2_5G:
88      return VNET_DPDK_PORT_TYPE_ETH_2_5G;
89    case ETH_SPEED_NUM_5G:
90      return VNET_DPDK_PORT_TYPE_ETH_5G;
91    case ETH_SPEED_NUM_10G:
92      return VNET_DPDK_PORT_TYPE_ETH_10G;
93    case ETH_SPEED_NUM_20G:
94      return VNET_DPDK_PORT_TYPE_ETH_20G;
95    case ETH_SPEED_NUM_25G:
96      return VNET_DPDK_PORT_TYPE_ETH_25G;
97    case ETH_SPEED_NUM_40G:
98      return VNET_DPDK_PORT_TYPE_ETH_40G;
99    case ETH_SPEED_NUM_50G:
100      return VNET_DPDK_PORT_TYPE_ETH_50G;
101    case ETH_SPEED_NUM_56G:
102      return VNET_DPDK_PORT_TYPE_ETH_56G;
103    case ETH_SPEED_NUM_100G:
104      return VNET_DPDK_PORT_TYPE_ETH_100G;
105    default:
106      return VNET_DPDK_PORT_TYPE_UNKNOWN;
107    }
108}
109
110static u32
111dpdk_flag_change (vnet_main_t * vnm, vnet_hw_interface_t * hi, u32 flags)
112{
113  dpdk_main_t *dm = &dpdk_main;
114  dpdk_device_t *xd = vec_elt_at_index (dm->devices, hi->dev_instance);
115  u32 old = 0;
116
117  if (ETHERNET_INTERFACE_FLAG_CONFIG_PROMISC (flags))
118    {
119      old = (xd->flags & DPDK_DEVICE_FLAG_PROMISC) != 0;
120
121      if (flags & ETHERNET_INTERFACE_FLAG_ACCEPT_ALL)
122	xd->flags |= DPDK_DEVICE_FLAG_PROMISC;
123      else
124	xd->flags &= ~DPDK_DEVICE_FLAG_PROMISC;
125
126      if (xd->flags & DPDK_DEVICE_FLAG_ADMIN_UP)
127	{
128	  if (xd->flags & DPDK_DEVICE_FLAG_PROMISC)
129	    rte_eth_promiscuous_enable (xd->port_id);
130	  else
131	    rte_eth_promiscuous_disable (xd->port_id);
132	}
133    }
134  else if (ETHERNET_INTERFACE_FLAG_CONFIG_MTU (flags))
135    {
136      xd->port_conf.rxmode.max_rx_pkt_len = hi->max_packet_bytes;
137      dpdk_device_setup (xd);
138    }
139  return old;
140}
141
142static void
143dpdk_device_lock_init (dpdk_device_t * xd)
144{
145  int q;
146  vec_validate (xd->lockp, xd->tx_q_used - 1);
147  for (q = 0; q < xd->tx_q_used; q++)
148    {
149      xd->lockp[q] = clib_mem_alloc_aligned (CLIB_CACHE_LINE_BYTES,
150					     CLIB_CACHE_LINE_BYTES);
151      clib_memset ((void *) xd->lockp[q], 0, CLIB_CACHE_LINE_BYTES);
152    }
153}
154
155static int
156dpdk_port_crc_strip_enabled (dpdk_device_t * xd)
157{
158  return !(xd->port_conf.rxmode.offloads & DEV_RX_OFFLOAD_KEEP_CRC);
159}
160
161/* The funciton check_l3cache helps check if Level 3 cache exists or not on current CPUs
162  return value 1: exist.
163  return value 0: not exist.
164*/
165static int
166check_l3cache ()
167{
168
169  struct dirent *dp;
170  clib_error_t *err;
171  const char *sys_cache_dir = "/sys/devices/system/cpu/cpu0/cache";
172  DIR *dir_cache = opendir (sys_cache_dir);
173
174  if (dir_cache == NULL)
175    return -1;
176
177  while ((dp = readdir (dir_cache)) != NULL)
178    {
179      if (dp->d_type == DT_DIR)
180	{
181	  u8 *p = NULL;
182	  int level_cache = -1;
183
184	  p = format (p, "%s/%s/%s", sys_cache_dir, dp->d_name, "level");
185	  if ((err = clib_sysfs_read ((char *) p, "%d", &level_cache)))
186	    clib_error_free (err);
187
188	  if (level_cache == 3)
189	    {
190	      closedir (dir_cache);
191	      return 1;
192	    }
193	}
194    }
195
196  if (dir_cache != NULL)
197    closedir (dir_cache);
198
199  return 0;
200}
201
202static clib_error_t *
203dpdk_lib_init (dpdk_main_t * dm)
204{
205  u32 nports;
206  u32 mtu, max_rx_frame;
207  int i;
208  clib_error_t *error;
209  vlib_main_t *vm = vlib_get_main ();
210  vnet_main_t *vnm = vnet_get_main ();
211  vlib_thread_main_t *tm = vlib_get_thread_main ();
212  vnet_device_main_t *vdm = &vnet_device_main;
213  vnet_sw_interface_t *sw;
214  vnet_hw_interface_t *hi;
215  dpdk_device_t *xd;
216  vlib_pci_addr_t last_pci_addr;
217  u32 last_pci_addr_port = 0;
218  vlib_thread_registration_t *tr_hqos;
219  uword *p_hqos;
220
221  u32 next_hqos_cpu = 0;
222  u8 af_packet_instance_num = 0;
223  last_pci_addr.as_u32 = ~0;
224
225  dm->hqos_cpu_first_index = 0;
226  dm->hqos_cpu_count = 0;
227
228  /* find out which cpus will be used for I/O TX */
229  p_hqos = hash_get_mem (tm->thread_registrations_by_name, "hqos-threads");
230  tr_hqos = p_hqos ? (vlib_thread_registration_t *) p_hqos[0] : 0;
231
232  if (tr_hqos && tr_hqos->count > 0)
233    {
234      dm->hqos_cpu_first_index = tr_hqos->first_index;
235      dm->hqos_cpu_count = tr_hqos->count;
236    }
237
238  vec_validate_aligned (dm->devices_by_hqos_cpu, tm->n_vlib_mains - 1,
239			CLIB_CACHE_LINE_BYTES);
240
241  nports = rte_eth_dev_count_avail ();
242
243  if (nports < 1)
244    {
245      dpdk_log_notice ("DPDK drivers found no ports...");
246    }
247
248  if (CLIB_DEBUG > 0)
249    dpdk_log_notice ("DPDK drivers found %d ports...", nports);
250
251  if (dm->conf->enable_tcp_udp_checksum)
252    dm->buffer_flags_template &= ~(VNET_BUFFER_F_L4_CHECKSUM_CORRECT
253				   | VNET_BUFFER_F_L4_CHECKSUM_COMPUTED);
254
255  /* vlib_buffer_t template */
256  vec_validate_aligned (dm->per_thread_data, tm->n_vlib_mains - 1,
257			CLIB_CACHE_LINE_BYTES);
258  for (i = 0; i < tm->n_vlib_mains; i++)
259    {
260      dpdk_per_thread_data_t *ptd = vec_elt_at_index (dm->per_thread_data, i);
261      clib_memset (&ptd->buffer_template, 0, sizeof (vlib_buffer_t));
262      ptd->buffer_template.flags = dm->buffer_flags_template;
263      vnet_buffer (&ptd->buffer_template)->sw_if_index[VLIB_TX] = (u32) ~ 0;
264    }
265
266  /* *INDENT-OFF* */
267  RTE_ETH_FOREACH_DEV(i)
268    {
269      u8 addr[6];
270      int vlan_off;
271      struct rte_eth_dev_info dev_info;
272      struct rte_pci_device *pci_dev;
273      struct rte_eth_link l;
274      dpdk_portid_t next_port_id;
275      dpdk_device_config_t *devconf = 0;
276      vlib_pci_addr_t pci_addr;
277      uword *p = 0;
278
279      if (!rte_eth_dev_is_valid_port(i))
280	continue;
281
282      rte_eth_link_get_nowait (i, &l);
283      rte_eth_dev_info_get (i, &dev_info);
284
285      if (dev_info.device == 0)
286	{
287	  clib_warning ("DPDK bug: missing device info. Skipping %s device",
288			dev_info.driver_name);
289	  continue;
290	}
291
292      pci_dev = dpdk_get_pci_device (&dev_info);
293
294      if (pci_dev)
295	{
296	  pci_addr.domain = pci_dev->addr.domain;
297	  pci_addr.bus = pci_dev->addr.bus;
298	  pci_addr.slot = pci_dev->addr.devid;
299	  pci_addr.function = pci_dev->addr.function;
300	  p = hash_get (dm->conf->device_config_index_by_pci_addr,
301			pci_addr.as_u32);
302	}
303
304
305      /* Create vnet interface */
306      vec_add2_aligned (dm->devices, xd, 1, CLIB_CACHE_LINE_BYTES);
307      xd->nb_rx_desc = DPDK_NB_RX_DESC_DEFAULT;
308      xd->nb_tx_desc = DPDK_NB_TX_DESC_DEFAULT;
309      xd->cpu_socket = (i8) rte_eth_dev_socket_id (i);
310
311      if (p)
312	{
313	  devconf = pool_elt_at_index (dm->conf->dev_confs, p[0]);
314	  xd->name = devconf->name;
315	}
316      else
317	devconf = &dm->conf->default_devconf;
318
319      /* Handle interface naming for devices with multiple ports sharing same PCI ID */
320      if (pci_dev &&
321	  ((next_port_id = rte_eth_find_next (i + 1)) != RTE_MAX_ETHPORTS))
322	{
323	  struct rte_eth_dev_info di = { 0 };
324	  struct rte_pci_device *next_pci_dev;
325	  rte_eth_dev_info_get (next_port_id, &di);
326	  next_pci_dev = di.device ? RTE_DEV_TO_PCI (di.device) : 0;
327	  if (next_pci_dev &&
328	      pci_addr.as_u32 != last_pci_addr.as_u32 &&
329	      memcmp (&pci_dev->addr, &next_pci_dev->addr,
330		      sizeof (struct rte_pci_addr)) == 0)
331	    {
332	      xd->interface_name_suffix = format (0, "0");
333	      last_pci_addr.as_u32 = pci_addr.as_u32;
334	      last_pci_addr_port = i;
335	    }
336	  else if (pci_addr.as_u32 == last_pci_addr.as_u32)
337	    {
338	      xd->interface_name_suffix =
339		format (0, "%u", i - last_pci_addr_port);
340	    }
341	  else
342	    {
343	      last_pci_addr.as_u32 = ~0;
344	    }
345	}
346      else
347	last_pci_addr.as_u32 = ~0;
348
349      clib_memcpy (&xd->tx_conf, &dev_info.default_txconf,
350		   sizeof (struct rte_eth_txconf));
351
352      if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_IPV4_CKSUM)
353	{
354	  xd->port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_IPV4_CKSUM;
355	  xd->flags |= DPDK_DEVICE_FLAG_RX_IP4_CKSUM;
356	}
357
358      if (dm->conf->no_multi_seg)
359	{
360	  xd->port_conf.txmode.offloads &= ~DEV_TX_OFFLOAD_MULTI_SEGS;
361	  xd->port_conf.rxmode.offloads &= ~DEV_RX_OFFLOAD_JUMBO_FRAME;
362	  xd->port_conf.rxmode.offloads &= ~DEV_RX_OFFLOAD_SCATTER;
363	}
364      else
365	{
366	  xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_MULTI_SEGS;
367	  xd->port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
368	  xd->port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_SCATTER;
369	  xd->flags |= DPDK_DEVICE_FLAG_MAYBE_MULTISEG;
370	}
371
372      xd->tx_q_used = clib_min (dev_info.max_tx_queues, tm->n_vlib_mains);
373
374      if (devconf->num_tx_queues > 0
375	  && devconf->num_tx_queues < xd->tx_q_used)
376	xd->tx_q_used = clib_min (xd->tx_q_used, devconf->num_tx_queues);
377
378      if (devconf->num_rx_queues > 1
379	  && dev_info.max_rx_queues >= devconf->num_rx_queues)
380	{
381	  xd->rx_q_used = devconf->num_rx_queues;
382	  xd->port_conf.rxmode.mq_mode = ETH_MQ_RX_RSS;
383	  if (devconf->rss_fn == 0)
384	    xd->port_conf.rx_adv_conf.rss_conf.rss_hf =
385	      ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP;
386	  else
387	    {
388	      u64 unsupported_bits;
389	      xd->port_conf.rx_adv_conf.rss_conf.rss_hf = devconf->rss_fn;
390	      unsupported_bits = xd->port_conf.rx_adv_conf.rss_conf.rss_hf;
391	      unsupported_bits &= ~dev_info.flow_type_rss_offloads;
392	      if (unsupported_bits)
393		dpdk_log_warn ("Unsupported RSS hash functions: %U",
394			       format_dpdk_rss_hf_name, unsupported_bits);
395	    }
396	  xd->port_conf.rx_adv_conf.rss_conf.rss_hf &=
397	    dev_info.flow_type_rss_offloads;
398	}
399      else
400	xd->rx_q_used = 1;
401
402      xd->flags |= DPDK_DEVICE_FLAG_PMD;
403
404      /* workaround for drivers not setting driver_name */
405      if ((!dev_info.driver_name) && (pci_dev))
406	dev_info.driver_name = pci_dev->driver->driver.name;
407
408      ASSERT (dev_info.driver_name);
409
410      if (!xd->pmd)
411	{
412
413
414#define _(s,f) else if (dev_info.driver_name &&                 \
415                        !strcmp(dev_info.driver_name, s))       \
416                 xd->pmd = VNET_DPDK_PMD_##f;
417	  if (0)
418	    ;
419	  foreach_dpdk_pmd
420#undef _
421	    else
422	    xd->pmd = VNET_DPDK_PMD_UNKNOWN;
423
424	  xd->port_type = VNET_DPDK_PORT_TYPE_UNKNOWN;
425	  xd->nb_rx_desc = DPDK_NB_RX_DESC_DEFAULT;
426	  xd->nb_tx_desc = DPDK_NB_TX_DESC_DEFAULT;
427
428	  switch (xd->pmd)
429	    {
430	      /* Drivers with valid speed_capa set */
431	    case VNET_DPDK_PMD_E1000EM:
432	    case VNET_DPDK_PMD_IGB:
433	    case VNET_DPDK_PMD_IXGBE:
434	    case VNET_DPDK_PMD_I40E:
435	    case VNET_DPDK_PMD_ICE:
436	      xd->port_type = port_type_from_speed_capa (&dev_info);
437	      xd->supported_flow_actions = VNET_FLOW_ACTION_MARK |
438		VNET_FLOW_ACTION_REDIRECT_TO_NODE |
439		VNET_FLOW_ACTION_REDIRECT_TO_QUEUE |
440		VNET_FLOW_ACTION_BUFFER_ADVANCE |
441		VNET_FLOW_ACTION_COUNT | VNET_FLOW_ACTION_DROP;
442
443	      if (dm->conf->no_tx_checksum_offload == 0)
444		{
445	          xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_TCP_CKSUM;
446	          xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_UDP_CKSUM;
447		  xd->flags |=
448		    DPDK_DEVICE_FLAG_TX_OFFLOAD |
449		    DPDK_DEVICE_FLAG_INTEL_PHDR_CKSUM;
450		}
451
452
453	      break;
454	    case VNET_DPDK_PMD_CXGBE:
455	    case VNET_DPDK_PMD_MLX4:
456	    case VNET_DPDK_PMD_MLX5:
457	    case VNET_DPDK_PMD_QEDE:
458	      xd->port_type = port_type_from_speed_capa (&dev_info);
459	      break;
460
461	      /* SR-IOV VFs */
462	    case VNET_DPDK_PMD_IGBVF:
463	    case VNET_DPDK_PMD_IXGBEVF:
464	    case VNET_DPDK_PMD_I40EVF:
465	      xd->port_type = VNET_DPDK_PORT_TYPE_ETH_VF;
466	      if (dm->conf->no_tx_checksum_offload == 0)
467		{
468	          xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_TCP_CKSUM;
469	          xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_UDP_CKSUM;
470		  xd->flags |=
471		    DPDK_DEVICE_FLAG_TX_OFFLOAD |
472		    DPDK_DEVICE_FLAG_INTEL_PHDR_CKSUM;
473		}
474	      break;
475
476	    case VNET_DPDK_PMD_THUNDERX:
477	      xd->port_type = VNET_DPDK_PORT_TYPE_ETH_VF;
478
479	      if (dm->conf->no_tx_checksum_offload == 0)
480		{
481	          xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_TCP_CKSUM;
482	          xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_UDP_CKSUM;
483		  xd->flags |= DPDK_DEVICE_FLAG_TX_OFFLOAD;
484		}
485	      break;
486
487	    case VNET_DPDK_PMD_ENA:
488	      xd->port_type = VNET_DPDK_PORT_TYPE_ETH_VF;
489	      xd->port_conf.rxmode.offloads &= ~DEV_RX_OFFLOAD_SCATTER;
490	      break;
491
492	    case VNET_DPDK_PMD_DPAA2:
493	      xd->port_type = VNET_DPDK_PORT_TYPE_ETH_10G;
494	      break;
495
496	      /* Cisco VIC */
497	    case VNET_DPDK_PMD_ENIC:
498	      xd->port_type = port_type_from_link_speed (l.link_speed);
499	      break;
500
501	      /* Intel Red Rock Canyon */
502	    case VNET_DPDK_PMD_FM10K:
503	      xd->port_type = VNET_DPDK_PORT_TYPE_ETH_SWITCH;
504	      break;
505
506	      /* virtio */
507	    case VNET_DPDK_PMD_VIRTIO:
508	      xd->port_type = VNET_DPDK_PORT_TYPE_ETH_1G;
509	      xd->nb_rx_desc = DPDK_NB_RX_DESC_VIRTIO;
510	      xd->nb_tx_desc = DPDK_NB_TX_DESC_VIRTIO;
511	      break;
512
513	      /* vmxnet3 */
514	    case VNET_DPDK_PMD_VMXNET3:
515	      xd->port_type = VNET_DPDK_PORT_TYPE_ETH_1G;
516	      xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_MULTI_SEGS;
517	      break;
518
519	    case VNET_DPDK_PMD_AF_PACKET:
520	      xd->port_type = VNET_DPDK_PORT_TYPE_AF_PACKET;
521	      xd->af_packet_instance_num = af_packet_instance_num++;
522	      break;
523
524	    case VNET_DPDK_PMD_VIRTIO_USER:
525	      xd->port_type = VNET_DPDK_PORT_TYPE_VIRTIO_USER;
526	      break;
527
528	    case VNET_DPDK_PMD_VHOST_ETHER:
529	      xd->port_type = VNET_DPDK_PORT_TYPE_VHOST_ETHER;
530	      break;
531
532	    case VNET_DPDK_PMD_LIOVF_ETHER:
533	      xd->port_type = VNET_DPDK_PORT_TYPE_ETH_VF;
534	      break;
535
536	    case VNET_DPDK_PMD_FAILSAFE:
537	      xd->port_type = VNET_DPDK_PORT_TYPE_FAILSAFE;
538	      xd->port_conf.intr_conf.lsc = 1;
539	      break;
540
541	    case VNET_DPDK_PMD_NETVSC:
542	      xd->port_type = port_type_from_link_speed (l.link_speed);
543	      break;
544
545	    default:
546	      xd->port_type = VNET_DPDK_PORT_TYPE_UNKNOWN;
547	    }
548
549	  if (devconf->num_rx_desc)
550	    xd->nb_rx_desc = devconf->num_rx_desc;
551          else {
552
553            /* If num_rx_desc is not specified by VPP user, the current CPU is working
554            with 2M page and has no L3 cache, default num_rx_desc is changed to 512
555            from original 1024 to help reduce TLB misses.
556            */
557            if ((clib_mem_get_default_hugepage_size () == 2 << 20)
558              && check_l3cache() == 0)
559              xd->nb_rx_desc = 512;
560          }
561
562	  if (devconf->num_tx_desc)
563	    xd->nb_tx_desc = devconf->num_tx_desc;
564          else {
565
566            /* If num_tx_desc is not specified by VPP user, the current CPU is working
567            with 2M page and has no L3 cache, default num_tx_desc is changed to 512
568            from original 1024 to help reduce TLB misses.
569            */
570            if ((clib_mem_get_default_hugepage_size () == 2 << 20)
571              && check_l3cache() == 0)
572              xd->nb_tx_desc = 512;
573	  }
574       }
575
576      if (xd->pmd == VNET_DPDK_PMD_AF_PACKET)
577	{
578	  f64 now = vlib_time_now (vm);
579	  u32 rnd;
580	  rnd = (u32) (now * 1e6);
581	  rnd = random_u32 (&rnd);
582	  clib_memcpy (addr + 2, &rnd, sizeof (rnd));
583	  addr[0] = 2;
584	  addr[1] = 0xfe;
585	}
586      else
587	rte_eth_macaddr_get (i, (void *) addr);
588
589      if (xd->tx_q_used < tm->n_vlib_mains)
590	dpdk_device_lock_init (xd);
591
592      xd->port_id = i;
593      xd->device_index = xd - dm->devices;
594      xd->per_interface_next_index = ~0;
595
596      /* assign interface to input thread */
597      int q;
598
599      if (devconf->hqos_enabled)
600	{
601	  xd->flags |= DPDK_DEVICE_FLAG_HQOS;
602
603	  int cpu;
604	  if (devconf->hqos.hqos_thread_valid)
605	    {
606	      if (devconf->hqos.hqos_thread >= dm->hqos_cpu_count)
607		return clib_error_return (0, "invalid HQoS thread index");
608
609	      cpu = dm->hqos_cpu_first_index + devconf->hqos.hqos_thread;
610	    }
611	  else
612	    {
613	      if (dm->hqos_cpu_count == 0)
614		return clib_error_return (0, "no HQoS threads available");
615
616	      cpu = dm->hqos_cpu_first_index + next_hqos_cpu;
617
618	      next_hqos_cpu++;
619	      if (next_hqos_cpu == dm->hqos_cpu_count)
620		next_hqos_cpu = 0;
621
622	      devconf->hqos.hqos_thread_valid = 1;
623	      devconf->hqos.hqos_thread = cpu;
624	    }
625
626	  dpdk_device_and_queue_t *dq;
627	  vec_add2 (dm->devices_by_hqos_cpu[cpu], dq, 1);
628	  dq->device = xd->device_index;
629	  dq->queue_id = 0;
630	}
631
632      error = ethernet_register_interface
633	(dm->vnet_main, dpdk_device_class.index, xd->device_index,
634	 /* ethernet address */ addr,
635	 &xd->hw_if_index, dpdk_flag_change);
636      if (error)
637	return error;
638
639      /*
640       * Ensure default mtu is not > the mtu read from the hardware.
641       * Otherwise rte_eth_dev_configure() will fail and the port will
642       * not be available.
643       * Calculate max_frame_size and mtu supported by NIC
644       */
645      if (ETHERNET_MAX_PACKET_BYTES > dev_info.max_rx_pktlen)
646	{
647	  /*
648	   * This device does not support the platforms's max frame
649	   * size. Use it's advertised mru instead.
650	   */
651	  max_rx_frame = dev_info.max_rx_pktlen;
652	  mtu = dev_info.max_rx_pktlen - sizeof (ethernet_header_t);
653	}
654      else
655	{
656	  /* VPP treats MTU and max_rx_pktlen both equal to
657	   * ETHERNET_MAX_PACKET_BYTES, if dev_info.max_rx_pktlen >=
658	   * ETHERNET_MAX_PACKET_BYTES + sizeof(ethernet_header_t)
659	   */
660	  if (dev_info.max_rx_pktlen >= (ETHERNET_MAX_PACKET_BYTES +
661					 sizeof (ethernet_header_t)))
662	    {
663	      mtu = ETHERNET_MAX_PACKET_BYTES;
664	      max_rx_frame = ETHERNET_MAX_PACKET_BYTES;
665
666	      /*
667	       * Some platforms do not account for Ethernet FCS (4 bytes) in
668	       * MTU calculations. To interop with them increase mru but only
669	       * if the device's settings can support it.
670	       */
671	      if (dpdk_port_crc_strip_enabled (xd) &&
672		  (dev_info.max_rx_pktlen >= (ETHERNET_MAX_PACKET_BYTES +
673					      sizeof (ethernet_header_t) +
674					      4)))
675		{
676		  max_rx_frame += 4;
677		}
678	    }
679	  else
680	    {
681	      max_rx_frame = ETHERNET_MAX_PACKET_BYTES;
682	      mtu = ETHERNET_MAX_PACKET_BYTES - sizeof (ethernet_header_t);
683
684	      if (dpdk_port_crc_strip_enabled (xd) &&
685		  (dev_info.max_rx_pktlen >= (ETHERNET_MAX_PACKET_BYTES + 4)))
686		{
687		  max_rx_frame += 4;
688		}
689	    }
690	}
691
692      if (xd->pmd == VNET_DPDK_PMD_FAILSAFE)
693	{
694	  /* failsafe device numerables are reported with active device only,
695	   * need to query the mtu for current device setup to overwrite
696	   * reported value.
697	   */
698	  uint16_t dev_mtu;
699	  if (!rte_eth_dev_get_mtu (i, &dev_mtu))
700	    {
701	      mtu = dev_mtu;
702	      max_rx_frame = mtu + sizeof (ethernet_header_t);
703
704	      if (dpdk_port_crc_strip_enabled (xd))
705		{
706		  max_rx_frame += 4;
707		}
708	    }
709	}
710
711      /*Set port rxmode config */
712      xd->port_conf.rxmode.max_rx_pkt_len = max_rx_frame;
713
714      sw = vnet_get_hw_sw_interface (dm->vnet_main, xd->hw_if_index);
715      xd->sw_if_index = sw->sw_if_index;
716      vnet_hw_interface_set_input_node (dm->vnet_main, xd->hw_if_index,
717					dpdk_input_node.index);
718
719      if (devconf->workers)
720	{
721	  int i;
722	  q = 0;
723	  clib_bitmap_foreach (i, devconf->workers, ({
724	    vnet_hw_interface_assign_rx_thread (dm->vnet_main, xd->hw_if_index, q++,
725					     vdm->first_worker_thread_index + i);
726	  }));
727	}
728      else
729	for (q = 0; q < xd->rx_q_used; q++)
730	  {
731	    vnet_hw_interface_assign_rx_thread (dm->vnet_main, xd->hw_if_index, q,	/* any */
732						~1);
733	  }
734
735      /*Get vnet hardware interface */
736      hi = vnet_get_hw_interface (dm->vnet_main, xd->hw_if_index);
737
738      /*Override default max_packet_bytes and max_supported_bytes set in
739       * ethernet_register_interface() above*/
740      if (hi)
741	{
742	  hi->max_packet_bytes = mtu;
743	  hi->max_supported_packet_bytes = max_rx_frame;
744	  hi->numa_node = xd->cpu_socket;
745	}
746
747      if (dm->conf->no_tx_checksum_offload == 0)
748	if (xd->flags & DPDK_DEVICE_FLAG_TX_OFFLOAD && hi != NULL)
749	  hi->flags |= VNET_HW_INTERFACE_FLAG_SUPPORTS_TX_L4_CKSUM_OFFLOAD;
750
751    if (devconf->tso == DPDK_DEVICE_TSO_ON)
752    {
753      if (xd->flags & DPDK_DEVICE_FLAG_TX_OFFLOAD && hi != NULL)
754      {
755        /*tcp_udp checksum must be enabled*/
756        if (hi->flags & VNET_HW_INTERFACE_FLAG_SUPPORTS_TX_L4_CKSUM_OFFLOAD)
757        {
758          hi->flags |= VNET_HW_INTERFACE_FLAG_SUPPORTS_GSO;
759          vnm->interface_main.gso_interface_count++;
760          xd->port_conf.txmode.offloads |= DEV_TX_OFFLOAD_TCP_TSO |
761                                   DEV_TX_OFFLOAD_UDP_TSO;
762        }
763        else
764          return clib_error_return (0, "TSO: TCP/UDP checksum offload must be enabled");
765      }
766    }
767
768      dpdk_device_setup (xd);
769
770      if (vec_len (xd->errors))
771	dpdk_log_err ("setup failed for device %U. Errors:\n  %U",
772		      format_dpdk_device_name, i,
773		      format_dpdk_device_errors, xd);
774
775      if (devconf->hqos_enabled)
776	{
777	  clib_error_t *rv;
778	  rv = dpdk_port_setup_hqos (xd, &devconf->hqos);
779	  if (rv)
780	    return rv;
781	}
782
783      /*
784       * A note on Cisco VIC (PMD_ENIC) and VLAN:
785       *
786       * With Cisco VIC vNIC, every ingress packet is tagged. On a
787       * trunk vNIC (C series "standalone" server), packets on no VLAN
788       * are tagged with vlan 0. On an access vNIC (standalone or B
789       * series "blade" server), packets on the default/native VLAN
790       * are tagged with that vNIC's VLAN. VPP expects these packets
791       * to be untagged, and previously enabled VLAN strip on VIC by
792       * default. But it also broke vlan sub-interfaces.
793       *
794       * The VIC adapter has "untag default vlan" ingress VLAN rewrite
795       * mode, which removes tags from these packets. VPP now includes
796       * a local patch for the enic driver to use this untag mode, so
797       * enabling vlan stripping is no longer needed. In future, the
798       * driver + dpdk will have an API to set the mode after
799       * rte_eal_init. Then, this note and local patch will be
800       * removed.
801       */
802
803      /*
804       * VLAN stripping: default to VLAN strip disabled, unless specified
805       * otherwise in the startup config.
806       */
807
808      vlan_off = rte_eth_dev_get_vlan_offload (xd->port_id);
809      if (devconf->vlan_strip_offload == DPDK_DEVICE_VLAN_STRIP_ON)
810	{
811	  vlan_off |= ETH_VLAN_STRIP_OFFLOAD;
812	  if (rte_eth_dev_set_vlan_offload (xd->port_id, vlan_off) >= 0)
813	    dpdk_log_info ("VLAN strip enabled for interface\n");
814	  else
815	    dpdk_log_warn ("VLAN strip cannot be supported by interface\n");
816	  xd->port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_STRIP;
817	}
818      else
819	{
820	  if (vlan_off & ETH_VLAN_STRIP_OFFLOAD)
821	    {
822	      vlan_off &= ~ETH_VLAN_STRIP_OFFLOAD;
823	      if (rte_eth_dev_set_vlan_offload (xd->port_id, vlan_off) >= 0)
824		dpdk_log_warn ("set VLAN offload failed\n");
825	    }
826	  xd->port_conf.rxmode.offloads &= ~DEV_RX_OFFLOAD_VLAN_STRIP;
827	}
828
829      if (hi)
830	hi->max_packet_bytes = xd->port_conf.rxmode.max_rx_pkt_len
831	  - sizeof (ethernet_header_t);
832      else
833	clib_warning ("hi NULL");
834
835      if (dm->conf->no_multi_seg)
836	mtu = mtu > ETHER_MAX_LEN ? ETHER_MAX_LEN : mtu;
837
838      rte_eth_dev_set_mtu (xd->port_id, mtu);
839    }
840  /* *INDENT-ON* */
841
842  return 0;
843}
844
845static void
846dpdk_bind_devices_to_uio (dpdk_config_main_t * conf)
847{
848  vlib_main_t *vm = vlib_get_main ();
849  clib_error_t *error;
850  u8 *pci_addr = 0;
851  int num_whitelisted = vec_len (conf->dev_confs);
852  vlib_pci_device_info_t *d = 0;
853  vlib_pci_addr_t *addr = 0, *addrs;
854  int i;
855
856  addrs = vlib_pci_get_all_dev_addrs ();
857  /* *INDENT-OFF* */
858  vec_foreach (addr, addrs)
859    {
860    dpdk_device_config_t * devconf = 0;
861    vec_reset_length (pci_addr);
862    pci_addr = format (pci_addr, "%U%c", format_vlib_pci_addr, addr, 0);
863    if (d)
864    {
865      vlib_pci_free_device_info (d);
866      d = 0;
867      }
868    d = vlib_pci_get_device_info (vm, addr, &error);
869    if (error)
870    {
871      clib_error_report (error);
872      continue;
873    }
874
875    if (d->device_class != PCI_CLASS_NETWORK_ETHERNET && d->device_class != PCI_CLASS_PROCESSOR_CO)
876      continue;
877
878    if (num_whitelisted)
879      {
880	uword * p = hash_get (conf->device_config_index_by_pci_addr, addr->as_u32);
881
882	if (!p)
883          {
884          skipped:
885            continue;
886          }
887
888	devconf = pool_elt_at_index (conf->dev_confs, p[0]);
889      }
890
891    /* Enforce Device blacklist by vendor and device */
892    for (i = 0; i < vec_len (conf->blacklist_by_pci_vendor_and_device); i++)
893      {
894        u16 vendor, device;
895        vendor = (u16)(conf->blacklist_by_pci_vendor_and_device[i] >> 16);
896        device = (u16)(conf->blacklist_by_pci_vendor_and_device[i] & 0xFFFF);
897        if (d->vendor_id == vendor && d->device_id == device)
898          {
899            /*
900             * Expected case: device isn't whitelisted,
901             * so blacklist it...
902             */
903            if (devconf == 0)
904              {
905                /* Device is blacklisted */
906                pool_get (conf->dev_confs, devconf);
907                hash_set (conf->device_config_index_by_pci_addr, addr->as_u32,
908                          devconf - conf->dev_confs);
909                devconf->pci_addr.as_u32 = addr->as_u32;
910                devconf->is_blacklisted = 1;
911                goto skipped;
912              }
913            else /* explicitly whitelisted, ignore the device blacklist  */
914              break;
915          }
916      }
917
918    /* virtio */
919    if (d->vendor_id == 0x1af4 &&
920            (d->device_id == VIRTIO_PCI_LEGACY_DEVICEID_NET ||
921             d->device_id == VIRTIO_PCI_MODERN_DEVICEID_NET))
922      ;
923    /* vmxnet3 */
924    else if (d->vendor_id == 0x15ad && d->device_id == 0x07b0)
925      {
926	/*
927	 * For vmxnet3 PCI, unless it is explicitly specified in the whitelist,
928	 * the default is to put it in the blacklist.
929	 */
930	if (devconf == 0)
931	  {
932	    pool_get (conf->dev_confs, devconf);
933	    hash_set (conf->device_config_index_by_pci_addr, addr->as_u32,
934		      devconf - conf->dev_confs);
935	    devconf->pci_addr.as_u32 = addr->as_u32;
936	    devconf->is_blacklisted = 1;
937	  }
938      }
939    /* all Intel network devices */
940    else if (d->vendor_id == 0x8086 && d->device_class == PCI_CLASS_NETWORK_ETHERNET)
941      ;
942    /* all Intel QAT devices VFs */
943    else if (d->vendor_id == 0x8086 && d->device_class == PCI_CLASS_PROCESSOR_CO &&
944        (d->device_id == 0x0443 || d->device_id == 0x18a1 || d->device_id == 0x19e3 ||
945        d->device_id == 0x37c9 || d->device_id == 0x6f55))
946      ;
947    /* Cisco VIC */
948    else if (d->vendor_id == 0x1137 &&
949        (d->device_id == 0x0043 || d->device_id == 0x0071))
950      ;
951    /* Chelsio T4/T5 */
952    else if (d->vendor_id == 0x1425 && (d->device_id & 0xe000) == 0x4000)
953      ;
954    /* Amazen Elastic Network Adapter */
955    else if (d->vendor_id == 0x1d0f && d->device_id >= 0xec20 && d->device_id <= 0xec21)
956      ;
957    /* Cavium Network Adapter */
958    else if (d->vendor_id == 0x177d && d->device_id == 0x9712)
959      ;
960    /* Cavium FastlinQ QL41000 Series */
961    else if (d->vendor_id == 0x1077 && d->device_id >= 0x8070 && d->device_id <= 0x8090)
962      ;
963    /* Mellanox mlx4 */
964    else if (d->vendor_id == 0x15b3 && d->device_id >= 0x1003 && d->device_id <= 0x1004)
965      {
966        continue;
967      }
968    /* Mellanox mlx5 */
969    else if (d->vendor_id == 0x15b3 && d->device_id >= 0x1013 && d->device_id <= 0x101a)
970      {
971        continue;
972      }
973    else
974      {
975        dpdk_log_warn ("Unsupported PCI device 0x%04x:0x%04x found "
976		      "at PCI address %s\n", (u16) d->vendor_id, (u16) d->device_id,
977		      pci_addr);
978        continue;
979      }
980
981    error = vlib_pci_bind_to_uio (vm, addr, (char *) conf->uio_driver_name);
982
983    if (error)
984      {
985	if (devconf == 0)
986	  {
987	    pool_get (conf->dev_confs, devconf);
988	    hash_set (conf->device_config_index_by_pci_addr, addr->as_u32,
989		      devconf - conf->dev_confs);
990	    devconf->pci_addr.as_u32 = addr->as_u32;
991	  }
992	devconf->is_blacklisted = 1;
993	clib_error_report (error);
994      }
995  }
996  /* *INDENT-ON* */
997  vec_free (pci_addr);
998  vlib_pci_free_device_info (d);
999}
1000
1001static void
1002dpdk_bind_vmbus_devices_to_uio (dpdk_config_main_t * conf)
1003{
1004  clib_error_t *error;
1005  vlib_vmbus_addr_t *addrs, *addr = 0;
1006
1007  addrs = vlib_vmbus_get_all_dev_addrs ();
1008
1009  /* *INDENT-OFF* */
1010  vec_foreach (addr, addrs)
1011    {
1012      error = vlib_vmbus_bind_to_uio (addr);
1013
1014      if (error)
1015	{
1016	  clib_error_report (error);
1017	}
1018    }
1019  /* *INDENT-ON* */
1020}
1021
1022static clib_error_t *
1023dpdk_device_config (dpdk_config_main_t * conf, vlib_pci_addr_t pci_addr,
1024		    unformat_input_t * input, u8 is_default)
1025{
1026  clib_error_t *error = 0;
1027  uword *p;
1028  dpdk_device_config_t *devconf;
1029  unformat_input_t sub_input;
1030
1031  if (is_default)
1032    {
1033      devconf = &conf->default_devconf;
1034    }
1035  else
1036    {
1037      p = hash_get (conf->device_config_index_by_pci_addr, pci_addr.as_u32);
1038
1039      if (!p)
1040	{
1041	  pool_get (conf->dev_confs, devconf);
1042	  hash_set (conf->device_config_index_by_pci_addr, pci_addr.as_u32,
1043		    devconf - conf->dev_confs);
1044	}
1045      else
1046	return clib_error_return (0,
1047				  "duplicate configuration for PCI address %U",
1048				  format_vlib_pci_addr, &pci_addr);
1049    }
1050
1051  devconf->pci_addr.as_u32 = pci_addr.as_u32;
1052  devconf->hqos_enabled = 0;
1053  devconf->tso = DPDK_DEVICE_TSO_DEFAULT;
1054#if 0
1055  dpdk_device_config_hqos_default (&devconf->hqos);
1056#endif
1057
1058  if (!input)
1059    return 0;
1060
1061  unformat_skip_white_space (input);
1062  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
1063    {
1064      if (unformat (input, "num-rx-queues %u", &devconf->num_rx_queues))
1065	;
1066      else if (unformat (input, "num-tx-queues %u", &devconf->num_tx_queues))
1067	;
1068      else if (unformat (input, "num-rx-desc %u", &devconf->num_rx_desc))
1069	;
1070      else if (unformat (input, "num-tx-desc %u", &devconf->num_tx_desc))
1071	;
1072      else if (unformat (input, "name %s", &devconf->name))
1073	;
1074      else if (unformat (input, "workers %U", unformat_bitmap_list,
1075			 &devconf->workers))
1076	;
1077      else
1078	if (unformat
1079	    (input, "rss %U", unformat_vlib_cli_sub_input, &sub_input))
1080	{
1081	  error = unformat_rss_fn (&sub_input, &devconf->rss_fn);
1082	  if (error)
1083	    break;
1084	}
1085      else if (unformat (input, "vlan-strip-offload off"))
1086	devconf->vlan_strip_offload = DPDK_DEVICE_VLAN_STRIP_OFF;
1087      else if (unformat (input, "vlan-strip-offload on"))
1088	devconf->vlan_strip_offload = DPDK_DEVICE_VLAN_STRIP_ON;
1089      else
1090	if (unformat
1091	    (input, "hqos %U", unformat_vlib_cli_sub_input, &sub_input))
1092	{
1093	  devconf->hqos_enabled = 1;
1094	  error = unformat_hqos (&sub_input, &devconf->hqos);
1095	  if (error)
1096	    break;
1097	}
1098      else if (unformat (input, "hqos"))
1099	{
1100	  devconf->hqos_enabled = 1;
1101	}
1102      else if (unformat (input, "tso on"))
1103	{
1104	  devconf->tso = DPDK_DEVICE_TSO_ON;
1105	}
1106      else if (unformat (input, "tso off"))
1107	{
1108	  devconf->tso = DPDK_DEVICE_TSO_OFF;
1109	}
1110      else
1111	{
1112	  error = clib_error_return (0, "unknown input `%U'",
1113				     format_unformat_error, input);
1114	  break;
1115	}
1116    }
1117
1118  if (error)
1119    return error;
1120
1121  if (devconf->workers && devconf->num_rx_queues == 0)
1122    devconf->num_rx_queues = clib_bitmap_count_set_bits (devconf->workers);
1123  else if (devconf->workers &&
1124	   clib_bitmap_count_set_bits (devconf->workers) !=
1125	   devconf->num_rx_queues)
1126    error =
1127      clib_error_return (0,
1128			 "%U: number of worker threads must be "
1129			 "equal to number of rx queues", format_vlib_pci_addr,
1130			 &pci_addr);
1131
1132  return error;
1133}
1134
1135static clib_error_t *
1136dpdk_log_read_ready (clib_file_t * uf)
1137{
1138  unformat_input_t input;
1139  u8 *line, *s = 0;
1140  int n, n_try;
1141
1142  n = n_try = 4096;
1143  while (n == n_try)
1144    {
1145      uword len = vec_len (s);
1146      vec_resize (s, len + n_try);
1147
1148      n = read (uf->file_descriptor, s + len, n_try);
1149      if (n < 0 && errno != EAGAIN)
1150	return clib_error_return_unix (0, "read");
1151      _vec_len (s) = len + (n < 0 ? 0 : n);
1152    }
1153
1154  unformat_init_vector (&input, s);
1155
1156  while (unformat_user (&input, unformat_line, &line))
1157    {
1158      dpdk_log_notice ("%v", line);
1159      vec_free (line);
1160    }
1161
1162  unformat_free (&input);
1163  return 0;
1164}
1165
1166static clib_error_t *
1167dpdk_config (vlib_main_t * vm, unformat_input_t * input)
1168{
1169  clib_error_t *error = 0;
1170  dpdk_config_main_t *conf = &dpdk_config_main;
1171  vlib_thread_main_t *tm = vlib_get_thread_main ();
1172  dpdk_device_config_t *devconf;
1173  vlib_pci_addr_t pci_addr;
1174  unformat_input_t sub_input;
1175  uword default_hugepage_sz, x;
1176  u8 *s, *tmp = 0;
1177  int ret, i;
1178  int num_whitelisted = 0;
1179  u8 no_pci = 0;
1180  u8 no_vmbus = 0;
1181  u8 file_prefix = 0;
1182  u8 *socket_mem = 0;
1183  u8 *huge_dir_path = 0;
1184  u32 vendor, device;
1185
1186  huge_dir_path =
1187    format (0, "%s/hugepages%c", vlib_unix_get_runtime_dir (), 0);
1188
1189  conf->device_config_index_by_pci_addr = hash_create (0, sizeof (uword));
1190
1191  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
1192    {
1193      /* Prime the pump */
1194      if (unformat (input, "no-hugetlb"))
1195	{
1196	  vec_add1 (conf->eal_init_args, (u8 *) "--no-huge");
1197	}
1198
1199      else if (unformat (input, "enable-tcp-udp-checksum"))
1200	conf->enable_tcp_udp_checksum = 1;
1201
1202      else if (unformat (input, "no-tx-checksum-offload"))
1203	conf->no_tx_checksum_offload = 1;
1204
1205      else if (unformat (input, "decimal-interface-names"))
1206	conf->interface_name_format_decimal = 1;
1207
1208      else if (unformat (input, "no-multi-seg"))
1209	conf->no_multi_seg = 1;
1210
1211      else if (unformat (input, "dev default %U", unformat_vlib_cli_sub_input,
1212			 &sub_input))
1213	{
1214	  error =
1215	    dpdk_device_config (conf, (vlib_pci_addr_t) (u32) ~ 1, &sub_input,
1216				1);
1217
1218	  if (error)
1219	    return error;
1220	}
1221      else
1222	if (unformat
1223	    (input, "dev %U %U", unformat_vlib_pci_addr, &pci_addr,
1224	     unformat_vlib_cli_sub_input, &sub_input))
1225	{
1226	  error = dpdk_device_config (conf, pci_addr, &sub_input, 0);
1227
1228	  if (error)
1229	    return error;
1230
1231	  num_whitelisted++;
1232	}
1233      else if (unformat (input, "dev %U", unformat_vlib_pci_addr, &pci_addr))
1234	{
1235	  error = dpdk_device_config (conf, pci_addr, 0, 0);
1236
1237	  if (error)
1238	    return error;
1239
1240	  num_whitelisted++;
1241	}
1242      else if (unformat (input, "num-mem-channels %d", &conf->nchannels))
1243	conf->nchannels_set_manually = 0;
1244      else if (unformat (input, "num-crypto-mbufs %d",
1245			 &conf->num_crypto_mbufs))
1246	;
1247      else if (unformat (input, "uio-driver %s", &conf->uio_driver_name))
1248	;
1249      else if (unformat (input, "socket-mem %s", &socket_mem))
1250	;
1251      else if (unformat (input, "no-pci"))
1252	{
1253	  no_pci = 1;
1254	  tmp = format (0, "--no-pci%c", 0);
1255	  vec_add1 (conf->eal_init_args, tmp);
1256	}
1257      else if (unformat (input, "blacklist %x:%x", &vendor, &device))
1258	{
1259	  u32 blacklist_entry;
1260	  if (vendor > 0xFFFF)
1261	    return clib_error_return (0, "blacklist PCI vendor out of range");
1262	  if (device > 0xFFFF)
1263	    return clib_error_return (0, "blacklist PCI device out of range");
1264	  blacklist_entry = (vendor << 16) | (device & 0xffff);
1265	  vec_add1 (conf->blacklist_by_pci_vendor_and_device,
1266		    blacklist_entry);
1267	}
1268      else if (unformat (input, "no-vmbus"))
1269	{
1270	  no_vmbus = 1;
1271	  tmp = format (0, "--no-vmbus%c", 0);
1272	  vec_add1 (conf->eal_init_args, tmp);
1273	}
1274
1275#define _(a)                                    \
1276      else if (unformat(input, #a))             \
1277        {                                       \
1278          tmp = format (0, "--%s%c", #a, 0);    \
1279          vec_add1 (conf->eal_init_args, tmp);    \
1280        }
1281      foreach_eal_double_hyphen_predicate_arg
1282#undef _
1283#define _(a)                                          \
1284	else if (unformat(input, #a " %s", &s))	      \
1285	  {					      \
1286            if (!strncmp(#a, "file-prefix", 11)) \
1287              file_prefix = 1;                        \
1288	    tmp = format (0, "--%s%c", #a, 0);	      \
1289	    vec_add1 (conf->eal_init_args, tmp);      \
1290	    vec_add1 (s, 0);			      \
1291            if (!strncmp(#a, "vdev", 4))              \
1292              if (strstr((char*)s, "af_packet"))      \
1293                clib_warning ("af_packet obsoleted. Use CLI 'create host-interface'."); \
1294	    vec_add1 (conf->eal_init_args, s);	      \
1295	  }
1296	foreach_eal_double_hyphen_arg
1297#undef _
1298#define _(a,b)						\
1299	  else if (unformat(input, #a " %s", &s))	\
1300	    {						\
1301	      tmp = format (0, "-%s%c", #b, 0);		\
1302	      vec_add1 (conf->eal_init_args, tmp);	\
1303	      vec_add1 (s, 0);				\
1304	      vec_add1 (conf->eal_init_args, s);	\
1305	    }
1306	foreach_eal_single_hyphen_arg
1307#undef _
1308#define _(a,b)						\
1309	    else if (unformat(input, #a " %s", &s))	\
1310	      {						\
1311		tmp = format (0, "-%s%c", #b, 0);	\
1312		vec_add1 (conf->eal_init_args, tmp);	\
1313		vec_add1 (s, 0);			\
1314		vec_add1 (conf->eal_init_args, s);	\
1315		conf->a##_set_manually = 1;		\
1316	      }
1317	foreach_eal_single_hyphen_mandatory_arg
1318#undef _
1319	else if (unformat (input, "default"))
1320	;
1321
1322      else if (unformat_skip_white_space (input))
1323	;
1324      else
1325	{
1326	  error = clib_error_return (0, "unknown input `%U'",
1327				     format_unformat_error, input);
1328	  goto done;
1329	}
1330    }
1331
1332  if (!conf->uio_driver_name)
1333    conf->uio_driver_name = format (0, "auto%c", 0);
1334
1335  default_hugepage_sz = clib_mem_get_default_hugepage_size ();
1336
1337  /* *INDENT-OFF* */
1338  clib_bitmap_foreach (x, tm->cpu_socket_bitmap, (
1339    {
1340      clib_error_t *e;
1341      uword n_pages;
1342      /* preallocate at least 16MB of hugepages per socket,
1343	 if more is needed it is up to consumer to preallocate more */
1344      n_pages = round_pow2 ((uword) 16 << 20, default_hugepage_sz);
1345      n_pages /= default_hugepage_sz;
1346
1347      if ((e = clib_sysfs_prealloc_hugepages(x, 0, n_pages)))
1348	clib_error_report (e);
1349  }));
1350  /* *INDENT-ON* */
1351
1352  if (!file_prefix)
1353    {
1354      tmp = format (0, "--file-prefix%c", 0);
1355      vec_add1 (conf->eal_init_args, tmp);
1356      tmp = format (0, "vpp%c", 0);
1357      vec_add1 (conf->eal_init_args, tmp);
1358    }
1359
1360  if (error)
1361    return error;
1362
1363  /* I'll bet that -c and -n must be the first and second args... */
1364  if (!conf->coremask_set_manually)
1365    {
1366      vlib_thread_registration_t *tr;
1367      uword *coremask = 0;
1368      int i;
1369
1370      /* main thread core */
1371      coremask = clib_bitmap_set (coremask, tm->main_lcore, 1);
1372
1373      for (i = 0; i < vec_len (tm->registrations); i++)
1374	{
1375	  tr = tm->registrations[i];
1376	  coremask = clib_bitmap_or (coremask, tr->coremask);
1377	}
1378
1379      vec_insert (conf->eal_init_args, 2, 1);
1380      conf->eal_init_args[1] = (u8 *) "-c";
1381      tmp = format (0, "%U%c", format_bitmap_hex, coremask, 0);
1382      conf->eal_init_args[2] = tmp;
1383      clib_bitmap_free (coremask);
1384    }
1385
1386  if (!conf->nchannels_set_manually)
1387    {
1388      vec_insert (conf->eal_init_args, 2, 3);
1389      conf->eal_init_args[3] = (u8 *) "-n";
1390      tmp = format (0, "%d", conf->nchannels);
1391      vec_terminate_c_string (tmp);
1392      conf->eal_init_args[4] = tmp;
1393    }
1394
1395  if (no_pci == 0 && geteuid () == 0)
1396    dpdk_bind_devices_to_uio (conf);
1397
1398  if (no_vmbus == 0 && geteuid () == 0)
1399    dpdk_bind_vmbus_devices_to_uio (conf);
1400
1401#define _(x) \
1402    if (devconf->x == 0 && conf->default_devconf.x > 0) \
1403      devconf->x = conf->default_devconf.x ;
1404
1405  /* *INDENT-OFF* */
1406  pool_foreach (devconf, conf->dev_confs, ({
1407
1408    /* default per-device config items */
1409    foreach_dpdk_device_config_item
1410
1411    /* copy vlan_strip config from default device */
1412	if (devconf->vlan_strip_offload == 0 &&
1413		conf->default_devconf.vlan_strip_offload > 0)
1414		devconf->vlan_strip_offload =
1415			conf->default_devconf.vlan_strip_offload;
1416
1417	/* copy tso config from default device */
1418	_(tso)
1419
1420    /* add DPDK EAL whitelist/blacklist entry */
1421    if (num_whitelisted > 0 && devconf->is_blacklisted == 0)
1422      {
1423	tmp = format (0, "-w%c", 0);
1424	vec_add1 (conf->eal_init_args, tmp);
1425	tmp = format (0, "%U%c", format_vlib_pci_addr, &devconf->pci_addr, 0);
1426	vec_add1 (conf->eal_init_args, tmp);
1427      }
1428    else if (num_whitelisted == 0 && devconf->is_blacklisted != 0)
1429      {
1430	tmp = format (0, "-b%c", 0);
1431	vec_add1 (conf->eal_init_args, tmp);
1432	tmp = format (0, "%U%c", format_vlib_pci_addr, &devconf->pci_addr, 0);
1433	vec_add1 (conf->eal_init_args, tmp);
1434      }
1435  }));
1436  /* *INDENT-ON* */
1437
1438#undef _
1439
1440  /* set master-lcore */
1441  tmp = format (0, "--master-lcore%c", 0);
1442  vec_add1 (conf->eal_init_args, tmp);
1443  tmp = format (0, "%u%c", tm->main_lcore, 0);
1444  vec_add1 (conf->eal_init_args, tmp);
1445
1446
1447  if (socket_mem)
1448    clib_warning ("socket-mem argument is deprecated");
1449
1450  /* NULL terminate the "argv" vector, in case of stupidity */
1451  vec_add1 (conf->eal_init_args, 0);
1452  _vec_len (conf->eal_init_args) -= 1;
1453
1454  /* Set up DPDK eal and packet mbuf pool early. */
1455
1456  int log_fds[2] = { 0 };
1457  if (pipe (log_fds) == 0)
1458    {
1459      if (fcntl (log_fds[1], F_SETFL, O_NONBLOCK) == 0)
1460	{
1461	  FILE *f = fdopen (log_fds[1], "a");
1462	  if (f && rte_openlog_stream (f) == 0)
1463	    {
1464	      clib_file_t t = { 0 };
1465	      t.read_function = dpdk_log_read_ready;
1466	      t.file_descriptor = log_fds[0];
1467	      t.description = format (0, "DPDK logging pipe");
1468	      clib_file_add (&file_main, &t);
1469	    }
1470	}
1471      else
1472	{
1473	  close (log_fds[0]);
1474	  close (log_fds[1]);
1475	}
1476    }
1477
1478  vm = vlib_get_main ();
1479
1480  /* make copy of args as rte_eal_init tends to mess up with arg array */
1481  for (i = 1; i < vec_len (conf->eal_init_args); i++)
1482    conf->eal_init_args_str = format (conf->eal_init_args_str, "%s ",
1483				      conf->eal_init_args[i]);
1484
1485  vec_terminate_c_string (conf->eal_init_args_str);
1486
1487  dpdk_log_warn ("EAL init args: %s", conf->eal_init_args_str);
1488  ret = rte_eal_init (vec_len (conf->eal_init_args),
1489		      (char **) conf->eal_init_args);
1490
1491  /* lazy umount hugepages */
1492  umount2 ((char *) huge_dir_path, MNT_DETACH);
1493  rmdir ((char *) huge_dir_path);
1494  vec_free (huge_dir_path);
1495
1496  if (ret < 0)
1497    return clib_error_return (0, "rte_eal_init returned %d", ret);
1498
1499  /* main thread 1st */
1500  if ((error = dpdk_buffer_pools_create (vm)))
1501    return error;
1502
1503done:
1504  return error;
1505}
1506
1507VLIB_CONFIG_FUNCTION (dpdk_config, "dpdk");
1508
1509void
1510dpdk_update_link_state (dpdk_device_t * xd, f64 now)
1511{
1512  vnet_main_t *vnm = vnet_get_main ();
1513  struct rte_eth_link prev_link = xd->link;
1514  u32 hw_flags = 0;
1515  u8 hw_flags_chg = 0;
1516
1517  /* only update link state for PMD interfaces */
1518  if ((xd->flags & DPDK_DEVICE_FLAG_PMD) == 0)
1519    return;
1520
1521  xd->time_last_link_update = now ? now : xd->time_last_link_update;
1522  clib_memset (&xd->link, 0, sizeof (xd->link));
1523  rte_eth_link_get_nowait (xd->port_id, &xd->link);
1524
1525  if (LINK_STATE_ELOGS)
1526    {
1527      vlib_main_t *vm = vlib_get_main ();
1528      ELOG_TYPE_DECLARE (e) =
1529      {
1530      .format =
1531	  "update-link-state: sw_if_index %d, admin_up %d,"
1532	  "old link_state %d new link_state %d",.format_args = "i4i1i1i1",};
1533
1534      struct
1535      {
1536	u32 sw_if_index;
1537	u8 admin_up;
1538	u8 old_link_state;
1539	u8 new_link_state;
1540      } *ed;
1541      ed = ELOG_DATA (&vm->elog_main, e);
1542      ed->sw_if_index = xd->sw_if_index;
1543      ed->admin_up = (xd->flags & DPDK_DEVICE_FLAG_ADMIN_UP) != 0;
1544      ed->old_link_state = (u8)
1545	vnet_hw_interface_is_link_up (vnm, xd->hw_if_index);
1546      ed->new_link_state = (u8) xd->link.link_status;
1547    }
1548
1549  if ((xd->link.link_duplex != prev_link.link_duplex))
1550    {
1551      hw_flags_chg = 1;
1552      switch (xd->link.link_duplex)
1553	{
1554	case ETH_LINK_HALF_DUPLEX:
1555	  hw_flags |= VNET_HW_INTERFACE_FLAG_HALF_DUPLEX;
1556	  break;
1557	case ETH_LINK_FULL_DUPLEX:
1558	  hw_flags |= VNET_HW_INTERFACE_FLAG_FULL_DUPLEX;
1559	  break;
1560	default:
1561	  break;
1562	}
1563    }
1564  if (xd->link.link_speed != prev_link.link_speed)
1565    vnet_hw_interface_set_link_speed (vnm, xd->hw_if_index,
1566				      xd->link.link_speed * 1000);
1567
1568  if (xd->link.link_status != prev_link.link_status)
1569    {
1570      hw_flags_chg = 1;
1571
1572      if (xd->link.link_status)
1573	hw_flags |= VNET_HW_INTERFACE_FLAG_LINK_UP;
1574    }
1575
1576  if (hw_flags_chg)
1577    {
1578      if (LINK_STATE_ELOGS)
1579	{
1580	  vlib_main_t *vm = vlib_get_main ();
1581
1582	  ELOG_TYPE_DECLARE (e) =
1583	  {
1584	  .format =
1585	      "update-link-state: sw_if_index %d, new flags %d",.format_args
1586	      = "i4i4",};
1587
1588	  struct
1589	  {
1590	    u32 sw_if_index;
1591	    u32 flags;
1592	  } *ed;
1593	  ed = ELOG_DATA (&vm->elog_main, e);
1594	  ed->sw_if_index = xd->sw_if_index;
1595	  ed->flags = hw_flags;
1596	}
1597      vnet_hw_interface_set_flags (vnm, xd->hw_if_index, hw_flags);
1598    }
1599}
1600
1601static uword
1602dpdk_process (vlib_main_t * vm, vlib_node_runtime_t * rt, vlib_frame_t * f)
1603{
1604  clib_error_t *error;
1605  dpdk_main_t *dm = &dpdk_main;
1606  dpdk_device_t *xd;
1607  vlib_thread_main_t *tm = vlib_get_thread_main ();
1608
1609  error = dpdk_lib_init (dm);
1610
1611  if (error)
1612    clib_error_report (error);
1613
1614  tm->worker_thread_release = 1;
1615
1616  f64 now = vlib_time_now (vm);
1617  vec_foreach (xd, dm->devices)
1618  {
1619    dpdk_update_link_state (xd, now);
1620  }
1621
1622  while (1)
1623    {
1624      /*
1625       * check each time through the loop in case intervals are changed
1626       */
1627      f64 min_wait = dm->link_state_poll_interval < dm->stat_poll_interval ?
1628	dm->link_state_poll_interval : dm->stat_poll_interval;
1629
1630      vlib_process_wait_for_event_or_clock (vm, min_wait);
1631
1632      if (dm->admin_up_down_in_progress)
1633	/* skip the poll if an admin up down is in progress (on any interface) */
1634	continue;
1635
1636      vec_foreach (xd, dm->devices)
1637      {
1638	f64 now = vlib_time_now (vm);
1639	if ((now - xd->time_last_stats_update) >= dm->stat_poll_interval)
1640	  dpdk_update_counters (xd, now);
1641	if ((now - xd->time_last_link_update) >= dm->link_state_poll_interval)
1642	  dpdk_update_link_state (xd, now);
1643
1644      }
1645    }
1646
1647  return 0;
1648}
1649
1650/* *INDENT-OFF* */
1651VLIB_REGISTER_NODE (dpdk_process_node,static) = {
1652    .function = dpdk_process,
1653    .type = VLIB_NODE_TYPE_PROCESS,
1654    .name = "dpdk-process",
1655    .process_log2_n_stack_bytes = 17,
1656};
1657/* *INDENT-ON* */
1658
1659static clib_error_t *
1660dpdk_init (vlib_main_t * vm)
1661{
1662  dpdk_main_t *dm = &dpdk_main;
1663  clib_error_t *error = 0;
1664
1665  /* verify that structs are cacheline aligned */
1666  STATIC_ASSERT (offsetof (dpdk_device_t, cacheline0) == 0,
1667		 "Cache line marker must be 1st element in dpdk_device_t");
1668  STATIC_ASSERT (offsetof (dpdk_device_t, cacheline1) ==
1669		 CLIB_CACHE_LINE_BYTES,
1670		 "Data in cache line 0 is bigger than cache line size");
1671  STATIC_ASSERT (offsetof (frame_queue_trace_t, cacheline0) == 0,
1672		 "Cache line marker must be 1st element in frame_queue_trace_t");
1673  STATIC_ASSERT (RTE_CACHE_LINE_SIZE == 1 << CLIB_LOG2_CACHE_LINE_BYTES,
1674		 "DPDK RTE CACHE LINE SIZE does not match with 1<<CLIB_LOG2_CACHE_LINE_BYTES");
1675
1676  dpdk_cli_reference ();
1677
1678  dm->vlib_main = vm;
1679  dm->vnet_main = vnet_get_main ();
1680  dm->conf = &dpdk_config_main;
1681
1682  dm->conf->nchannels = 4;
1683  vec_add1 (dm->conf->eal_init_args, (u8 *) "vnet");
1684  vec_add1 (dm->conf->eal_init_args, (u8 *) "--in-memory");
1685
1686  /* Default vlib_buffer_t flags, DISABLES tcp/udp checksumming... */
1687  dm->buffer_flags_template = (VLIB_BUFFER_TOTAL_LENGTH_VALID |
1688			       VLIB_BUFFER_EXT_HDR_VALID |
1689			       VNET_BUFFER_F_L4_CHECKSUM_COMPUTED |
1690			       VNET_BUFFER_F_L4_CHECKSUM_CORRECT);
1691
1692  dm->stat_poll_interval = DPDK_STATS_POLL_INTERVAL;
1693  dm->link_state_poll_interval = DPDK_LINK_POLL_INTERVAL;
1694
1695  dm->log_default = vlib_log_register_class ("dpdk", 0);
1696
1697  return error;
1698}
1699
1700VLIB_INIT_FUNCTION (dpdk_init);
1701
1702/*
1703 * fd.io coding-style-patch-verification: ON
1704 *
1705 * Local Variables:
1706 * eval: (c-set-style "gnu")
1707 * End:
1708 */
1709