drivers/net/bonding/rte_eth_bond_api.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2017 Intel Corporation
   3  */
   4
   5 #include <string.h>
   6
   7 #include <rte_mbuf.h>
   8 #include <rte_malloc.h>
   9 #include <ethdev_driver.h>
  10 #include <rte_tcp.h>
  11 #include <rte_bus_vdev.h>
  12 #include <rte_kvargs.h>
  13
  14 #include "rte_eth_bond.h"
  15 #include "eth_bond_private.h"
  16 #include "eth_bond_8023ad_private.h"
  17
  18 int
  19 check_for_bonded_ethdev(const struct rte_eth_dev *eth_dev)
  20 {
  21         /* Check valid pointer */
  22         if (eth_dev == NULL ||
  23                 eth_dev->device == NULL ||
  24                 eth_dev->device->driver == NULL ||
  25                 eth_dev->device->driver->name == NULL)
  26                 return -1;
  27
  28         /* return 0 if driver name matches */
  29         return eth_dev->device->driver->name != pmd_bond_drv.driver.name;
  30 }
  31
  32 int
  33 valid_bonded_port_id(uint16_t port_id)
  34 {
  35         RTE_ETH_VALID_PORTID_OR_ERR_RET(port_id, -1);
  36         return check_for_bonded_ethdev(&rte_eth_devices[port_id]);
  37 }
  38
  39 int
  40 check_for_master_bonded_ethdev(const struct rte_eth_dev *eth_dev)
  41 {
  42         int i;
  43         struct bond_dev_private *internals;
  44
  45         if (check_for_bonded_ethdev(eth_dev) != 0)
  46                 return 0;
  47
  48         internals = eth_dev->data->dev_private;
  49
  50         /* Check if any of slave devices is a bonded device */
  51         for (i = 0; i < internals->slave_count; i++)
  52                 if (valid_bonded_port_id(internals->slaves[i].port_id) == 0)
  53                         return 1;
  54
  55         return 0;
  56 }
  57
  58 int
  59 valid_slave_port_id(struct bond_dev_private *internals, uint16_t slave_port_id)
  60 {
  61         RTE_ETH_VALID_PORTID_OR_ERR_RET(slave_port_id, -1);
  62
  63         /* Verify that slave_port_id refers to a non bonded port */
  64         if (check_for_bonded_ethdev(&rte_eth_devices[slave_port_id]) == 0 &&
  65                         internals->mode == BONDING_MODE_8023AD) {
  66                 RTE_BOND_LOG(ERR, "Cannot add slave to bonded device in 802.3ad"
  67                                 " mode as slave is also a bonded device, only "
  68                                 "physical devices can be support in this mode.");
  69                 return -1;
  70         }
  71
  72         if (internals->port_id == slave_port_id) {
  73                 RTE_BOND_LOG(ERR,
  74                         "Cannot add the bonded device itself as its slave.");
  75                 return -1;
  76         }
  77
  78         return 0;
  79 }
  80
  81 void
  82 activate_slave(struct rte_eth_dev *eth_dev, uint16_t port_id)
  83 {
  84         struct bond_dev_private *internals = eth_dev->data->dev_private;
  85         uint16_t active_count = internals->active_slave_count;
  86
  87         if (internals->mode == BONDING_MODE_8023AD)
  88                 bond_mode_8023ad_activate_slave(eth_dev, port_id);
  89
  90         if (internals->mode == BONDING_MODE_TLB
  91                         || internals->mode == BONDING_MODE_ALB) {
  92
  93                 internals->tlb_slaves_order[active_count] = port_id;
  94         }
  95
  96         RTE_ASSERT(internals->active_slave_count <
  97                         (RTE_DIM(internals->active_slaves) - 1));
  98
  99         internals->active_slaves[internals->active_slave_count] = port_id;
 100         internals->active_slave_count++;
 101
 102         if (internals->mode == BONDING_MODE_TLB)
 103                 bond_tlb_activate_slave(internals);
 104         if (internals->mode == BONDING_MODE_ALB)
 105                 bond_mode_alb_client_list_upd(eth_dev);
 106 }
 107
 108 void
 109 deactivate_slave(struct rte_eth_dev *eth_dev, uint16_t port_id)
 110 {
 111         uint16_t slave_pos;
 112         struct bond_dev_private *internals = eth_dev->data->dev_private;
 113         uint16_t active_count = internals->active_slave_count;
 114
 115         if (internals->mode == BONDING_MODE_8023AD) {
 116                 bond_mode_8023ad_stop(eth_dev);
 117                 bond_mode_8023ad_deactivate_slave(eth_dev, port_id);
 118         } else if (internals->mode == BONDING_MODE_TLB
 119                         || internals->mode == BONDING_MODE_ALB)
 120                 bond_tlb_disable(internals);
 121
 122         slave_pos = find_slave_by_id(internals->active_slaves, active_count,
 123                         port_id);
 124
 125         /* If slave was not at the end of the list
 126          * shift active slaves up active array list */
 127         if (slave_pos < active_count) {
 128                 active_count--;
 129                 memmove(internals->active_slaves + slave_pos,
 130                                 internals->active_slaves + slave_pos + 1,
 131                                 (active_count - slave_pos) *
 132                                         sizeof(internals->active_slaves[0]));
 133         }
 134
 135         RTE_ASSERT(active_count < RTE_DIM(internals->active_slaves));
 136         internals->active_slave_count = active_count;
 137
 138         if (eth_dev->data->dev_started) {
 139                 if (internals->mode == BONDING_MODE_8023AD) {
 140                         bond_mode_8023ad_start(eth_dev);
 141                 } else if (internals->mode == BONDING_MODE_TLB) {
 142                         bond_tlb_enable(internals);
 143                 } else if (internals->mode == BONDING_MODE_ALB) {
 144                         bond_tlb_enable(internals);
 145                         bond_mode_alb_client_list_upd(eth_dev);
 146                 }
 147         }
 148 }
 149
 150 int
 151 rte_eth_bond_create(const char *name, uint8_t mode, uint8_t socket_id)
 152 {
 153         struct bond_dev_private *internals;
 154         char devargs[52];
 155         uint16_t port_id;
 156         int ret;
 157
 158         if (name == NULL) {
 159                 RTE_BOND_LOG(ERR, "Invalid name specified");
 160                 return -EINVAL;
 161         }
 162
 163         ret = snprintf(devargs, sizeof(devargs),
 164                 "driver=net_bonding,mode=%d,socket_id=%d", mode, socket_id);
 165         if (ret < 0 || ret >= (int)sizeof(devargs))
 166                 return -ENOMEM;
 167
 168         ret = rte_vdev_init(name, devargs);
 169         if (ret)
 170                 return ret;
 171
 172         ret = rte_eth_dev_get_port_by_name(name, &port_id);
 173         RTE_ASSERT(!ret);
 174
 175         /*
 176          * To make bond_ethdev_configure() happy we need to free the
 177          * internals->kvlist here.
 178          *
 179          * Also see comment in bond_ethdev_configure().
 180          */
 181         internals = rte_eth_devices[port_id].data->dev_private;
 182         rte_kvargs_free(internals->kvlist);
 183         internals->kvlist = NULL;
 184
 185         return port_id;
 186 }
 187
 188 int
 189 rte_eth_bond_free(const char *name)
 190 {
 191         return rte_vdev_uninit(name);
 192 }
 193
 194 static int
 195 slave_vlan_filter_set(uint16_t bonded_port_id, uint16_t slave_port_id)
 196 {
 197         struct rte_eth_dev *bonded_eth_dev;
 198         struct bond_dev_private *internals;
 199         int found;
 200         int res = 0;
 201         uint64_t slab = 0;
 202         uint32_t pos = 0;
 203         uint16_t first;
 204
 205         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 206         if ((bonded_eth_dev->data->dev_conf.rxmode.offloads &
 207                         DEV_RX_OFFLOAD_VLAN_FILTER) == 0)
 208                 return 0;
 209
 210         internals = bonded_eth_dev->data->dev_private;
 211         found = rte_bitmap_scan(internals->vlan_filter_bmp, &pos, &slab);
 212         first = pos;
 213
 214         if (!found)
 215                 return 0;
 216
 217         do {
 218                 uint32_t i;
 219                 uint64_t mask;
 220
 221                 for (i = 0, mask = 1;
 222                      i < RTE_BITMAP_SLAB_BIT_SIZE;
 223                      i ++, mask <<= 1) {
 224                         if (unlikely(slab & mask)) {
 225                                 uint16_t vlan_id = pos + i;
 226
 227                                 res = rte_eth_dev_vlan_filter(slave_port_id,
 228                                                               vlan_id, 1);
 229                         }
 230                 }
 231                 found = rte_bitmap_scan(internals->vlan_filter_bmp,
 232                                         &pos, &slab);
 233         } while (found && first != pos && res == 0);
 234
 235         return res;
 236 }
 237
 238 static int
 239 slave_rte_flow_prepare(uint16_t slave_id, struct bond_dev_private *internals)
 240 {
 241         struct rte_flow *flow;
 242         struct rte_flow_error ferror;
 243         uint16_t slave_port_id = internals->slaves[slave_id].port_id;
 244
 245         if (internals->flow_isolated_valid != 0) {
 246                 if (rte_eth_dev_stop(slave_port_id) != 0) {
 247                         RTE_BOND_LOG(ERR, "Failed to stop device on port %u",
 248                                      slave_port_id);
 249                         return -1;
 250                 }
 251
 252                 if (rte_flow_isolate(slave_port_id, internals->flow_isolated,
 253                     &ferror)) {
 254                         RTE_BOND_LOG(ERR, "rte_flow_isolate failed for slave"
 255                                      " %d: %s", slave_id, ferror.message ?
 256                                      ferror.message : "(no stated reason)");
 257                         return -1;
 258                 }
 259         }
 260         TAILQ_FOREACH(flow, &internals->flow_list, next) {
 261                 flow->flows[slave_id] = rte_flow_create(slave_port_id,
 262                                                         flow->rule.attr,
 263                                                         flow->rule.pattern,
 264                                                         flow->rule.actions,
 265                                                         &ferror);
 266                 if (flow->flows[slave_id] == NULL) {
 267                         RTE_BOND_LOG(ERR, "Cannot create flow for slave"
 268                                      " %d: %s", slave_id,
 269                                      ferror.message ? ferror.message :
 270                                      "(no stated reason)");
 271                         /* Destroy successful bond flows from the slave */
 272                         TAILQ_FOREACH(flow, &internals->flow_list, next) {
 273                                 if (flow->flows[slave_id] != NULL) {
 274                                         rte_flow_destroy(slave_port_id,
 275                                                          flow->flows[slave_id],
 276                                                          &ferror);
 277                                         flow->flows[slave_id] = NULL;
 278                                 }
 279                         }
 280                         return -1;
 281                 }
 282         }
 283         return 0;
 284 }
 285
 286 static void
 287 eth_bond_slave_inherit_dev_info_rx_first(struct bond_dev_private *internals,
 288                                          const struct rte_eth_dev_info *di)
 289 {
 290         struct rte_eth_rxconf *rxconf_i = &internals->default_rxconf;
 291
 292         internals->reta_size = di->reta_size;
 293
 294         /* Inherit Rx offload capabilities from the first slave device */
 295         internals->rx_offload_capa = di->rx_offload_capa;
 296         internals->rx_queue_offload_capa = di->rx_queue_offload_capa;
 297         internals->flow_type_rss_offloads = di->flow_type_rss_offloads;
 298
 299         /* Inherit maximum Rx packet size from the first slave device */
 300         internals->candidate_max_rx_pktlen = di->max_rx_pktlen;
 301
 302         /* Inherit default Rx queue settings from the first slave device */
 303         memcpy(rxconf_i, &di->default_rxconf, sizeof(*rxconf_i));
 304
 305         /*
 306          * Turn off descriptor prefetch and writeback by default for all
 307          * slave devices. Applications may tweak this setting if need be.
 308          */
 309         rxconf_i->rx_thresh.pthresh = 0;
 310         rxconf_i->rx_thresh.hthresh = 0;
 311         rxconf_i->rx_thresh.wthresh = 0;
 312
 313         /* Setting this to zero should effectively enable default values */
 314         rxconf_i->rx_free_thresh = 0;
 315
 316         /* Disable deferred start by default for all slave devices */
 317         rxconf_i->rx_deferred_start = 0;
 318 }
 319
 320 static void
 321 eth_bond_slave_inherit_dev_info_tx_first(struct bond_dev_private *internals,
 322                                          const struct rte_eth_dev_info *di)
 323 {
 324         struct rte_eth_txconf *txconf_i = &internals->default_txconf;
 325
 326         /* Inherit Tx offload capabilities from the first slave device */
 327         internals->tx_offload_capa = di->tx_offload_capa;
 328         internals->tx_queue_offload_capa = di->tx_queue_offload_capa;
 329
 330         /* Inherit default Tx queue settings from the first slave device */
 331         memcpy(txconf_i, &di->default_txconf, sizeof(*txconf_i));
 332
 333         /*
 334          * Turn off descriptor prefetch and writeback by default for all
 335          * slave devices. Applications may tweak this setting if need be.
 336          */
 337         txconf_i->tx_thresh.pthresh = 0;
 338         txconf_i->tx_thresh.hthresh = 0;
 339         txconf_i->tx_thresh.wthresh = 0;
 340
 341         /*
 342          * Setting these parameters to zero assumes that default
 343          * values will be configured implicitly by slave devices.
 344          */
 345         txconf_i->tx_free_thresh = 0;
 346         txconf_i->tx_rs_thresh = 0;
 347
 348         /* Disable deferred start by default for all slave devices */
 349         txconf_i->tx_deferred_start = 0;
 350 }
 351
 352 static void
 353 eth_bond_slave_inherit_dev_info_rx_next(struct bond_dev_private *internals,
 354                                         const struct rte_eth_dev_info *di)
 355 {
 356         struct rte_eth_rxconf *rxconf_i = &internals->default_rxconf;
 357         const struct rte_eth_rxconf *rxconf = &di->default_rxconf;
 358
 359         internals->rx_offload_capa &= di->rx_offload_capa;
 360         internals->rx_queue_offload_capa &= di->rx_queue_offload_capa;
 361         internals->flow_type_rss_offloads &= di->flow_type_rss_offloads;
 362
 363         /*
 364          * If at least one slave device suggests enabling this
 365          * setting by default, enable it for all slave devices
 366          * since disabling it may not be necessarily supported.
 367          */
 368         if (rxconf->rx_drop_en == 1)
 369                 rxconf_i->rx_drop_en = 1;
 370
 371         /*
 372          * Adding a new slave device may cause some of previously inherited
 373          * offloads to be withdrawn from the internal rx_queue_offload_capa
 374          * value. Thus, the new internal value of default Rx queue offloads
 375          * has to be masked by rx_queue_offload_capa to make sure that only
 376          * commonly supported offloads are preserved from both the previous
 377          * value and the value being inhereted from the new slave device.
 378          */
 379         rxconf_i->offloads = (rxconf_i->offloads | rxconf->offloads) &
 380                              internals->rx_queue_offload_capa;
 381
 382         /*
 383          * RETA size is GCD of all slaves RETA sizes, so, if all sizes will be
 384          * the power of 2, the lower one is GCD
 385          */
 386         if (internals->reta_size > di->reta_size)
 387                 internals->reta_size = di->reta_size;
 388
 389         if (!internals->max_rx_pktlen &&
 390             di->max_rx_pktlen < internals->candidate_max_rx_pktlen)
 391                 internals->candidate_max_rx_pktlen = di->max_rx_pktlen;
 392 }
 393
 394 static void
 395 eth_bond_slave_inherit_dev_info_tx_next(struct bond_dev_private *internals,
 396                                         const struct rte_eth_dev_info *di)
 397 {
 398         struct rte_eth_txconf *txconf_i = &internals->default_txconf;
 399         const struct rte_eth_txconf *txconf = &di->default_txconf;
 400
 401         internals->tx_offload_capa &= di->tx_offload_capa;
 402         internals->tx_queue_offload_capa &= di->tx_queue_offload_capa;
 403
 404         /*
 405          * Adding a new slave device may cause some of previously inherited
 406          * offloads to be withdrawn from the internal tx_queue_offload_capa
 407          * value. Thus, the new internal value of default Tx queue offloads
 408          * has to be masked by tx_queue_offload_capa to make sure that only
 409          * commonly supported offloads are preserved from both the previous
 410          * value and the value being inhereted from the new slave device.
 411          */
 412         txconf_i->offloads = (txconf_i->offloads | txconf->offloads) &
 413                              internals->tx_queue_offload_capa;
 414 }
 415
 416 static void
 417 eth_bond_slave_inherit_desc_lim_first(struct rte_eth_desc_lim *bond_desc_lim,
 418                 const struct rte_eth_desc_lim *slave_desc_lim)
 419 {
 420         memcpy(bond_desc_lim, slave_desc_lim, sizeof(*bond_desc_lim));
 421 }
 422
 423 static int
 424 eth_bond_slave_inherit_desc_lim_next(struct rte_eth_desc_lim *bond_desc_lim,
 425                 const struct rte_eth_desc_lim *slave_desc_lim)
 426 {
 427         bond_desc_lim->nb_max = RTE_MIN(bond_desc_lim->nb_max,
 428                                         slave_desc_lim->nb_max);
 429         bond_desc_lim->nb_min = RTE_MAX(bond_desc_lim->nb_min,
 430                                         slave_desc_lim->nb_min);
 431         bond_desc_lim->nb_align = RTE_MAX(bond_desc_lim->nb_align,
 432                                           slave_desc_lim->nb_align);
 433
 434         if (bond_desc_lim->nb_min > bond_desc_lim->nb_max ||
 435             bond_desc_lim->nb_align > bond_desc_lim->nb_max) {
 436                 RTE_BOND_LOG(ERR, "Failed to inherit descriptor limits");
 437                 return -EINVAL;
 438         }
 439
 440         /* Treat maximum number of segments equal to 0 as unspecified */
 441         if (slave_desc_lim->nb_seg_max != 0 &&
 442             (bond_desc_lim->nb_seg_max == 0 ||
 443              slave_desc_lim->nb_seg_max < bond_desc_lim->nb_seg_max))
 444                 bond_desc_lim->nb_seg_max = slave_desc_lim->nb_seg_max;
 445         if (slave_desc_lim->nb_mtu_seg_max != 0 &&
 446             (bond_desc_lim->nb_mtu_seg_max == 0 ||
 447              slave_desc_lim->nb_mtu_seg_max < bond_desc_lim->nb_mtu_seg_max))
 448                 bond_desc_lim->nb_mtu_seg_max = slave_desc_lim->nb_mtu_seg_max;
 449
 450         return 0;
 451 }
 452
 453 static int
 454 __eth_bond_slave_add_lock_free(uint16_t bonded_port_id, uint16_t slave_port_id)
 455 {
 456         struct rte_eth_dev *bonded_eth_dev, *slave_eth_dev;
 457         struct bond_dev_private *internals;
 458         struct rte_eth_link link_props;
 459         struct rte_eth_dev_info dev_info;
 460         int ret;
 461
 462         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 463         internals = bonded_eth_dev->data->dev_private;
 464
 465         if (valid_slave_port_id(internals, slave_port_id) != 0)
 466                 return -1;
 467
 468         slave_eth_dev = &rte_eth_devices[slave_port_id];
 469         if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_BONDED_SLAVE) {
 470                 RTE_BOND_LOG(ERR, "Slave device is already a slave of a bonded device");
 471                 return -1;
 472         }
 473
 474         ret = rte_eth_dev_info_get(slave_port_id, &dev_info);
 475         if (ret != 0) {
 476                 RTE_BOND_LOG(ERR,
 477                         "%s: Error during getting device (port %u) info: %s\n",
 478                         __func__, slave_port_id, strerror(-ret));
 479
 480                 return ret;
 481         }
 482         if (dev_info.max_rx_pktlen < internals->max_rx_pktlen) {
 483                 RTE_BOND_LOG(ERR, "Slave (port %u) max_rx_pktlen too small",
 484                              slave_port_id);
 485                 return -1;
 486         }
 487
 488         slave_add(internals, slave_eth_dev);
 489
 490         /* We need to store slaves reta_size to be able to synchronize RETA for all
 491          * slave devices even if its sizes are different.
 492          */
 493         internals->slaves[internals->slave_count].reta_size = dev_info.reta_size;
 494
 495         if (internals->slave_count < 1) {
 496                 /* if MAC is not user defined then use MAC of first slave add to
 497                  * bonded device */
 498                 if (!internals->user_defined_mac) {
 499                         if (mac_address_set(bonded_eth_dev,
 500                                             slave_eth_dev->data->mac_addrs)) {
 501                                 RTE_BOND_LOG(ERR, "Failed to set MAC address");
 502                                 return -1;
 503                         }
 504                 }
 505
 506                 /* Make primary slave */
 507                 internals->primary_port = slave_port_id;
 508                 internals->current_primary_port = slave_port_id;
 509
 510                 /* Inherit queues settings from first slave */
 511                 internals->nb_rx_queues = slave_eth_dev->data->nb_rx_queues;
 512                 internals->nb_tx_queues = slave_eth_dev->data->nb_tx_queues;
 513
 514                 eth_bond_slave_inherit_dev_info_rx_first(internals, &dev_info);
 515                 eth_bond_slave_inherit_dev_info_tx_first(internals, &dev_info);
 516
 517                 eth_bond_slave_inherit_desc_lim_first(&internals->rx_desc_lim,
 518                                                       &dev_info.rx_desc_lim);
 519                 eth_bond_slave_inherit_desc_lim_first(&internals->tx_desc_lim,
 520                                                       &dev_info.tx_desc_lim);
 521         } else {
 522                 int ret;
 523
 524                 eth_bond_slave_inherit_dev_info_rx_next(internals, &dev_info);
 525                 eth_bond_slave_inherit_dev_info_tx_next(internals, &dev_info);
 526
 527                 ret = eth_bond_slave_inherit_desc_lim_next(
 528                                 &internals->rx_desc_lim, &dev_info.rx_desc_lim);
 529                 if (ret != 0)
 530                         return ret;
 531
 532                 ret = eth_bond_slave_inherit_desc_lim_next(
 533                                 &internals->tx_desc_lim, &dev_info.tx_desc_lim);
 534                 if (ret != 0)
 535                         return ret;
 536         }
 537
 538         bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf &=
 539                         internals->flow_type_rss_offloads;
 540
 541         if (slave_rte_flow_prepare(internals->slave_count, internals) != 0) {
 542                 RTE_BOND_LOG(ERR, "Failed to prepare new slave flows: port=%d",
 543                              slave_port_id);
 544                 return -1;
 545         }
 546
 547         /* Add additional MAC addresses to the slave */
 548         if (slave_add_mac_addresses(bonded_eth_dev, slave_port_id) != 0) {
 549                 RTE_BOND_LOG(ERR, "Failed to add mac address(es) to slave %hu",
 550                                 slave_port_id);
 551                 return -1;
 552         }
 553
 554         internals->slave_count++;
 555
 556         if (bonded_eth_dev->data->dev_started) {
 557                 if (slave_configure(bonded_eth_dev, slave_eth_dev) != 0) {
 558                         internals->slave_count--;
 559                         RTE_BOND_LOG(ERR, "rte_bond_slaves_configure: port=%d",
 560                                         slave_port_id);
 561                         return -1;
 562                 }
 563         }
 564
 565         /* Update all slave devices MACs */
 566         mac_address_slaves_update(bonded_eth_dev);
 567
 568         /* Register link status change callback with bonded device pointer as
 569          * argument*/
 570         rte_eth_dev_callback_register(slave_port_id, RTE_ETH_EVENT_INTR_LSC,
 571                         bond_ethdev_lsc_event_callback, &bonded_eth_dev->data->port_id);
 572
 573         /* If bonded device is started then we can add the slave to our active
 574          * slave array */
 575         if (bonded_eth_dev->data->dev_started) {
 576                 ret = rte_eth_link_get_nowait(slave_port_id, &link_props);
 577                 if (ret < 0) {
 578                         rte_eth_dev_callback_unregister(slave_port_id,
 579                                         RTE_ETH_EVENT_INTR_LSC,
 580                                         bond_ethdev_lsc_event_callback,
 581                                         &bonded_eth_dev->data->port_id);
 582                         internals->slave_count--;
 583                         RTE_BOND_LOG(ERR,
 584                                 "Slave (port %u) link get failed: %s\n",
 585                                 slave_port_id, rte_strerror(-ret));
 586                         return -1;
 587                 }
 588
 589                  if (link_props.link_status == ETH_LINK_UP) {
 590                         if (internals->active_slave_count == 0 &&
 591                             !internals->user_defined_primary_port)
 592                                 bond_ethdev_primary_set(internals,
 593                                                         slave_port_id);
 594                 }
 595         }
 596
 597         /* Add slave details to bonded device */
 598         slave_eth_dev->data->dev_flags |= RTE_ETH_DEV_BONDED_SLAVE;
 599
 600         slave_vlan_filter_set(bonded_port_id, slave_port_id);
 601
 602         return 0;
 603
 604 }
 605
 606 int
 607 rte_eth_bond_slave_add(uint16_t bonded_port_id, uint16_t slave_port_id)
 608 {
 609         struct rte_eth_dev *bonded_eth_dev;
 610         struct bond_dev_private *internals;
 611
 612         int retval;
 613
 614         if (valid_bonded_port_id(bonded_port_id) != 0)
 615                 return -1;
 616
 617         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 618         internals = bonded_eth_dev->data->dev_private;
 619
 620         if (valid_slave_port_id(internals, slave_port_id) != 0)
 621                 return -1;
 622
 623         rte_spinlock_lock(&internals->lock);
 624
 625         retval = __eth_bond_slave_add_lock_free(bonded_port_id, slave_port_id);
 626
 627         rte_spinlock_unlock(&internals->lock);
 628
 629         return retval;
 630 }
 631
 632 static int
 633 __eth_bond_slave_remove_lock_free(uint16_t bonded_port_id,
 634                                    uint16_t slave_port_id)
 635 {
 636         struct rte_eth_dev *bonded_eth_dev;
 637         struct bond_dev_private *internals;
 638         struct rte_eth_dev *slave_eth_dev;
 639         struct rte_flow_error flow_error;
 640         struct rte_flow *flow;
 641         int i, slave_idx;
 642
 643         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 644         internals = bonded_eth_dev->data->dev_private;
 645
 646         if (valid_slave_port_id(internals, slave_port_id) < 0)
 647                 return -1;
 648
 649         /* first remove from active slave list */
 650         slave_idx = find_slave_by_id(internals->active_slaves,
 651                 internals->active_slave_count, slave_port_id);
 652
 653         if (slave_idx < internals->active_slave_count)
 654                 deactivate_slave(bonded_eth_dev, slave_port_id);
 655
 656         slave_idx = -1;
 657         /* now find in slave list */
 658         for (i = 0; i < internals->slave_count; i++)
 659                 if (internals->slaves[i].port_id == slave_port_id) {
 660                         slave_idx = i;
 661                         break;
 662                 }
 663
 664         if (slave_idx < 0) {
 665                 RTE_BOND_LOG(ERR, "Couldn't find slave in port list, slave count %d",
 666                                 internals->slave_count);
 667                 return -1;
 668         }
 669
 670         /* Un-register link status change callback with bonded device pointer as
 671          * argument*/
 672         rte_eth_dev_callback_unregister(slave_port_id, RTE_ETH_EVENT_INTR_LSC,
 673                         bond_ethdev_lsc_event_callback,
 674                         &rte_eth_devices[bonded_port_id].data->port_id);
 675
 676         /* Restore original MAC address of slave device */
 677         rte_eth_dev_default_mac_addr_set(slave_port_id,
 678                         &(internals->slaves[slave_idx].persisted_mac_addr));
 679
 680         /* remove additional MAC addresses from the slave */
 681         slave_remove_mac_addresses(bonded_eth_dev, slave_port_id);
 682
 683         /*
 684          * Remove bond device flows from slave device.
 685          * Note: don't restore flow isolate mode.
 686          */
 687         TAILQ_FOREACH(flow, &internals->flow_list, next) {
 688                 if (flow->flows[slave_idx] != NULL) {
 689                         rte_flow_destroy(slave_port_id, flow->flows[slave_idx],
 690                                          &flow_error);
 691                         flow->flows[slave_idx] = NULL;
 692                 }
 693         }
 694
 695         slave_eth_dev = &rte_eth_devices[slave_port_id];
 696         slave_remove(internals, slave_eth_dev);
 697         slave_eth_dev->data->dev_flags &= (~RTE_ETH_DEV_BONDED_SLAVE);
 698
 699         /*  first slave in the active list will be the primary by default,
 700          *  otherwise use first device in list */
 701         if (internals->current_primary_port == slave_port_id) {
 702                 if (internals->active_slave_count > 0)
 703                         internals->current_primary_port = internals->active_slaves[0];
 704                 else if (internals->slave_count > 0)
 705                         internals->current_primary_port = internals->slaves[0].port_id;
 706                 else
 707                         internals->primary_port = 0;
 708                 mac_address_slaves_update(bonded_eth_dev);
 709         }
 710
 711         if (internals->active_slave_count < 1) {
 712                 /* if no slaves are any longer attached to bonded device and MAC is not
 713                  * user defined then clear MAC of bonded device as it will be reset
 714                  * when a new slave is added */
 715                 if (internals->slave_count < 1 && !internals->user_defined_mac)
 716                         memset(rte_eth_devices[bonded_port_id].data->mac_addrs, 0,
 717                                         sizeof(*(rte_eth_devices[bonded_port_id].data->mac_addrs)));
 718         }
 719         if (internals->slave_count == 0) {
 720                 internals->rx_offload_capa = 0;
 721                 internals->tx_offload_capa = 0;
 722                 internals->rx_queue_offload_capa = 0;
 723                 internals->tx_queue_offload_capa = 0;
 724                 internals->flow_type_rss_offloads = ETH_RSS_PROTO_MASK;
 725                 internals->reta_size = 0;
 726                 internals->candidate_max_rx_pktlen = 0;
 727                 internals->max_rx_pktlen = 0;
 728         }
 729         return 0;
 730 }
 731
 732 int
 733 rte_eth_bond_slave_remove(uint16_t bonded_port_id, uint16_t slave_port_id)
 734 {
 735         struct rte_eth_dev *bonded_eth_dev;
 736         struct bond_dev_private *internals;
 737         int retval;
 738
 739         if (valid_bonded_port_id(bonded_port_id) != 0)
 740                 return -1;
 741
 742         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 743         internals = bonded_eth_dev->data->dev_private;
 744
 745         rte_spinlock_lock(&internals->lock);
 746
 747         retval = __eth_bond_slave_remove_lock_free(bonded_port_id, slave_port_id);
 748
 749         rte_spinlock_unlock(&internals->lock);
 750
 751         return retval;
 752 }
 753
 754 int
 755 rte_eth_bond_mode_set(uint16_t bonded_port_id, uint8_t mode)
 756 {
 757         struct rte_eth_dev *bonded_eth_dev;
 758
 759         if (valid_bonded_port_id(bonded_port_id) != 0)
 760                 return -1;
 761
 762         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 763
 764         if (check_for_master_bonded_ethdev(bonded_eth_dev) != 0 &&
 765                         mode == BONDING_MODE_8023AD)
 766                 return -1;
 767
 768         return bond_ethdev_mode_set(bonded_eth_dev, mode);
 769 }
 770
 771 int
 772 rte_eth_bond_mode_get(uint16_t bonded_port_id)
 773 {
 774         struct bond_dev_private *internals;
 775
 776         if (valid_bonded_port_id(bonded_port_id) != 0)
 777                 return -1;
 778
 779         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 780
 781         return internals->mode;
 782 }
 783
 784 int
 785 rte_eth_bond_primary_set(uint16_t bonded_port_id, uint16_t slave_port_id)
 786 {
 787         struct bond_dev_private *internals;
 788
 789         if (valid_bonded_port_id(bonded_port_id) != 0)
 790                 return -1;
 791
 792         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 793
 794         if (valid_slave_port_id(internals, slave_port_id) != 0)
 795                 return -1;
 796
 797         internals->user_defined_primary_port = 1;
 798         internals->primary_port = slave_port_id;
 799
 800         bond_ethdev_primary_set(internals, slave_port_id);
 801
 802         return 0;
 803 }
 804
 805 int
 806 rte_eth_bond_primary_get(uint16_t bonded_port_id)
 807 {
 808         struct bond_dev_private *internals;
 809
 810         if (valid_bonded_port_id(bonded_port_id) != 0)
 811                 return -1;
 812
 813         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 814
 815         if (internals->slave_count < 1)
 816                 return -1;
 817
 818         return internals->current_primary_port;
 819 }
 820
 821 int
 822 rte_eth_bond_slaves_get(uint16_t bonded_port_id, uint16_t slaves[],
 823                         uint16_t len)
 824 {
 825         struct bond_dev_private *internals;
 826         uint16_t i;
 827
 828         if (valid_bonded_port_id(bonded_port_id) != 0)
 829                 return -1;
 830
 831         if (slaves == NULL)
 832                 return -1;
 833
 834         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 835
 836         if (internals->slave_count > len)
 837                 return -1;
 838
 839         for (i = 0; i < internals->slave_count; i++)
 840                 slaves[i] = internals->slaves[i].port_id;
 841
 842         return internals->slave_count;
 843 }
 844
 845 int
 846 rte_eth_bond_active_slaves_get(uint16_t bonded_port_id, uint16_t slaves[],
 847                 uint16_t len)
 848 {
 849         struct bond_dev_private *internals;
 850
 851         if (valid_bonded_port_id(bonded_port_id) != 0)
 852                 return -1;
 853
 854         if (slaves == NULL)
 855                 return -1;
 856
 857         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 858
 859         if (internals->active_slave_count > len)
 860                 return -1;
 861
 862         memcpy(slaves, internals->active_slaves,
 863         internals->active_slave_count * sizeof(internals->active_slaves[0]));
 864
 865         return internals->active_slave_count;
 866 }
 867
 868 int
 869 rte_eth_bond_mac_address_set(uint16_t bonded_port_id,
 870                 struct rte_ether_addr *mac_addr)
 871 {
 872         struct rte_eth_dev *bonded_eth_dev;
 873         struct bond_dev_private *internals;
 874
 875         if (valid_bonded_port_id(bonded_port_id) != 0)
 876                 return -1;
 877
 878         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 879         internals = bonded_eth_dev->data->dev_private;
 880
 881         /* Set MAC Address of Bonded Device */
 882         if (mac_address_set(bonded_eth_dev, mac_addr))
 883                 return -1;
 884
 885         internals->user_defined_mac = 1;
 886
 887         /* Update all slave devices MACs*/
 888         if (internals->slave_count > 0)
 889                 return mac_address_slaves_update(bonded_eth_dev);
 890
 891         return 0;
 892 }
 893
 894 int
 895 rte_eth_bond_mac_address_reset(uint16_t bonded_port_id)
 896 {
 897         struct rte_eth_dev *bonded_eth_dev;
 898         struct bond_dev_private *internals;
 899
 900         if (valid_bonded_port_id(bonded_port_id) != 0)
 901                 return -1;
 902
 903         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 904         internals = bonded_eth_dev->data->dev_private;
 905
 906         internals->user_defined_mac = 0;
 907
 908         if (internals->slave_count > 0) {
 909                 int slave_port;
 910                 /* Get the primary slave location based on the primary port
 911                  * number as, while slave_add(), we will keep the primary
 912                  * slave based on slave_count,but not based on the primary port.
 913                  */
 914                 for (slave_port = 0; slave_port < internals->slave_count;
 915                      slave_port++) {
 916                         if (internals->slaves[slave_port].port_id ==
 917                             internals->primary_port)
 918                                 break;
 919                 }
 920
 921                 /* Set MAC Address of Bonded Device */
 922                 if (mac_address_set(bonded_eth_dev,
 923                         &internals->slaves[slave_port].persisted_mac_addr)
 924                                 != 0) {
 925                         RTE_BOND_LOG(ERR, "Failed to set MAC address on bonded device");
 926                         return -1;
 927                 }
 928                 /* Update all slave devices MAC addresses */
 929                 return mac_address_slaves_update(bonded_eth_dev);
 930         }
 931         /* No need to update anything as no slaves present */
 932         return 0;
 933 }
 934
 935 int
 936 rte_eth_bond_xmit_policy_set(uint16_t bonded_port_id, uint8_t policy)
 937 {
 938         struct bond_dev_private *internals;
 939
 940         if (valid_bonded_port_id(bonded_port_id) != 0)
 941                 return -1;
 942
 943         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 944
 945         switch (policy) {
 946         case BALANCE_XMIT_POLICY_LAYER2:
 947                 internals->balance_xmit_policy = policy;
 948                 internals->burst_xmit_hash = burst_xmit_l2_hash;
 949                 break;
 950         case BALANCE_XMIT_POLICY_LAYER23:
 951                 internals->balance_xmit_policy = policy;
 952                 internals->burst_xmit_hash = burst_xmit_l23_hash;
 953                 break;
 954         case BALANCE_XMIT_POLICY_LAYER34:
 955                 internals->balance_xmit_policy = policy;
 956                 internals->burst_xmit_hash = burst_xmit_l34_hash;
 957                 break;
 958
 959         default:
 960                 return -1;
 961         }
 962         return 0;
 963 }
 964
 965 int
 966 rte_eth_bond_xmit_policy_get(uint16_t bonded_port_id)
 967 {
 968         struct bond_dev_private *internals;
 969
 970         if (valid_bonded_port_id(bonded_port_id) != 0)
 971                 return -1;
 972
 973         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 974
 975         return internals->balance_xmit_policy;
 976 }
 977
 978 int
 979 rte_eth_bond_link_monitoring_set(uint16_t bonded_port_id, uint32_t internal_ms)
 980 {
 981         struct bond_dev_private *internals;
 982
 983         if (valid_bonded_port_id(bonded_port_id) != 0)
 984                 return -1;
 985
 986         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 987         internals->link_status_polling_interval_ms = internal_ms;
 988
 989         return 0;
 990 }
 991
 992 int
 993 rte_eth_bond_link_monitoring_get(uint16_t bonded_port_id)
 994 {
 995         struct bond_dev_private *internals;
 996
 997         if (valid_bonded_port_id(bonded_port_id) != 0)
 998                 return -1;
 999
1000         internals = rte_eth_devices[bonded_port_id].data->dev_private;
1001
1002         return internals->link_status_polling_interval_ms;
1003 }
1004
1005 int
1006 rte_eth_bond_link_down_prop_delay_set(uint16_t bonded_port_id,
1007                                        uint32_t delay_ms)
1008
1009 {
1010         struct bond_dev_private *internals;
1011
1012         if (valid_bonded_port_id(bonded_port_id) != 0)
1013                 return -1;
1014
1015         internals = rte_eth_devices[bonded_port_id].data->dev_private;
1016         internals->link_down_delay_ms = delay_ms;
1017
1018         return 0;
1019 }
1020
1021 int
1022 rte_eth_bond_link_down_prop_delay_get(uint16_t bonded_port_id)
1023 {
1024         struct bond_dev_private *internals;
1025
1026         if (valid_bonded_port_id(bonded_port_id) != 0)
1027                 return -1;
1028
1029         internals = rte_eth_devices[bonded_port_id].data->dev_private;
1030
1031         return internals->link_down_delay_ms;
1032 }
1033
1034 int
1035 rte_eth_bond_link_up_prop_delay_set(uint16_t bonded_port_id, uint32_t delay_ms)
1036
1037 {
1038         struct bond_dev_private *internals;
1039
1040         if (valid_bonded_port_id(bonded_port_id) != 0)
1041                 return -1;
1042
1043         internals = rte_eth_devices[bonded_port_id].data->dev_private;
1044         internals->link_up_delay_ms = delay_ms;
1045
1046         return 0;
1047 }
1048
1049 int
1050 rte_eth_bond_link_up_prop_delay_get(uint16_t bonded_port_id)
1051 {
1052         struct bond_dev_private *internals;
1053
1054         if (valid_bonded_port_id(bonded_port_id) != 0)
1055                 return -1;
1056
1057         internals = rte_eth_devices[bonded_port_id].data->dev_private;
1058
1059         return internals->link_up_delay_ms;
1060 }