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