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                         RTE_ETH_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         internals->rss_key_len = di->hash_key_size;
 294
 295         /* Inherit Rx offload capabilities from the first slave device */
 296         internals->rx_offload_capa = di->rx_offload_capa;
 297         internals->rx_queue_offload_capa = di->rx_queue_offload_capa;
 298         internals->flow_type_rss_offloads = di->flow_type_rss_offloads;
 299
 300         /* Inherit maximum Rx packet size from the first slave device */
 301         internals->candidate_max_rx_pktlen = di->max_rx_pktlen;
 302
 303         /* Inherit default Rx queue settings from the first slave device */
 304         memcpy(rxconf_i, &di->default_rxconf, sizeof(*rxconf_i));
 305
 306         /*
 307          * Turn off descriptor prefetch and writeback by default for all
 308          * slave devices. Applications may tweak this setting if need be.
 309          */
 310         rxconf_i->rx_thresh.pthresh = 0;
 311         rxconf_i->rx_thresh.hthresh = 0;
 312         rxconf_i->rx_thresh.wthresh = 0;
 313
 314         /* Setting this to zero should effectively enable default values */
 315         rxconf_i->rx_free_thresh = 0;
 316
 317         /* Disable deferred start by default for all slave devices */
 318         rxconf_i->rx_deferred_start = 0;
 319 }
 320
 321 static void
 322 eth_bond_slave_inherit_dev_info_tx_first(struct bond_dev_private *internals,
 323                                          const struct rte_eth_dev_info *di)
 324 {
 325         struct rte_eth_txconf *txconf_i = &internals->default_txconf;
 326
 327         /* Inherit Tx offload capabilities from the first slave device */
 328         internals->tx_offload_capa = di->tx_offload_capa;
 329         internals->tx_queue_offload_capa = di->tx_queue_offload_capa;
 330
 331         /* Inherit default Tx queue settings from the first slave device */
 332         memcpy(txconf_i, &di->default_txconf, sizeof(*txconf_i));
 333
 334         /*
 335          * Turn off descriptor prefetch and writeback by default for all
 336          * slave devices. Applications may tweak this setting if need be.
 337          */
 338         txconf_i->tx_thresh.pthresh = 0;
 339         txconf_i->tx_thresh.hthresh = 0;
 340         txconf_i->tx_thresh.wthresh = 0;
 341
 342         /*
 343          * Setting these parameters to zero assumes that default
 344          * values will be configured implicitly by slave devices.
 345          */
 346         txconf_i->tx_free_thresh = 0;
 347         txconf_i->tx_rs_thresh = 0;
 348
 349         /* Disable deferred start by default for all slave devices */
 350         txconf_i->tx_deferred_start = 0;
 351 }
 352
 353 static void
 354 eth_bond_slave_inherit_dev_info_rx_next(struct bond_dev_private *internals,
 355                                         const struct rte_eth_dev_info *di)
 356 {
 357         struct rte_eth_rxconf *rxconf_i = &internals->default_rxconf;
 358         const struct rte_eth_rxconf *rxconf = &di->default_rxconf;
 359
 360         internals->rx_offload_capa &= di->rx_offload_capa;
 361         internals->rx_queue_offload_capa &= di->rx_queue_offload_capa;
 362         internals->flow_type_rss_offloads &= di->flow_type_rss_offloads;
 363
 364         /*
 365          * If at least one slave device suggests enabling this
 366          * setting by default, enable it for all slave devices
 367          * since disabling it may not be necessarily supported.
 368          */
 369         if (rxconf->rx_drop_en == 1)
 370                 rxconf_i->rx_drop_en = 1;
 371
 372         /*
 373          * Adding a new slave device may cause some of previously inherited
 374          * offloads to be withdrawn from the internal rx_queue_offload_capa
 375          * value. Thus, the new internal value of default Rx queue offloads
 376          * has to be masked by rx_queue_offload_capa to make sure that only
 377          * commonly supported offloads are preserved from both the previous
 378          * value and the value being inhereted from the new slave device.
 379          */
 380         rxconf_i->offloads = (rxconf_i->offloads | rxconf->offloads) &
 381                              internals->rx_queue_offload_capa;
 382
 383         /*
 384          * RETA size is GCD of all slaves RETA sizes, so, if all sizes will be
 385          * the power of 2, the lower one is GCD
 386          */
 387         if (internals->reta_size > di->reta_size)
 388                 internals->reta_size = di->reta_size;
 389         if (internals->rss_key_len > di->hash_key_size) {
 390                 RTE_BOND_LOG(WARNING, "slave has different rss key size, "
 391                                 "configuring rss may fail");
 392                 internals->rss_key_len = di->hash_key_size;
 393         }
 394
 395         if (!internals->max_rx_pktlen &&
 396             di->max_rx_pktlen < internals->candidate_max_rx_pktlen)
 397                 internals->candidate_max_rx_pktlen = di->max_rx_pktlen;
 398 }
 399
 400 static void
 401 eth_bond_slave_inherit_dev_info_tx_next(struct bond_dev_private *internals,
 402                                         const struct rte_eth_dev_info *di)
 403 {
 404         struct rte_eth_txconf *txconf_i = &internals->default_txconf;
 405         const struct rte_eth_txconf *txconf = &di->default_txconf;
 406
 407         internals->tx_offload_capa &= di->tx_offload_capa;
 408         internals->tx_queue_offload_capa &= di->tx_queue_offload_capa;
 409
 410         /*
 411          * Adding a new slave device may cause some of previously inherited
 412          * offloads to be withdrawn from the internal tx_queue_offload_capa
 413          * value. Thus, the new internal value of default Tx queue offloads
 414          * has to be masked by tx_queue_offload_capa to make sure that only
 415          * commonly supported offloads are preserved from both the previous
 416          * value and the value being inhereted from the new slave device.
 417          */
 418         txconf_i->offloads = (txconf_i->offloads | txconf->offloads) &
 419                              internals->tx_queue_offload_capa;
 420 }
 421
 422 static void
 423 eth_bond_slave_inherit_desc_lim_first(struct rte_eth_desc_lim *bond_desc_lim,
 424                 const struct rte_eth_desc_lim *slave_desc_lim)
 425 {
 426         memcpy(bond_desc_lim, slave_desc_lim, sizeof(*bond_desc_lim));
 427 }
 428
 429 static int
 430 eth_bond_slave_inherit_desc_lim_next(struct rte_eth_desc_lim *bond_desc_lim,
 431                 const struct rte_eth_desc_lim *slave_desc_lim)
 432 {
 433         bond_desc_lim->nb_max = RTE_MIN(bond_desc_lim->nb_max,
 434                                         slave_desc_lim->nb_max);
 435         bond_desc_lim->nb_min = RTE_MAX(bond_desc_lim->nb_min,
 436                                         slave_desc_lim->nb_min);
 437         bond_desc_lim->nb_align = RTE_MAX(bond_desc_lim->nb_align,
 438                                           slave_desc_lim->nb_align);
 439
 440         if (bond_desc_lim->nb_min > bond_desc_lim->nb_max ||
 441             bond_desc_lim->nb_align > bond_desc_lim->nb_max) {
 442                 RTE_BOND_LOG(ERR, "Failed to inherit descriptor limits");
 443                 return -EINVAL;
 444         }
 445
 446         /* Treat maximum number of segments equal to 0 as unspecified */
 447         if (slave_desc_lim->nb_seg_max != 0 &&
 448             (bond_desc_lim->nb_seg_max == 0 ||
 449              slave_desc_lim->nb_seg_max < bond_desc_lim->nb_seg_max))
 450                 bond_desc_lim->nb_seg_max = slave_desc_lim->nb_seg_max;
 451         if (slave_desc_lim->nb_mtu_seg_max != 0 &&
 452             (bond_desc_lim->nb_mtu_seg_max == 0 ||
 453              slave_desc_lim->nb_mtu_seg_max < bond_desc_lim->nb_mtu_seg_max))
 454                 bond_desc_lim->nb_mtu_seg_max = slave_desc_lim->nb_mtu_seg_max;
 455
 456         return 0;
 457 }
 458
 459 static int
 460 __eth_bond_slave_add_lock_free(uint16_t bonded_port_id, uint16_t slave_port_id)
 461 {
 462         struct rte_eth_dev *bonded_eth_dev, *slave_eth_dev;
 463         struct bond_dev_private *internals;
 464         struct rte_eth_link link_props;
 465         struct rte_eth_dev_info dev_info;
 466         int ret;
 467
 468         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 469         internals = bonded_eth_dev->data->dev_private;
 470
 471         if (valid_slave_port_id(internals, slave_port_id) != 0)
 472                 return -1;
 473
 474         slave_eth_dev = &rte_eth_devices[slave_port_id];
 475         if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_BONDED_SLAVE) {
 476                 RTE_BOND_LOG(ERR, "Slave device is already a slave of a bonded device");
 477                 return -1;
 478         }
 479
 480         ret = rte_eth_dev_info_get(slave_port_id, &dev_info);
 481         if (ret != 0) {
 482                 RTE_BOND_LOG(ERR,
 483                         "%s: Error during getting device (port %u) info: %s\n",
 484                         __func__, slave_port_id, strerror(-ret));
 485
 486                 return ret;
 487         }
 488         if (dev_info.max_rx_pktlen < internals->max_rx_pktlen) {
 489                 RTE_BOND_LOG(ERR, "Slave (port %u) max_rx_pktlen too small",
 490                              slave_port_id);
 491                 return -1;
 492         }
 493
 494         slave_add(internals, slave_eth_dev);
 495
 496         /* We need to store slaves reta_size to be able to synchronize RETA for all
 497          * slave devices even if its sizes are different.
 498          */
 499         internals->slaves[internals->slave_count].reta_size = dev_info.reta_size;
 500
 501         if (internals->slave_count < 1) {
 502                 /* if MAC is not user defined then use MAC of first slave add to
 503                  * bonded device */
 504                 if (!internals->user_defined_mac) {
 505                         if (mac_address_set(bonded_eth_dev,
 506                                             slave_eth_dev->data->mac_addrs)) {
 507                                 RTE_BOND_LOG(ERR, "Failed to set MAC address");
 508                                 return -1;
 509                         }
 510                 }
 511
 512                 /* Make primary slave */
 513                 internals->primary_port = slave_port_id;
 514                 internals->current_primary_port = slave_port_id;
 515
 516                 /* Inherit queues settings from first slave */
 517                 internals->nb_rx_queues = slave_eth_dev->data->nb_rx_queues;
 518                 internals->nb_tx_queues = slave_eth_dev->data->nb_tx_queues;
 519
 520                 eth_bond_slave_inherit_dev_info_rx_first(internals, &dev_info);
 521                 eth_bond_slave_inherit_dev_info_tx_first(internals, &dev_info);
 522
 523                 eth_bond_slave_inherit_desc_lim_first(&internals->rx_desc_lim,
 524                                                       &dev_info.rx_desc_lim);
 525                 eth_bond_slave_inherit_desc_lim_first(&internals->tx_desc_lim,
 526                                                       &dev_info.tx_desc_lim);
 527         } else {
 528                 int ret;
 529
 530                 eth_bond_slave_inherit_dev_info_rx_next(internals, &dev_info);
 531                 eth_bond_slave_inherit_dev_info_tx_next(internals, &dev_info);
 532
 533                 ret = eth_bond_slave_inherit_desc_lim_next(
 534                                 &internals->rx_desc_lim, &dev_info.rx_desc_lim);
 535                 if (ret != 0)
 536                         return ret;
 537
 538                 ret = eth_bond_slave_inherit_desc_lim_next(
 539                                 &internals->tx_desc_lim, &dev_info.tx_desc_lim);
 540                 if (ret != 0)
 541                         return ret;
 542         }
 543
 544         bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf &=
 545                         internals->flow_type_rss_offloads;
 546
 547         if (slave_rte_flow_prepare(internals->slave_count, internals) != 0) {
 548                 RTE_BOND_LOG(ERR, "Failed to prepare new slave flows: port=%d",
 549                              slave_port_id);
 550                 return -1;
 551         }
 552
 553         /* Add additional MAC addresses to the slave */
 554         if (slave_add_mac_addresses(bonded_eth_dev, slave_port_id) != 0) {
 555                 RTE_BOND_LOG(ERR, "Failed to add mac address(es) to slave %hu",
 556                                 slave_port_id);
 557                 return -1;
 558         }
 559
 560         internals->slave_count++;
 561
 562         if (bonded_eth_dev->data->dev_started) {
 563                 if (slave_configure(bonded_eth_dev, slave_eth_dev) != 0) {
 564                         internals->slave_count--;
 565                         RTE_BOND_LOG(ERR, "rte_bond_slaves_configure: port=%d",
 566                                         slave_port_id);
 567                         return -1;
 568                 }
 569         }
 570
 571         /* Update all slave devices MACs */
 572         mac_address_slaves_update(bonded_eth_dev);
 573
 574         /* Register link status change callback with bonded device pointer as
 575          * argument*/
 576         rte_eth_dev_callback_register(slave_port_id, RTE_ETH_EVENT_INTR_LSC,
 577                         bond_ethdev_lsc_event_callback, &bonded_eth_dev->data->port_id);
 578
 579         /* If bonded device is started then we can add the slave to our active
 580          * slave array */
 581         if (bonded_eth_dev->data->dev_started) {
 582                 ret = rte_eth_link_get_nowait(slave_port_id, &link_props);
 583                 if (ret < 0) {
 584                         rte_eth_dev_callback_unregister(slave_port_id,
 585                                         RTE_ETH_EVENT_INTR_LSC,
 586                                         bond_ethdev_lsc_event_callback,
 587                                         &bonded_eth_dev->data->port_id);
 588                         internals->slave_count--;
 589                         RTE_BOND_LOG(ERR,
 590                                 "Slave (port %u) link get failed: %s\n",
 591                                 slave_port_id, rte_strerror(-ret));
 592                         return -1;
 593                 }
 594
 595                 if (link_props.link_status == RTE_ETH_LINK_UP) {
 596                         if (internals->active_slave_count == 0 &&
 597                             !internals->user_defined_primary_port)
 598                                 bond_ethdev_primary_set(internals,
 599                                                         slave_port_id);
 600                 }
 601         }
 602
 603         /* Add slave details to bonded device */
 604         slave_eth_dev->data->dev_flags |= RTE_ETH_DEV_BONDED_SLAVE;
 605
 606         slave_vlan_filter_set(bonded_port_id, slave_port_id);
 607
 608         return 0;
 609
 610 }
 611
 612 int
 613 rte_eth_bond_slave_add(uint16_t bonded_port_id, uint16_t slave_port_id)
 614 {
 615         struct rte_eth_dev *bonded_eth_dev;
 616         struct bond_dev_private *internals;
 617
 618         int retval;
 619
 620         if (valid_bonded_port_id(bonded_port_id) != 0)
 621                 return -1;
 622
 623         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 624         internals = bonded_eth_dev->data->dev_private;
 625
 626         if (valid_slave_port_id(internals, slave_port_id) != 0)
 627                 return -1;
 628
 629         rte_spinlock_lock(&internals->lock);
 630
 631         retval = __eth_bond_slave_add_lock_free(bonded_port_id, slave_port_id);
 632
 633         rte_spinlock_unlock(&internals->lock);
 634
 635         return retval;
 636 }
 637
 638 static int
 639 __eth_bond_slave_remove_lock_free(uint16_t bonded_port_id,
 640                                    uint16_t slave_port_id)
 641 {
 642         struct rte_eth_dev *bonded_eth_dev;
 643         struct bond_dev_private *internals;
 644         struct rte_eth_dev *slave_eth_dev;
 645         struct rte_flow_error flow_error;
 646         struct rte_flow *flow;
 647         int i, slave_idx;
 648
 649         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 650         internals = bonded_eth_dev->data->dev_private;
 651
 652         if (valid_slave_port_id(internals, slave_port_id) < 0)
 653                 return -1;
 654
 655         /* first remove from active slave list */
 656         slave_idx = find_slave_by_id(internals->active_slaves,
 657                 internals->active_slave_count, slave_port_id);
 658
 659         if (slave_idx < internals->active_slave_count)
 660                 deactivate_slave(bonded_eth_dev, slave_port_id);
 661
 662         slave_idx = -1;
 663         /* now find in slave list */
 664         for (i = 0; i < internals->slave_count; i++)
 665                 if (internals->slaves[i].port_id == slave_port_id) {
 666                         slave_idx = i;
 667                         break;
 668                 }
 669
 670         if (slave_idx < 0) {
 671                 RTE_BOND_LOG(ERR, "Couldn't find slave in port list, slave count %d",
 672                                 internals->slave_count);
 673                 return -1;
 674         }
 675
 676         /* Un-register link status change callback with bonded device pointer as
 677          * argument*/
 678         rte_eth_dev_callback_unregister(slave_port_id, RTE_ETH_EVENT_INTR_LSC,
 679                         bond_ethdev_lsc_event_callback,
 680                         &rte_eth_devices[bonded_port_id].data->port_id);
 681
 682         /* Restore original MAC address of slave device */
 683         rte_eth_dev_default_mac_addr_set(slave_port_id,
 684                         &(internals->slaves[slave_idx].persisted_mac_addr));
 685
 686         /* remove additional MAC addresses from the slave */
 687         slave_remove_mac_addresses(bonded_eth_dev, slave_port_id);
 688
 689         /*
 690          * Remove bond device flows from slave device.
 691          * Note: don't restore flow isolate mode.
 692          */
 693         TAILQ_FOREACH(flow, &internals->flow_list, next) {
 694                 if (flow->flows[slave_idx] != NULL) {
 695                         rte_flow_destroy(slave_port_id, flow->flows[slave_idx],
 696                                          &flow_error);
 697                         flow->flows[slave_idx] = NULL;
 698                 }
 699         }
 700
 701         slave_eth_dev = &rte_eth_devices[slave_port_id];
 702         slave_remove(internals, slave_eth_dev);
 703         slave_eth_dev->data->dev_flags &= (~RTE_ETH_DEV_BONDED_SLAVE);
 704
 705         /*  first slave in the active list will be the primary by default,
 706          *  otherwise use first device in list */
 707         if (internals->current_primary_port == slave_port_id) {
 708                 if (internals->active_slave_count > 0)
 709                         internals->current_primary_port = internals->active_slaves[0];
 710                 else if (internals->slave_count > 0)
 711                         internals->current_primary_port = internals->slaves[0].port_id;
 712                 else
 713                         internals->primary_port = 0;
 714                 mac_address_slaves_update(bonded_eth_dev);
 715         }
 716
 717         if (internals->active_slave_count < 1) {
 718                 /* if no slaves are any longer attached to bonded device and MAC is not
 719                  * user defined then clear MAC of bonded device as it will be reset
 720                  * when a new slave is added */
 721                 if (internals->slave_count < 1 && !internals->user_defined_mac)
 722                         memset(rte_eth_devices[bonded_port_id].data->mac_addrs, 0,
 723                                         sizeof(*(rte_eth_devices[bonded_port_id].data->mac_addrs)));
 724         }
 725         if (internals->slave_count == 0) {
 726                 internals->rx_offload_capa = 0;
 727                 internals->tx_offload_capa = 0;
 728                 internals->rx_queue_offload_capa = 0;
 729                 internals->tx_queue_offload_capa = 0;
 730                 internals->flow_type_rss_offloads = RTE_ETH_RSS_PROTO_MASK;
 731                 internals->reta_size = 0;
 732                 internals->candidate_max_rx_pktlen = 0;
 733                 internals->max_rx_pktlen = 0;
 734         }
 735         return 0;
 736 }
 737
 738 int
 739 rte_eth_bond_slave_remove(uint16_t bonded_port_id, uint16_t slave_port_id)
 740 {
 741         struct rte_eth_dev *bonded_eth_dev;
 742         struct bond_dev_private *internals;
 743         int retval;
 744
 745         if (valid_bonded_port_id(bonded_port_id) != 0)
 746                 return -1;
 747
 748         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 749         internals = bonded_eth_dev->data->dev_private;
 750
 751         rte_spinlock_lock(&internals->lock);
 752
 753         retval = __eth_bond_slave_remove_lock_free(bonded_port_id, slave_port_id);
 754
 755         rte_spinlock_unlock(&internals->lock);
 756
 757         return retval;
 758 }
 759
 760 int
 761 rte_eth_bond_mode_set(uint16_t bonded_port_id, uint8_t mode)
 762 {
 763         struct rte_eth_dev *bonded_eth_dev;
 764
 765         if (valid_bonded_port_id(bonded_port_id) != 0)
 766                 return -1;
 767
 768         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 769
 770         if (check_for_master_bonded_ethdev(bonded_eth_dev) != 0 &&
 771                         mode == BONDING_MODE_8023AD)
 772                 return -1;
 773
 774         return bond_ethdev_mode_set(bonded_eth_dev, mode);
 775 }
 776
 777 int
 778 rte_eth_bond_mode_get(uint16_t bonded_port_id)
 779 {
 780         struct bond_dev_private *internals;
 781
 782         if (valid_bonded_port_id(bonded_port_id) != 0)
 783                 return -1;
 784
 785         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 786
 787         return internals->mode;
 788 }
 789
 790 int
 791 rte_eth_bond_primary_set(uint16_t bonded_port_id, uint16_t slave_port_id)
 792 {
 793         struct bond_dev_private *internals;
 794
 795         if (valid_bonded_port_id(bonded_port_id) != 0)
 796                 return -1;
 797
 798         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 799
 800         if (valid_slave_port_id(internals, slave_port_id) != 0)
 801                 return -1;
 802
 803         internals->user_defined_primary_port = 1;
 804         internals->primary_port = slave_port_id;
 805
 806         bond_ethdev_primary_set(internals, slave_port_id);
 807
 808         return 0;
 809 }
 810
 811 int
 812 rte_eth_bond_primary_get(uint16_t bonded_port_id)
 813 {
 814         struct bond_dev_private *internals;
 815
 816         if (valid_bonded_port_id(bonded_port_id) != 0)
 817                 return -1;
 818
 819         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 820
 821         if (internals->slave_count < 1)
 822                 return -1;
 823
 824         return internals->current_primary_port;
 825 }
 826
 827 int
 828 rte_eth_bond_slaves_get(uint16_t bonded_port_id, uint16_t slaves[],
 829                         uint16_t len)
 830 {
 831         struct bond_dev_private *internals;
 832         uint16_t i;
 833
 834         if (valid_bonded_port_id(bonded_port_id) != 0)
 835                 return -1;
 836
 837         if (slaves == NULL)
 838                 return -1;
 839
 840         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 841
 842         if (internals->slave_count > len)
 843                 return -1;
 844
 845         for (i = 0; i < internals->slave_count; i++)
 846                 slaves[i] = internals->slaves[i].port_id;
 847
 848         return internals->slave_count;
 849 }
 850
 851 int
 852 rte_eth_bond_active_slaves_get(uint16_t bonded_port_id, uint16_t slaves[],
 853                 uint16_t len)
 854 {
 855         struct bond_dev_private *internals;
 856
 857         if (valid_bonded_port_id(bonded_port_id) != 0)
 858                 return -1;
 859
 860         if (slaves == NULL)
 861                 return -1;
 862
 863         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 864
 865         if (internals->active_slave_count > len)
 866                 return -1;
 867
 868         memcpy(slaves, internals->active_slaves,
 869         internals->active_slave_count * sizeof(internals->active_slaves[0]));
 870
 871         return internals->active_slave_count;
 872 }
 873
 874 int
 875 rte_eth_bond_mac_address_set(uint16_t bonded_port_id,
 876                 struct rte_ether_addr *mac_addr)
 877 {
 878         struct rte_eth_dev *bonded_eth_dev;
 879         struct bond_dev_private *internals;
 880
 881         if (valid_bonded_port_id(bonded_port_id) != 0)
 882                 return -1;
 883
 884         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 885         internals = bonded_eth_dev->data->dev_private;
 886
 887         /* Set MAC Address of Bonded Device */
 888         if (mac_address_set(bonded_eth_dev, mac_addr))
 889                 return -1;
 890
 891         internals->user_defined_mac = 1;
 892
 893         /* Update all slave devices MACs*/
 894         if (internals->slave_count > 0)
 895                 return mac_address_slaves_update(bonded_eth_dev);
 896
 897         return 0;
 898 }
 899
 900 int
 901 rte_eth_bond_mac_address_reset(uint16_t bonded_port_id)
 902 {
 903         struct rte_eth_dev *bonded_eth_dev;
 904         struct bond_dev_private *internals;
 905
 906         if (valid_bonded_port_id(bonded_port_id) != 0)
 907                 return -1;
 908
 909         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 910         internals = bonded_eth_dev->data->dev_private;
 911
 912         internals->user_defined_mac = 0;
 913
 914         if (internals->slave_count > 0) {
 915                 int slave_port;
 916                 /* Get the primary slave location based on the primary port
 917                  * number as, while slave_add(), we will keep the primary
 918                  * slave based on slave_count,but not based on the primary port.
 919                  */
 920                 for (slave_port = 0; slave_port < internals->slave_count;
 921                      slave_port++) {
 922                         if (internals->slaves[slave_port].port_id ==
 923                             internals->primary_port)
 924                                 break;
 925                 }
 926
 927                 /* Set MAC Address of Bonded Device */
 928                 if (mac_address_set(bonded_eth_dev,
 929                         &internals->slaves[slave_port].persisted_mac_addr)
 930                                 != 0) {
 931                         RTE_BOND_LOG(ERR, "Failed to set MAC address on bonded device");
 932                         return -1;
 933                 }
 934                 /* Update all slave devices MAC addresses */
 935                 return mac_address_slaves_update(bonded_eth_dev);
 936         }
 937         /* No need to update anything as no slaves present */
 938         return 0;
 939 }
 940
 941 int
 942 rte_eth_bond_xmit_policy_set(uint16_t bonded_port_id, uint8_t policy)
 943 {
 944         struct bond_dev_private *internals;
 945
 946         if (valid_bonded_port_id(bonded_port_id) != 0)
 947                 return -1;
 948
 949         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 950
 951         switch (policy) {
 952         case BALANCE_XMIT_POLICY_LAYER2:
 953                 internals->balance_xmit_policy = policy;
 954                 internals->burst_xmit_hash = burst_xmit_l2_hash;
 955                 break;
 956         case BALANCE_XMIT_POLICY_LAYER23:
 957                 internals->balance_xmit_policy = policy;
 958                 internals->burst_xmit_hash = burst_xmit_l23_hash;
 959                 break;
 960         case BALANCE_XMIT_POLICY_LAYER34:
 961                 internals->balance_xmit_policy = policy;
 962                 internals->burst_xmit_hash = burst_xmit_l34_hash;
 963                 break;
 964
 965         default:
 966                 return -1;
 967         }
 968         return 0;
 969 }
 970
 971 int
 972 rte_eth_bond_xmit_policy_get(uint16_t bonded_port_id)
 973 {
 974         struct bond_dev_private *internals;
 975
 976         if (valid_bonded_port_id(bonded_port_id) != 0)
 977                 return -1;
 978
 979         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 980
 981         return internals->balance_xmit_policy;
 982 }
 983
 984 int
 985 rte_eth_bond_link_monitoring_set(uint16_t bonded_port_id, uint32_t internal_ms)
 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         internals->link_status_polling_interval_ms = internal_ms;
 994
 995         return 0;
 996 }
 997
 998 int
 999 rte_eth_bond_link_monitoring_get(uint16_t bonded_port_id)
1000 {
1001         struct bond_dev_private *internals;
1002
1003         if (valid_bonded_port_id(bonded_port_id) != 0)
1004                 return -1;
1005
1006         internals = rte_eth_devices[bonded_port_id].data->dev_private;
1007
1008         return internals->link_status_polling_interval_ms;
1009 }
1010
1011 int
1012 rte_eth_bond_link_down_prop_delay_set(uint16_t bonded_port_id,
1013                                        uint32_t delay_ms)
1014
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         internals->link_down_delay_ms = delay_ms;
1023
1024         return 0;
1025 }
1026
1027 int
1028 rte_eth_bond_link_down_prop_delay_get(uint16_t bonded_port_id)
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
1037         return internals->link_down_delay_ms;
1038 }
1039
1040 int
1041 rte_eth_bond_link_up_prop_delay_set(uint16_t bonded_port_id, uint32_t delay_ms)
1042
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         internals->link_up_delay_ms = delay_ms;
1051
1052         return 0;
1053 }
1054
1055 int
1056 rte_eth_bond_link_up_prop_delay_get(uint16_t bonded_port_id)
1057 {
1058         struct bond_dev_private *internals;
1059
1060         if (valid_bonded_port_id(bonded_port_id) != 0)
1061                 return -1;
1062
1063         internals = rte_eth_devices[bonded_port_id].data->dev_private;
1064
1065         return internals->link_up_delay_ms;
1066 }