drivers/net/bonding/rte_eth_bond_pmd.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2017 Intel Corporation
   3  */
   4 #include <stdlib.h>
   5 #include <stdbool.h>
   6 #include <netinet/in.h>
   7
   8 #include <rte_mbuf.h>
   9 #include <rte_malloc.h>
  10 #include <ethdev_driver.h>
  11 #include <ethdev_vdev.h>
  12 #include <rte_tcp.h>
  13 #include <rte_udp.h>
  14 #include <rte_ip.h>
  15 #include <rte_ip_frag.h>
  16 #include <rte_devargs.h>
  17 #include <rte_kvargs.h>
  18 #include <rte_bus_vdev.h>
  19 #include <rte_alarm.h>
  20 #include <rte_cycles.h>
  21 #include <rte_string_fns.h>
  22
  23 #include "rte_eth_bond.h"
  24 #include "eth_bond_private.h"
  25 #include "eth_bond_8023ad_private.h"
  26
  27 #define REORDER_PERIOD_MS 10
  28 #define DEFAULT_POLLING_INTERVAL_10_MS (10)
  29 #define BOND_MAX_MAC_ADDRS 16
  30
  31 #define HASH_L4_PORTS(h) ((h)->src_port ^ (h)->dst_port)
  32
  33 /* Table for statistics in mode 5 TLB */
  34 static uint64_t tlb_last_obytets[RTE_MAX_ETHPORTS];
  35
  36 static inline size_t
  37 get_vlan_offset(struct rte_ether_hdr *eth_hdr, uint16_t *proto)
  38 {
  39         size_t vlan_offset = 0;
  40
  41         if (rte_cpu_to_be_16(RTE_ETHER_TYPE_VLAN) == *proto ||
  42                 rte_cpu_to_be_16(RTE_ETHER_TYPE_QINQ) == *proto) {
  43                 struct rte_vlan_hdr *vlan_hdr =
  44                         (struct rte_vlan_hdr *)(eth_hdr + 1);
  45
  46                 vlan_offset = sizeof(struct rte_vlan_hdr);
  47                 *proto = vlan_hdr->eth_proto;
  48
  49                 if (rte_cpu_to_be_16(RTE_ETHER_TYPE_VLAN) == *proto) {
  50                         vlan_hdr = vlan_hdr + 1;
  51                         *proto = vlan_hdr->eth_proto;
  52                         vlan_offset += sizeof(struct rte_vlan_hdr);
  53                 }
  54         }
  55         return vlan_offset;
  56 }
  57
  58 static uint16_t
  59 bond_ethdev_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
  60 {
  61         struct bond_dev_private *internals;
  62
  63         uint16_t num_rx_total = 0;
  64         uint16_t slave_count;
  65         uint16_t active_slave;
  66         int i;
  67
  68         /* Cast to structure, containing bonded device's port id and queue id */
  69         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
  70         internals = bd_rx_q->dev_private;
  71         slave_count = internals->active_slave_count;
  72         active_slave = bd_rx_q->active_slave;
  73
  74         for (i = 0; i < slave_count && nb_pkts; i++) {
  75                 uint16_t num_rx_slave;
  76
  77                 /* Offset of pointer to *bufs increases as packets are received
  78                  * from other slaves */
  79                 num_rx_slave =
  80                         rte_eth_rx_burst(internals->active_slaves[active_slave],
  81                                          bd_rx_q->queue_id,
  82                                          bufs + num_rx_total, nb_pkts);
  83                 num_rx_total += num_rx_slave;
  84                 nb_pkts -= num_rx_slave;
  85                 if (++active_slave == slave_count)
  86                         active_slave = 0;
  87         }
  88
  89         if (++bd_rx_q->active_slave >= slave_count)
  90                 bd_rx_q->active_slave = 0;
  91         return num_rx_total;
  92 }
  93
  94 static uint16_t
  95 bond_ethdev_rx_burst_active_backup(void *queue, struct rte_mbuf **bufs,
  96                 uint16_t nb_pkts)
  97 {
  98         struct bond_dev_private *internals;
  99
 100         /* Cast to structure, containing bonded device's port id and queue id */
 101         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
 102
 103         internals = bd_rx_q->dev_private;
 104
 105         return rte_eth_rx_burst(internals->current_primary_port,
 106                         bd_rx_q->queue_id, bufs, nb_pkts);
 107 }
 108
 109 static inline uint8_t
 110 is_lacp_packets(uint16_t ethertype, uint8_t subtype, struct rte_mbuf *mbuf)
 111 {
 112         const uint16_t ether_type_slow_be =
 113                 rte_be_to_cpu_16(RTE_ETHER_TYPE_SLOW);
 114
 115         return !((mbuf->ol_flags & RTE_MBUF_F_RX_VLAN) ? mbuf->vlan_tci : 0) &&
 116                 (ethertype == ether_type_slow_be &&
 117                 (subtype == SLOW_SUBTYPE_MARKER || subtype == SLOW_SUBTYPE_LACP));
 118 }
 119
 120 /*****************************************************************************
 121  * Flow director's setup for mode 4 optimization
 122  */
 123
 124 static struct rte_flow_item_eth flow_item_eth_type_8023ad = {
 125         .dst.addr_bytes = { 0 },
 126         .src.addr_bytes = { 0 },
 127         .type = RTE_BE16(RTE_ETHER_TYPE_SLOW),
 128 };
 129
 130 static struct rte_flow_item_eth flow_item_eth_mask_type_8023ad = {
 131         .dst.addr_bytes = { 0 },
 132         .src.addr_bytes = { 0 },
 133         .type = 0xFFFF,
 134 };
 135
 136 static struct rte_flow_item flow_item_8023ad[] = {
 137         {
 138                 .type = RTE_FLOW_ITEM_TYPE_ETH,
 139                 .spec = &flow_item_eth_type_8023ad,
 140                 .last = NULL,
 141                 .mask = &flow_item_eth_mask_type_8023ad,
 142         },
 143         {
 144                 .type = RTE_FLOW_ITEM_TYPE_END,
 145                 .spec = NULL,
 146                 .last = NULL,
 147                 .mask = NULL,
 148         }
 149 };
 150
 151 const struct rte_flow_attr flow_attr_8023ad = {
 152         .group = 0,
 153         .priority = 0,
 154         .ingress = 1,
 155         .egress = 0,
 156         .reserved = 0,
 157 };
 158
 159 int
 160 bond_ethdev_8023ad_flow_verify(struct rte_eth_dev *bond_dev,
 161                 uint16_t slave_port) {
 162         struct rte_eth_dev_info slave_info;
 163         struct rte_flow_error error;
 164         struct bond_dev_private *internals = bond_dev->data->dev_private;
 165
 166         const struct rte_flow_action_queue lacp_queue_conf = {
 167                 .index = 0,
 168         };
 169
 170         const struct rte_flow_action actions[] = {
 171                 {
 172                         .type = RTE_FLOW_ACTION_TYPE_QUEUE,
 173                         .conf = &lacp_queue_conf
 174                 },
 175                 {
 176                         .type = RTE_FLOW_ACTION_TYPE_END,
 177                 }
 178         };
 179
 180         int ret = rte_flow_validate(slave_port, &flow_attr_8023ad,
 181                         flow_item_8023ad, actions, &error);
 182         if (ret < 0) {
 183                 RTE_BOND_LOG(ERR, "%s: %s (slave_port=%d queue_id=%d)",
 184                                 __func__, error.message, slave_port,
 185                                 internals->mode4.dedicated_queues.rx_qid);
 186                 return -1;
 187         }
 188
 189         ret = rte_eth_dev_info_get(slave_port, &slave_info);
 190         if (ret != 0) {
 191                 RTE_BOND_LOG(ERR,
 192                         "%s: Error during getting device (port %u) info: %s\n",
 193                         __func__, slave_port, strerror(-ret));
 194
 195                 return ret;
 196         }
 197
 198         if (slave_info.max_rx_queues < bond_dev->data->nb_rx_queues ||
 199                         slave_info.max_tx_queues < bond_dev->data->nb_tx_queues) {
 200                 RTE_BOND_LOG(ERR,
 201                         "%s: Slave %d capabilities doesn't allow to allocate additional queues",
 202                         __func__, slave_port);
 203                 return -1;
 204         }
 205
 206         return 0;
 207 }
 208
 209 int
 210 bond_8023ad_slow_pkt_hw_filter_supported(uint16_t port_id) {
 211         struct rte_eth_dev *bond_dev = &rte_eth_devices[port_id];
 212         struct bond_dev_private *internals = bond_dev->data->dev_private;
 213         struct rte_eth_dev_info bond_info;
 214         uint16_t idx;
 215         int ret;
 216
 217         /* Verify if all slaves in bonding supports flow director and */
 218         if (internals->slave_count > 0) {
 219                 ret = rte_eth_dev_info_get(bond_dev->data->port_id, &bond_info);
 220                 if (ret != 0) {
 221                         RTE_BOND_LOG(ERR,
 222                                 "%s: Error during getting device (port %u) info: %s\n",
 223                                 __func__, bond_dev->data->port_id,
 224                                 strerror(-ret));
 225
 226                         return ret;
 227                 }
 228
 229                 internals->mode4.dedicated_queues.rx_qid = bond_info.nb_rx_queues;
 230                 internals->mode4.dedicated_queues.tx_qid = bond_info.nb_tx_queues;
 231
 232                 for (idx = 0; idx < internals->slave_count; idx++) {
 233                         if (bond_ethdev_8023ad_flow_verify(bond_dev,
 234                                         internals->slaves[idx].port_id) != 0)
 235                                 return -1;
 236                 }
 237         }
 238
 239         return 0;
 240 }
 241
 242 int
 243 bond_ethdev_8023ad_flow_set(struct rte_eth_dev *bond_dev, uint16_t slave_port) {
 244
 245         struct rte_flow_error error;
 246         struct bond_dev_private *internals = bond_dev->data->dev_private;
 247         struct rte_flow_action_queue lacp_queue_conf = {
 248                 .index = internals->mode4.dedicated_queues.rx_qid,
 249         };
 250
 251         const struct rte_flow_action actions[] = {
 252                 {
 253                         .type = RTE_FLOW_ACTION_TYPE_QUEUE,
 254                         .conf = &lacp_queue_conf
 255                 },
 256                 {
 257                         .type = RTE_FLOW_ACTION_TYPE_END,
 258                 }
 259         };
 260
 261         internals->mode4.dedicated_queues.flow[slave_port] = rte_flow_create(slave_port,
 262                         &flow_attr_8023ad, flow_item_8023ad, actions, &error);
 263         if (internals->mode4.dedicated_queues.flow[slave_port] == NULL) {
 264                 RTE_BOND_LOG(ERR, "bond_ethdev_8023ad_flow_set: %s "
 265                                 "(slave_port=%d queue_id=%d)",
 266                                 error.message, slave_port,
 267                                 internals->mode4.dedicated_queues.rx_qid);
 268                 return -1;
 269         }
 270
 271         return 0;
 272 }
 273
 274 static inline uint16_t
 275 rx_burst_8023ad(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts,
 276                 bool dedicated_rxq)
 277 {
 278         /* Cast to structure, containing bonded device's port id and queue id */
 279         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
 280         struct bond_dev_private *internals = bd_rx_q->dev_private;
 281         struct rte_eth_dev *bonded_eth_dev =
 282                                         &rte_eth_devices[internals->port_id];
 283         struct rte_ether_addr *bond_mac = bonded_eth_dev->data->mac_addrs;
 284         struct rte_ether_hdr *hdr;
 285
 286         const uint16_t ether_type_slow_be =
 287                 rte_be_to_cpu_16(RTE_ETHER_TYPE_SLOW);
 288         uint16_t num_rx_total = 0;      /* Total number of received packets */
 289         uint16_t slaves[RTE_MAX_ETHPORTS];
 290         uint16_t slave_count, idx;
 291
 292         uint8_t collecting;  /* current slave collecting status */
 293         const uint8_t promisc = rte_eth_promiscuous_get(internals->port_id);
 294         const uint8_t allmulti = rte_eth_allmulticast_get(internals->port_id);
 295         uint8_t subtype;
 296         uint16_t i;
 297         uint16_t j;
 298         uint16_t k;
 299
 300         /* Copy slave list to protect against slave up/down changes during tx
 301          * bursting */
 302         slave_count = internals->active_slave_count;
 303         memcpy(slaves, internals->active_slaves,
 304                         sizeof(internals->active_slaves[0]) * slave_count);
 305
 306         idx = bd_rx_q->active_slave;
 307         if (idx >= slave_count) {
 308                 bd_rx_q->active_slave = 0;
 309                 idx = 0;
 310         }
 311         for (i = 0; i < slave_count && num_rx_total < nb_pkts; i++) {
 312                 j = num_rx_total;
 313                 collecting = ACTOR_STATE(&bond_mode_8023ad_ports[slaves[idx]],
 314                                          COLLECTING);
 315
 316                 /* Read packets from this slave */
 317                 num_rx_total += rte_eth_rx_burst(slaves[idx], bd_rx_q->queue_id,
 318                                 &bufs[num_rx_total], nb_pkts - num_rx_total);
 319
 320                 for (k = j; k < 2 && k < num_rx_total; k++)
 321                         rte_prefetch0(rte_pktmbuf_mtod(bufs[k], void *));
 322
 323                 /* Handle slow protocol packets. */
 324                 while (j < num_rx_total) {
 325                         if (j + 3 < num_rx_total)
 326                                 rte_prefetch0(rte_pktmbuf_mtod(bufs[j + 3], void *));
 327
 328                         hdr = rte_pktmbuf_mtod(bufs[j], struct rte_ether_hdr *);
 329                         subtype = ((struct slow_protocol_frame *)hdr)->slow_protocol.subtype;
 330
 331                         /* Remove packet from array if:
 332                          * - it is slow packet but no dedicated rxq is present,
 333                          * - slave is not in collecting state,
 334                          * - bonding interface is not in promiscuous mode:
 335                          *   - packet is unicast and address does not match,
 336                          *   - packet is multicast and bonding interface
 337                          *     is not in allmulti,
 338                          */
 339                         if (unlikely(
 340                                 (!dedicated_rxq &&
 341                                  is_lacp_packets(hdr->ether_type, subtype,
 342                                                  bufs[j])) ||
 343                                 !collecting ||
 344                                 (!promisc &&
 345                                  ((rte_is_unicast_ether_addr(&hdr->dst_addr) &&
 346                                    !rte_is_same_ether_addr(bond_mac,
 347                                                        &hdr->dst_addr)) ||
 348                                   (!allmulti &&
 349                                    rte_is_multicast_ether_addr(&hdr->dst_addr)))))) {
 350
 351                                 if (hdr->ether_type == ether_type_slow_be) {
 352                                         bond_mode_8023ad_handle_slow_pkt(
 353                                             internals, slaves[idx], bufs[j]);
 354                                 } else
 355                                         rte_pktmbuf_free(bufs[j]);
 356
 357                                 /* Packet is managed by mode 4 or dropped, shift the array */
 358                                 num_rx_total--;
 359                                 if (j < num_rx_total) {
 360                                         memmove(&bufs[j], &bufs[j + 1], sizeof(bufs[0]) *
 361                                                 (num_rx_total - j));
 362                                 }
 363                         } else
 364                                 j++;
 365                 }
 366                 if (unlikely(++idx == slave_count))
 367                         idx = 0;
 368         }
 369
 370         if (++bd_rx_q->active_slave >= slave_count)
 371                 bd_rx_q->active_slave = 0;
 372
 373         return num_rx_total;
 374 }
 375
 376 static uint16_t
 377 bond_ethdev_rx_burst_8023ad(void *queue, struct rte_mbuf **bufs,
 378                 uint16_t nb_pkts)
 379 {
 380         return rx_burst_8023ad(queue, bufs, nb_pkts, false);
 381 }
 382
 383 static uint16_t
 384 bond_ethdev_rx_burst_8023ad_fast_queue(void *queue, struct rte_mbuf **bufs,
 385                 uint16_t nb_pkts)
 386 {
 387         return rx_burst_8023ad(queue, bufs, nb_pkts, true);
 388 }
 389
 390 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 391 uint32_t burstnumberRX;
 392 uint32_t burstnumberTX;
 393
 394 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 395
 396 static void
 397 arp_op_name(uint16_t arp_op, char *buf, size_t buf_len)
 398 {
 399         switch (arp_op) {
 400         case RTE_ARP_OP_REQUEST:
 401                 strlcpy(buf, "ARP Request", buf_len);
 402                 return;
 403         case RTE_ARP_OP_REPLY:
 404                 strlcpy(buf, "ARP Reply", buf_len);
 405                 return;
 406         case RTE_ARP_OP_REVREQUEST:
 407                 strlcpy(buf, "Reverse ARP Request", buf_len);
 408                 return;
 409         case RTE_ARP_OP_REVREPLY:
 410                 strlcpy(buf, "Reverse ARP Reply", buf_len);
 411                 return;
 412         case RTE_ARP_OP_INVREQUEST:
 413                 strlcpy(buf, "Peer Identify Request", buf_len);
 414                 return;
 415         case RTE_ARP_OP_INVREPLY:
 416                 strlcpy(buf, "Peer Identify Reply", buf_len);
 417                 return;
 418         default:
 419                 break;
 420         }
 421         strlcpy(buf, "Unknown", buf_len);
 422         return;
 423 }
 424 #endif
 425 #define MaxIPv4String   16
 426 static void
 427 ipv4_addr_to_dot(uint32_t be_ipv4_addr, char *buf, uint8_t buf_size)
 428 {
 429         uint32_t ipv4_addr;
 430
 431         ipv4_addr = rte_be_to_cpu_32(be_ipv4_addr);
 432         snprintf(buf, buf_size, "%d.%d.%d.%d", (ipv4_addr >> 24) & 0xFF,
 433                 (ipv4_addr >> 16) & 0xFF, (ipv4_addr >> 8) & 0xFF,
 434                 ipv4_addr & 0xFF);
 435 }
 436
 437 #define MAX_CLIENTS_NUMBER      128
 438 uint8_t active_clients;
 439 struct client_stats_t {
 440         uint16_t port;
 441         uint32_t ipv4_addr;
 442         uint32_t ipv4_rx_packets;
 443         uint32_t ipv4_tx_packets;
 444 };
 445 struct client_stats_t client_stats[MAX_CLIENTS_NUMBER];
 446
 447 static void
 448 update_client_stats(uint32_t addr, uint16_t port, uint32_t *TXorRXindicator)
 449 {
 450         int i = 0;
 451
 452         for (; i < MAX_CLIENTS_NUMBER; i++)     {
 453                 if ((client_stats[i].ipv4_addr == addr) && (client_stats[i].port == port))      {
 454                         /* Just update RX packets number for this client */
 455                         if (TXorRXindicator == &burstnumberRX)
 456                                 client_stats[i].ipv4_rx_packets++;
 457                         else
 458                                 client_stats[i].ipv4_tx_packets++;
 459                         return;
 460                 }
 461         }
 462         /* We have a new client. Insert him to the table, and increment stats */
 463         if (TXorRXindicator == &burstnumberRX)
 464                 client_stats[active_clients].ipv4_rx_packets++;
 465         else
 466                 client_stats[active_clients].ipv4_tx_packets++;
 467         client_stats[active_clients].ipv4_addr = addr;
 468         client_stats[active_clients].port = port;
 469         active_clients++;
 470
 471 }
 472
 473 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 474 #define MODE6_DEBUG(info, src_ip, dst_ip, eth_h, arp_op, port, burstnumber) \
 475         rte_log(RTE_LOG_DEBUG, bond_logtype,                            \
 476                 "%s port:%d SrcMAC:" RTE_ETHER_ADDR_PRT_FMT " SrcIP:%s " \
 477                 "DstMAC:" RTE_ETHER_ADDR_PRT_FMT " DstIP:%s %s %d\n", \
 478                 info,                                                   \
 479                 port,                                                   \
 480                 RTE_ETHER_ADDR_BYTES(&eth_h->src_addr),                  \
 481                 src_ip,                                                 \
 482                 RTE_ETHER_ADDR_BYTES(&eth_h->dst_addr),                  \
 483                 dst_ip,                                                 \
 484                 arp_op, ++burstnumber)
 485 #endif
 486
 487 static void
 488 mode6_debug(const char __rte_unused *info,
 489         struct rte_ether_hdr *eth_h, uint16_t port,
 490         uint32_t __rte_unused *burstnumber)
 491 {
 492         struct rte_ipv4_hdr *ipv4_h;
 493 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 494         struct rte_arp_hdr *arp_h;
 495         char dst_ip[16];
 496         char ArpOp[24];
 497         char buf[16];
 498 #endif
 499         char src_ip[16];
 500
 501         uint16_t ether_type = eth_h->ether_type;
 502         uint16_t offset = get_vlan_offset(eth_h, &ether_type);
 503
 504 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 505         strlcpy(buf, info, 16);
 506 #endif
 507
 508         if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
 509                 ipv4_h = (struct rte_ipv4_hdr *)((char *)(eth_h + 1) + offset);
 510                 ipv4_addr_to_dot(ipv4_h->src_addr, src_ip, MaxIPv4String);
 511 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 512                 ipv4_addr_to_dot(ipv4_h->dst_addr, dst_ip, MaxIPv4String);
 513                 MODE6_DEBUG(buf, src_ip, dst_ip, eth_h, "", port, *burstnumber);
 514 #endif
 515                 update_client_stats(ipv4_h->src_addr, port, burstnumber);
 516         }
 517 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 518         else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_ARP)) {
 519                 arp_h = (struct rte_arp_hdr *)((char *)(eth_h + 1) + offset);
 520                 ipv4_addr_to_dot(arp_h->arp_data.arp_sip, src_ip, MaxIPv4String);
 521                 ipv4_addr_to_dot(arp_h->arp_data.arp_tip, dst_ip, MaxIPv4String);
 522                 arp_op_name(rte_be_to_cpu_16(arp_h->arp_opcode),
 523                                 ArpOp, sizeof(ArpOp));
 524                 MODE6_DEBUG(buf, src_ip, dst_ip, eth_h, ArpOp, port, *burstnumber);
 525         }
 526 #endif
 527 }
 528 #endif
 529
 530 static uint16_t
 531 bond_ethdev_rx_burst_alb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 532 {
 533         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
 534         struct bond_dev_private *internals = bd_rx_q->dev_private;
 535         struct rte_ether_hdr *eth_h;
 536         uint16_t ether_type, offset;
 537         uint16_t nb_recv_pkts;
 538         int i;
 539
 540         nb_recv_pkts = bond_ethdev_rx_burst(queue, bufs, nb_pkts);
 541
 542         for (i = 0; i < nb_recv_pkts; i++) {
 543                 eth_h = rte_pktmbuf_mtod(bufs[i], struct rte_ether_hdr *);
 544                 ether_type = eth_h->ether_type;
 545                 offset = get_vlan_offset(eth_h, &ether_type);
 546
 547                 if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_ARP)) {
 548 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 549                         mode6_debug("RX ARP:", eth_h, bufs[i]->port, &burstnumberRX);
 550 #endif
 551                         bond_mode_alb_arp_recv(eth_h, offset, internals);
 552                 }
 553 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 554                 else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
 555                         mode6_debug("RX IPv4:", eth_h, bufs[i]->port, &burstnumberRX);
 556 #endif
 557         }
 558
 559         return nb_recv_pkts;
 560 }
 561
 562 static uint16_t
 563 bond_ethdev_tx_burst_round_robin(void *queue, struct rte_mbuf **bufs,
 564                 uint16_t nb_pkts)
 565 {
 566         struct bond_dev_private *internals;
 567         struct bond_tx_queue *bd_tx_q;
 568
 569         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_pkts];
 570         uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
 571
 572         uint16_t num_of_slaves;
 573         uint16_t slaves[RTE_MAX_ETHPORTS];
 574
 575         uint16_t num_tx_total = 0, num_tx_slave;
 576
 577         static int slave_idx = 0;
 578         int i, cslave_idx = 0, tx_fail_total = 0;
 579
 580         bd_tx_q = (struct bond_tx_queue *)queue;
 581         internals = bd_tx_q->dev_private;
 582
 583         /* Copy slave list to protect against slave up/down changes during tx
 584          * bursting */
 585         num_of_slaves = internals->active_slave_count;
 586         memcpy(slaves, internals->active_slaves,
 587                         sizeof(internals->active_slaves[0]) * num_of_slaves);
 588
 589         if (num_of_slaves < 1)
 590                 return num_tx_total;
 591
 592         /* Populate slaves mbuf with which packets are to be sent on it  */
 593         for (i = 0; i < nb_pkts; i++) {
 594                 cslave_idx = (slave_idx + i) % num_of_slaves;
 595                 slave_bufs[cslave_idx][(slave_nb_pkts[cslave_idx])++] = bufs[i];
 596         }
 597
 598         /* increment current slave index so the next call to tx burst starts on the
 599          * next slave */
 600         slave_idx = ++cslave_idx;
 601
 602         /* Send packet burst on each slave device */
 603         for (i = 0; i < num_of_slaves; i++) {
 604                 if (slave_nb_pkts[i] > 0) {
 605                         num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
 606                                         slave_bufs[i], slave_nb_pkts[i]);
 607
 608                         /* if tx burst fails move packets to end of bufs */
 609                         if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
 610                                 int tx_fail_slave = slave_nb_pkts[i] - num_tx_slave;
 611
 612                                 tx_fail_total += tx_fail_slave;
 613
 614                                 memcpy(&bufs[nb_pkts - tx_fail_total],
 615                                        &slave_bufs[i][num_tx_slave],
 616                                        tx_fail_slave * sizeof(bufs[0]));
 617                         }
 618                         num_tx_total += num_tx_slave;
 619                 }
 620         }
 621
 622         return num_tx_total;
 623 }
 624
 625 static uint16_t
 626 bond_ethdev_tx_burst_active_backup(void *queue,
 627                 struct rte_mbuf **bufs, uint16_t nb_pkts)
 628 {
 629         struct bond_dev_private *internals;
 630         struct bond_tx_queue *bd_tx_q;
 631
 632         bd_tx_q = (struct bond_tx_queue *)queue;
 633         internals = bd_tx_q->dev_private;
 634
 635         if (internals->active_slave_count < 1)
 636                 return 0;
 637
 638         return rte_eth_tx_burst(internals->current_primary_port, bd_tx_q->queue_id,
 639                         bufs, nb_pkts);
 640 }
 641
 642 static inline uint16_t
 643 ether_hash(struct rte_ether_hdr *eth_hdr)
 644 {
 645         unaligned_uint16_t *word_src_addr =
 646                 (unaligned_uint16_t *)eth_hdr->src_addr.addr_bytes;
 647         unaligned_uint16_t *word_dst_addr =
 648                 (unaligned_uint16_t *)eth_hdr->dst_addr.addr_bytes;
 649
 650         return (word_src_addr[0] ^ word_dst_addr[0]) ^
 651                         (word_src_addr[1] ^ word_dst_addr[1]) ^
 652                         (word_src_addr[2] ^ word_dst_addr[2]);
 653 }
 654
 655 static inline uint32_t
 656 ipv4_hash(struct rte_ipv4_hdr *ipv4_hdr)
 657 {
 658         return ipv4_hdr->src_addr ^ ipv4_hdr->dst_addr;
 659 }
 660
 661 static inline uint32_t
 662 ipv6_hash(struct rte_ipv6_hdr *ipv6_hdr)
 663 {
 664         unaligned_uint32_t *word_src_addr =
 665                 (unaligned_uint32_t *)&(ipv6_hdr->src_addr[0]);
 666         unaligned_uint32_t *word_dst_addr =
 667                 (unaligned_uint32_t *)&(ipv6_hdr->dst_addr[0]);
 668
 669         return (word_src_addr[0] ^ word_dst_addr[0]) ^
 670                         (word_src_addr[1] ^ word_dst_addr[1]) ^
 671                         (word_src_addr[2] ^ word_dst_addr[2]) ^
 672                         (word_src_addr[3] ^ word_dst_addr[3]);
 673 }
 674
 675
 676 void
 677 burst_xmit_l2_hash(struct rte_mbuf **buf, uint16_t nb_pkts,
 678                 uint16_t slave_count, uint16_t *slaves)
 679 {
 680         struct rte_ether_hdr *eth_hdr;
 681         uint32_t hash;
 682         int i;
 683
 684         for (i = 0; i < nb_pkts; i++) {
 685                 eth_hdr = rte_pktmbuf_mtod(buf[i], struct rte_ether_hdr *);
 686
 687                 hash = ether_hash(eth_hdr);
 688
 689                 slaves[i] = (hash ^= hash >> 8) % slave_count;
 690         }
 691 }
 692
 693 void
 694 burst_xmit_l23_hash(struct rte_mbuf **buf, uint16_t nb_pkts,
 695                 uint16_t slave_count, uint16_t *slaves)
 696 {
 697         uint16_t i;
 698         struct rte_ether_hdr *eth_hdr;
 699         uint16_t proto;
 700         size_t vlan_offset;
 701         uint32_t hash, l3hash;
 702
 703         for (i = 0; i < nb_pkts; i++) {
 704                 eth_hdr = rte_pktmbuf_mtod(buf[i], struct rte_ether_hdr *);
 705                 l3hash = 0;
 706
 707                 proto = eth_hdr->ether_type;
 708                 hash = ether_hash(eth_hdr);
 709
 710                 vlan_offset = get_vlan_offset(eth_hdr, &proto);
 711
 712                 if (rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) == proto) {
 713                         struct rte_ipv4_hdr *ipv4_hdr = (struct rte_ipv4_hdr *)
 714                                         ((char *)(eth_hdr + 1) + vlan_offset);
 715                         l3hash = ipv4_hash(ipv4_hdr);
 716
 717                 } else if (rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) == proto) {
 718                         struct rte_ipv6_hdr *ipv6_hdr = (struct rte_ipv6_hdr *)
 719                                         ((char *)(eth_hdr + 1) + vlan_offset);
 720                         l3hash = ipv6_hash(ipv6_hdr);
 721                 }
 722
 723                 hash = hash ^ l3hash;
 724                 hash ^= hash >> 16;
 725                 hash ^= hash >> 8;
 726
 727                 slaves[i] = hash % slave_count;
 728         }
 729 }
 730
 731 void
 732 burst_xmit_l34_hash(struct rte_mbuf **buf, uint16_t nb_pkts,
 733                 uint16_t slave_count, uint16_t *slaves)
 734 {
 735         struct rte_ether_hdr *eth_hdr;
 736         uint16_t proto;
 737         size_t vlan_offset;
 738         int i;
 739
 740         struct rte_udp_hdr *udp_hdr;
 741         struct rte_tcp_hdr *tcp_hdr;
 742         uint32_t hash, l3hash, l4hash;
 743
 744         for (i = 0; i < nb_pkts; i++) {
 745                 eth_hdr = rte_pktmbuf_mtod(buf[i], struct rte_ether_hdr *);
 746                 size_t pkt_end = (size_t)eth_hdr + rte_pktmbuf_data_len(buf[i]);
 747                 proto = eth_hdr->ether_type;
 748                 vlan_offset = get_vlan_offset(eth_hdr, &proto);
 749                 l3hash = 0;
 750                 l4hash = 0;
 751
 752                 if (rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) == proto) {
 753                         struct rte_ipv4_hdr *ipv4_hdr = (struct rte_ipv4_hdr *)
 754                                         ((char *)(eth_hdr + 1) + vlan_offset);
 755                         size_t ip_hdr_offset;
 756
 757                         l3hash = ipv4_hash(ipv4_hdr);
 758
 759                         /* there is no L4 header in fragmented packet */
 760                         if (likely(rte_ipv4_frag_pkt_is_fragmented(ipv4_hdr)
 761                                                                 == 0)) {
 762                                 ip_hdr_offset = (ipv4_hdr->version_ihl
 763                                         & RTE_IPV4_HDR_IHL_MASK) *
 764                                         RTE_IPV4_IHL_MULTIPLIER;
 765
 766                                 if (ipv4_hdr->next_proto_id == IPPROTO_TCP) {
 767                                         tcp_hdr = (struct rte_tcp_hdr *)
 768                                                 ((char *)ipv4_hdr +
 769                                                         ip_hdr_offset);
 770                                         if ((size_t)tcp_hdr + sizeof(*tcp_hdr)
 771                                                         < pkt_end)
 772                                                 l4hash = HASH_L4_PORTS(tcp_hdr);
 773                                 } else if (ipv4_hdr->next_proto_id ==
 774                                                                 IPPROTO_UDP) {
 775                                         udp_hdr = (struct rte_udp_hdr *)
 776                                                 ((char *)ipv4_hdr +
 777                                                         ip_hdr_offset);
 778                                         if ((size_t)udp_hdr + sizeof(*udp_hdr)
 779                                                         < pkt_end)
 780                                                 l4hash = HASH_L4_PORTS(udp_hdr);
 781                                 }
 782                         }
 783                 } else if  (rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) == proto) {
 784                         struct rte_ipv6_hdr *ipv6_hdr = (struct rte_ipv6_hdr *)
 785                                         ((char *)(eth_hdr + 1) + vlan_offset);
 786                         l3hash = ipv6_hash(ipv6_hdr);
 787
 788                         if (ipv6_hdr->proto == IPPROTO_TCP) {
 789                                 tcp_hdr = (struct rte_tcp_hdr *)(ipv6_hdr + 1);
 790                                 l4hash = HASH_L4_PORTS(tcp_hdr);
 791                         } else if (ipv6_hdr->proto == IPPROTO_UDP) {
 792                                 udp_hdr = (struct rte_udp_hdr *)(ipv6_hdr + 1);
 793                                 l4hash = HASH_L4_PORTS(udp_hdr);
 794                         }
 795                 }
 796
 797                 hash = l3hash ^ l4hash;
 798                 hash ^= hash >> 16;
 799                 hash ^= hash >> 8;
 800
 801                 slaves[i] = hash % slave_count;
 802         }
 803 }
 804
 805 struct bwg_slave {
 806         uint64_t bwg_left_int;
 807         uint64_t bwg_left_remainder;
 808         uint16_t slave;
 809 };
 810
 811 void
 812 bond_tlb_activate_slave(struct bond_dev_private *internals) {
 813         int i;
 814
 815         for (i = 0; i < internals->active_slave_count; i++) {
 816                 tlb_last_obytets[internals->active_slaves[i]] = 0;
 817         }
 818 }
 819
 820 static int
 821 bandwidth_cmp(const void *a, const void *b)
 822 {
 823         const struct bwg_slave *bwg_a = a;
 824         const struct bwg_slave *bwg_b = b;
 825         int64_t diff = (int64_t)bwg_b->bwg_left_int - (int64_t)bwg_a->bwg_left_int;
 826         int64_t diff2 = (int64_t)bwg_b->bwg_left_remainder -
 827                         (int64_t)bwg_a->bwg_left_remainder;
 828         if (diff > 0)
 829                 return 1;
 830         else if (diff < 0)
 831                 return -1;
 832         else if (diff2 > 0)
 833                 return 1;
 834         else if (diff2 < 0)
 835                 return -1;
 836         else
 837                 return 0;
 838 }
 839
 840 static void
 841 bandwidth_left(uint16_t port_id, uint64_t load, uint8_t update_idx,
 842                 struct bwg_slave *bwg_slave)
 843 {
 844         struct rte_eth_link link_status;
 845         int ret;
 846
 847         ret = rte_eth_link_get_nowait(port_id, &link_status);
 848         if (ret < 0) {
 849                 RTE_BOND_LOG(ERR, "Slave (port %u) link get failed: %s",
 850                              port_id, rte_strerror(-ret));
 851                 return;
 852         }
 853         uint64_t link_bwg = link_status.link_speed * 1000000ULL / 8;
 854         if (link_bwg == 0)
 855                 return;
 856         link_bwg = link_bwg * (update_idx+1) * REORDER_PERIOD_MS;
 857         bwg_slave->bwg_left_int = (link_bwg - 1000*load) / link_bwg;
 858         bwg_slave->bwg_left_remainder = (link_bwg - 1000*load) % link_bwg;
 859 }
 860
 861 static void
 862 bond_ethdev_update_tlb_slave_cb(void *arg)
 863 {
 864         struct bond_dev_private *internals = arg;
 865         struct rte_eth_stats slave_stats;
 866         struct bwg_slave bwg_array[RTE_MAX_ETHPORTS];
 867         uint16_t slave_count;
 868         uint64_t tx_bytes;
 869
 870         uint8_t update_stats = 0;
 871         uint16_t slave_id;
 872         uint16_t i;
 873
 874         internals->slave_update_idx++;
 875
 876
 877         if (internals->slave_update_idx >= REORDER_PERIOD_MS)
 878                 update_stats = 1;
 879
 880         for (i = 0; i < internals->active_slave_count; i++) {
 881                 slave_id = internals->active_slaves[i];
 882                 rte_eth_stats_get(slave_id, &slave_stats);
 883                 tx_bytes = slave_stats.obytes - tlb_last_obytets[slave_id];
 884                 bandwidth_left(slave_id, tx_bytes,
 885                                 internals->slave_update_idx, &bwg_array[i]);
 886                 bwg_array[i].slave = slave_id;
 887
 888                 if (update_stats) {
 889                         tlb_last_obytets[slave_id] = slave_stats.obytes;
 890                 }
 891         }
 892
 893         if (update_stats == 1)
 894                 internals->slave_update_idx = 0;
 895
 896         slave_count = i;
 897         qsort(bwg_array, slave_count, sizeof(bwg_array[0]), bandwidth_cmp);
 898         for (i = 0; i < slave_count; i++)
 899                 internals->tlb_slaves_order[i] = bwg_array[i].slave;
 900
 901         rte_eal_alarm_set(REORDER_PERIOD_MS * 1000, bond_ethdev_update_tlb_slave_cb,
 902                         (struct bond_dev_private *)internals);
 903 }
 904
 905 static uint16_t
 906 bond_ethdev_tx_burst_tlb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 907 {
 908         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
 909         struct bond_dev_private *internals = bd_tx_q->dev_private;
 910
 911         struct rte_eth_dev *primary_port =
 912                         &rte_eth_devices[internals->primary_port];
 913         uint16_t num_tx_total = 0;
 914         uint16_t i, j;
 915
 916         uint16_t num_of_slaves = internals->active_slave_count;
 917         uint16_t slaves[RTE_MAX_ETHPORTS];
 918
 919         struct rte_ether_hdr *ether_hdr;
 920         struct rte_ether_addr primary_slave_addr;
 921         struct rte_ether_addr active_slave_addr;
 922
 923         if (num_of_slaves < 1)
 924                 return num_tx_total;
 925
 926         memcpy(slaves, internals->tlb_slaves_order,
 927                                 sizeof(internals->tlb_slaves_order[0]) * num_of_slaves);
 928
 929
 930         rte_ether_addr_copy(primary_port->data->mac_addrs, &primary_slave_addr);
 931
 932         if (nb_pkts > 3) {
 933                 for (i = 0; i < 3; i++)
 934                         rte_prefetch0(rte_pktmbuf_mtod(bufs[i], void*));
 935         }
 936
 937         for (i = 0; i < num_of_slaves; i++) {
 938                 rte_eth_macaddr_get(slaves[i], &active_slave_addr);
 939                 for (j = num_tx_total; j < nb_pkts; j++) {
 940                         if (j + 3 < nb_pkts)
 941                                 rte_prefetch0(rte_pktmbuf_mtod(bufs[j+3], void*));
 942
 943                         ether_hdr = rte_pktmbuf_mtod(bufs[j],
 944                                                 struct rte_ether_hdr *);
 945                         if (rte_is_same_ether_addr(&ether_hdr->src_addr,
 946                                                         &primary_slave_addr))
 947                                 rte_ether_addr_copy(&active_slave_addr,
 948                                                 &ether_hdr->src_addr);
 949 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 950                                         mode6_debug("TX IPv4:", ether_hdr, slaves[i], &burstnumberTX);
 951 #endif
 952                 }
 953
 954                 num_tx_total += rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
 955                                 bufs + num_tx_total, nb_pkts - num_tx_total);
 956
 957                 if (num_tx_total == nb_pkts)
 958                         break;
 959         }
 960
 961         return num_tx_total;
 962 }
 963
 964 void
 965 bond_tlb_disable(struct bond_dev_private *internals)
 966 {
 967         rte_eal_alarm_cancel(bond_ethdev_update_tlb_slave_cb, internals);
 968 }
 969
 970 void
 971 bond_tlb_enable(struct bond_dev_private *internals)
 972 {
 973         bond_ethdev_update_tlb_slave_cb(internals);
 974 }
 975
 976 static uint16_t
 977 bond_ethdev_tx_burst_alb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 978 {
 979         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
 980         struct bond_dev_private *internals = bd_tx_q->dev_private;
 981
 982         struct rte_ether_hdr *eth_h;
 983         uint16_t ether_type, offset;
 984
 985         struct client_data *client_info;
 986
 987         /*
 988          * We create transmit buffers for every slave and one additional to send
 989          * through tlb. In worst case every packet will be send on one port.
 990          */
 991         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS + 1][nb_pkts];
 992         uint16_t slave_bufs_pkts[RTE_MAX_ETHPORTS + 1] = { 0 };
 993
 994         /*
 995          * We create separate transmit buffers for update packets as they won't
 996          * be counted in num_tx_total.
 997          */
 998         struct rte_mbuf *update_bufs[RTE_MAX_ETHPORTS][ALB_HASH_TABLE_SIZE];
 999         uint16_t update_bufs_pkts[RTE_MAX_ETHPORTS] = { 0 };
1000
1001         struct rte_mbuf *upd_pkt;
1002         size_t pkt_size;
1003
1004         uint16_t num_send, num_not_send = 0;
1005         uint16_t num_tx_total = 0;
1006         uint16_t slave_idx;
1007
1008         int i, j;
1009
1010         /* Search tx buffer for ARP packets and forward them to alb */
1011         for (i = 0; i < nb_pkts; i++) {
1012                 eth_h = rte_pktmbuf_mtod(bufs[i], struct rte_ether_hdr *);
1013                 ether_type = eth_h->ether_type;
1014                 offset = get_vlan_offset(eth_h, &ether_type);
1015
1016                 if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_ARP)) {
1017                         slave_idx = bond_mode_alb_arp_xmit(eth_h, offset, internals);
1018
1019                         /* Change src mac in eth header */
1020                         rte_eth_macaddr_get(slave_idx, &eth_h->src_addr);
1021
1022                         /* Add packet to slave tx buffer */
1023                         slave_bufs[slave_idx][slave_bufs_pkts[slave_idx]] = bufs[i];
1024                         slave_bufs_pkts[slave_idx]++;
1025                 } else {
1026                         /* If packet is not ARP, send it with TLB policy */
1027                         slave_bufs[RTE_MAX_ETHPORTS][slave_bufs_pkts[RTE_MAX_ETHPORTS]] =
1028                                         bufs[i];
1029                         slave_bufs_pkts[RTE_MAX_ETHPORTS]++;
1030                 }
1031         }
1032
1033         /* Update connected client ARP tables */
1034         if (internals->mode6.ntt) {
1035                 for (i = 0; i < ALB_HASH_TABLE_SIZE; i++) {
1036                         client_info = &internals->mode6.client_table[i];
1037
1038                         if (client_info->in_use) {
1039                                 /* Allocate new packet to send ARP update on current slave */
1040                                 upd_pkt = rte_pktmbuf_alloc(internals->mode6.mempool);
1041                                 if (upd_pkt == NULL) {
1042                                         RTE_BOND_LOG(ERR,
1043                                                      "Failed to allocate ARP packet from pool");
1044                                         continue;
1045                                 }
1046                                 pkt_size = sizeof(struct rte_ether_hdr) +
1047                                         sizeof(struct rte_arp_hdr) +
1048                                         client_info->vlan_count *
1049                                         sizeof(struct rte_vlan_hdr);
1050                                 upd_pkt->data_len = pkt_size;
1051                                 upd_pkt->pkt_len = pkt_size;
1052
1053                                 slave_idx = bond_mode_alb_arp_upd(client_info, upd_pkt,
1054                                                 internals);
1055
1056                                 /* Add packet to update tx buffer */
1057                                 update_bufs[slave_idx][update_bufs_pkts[slave_idx]] = upd_pkt;
1058                                 update_bufs_pkts[slave_idx]++;
1059                         }
1060                 }
1061                 internals->mode6.ntt = 0;
1062         }
1063
1064         /* Send ARP packets on proper slaves */
1065         for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
1066                 if (slave_bufs_pkts[i] > 0) {
1067                         num_send = rte_eth_tx_burst(i, bd_tx_q->queue_id,
1068                                         slave_bufs[i], slave_bufs_pkts[i]);
1069                         for (j = 0; j < slave_bufs_pkts[i] - num_send; j++) {
1070                                 bufs[nb_pkts - 1 - num_not_send - j] =
1071                                                 slave_bufs[i][nb_pkts - 1 - j];
1072                         }
1073
1074                         num_tx_total += num_send;
1075                         num_not_send += slave_bufs_pkts[i] - num_send;
1076
1077 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
1078         /* Print TX stats including update packets */
1079                         for (j = 0; j < slave_bufs_pkts[i]; j++) {
1080                                 eth_h = rte_pktmbuf_mtod(slave_bufs[i][j],
1081                                                         struct rte_ether_hdr *);
1082                                 mode6_debug("TX ARP:", eth_h, i, &burstnumberTX);
1083                         }
1084 #endif
1085                 }
1086         }
1087
1088         /* Send update packets on proper slaves */
1089         for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
1090                 if (update_bufs_pkts[i] > 0) {
1091                         num_send = rte_eth_tx_burst(i, bd_tx_q->queue_id, update_bufs[i],
1092                                         update_bufs_pkts[i]);
1093                         for (j = num_send; j < update_bufs_pkts[i]; j++) {
1094                                 rte_pktmbuf_free(update_bufs[i][j]);
1095                         }
1096 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
1097                         for (j = 0; j < update_bufs_pkts[i]; j++) {
1098                                 eth_h = rte_pktmbuf_mtod(update_bufs[i][j],
1099                                                         struct rte_ether_hdr *);
1100                                 mode6_debug("TX ARPupd:", eth_h, i, &burstnumberTX);
1101                         }
1102 #endif
1103                 }
1104         }
1105
1106         /* Send non-ARP packets using tlb policy */
1107         if (slave_bufs_pkts[RTE_MAX_ETHPORTS] > 0) {
1108                 num_send = bond_ethdev_tx_burst_tlb(queue,
1109                                 slave_bufs[RTE_MAX_ETHPORTS],
1110                                 slave_bufs_pkts[RTE_MAX_ETHPORTS]);
1111
1112                 for (j = 0; j < slave_bufs_pkts[RTE_MAX_ETHPORTS]; j++) {
1113                         bufs[nb_pkts - 1 - num_not_send - j] =
1114                                         slave_bufs[RTE_MAX_ETHPORTS][nb_pkts - 1 - j];
1115                 }
1116
1117                 num_tx_total += num_send;
1118         }
1119
1120         return num_tx_total;
1121 }
1122
1123 static inline uint16_t
1124 tx_burst_balance(void *queue, struct rte_mbuf **bufs, uint16_t nb_bufs,
1125                  uint16_t *slave_port_ids, uint16_t slave_count)
1126 {
1127         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
1128         struct bond_dev_private *internals = bd_tx_q->dev_private;
1129
1130         /* Array to sort mbufs for transmission on each slave into */
1131         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_bufs];
1132         /* Number of mbufs for transmission on each slave */
1133         uint16_t slave_nb_bufs[RTE_MAX_ETHPORTS] = { 0 };
1134         /* Mapping array generated by hash function to map mbufs to slaves */
1135         uint16_t bufs_slave_port_idxs[nb_bufs];
1136
1137         uint16_t slave_tx_count;
1138         uint16_t total_tx_count = 0, total_tx_fail_count = 0;
1139
1140         uint16_t i;
1141
1142         /*
1143          * Populate slaves mbuf with the packets which are to be sent on it
1144          * selecting output slave using hash based on xmit policy
1145          */
1146         internals->burst_xmit_hash(bufs, nb_bufs, slave_count,
1147                         bufs_slave_port_idxs);
1148
1149         for (i = 0; i < nb_bufs; i++) {
1150                 /* Populate slave mbuf arrays with mbufs for that slave. */
1151                 uint16_t slave_idx = bufs_slave_port_idxs[i];
1152
1153                 slave_bufs[slave_idx][slave_nb_bufs[slave_idx]++] = bufs[i];
1154         }
1155
1156         /* Send packet burst on each slave device */
1157         for (i = 0; i < slave_count; i++) {
1158                 if (slave_nb_bufs[i] == 0)
1159                         continue;
1160
1161                 slave_tx_count = rte_eth_tx_burst(slave_port_ids[i],
1162                                 bd_tx_q->queue_id, slave_bufs[i],
1163                                 slave_nb_bufs[i]);
1164
1165                 total_tx_count += slave_tx_count;
1166
1167                 /* If tx burst fails move packets to end of bufs */
1168                 if (unlikely(slave_tx_count < slave_nb_bufs[i])) {
1169                         int slave_tx_fail_count = slave_nb_bufs[i] -
1170                                         slave_tx_count;
1171                         total_tx_fail_count += slave_tx_fail_count;
1172                         memcpy(&bufs[nb_bufs - total_tx_fail_count],
1173                                &slave_bufs[i][slave_tx_count],
1174                                slave_tx_fail_count * sizeof(bufs[0]));
1175                 }
1176         }
1177
1178         return total_tx_count;
1179 }
1180
1181 static uint16_t
1182 bond_ethdev_tx_burst_balance(void *queue, struct rte_mbuf **bufs,
1183                 uint16_t nb_bufs)
1184 {
1185         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
1186         struct bond_dev_private *internals = bd_tx_q->dev_private;
1187
1188         uint16_t slave_port_ids[RTE_MAX_ETHPORTS];
1189         uint16_t slave_count;
1190
1191         if (unlikely(nb_bufs == 0))
1192                 return 0;
1193
1194         /* Copy slave list to protect against slave up/down changes during tx
1195          * bursting
1196          */
1197         slave_count = internals->active_slave_count;
1198         if (unlikely(slave_count < 1))
1199                 return 0;
1200
1201         memcpy(slave_port_ids, internals->active_slaves,
1202                         sizeof(slave_port_ids[0]) * slave_count);
1203         return tx_burst_balance(queue, bufs, nb_bufs, slave_port_ids,
1204                                 slave_count);
1205 }
1206
1207 static inline uint16_t
1208 tx_burst_8023ad(void *queue, struct rte_mbuf **bufs, uint16_t nb_bufs,
1209                 bool dedicated_txq)
1210 {
1211         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
1212         struct bond_dev_private *internals = bd_tx_q->dev_private;
1213
1214         uint16_t slave_port_ids[RTE_MAX_ETHPORTS];
1215         uint16_t slave_count;
1216
1217         uint16_t dist_slave_port_ids[RTE_MAX_ETHPORTS];
1218         uint16_t dist_slave_count;
1219
1220         uint16_t slave_tx_count;
1221
1222         uint16_t i;
1223
1224         /* Copy slave list to protect against slave up/down changes during tx
1225          * bursting */
1226         slave_count = internals->active_slave_count;
1227         if (unlikely(slave_count < 1))
1228                 return 0;
1229
1230         memcpy(slave_port_ids, internals->active_slaves,
1231                         sizeof(slave_port_ids[0]) * slave_count);
1232
1233         if (dedicated_txq)
1234                 goto skip_tx_ring;
1235
1236         /* Check for LACP control packets and send if available */
1237         for (i = 0; i < slave_count; i++) {
1238                 struct port *port = &bond_mode_8023ad_ports[slave_port_ids[i]];
1239                 struct rte_mbuf *ctrl_pkt = NULL;
1240
1241                 if (likely(rte_ring_empty(port->tx_ring)))
1242                         continue;
1243
1244                 if (rte_ring_dequeue(port->tx_ring,
1245                                      (void **)&ctrl_pkt) != -ENOENT) {
1246                         slave_tx_count = rte_eth_tx_burst(slave_port_ids[i],
1247                                         bd_tx_q->queue_id, &ctrl_pkt, 1);
1248                         /*
1249                          * re-enqueue LAG control plane packets to buffering
1250                          * ring if transmission fails so the packet isn't lost.
1251                          */
1252                         if (slave_tx_count != 1)
1253                                 rte_ring_enqueue(port->tx_ring, ctrl_pkt);
1254                 }
1255         }
1256
1257 skip_tx_ring:
1258         if (unlikely(nb_bufs == 0))
1259                 return 0;
1260
1261         dist_slave_count = 0;
1262         for (i = 0; i < slave_count; i++) {
1263                 struct port *port = &bond_mode_8023ad_ports[slave_port_ids[i]];
1264
1265                 if (ACTOR_STATE(port, DISTRIBUTING))
1266                         dist_slave_port_ids[dist_slave_count++] =
1267                                         slave_port_ids[i];
1268         }
1269
1270         if (unlikely(dist_slave_count < 1))
1271                 return 0;
1272
1273         return tx_burst_balance(queue, bufs, nb_bufs, dist_slave_port_ids,
1274                                 dist_slave_count);
1275 }
1276
1277 static uint16_t
1278 bond_ethdev_tx_burst_8023ad(void *queue, struct rte_mbuf **bufs,
1279                 uint16_t nb_bufs)
1280 {
1281         return tx_burst_8023ad(queue, bufs, nb_bufs, false);
1282 }
1283
1284 static uint16_t
1285 bond_ethdev_tx_burst_8023ad_fast_queue(void *queue, struct rte_mbuf **bufs,
1286                 uint16_t nb_bufs)
1287 {
1288         return tx_burst_8023ad(queue, bufs, nb_bufs, true);
1289 }
1290
1291 static uint16_t
1292 bond_ethdev_tx_burst_broadcast(void *queue, struct rte_mbuf **bufs,
1293                 uint16_t nb_pkts)
1294 {
1295         struct bond_dev_private *internals;
1296         struct bond_tx_queue *bd_tx_q;
1297
1298         uint16_t slaves[RTE_MAX_ETHPORTS];
1299         uint8_t tx_failed_flag = 0;
1300         uint16_t num_of_slaves;
1301
1302         uint16_t max_nb_of_tx_pkts = 0;
1303
1304         int slave_tx_total[RTE_MAX_ETHPORTS];
1305         int i, most_successful_tx_slave = -1;
1306
1307         bd_tx_q = (struct bond_tx_queue *)queue;
1308         internals = bd_tx_q->dev_private;
1309
1310         /* Copy slave list to protect against slave up/down changes during tx
1311          * bursting */
1312         num_of_slaves = internals->active_slave_count;
1313         memcpy(slaves, internals->active_slaves,
1314                         sizeof(internals->active_slaves[0]) * num_of_slaves);
1315
1316         if (num_of_slaves < 1)
1317                 return 0;
1318
1319         /* Increment reference count on mbufs */
1320         for (i = 0; i < nb_pkts; i++)
1321                 rte_mbuf_refcnt_update(bufs[i], num_of_slaves - 1);
1322
1323         /* Transmit burst on each active slave */
1324         for (i = 0; i < num_of_slaves; i++) {
1325                 slave_tx_total[i] = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
1326                                         bufs, nb_pkts);
1327
1328                 if (unlikely(slave_tx_total[i] < nb_pkts))
1329                         tx_failed_flag = 1;
1330
1331                 /* record the value and slave index for the slave which transmits the
1332                  * maximum number of packets */
1333                 if (slave_tx_total[i] > max_nb_of_tx_pkts) {
1334                         max_nb_of_tx_pkts = slave_tx_total[i];
1335                         most_successful_tx_slave = i;
1336                 }
1337         }
1338
1339         /* if slaves fail to transmit packets from burst, the calling application
1340          * is not expected to know about multiple references to packets so we must
1341          * handle failures of all packets except those of the most successful slave
1342          */
1343         if (unlikely(tx_failed_flag))
1344                 for (i = 0; i < num_of_slaves; i++)
1345                         if (i != most_successful_tx_slave)
1346                                 while (slave_tx_total[i] < nb_pkts)
1347                                         rte_pktmbuf_free(bufs[slave_tx_total[i]++]);
1348
1349         return max_nb_of_tx_pkts;
1350 }
1351
1352 static void
1353 link_properties_set(struct rte_eth_dev *ethdev, struct rte_eth_link *slave_link)
1354 {
1355         struct bond_dev_private *bond_ctx = ethdev->data->dev_private;
1356
1357         if (bond_ctx->mode == BONDING_MODE_8023AD) {
1358                 /**
1359                  * If in mode 4 then save the link properties of the first
1360                  * slave, all subsequent slaves must match these properties
1361                  */
1362                 struct rte_eth_link *bond_link = &bond_ctx->mode4.slave_link;
1363
1364                 bond_link->link_autoneg = slave_link->link_autoneg;
1365                 bond_link->link_duplex = slave_link->link_duplex;
1366                 bond_link->link_speed = slave_link->link_speed;
1367         } else {
1368                 /**
1369                  * In any other mode the link properties are set to default
1370                  * values of AUTONEG/DUPLEX
1371                  */
1372                 ethdev->data->dev_link.link_autoneg = RTE_ETH_LINK_AUTONEG;
1373                 ethdev->data->dev_link.link_duplex = RTE_ETH_LINK_FULL_DUPLEX;
1374         }
1375 }
1376
1377 static int
1378 link_properties_valid(struct rte_eth_dev *ethdev,
1379                 struct rte_eth_link *slave_link)
1380 {
1381         struct bond_dev_private *bond_ctx = ethdev->data->dev_private;
1382
1383         if (bond_ctx->mode == BONDING_MODE_8023AD) {
1384                 struct rte_eth_link *bond_link = &bond_ctx->mode4.slave_link;
1385
1386                 if (bond_link->link_duplex != slave_link->link_duplex ||
1387                         bond_link->link_autoneg != slave_link->link_autoneg ||
1388                         bond_link->link_speed != slave_link->link_speed)
1389                         return -1;
1390         }
1391
1392         return 0;
1393 }
1394
1395 int
1396 mac_address_get(struct rte_eth_dev *eth_dev,
1397                 struct rte_ether_addr *dst_mac_addr)
1398 {
1399         struct rte_ether_addr *mac_addr;
1400
1401         if (eth_dev == NULL) {
1402                 RTE_BOND_LOG(ERR, "NULL pointer eth_dev specified");
1403                 return -1;
1404         }
1405
1406         if (dst_mac_addr == NULL) {
1407                 RTE_BOND_LOG(ERR, "NULL pointer MAC specified");
1408                 return -1;
1409         }
1410
1411         mac_addr = eth_dev->data->mac_addrs;
1412
1413         rte_ether_addr_copy(mac_addr, dst_mac_addr);
1414         return 0;
1415 }
1416
1417 int
1418 mac_address_set(struct rte_eth_dev *eth_dev,
1419                 struct rte_ether_addr *new_mac_addr)
1420 {
1421         struct rte_ether_addr *mac_addr;
1422
1423         if (eth_dev == NULL) {
1424                 RTE_BOND_LOG(ERR, "NULL pointer eth_dev specified");
1425                 return -1;
1426         }
1427
1428         if (new_mac_addr == NULL) {
1429                 RTE_BOND_LOG(ERR, "NULL pointer MAC specified");
1430                 return -1;
1431         }
1432
1433         mac_addr = eth_dev->data->mac_addrs;
1434
1435         /* If new MAC is different to current MAC then update */
1436         if (memcmp(mac_addr, new_mac_addr, sizeof(*mac_addr)) != 0)
1437                 memcpy(mac_addr, new_mac_addr, sizeof(*mac_addr));
1438
1439         return 0;
1440 }
1441
1442 static const struct rte_ether_addr null_mac_addr;
1443
1444 /*
1445  * Add additional MAC addresses to the slave
1446  */
1447 int
1448 slave_add_mac_addresses(struct rte_eth_dev *bonded_eth_dev,
1449                 uint16_t slave_port_id)
1450 {
1451         int i, ret;
1452         struct rte_ether_addr *mac_addr;
1453
1454         for (i = 1; i < BOND_MAX_MAC_ADDRS; i++) {
1455                 mac_addr = &bonded_eth_dev->data->mac_addrs[i];
1456                 if (rte_is_same_ether_addr(mac_addr, &null_mac_addr))
1457                         break;
1458
1459                 ret = rte_eth_dev_mac_addr_add(slave_port_id, mac_addr, 0);
1460                 if (ret < 0) {
1461                         /* rollback */
1462                         for (i--; i > 0; i--)
1463                                 rte_eth_dev_mac_addr_remove(slave_port_id,
1464                                         &bonded_eth_dev->data->mac_addrs[i]);
1465                         return ret;
1466                 }
1467         }
1468
1469         return 0;
1470 }
1471
1472 /*
1473  * Remove additional MAC addresses from the slave
1474  */
1475 int
1476 slave_remove_mac_addresses(struct rte_eth_dev *bonded_eth_dev,
1477                 uint16_t slave_port_id)
1478 {
1479         int i, rc, ret;
1480         struct rte_ether_addr *mac_addr;
1481
1482         rc = 0;
1483         for (i = 1; i < BOND_MAX_MAC_ADDRS; i++) {
1484                 mac_addr = &bonded_eth_dev->data->mac_addrs[i];
1485                 if (rte_is_same_ether_addr(mac_addr, &null_mac_addr))
1486                         break;
1487
1488                 ret = rte_eth_dev_mac_addr_remove(slave_port_id, mac_addr);
1489                 /* save only the first error */
1490                 if (ret < 0 && rc == 0)
1491                         rc = ret;
1492         }
1493
1494         return rc;
1495 }
1496
1497 int
1498 mac_address_slaves_update(struct rte_eth_dev *bonded_eth_dev)
1499 {
1500         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1501         bool set;
1502         int i;
1503
1504         /* Update slave devices MAC addresses */
1505         if (internals->slave_count < 1)
1506                 return -1;
1507
1508         switch (internals->mode) {
1509         case BONDING_MODE_ROUND_ROBIN:
1510         case BONDING_MODE_BALANCE:
1511         case BONDING_MODE_BROADCAST:
1512                 for (i = 0; i < internals->slave_count; i++) {
1513                         if (rte_eth_dev_default_mac_addr_set(
1514                                         internals->slaves[i].port_id,
1515                                         bonded_eth_dev->data->mac_addrs)) {
1516                                 RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1517                                                 internals->slaves[i].port_id);
1518                                 return -1;
1519                         }
1520                 }
1521                 break;
1522         case BONDING_MODE_8023AD:
1523                 bond_mode_8023ad_mac_address_update(bonded_eth_dev);
1524                 break;
1525         case BONDING_MODE_ACTIVE_BACKUP:
1526         case BONDING_MODE_TLB:
1527         case BONDING_MODE_ALB:
1528         default:
1529                 set = true;
1530                 for (i = 0; i < internals->slave_count; i++) {
1531                         if (internals->slaves[i].port_id ==
1532                                         internals->current_primary_port) {
1533                                 if (rte_eth_dev_default_mac_addr_set(
1534                                                 internals->current_primary_port,
1535                                                 bonded_eth_dev->data->mac_addrs)) {
1536                                         RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1537                                                         internals->current_primary_port);
1538                                         set = false;
1539                                 }
1540                         } else {
1541                                 if (rte_eth_dev_default_mac_addr_set(
1542                                                 internals->slaves[i].port_id,
1543                                                 &internals->slaves[i].persisted_mac_addr)) {
1544                                         RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1545                                                         internals->slaves[i].port_id);
1546                                 }
1547                         }
1548                 }
1549                 if (!set)
1550                         return -1;
1551         }
1552
1553         return 0;
1554 }
1555
1556 int
1557 bond_ethdev_mode_set(struct rte_eth_dev *eth_dev, int mode)
1558 {
1559         struct bond_dev_private *internals;
1560
1561         internals = eth_dev->data->dev_private;
1562
1563         switch (mode) {
1564         case BONDING_MODE_ROUND_ROBIN:
1565                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_round_robin;
1566                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1567                 break;
1568         case BONDING_MODE_ACTIVE_BACKUP:
1569                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_active_backup;
1570                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup;
1571                 break;
1572         case BONDING_MODE_BALANCE:
1573                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_balance;
1574                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1575                 break;
1576         case BONDING_MODE_BROADCAST:
1577                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_broadcast;
1578                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1579                 break;
1580         case BONDING_MODE_8023AD:
1581                 if (bond_mode_8023ad_enable(eth_dev) != 0)
1582                         return -1;
1583
1584                 if (internals->mode4.dedicated_queues.enabled == 0) {
1585                         eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_8023ad;
1586                         eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_8023ad;
1587                         RTE_BOND_LOG(WARNING,
1588                                 "Using mode 4, it is necessary to do TX burst "
1589                                 "and RX burst at least every 100ms.");
1590                 } else {
1591                         /* Use flow director's optimization */
1592                         eth_dev->rx_pkt_burst =
1593                                         bond_ethdev_rx_burst_8023ad_fast_queue;
1594                         eth_dev->tx_pkt_burst =
1595                                         bond_ethdev_tx_burst_8023ad_fast_queue;
1596                 }
1597                 break;
1598         case BONDING_MODE_TLB:
1599                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_tlb;
1600                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup;
1601                 break;
1602         case BONDING_MODE_ALB:
1603                 if (bond_mode_alb_enable(eth_dev) != 0)
1604                         return -1;
1605
1606                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_alb;
1607                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_alb;
1608                 break;
1609         default:
1610                 return -1;
1611         }
1612
1613         internals->mode = mode;
1614
1615         return 0;
1616 }
1617
1618
1619 static int
1620 slave_configure_slow_queue(struct rte_eth_dev *bonded_eth_dev,
1621                 struct rte_eth_dev *slave_eth_dev)
1622 {
1623         int errval = 0;
1624         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1625         struct port *port = &bond_mode_8023ad_ports[slave_eth_dev->data->port_id];
1626
1627         if (port->slow_pool == NULL) {
1628                 char mem_name[256];
1629                 int slave_id = slave_eth_dev->data->port_id;
1630
1631                 snprintf(mem_name, RTE_DIM(mem_name), "slave_port%u_slow_pool",
1632                                 slave_id);
1633                 port->slow_pool = rte_pktmbuf_pool_create(mem_name, 8191,
1634                         250, 0, RTE_MBUF_DEFAULT_BUF_SIZE,
1635                         slave_eth_dev->data->numa_node);
1636
1637                 /* Any memory allocation failure in initialization is critical because
1638                  * resources can't be free, so reinitialization is impossible. */
1639                 if (port->slow_pool == NULL) {
1640                         rte_panic("Slave %u: Failed to create memory pool '%s': %s\n",
1641                                 slave_id, mem_name, rte_strerror(rte_errno));
1642                 }
1643         }
1644
1645         if (internals->mode4.dedicated_queues.enabled == 1) {
1646                 /* Configure slow Rx queue */
1647
1648                 errval = rte_eth_rx_queue_setup(slave_eth_dev->data->port_id,
1649                                 internals->mode4.dedicated_queues.rx_qid, 128,
1650                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1651                                 NULL, port->slow_pool);
1652                 if (errval != 0) {
1653                         RTE_BOND_LOG(ERR,
1654                                         "rte_eth_rx_queue_setup: port=%d queue_id %d, err (%d)",
1655                                         slave_eth_dev->data->port_id,
1656                                         internals->mode4.dedicated_queues.rx_qid,
1657                                         errval);
1658                         return errval;
1659                 }
1660
1661                 errval = rte_eth_tx_queue_setup(slave_eth_dev->data->port_id,
1662                                 internals->mode4.dedicated_queues.tx_qid, 512,
1663                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1664                                 NULL);
1665                 if (errval != 0) {
1666                         RTE_BOND_LOG(ERR,
1667                                 "rte_eth_tx_queue_setup: port=%d queue_id %d, err (%d)",
1668                                 slave_eth_dev->data->port_id,
1669                                 internals->mode4.dedicated_queues.tx_qid,
1670                                 errval);
1671                         return errval;
1672                 }
1673         }
1674         return 0;
1675 }
1676
1677 int
1678 slave_configure(struct rte_eth_dev *bonded_eth_dev,
1679                 struct rte_eth_dev *slave_eth_dev)
1680 {
1681         struct bond_rx_queue *bd_rx_q;
1682         struct bond_tx_queue *bd_tx_q;
1683         uint16_t nb_rx_queues;
1684         uint16_t nb_tx_queues;
1685
1686         int errval;
1687         uint16_t q_id;
1688         struct rte_flow_error flow_error;
1689
1690         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1691
1692         /* Stop slave */
1693         errval = rte_eth_dev_stop(slave_eth_dev->data->port_id);
1694         if (errval != 0)
1695                 RTE_BOND_LOG(ERR, "rte_eth_dev_stop: port %u, err (%d)",
1696                              slave_eth_dev->data->port_id, errval);
1697
1698         /* Enable interrupts on slave device if supported */
1699         if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1700                 slave_eth_dev->data->dev_conf.intr_conf.lsc = 1;
1701
1702         /* If RSS is enabled for bonding, try to enable it for slaves  */
1703         if (bonded_eth_dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG) {
1704                 /* rss_key won't be empty if RSS is configured in bonded dev */
1705                 slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len =
1706                                         internals->rss_key_len;
1707                 slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key =
1708                                         internals->rss_key;
1709
1710                 slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf =
1711                                 bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
1712                 slave_eth_dev->data->dev_conf.rxmode.mq_mode =
1713                                 bonded_eth_dev->data->dev_conf.rxmode.mq_mode;
1714         }
1715
1716         if (bonded_eth_dev->data->dev_conf.rxmode.offloads &
1717                         RTE_ETH_RX_OFFLOAD_VLAN_FILTER)
1718                 slave_eth_dev->data->dev_conf.rxmode.offloads |=
1719                                 RTE_ETH_RX_OFFLOAD_VLAN_FILTER;
1720         else
1721                 slave_eth_dev->data->dev_conf.rxmode.offloads &=
1722                                 ~RTE_ETH_RX_OFFLOAD_VLAN_FILTER;
1723
1724         slave_eth_dev->data->dev_conf.rxmode.mtu =
1725                         bonded_eth_dev->data->dev_conf.rxmode.mtu;
1726
1727         nb_rx_queues = bonded_eth_dev->data->nb_rx_queues;
1728         nb_tx_queues = bonded_eth_dev->data->nb_tx_queues;
1729
1730         if (internals->mode == BONDING_MODE_8023AD) {
1731                 if (internals->mode4.dedicated_queues.enabled == 1) {
1732                         nb_rx_queues++;
1733                         nb_tx_queues++;
1734                 }
1735         }
1736
1737         errval = rte_eth_dev_set_mtu(slave_eth_dev->data->port_id,
1738                                      bonded_eth_dev->data->mtu);
1739         if (errval != 0 && errval != -ENOTSUP) {
1740                 RTE_BOND_LOG(ERR, "rte_eth_dev_set_mtu: port %u, err (%d)",
1741                                 slave_eth_dev->data->port_id, errval);
1742                 return errval;
1743         }
1744
1745         /* Configure device */
1746         errval = rte_eth_dev_configure(slave_eth_dev->data->port_id,
1747                         nb_rx_queues, nb_tx_queues,
1748                         &(slave_eth_dev->data->dev_conf));
1749         if (errval != 0) {
1750                 RTE_BOND_LOG(ERR, "Cannot configure slave device: port %u, err (%d)",
1751                                 slave_eth_dev->data->port_id, errval);
1752                 return errval;
1753         }
1754
1755         /* Setup Rx Queues */
1756         for (q_id = 0; q_id < bonded_eth_dev->data->nb_rx_queues; q_id++) {
1757                 bd_rx_q = (struct bond_rx_queue *)bonded_eth_dev->data->rx_queues[q_id];
1758
1759                 errval = rte_eth_rx_queue_setup(slave_eth_dev->data->port_id, q_id,
1760                                 bd_rx_q->nb_rx_desc,
1761                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1762                                 &(bd_rx_q->rx_conf), bd_rx_q->mb_pool);
1763                 if (errval != 0) {
1764                         RTE_BOND_LOG(ERR,
1765                                         "rte_eth_rx_queue_setup: port=%d queue_id %d, err (%d)",
1766                                         slave_eth_dev->data->port_id, q_id, errval);
1767                         return errval;
1768                 }
1769         }
1770
1771         /* Setup Tx Queues */
1772         for (q_id = 0; q_id < bonded_eth_dev->data->nb_tx_queues; q_id++) {
1773                 bd_tx_q = (struct bond_tx_queue *)bonded_eth_dev->data->tx_queues[q_id];
1774
1775                 errval = rte_eth_tx_queue_setup(slave_eth_dev->data->port_id, q_id,
1776                                 bd_tx_q->nb_tx_desc,
1777                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1778                                 &bd_tx_q->tx_conf);
1779                 if (errval != 0) {
1780                         RTE_BOND_LOG(ERR,
1781                                 "rte_eth_tx_queue_setup: port=%d queue_id %d, err (%d)",
1782                                 slave_eth_dev->data->port_id, q_id, errval);
1783                         return errval;
1784                 }
1785         }
1786
1787         if (internals->mode == BONDING_MODE_8023AD &&
1788                         internals->mode4.dedicated_queues.enabled == 1) {
1789                 if (slave_configure_slow_queue(bonded_eth_dev, slave_eth_dev)
1790                                 != 0)
1791                         return errval;
1792
1793                 errval = bond_ethdev_8023ad_flow_verify(bonded_eth_dev,
1794                                 slave_eth_dev->data->port_id);
1795                 if (errval != 0) {
1796                         RTE_BOND_LOG(ERR,
1797                                 "bond_ethdev_8023ad_flow_verify: port=%d, err (%d)",
1798                                 slave_eth_dev->data->port_id, errval);
1799                         return errval;
1800                 }
1801
1802                 if (internals->mode4.dedicated_queues.flow[slave_eth_dev->data->port_id] != NULL)
1803                         rte_flow_destroy(slave_eth_dev->data->port_id,
1804                                         internals->mode4.dedicated_queues.flow[slave_eth_dev->data->port_id],
1805                                         &flow_error);
1806
1807                 errval = bond_ethdev_8023ad_flow_set(bonded_eth_dev,
1808                                 slave_eth_dev->data->port_id);
1809                 if (errval != 0) {
1810                         RTE_BOND_LOG(ERR,
1811                                 "bond_ethdev_8023ad_flow_set: port=%d, err (%d)",
1812                                 slave_eth_dev->data->port_id, errval);
1813                         return errval;
1814                 }
1815         }
1816
1817         /* Start device */
1818         errval = rte_eth_dev_start(slave_eth_dev->data->port_id);
1819         if (errval != 0) {
1820                 RTE_BOND_LOG(ERR, "rte_eth_dev_start: port=%u, err (%d)",
1821                                 slave_eth_dev->data->port_id, errval);
1822                 return -1;
1823         }
1824
1825         /* If RSS is enabled for bonding, synchronize RETA */
1826         if (bonded_eth_dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS) {
1827                 int i;
1828                 struct bond_dev_private *internals;
1829
1830                 internals = bonded_eth_dev->data->dev_private;
1831
1832                 for (i = 0; i < internals->slave_count; i++) {
1833                         if (internals->slaves[i].port_id == slave_eth_dev->data->port_id) {
1834                                 errval = rte_eth_dev_rss_reta_update(
1835                                                 slave_eth_dev->data->port_id,
1836                                                 &internals->reta_conf[0],
1837                                                 internals->slaves[i].reta_size);
1838                                 if (errval != 0) {
1839                                         RTE_BOND_LOG(WARNING,
1840                                                      "rte_eth_dev_rss_reta_update on slave port %d fails (err %d)."
1841                                                      " RSS Configuration for bonding may be inconsistent.",
1842                                                      slave_eth_dev->data->port_id, errval);
1843                                 }
1844                                 break;
1845                         }
1846                 }
1847         }
1848
1849         /* If lsc interrupt is set, check initial slave's link status */
1850         if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
1851                 slave_eth_dev->dev_ops->link_update(slave_eth_dev, 0);
1852                 bond_ethdev_lsc_event_callback(slave_eth_dev->data->port_id,
1853                         RTE_ETH_EVENT_INTR_LSC, &bonded_eth_dev->data->port_id,
1854                         NULL);
1855         }
1856
1857         return 0;
1858 }
1859
1860 void
1861 slave_remove(struct bond_dev_private *internals,
1862                 struct rte_eth_dev *slave_eth_dev)
1863 {
1864         uint16_t i;
1865
1866         for (i = 0; i < internals->slave_count; i++)
1867                 if (internals->slaves[i].port_id ==
1868                                 slave_eth_dev->data->port_id)
1869                         break;
1870
1871         if (i < (internals->slave_count - 1)) {
1872                 struct rte_flow *flow;
1873
1874                 memmove(&internals->slaves[i], &internals->slaves[i + 1],
1875                                 sizeof(internals->slaves[0]) *
1876                                 (internals->slave_count - i - 1));
1877                 TAILQ_FOREACH(flow, &internals->flow_list, next) {
1878                         memmove(&flow->flows[i], &flow->flows[i + 1],
1879                                 sizeof(flow->flows[0]) *
1880                                 (internals->slave_count - i - 1));
1881                         flow->flows[internals->slave_count - 1] = NULL;
1882                 }
1883         }
1884
1885         internals->slave_count--;
1886
1887         /* force reconfiguration of slave interfaces */
1888         rte_eth_dev_internal_reset(slave_eth_dev);
1889 }
1890
1891 static void
1892 bond_ethdev_slave_link_status_change_monitor(void *cb_arg);
1893
1894 void
1895 slave_add(struct bond_dev_private *internals,
1896                 struct rte_eth_dev *slave_eth_dev)
1897 {
1898         struct bond_slave_details *slave_details =
1899                         &internals->slaves[internals->slave_count];
1900
1901         slave_details->port_id = slave_eth_dev->data->port_id;
1902         slave_details->last_link_status = 0;
1903
1904         /* Mark slave devices that don't support interrupts so we can
1905          * compensate when we start the bond
1906          */
1907         if (!(slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1908                 slave_details->link_status_poll_enabled = 1;
1909         }
1910
1911         slave_details->link_status_wait_to_complete = 0;
1912         /* clean tlb_last_obytes when adding port for bonding device */
1913         memcpy(&(slave_details->persisted_mac_addr), slave_eth_dev->data->mac_addrs,
1914                         sizeof(struct rte_ether_addr));
1915 }
1916
1917 void
1918 bond_ethdev_primary_set(struct bond_dev_private *internals,
1919                 uint16_t slave_port_id)
1920 {
1921         int i;
1922
1923         if (internals->active_slave_count < 1)
1924                 internals->current_primary_port = slave_port_id;
1925         else
1926                 /* Search bonded device slave ports for new proposed primary port */
1927                 for (i = 0; i < internals->active_slave_count; i++) {
1928                         if (internals->active_slaves[i] == slave_port_id)
1929                                 internals->current_primary_port = slave_port_id;
1930                 }
1931 }
1932
1933 static int
1934 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev);
1935
1936 static int
1937 bond_ethdev_start(struct rte_eth_dev *eth_dev)
1938 {
1939         struct bond_dev_private *internals;
1940         int i;
1941
1942         /* slave eth dev will be started by bonded device */
1943         if (check_for_bonded_ethdev(eth_dev)) {
1944                 RTE_BOND_LOG(ERR, "User tried to explicitly start a slave eth_dev (%d)",
1945                                 eth_dev->data->port_id);
1946                 return -1;
1947         }
1948
1949         eth_dev->data->dev_link.link_status = RTE_ETH_LINK_DOWN;
1950         eth_dev->data->dev_started = 1;
1951
1952         internals = eth_dev->data->dev_private;
1953
1954         if (internals->slave_count == 0) {
1955                 RTE_BOND_LOG(ERR, "Cannot start port since there are no slave devices");
1956                 goto out_err;
1957         }
1958
1959         if (internals->user_defined_mac == 0) {
1960                 struct rte_ether_addr *new_mac_addr = NULL;
1961
1962                 for (i = 0; i < internals->slave_count; i++)
1963                         if (internals->slaves[i].port_id == internals->primary_port)
1964                                 new_mac_addr = &internals->slaves[i].persisted_mac_addr;
1965
1966                 if (new_mac_addr == NULL)
1967                         goto out_err;
1968
1969                 if (mac_address_set(eth_dev, new_mac_addr) != 0) {
1970                         RTE_BOND_LOG(ERR, "bonded port (%d) failed to update MAC address",
1971                                         eth_dev->data->port_id);
1972                         goto out_err;
1973                 }
1974         }
1975
1976         if (internals->mode == BONDING_MODE_8023AD) {
1977                 if (internals->mode4.dedicated_queues.enabled == 1) {
1978                         internals->mode4.dedicated_queues.rx_qid =
1979                                         eth_dev->data->nb_rx_queues;
1980                         internals->mode4.dedicated_queues.tx_qid =
1981                                         eth_dev->data->nb_tx_queues;
1982                 }
1983         }
1984
1985
1986         /* Reconfigure each slave device if starting bonded device */
1987         for (i = 0; i < internals->slave_count; i++) {
1988                 struct rte_eth_dev *slave_ethdev =
1989                                 &(rte_eth_devices[internals->slaves[i].port_id]);
1990                 if (slave_configure(eth_dev, slave_ethdev) != 0) {
1991                         RTE_BOND_LOG(ERR,
1992                                 "bonded port (%d) failed to reconfigure slave device (%d)",
1993                                 eth_dev->data->port_id,
1994                                 internals->slaves[i].port_id);
1995                         goto out_err;
1996                 }
1997                 /* We will need to poll for link status if any slave doesn't
1998                  * support interrupts
1999                  */
2000                 if (internals->slaves[i].link_status_poll_enabled)
2001                         internals->link_status_polling_enabled = 1;
2002         }
2003
2004         /* start polling if needed */
2005         if (internals->link_status_polling_enabled) {
2006                 rte_eal_alarm_set(
2007                         internals->link_status_polling_interval_ms * 1000,
2008                         bond_ethdev_slave_link_status_change_monitor,
2009                         (void *)&rte_eth_devices[internals->port_id]);
2010         }
2011
2012         /* Update all slave devices MACs*/
2013         if (mac_address_slaves_update(eth_dev) != 0)
2014                 goto out_err;
2015
2016         if (internals->user_defined_primary_port)
2017                 bond_ethdev_primary_set(internals, internals->primary_port);
2018
2019         if (internals->mode == BONDING_MODE_8023AD)
2020                 bond_mode_8023ad_start(eth_dev);
2021
2022         if (internals->mode == BONDING_MODE_TLB ||
2023                         internals->mode == BONDING_MODE_ALB)
2024                 bond_tlb_enable(internals);
2025
2026         return 0;
2027
2028 out_err:
2029         eth_dev->data->dev_started = 0;
2030         return -1;
2031 }
2032
2033 static void
2034 bond_ethdev_free_queues(struct rte_eth_dev *dev)
2035 {
2036         uint16_t i;
2037
2038         if (dev->data->rx_queues != NULL) {
2039                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2040                         rte_free(dev->data->rx_queues[i]);
2041                         dev->data->rx_queues[i] = NULL;
2042                 }
2043                 dev->data->nb_rx_queues = 0;
2044         }
2045
2046         if (dev->data->tx_queues != NULL) {
2047                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2048                         rte_free(dev->data->tx_queues[i]);
2049                         dev->data->tx_queues[i] = NULL;
2050                 }
2051                 dev->data->nb_tx_queues = 0;
2052         }
2053 }
2054
2055 int
2056 bond_ethdev_stop(struct rte_eth_dev *eth_dev)
2057 {
2058         struct bond_dev_private *internals = eth_dev->data->dev_private;
2059         uint16_t i;
2060         int ret;
2061
2062         if (internals->mode == BONDING_MODE_8023AD) {
2063                 struct port *port;
2064                 void *pkt = NULL;
2065
2066                 bond_mode_8023ad_stop(eth_dev);
2067
2068                 /* Discard all messages to/from mode 4 state machines */
2069                 for (i = 0; i < internals->active_slave_count; i++) {
2070                         port = &bond_mode_8023ad_ports[internals->active_slaves[i]];
2071
2072                         RTE_ASSERT(port->rx_ring != NULL);
2073                         while (rte_ring_dequeue(port->rx_ring, &pkt) != -ENOENT)
2074                                 rte_pktmbuf_free(pkt);
2075
2076                         RTE_ASSERT(port->tx_ring != NULL);
2077                         while (rte_ring_dequeue(port->tx_ring, &pkt) != -ENOENT)
2078                                 rte_pktmbuf_free(pkt);
2079                 }
2080         }
2081
2082         if (internals->mode == BONDING_MODE_TLB ||
2083                         internals->mode == BONDING_MODE_ALB) {
2084                 bond_tlb_disable(internals);
2085                 for (i = 0; i < internals->active_slave_count; i++)
2086                         tlb_last_obytets[internals->active_slaves[i]] = 0;
2087         }
2088
2089         eth_dev->data->dev_link.link_status = RTE_ETH_LINK_DOWN;
2090         eth_dev->data->dev_started = 0;
2091
2092         internals->link_status_polling_enabled = 0;
2093         for (i = 0; i < internals->slave_count; i++) {
2094                 uint16_t slave_id = internals->slaves[i].port_id;
2095                 if (find_slave_by_id(internals->active_slaves,
2096                                 internals->active_slave_count, slave_id) !=
2097                                                 internals->active_slave_count) {
2098                         internals->slaves[i].last_link_status = 0;
2099                         ret = rte_eth_dev_stop(slave_id);
2100                         if (ret != 0) {
2101                                 RTE_BOND_LOG(ERR, "Failed to stop device on port %u",
2102                                              slave_id);
2103                                 return ret;
2104                         }
2105                         deactivate_slave(eth_dev, slave_id);
2106                 }
2107         }
2108
2109         return 0;
2110 }
2111
2112 int
2113 bond_ethdev_close(struct rte_eth_dev *dev)
2114 {
2115         struct bond_dev_private *internals = dev->data->dev_private;
2116         uint16_t bond_port_id = internals->port_id;
2117         int skipped = 0;
2118         struct rte_flow_error ferror;
2119
2120         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2121                 return 0;
2122
2123         RTE_BOND_LOG(INFO, "Closing bonded device %s", dev->device->name);
2124         while (internals->slave_count != skipped) {
2125                 uint16_t port_id = internals->slaves[skipped].port_id;
2126
2127                 if (rte_eth_dev_stop(port_id) != 0) {
2128                         RTE_BOND_LOG(ERR, "Failed to stop device on port %u",
2129                                      port_id);
2130                         skipped++;
2131                 }
2132
2133                 if (rte_eth_bond_slave_remove(bond_port_id, port_id) != 0) {
2134                         RTE_BOND_LOG(ERR,
2135                                      "Failed to remove port %d from bonded device %s",
2136                                      port_id, dev->device->name);
2137                         skipped++;
2138                 }
2139         }
2140         bond_flow_ops.flush(dev, &ferror);
2141         bond_ethdev_free_queues(dev);
2142         rte_bitmap_reset(internals->vlan_filter_bmp);
2143         rte_bitmap_free(internals->vlan_filter_bmp);
2144         rte_free(internals->vlan_filter_bmpmem);
2145
2146         /* Try to release mempool used in mode6. If the bond
2147          * device is not mode6, free the NULL is not problem.
2148          */
2149         rte_mempool_free(internals->mode6.mempool);
2150
2151         if (internals->kvlist != NULL)
2152                 rte_kvargs_free(internals->kvlist);
2153
2154         return 0;
2155 }
2156
2157 /* forward declaration */
2158 static int bond_ethdev_configure(struct rte_eth_dev *dev);
2159
2160 static int
2161 bond_ethdev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2162 {
2163         struct bond_dev_private *internals = dev->data->dev_private;
2164         struct bond_slave_details slave;
2165         int ret;
2166
2167         uint16_t max_nb_rx_queues = UINT16_MAX;
2168         uint16_t max_nb_tx_queues = UINT16_MAX;
2169         uint16_t max_rx_desc_lim = UINT16_MAX;
2170         uint16_t max_tx_desc_lim = UINT16_MAX;
2171
2172         dev_info->max_mac_addrs = BOND_MAX_MAC_ADDRS;
2173
2174         dev_info->max_rx_pktlen = internals->candidate_max_rx_pktlen ?
2175                         internals->candidate_max_rx_pktlen :
2176                         RTE_ETHER_MAX_JUMBO_FRAME_LEN;
2177
2178         /* Max number of tx/rx queues that the bonded device can support is the
2179          * minimum values of the bonded slaves, as all slaves must be capable
2180          * of supporting the same number of tx/rx queues.
2181          */
2182         if (internals->slave_count > 0) {
2183                 struct rte_eth_dev_info slave_info;
2184                 uint16_t idx;
2185
2186                 for (idx = 0; idx < internals->slave_count; idx++) {
2187                         slave = internals->slaves[idx];
2188                         ret = rte_eth_dev_info_get(slave.port_id, &slave_info);
2189                         if (ret != 0) {
2190                                 RTE_BOND_LOG(ERR,
2191                                         "%s: Error during getting device (port %u) info: %s\n",
2192                                         __func__,
2193                                         slave.port_id,
2194                                         strerror(-ret));
2195
2196                                 return ret;
2197                         }
2198
2199                         if (slave_info.max_rx_queues < max_nb_rx_queues)
2200                                 max_nb_rx_queues = slave_info.max_rx_queues;
2201
2202                         if (slave_info.max_tx_queues < max_nb_tx_queues)
2203                                 max_nb_tx_queues = slave_info.max_tx_queues;
2204
2205                         if (slave_info.rx_desc_lim.nb_max < max_rx_desc_lim)
2206                                 max_rx_desc_lim = slave_info.rx_desc_lim.nb_max;
2207
2208                         if (slave_info.tx_desc_lim.nb_max < max_tx_desc_lim)
2209                                 max_tx_desc_lim = slave_info.tx_desc_lim.nb_max;
2210                 }
2211         }
2212
2213         dev_info->max_rx_queues = max_nb_rx_queues;
2214         dev_info->max_tx_queues = max_nb_tx_queues;
2215
2216         memcpy(&dev_info->default_rxconf, &internals->default_rxconf,
2217                sizeof(dev_info->default_rxconf));
2218         memcpy(&dev_info->default_txconf, &internals->default_txconf,
2219                sizeof(dev_info->default_txconf));
2220
2221         dev_info->rx_desc_lim.nb_max = max_rx_desc_lim;
2222         dev_info->tx_desc_lim.nb_max = max_tx_desc_lim;
2223
2224         /**
2225          * If dedicated hw queues enabled for link bonding device in LACP mode
2226          * then we need to reduce the maximum number of data path queues by 1.
2227          */
2228         if (internals->mode == BONDING_MODE_8023AD &&
2229                 internals->mode4.dedicated_queues.enabled == 1) {
2230                 dev_info->max_rx_queues--;
2231                 dev_info->max_tx_queues--;
2232         }
2233
2234         dev_info->min_rx_bufsize = 0;
2235
2236         dev_info->rx_offload_capa = internals->rx_offload_capa;
2237         dev_info->tx_offload_capa = internals->tx_offload_capa;
2238         dev_info->rx_queue_offload_capa = internals->rx_queue_offload_capa;
2239         dev_info->tx_queue_offload_capa = internals->tx_queue_offload_capa;
2240         dev_info->flow_type_rss_offloads = internals->flow_type_rss_offloads;
2241
2242         dev_info->reta_size = internals->reta_size;
2243         dev_info->hash_key_size = internals->rss_key_len;
2244
2245         return 0;
2246 }
2247
2248 static int
2249 bond_ethdev_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
2250 {
2251         int res;
2252         uint16_t i;
2253         struct bond_dev_private *internals = dev->data->dev_private;
2254
2255         /* don't do this while a slave is being added */
2256         rte_spinlock_lock(&internals->lock);
2257
2258         if (on)
2259                 rte_bitmap_set(internals->vlan_filter_bmp, vlan_id);
2260         else
2261                 rte_bitmap_clear(internals->vlan_filter_bmp, vlan_id);
2262
2263         for (i = 0; i < internals->slave_count; i++) {
2264                 uint16_t port_id = internals->slaves[i].port_id;
2265
2266                 res = rte_eth_dev_vlan_filter(port_id, vlan_id, on);
2267                 if (res == ENOTSUP)
2268                         RTE_BOND_LOG(WARNING,
2269                                      "Setting VLAN filter on slave port %u not supported.",
2270                                      port_id);
2271         }
2272
2273         rte_spinlock_unlock(&internals->lock);
2274         return 0;
2275 }
2276
2277 static int
2278 bond_ethdev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
2279                 uint16_t nb_rx_desc, unsigned int socket_id __rte_unused,
2280                 const struct rte_eth_rxconf *rx_conf, struct rte_mempool *mb_pool)
2281 {
2282         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)
2283                         rte_zmalloc_socket(NULL, sizeof(struct bond_rx_queue),
2284                                         0, dev->data->numa_node);
2285         if (bd_rx_q == NULL)
2286                 return -1;
2287
2288         bd_rx_q->queue_id = rx_queue_id;
2289         bd_rx_q->dev_private = dev->data->dev_private;
2290
2291         bd_rx_q->nb_rx_desc = nb_rx_desc;
2292
2293         memcpy(&(bd_rx_q->rx_conf), rx_conf, sizeof(struct rte_eth_rxconf));
2294         bd_rx_q->mb_pool = mb_pool;
2295
2296         dev->data->rx_queues[rx_queue_id] = bd_rx_q;
2297
2298         return 0;
2299 }
2300
2301 static int
2302 bond_ethdev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
2303                 uint16_t nb_tx_desc, unsigned int socket_id __rte_unused,
2304                 const struct rte_eth_txconf *tx_conf)
2305 {
2306         struct bond_tx_queue *bd_tx_q  = (struct bond_tx_queue *)
2307                         rte_zmalloc_socket(NULL, sizeof(struct bond_tx_queue),
2308                                         0, dev->data->numa_node);
2309
2310         if (bd_tx_q == NULL)
2311                 return -1;
2312
2313         bd_tx_q->queue_id = tx_queue_id;
2314         bd_tx_q->dev_private = dev->data->dev_private;
2315
2316         bd_tx_q->nb_tx_desc = nb_tx_desc;
2317         memcpy(&(bd_tx_q->tx_conf), tx_conf, sizeof(bd_tx_q->tx_conf));
2318
2319         dev->data->tx_queues[tx_queue_id] = bd_tx_q;
2320
2321         return 0;
2322 }
2323
2324 static void
2325 bond_ethdev_rx_queue_release(struct rte_eth_dev *dev, uint16_t queue_id)
2326 {
2327         void *queue = dev->data->rx_queues[queue_id];
2328
2329         if (queue == NULL)
2330                 return;
2331
2332         rte_free(queue);
2333 }
2334
2335 static void
2336 bond_ethdev_tx_queue_release(struct rte_eth_dev *dev, uint16_t queue_id)
2337 {
2338         void *queue = dev->data->tx_queues[queue_id];
2339
2340         if (queue == NULL)
2341                 return;
2342
2343         rte_free(queue);
2344 }
2345
2346 static void
2347 bond_ethdev_slave_link_status_change_monitor(void *cb_arg)
2348 {
2349         struct rte_eth_dev *bonded_ethdev, *slave_ethdev;
2350         struct bond_dev_private *internals;
2351
2352         /* Default value for polling slave found is true as we don't want to
2353          * disable the polling thread if we cannot get the lock */
2354         int i, polling_slave_found = 1;
2355
2356         if (cb_arg == NULL)
2357                 return;
2358
2359         bonded_ethdev = cb_arg;
2360         internals = bonded_ethdev->data->dev_private;
2361
2362         if (!bonded_ethdev->data->dev_started ||
2363                 !internals->link_status_polling_enabled)
2364                 return;
2365
2366         /* If device is currently being configured then don't check slaves link
2367          * status, wait until next period */
2368         if (rte_spinlock_trylock(&internals->lock)) {
2369                 if (internals->slave_count > 0)
2370                         polling_slave_found = 0;
2371
2372                 for (i = 0; i < internals->slave_count; i++) {
2373                         if (!internals->slaves[i].link_status_poll_enabled)
2374                                 continue;
2375
2376                         slave_ethdev = &rte_eth_devices[internals->slaves[i].port_id];
2377                         polling_slave_found = 1;
2378
2379                         /* Update slave link status */
2380                         (*slave_ethdev->dev_ops->link_update)(slave_ethdev,
2381                                         internals->slaves[i].link_status_wait_to_complete);
2382
2383                         /* if link status has changed since last checked then call lsc
2384                          * event callback */
2385                         if (slave_ethdev->data->dev_link.link_status !=
2386                                         internals->slaves[i].last_link_status) {
2387                                 internals->slaves[i].last_link_status =
2388                                                 slave_ethdev->data->dev_link.link_status;
2389
2390                                 bond_ethdev_lsc_event_callback(internals->slaves[i].port_id,
2391                                                 RTE_ETH_EVENT_INTR_LSC,
2392                                                 &bonded_ethdev->data->port_id,
2393                                                 NULL);
2394                         }
2395                 }
2396                 rte_spinlock_unlock(&internals->lock);
2397         }
2398
2399         if (polling_slave_found)
2400                 /* Set alarm to continue monitoring link status of slave ethdev's */
2401                 rte_eal_alarm_set(internals->link_status_polling_interval_ms * 1000,
2402                                 bond_ethdev_slave_link_status_change_monitor, cb_arg);
2403 }
2404
2405 static int
2406 bond_ethdev_link_update(struct rte_eth_dev *ethdev, int wait_to_complete)
2407 {
2408         int (*link_update)(uint16_t port_id, struct rte_eth_link *eth_link);
2409
2410         struct bond_dev_private *bond_ctx;
2411         struct rte_eth_link slave_link;
2412
2413         bool one_link_update_succeeded;
2414         uint32_t idx;
2415         int ret;
2416
2417         bond_ctx = ethdev->data->dev_private;
2418
2419         ethdev->data->dev_link.link_speed = RTE_ETH_SPEED_NUM_NONE;
2420
2421         if (ethdev->data->dev_started == 0 ||
2422                         bond_ctx->active_slave_count == 0) {
2423                 ethdev->data->dev_link.link_status = RTE_ETH_LINK_DOWN;
2424                 return 0;
2425         }
2426
2427         ethdev->data->dev_link.link_status = RTE_ETH_LINK_UP;
2428
2429         if (wait_to_complete)
2430                 link_update = rte_eth_link_get;
2431         else
2432                 link_update = rte_eth_link_get_nowait;
2433
2434         switch (bond_ctx->mode) {
2435         case BONDING_MODE_BROADCAST:
2436                 /**
2437                  * Setting link speed to UINT32_MAX to ensure we pick up the
2438                  * value of the first active slave
2439                  */
2440                 ethdev->data->dev_link.link_speed = UINT32_MAX;
2441
2442                 /**
2443                  * link speed is minimum value of all the slaves link speed as
2444                  * packet loss will occur on this slave if transmission at rates
2445                  * greater than this are attempted
2446                  */
2447                 for (idx = 0; idx < bond_ctx->active_slave_count; idx++) {
2448                         ret = link_update(bond_ctx->active_slaves[idx],
2449                                           &slave_link);
2450                         if (ret < 0) {
2451                                 ethdev->data->dev_link.link_speed =
2452                                         RTE_ETH_SPEED_NUM_NONE;
2453                                 RTE_BOND_LOG(ERR,
2454                                         "Slave (port %u) link get failed: %s",
2455                                         bond_ctx->active_slaves[idx],
2456                                         rte_strerror(-ret));
2457                                 return 0;
2458                         }
2459
2460                         if (slave_link.link_speed <
2461                                         ethdev->data->dev_link.link_speed)
2462                                 ethdev->data->dev_link.link_speed =
2463                                                 slave_link.link_speed;
2464                 }
2465                 break;
2466         case BONDING_MODE_ACTIVE_BACKUP:
2467                 /* Current primary slave */
2468                 ret = link_update(bond_ctx->current_primary_port, &slave_link);
2469                 if (ret < 0) {
2470                         RTE_BOND_LOG(ERR, "Slave (port %u) link get failed: %s",
2471                                 bond_ctx->current_primary_port,
2472                                 rte_strerror(-ret));
2473                         return 0;
2474                 }
2475
2476                 ethdev->data->dev_link.link_speed = slave_link.link_speed;
2477                 break;
2478         case BONDING_MODE_8023AD:
2479                 ethdev->data->dev_link.link_autoneg =
2480                                 bond_ctx->mode4.slave_link.link_autoneg;
2481                 ethdev->data->dev_link.link_duplex =
2482                                 bond_ctx->mode4.slave_link.link_duplex;
2483                 /* fall through */
2484                 /* to update link speed */
2485         case BONDING_MODE_ROUND_ROBIN:
2486         case BONDING_MODE_BALANCE:
2487         case BONDING_MODE_TLB:
2488         case BONDING_MODE_ALB:
2489         default:
2490                 /**
2491                  * In theses mode the maximum theoretical link speed is the sum
2492                  * of all the slaves
2493                  */
2494                 ethdev->data->dev_link.link_speed = RTE_ETH_SPEED_NUM_NONE;
2495                 one_link_update_succeeded = false;
2496
2497                 for (idx = 0; idx < bond_ctx->active_slave_count; idx++) {
2498                         ret = link_update(bond_ctx->active_slaves[idx],
2499                                         &slave_link);
2500                         if (ret < 0) {
2501                                 RTE_BOND_LOG(ERR,
2502                                         "Slave (port %u) link get failed: %s",
2503                                         bond_ctx->active_slaves[idx],
2504                                         rte_strerror(-ret));
2505                                 continue;
2506                         }
2507
2508                         one_link_update_succeeded = true;
2509                         ethdev->data->dev_link.link_speed +=
2510                                         slave_link.link_speed;
2511                 }
2512
2513                 if (!one_link_update_succeeded) {
2514                         RTE_BOND_LOG(ERR, "All slaves link get failed");
2515                         return 0;
2516                 }
2517         }
2518
2519
2520         return 0;
2521 }
2522
2523
2524 static int
2525 bond_ethdev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
2526 {
2527         struct bond_dev_private *internals = dev->data->dev_private;
2528         struct rte_eth_stats slave_stats;
2529         int i, j;
2530
2531         for (i = 0; i < internals->slave_count; i++) {
2532                 rte_eth_stats_get(internals->slaves[i].port_id, &slave_stats);
2533
2534                 stats->ipackets += slave_stats.ipackets;
2535                 stats->opackets += slave_stats.opackets;
2536                 stats->ibytes += slave_stats.ibytes;
2537                 stats->obytes += slave_stats.obytes;
2538                 stats->imissed += slave_stats.imissed;
2539                 stats->ierrors += slave_stats.ierrors;
2540                 stats->oerrors += slave_stats.oerrors;
2541                 stats->rx_nombuf += slave_stats.rx_nombuf;
2542
2543                 for (j = 0; j < RTE_ETHDEV_QUEUE_STAT_CNTRS; j++) {
2544                         stats->q_ipackets[j] += slave_stats.q_ipackets[j];
2545                         stats->q_opackets[j] += slave_stats.q_opackets[j];
2546                         stats->q_ibytes[j] += slave_stats.q_ibytes[j];
2547                         stats->q_obytes[j] += slave_stats.q_obytes[j];
2548                         stats->q_errors[j] += slave_stats.q_errors[j];
2549                 }
2550
2551         }
2552
2553         return 0;
2554 }
2555
2556 static int
2557 bond_ethdev_stats_reset(struct rte_eth_dev *dev)
2558 {
2559         struct bond_dev_private *internals = dev->data->dev_private;
2560         int i;
2561         int err;
2562         int ret;
2563
2564         for (i = 0, err = 0; i < internals->slave_count; i++) {
2565                 ret = rte_eth_stats_reset(internals->slaves[i].port_id);
2566                 if (ret != 0)
2567                         err = ret;
2568         }
2569
2570         return err;
2571 }
2572
2573 static int
2574 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev)
2575 {
2576         struct bond_dev_private *internals = eth_dev->data->dev_private;
2577         int i;
2578         int ret = 0;
2579         uint16_t port_id;
2580
2581         switch (internals->mode) {
2582         /* Promiscuous mode is propagated to all slaves */
2583         case BONDING_MODE_ROUND_ROBIN:
2584         case BONDING_MODE_BALANCE:
2585         case BONDING_MODE_BROADCAST:
2586         case BONDING_MODE_8023AD: {
2587                 unsigned int slave_ok = 0;
2588
2589                 for (i = 0; i < internals->slave_count; i++) {
2590                         port_id = internals->slaves[i].port_id;
2591
2592                         ret = rte_eth_promiscuous_enable(port_id);
2593                         if (ret != 0)
2594                                 RTE_BOND_LOG(ERR,
2595                                         "Failed to enable promiscuous mode for port %u: %s",
2596                                         port_id, rte_strerror(-ret));
2597                         else
2598                                 slave_ok++;
2599                 }
2600                 /*
2601                  * Report success if operation is successful on at least
2602                  * on one slave. Otherwise return last error code.
2603                  */
2604                 if (slave_ok > 0)
2605                         ret = 0;
2606                 break;
2607         }
2608         /* Promiscuous mode is propagated only to primary slave */
2609         case BONDING_MODE_ACTIVE_BACKUP:
2610         case BONDING_MODE_TLB:
2611         case BONDING_MODE_ALB:
2612         default:
2613                 /* Do not touch promisc when there cannot be primary ports */
2614                 if (internals->slave_count == 0)
2615                         break;
2616                 port_id = internals->current_primary_port;
2617                 ret = rte_eth_promiscuous_enable(port_id);
2618                 if (ret != 0)
2619                         RTE_BOND_LOG(ERR,
2620                                 "Failed to enable promiscuous mode for port %u: %s",
2621                                 port_id, rte_strerror(-ret));
2622         }
2623
2624         return ret;
2625 }
2626
2627 static int
2628 bond_ethdev_promiscuous_disable(struct rte_eth_dev *dev)
2629 {
2630         struct bond_dev_private *internals = dev->data->dev_private;
2631         int i;
2632         int ret = 0;
2633         uint16_t port_id;
2634
2635         switch (internals->mode) {
2636         /* Promiscuous mode is propagated to all slaves */
2637         case BONDING_MODE_ROUND_ROBIN:
2638         case BONDING_MODE_BALANCE:
2639         case BONDING_MODE_BROADCAST:
2640         case BONDING_MODE_8023AD: {
2641                 unsigned int slave_ok = 0;
2642
2643                 for (i = 0; i < internals->slave_count; i++) {
2644                         port_id = internals->slaves[i].port_id;
2645
2646                         if (internals->mode == BONDING_MODE_8023AD &&
2647                             bond_mode_8023ad_ports[port_id].forced_rx_flags ==
2648                                         BOND_8023AD_FORCED_PROMISC) {
2649                                 slave_ok++;
2650                                 continue;
2651                         }
2652                         ret = rte_eth_promiscuous_disable(port_id);
2653                         if (ret != 0)
2654                                 RTE_BOND_LOG(ERR,
2655                                         "Failed to disable promiscuous mode for port %u: %s",
2656                                         port_id, rte_strerror(-ret));
2657                         else
2658                                 slave_ok++;
2659                 }
2660                 /*
2661                  * Report success if operation is successful on at least
2662                  * on one slave. Otherwise return last error code.
2663                  */
2664                 if (slave_ok > 0)
2665                         ret = 0;
2666                 break;
2667         }
2668         /* Promiscuous mode is propagated only to primary slave */
2669         case BONDING_MODE_ACTIVE_BACKUP:
2670         case BONDING_MODE_TLB:
2671         case BONDING_MODE_ALB:
2672         default:
2673                 /* Do not touch promisc when there cannot be primary ports */
2674                 if (internals->slave_count == 0)
2675                         break;
2676                 port_id = internals->current_primary_port;
2677                 ret = rte_eth_promiscuous_disable(port_id);
2678                 if (ret != 0)
2679                         RTE_BOND_LOG(ERR,
2680                                 "Failed to disable promiscuous mode for port %u: %s",
2681                                 port_id, rte_strerror(-ret));
2682         }
2683
2684         return ret;
2685 }
2686
2687 static int
2688 bond_ethdev_allmulticast_enable(struct rte_eth_dev *eth_dev)
2689 {
2690         struct bond_dev_private *internals = eth_dev->data->dev_private;
2691         int i;
2692         int ret = 0;
2693         uint16_t port_id;
2694
2695         switch (internals->mode) {
2696         /* allmulti mode is propagated to all slaves */
2697         case BONDING_MODE_ROUND_ROBIN:
2698         case BONDING_MODE_BALANCE:
2699         case BONDING_MODE_BROADCAST:
2700         case BONDING_MODE_8023AD: {
2701                 unsigned int slave_ok = 0;
2702
2703                 for (i = 0; i < internals->slave_count; i++) {
2704                         port_id = internals->slaves[i].port_id;
2705
2706                         ret = rte_eth_allmulticast_enable(port_id);
2707                         if (ret != 0)
2708                                 RTE_BOND_LOG(ERR,
2709                                         "Failed to enable allmulti mode for port %u: %s",
2710                                         port_id, rte_strerror(-ret));
2711                         else
2712                                 slave_ok++;
2713                 }
2714                 /*
2715                  * Report success if operation is successful on at least
2716                  * on one slave. Otherwise return last error code.
2717                  */
2718                 if (slave_ok > 0)
2719                         ret = 0;
2720                 break;
2721         }
2722         /* allmulti mode is propagated only to primary slave */
2723         case BONDING_MODE_ACTIVE_BACKUP:
2724         case BONDING_MODE_TLB:
2725         case BONDING_MODE_ALB:
2726         default:
2727                 /* Do not touch allmulti when there cannot be primary ports */
2728                 if (internals->slave_count == 0)
2729                         break;
2730                 port_id = internals->current_primary_port;
2731                 ret = rte_eth_allmulticast_enable(port_id);
2732                 if (ret != 0)
2733                         RTE_BOND_LOG(ERR,
2734                                 "Failed to enable allmulti mode for port %u: %s",
2735                                 port_id, rte_strerror(-ret));
2736         }
2737
2738         return ret;
2739 }
2740
2741 static int
2742 bond_ethdev_allmulticast_disable(struct rte_eth_dev *eth_dev)
2743 {
2744         struct bond_dev_private *internals = eth_dev->data->dev_private;
2745         int i;
2746         int ret = 0;
2747         uint16_t port_id;
2748
2749         switch (internals->mode) {
2750         /* allmulti mode is propagated to all slaves */
2751         case BONDING_MODE_ROUND_ROBIN:
2752         case BONDING_MODE_BALANCE:
2753         case BONDING_MODE_BROADCAST:
2754         case BONDING_MODE_8023AD: {
2755                 unsigned int slave_ok = 0;
2756
2757                 for (i = 0; i < internals->slave_count; i++) {
2758                         uint16_t port_id = internals->slaves[i].port_id;
2759
2760                         if (internals->mode == BONDING_MODE_8023AD &&
2761                             bond_mode_8023ad_ports[port_id].forced_rx_flags ==
2762                                         BOND_8023AD_FORCED_ALLMULTI)
2763                                 continue;
2764
2765                         ret = rte_eth_allmulticast_disable(port_id);
2766                         if (ret != 0)
2767                                 RTE_BOND_LOG(ERR,
2768                                         "Failed to disable allmulti mode for port %u: %s",
2769                                         port_id, rte_strerror(-ret));
2770                         else
2771                                 slave_ok++;
2772                 }
2773                 /*
2774                  * Report success if operation is successful on at least
2775                  * on one slave. Otherwise return last error code.
2776                  */
2777                 if (slave_ok > 0)
2778                         ret = 0;
2779                 break;
2780         }
2781         /* allmulti mode is propagated only to primary slave */
2782         case BONDING_MODE_ACTIVE_BACKUP:
2783         case BONDING_MODE_TLB:
2784         case BONDING_MODE_ALB:
2785         default:
2786                 /* Do not touch allmulti when there cannot be primary ports */
2787                 if (internals->slave_count == 0)
2788                         break;
2789                 port_id = internals->current_primary_port;
2790                 ret = rte_eth_allmulticast_disable(port_id);
2791                 if (ret != 0)
2792                         RTE_BOND_LOG(ERR,
2793                                 "Failed to disable allmulti mode for port %u: %s",
2794                                 port_id, rte_strerror(-ret));
2795         }
2796
2797         return ret;
2798 }
2799
2800 static void
2801 bond_ethdev_delayed_lsc_propagation(void *arg)
2802 {
2803         if (arg == NULL)
2804                 return;
2805
2806         rte_eth_dev_callback_process((struct rte_eth_dev *)arg,
2807                         RTE_ETH_EVENT_INTR_LSC, NULL);
2808 }
2809
2810 int
2811 bond_ethdev_lsc_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2812                 void *param, void *ret_param __rte_unused)
2813 {
2814         struct rte_eth_dev *bonded_eth_dev;
2815         struct bond_dev_private *internals;
2816         struct rte_eth_link link;
2817         int rc = -1;
2818         int ret;
2819
2820         uint8_t lsc_flag = 0;
2821         int valid_slave = 0;
2822         uint16_t active_pos;
2823         uint16_t i;
2824
2825         if (type != RTE_ETH_EVENT_INTR_LSC || param == NULL)
2826                 return rc;
2827
2828         bonded_eth_dev = &rte_eth_devices[*(uint16_t *)param];
2829
2830         if (check_for_bonded_ethdev(bonded_eth_dev))
2831                 return rc;
2832
2833         internals = bonded_eth_dev->data->dev_private;
2834
2835         /* If the device isn't started don't handle interrupts */
2836         if (!bonded_eth_dev->data->dev_started)
2837                 return rc;
2838
2839         /* verify that port_id is a valid slave of bonded port */
2840         for (i = 0; i < internals->slave_count; i++) {
2841                 if (internals->slaves[i].port_id == port_id) {
2842                         valid_slave = 1;
2843                         break;
2844                 }
2845         }
2846
2847         if (!valid_slave)
2848                 return rc;
2849
2850         /* Synchronize lsc callback parallel calls either by real link event
2851          * from the slaves PMDs or by the bonding PMD itself.
2852          */
2853         rte_spinlock_lock(&internals->lsc_lock);
2854
2855         /* Search for port in active port list */
2856         active_pos = find_slave_by_id(internals->active_slaves,
2857                         internals->active_slave_count, port_id);
2858
2859         ret = rte_eth_link_get_nowait(port_id, &link);
2860         if (ret < 0)
2861                 RTE_BOND_LOG(ERR, "Slave (port %u) link get failed", port_id);
2862
2863         if (ret == 0 && link.link_status) {
2864                 if (active_pos < internals->active_slave_count)
2865                         goto link_update;
2866
2867                 /* check link state properties if bonded link is up*/
2868                 if (bonded_eth_dev->data->dev_link.link_status == RTE_ETH_LINK_UP) {
2869                         if (link_properties_valid(bonded_eth_dev, &link) != 0)
2870                                 RTE_BOND_LOG(ERR, "Invalid link properties "
2871                                              "for slave %d in bonding mode %d",
2872                                              port_id, internals->mode);
2873                 } else {
2874                         /* inherit slave link properties */
2875                         link_properties_set(bonded_eth_dev, &link);
2876                 }
2877
2878                 /* If no active slave ports then set this port to be
2879                  * the primary port.
2880                  */
2881                 if (internals->active_slave_count < 1) {
2882                         /* If first active slave, then change link status */
2883                         bonded_eth_dev->data->dev_link.link_status =
2884                                                                 RTE_ETH_LINK_UP;
2885                         internals->current_primary_port = port_id;
2886                         lsc_flag = 1;
2887
2888                         mac_address_slaves_update(bonded_eth_dev);
2889                 }
2890
2891                 activate_slave(bonded_eth_dev, port_id);
2892
2893                 /* If the user has defined the primary port then default to
2894                  * using it.
2895                  */
2896                 if (internals->user_defined_primary_port &&
2897                                 internals->primary_port == port_id)
2898                         bond_ethdev_primary_set(internals, port_id);
2899         } else {
2900                 if (active_pos == internals->active_slave_count)
2901                         goto link_update;
2902
2903                 /* Remove from active slave list */
2904                 deactivate_slave(bonded_eth_dev, port_id);
2905
2906                 if (internals->active_slave_count < 1)
2907                         lsc_flag = 1;
2908
2909                 /* Update primary id, take first active slave from list or if none
2910                  * available set to -1 */
2911                 if (port_id == internals->current_primary_port) {
2912                         if (internals->active_slave_count > 0)
2913                                 bond_ethdev_primary_set(internals,
2914                                                 internals->active_slaves[0]);
2915                         else
2916                                 internals->current_primary_port = internals->primary_port;
2917                         mac_address_slaves_update(bonded_eth_dev);
2918                 }
2919         }
2920
2921 link_update:
2922         /**
2923          * Update bonded device link properties after any change to active
2924          * slaves
2925          */
2926         bond_ethdev_link_update(bonded_eth_dev, 0);
2927
2928         if (lsc_flag) {
2929                 /* Cancel any possible outstanding interrupts if delays are enabled */
2930                 if (internals->link_up_delay_ms > 0 ||
2931                         internals->link_down_delay_ms > 0)
2932                         rte_eal_alarm_cancel(bond_ethdev_delayed_lsc_propagation,
2933                                         bonded_eth_dev);
2934
2935                 if (bonded_eth_dev->data->dev_link.link_status) {
2936                         if (internals->link_up_delay_ms > 0)
2937                                 rte_eal_alarm_set(internals->link_up_delay_ms * 1000,
2938                                                 bond_ethdev_delayed_lsc_propagation,
2939                                                 (void *)bonded_eth_dev);
2940                         else
2941                                 rte_eth_dev_callback_process(bonded_eth_dev,
2942                                                 RTE_ETH_EVENT_INTR_LSC,
2943                                                 NULL);
2944
2945                 } else {
2946                         if (internals->link_down_delay_ms > 0)
2947                                 rte_eal_alarm_set(internals->link_down_delay_ms * 1000,
2948                                                 bond_ethdev_delayed_lsc_propagation,
2949                                                 (void *)bonded_eth_dev);
2950                         else
2951                                 rte_eth_dev_callback_process(bonded_eth_dev,
2952                                                 RTE_ETH_EVENT_INTR_LSC,
2953                                                 NULL);
2954                 }
2955         }
2956
2957         rte_spinlock_unlock(&internals->lsc_lock);
2958
2959         return rc;
2960 }
2961
2962 static int
2963 bond_ethdev_rss_reta_update(struct rte_eth_dev *dev,
2964                 struct rte_eth_rss_reta_entry64 *reta_conf, uint16_t reta_size)
2965 {
2966         unsigned i, j;
2967         int result = 0;
2968         int slave_reta_size;
2969         unsigned reta_count;
2970         struct bond_dev_private *internals = dev->data->dev_private;
2971
2972         if (reta_size != internals->reta_size)
2973                 return -EINVAL;
2974
2975          /* Copy RETA table */
2976         reta_count = (reta_size + RTE_ETH_RETA_GROUP_SIZE - 1) /
2977                         RTE_ETH_RETA_GROUP_SIZE;
2978
2979         for (i = 0; i < reta_count; i++) {
2980                 internals->reta_conf[i].mask = reta_conf[i].mask;
2981                 for (j = 0; j < RTE_ETH_RETA_GROUP_SIZE; j++)
2982                         if ((reta_conf[i].mask >> j) & 0x01)
2983                                 internals->reta_conf[i].reta[j] = reta_conf[i].reta[j];
2984         }
2985
2986         /* Fill rest of array */
2987         for (; i < RTE_DIM(internals->reta_conf); i += reta_count)
2988                 memcpy(&internals->reta_conf[i], &internals->reta_conf[0],
2989                                 sizeof(internals->reta_conf[0]) * reta_count);
2990
2991         /* Propagate RETA over slaves */
2992         for (i = 0; i < internals->slave_count; i++) {
2993                 slave_reta_size = internals->slaves[i].reta_size;
2994                 result = rte_eth_dev_rss_reta_update(internals->slaves[i].port_id,
2995                                 &internals->reta_conf[0], slave_reta_size);
2996                 if (result < 0)
2997                         return result;
2998         }
2999
3000         return 0;
3001 }
3002
3003 static int
3004 bond_ethdev_rss_reta_query(struct rte_eth_dev *dev,
3005                 struct rte_eth_rss_reta_entry64 *reta_conf, uint16_t reta_size)
3006 {
3007         int i, j;
3008         struct bond_dev_private *internals = dev->data->dev_private;
3009
3010         if (reta_size != internals->reta_size)
3011                 return -EINVAL;
3012
3013          /* Copy RETA table */
3014         for (i = 0; i < reta_size / RTE_ETH_RETA_GROUP_SIZE; i++)
3015                 for (j = 0; j < RTE_ETH_RETA_GROUP_SIZE; j++)
3016                         if ((reta_conf[i].mask >> j) & 0x01)
3017                                 reta_conf[i].reta[j] = internals->reta_conf[i].reta[j];
3018
3019         return 0;
3020 }
3021
3022 static int
3023 bond_ethdev_rss_hash_update(struct rte_eth_dev *dev,
3024                 struct rte_eth_rss_conf *rss_conf)
3025 {
3026         int i, result = 0;
3027         struct bond_dev_private *internals = dev->data->dev_private;
3028         struct rte_eth_rss_conf bond_rss_conf;
3029
3030         memcpy(&bond_rss_conf, rss_conf, sizeof(struct rte_eth_rss_conf));
3031
3032         bond_rss_conf.rss_hf &= internals->flow_type_rss_offloads;
3033
3034         if (bond_rss_conf.rss_hf != 0)
3035                 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf = bond_rss_conf.rss_hf;
3036
3037         if (bond_rss_conf.rss_key) {
3038                 if (bond_rss_conf.rss_key_len < internals->rss_key_len)
3039                         return -EINVAL;
3040                 else if (bond_rss_conf.rss_key_len > internals->rss_key_len)
3041                         RTE_BOND_LOG(WARNING, "rss_key will be truncated");
3042
3043                 memcpy(internals->rss_key, bond_rss_conf.rss_key,
3044                                 internals->rss_key_len);
3045                 bond_rss_conf.rss_key_len = internals->rss_key_len;
3046         }
3047
3048         for (i = 0; i < internals->slave_count; i++) {
3049                 result = rte_eth_dev_rss_hash_update(internals->slaves[i].port_id,
3050                                 &bond_rss_conf);
3051                 if (result < 0)
3052                         return result;
3053         }
3054
3055         return 0;
3056 }
3057
3058 static int
3059 bond_ethdev_rss_hash_conf_get(struct rte_eth_dev *dev,
3060                 struct rte_eth_rss_conf *rss_conf)
3061 {
3062         struct bond_dev_private *internals = dev->data->dev_private;
3063
3064         rss_conf->rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
3065         rss_conf->rss_key_len = internals->rss_key_len;
3066         if (rss_conf->rss_key)
3067                 memcpy(rss_conf->rss_key, internals->rss_key, internals->rss_key_len);
3068
3069         return 0;
3070 }
3071
3072 static int
3073 bond_ethdev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
3074 {
3075         struct rte_eth_dev *slave_eth_dev;
3076         struct bond_dev_private *internals = dev->data->dev_private;
3077         int ret, i;
3078
3079         rte_spinlock_lock(&internals->lock);
3080
3081         for (i = 0; i < internals->slave_count; i++) {
3082                 slave_eth_dev = &rte_eth_devices[internals->slaves[i].port_id];
3083                 if (*slave_eth_dev->dev_ops->mtu_set == NULL) {
3084                         rte_spinlock_unlock(&internals->lock);
3085                         return -ENOTSUP;
3086                 }
3087         }
3088         for (i = 0; i < internals->slave_count; i++) {
3089                 ret = rte_eth_dev_set_mtu(internals->slaves[i].port_id, mtu);
3090                 if (ret < 0) {
3091                         rte_spinlock_unlock(&internals->lock);
3092                         return ret;
3093                 }
3094         }
3095
3096         rte_spinlock_unlock(&internals->lock);
3097         return 0;
3098 }
3099
3100 static int
3101 bond_ethdev_mac_address_set(struct rte_eth_dev *dev,
3102                         struct rte_ether_addr *addr)
3103 {
3104         if (mac_address_set(dev, addr)) {
3105                 RTE_BOND_LOG(ERR, "Failed to update MAC address");
3106                 return -EINVAL;
3107         }
3108
3109         return 0;
3110 }
3111
3112 static int
3113 bond_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
3114                   const struct rte_flow_ops **ops)
3115 {
3116         *ops = &bond_flow_ops;
3117         return 0;
3118 }
3119
3120 static int
3121 bond_ethdev_mac_addr_add(struct rte_eth_dev *dev,
3122                         struct rte_ether_addr *mac_addr,
3123                         __rte_unused uint32_t index, uint32_t vmdq)
3124 {
3125         struct rte_eth_dev *slave_eth_dev;
3126         struct bond_dev_private *internals = dev->data->dev_private;
3127         int ret, i;
3128
3129         rte_spinlock_lock(&internals->lock);
3130
3131         for (i = 0; i < internals->slave_count; i++) {
3132                 slave_eth_dev = &rte_eth_devices[internals->slaves[i].port_id];
3133                 if (*slave_eth_dev->dev_ops->mac_addr_add == NULL ||
3134                          *slave_eth_dev->dev_ops->mac_addr_remove == NULL) {
3135                         ret = -ENOTSUP;
3136                         goto end;
3137                 }
3138         }
3139
3140         for (i = 0; i < internals->slave_count; i++) {
3141                 ret = rte_eth_dev_mac_addr_add(internals->slaves[i].port_id,
3142                                 mac_addr, vmdq);
3143                 if (ret < 0) {
3144                         /* rollback */
3145                         for (i--; i >= 0; i--)
3146                                 rte_eth_dev_mac_addr_remove(
3147                                         internals->slaves[i].port_id, mac_addr);
3148                         goto end;
3149                 }
3150         }
3151
3152         ret = 0;
3153 end:
3154         rte_spinlock_unlock(&internals->lock);
3155         return ret;
3156 }
3157
3158 static void
3159 bond_ethdev_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
3160 {
3161         struct rte_eth_dev *slave_eth_dev;
3162         struct bond_dev_private *internals = dev->data->dev_private;
3163         int i;
3164
3165         rte_spinlock_lock(&internals->lock);
3166
3167         for (i = 0; i < internals->slave_count; i++) {
3168                 slave_eth_dev = &rte_eth_devices[internals->slaves[i].port_id];
3169                 if (*slave_eth_dev->dev_ops->mac_addr_remove == NULL)
3170                         goto end;
3171         }
3172
3173         struct rte_ether_addr *mac_addr = &dev->data->mac_addrs[index];
3174
3175         for (i = 0; i < internals->slave_count; i++)
3176                 rte_eth_dev_mac_addr_remove(internals->slaves[i].port_id,
3177                                 mac_addr);
3178
3179 end:
3180         rte_spinlock_unlock(&internals->lock);
3181 }
3182
3183 const struct eth_dev_ops default_dev_ops = {
3184         .dev_start            = bond_ethdev_start,
3185         .dev_stop             = bond_ethdev_stop,
3186         .dev_close            = bond_ethdev_close,
3187         .dev_configure        = bond_ethdev_configure,
3188         .dev_infos_get        = bond_ethdev_info,
3189         .vlan_filter_set      = bond_ethdev_vlan_filter_set,
3190         .rx_queue_setup       = bond_ethdev_rx_queue_setup,
3191         .tx_queue_setup       = bond_ethdev_tx_queue_setup,
3192         .rx_queue_release     = bond_ethdev_rx_queue_release,
3193         .tx_queue_release     = bond_ethdev_tx_queue_release,
3194         .link_update          = bond_ethdev_link_update,
3195         .stats_get            = bond_ethdev_stats_get,
3196         .stats_reset          = bond_ethdev_stats_reset,
3197         .promiscuous_enable   = bond_ethdev_promiscuous_enable,
3198         .promiscuous_disable  = bond_ethdev_promiscuous_disable,
3199         .allmulticast_enable  = bond_ethdev_allmulticast_enable,
3200         .allmulticast_disable = bond_ethdev_allmulticast_disable,
3201         .reta_update          = bond_ethdev_rss_reta_update,
3202         .reta_query           = bond_ethdev_rss_reta_query,
3203         .rss_hash_update      = bond_ethdev_rss_hash_update,
3204         .rss_hash_conf_get    = bond_ethdev_rss_hash_conf_get,
3205         .mtu_set              = bond_ethdev_mtu_set,
3206         .mac_addr_set         = bond_ethdev_mac_address_set,
3207         .mac_addr_add         = bond_ethdev_mac_addr_add,
3208         .mac_addr_remove      = bond_ethdev_mac_addr_remove,
3209         .flow_ops_get         = bond_flow_ops_get
3210 };
3211
3212 static int
3213 bond_alloc(struct rte_vdev_device *dev, uint8_t mode)
3214 {
3215         const char *name = rte_vdev_device_name(dev);
3216         uint8_t socket_id = dev->device.numa_node;
3217         struct bond_dev_private *internals = NULL;
3218         struct rte_eth_dev *eth_dev = NULL;
3219         uint32_t vlan_filter_bmp_size;
3220
3221         /* now do all data allocation - for eth_dev structure, dummy pci driver
3222          * and internal (private) data
3223          */
3224
3225         /* reserve an ethdev entry */
3226         eth_dev = rte_eth_vdev_allocate(dev, sizeof(*internals));
3227         if (eth_dev == NULL) {
3228                 RTE_BOND_LOG(ERR, "Unable to allocate rte_eth_dev");
3229                 goto err;
3230         }
3231
3232         internals = eth_dev->data->dev_private;
3233         eth_dev->data->nb_rx_queues = (uint16_t)1;
3234         eth_dev->data->nb_tx_queues = (uint16_t)1;
3235
3236         /* Allocate memory for storing MAC addresses */
3237         eth_dev->data->mac_addrs = rte_zmalloc_socket(name, RTE_ETHER_ADDR_LEN *
3238                         BOND_MAX_MAC_ADDRS, 0, socket_id);
3239         if (eth_dev->data->mac_addrs == NULL) {
3240                 RTE_BOND_LOG(ERR,
3241                              "Failed to allocate %u bytes needed to store MAC addresses",
3242                              RTE_ETHER_ADDR_LEN * BOND_MAX_MAC_ADDRS);
3243                 goto err;
3244         }
3245
3246         eth_dev->dev_ops = &default_dev_ops;
3247         eth_dev->data->dev_flags = RTE_ETH_DEV_INTR_LSC |
3248                                         RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
3249
3250         rte_spinlock_init(&internals->lock);
3251         rte_spinlock_init(&internals->lsc_lock);
3252
3253         internals->port_id = eth_dev->data->port_id;
3254         internals->mode = BONDING_MODE_INVALID;
3255         internals->current_primary_port = RTE_MAX_ETHPORTS + 1;
3256         internals->balance_xmit_policy = BALANCE_XMIT_POLICY_LAYER2;
3257         internals->burst_xmit_hash = burst_xmit_l2_hash;
3258         internals->user_defined_mac = 0;
3259
3260         internals->link_status_polling_enabled = 0;
3261
3262         internals->link_status_polling_interval_ms =
3263                 DEFAULT_POLLING_INTERVAL_10_MS;
3264         internals->link_down_delay_ms = 0;
3265         internals->link_up_delay_ms = 0;
3266
3267         internals->slave_count = 0;
3268         internals->active_slave_count = 0;
3269         internals->rx_offload_capa = 0;
3270         internals->tx_offload_capa = 0;
3271         internals->rx_queue_offload_capa = 0;
3272         internals->tx_queue_offload_capa = 0;
3273         internals->candidate_max_rx_pktlen = 0;
3274         internals->max_rx_pktlen = 0;
3275
3276         /* Initially allow to choose any offload type */
3277         internals->flow_type_rss_offloads = RTE_ETH_RSS_PROTO_MASK;
3278
3279         memset(&internals->default_rxconf, 0,
3280                sizeof(internals->default_rxconf));
3281         memset(&internals->default_txconf, 0,
3282                sizeof(internals->default_txconf));
3283
3284         memset(&internals->rx_desc_lim, 0, sizeof(internals->rx_desc_lim));
3285         memset(&internals->tx_desc_lim, 0, sizeof(internals->tx_desc_lim));
3286
3287         memset(internals->active_slaves, 0, sizeof(internals->active_slaves));
3288         memset(internals->slaves, 0, sizeof(internals->slaves));
3289
3290         TAILQ_INIT(&internals->flow_list);
3291         internals->flow_isolated_valid = 0;
3292
3293         /* Set mode 4 default configuration */
3294         bond_mode_8023ad_setup(eth_dev, NULL);
3295         if (bond_ethdev_mode_set(eth_dev, mode)) {
3296                 RTE_BOND_LOG(ERR, "Failed to set bonded device %d mode to %d",
3297                                  eth_dev->data->port_id, mode);
3298                 goto err;
3299         }
3300
3301         vlan_filter_bmp_size =
3302                 rte_bitmap_get_memory_footprint(RTE_ETHER_MAX_VLAN_ID + 1);
3303         internals->vlan_filter_bmpmem = rte_malloc(name, vlan_filter_bmp_size,
3304                                                    RTE_CACHE_LINE_SIZE);
3305         if (internals->vlan_filter_bmpmem == NULL) {
3306                 RTE_BOND_LOG(ERR,
3307                              "Failed to allocate vlan bitmap for bonded device %u",
3308                              eth_dev->data->port_id);
3309                 goto err;
3310         }
3311
3312         internals->vlan_filter_bmp = rte_bitmap_init(RTE_ETHER_MAX_VLAN_ID + 1,
3313                         internals->vlan_filter_bmpmem, vlan_filter_bmp_size);
3314         if (internals->vlan_filter_bmp == NULL) {
3315                 RTE_BOND_LOG(ERR,
3316                              "Failed to init vlan bitmap for bonded device %u",
3317                              eth_dev->data->port_id);
3318                 rte_free(internals->vlan_filter_bmpmem);
3319                 goto err;
3320         }
3321
3322         return eth_dev->data->port_id;
3323
3324 err:
3325         rte_free(internals);
3326         if (eth_dev != NULL)
3327                 eth_dev->data->dev_private = NULL;
3328         rte_eth_dev_release_port(eth_dev);
3329         return -1;
3330 }
3331
3332 static int
3333 bond_probe(struct rte_vdev_device *dev)
3334 {
3335         const char *name;
3336         struct bond_dev_private *internals;
3337         struct rte_kvargs *kvlist;
3338         uint8_t bonding_mode;
3339         int arg_count, port_id;
3340         int socket_id;
3341         uint8_t agg_mode;
3342         struct rte_eth_dev *eth_dev;
3343
3344         if (!dev)
3345                 return -EINVAL;
3346
3347         name = rte_vdev_device_name(dev);
3348         RTE_BOND_LOG(INFO, "Initializing pmd_bond for %s", name);
3349
3350         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
3351                 eth_dev = rte_eth_dev_attach_secondary(name);
3352                 if (!eth_dev) {
3353                         RTE_BOND_LOG(ERR, "Failed to probe %s", name);
3354                         return -1;
3355                 }
3356                 /* TODO: request info from primary to set up Rx and Tx */
3357                 eth_dev->dev_ops = &default_dev_ops;
3358                 eth_dev->device = &dev->device;
3359                 rte_eth_dev_probing_finish(eth_dev);
3360                 return 0;
3361         }
3362
3363         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev),
3364                 pmd_bond_init_valid_arguments);
3365         if (kvlist == NULL)
3366                 return -1;
3367
3368         /* Parse link bonding mode */
3369         if (rte_kvargs_count(kvlist, PMD_BOND_MODE_KVARG) == 1) {
3370                 if (rte_kvargs_process(kvlist, PMD_BOND_MODE_KVARG,
3371                                 &bond_ethdev_parse_slave_mode_kvarg,
3372                                 &bonding_mode) != 0) {
3373                         RTE_BOND_LOG(ERR, "Invalid mode for bonded device %s",
3374                                         name);
3375                         goto parse_error;
3376                 }
3377         } else {
3378                 RTE_BOND_LOG(ERR, "Mode must be specified only once for bonded "
3379                                 "device %s", name);
3380                 goto parse_error;
3381         }
3382
3383         /* Parse socket id to create bonding device on */
3384         arg_count = rte_kvargs_count(kvlist, PMD_BOND_SOCKET_ID_KVARG);
3385         if (arg_count == 1) {
3386                 if (rte_kvargs_process(kvlist, PMD_BOND_SOCKET_ID_KVARG,
3387                                 &bond_ethdev_parse_socket_id_kvarg, &socket_id)
3388                                 != 0) {
3389                         RTE_BOND_LOG(ERR, "Invalid socket Id specified for "
3390                                         "bonded device %s", name);
3391                         goto parse_error;
3392                 }
3393         } else if (arg_count > 1) {
3394                 RTE_BOND_LOG(ERR, "Socket Id can be specified only once for "
3395                                 "bonded device %s", name);
3396                 goto parse_error;
3397         } else {
3398                 socket_id = rte_socket_id();
3399         }
3400
3401         dev->device.numa_node = socket_id;
3402
3403         /* Create link bonding eth device */
3404         port_id = bond_alloc(dev, bonding_mode);
3405         if (port_id < 0) {
3406                 RTE_BOND_LOG(ERR, "Failed to create socket %s in mode %u on "
3407                                 "socket %u.",   name, bonding_mode, socket_id);
3408                 goto parse_error;
3409         }
3410         internals = rte_eth_devices[port_id].data->dev_private;
3411         internals->kvlist = kvlist;
3412
3413         if (rte_kvargs_count(kvlist, PMD_BOND_AGG_MODE_KVARG) == 1) {
3414                 if (rte_kvargs_process(kvlist,
3415                                 PMD_BOND_AGG_MODE_KVARG,
3416                                 &bond_ethdev_parse_slave_agg_mode_kvarg,
3417                                 &agg_mode) != 0) {
3418                         RTE_BOND_LOG(ERR,
3419                                         "Failed to parse agg selection mode for bonded device %s",
3420                                         name);
3421                         goto parse_error;
3422                 }
3423
3424                 if (internals->mode == BONDING_MODE_8023AD)
3425                         internals->mode4.agg_selection = agg_mode;
3426         } else {
3427                 internals->mode4.agg_selection = AGG_STABLE;
3428         }
3429
3430         rte_eth_dev_probing_finish(&rte_eth_devices[port_id]);
3431         RTE_BOND_LOG(INFO, "Create bonded device %s on port %d in mode %u on "
3432                         "socket %u.",   name, port_id, bonding_mode, socket_id);
3433         return 0;
3434
3435 parse_error:
3436         rte_kvargs_free(kvlist);
3437
3438         return -1;
3439 }
3440
3441 static int
3442 bond_remove(struct rte_vdev_device *dev)
3443 {
3444         struct rte_eth_dev *eth_dev;
3445         struct bond_dev_private *internals;
3446         const char *name;
3447         int ret = 0;
3448
3449         if (!dev)
3450                 return -EINVAL;
3451
3452         name = rte_vdev_device_name(dev);
3453         RTE_BOND_LOG(INFO, "Uninitializing pmd_bond for %s", name);
3454
3455         /* find an ethdev entry */
3456         eth_dev = rte_eth_dev_allocated(name);
3457         if (eth_dev == NULL)
3458                 return 0; /* port already released */
3459
3460         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
3461                 return rte_eth_dev_release_port(eth_dev);
3462
3463         RTE_ASSERT(eth_dev->device == &dev->device);
3464
3465         internals = eth_dev->data->dev_private;
3466         if (internals->slave_count != 0)
3467                 return -EBUSY;
3468
3469         if (eth_dev->data->dev_started == 1) {
3470                 ret = bond_ethdev_stop(eth_dev);
3471                 bond_ethdev_close(eth_dev);
3472         }
3473         rte_eth_dev_release_port(eth_dev);
3474
3475         return ret;
3476 }
3477
3478 /* this part will resolve the slave portids after all the other pdev and vdev
3479  * have been allocated */
3480 static int
3481 bond_ethdev_configure(struct rte_eth_dev *dev)
3482 {
3483         const char *name = dev->device->name;
3484         struct bond_dev_private *internals = dev->data->dev_private;
3485         struct rte_kvargs *kvlist = internals->kvlist;
3486         int arg_count;
3487         uint16_t port_id = dev - rte_eth_devices;
3488         uint8_t agg_mode;
3489
3490         static const uint8_t default_rss_key[40] = {
3491                 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2, 0x41, 0x67, 0x25, 0x3D,
3492                 0x43, 0xA3, 0x8F, 0xB0, 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
3493                 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C, 0x6A, 0x42, 0xB7, 0x3B,
3494                 0xBE, 0xAC, 0x01, 0xFA
3495         };
3496
3497         unsigned i, j;
3498
3499         /*
3500          * If RSS is enabled, fill table with default values and
3501          * set key to the the value specified in port RSS configuration.
3502          * Fall back to default RSS key if the key is not specified
3503          */
3504         if (dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS) {
3505                 struct rte_eth_rss_conf *rss_conf =
3506                         &dev->data->dev_conf.rx_adv_conf.rss_conf;
3507                 if (rss_conf->rss_key != NULL) {
3508                         if (internals->rss_key_len > rss_conf->rss_key_len) {
3509                                 RTE_BOND_LOG(ERR, "Invalid rss key length(%u)",
3510                                                 rss_conf->rss_key_len);
3511                                 return -EINVAL;
3512                         }
3513
3514                         memcpy(internals->rss_key, rss_conf->rss_key,
3515                                internals->rss_key_len);
3516                 } else {
3517                         if (internals->rss_key_len > sizeof(default_rss_key)) {
3518                                 RTE_BOND_LOG(ERR,
3519                                        "There is no suitable default hash key");
3520                                 return -EINVAL;
3521                         }
3522
3523                         memcpy(internals->rss_key, default_rss_key,
3524                                internals->rss_key_len);
3525                 }
3526
3527                 for (i = 0; i < RTE_DIM(internals->reta_conf); i++) {
3528                         internals->reta_conf[i].mask = ~0LL;
3529                         for (j = 0; j < RTE_ETH_RETA_GROUP_SIZE; j++)
3530                                 internals->reta_conf[i].reta[j] =
3531                                                 (i * RTE_ETH_RETA_GROUP_SIZE + j) %
3532                                                 dev->data->nb_rx_queues;
3533                 }
3534         }
3535
3536         /* set the max_rx_pktlen */
3537         internals->max_rx_pktlen = internals->candidate_max_rx_pktlen;
3538
3539         /*
3540          * if no kvlist, it means that this bonded device has been created
3541          * through the bonding api.
3542          */
3543         if (!kvlist)
3544                 return 0;
3545
3546         /* Parse MAC address for bonded device */
3547         arg_count = rte_kvargs_count(kvlist, PMD_BOND_MAC_ADDR_KVARG);
3548         if (arg_count == 1) {
3549                 struct rte_ether_addr bond_mac;
3550
3551                 if (rte_kvargs_process(kvlist, PMD_BOND_MAC_ADDR_KVARG,
3552                                        &bond_ethdev_parse_bond_mac_addr_kvarg, &bond_mac) < 0) {
3553                         RTE_BOND_LOG(INFO, "Invalid mac address for bonded device %s",
3554                                      name);
3555                         return -1;
3556                 }
3557
3558                 /* Set MAC address */
3559                 if (rte_eth_bond_mac_address_set(port_id, &bond_mac) != 0) {
3560                         RTE_BOND_LOG(ERR,
3561                                      "Failed to set mac address on bonded device %s",
3562                                      name);
3563                         return -1;
3564                 }
3565         } else if (arg_count > 1) {
3566                 RTE_BOND_LOG(ERR,
3567                              "MAC address can be specified only once for bonded device %s",
3568                              name);
3569                 return -1;
3570         }
3571
3572         /* Parse/set balance mode transmit policy */
3573         arg_count = rte_kvargs_count(kvlist, PMD_BOND_XMIT_POLICY_KVARG);
3574         if (arg_count == 1) {
3575                 uint8_t xmit_policy;
3576
3577                 if (rte_kvargs_process(kvlist, PMD_BOND_XMIT_POLICY_KVARG,
3578                                        &bond_ethdev_parse_balance_xmit_policy_kvarg, &xmit_policy) !=
3579                     0) {
3580                         RTE_BOND_LOG(INFO,
3581                                      "Invalid xmit policy specified for bonded device %s",
3582                                      name);
3583                         return -1;
3584                 }
3585
3586                 /* Set balance mode transmit policy*/
3587                 if (rte_eth_bond_xmit_policy_set(port_id, xmit_policy) != 0) {
3588                         RTE_BOND_LOG(ERR,
3589                                      "Failed to set balance xmit policy on bonded device %s",
3590                                      name);
3591                         return -1;
3592                 }
3593         } else if (arg_count > 1) {
3594                 RTE_BOND_LOG(ERR,
3595                              "Transmit policy can be specified only once for bonded device %s",
3596                              name);
3597                 return -1;
3598         }
3599
3600         if (rte_kvargs_count(kvlist, PMD_BOND_AGG_MODE_KVARG) == 1) {
3601                 if (rte_kvargs_process(kvlist,
3602                                        PMD_BOND_AGG_MODE_KVARG,
3603                                        &bond_ethdev_parse_slave_agg_mode_kvarg,
3604                                        &agg_mode) != 0) {
3605                         RTE_BOND_LOG(ERR,
3606                                      "Failed to parse agg selection mode for bonded device %s",
3607                                      name);
3608                 }
3609                 if (internals->mode == BONDING_MODE_8023AD) {
3610                         int ret = rte_eth_bond_8023ad_agg_selection_set(port_id,
3611                                         agg_mode);
3612                         if (ret < 0) {
3613                                 RTE_BOND_LOG(ERR,
3614                                         "Invalid args for agg selection set for bonded device %s",
3615                                         name);
3616                                 return -1;
3617                         }
3618                 }
3619         }
3620
3621         /* Parse/add slave ports to bonded device */
3622         if (rte_kvargs_count(kvlist, PMD_BOND_SLAVE_PORT_KVARG) > 0) {
3623                 struct bond_ethdev_slave_ports slave_ports;
3624                 unsigned i;
3625
3626                 memset(&slave_ports, 0, sizeof(slave_ports));
3627
3628                 if (rte_kvargs_process(kvlist, PMD_BOND_SLAVE_PORT_KVARG,
3629                                        &bond_ethdev_parse_slave_port_kvarg, &slave_ports) != 0) {
3630                         RTE_BOND_LOG(ERR,
3631                                      "Failed to parse slave ports for bonded device %s",
3632                                      name);
3633                         return -1;
3634                 }
3635
3636                 for (i = 0; i < slave_ports.slave_count; i++) {
3637                         if (rte_eth_bond_slave_add(port_id, slave_ports.slaves[i]) != 0) {
3638                                 RTE_BOND_LOG(ERR,
3639                                              "Failed to add port %d as slave to bonded device %s",
3640                                              slave_ports.slaves[i], name);
3641                         }
3642                 }
3643
3644         } else {
3645                 RTE_BOND_LOG(INFO, "No slaves specified for bonded device %s", name);
3646                 return -1;
3647         }
3648
3649         /* Parse/set primary slave port id*/
3650         arg_count = rte_kvargs_count(kvlist, PMD_BOND_PRIMARY_SLAVE_KVARG);
3651         if (arg_count == 1) {
3652                 uint16_t primary_slave_port_id;
3653
3654                 if (rte_kvargs_process(kvlist,
3655                                        PMD_BOND_PRIMARY_SLAVE_KVARG,
3656                                        &bond_ethdev_parse_primary_slave_port_id_kvarg,
3657                                        &primary_slave_port_id) < 0) {
3658                         RTE_BOND_LOG(INFO,
3659                                      "Invalid primary slave port id specified for bonded device %s",
3660                                      name);
3661                         return -1;
3662                 }
3663
3664                 /* Set balance mode transmit policy*/
3665                 if (rte_eth_bond_primary_set(port_id, primary_slave_port_id)
3666                     != 0) {
3667                         RTE_BOND_LOG(ERR,
3668                                      "Failed to set primary slave port %d on bonded device %s",
3669                                      primary_slave_port_id, name);
3670                         return -1;
3671                 }
3672         } else if (arg_count > 1) {
3673                 RTE_BOND_LOG(INFO,
3674                              "Primary slave can be specified only once for bonded device %s",
3675                              name);
3676                 return -1;
3677         }
3678
3679         /* Parse link status monitor polling interval */
3680         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LSC_POLL_PERIOD_KVARG);
3681         if (arg_count == 1) {
3682                 uint32_t lsc_poll_interval_ms;
3683
3684                 if (rte_kvargs_process(kvlist,
3685                                        PMD_BOND_LSC_POLL_PERIOD_KVARG,
3686                                        &bond_ethdev_parse_time_ms_kvarg,
3687                                        &lsc_poll_interval_ms) < 0) {
3688                         RTE_BOND_LOG(INFO,
3689                                      "Invalid lsc polling interval value specified for bonded"
3690                                      " device %s", name);
3691                         return -1;
3692                 }
3693
3694                 if (rte_eth_bond_link_monitoring_set(port_id, lsc_poll_interval_ms)
3695                     != 0) {
3696                         RTE_BOND_LOG(ERR,
3697                                      "Failed to set lsc monitor polling interval (%u ms) on bonded device %s",
3698                                      lsc_poll_interval_ms, name);
3699                         return -1;
3700                 }
3701         } else if (arg_count > 1) {
3702                 RTE_BOND_LOG(INFO,
3703                              "LSC polling interval can be specified only once for bonded"
3704                              " device %s", name);
3705                 return -1;
3706         }
3707
3708         /* Parse link up interrupt propagation delay */
3709         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_UP_PROP_DELAY_KVARG);
3710         if (arg_count == 1) {
3711                 uint32_t link_up_delay_ms;
3712
3713                 if (rte_kvargs_process(kvlist,
3714                                        PMD_BOND_LINK_UP_PROP_DELAY_KVARG,
3715                                        &bond_ethdev_parse_time_ms_kvarg,
3716                                        &link_up_delay_ms) < 0) {
3717                         RTE_BOND_LOG(INFO,
3718                                      "Invalid link up propagation delay value specified for"
3719                                      " bonded device %s", name);
3720                         return -1;
3721                 }
3722
3723                 /* Set balance mode transmit policy*/
3724                 if (rte_eth_bond_link_up_prop_delay_set(port_id, link_up_delay_ms)
3725                     != 0) {
3726                         RTE_BOND_LOG(ERR,
3727                                      "Failed to set link up propagation delay (%u ms) on bonded"
3728                                      " device %s", link_up_delay_ms, name);
3729                         return -1;
3730                 }
3731         } else if (arg_count > 1) {
3732                 RTE_BOND_LOG(INFO,
3733                              "Link up propagation delay can be specified only once for"
3734                              " bonded device %s", name);
3735                 return -1;
3736         }
3737
3738         /* Parse link down interrupt propagation delay */
3739         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG);
3740         if (arg_count == 1) {
3741                 uint32_t link_down_delay_ms;
3742
3743                 if (rte_kvargs_process(kvlist,
3744                                        PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG,
3745                                        &bond_ethdev_parse_time_ms_kvarg,
3746                                        &link_down_delay_ms) < 0) {
3747                         RTE_BOND_LOG(INFO,
3748                                      "Invalid link down propagation delay value specified for"
3749                                      " bonded device %s", name);
3750                         return -1;
3751                 }
3752
3753                 /* Set balance mode transmit policy*/
3754                 if (rte_eth_bond_link_down_prop_delay_set(port_id, link_down_delay_ms)
3755                     != 0) {
3756                         RTE_BOND_LOG(ERR,
3757                                      "Failed to set link down propagation delay (%u ms) on bonded device %s",
3758                                      link_down_delay_ms, name);
3759                         return -1;
3760                 }
3761         } else if (arg_count > 1) {
3762                 RTE_BOND_LOG(INFO,
3763                              "Link down propagation delay can be specified only once for  bonded device %s",
3764                              name);
3765                 return -1;
3766         }
3767
3768         return 0;
3769 }
3770
3771 struct rte_vdev_driver pmd_bond_drv = {
3772         .probe = bond_probe,
3773         .remove = bond_remove,
3774 };
3775
3776 RTE_PMD_REGISTER_VDEV(net_bonding, pmd_bond_drv);
3777 RTE_PMD_REGISTER_ALIAS(net_bonding, eth_bond);
3778
3779 RTE_PMD_REGISTER_PARAM_STRING(net_bonding,
3780         "slave=<ifc> "
3781         "primary=<ifc> "
3782         "mode=[0-6] "
3783         "xmit_policy=[l2 | l23 | l34] "
3784         "agg_mode=[count | stable | bandwidth] "
3785         "socket_id=<int> "
3786         "mac=<mac addr> "
3787         "lsc_poll_period_ms=<int> "
3788         "up_delay=<int> "
3789         "down_delay=<int>");
3790
3791 /* We can't use RTE_LOG_REGISTER_DEFAULT because of the forced name for
3792  * this library, see meson.build.
3793  */
3794 RTE_LOG_REGISTER(bond_logtype, pmd.net.bonding, NOTICE);