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 & PKT_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->d_addr) &&
 346                                    !rte_is_same_ether_addr(bond_mac,
 347                                                        &hdr->d_addr)) ||
 348                                   (!allmulti &&
 349                                    rte_is_multicast_ether_addr(&hdr->d_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:%02X:%02X:%02X:%02X:%02X:%02X SrcIP:%s " \
 477                 "DstMAC:%02X:%02X:%02X:%02X:%02X:%02X DstIP:%s %s %d\n", \
 478                 info,                                                   \
 479                 port,                                                   \
 480                 eth_h->s_addr.addr_bytes[0], eth_h->s_addr.addr_bytes[1], \
 481                 eth_h->s_addr.addr_bytes[2], eth_h->s_addr.addr_bytes[3], \
 482                 eth_h->s_addr.addr_bytes[4], eth_h->s_addr.addr_bytes[5], \
 483                 src_ip,                                                 \
 484                 eth_h->d_addr.addr_bytes[0], eth_h->d_addr.addr_bytes[1], \
 485                 eth_h->d_addr.addr_bytes[2], eth_h->d_addr.addr_bytes[3], \
 486                 eth_h->d_addr.addr_bytes[4], eth_h->d_addr.addr_bytes[5], \
 487                 dst_ip,                                                 \
 488                 arp_op, ++burstnumber)
 489 #endif
 490
 491 static void
 492 mode6_debug(const char __rte_unused *info,
 493         struct rte_ether_hdr *eth_h, uint16_t port,
 494         uint32_t __rte_unused *burstnumber)
 495 {
 496         struct rte_ipv4_hdr *ipv4_h;
 497 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 498         struct rte_arp_hdr *arp_h;
 499         char dst_ip[16];
 500         char ArpOp[24];
 501         char buf[16];
 502 #endif
 503         char src_ip[16];
 504
 505         uint16_t ether_type = eth_h->ether_type;
 506         uint16_t offset = get_vlan_offset(eth_h, &ether_type);
 507
 508 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 509         strlcpy(buf, info, 16);
 510 #endif
 511
 512         if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
 513                 ipv4_h = (struct rte_ipv4_hdr *)((char *)(eth_h + 1) + offset);
 514                 ipv4_addr_to_dot(ipv4_h->src_addr, src_ip, MaxIPv4String);
 515 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 516                 ipv4_addr_to_dot(ipv4_h->dst_addr, dst_ip, MaxIPv4String);
 517                 MODE6_DEBUG(buf, src_ip, dst_ip, eth_h, "", port, *burstnumber);
 518 #endif
 519                 update_client_stats(ipv4_h->src_addr, port, burstnumber);
 520         }
 521 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 522         else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_ARP)) {
 523                 arp_h = (struct rte_arp_hdr *)((char *)(eth_h + 1) + offset);
 524                 ipv4_addr_to_dot(arp_h->arp_data.arp_sip, src_ip, MaxIPv4String);
 525                 ipv4_addr_to_dot(arp_h->arp_data.arp_tip, dst_ip, MaxIPv4String);
 526                 arp_op_name(rte_be_to_cpu_16(arp_h->arp_opcode),
 527                                 ArpOp, sizeof(ArpOp));
 528                 MODE6_DEBUG(buf, src_ip, dst_ip, eth_h, ArpOp, port, *burstnumber);
 529         }
 530 #endif
 531 }
 532 #endif
 533
 534 static uint16_t
 535 bond_ethdev_rx_burst_alb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 536 {
 537         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
 538         struct bond_dev_private *internals = bd_rx_q->dev_private;
 539         struct rte_ether_hdr *eth_h;
 540         uint16_t ether_type, offset;
 541         uint16_t nb_recv_pkts;
 542         int i;
 543
 544         nb_recv_pkts = bond_ethdev_rx_burst(queue, bufs, nb_pkts);
 545
 546         for (i = 0; i < nb_recv_pkts; i++) {
 547                 eth_h = rte_pktmbuf_mtod(bufs[i], struct rte_ether_hdr *);
 548                 ether_type = eth_h->ether_type;
 549                 offset = get_vlan_offset(eth_h, &ether_type);
 550
 551                 if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_ARP)) {
 552 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 553                         mode6_debug("RX ARP:", eth_h, bufs[i]->port, &burstnumberRX);
 554 #endif
 555                         bond_mode_alb_arp_recv(eth_h, offset, internals);
 556                 }
 557 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 558                 else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
 559                         mode6_debug("RX IPv4:", eth_h, bufs[i]->port, &burstnumberRX);
 560 #endif
 561         }
 562
 563         return nb_recv_pkts;
 564 }
 565
 566 static uint16_t
 567 bond_ethdev_tx_burst_round_robin(void *queue, struct rte_mbuf **bufs,
 568                 uint16_t nb_pkts)
 569 {
 570         struct bond_dev_private *internals;
 571         struct bond_tx_queue *bd_tx_q;
 572
 573         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_pkts];
 574         uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
 575
 576         uint16_t num_of_slaves;
 577         uint16_t slaves[RTE_MAX_ETHPORTS];
 578
 579         uint16_t num_tx_total = 0, num_tx_slave;
 580
 581         static int slave_idx = 0;
 582         int i, cslave_idx = 0, tx_fail_total = 0;
 583
 584         bd_tx_q = (struct bond_tx_queue *)queue;
 585         internals = bd_tx_q->dev_private;
 586
 587         /* Copy slave list to protect against slave up/down changes during tx
 588          * bursting */
 589         num_of_slaves = internals->active_slave_count;
 590         memcpy(slaves, internals->active_slaves,
 591                         sizeof(internals->active_slaves[0]) * num_of_slaves);
 592
 593         if (num_of_slaves < 1)
 594                 return num_tx_total;
 595
 596         /* Populate slaves mbuf with which packets are to be sent on it  */
 597         for (i = 0; i < nb_pkts; i++) {
 598                 cslave_idx = (slave_idx + i) % num_of_slaves;
 599                 slave_bufs[cslave_idx][(slave_nb_pkts[cslave_idx])++] = bufs[i];
 600         }
 601
 602         /* increment current slave index so the next call to tx burst starts on the
 603          * next slave */
 604         slave_idx = ++cslave_idx;
 605
 606         /* Send packet burst on each slave device */
 607         for (i = 0; i < num_of_slaves; i++) {
 608                 if (slave_nb_pkts[i] > 0) {
 609                         num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
 610                                         slave_bufs[i], slave_nb_pkts[i]);
 611
 612                         /* if tx burst fails move packets to end of bufs */
 613                         if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
 614                                 int tx_fail_slave = slave_nb_pkts[i] - num_tx_slave;
 615
 616                                 tx_fail_total += tx_fail_slave;
 617
 618                                 memcpy(&bufs[nb_pkts - tx_fail_total],
 619                                        &slave_bufs[i][num_tx_slave],
 620                                        tx_fail_slave * sizeof(bufs[0]));
 621                         }
 622                         num_tx_total += num_tx_slave;
 623                 }
 624         }
 625
 626         return num_tx_total;
 627 }
 628
 629 static uint16_t
 630 bond_ethdev_tx_burst_active_backup(void *queue,
 631                 struct rte_mbuf **bufs, uint16_t nb_pkts)
 632 {
 633         struct bond_dev_private *internals;
 634         struct bond_tx_queue *bd_tx_q;
 635
 636         bd_tx_q = (struct bond_tx_queue *)queue;
 637         internals = bd_tx_q->dev_private;
 638
 639         if (internals->active_slave_count < 1)
 640                 return 0;
 641
 642         return rte_eth_tx_burst(internals->current_primary_port, bd_tx_q->queue_id,
 643                         bufs, nb_pkts);
 644 }
 645
 646 static inline uint16_t
 647 ether_hash(struct rte_ether_hdr *eth_hdr)
 648 {
 649         unaligned_uint16_t *word_src_addr =
 650                 (unaligned_uint16_t *)eth_hdr->s_addr.addr_bytes;
 651         unaligned_uint16_t *word_dst_addr =
 652                 (unaligned_uint16_t *)eth_hdr->d_addr.addr_bytes;
 653
 654         return (word_src_addr[0] ^ word_dst_addr[0]) ^
 655                         (word_src_addr[1] ^ word_dst_addr[1]) ^
 656                         (word_src_addr[2] ^ word_dst_addr[2]);
 657 }
 658
 659 static inline uint32_t
 660 ipv4_hash(struct rte_ipv4_hdr *ipv4_hdr)
 661 {
 662         return ipv4_hdr->src_addr ^ ipv4_hdr->dst_addr;
 663 }
 664
 665 static inline uint32_t
 666 ipv6_hash(struct rte_ipv6_hdr *ipv6_hdr)
 667 {
 668         unaligned_uint32_t *word_src_addr =
 669                 (unaligned_uint32_t *)&(ipv6_hdr->src_addr[0]);
 670         unaligned_uint32_t *word_dst_addr =
 671                 (unaligned_uint32_t *)&(ipv6_hdr->dst_addr[0]);
 672
 673         return (word_src_addr[0] ^ word_dst_addr[0]) ^
 674                         (word_src_addr[1] ^ word_dst_addr[1]) ^
 675                         (word_src_addr[2] ^ word_dst_addr[2]) ^
 676                         (word_src_addr[3] ^ word_dst_addr[3]);
 677 }
 678
 679
 680 void
 681 burst_xmit_l2_hash(struct rte_mbuf **buf, uint16_t nb_pkts,
 682                 uint16_t slave_count, uint16_t *slaves)
 683 {
 684         struct rte_ether_hdr *eth_hdr;
 685         uint32_t hash;
 686         int i;
 687
 688         for (i = 0; i < nb_pkts; i++) {
 689                 eth_hdr = rte_pktmbuf_mtod(buf[i], struct rte_ether_hdr *);
 690
 691                 hash = ether_hash(eth_hdr);
 692
 693                 slaves[i] = (hash ^= hash >> 8) % slave_count;
 694         }
 695 }
 696
 697 void
 698 burst_xmit_l23_hash(struct rte_mbuf **buf, uint16_t nb_pkts,
 699                 uint16_t slave_count, uint16_t *slaves)
 700 {
 701         uint16_t i;
 702         struct rte_ether_hdr *eth_hdr;
 703         uint16_t proto;
 704         size_t vlan_offset;
 705         uint32_t hash, l3hash;
 706
 707         for (i = 0; i < nb_pkts; i++) {
 708                 eth_hdr = rte_pktmbuf_mtod(buf[i], struct rte_ether_hdr *);
 709                 l3hash = 0;
 710
 711                 proto = eth_hdr->ether_type;
 712                 hash = ether_hash(eth_hdr);
 713
 714                 vlan_offset = get_vlan_offset(eth_hdr, &proto);
 715
 716                 if (rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) == proto) {
 717                         struct rte_ipv4_hdr *ipv4_hdr = (struct rte_ipv4_hdr *)
 718                                         ((char *)(eth_hdr + 1) + vlan_offset);
 719                         l3hash = ipv4_hash(ipv4_hdr);
 720
 721                 } else if (rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) == proto) {
 722                         struct rte_ipv6_hdr *ipv6_hdr = (struct rte_ipv6_hdr *)
 723                                         ((char *)(eth_hdr + 1) + vlan_offset);
 724                         l3hash = ipv6_hash(ipv6_hdr);
 725                 }
 726
 727                 hash = hash ^ l3hash;
 728                 hash ^= hash >> 16;
 729                 hash ^= hash >> 8;
 730
 731                 slaves[i] = hash % slave_count;
 732         }
 733 }
 734
 735 void
 736 burst_xmit_l34_hash(struct rte_mbuf **buf, uint16_t nb_pkts,
 737                 uint16_t slave_count, uint16_t *slaves)
 738 {
 739         struct rte_ether_hdr *eth_hdr;
 740         uint16_t proto;
 741         size_t vlan_offset;
 742         int i;
 743
 744         struct rte_udp_hdr *udp_hdr;
 745         struct rte_tcp_hdr *tcp_hdr;
 746         uint32_t hash, l3hash, l4hash;
 747
 748         for (i = 0; i < nb_pkts; i++) {
 749                 eth_hdr = rte_pktmbuf_mtod(buf[i], struct rte_ether_hdr *);
 750                 size_t pkt_end = (size_t)eth_hdr + rte_pktmbuf_data_len(buf[i]);
 751                 proto = eth_hdr->ether_type;
 752                 vlan_offset = get_vlan_offset(eth_hdr, &proto);
 753                 l3hash = 0;
 754                 l4hash = 0;
 755
 756                 if (rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) == proto) {
 757                         struct rte_ipv4_hdr *ipv4_hdr = (struct rte_ipv4_hdr *)
 758                                         ((char *)(eth_hdr + 1) + vlan_offset);
 759                         size_t ip_hdr_offset;
 760
 761                         l3hash = ipv4_hash(ipv4_hdr);
 762
 763                         /* there is no L4 header in fragmented packet */
 764                         if (likely(rte_ipv4_frag_pkt_is_fragmented(ipv4_hdr)
 765                                                                 == 0)) {
 766                                 ip_hdr_offset = (ipv4_hdr->version_ihl
 767                                         & RTE_IPV4_HDR_IHL_MASK) *
 768                                         RTE_IPV4_IHL_MULTIPLIER;
 769
 770                                 if (ipv4_hdr->next_proto_id == IPPROTO_TCP) {
 771                                         tcp_hdr = (struct rte_tcp_hdr *)
 772                                                 ((char *)ipv4_hdr +
 773                                                         ip_hdr_offset);
 774                                         if ((size_t)tcp_hdr + sizeof(*tcp_hdr)
 775                                                         < pkt_end)
 776                                                 l4hash = HASH_L4_PORTS(tcp_hdr);
 777                                 } else if (ipv4_hdr->next_proto_id ==
 778                                                                 IPPROTO_UDP) {
 779                                         udp_hdr = (struct rte_udp_hdr *)
 780                                                 ((char *)ipv4_hdr +
 781                                                         ip_hdr_offset);
 782                                         if ((size_t)udp_hdr + sizeof(*udp_hdr)
 783                                                         < pkt_end)
 784                                                 l4hash = HASH_L4_PORTS(udp_hdr);
 785                                 }
 786                         }
 787                 } else if  (rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) == proto) {
 788                         struct rte_ipv6_hdr *ipv6_hdr = (struct rte_ipv6_hdr *)
 789                                         ((char *)(eth_hdr + 1) + vlan_offset);
 790                         l3hash = ipv6_hash(ipv6_hdr);
 791
 792                         if (ipv6_hdr->proto == IPPROTO_TCP) {
 793                                 tcp_hdr = (struct rte_tcp_hdr *)(ipv6_hdr + 1);
 794                                 l4hash = HASH_L4_PORTS(tcp_hdr);
 795                         } else if (ipv6_hdr->proto == IPPROTO_UDP) {
 796                                 udp_hdr = (struct rte_udp_hdr *)(ipv6_hdr + 1);
 797                                 l4hash = HASH_L4_PORTS(udp_hdr);
 798                         }
 799                 }
 800
 801                 hash = l3hash ^ l4hash;
 802                 hash ^= hash >> 16;
 803                 hash ^= hash >> 8;
 804
 805                 slaves[i] = hash % slave_count;
 806         }
 807 }
 808
 809 struct bwg_slave {
 810         uint64_t bwg_left_int;
 811         uint64_t bwg_left_remainder;
 812         uint16_t slave;
 813 };
 814
 815 void
 816 bond_tlb_activate_slave(struct bond_dev_private *internals) {
 817         int i;
 818
 819         for (i = 0; i < internals->active_slave_count; i++) {
 820                 tlb_last_obytets[internals->active_slaves[i]] = 0;
 821         }
 822 }
 823
 824 static int
 825 bandwidth_cmp(const void *a, const void *b)
 826 {
 827         const struct bwg_slave *bwg_a = a;
 828         const struct bwg_slave *bwg_b = b;
 829         int64_t diff = (int64_t)bwg_b->bwg_left_int - (int64_t)bwg_a->bwg_left_int;
 830         int64_t diff2 = (int64_t)bwg_b->bwg_left_remainder -
 831                         (int64_t)bwg_a->bwg_left_remainder;
 832         if (diff > 0)
 833                 return 1;
 834         else if (diff < 0)
 835                 return -1;
 836         else if (diff2 > 0)
 837                 return 1;
 838         else if (diff2 < 0)
 839                 return -1;
 840         else
 841                 return 0;
 842 }
 843
 844 static void
 845 bandwidth_left(uint16_t port_id, uint64_t load, uint8_t update_idx,
 846                 struct bwg_slave *bwg_slave)
 847 {
 848         struct rte_eth_link link_status;
 849         int ret;
 850
 851         ret = rte_eth_link_get_nowait(port_id, &link_status);
 852         if (ret < 0) {
 853                 RTE_BOND_LOG(ERR, "Slave (port %u) link get failed: %s",
 854                              port_id, rte_strerror(-ret));
 855                 return;
 856         }
 857         uint64_t link_bwg = link_status.link_speed * 1000000ULL / 8;
 858         if (link_bwg == 0)
 859                 return;
 860         link_bwg = link_bwg * (update_idx+1) * REORDER_PERIOD_MS;
 861         bwg_slave->bwg_left_int = (link_bwg - 1000*load) / link_bwg;
 862         bwg_slave->bwg_left_remainder = (link_bwg - 1000*load) % link_bwg;
 863 }
 864
 865 static void
 866 bond_ethdev_update_tlb_slave_cb(void *arg)
 867 {
 868         struct bond_dev_private *internals = arg;
 869         struct rte_eth_stats slave_stats;
 870         struct bwg_slave bwg_array[RTE_MAX_ETHPORTS];
 871         uint16_t slave_count;
 872         uint64_t tx_bytes;
 873
 874         uint8_t update_stats = 0;
 875         uint16_t slave_id;
 876         uint16_t i;
 877
 878         internals->slave_update_idx++;
 879
 880
 881         if (internals->slave_update_idx >= REORDER_PERIOD_MS)
 882                 update_stats = 1;
 883
 884         for (i = 0; i < internals->active_slave_count; i++) {
 885                 slave_id = internals->active_slaves[i];
 886                 rte_eth_stats_get(slave_id, &slave_stats);
 887                 tx_bytes = slave_stats.obytes - tlb_last_obytets[slave_id];
 888                 bandwidth_left(slave_id, tx_bytes,
 889                                 internals->slave_update_idx, &bwg_array[i]);
 890                 bwg_array[i].slave = slave_id;
 891
 892                 if (update_stats) {
 893                         tlb_last_obytets[slave_id] = slave_stats.obytes;
 894                 }
 895         }
 896
 897         if (update_stats == 1)
 898                 internals->slave_update_idx = 0;
 899
 900         slave_count = i;
 901         qsort(bwg_array, slave_count, sizeof(bwg_array[0]), bandwidth_cmp);
 902         for (i = 0; i < slave_count; i++)
 903                 internals->tlb_slaves_order[i] = bwg_array[i].slave;
 904
 905         rte_eal_alarm_set(REORDER_PERIOD_MS * 1000, bond_ethdev_update_tlb_slave_cb,
 906                         (struct bond_dev_private *)internals);
 907 }
 908
 909 static uint16_t
 910 bond_ethdev_tx_burst_tlb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 911 {
 912         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
 913         struct bond_dev_private *internals = bd_tx_q->dev_private;
 914
 915         struct rte_eth_dev *primary_port =
 916                         &rte_eth_devices[internals->primary_port];
 917         uint16_t num_tx_total = 0;
 918         uint16_t i, j;
 919
 920         uint16_t num_of_slaves = internals->active_slave_count;
 921         uint16_t slaves[RTE_MAX_ETHPORTS];
 922
 923         struct rte_ether_hdr *ether_hdr;
 924         struct rte_ether_addr primary_slave_addr;
 925         struct rte_ether_addr active_slave_addr;
 926
 927         if (num_of_slaves < 1)
 928                 return num_tx_total;
 929
 930         memcpy(slaves, internals->tlb_slaves_order,
 931                                 sizeof(internals->tlb_slaves_order[0]) * num_of_slaves);
 932
 933
 934         rte_ether_addr_copy(primary_port->data->mac_addrs, &primary_slave_addr);
 935
 936         if (nb_pkts > 3) {
 937                 for (i = 0; i < 3; i++)
 938                         rte_prefetch0(rte_pktmbuf_mtod(bufs[i], void*));
 939         }
 940
 941         for (i = 0; i < num_of_slaves; i++) {
 942                 rte_eth_macaddr_get(slaves[i], &active_slave_addr);
 943                 for (j = num_tx_total; j < nb_pkts; j++) {
 944                         if (j + 3 < nb_pkts)
 945                                 rte_prefetch0(rte_pktmbuf_mtod(bufs[j+3], void*));
 946
 947                         ether_hdr = rte_pktmbuf_mtod(bufs[j],
 948                                                 struct rte_ether_hdr *);
 949                         if (rte_is_same_ether_addr(&ether_hdr->s_addr,
 950                                                         &primary_slave_addr))
 951                                 rte_ether_addr_copy(&active_slave_addr,
 952                                                 &ether_hdr->s_addr);
 953 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 954                                         mode6_debug("TX IPv4:", ether_hdr, slaves[i], &burstnumberTX);
 955 #endif
 956                 }
 957
 958                 num_tx_total += rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
 959                                 bufs + num_tx_total, nb_pkts - num_tx_total);
 960
 961                 if (num_tx_total == nb_pkts)
 962                         break;
 963         }
 964
 965         return num_tx_total;
 966 }
 967
 968 void
 969 bond_tlb_disable(struct bond_dev_private *internals)
 970 {
 971         rte_eal_alarm_cancel(bond_ethdev_update_tlb_slave_cb, internals);
 972 }
 973
 974 void
 975 bond_tlb_enable(struct bond_dev_private *internals)
 976 {
 977         bond_ethdev_update_tlb_slave_cb(internals);
 978 }
 979
 980 static uint16_t
 981 bond_ethdev_tx_burst_alb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 982 {
 983         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
 984         struct bond_dev_private *internals = bd_tx_q->dev_private;
 985
 986         struct rte_ether_hdr *eth_h;
 987         uint16_t ether_type, offset;
 988
 989         struct client_data *client_info;
 990
 991         /*
 992          * We create transmit buffers for every slave and one additional to send
 993          * through tlb. In worst case every packet will be send on one port.
 994          */
 995         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS + 1][nb_pkts];
 996         uint16_t slave_bufs_pkts[RTE_MAX_ETHPORTS + 1] = { 0 };
 997
 998         /*
 999          * We create separate transmit buffers for update packets as they won't
1000          * be counted in num_tx_total.
1001          */
1002         struct rte_mbuf *update_bufs[RTE_MAX_ETHPORTS][ALB_HASH_TABLE_SIZE];
1003         uint16_t update_bufs_pkts[RTE_MAX_ETHPORTS] = { 0 };
1004
1005         struct rte_mbuf *upd_pkt;
1006         size_t pkt_size;
1007
1008         uint16_t num_send, num_not_send = 0;
1009         uint16_t num_tx_total = 0;
1010         uint16_t slave_idx;
1011
1012         int i, j;
1013
1014         /* Search tx buffer for ARP packets and forward them to alb */
1015         for (i = 0; i < nb_pkts; i++) {
1016                 eth_h = rte_pktmbuf_mtod(bufs[i], struct rte_ether_hdr *);
1017                 ether_type = eth_h->ether_type;
1018                 offset = get_vlan_offset(eth_h, &ether_type);
1019
1020                 if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_ARP)) {
1021                         slave_idx = bond_mode_alb_arp_xmit(eth_h, offset, internals);
1022
1023                         /* Change src mac in eth header */
1024                         rte_eth_macaddr_get(slave_idx, &eth_h->s_addr);
1025
1026                         /* Add packet to slave tx buffer */
1027                         slave_bufs[slave_idx][slave_bufs_pkts[slave_idx]] = bufs[i];
1028                         slave_bufs_pkts[slave_idx]++;
1029                 } else {
1030                         /* If packet is not ARP, send it with TLB policy */
1031                         slave_bufs[RTE_MAX_ETHPORTS][slave_bufs_pkts[RTE_MAX_ETHPORTS]] =
1032                                         bufs[i];
1033                         slave_bufs_pkts[RTE_MAX_ETHPORTS]++;
1034                 }
1035         }
1036
1037         /* Update connected client ARP tables */
1038         if (internals->mode6.ntt) {
1039                 for (i = 0; i < ALB_HASH_TABLE_SIZE; i++) {
1040                         client_info = &internals->mode6.client_table[i];
1041
1042                         if (client_info->in_use) {
1043                                 /* Allocate new packet to send ARP update on current slave */
1044                                 upd_pkt = rte_pktmbuf_alloc(internals->mode6.mempool);
1045                                 if (upd_pkt == NULL) {
1046                                         RTE_BOND_LOG(ERR,
1047                                                      "Failed to allocate ARP packet from pool");
1048                                         continue;
1049                                 }
1050                                 pkt_size = sizeof(struct rte_ether_hdr) +
1051                                         sizeof(struct rte_arp_hdr) +
1052                                         client_info->vlan_count *
1053                                         sizeof(struct rte_vlan_hdr);
1054                                 upd_pkt->data_len = pkt_size;
1055                                 upd_pkt->pkt_len = pkt_size;
1056
1057                                 slave_idx = bond_mode_alb_arp_upd(client_info, upd_pkt,
1058                                                 internals);
1059
1060                                 /* Add packet to update tx buffer */
1061                                 update_bufs[slave_idx][update_bufs_pkts[slave_idx]] = upd_pkt;
1062                                 update_bufs_pkts[slave_idx]++;
1063                         }
1064                 }
1065                 internals->mode6.ntt = 0;
1066         }
1067
1068         /* Send ARP packets on proper slaves */
1069         for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
1070                 if (slave_bufs_pkts[i] > 0) {
1071                         num_send = rte_eth_tx_burst(i, bd_tx_q->queue_id,
1072                                         slave_bufs[i], slave_bufs_pkts[i]);
1073                         for (j = 0; j < slave_bufs_pkts[i] - num_send; j++) {
1074                                 bufs[nb_pkts - 1 - num_not_send - j] =
1075                                                 slave_bufs[i][nb_pkts - 1 - j];
1076                         }
1077
1078                         num_tx_total += num_send;
1079                         num_not_send += slave_bufs_pkts[i] - num_send;
1080
1081 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
1082         /* Print TX stats including update packets */
1083                         for (j = 0; j < slave_bufs_pkts[i]; j++) {
1084                                 eth_h = rte_pktmbuf_mtod(slave_bufs[i][j],
1085                                                         struct rte_ether_hdr *);
1086                                 mode6_debug("TX ARP:", eth_h, i, &burstnumberTX);
1087                         }
1088 #endif
1089                 }
1090         }
1091
1092         /* Send update packets on proper slaves */
1093         for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
1094                 if (update_bufs_pkts[i] > 0) {
1095                         num_send = rte_eth_tx_burst(i, bd_tx_q->queue_id, update_bufs[i],
1096                                         update_bufs_pkts[i]);
1097                         for (j = num_send; j < update_bufs_pkts[i]; j++) {
1098                                 rte_pktmbuf_free(update_bufs[i][j]);
1099                         }
1100 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
1101                         for (j = 0; j < update_bufs_pkts[i]; j++) {
1102                                 eth_h = rte_pktmbuf_mtod(update_bufs[i][j],
1103                                                         struct rte_ether_hdr *);
1104                                 mode6_debug("TX ARPupd:", eth_h, i, &burstnumberTX);
1105                         }
1106 #endif
1107                 }
1108         }
1109
1110         /* Send non-ARP packets using tlb policy */
1111         if (slave_bufs_pkts[RTE_MAX_ETHPORTS] > 0) {
1112                 num_send = bond_ethdev_tx_burst_tlb(queue,
1113                                 slave_bufs[RTE_MAX_ETHPORTS],
1114                                 slave_bufs_pkts[RTE_MAX_ETHPORTS]);
1115
1116                 for (j = 0; j < slave_bufs_pkts[RTE_MAX_ETHPORTS]; j++) {
1117                         bufs[nb_pkts - 1 - num_not_send - j] =
1118                                         slave_bufs[RTE_MAX_ETHPORTS][nb_pkts - 1 - j];
1119                 }
1120
1121                 num_tx_total += num_send;
1122         }
1123
1124         return num_tx_total;
1125 }
1126
1127 static inline uint16_t
1128 tx_burst_balance(void *queue, struct rte_mbuf **bufs, uint16_t nb_bufs,
1129                  uint16_t *slave_port_ids, uint16_t slave_count)
1130 {
1131         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
1132         struct bond_dev_private *internals = bd_tx_q->dev_private;
1133
1134         /* Array to sort mbufs for transmission on each slave into */
1135         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_bufs];
1136         /* Number of mbufs for transmission on each slave */
1137         uint16_t slave_nb_bufs[RTE_MAX_ETHPORTS] = { 0 };
1138         /* Mapping array generated by hash function to map mbufs to slaves */
1139         uint16_t bufs_slave_port_idxs[nb_bufs];
1140
1141         uint16_t slave_tx_count;
1142         uint16_t total_tx_count = 0, total_tx_fail_count = 0;
1143
1144         uint16_t i;
1145
1146         /*
1147          * Populate slaves mbuf with the packets which are to be sent on it
1148          * selecting output slave using hash based on xmit policy
1149          */
1150         internals->burst_xmit_hash(bufs, nb_bufs, slave_count,
1151                         bufs_slave_port_idxs);
1152
1153         for (i = 0; i < nb_bufs; i++) {
1154                 /* Populate slave mbuf arrays with mbufs for that slave. */
1155                 uint16_t slave_idx = bufs_slave_port_idxs[i];
1156
1157                 slave_bufs[slave_idx][slave_nb_bufs[slave_idx]++] = bufs[i];
1158         }
1159
1160         /* Send packet burst on each slave device */
1161         for (i = 0; i < slave_count; i++) {
1162                 if (slave_nb_bufs[i] == 0)
1163                         continue;
1164
1165                 slave_tx_count = rte_eth_tx_burst(slave_port_ids[i],
1166                                 bd_tx_q->queue_id, slave_bufs[i],
1167                                 slave_nb_bufs[i]);
1168
1169                 total_tx_count += slave_tx_count;
1170
1171                 /* If tx burst fails move packets to end of bufs */
1172                 if (unlikely(slave_tx_count < slave_nb_bufs[i])) {
1173                         int slave_tx_fail_count = slave_nb_bufs[i] -
1174                                         slave_tx_count;
1175                         total_tx_fail_count += slave_tx_fail_count;
1176                         memcpy(&bufs[nb_bufs - total_tx_fail_count],
1177                                &slave_bufs[i][slave_tx_count],
1178                                slave_tx_fail_count * sizeof(bufs[0]));
1179                 }
1180         }
1181
1182         return total_tx_count;
1183 }
1184
1185 static uint16_t
1186 bond_ethdev_tx_burst_balance(void *queue, struct rte_mbuf **bufs,
1187                 uint16_t nb_bufs)
1188 {
1189         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
1190         struct bond_dev_private *internals = bd_tx_q->dev_private;
1191
1192         uint16_t slave_port_ids[RTE_MAX_ETHPORTS];
1193         uint16_t slave_count;
1194
1195         if (unlikely(nb_bufs == 0))
1196                 return 0;
1197
1198         /* Copy slave list to protect against slave up/down changes during tx
1199          * bursting
1200          */
1201         slave_count = internals->active_slave_count;
1202         if (unlikely(slave_count < 1))
1203                 return 0;
1204
1205         memcpy(slave_port_ids, internals->active_slaves,
1206                         sizeof(slave_port_ids[0]) * slave_count);
1207         return tx_burst_balance(queue, bufs, nb_bufs, slave_port_ids,
1208                                 slave_count);
1209 }
1210
1211 static inline uint16_t
1212 tx_burst_8023ad(void *queue, struct rte_mbuf **bufs, uint16_t nb_bufs,
1213                 bool dedicated_txq)
1214 {
1215         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
1216         struct bond_dev_private *internals = bd_tx_q->dev_private;
1217
1218         uint16_t slave_port_ids[RTE_MAX_ETHPORTS];
1219         uint16_t slave_count;
1220
1221         uint16_t dist_slave_port_ids[RTE_MAX_ETHPORTS];
1222         uint16_t dist_slave_count;
1223
1224         uint16_t slave_tx_count;
1225
1226         uint16_t i;
1227
1228         /* Copy slave list to protect against slave up/down changes during tx
1229          * bursting */
1230         slave_count = internals->active_slave_count;
1231         if (unlikely(slave_count < 1))
1232                 return 0;
1233
1234         memcpy(slave_port_ids, internals->active_slaves,
1235                         sizeof(slave_port_ids[0]) * slave_count);
1236
1237         if (dedicated_txq)
1238                 goto skip_tx_ring;
1239
1240         /* Check for LACP control packets and send if available */
1241         for (i = 0; i < slave_count; i++) {
1242                 struct port *port = &bond_mode_8023ad_ports[slave_port_ids[i]];
1243                 struct rte_mbuf *ctrl_pkt = NULL;
1244
1245                 if (likely(rte_ring_empty(port->tx_ring)))
1246                         continue;
1247
1248                 if (rte_ring_dequeue(port->tx_ring,
1249                                      (void **)&ctrl_pkt) != -ENOENT) {
1250                         slave_tx_count = rte_eth_tx_burst(slave_port_ids[i],
1251                                         bd_tx_q->queue_id, &ctrl_pkt, 1);
1252                         /*
1253                          * re-enqueue LAG control plane packets to buffering
1254                          * ring if transmission fails so the packet isn't lost.
1255                          */
1256                         if (slave_tx_count != 1)
1257                                 rte_ring_enqueue(port->tx_ring, ctrl_pkt);
1258                 }
1259         }
1260
1261 skip_tx_ring:
1262         if (unlikely(nb_bufs == 0))
1263                 return 0;
1264
1265         dist_slave_count = 0;
1266         for (i = 0; i < slave_count; i++) {
1267                 struct port *port = &bond_mode_8023ad_ports[slave_port_ids[i]];
1268
1269                 if (ACTOR_STATE(port, DISTRIBUTING))
1270                         dist_slave_port_ids[dist_slave_count++] =
1271                                         slave_port_ids[i];
1272         }
1273
1274         if (unlikely(dist_slave_count < 1))
1275                 return 0;
1276
1277         return tx_burst_balance(queue, bufs, nb_bufs, dist_slave_port_ids,
1278                                 dist_slave_count);
1279 }
1280
1281 static uint16_t
1282 bond_ethdev_tx_burst_8023ad(void *queue, struct rte_mbuf **bufs,
1283                 uint16_t nb_bufs)
1284 {
1285         return tx_burst_8023ad(queue, bufs, nb_bufs, false);
1286 }
1287
1288 static uint16_t
1289 bond_ethdev_tx_burst_8023ad_fast_queue(void *queue, struct rte_mbuf **bufs,
1290                 uint16_t nb_bufs)
1291 {
1292         return tx_burst_8023ad(queue, bufs, nb_bufs, true);
1293 }
1294
1295 static uint16_t
1296 bond_ethdev_tx_burst_broadcast(void *queue, struct rte_mbuf **bufs,
1297                 uint16_t nb_pkts)
1298 {
1299         struct bond_dev_private *internals;
1300         struct bond_tx_queue *bd_tx_q;
1301
1302         uint16_t slaves[RTE_MAX_ETHPORTS];
1303         uint8_t tx_failed_flag = 0;
1304         uint16_t num_of_slaves;
1305
1306         uint16_t max_nb_of_tx_pkts = 0;
1307
1308         int slave_tx_total[RTE_MAX_ETHPORTS];
1309         int i, most_successful_tx_slave = -1;
1310
1311         bd_tx_q = (struct bond_tx_queue *)queue;
1312         internals = bd_tx_q->dev_private;
1313
1314         /* Copy slave list to protect against slave up/down changes during tx
1315          * bursting */
1316         num_of_slaves = internals->active_slave_count;
1317         memcpy(slaves, internals->active_slaves,
1318                         sizeof(internals->active_slaves[0]) * num_of_slaves);
1319
1320         if (num_of_slaves < 1)
1321                 return 0;
1322
1323         /* Increment reference count on mbufs */
1324         for (i = 0; i < nb_pkts; i++)
1325                 rte_mbuf_refcnt_update(bufs[i], num_of_slaves - 1);
1326
1327         /* Transmit burst on each active slave */
1328         for (i = 0; i < num_of_slaves; i++) {
1329                 slave_tx_total[i] = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
1330                                         bufs, nb_pkts);
1331
1332                 if (unlikely(slave_tx_total[i] < nb_pkts))
1333                         tx_failed_flag = 1;
1334
1335                 /* record the value and slave index for the slave which transmits the
1336                  * maximum number of packets */
1337                 if (slave_tx_total[i] > max_nb_of_tx_pkts) {
1338                         max_nb_of_tx_pkts = slave_tx_total[i];
1339                         most_successful_tx_slave = i;
1340                 }
1341         }
1342
1343         /* if slaves fail to transmit packets from burst, the calling application
1344          * is not expected to know about multiple references to packets so we must
1345          * handle failures of all packets except those of the most successful slave
1346          */
1347         if (unlikely(tx_failed_flag))
1348                 for (i = 0; i < num_of_slaves; i++)
1349                         if (i != most_successful_tx_slave)
1350                                 while (slave_tx_total[i] < nb_pkts)
1351                                         rte_pktmbuf_free(bufs[slave_tx_total[i]++]);
1352
1353         return max_nb_of_tx_pkts;
1354 }
1355
1356 static void
1357 link_properties_set(struct rte_eth_dev *ethdev, struct rte_eth_link *slave_link)
1358 {
1359         struct bond_dev_private *bond_ctx = ethdev->data->dev_private;
1360
1361         if (bond_ctx->mode == BONDING_MODE_8023AD) {
1362                 /**
1363                  * If in mode 4 then save the link properties of the first
1364                  * slave, all subsequent slaves must match these properties
1365                  */
1366                 struct rte_eth_link *bond_link = &bond_ctx->mode4.slave_link;
1367
1368                 bond_link->link_autoneg = slave_link->link_autoneg;
1369                 bond_link->link_duplex = slave_link->link_duplex;
1370                 bond_link->link_speed = slave_link->link_speed;
1371         } else {
1372                 /**
1373                  * In any other mode the link properties are set to default
1374                  * values of AUTONEG/DUPLEX
1375                  */
1376                 ethdev->data->dev_link.link_autoneg = ETH_LINK_AUTONEG;
1377                 ethdev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
1378         }
1379 }
1380
1381 static int
1382 link_properties_valid(struct rte_eth_dev *ethdev,
1383                 struct rte_eth_link *slave_link)
1384 {
1385         struct bond_dev_private *bond_ctx = ethdev->data->dev_private;
1386
1387         if (bond_ctx->mode == BONDING_MODE_8023AD) {
1388                 struct rte_eth_link *bond_link = &bond_ctx->mode4.slave_link;
1389
1390                 if (bond_link->link_duplex != slave_link->link_duplex ||
1391                         bond_link->link_autoneg != slave_link->link_autoneg ||
1392                         bond_link->link_speed != slave_link->link_speed)
1393                         return -1;
1394         }
1395
1396         return 0;
1397 }
1398
1399 int
1400 mac_address_get(struct rte_eth_dev *eth_dev,
1401                 struct rte_ether_addr *dst_mac_addr)
1402 {
1403         struct rte_ether_addr *mac_addr;
1404
1405         if (eth_dev == NULL) {
1406                 RTE_BOND_LOG(ERR, "NULL pointer eth_dev specified");
1407                 return -1;
1408         }
1409
1410         if (dst_mac_addr == NULL) {
1411                 RTE_BOND_LOG(ERR, "NULL pointer MAC specified");
1412                 return -1;
1413         }
1414
1415         mac_addr = eth_dev->data->mac_addrs;
1416
1417         rte_ether_addr_copy(mac_addr, dst_mac_addr);
1418         return 0;
1419 }
1420
1421 int
1422 mac_address_set(struct rte_eth_dev *eth_dev,
1423                 struct rte_ether_addr *new_mac_addr)
1424 {
1425         struct rte_ether_addr *mac_addr;
1426
1427         if (eth_dev == NULL) {
1428                 RTE_BOND_LOG(ERR, "NULL pointer eth_dev specified");
1429                 return -1;
1430         }
1431
1432         if (new_mac_addr == NULL) {
1433                 RTE_BOND_LOG(ERR, "NULL pointer MAC specified");
1434                 return -1;
1435         }
1436
1437         mac_addr = eth_dev->data->mac_addrs;
1438
1439         /* If new MAC is different to current MAC then update */
1440         if (memcmp(mac_addr, new_mac_addr, sizeof(*mac_addr)) != 0)
1441                 memcpy(mac_addr, new_mac_addr, sizeof(*mac_addr));
1442
1443         return 0;
1444 }
1445
1446 static const struct rte_ether_addr null_mac_addr;
1447
1448 /*
1449  * Add additional MAC addresses to the slave
1450  */
1451 int
1452 slave_add_mac_addresses(struct rte_eth_dev *bonded_eth_dev,
1453                 uint16_t slave_port_id)
1454 {
1455         int i, ret;
1456         struct rte_ether_addr *mac_addr;
1457
1458         for (i = 1; i < BOND_MAX_MAC_ADDRS; i++) {
1459                 mac_addr = &bonded_eth_dev->data->mac_addrs[i];
1460                 if (rte_is_same_ether_addr(mac_addr, &null_mac_addr))
1461                         break;
1462
1463                 ret = rte_eth_dev_mac_addr_add(slave_port_id, mac_addr, 0);
1464                 if (ret < 0) {
1465                         /* rollback */
1466                         for (i--; i > 0; i--)
1467                                 rte_eth_dev_mac_addr_remove(slave_port_id,
1468                                         &bonded_eth_dev->data->mac_addrs[i]);
1469                         return ret;
1470                 }
1471         }
1472
1473         return 0;
1474 }
1475
1476 /*
1477  * Remove additional MAC addresses from the slave
1478  */
1479 int
1480 slave_remove_mac_addresses(struct rte_eth_dev *bonded_eth_dev,
1481                 uint16_t slave_port_id)
1482 {
1483         int i, rc, ret;
1484         struct rte_ether_addr *mac_addr;
1485
1486         rc = 0;
1487         for (i = 1; i < BOND_MAX_MAC_ADDRS; i++) {
1488                 mac_addr = &bonded_eth_dev->data->mac_addrs[i];
1489                 if (rte_is_same_ether_addr(mac_addr, &null_mac_addr))
1490                         break;
1491
1492                 ret = rte_eth_dev_mac_addr_remove(slave_port_id, mac_addr);
1493                 /* save only the first error */
1494                 if (ret < 0 && rc == 0)
1495                         rc = ret;
1496         }
1497
1498         return rc;
1499 }
1500
1501 int
1502 mac_address_slaves_update(struct rte_eth_dev *bonded_eth_dev)
1503 {
1504         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1505         bool set;
1506         int i;
1507
1508         /* Update slave devices MAC addresses */
1509         if (internals->slave_count < 1)
1510                 return -1;
1511
1512         switch (internals->mode) {
1513         case BONDING_MODE_ROUND_ROBIN:
1514         case BONDING_MODE_BALANCE:
1515         case BONDING_MODE_BROADCAST:
1516                 for (i = 0; i < internals->slave_count; i++) {
1517                         if (rte_eth_dev_default_mac_addr_set(
1518                                         internals->slaves[i].port_id,
1519                                         bonded_eth_dev->data->mac_addrs)) {
1520                                 RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1521                                                 internals->slaves[i].port_id);
1522                                 return -1;
1523                         }
1524                 }
1525                 break;
1526         case BONDING_MODE_8023AD:
1527                 bond_mode_8023ad_mac_address_update(bonded_eth_dev);
1528                 break;
1529         case BONDING_MODE_ACTIVE_BACKUP:
1530         case BONDING_MODE_TLB:
1531         case BONDING_MODE_ALB:
1532         default:
1533                 set = true;
1534                 for (i = 0; i < internals->slave_count; i++) {
1535                         if (internals->slaves[i].port_id ==
1536                                         internals->current_primary_port) {
1537                                 if (rte_eth_dev_default_mac_addr_set(
1538                                                 internals->current_primary_port,
1539                                                 bonded_eth_dev->data->mac_addrs)) {
1540                                         RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1541                                                         internals->current_primary_port);
1542                                         set = false;
1543                                 }
1544                         } else {
1545                                 if (rte_eth_dev_default_mac_addr_set(
1546                                                 internals->slaves[i].port_id,
1547                                                 &internals->slaves[i].persisted_mac_addr)) {
1548                                         RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1549                                                         internals->slaves[i].port_id);
1550                                 }
1551                         }
1552                 }
1553                 if (!set)
1554                         return -1;
1555         }
1556
1557         return 0;
1558 }
1559
1560 int
1561 bond_ethdev_mode_set(struct rte_eth_dev *eth_dev, int mode)
1562 {
1563         struct bond_dev_private *internals;
1564
1565         internals = eth_dev->data->dev_private;
1566
1567         switch (mode) {
1568         case BONDING_MODE_ROUND_ROBIN:
1569                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_round_robin;
1570                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1571                 break;
1572         case BONDING_MODE_ACTIVE_BACKUP:
1573                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_active_backup;
1574                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup;
1575                 break;
1576         case BONDING_MODE_BALANCE:
1577                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_balance;
1578                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1579                 break;
1580         case BONDING_MODE_BROADCAST:
1581                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_broadcast;
1582                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1583                 break;
1584         case BONDING_MODE_8023AD:
1585                 if (bond_mode_8023ad_enable(eth_dev) != 0)
1586                         return -1;
1587
1588                 if (internals->mode4.dedicated_queues.enabled == 0) {
1589                         eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_8023ad;
1590                         eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_8023ad;
1591                         RTE_BOND_LOG(WARNING,
1592                                 "Using mode 4, it is necessary to do TX burst "
1593                                 "and RX burst at least every 100ms.");
1594                 } else {
1595                         /* Use flow director's optimization */
1596                         eth_dev->rx_pkt_burst =
1597                                         bond_ethdev_rx_burst_8023ad_fast_queue;
1598                         eth_dev->tx_pkt_burst =
1599                                         bond_ethdev_tx_burst_8023ad_fast_queue;
1600                 }
1601                 break;
1602         case BONDING_MODE_TLB:
1603                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_tlb;
1604                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup;
1605                 break;
1606         case BONDING_MODE_ALB:
1607                 if (bond_mode_alb_enable(eth_dev) != 0)
1608                         return -1;
1609
1610                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_alb;
1611                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_alb;
1612                 break;
1613         default:
1614                 return -1;
1615         }
1616
1617         internals->mode = mode;
1618
1619         return 0;
1620 }
1621
1622
1623 static int
1624 slave_configure_slow_queue(struct rte_eth_dev *bonded_eth_dev,
1625                 struct rte_eth_dev *slave_eth_dev)
1626 {
1627         int errval = 0;
1628         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1629         struct port *port = &bond_mode_8023ad_ports[slave_eth_dev->data->port_id];
1630
1631         if (port->slow_pool == NULL) {
1632                 char mem_name[256];
1633                 int slave_id = slave_eth_dev->data->port_id;
1634
1635                 snprintf(mem_name, RTE_DIM(mem_name), "slave_port%u_slow_pool",
1636                                 slave_id);
1637                 port->slow_pool = rte_pktmbuf_pool_create(mem_name, 8191,
1638                         250, 0, RTE_MBUF_DEFAULT_BUF_SIZE,
1639                         slave_eth_dev->data->numa_node);
1640
1641                 /* Any memory allocation failure in initialization is critical because
1642                  * resources can't be free, so reinitialization is impossible. */
1643                 if (port->slow_pool == NULL) {
1644                         rte_panic("Slave %u: Failed to create memory pool '%s': %s\n",
1645                                 slave_id, mem_name, rte_strerror(rte_errno));
1646                 }
1647         }
1648
1649         if (internals->mode4.dedicated_queues.enabled == 1) {
1650                 /* Configure slow Rx queue */
1651
1652                 errval = rte_eth_rx_queue_setup(slave_eth_dev->data->port_id,
1653                                 internals->mode4.dedicated_queues.rx_qid, 128,
1654                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1655                                 NULL, port->slow_pool);
1656                 if (errval != 0) {
1657                         RTE_BOND_LOG(ERR,
1658                                         "rte_eth_rx_queue_setup: port=%d queue_id %d, err (%d)",
1659                                         slave_eth_dev->data->port_id,
1660                                         internals->mode4.dedicated_queues.rx_qid,
1661                                         errval);
1662                         return errval;
1663                 }
1664
1665                 errval = rte_eth_tx_queue_setup(slave_eth_dev->data->port_id,
1666                                 internals->mode4.dedicated_queues.tx_qid, 512,
1667                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1668                                 NULL);
1669                 if (errval != 0) {
1670                         RTE_BOND_LOG(ERR,
1671                                 "rte_eth_tx_queue_setup: port=%d queue_id %d, err (%d)",
1672                                 slave_eth_dev->data->port_id,
1673                                 internals->mode4.dedicated_queues.tx_qid,
1674                                 errval);
1675                         return errval;
1676                 }
1677         }
1678         return 0;
1679 }
1680
1681 int
1682 slave_configure(struct rte_eth_dev *bonded_eth_dev,
1683                 struct rte_eth_dev *slave_eth_dev)
1684 {
1685         struct bond_rx_queue *bd_rx_q;
1686         struct bond_tx_queue *bd_tx_q;
1687         uint16_t nb_rx_queues;
1688         uint16_t nb_tx_queues;
1689
1690         int errval;
1691         uint16_t q_id;
1692         struct rte_flow_error flow_error;
1693
1694         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1695
1696         /* Stop slave */
1697         errval = rte_eth_dev_stop(slave_eth_dev->data->port_id);
1698         if (errval != 0)
1699                 RTE_BOND_LOG(ERR, "rte_eth_dev_stop: port %u, err (%d)",
1700                              slave_eth_dev->data->port_id, errval);
1701
1702         /* Enable interrupts on slave device if supported */
1703         if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1704                 slave_eth_dev->data->dev_conf.intr_conf.lsc = 1;
1705
1706         /* If RSS is enabled for bonding, try to enable it for slaves  */
1707         if (bonded_eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG) {
1708                 if (internals->rss_key_len != 0) {
1709                         slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len =
1710                                         internals->rss_key_len;
1711                         slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key =
1712                                         internals->rss_key;
1713                 } else {
1714                         slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
1715                 }
1716
1717                 slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf =
1718                                 bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
1719                 slave_eth_dev->data->dev_conf.rxmode.mq_mode =
1720                                 bonded_eth_dev->data->dev_conf.rxmode.mq_mode;
1721         }
1722
1723         if (bonded_eth_dev->data->dev_conf.rxmode.offloads &
1724                         DEV_RX_OFFLOAD_VLAN_FILTER)
1725                 slave_eth_dev->data->dev_conf.rxmode.offloads |=
1726                                 DEV_RX_OFFLOAD_VLAN_FILTER;
1727         else
1728                 slave_eth_dev->data->dev_conf.rxmode.offloads &=
1729                                 ~DEV_RX_OFFLOAD_VLAN_FILTER;
1730
1731         slave_eth_dev->data->dev_conf.rxmode.max_rx_pkt_len =
1732                         bonded_eth_dev->data->dev_conf.rxmode.max_rx_pkt_len;
1733
1734         if (bonded_eth_dev->data->dev_conf.rxmode.offloads &
1735                         DEV_RX_OFFLOAD_JUMBO_FRAME)
1736                 slave_eth_dev->data->dev_conf.rxmode.offloads |=
1737                                 DEV_RX_OFFLOAD_JUMBO_FRAME;
1738         else
1739                 slave_eth_dev->data->dev_conf.rxmode.offloads &=
1740                                 ~DEV_RX_OFFLOAD_JUMBO_FRAME;
1741
1742         nb_rx_queues = bonded_eth_dev->data->nb_rx_queues;
1743         nb_tx_queues = bonded_eth_dev->data->nb_tx_queues;
1744
1745         if (internals->mode == BONDING_MODE_8023AD) {
1746                 if (internals->mode4.dedicated_queues.enabled == 1) {
1747                         nb_rx_queues++;
1748                         nb_tx_queues++;
1749                 }
1750         }
1751
1752         errval = rte_eth_dev_set_mtu(slave_eth_dev->data->port_id,
1753                                      bonded_eth_dev->data->mtu);
1754         if (errval != 0 && errval != -ENOTSUP) {
1755                 RTE_BOND_LOG(ERR, "rte_eth_dev_set_mtu: port %u, err (%d)",
1756                                 slave_eth_dev->data->port_id, errval);
1757                 return errval;
1758         }
1759
1760         /* Configure device */
1761         errval = rte_eth_dev_configure(slave_eth_dev->data->port_id,
1762                         nb_rx_queues, nb_tx_queues,
1763                         &(slave_eth_dev->data->dev_conf));
1764         if (errval != 0) {
1765                 RTE_BOND_LOG(ERR, "Cannot configure slave device: port %u, err (%d)",
1766                                 slave_eth_dev->data->port_id, errval);
1767                 return errval;
1768         }
1769
1770         /* Setup Rx Queues */
1771         for (q_id = 0; q_id < bonded_eth_dev->data->nb_rx_queues; q_id++) {
1772                 bd_rx_q = (struct bond_rx_queue *)bonded_eth_dev->data->rx_queues[q_id];
1773
1774                 errval = rte_eth_rx_queue_setup(slave_eth_dev->data->port_id, q_id,
1775                                 bd_rx_q->nb_rx_desc,
1776                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1777                                 &(bd_rx_q->rx_conf), bd_rx_q->mb_pool);
1778                 if (errval != 0) {
1779                         RTE_BOND_LOG(ERR,
1780                                         "rte_eth_rx_queue_setup: port=%d queue_id %d, err (%d)",
1781                                         slave_eth_dev->data->port_id, q_id, errval);
1782                         return errval;
1783                 }
1784         }
1785
1786         /* Setup Tx Queues */
1787         for (q_id = 0; q_id < bonded_eth_dev->data->nb_tx_queues; q_id++) {
1788                 bd_tx_q = (struct bond_tx_queue *)bonded_eth_dev->data->tx_queues[q_id];
1789
1790                 errval = rte_eth_tx_queue_setup(slave_eth_dev->data->port_id, q_id,
1791                                 bd_tx_q->nb_tx_desc,
1792                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1793                                 &bd_tx_q->tx_conf);
1794                 if (errval != 0) {
1795                         RTE_BOND_LOG(ERR,
1796                                 "rte_eth_tx_queue_setup: port=%d queue_id %d, err (%d)",
1797                                 slave_eth_dev->data->port_id, q_id, errval);
1798                         return errval;
1799                 }
1800         }
1801
1802         if (internals->mode == BONDING_MODE_8023AD &&
1803                         internals->mode4.dedicated_queues.enabled == 1) {
1804                 if (slave_configure_slow_queue(bonded_eth_dev, slave_eth_dev)
1805                                 != 0)
1806                         return errval;
1807
1808                 errval = bond_ethdev_8023ad_flow_verify(bonded_eth_dev,
1809                                 slave_eth_dev->data->port_id);
1810                 if (errval != 0) {
1811                         RTE_BOND_LOG(ERR,
1812                                 "bond_ethdev_8023ad_flow_verify: port=%d, err (%d)",
1813                                 slave_eth_dev->data->port_id, errval);
1814                         return errval;
1815                 }
1816
1817                 if (internals->mode4.dedicated_queues.flow[slave_eth_dev->data->port_id] != NULL)
1818                         rte_flow_destroy(slave_eth_dev->data->port_id,
1819                                         internals->mode4.dedicated_queues.flow[slave_eth_dev->data->port_id],
1820                                         &flow_error);
1821
1822                 bond_ethdev_8023ad_flow_set(bonded_eth_dev,
1823                                 slave_eth_dev->data->port_id);
1824         }
1825
1826         /* Start device */
1827         errval = rte_eth_dev_start(slave_eth_dev->data->port_id);
1828         if (errval != 0) {
1829                 RTE_BOND_LOG(ERR, "rte_eth_dev_start: port=%u, err (%d)",
1830                                 slave_eth_dev->data->port_id, errval);
1831                 return -1;
1832         }
1833
1834         /* If RSS is enabled for bonding, synchronize RETA */
1835         if (bonded_eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS) {
1836                 int i;
1837                 struct bond_dev_private *internals;
1838
1839                 internals = bonded_eth_dev->data->dev_private;
1840
1841                 for (i = 0; i < internals->slave_count; i++) {
1842                         if (internals->slaves[i].port_id == slave_eth_dev->data->port_id) {
1843                                 errval = rte_eth_dev_rss_reta_update(
1844                                                 slave_eth_dev->data->port_id,
1845                                                 &internals->reta_conf[0],
1846                                                 internals->slaves[i].reta_size);
1847                                 if (errval != 0) {
1848                                         RTE_BOND_LOG(WARNING,
1849                                                      "rte_eth_dev_rss_reta_update on slave port %d fails (err %d)."
1850                                                      " RSS Configuration for bonding may be inconsistent.",
1851                                                      slave_eth_dev->data->port_id, errval);
1852                                 }
1853                                 break;
1854                         }
1855                 }
1856         }
1857
1858         /* If lsc interrupt is set, check initial slave's link status */
1859         if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
1860                 slave_eth_dev->dev_ops->link_update(slave_eth_dev, 0);
1861                 bond_ethdev_lsc_event_callback(slave_eth_dev->data->port_id,
1862                         RTE_ETH_EVENT_INTR_LSC, &bonded_eth_dev->data->port_id,
1863                         NULL);
1864         }
1865
1866         return 0;
1867 }
1868
1869 void
1870 slave_remove(struct bond_dev_private *internals,
1871                 struct rte_eth_dev *slave_eth_dev)
1872 {
1873         uint16_t i;
1874
1875         for (i = 0; i < internals->slave_count; i++)
1876                 if (internals->slaves[i].port_id ==
1877                                 slave_eth_dev->data->port_id)
1878                         break;
1879
1880         if (i < (internals->slave_count - 1)) {
1881                 struct rte_flow *flow;
1882
1883                 memmove(&internals->slaves[i], &internals->slaves[i + 1],
1884                                 sizeof(internals->slaves[0]) *
1885                                 (internals->slave_count - i - 1));
1886                 TAILQ_FOREACH(flow, &internals->flow_list, next) {
1887                         memmove(&flow->flows[i], &flow->flows[i + 1],
1888                                 sizeof(flow->flows[0]) *
1889                                 (internals->slave_count - i - 1));
1890                         flow->flows[internals->slave_count - 1] = NULL;
1891                 }
1892         }
1893
1894         internals->slave_count--;
1895
1896         /* force reconfiguration of slave interfaces */
1897         rte_eth_dev_internal_reset(slave_eth_dev);
1898 }
1899
1900 static void
1901 bond_ethdev_slave_link_status_change_monitor(void *cb_arg);
1902
1903 void
1904 slave_add(struct bond_dev_private *internals,
1905                 struct rte_eth_dev *slave_eth_dev)
1906 {
1907         struct bond_slave_details *slave_details =
1908                         &internals->slaves[internals->slave_count];
1909
1910         slave_details->port_id = slave_eth_dev->data->port_id;
1911         slave_details->last_link_status = 0;
1912
1913         /* Mark slave devices that don't support interrupts so we can
1914          * compensate when we start the bond
1915          */
1916         if (!(slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1917                 slave_details->link_status_poll_enabled = 1;
1918         }
1919
1920         slave_details->link_status_wait_to_complete = 0;
1921         /* clean tlb_last_obytes when adding port for bonding device */
1922         memcpy(&(slave_details->persisted_mac_addr), slave_eth_dev->data->mac_addrs,
1923                         sizeof(struct rte_ether_addr));
1924 }
1925
1926 void
1927 bond_ethdev_primary_set(struct bond_dev_private *internals,
1928                 uint16_t slave_port_id)
1929 {
1930         int i;
1931
1932         if (internals->active_slave_count < 1)
1933                 internals->current_primary_port = slave_port_id;
1934         else
1935                 /* Search bonded device slave ports for new proposed primary port */
1936                 for (i = 0; i < internals->active_slave_count; i++) {
1937                         if (internals->active_slaves[i] == slave_port_id)
1938                                 internals->current_primary_port = slave_port_id;
1939                 }
1940 }
1941
1942 static int
1943 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev);
1944
1945 static int
1946 bond_ethdev_start(struct rte_eth_dev *eth_dev)
1947 {
1948         struct bond_dev_private *internals;
1949         int i;
1950
1951         /* slave eth dev will be started by bonded device */
1952         if (check_for_bonded_ethdev(eth_dev)) {
1953                 RTE_BOND_LOG(ERR, "User tried to explicitly start a slave eth_dev (%d)",
1954                                 eth_dev->data->port_id);
1955                 return -1;
1956         }
1957
1958         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
1959         eth_dev->data->dev_started = 1;
1960
1961         internals = eth_dev->data->dev_private;
1962
1963         if (internals->slave_count == 0) {
1964                 RTE_BOND_LOG(ERR, "Cannot start port since there are no slave devices");
1965                 goto out_err;
1966         }
1967
1968         if (internals->user_defined_mac == 0) {
1969                 struct rte_ether_addr *new_mac_addr = NULL;
1970
1971                 for (i = 0; i < internals->slave_count; i++)
1972                         if (internals->slaves[i].port_id == internals->primary_port)
1973                                 new_mac_addr = &internals->slaves[i].persisted_mac_addr;
1974
1975                 if (new_mac_addr == NULL)
1976                         goto out_err;
1977
1978                 if (mac_address_set(eth_dev, new_mac_addr) != 0) {
1979                         RTE_BOND_LOG(ERR, "bonded port (%d) failed to update MAC address",
1980                                         eth_dev->data->port_id);
1981                         goto out_err;
1982                 }
1983         }
1984
1985         if (internals->mode == BONDING_MODE_8023AD) {
1986                 if (internals->mode4.dedicated_queues.enabled == 1) {
1987                         internals->mode4.dedicated_queues.rx_qid =
1988                                         eth_dev->data->nb_rx_queues;
1989                         internals->mode4.dedicated_queues.tx_qid =
1990                                         eth_dev->data->nb_tx_queues;
1991                 }
1992         }
1993
1994
1995         /* Reconfigure each slave device if starting bonded device */
1996         for (i = 0; i < internals->slave_count; i++) {
1997                 struct rte_eth_dev *slave_ethdev =
1998                                 &(rte_eth_devices[internals->slaves[i].port_id]);
1999                 if (slave_configure(eth_dev, slave_ethdev) != 0) {
2000                         RTE_BOND_LOG(ERR,
2001                                 "bonded port (%d) failed to reconfigure slave device (%d)",
2002                                 eth_dev->data->port_id,
2003                                 internals->slaves[i].port_id);
2004                         goto out_err;
2005                 }
2006                 /* We will need to poll for link status if any slave doesn't
2007                  * support interrupts
2008                  */
2009                 if (internals->slaves[i].link_status_poll_enabled)
2010                         internals->link_status_polling_enabled = 1;
2011         }
2012
2013         /* start polling if needed */
2014         if (internals->link_status_polling_enabled) {
2015                 rte_eal_alarm_set(
2016                         internals->link_status_polling_interval_ms * 1000,
2017                         bond_ethdev_slave_link_status_change_monitor,
2018                         (void *)&rte_eth_devices[internals->port_id]);
2019         }
2020
2021         /* Update all slave devices MACs*/
2022         if (mac_address_slaves_update(eth_dev) != 0)
2023                 goto out_err;
2024
2025         if (internals->user_defined_primary_port)
2026                 bond_ethdev_primary_set(internals, internals->primary_port);
2027
2028         if (internals->mode == BONDING_MODE_8023AD)
2029                 bond_mode_8023ad_start(eth_dev);
2030
2031         if (internals->mode == BONDING_MODE_TLB ||
2032                         internals->mode == BONDING_MODE_ALB)
2033                 bond_tlb_enable(internals);
2034
2035         return 0;
2036
2037 out_err:
2038         eth_dev->data->dev_started = 0;
2039         return -1;
2040 }
2041
2042 static void
2043 bond_ethdev_free_queues(struct rte_eth_dev *dev)
2044 {
2045         uint16_t i;
2046
2047         if (dev->data->rx_queues != NULL) {
2048                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2049                         rte_free(dev->data->rx_queues[i]);
2050                         dev->data->rx_queues[i] = NULL;
2051                 }
2052                 dev->data->nb_rx_queues = 0;
2053         }
2054
2055         if (dev->data->tx_queues != NULL) {
2056                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2057                         rte_free(dev->data->tx_queues[i]);
2058                         dev->data->tx_queues[i] = NULL;
2059                 }
2060                 dev->data->nb_tx_queues = 0;
2061         }
2062 }
2063
2064 int
2065 bond_ethdev_stop(struct rte_eth_dev *eth_dev)
2066 {
2067         struct bond_dev_private *internals = eth_dev->data->dev_private;
2068         uint16_t i;
2069         int ret;
2070
2071         if (internals->mode == BONDING_MODE_8023AD) {
2072                 struct port *port;
2073                 void *pkt = NULL;
2074
2075                 bond_mode_8023ad_stop(eth_dev);
2076
2077                 /* Discard all messages to/from mode 4 state machines */
2078                 for (i = 0; i < internals->active_slave_count; i++) {
2079                         port = &bond_mode_8023ad_ports[internals->active_slaves[i]];
2080
2081                         RTE_ASSERT(port->rx_ring != NULL);
2082                         while (rte_ring_dequeue(port->rx_ring, &pkt) != -ENOENT)
2083                                 rte_pktmbuf_free(pkt);
2084
2085                         RTE_ASSERT(port->tx_ring != NULL);
2086                         while (rte_ring_dequeue(port->tx_ring, &pkt) != -ENOENT)
2087                                 rte_pktmbuf_free(pkt);
2088                 }
2089         }
2090
2091         if (internals->mode == BONDING_MODE_TLB ||
2092                         internals->mode == BONDING_MODE_ALB) {
2093                 bond_tlb_disable(internals);
2094                 for (i = 0; i < internals->active_slave_count; i++)
2095                         tlb_last_obytets[internals->active_slaves[i]] = 0;
2096         }
2097
2098         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
2099         eth_dev->data->dev_started = 0;
2100
2101         internals->link_status_polling_enabled = 0;
2102         for (i = 0; i < internals->slave_count; i++) {
2103                 uint16_t slave_id = internals->slaves[i].port_id;
2104                 if (find_slave_by_id(internals->active_slaves,
2105                                 internals->active_slave_count, slave_id) !=
2106                                                 internals->active_slave_count) {
2107                         internals->slaves[i].last_link_status = 0;
2108                         ret = rte_eth_dev_stop(slave_id);
2109                         if (ret != 0) {
2110                                 RTE_BOND_LOG(ERR, "Failed to stop device on port %u",
2111                                              slave_id);
2112                                 return ret;
2113                         }
2114                         deactivate_slave(eth_dev, slave_id);
2115                 }
2116         }
2117
2118         return 0;
2119 }
2120
2121 int
2122 bond_ethdev_close(struct rte_eth_dev *dev)
2123 {
2124         struct bond_dev_private *internals = dev->data->dev_private;
2125         uint16_t bond_port_id = internals->port_id;
2126         int skipped = 0;
2127         struct rte_flow_error ferror;
2128
2129         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2130                 return 0;
2131
2132         RTE_BOND_LOG(INFO, "Closing bonded device %s", dev->device->name);
2133         while (internals->slave_count != skipped) {
2134                 uint16_t port_id = internals->slaves[skipped].port_id;
2135
2136                 if (rte_eth_dev_stop(port_id) != 0) {
2137                         RTE_BOND_LOG(ERR, "Failed to stop device on port %u",
2138                                      port_id);
2139                         skipped++;
2140                 }
2141
2142                 if (rte_eth_bond_slave_remove(bond_port_id, port_id) != 0) {
2143                         RTE_BOND_LOG(ERR,
2144                                      "Failed to remove port %d from bonded device %s",
2145                                      port_id, dev->device->name);
2146                         skipped++;
2147                 }
2148         }
2149         bond_flow_ops.flush(dev, &ferror);
2150         bond_ethdev_free_queues(dev);
2151         rte_bitmap_reset(internals->vlan_filter_bmp);
2152         rte_bitmap_free(internals->vlan_filter_bmp);
2153         rte_free(internals->vlan_filter_bmpmem);
2154
2155         /* Try to release mempool used in mode6. If the bond
2156          * device is not mode6, free the NULL is not problem.
2157          */
2158         rte_mempool_free(internals->mode6.mempool);
2159
2160         return 0;
2161 }
2162
2163 /* forward declaration */
2164 static int bond_ethdev_configure(struct rte_eth_dev *dev);
2165
2166 static int
2167 bond_ethdev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2168 {
2169         struct bond_dev_private *internals = dev->data->dev_private;
2170         struct bond_slave_details slave;
2171         int ret;
2172
2173         uint16_t max_nb_rx_queues = UINT16_MAX;
2174         uint16_t max_nb_tx_queues = UINT16_MAX;
2175         uint16_t max_rx_desc_lim = UINT16_MAX;
2176         uint16_t max_tx_desc_lim = UINT16_MAX;
2177
2178         dev_info->max_mac_addrs = BOND_MAX_MAC_ADDRS;
2179
2180         dev_info->max_rx_pktlen = internals->candidate_max_rx_pktlen ?
2181                         internals->candidate_max_rx_pktlen :
2182                         RTE_ETHER_MAX_JUMBO_FRAME_LEN;
2183
2184         /* Max number of tx/rx queues that the bonded device can support is the
2185          * minimum values of the bonded slaves, as all slaves must be capable
2186          * of supporting the same number of tx/rx queues.
2187          */
2188         if (internals->slave_count > 0) {
2189                 struct rte_eth_dev_info slave_info;
2190                 uint16_t idx;
2191
2192                 for (idx = 0; idx < internals->slave_count; idx++) {
2193                         slave = internals->slaves[idx];
2194                         ret = rte_eth_dev_info_get(slave.port_id, &slave_info);
2195                         if (ret != 0) {
2196                                 RTE_BOND_LOG(ERR,
2197                                         "%s: Error during getting device (port %u) info: %s\n",
2198                                         __func__,
2199                                         slave.port_id,
2200                                         strerror(-ret));
2201
2202                                 return ret;
2203                         }
2204
2205                         if (slave_info.max_rx_queues < max_nb_rx_queues)
2206                                 max_nb_rx_queues = slave_info.max_rx_queues;
2207
2208                         if (slave_info.max_tx_queues < max_nb_tx_queues)
2209                                 max_nb_tx_queues = slave_info.max_tx_queues;
2210
2211                         if (slave_info.rx_desc_lim.nb_max < max_rx_desc_lim)
2212                                 max_rx_desc_lim = slave_info.rx_desc_lim.nb_max;
2213
2214                         if (slave_info.tx_desc_lim.nb_max < max_tx_desc_lim)
2215                                 max_tx_desc_lim = slave_info.tx_desc_lim.nb_max;
2216                 }
2217         }
2218
2219         dev_info->max_rx_queues = max_nb_rx_queues;
2220         dev_info->max_tx_queues = max_nb_tx_queues;
2221
2222         memcpy(&dev_info->default_rxconf, &internals->default_rxconf,
2223                sizeof(dev_info->default_rxconf));
2224         memcpy(&dev_info->default_txconf, &internals->default_txconf,
2225                sizeof(dev_info->default_txconf));
2226
2227         dev_info->rx_desc_lim.nb_max = max_rx_desc_lim;
2228         dev_info->tx_desc_lim.nb_max = max_tx_desc_lim;
2229
2230         /**
2231          * If dedicated hw queues enabled for link bonding device in LACP mode
2232          * then we need to reduce the maximum number of data path queues by 1.
2233          */
2234         if (internals->mode == BONDING_MODE_8023AD &&
2235                 internals->mode4.dedicated_queues.enabled == 1) {
2236                 dev_info->max_rx_queues--;
2237                 dev_info->max_tx_queues--;
2238         }
2239
2240         dev_info->min_rx_bufsize = 0;
2241
2242         dev_info->rx_offload_capa = internals->rx_offload_capa;
2243         dev_info->tx_offload_capa = internals->tx_offload_capa;
2244         dev_info->rx_queue_offload_capa = internals->rx_queue_offload_capa;
2245         dev_info->tx_queue_offload_capa = internals->tx_queue_offload_capa;
2246         dev_info->flow_type_rss_offloads = internals->flow_type_rss_offloads;
2247
2248         dev_info->reta_size = internals->reta_size;
2249
2250         return 0;
2251 }
2252
2253 static int
2254 bond_ethdev_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
2255 {
2256         int res;
2257         uint16_t i;
2258         struct bond_dev_private *internals = dev->data->dev_private;
2259
2260         /* don't do this while a slave is being added */
2261         rte_spinlock_lock(&internals->lock);
2262
2263         if (on)
2264                 rte_bitmap_set(internals->vlan_filter_bmp, vlan_id);
2265         else
2266                 rte_bitmap_clear(internals->vlan_filter_bmp, vlan_id);
2267
2268         for (i = 0; i < internals->slave_count; i++) {
2269                 uint16_t port_id = internals->slaves[i].port_id;
2270
2271                 res = rte_eth_dev_vlan_filter(port_id, vlan_id, on);
2272                 if (res == ENOTSUP)
2273                         RTE_BOND_LOG(WARNING,
2274                                      "Setting VLAN filter on slave port %u not supported.",
2275                                      port_id);
2276         }
2277
2278         rte_spinlock_unlock(&internals->lock);
2279         return 0;
2280 }
2281
2282 static int
2283 bond_ethdev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
2284                 uint16_t nb_rx_desc, unsigned int socket_id __rte_unused,
2285                 const struct rte_eth_rxconf *rx_conf, struct rte_mempool *mb_pool)
2286 {
2287         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)
2288                         rte_zmalloc_socket(NULL, sizeof(struct bond_rx_queue),
2289                                         0, dev->data->numa_node);
2290         if (bd_rx_q == NULL)
2291                 return -1;
2292
2293         bd_rx_q->queue_id = rx_queue_id;
2294         bd_rx_q->dev_private = dev->data->dev_private;
2295
2296         bd_rx_q->nb_rx_desc = nb_rx_desc;
2297
2298         memcpy(&(bd_rx_q->rx_conf), rx_conf, sizeof(struct rte_eth_rxconf));
2299         bd_rx_q->mb_pool = mb_pool;
2300
2301         dev->data->rx_queues[rx_queue_id] = bd_rx_q;
2302
2303         return 0;
2304 }
2305
2306 static int
2307 bond_ethdev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
2308                 uint16_t nb_tx_desc, unsigned int socket_id __rte_unused,
2309                 const struct rte_eth_txconf *tx_conf)
2310 {
2311         struct bond_tx_queue *bd_tx_q  = (struct bond_tx_queue *)
2312                         rte_zmalloc_socket(NULL, sizeof(struct bond_tx_queue),
2313                                         0, dev->data->numa_node);
2314
2315         if (bd_tx_q == NULL)
2316                 return -1;
2317
2318         bd_tx_q->queue_id = tx_queue_id;
2319         bd_tx_q->dev_private = dev->data->dev_private;
2320
2321         bd_tx_q->nb_tx_desc = nb_tx_desc;
2322         memcpy(&(bd_tx_q->tx_conf), tx_conf, sizeof(bd_tx_q->tx_conf));
2323
2324         dev->data->tx_queues[tx_queue_id] = bd_tx_q;
2325
2326         return 0;
2327 }
2328
2329 static void
2330 bond_ethdev_rx_queue_release(void *queue)
2331 {
2332         if (queue == NULL)
2333                 return;
2334
2335         rte_free(queue);
2336 }
2337
2338 static void
2339 bond_ethdev_tx_queue_release(void *queue)
2340 {
2341         if (queue == NULL)
2342                 return;
2343
2344         rte_free(queue);
2345 }
2346
2347 static void
2348 bond_ethdev_slave_link_status_change_monitor(void *cb_arg)
2349 {
2350         struct rte_eth_dev *bonded_ethdev, *slave_ethdev;
2351         struct bond_dev_private *internals;
2352
2353         /* Default value for polling slave found is true as we don't want to
2354          * disable the polling thread if we cannot get the lock */
2355         int i, polling_slave_found = 1;
2356
2357         if (cb_arg == NULL)
2358                 return;
2359
2360         bonded_ethdev = cb_arg;
2361         internals = bonded_ethdev->data->dev_private;
2362
2363         if (!bonded_ethdev->data->dev_started ||
2364                 !internals->link_status_polling_enabled)
2365                 return;
2366
2367         /* If device is currently being configured then don't check slaves link
2368          * status, wait until next period */
2369         if (rte_spinlock_trylock(&internals->lock)) {
2370                 if (internals->slave_count > 0)
2371                         polling_slave_found = 0;
2372
2373                 for (i = 0; i < internals->slave_count; i++) {
2374                         if (!internals->slaves[i].link_status_poll_enabled)
2375                                 continue;
2376
2377                         slave_ethdev = &rte_eth_devices[internals->slaves[i].port_id];
2378                         polling_slave_found = 1;
2379
2380                         /* Update slave link status */
2381                         (*slave_ethdev->dev_ops->link_update)(slave_ethdev,
2382                                         internals->slaves[i].link_status_wait_to_complete);
2383
2384                         /* if link status has changed since last checked then call lsc
2385                          * event callback */
2386                         if (slave_ethdev->data->dev_link.link_status !=
2387                                         internals->slaves[i].last_link_status) {
2388                                 internals->slaves[i].last_link_status =
2389                                                 slave_ethdev->data->dev_link.link_status;
2390
2391                                 bond_ethdev_lsc_event_callback(internals->slaves[i].port_id,
2392                                                 RTE_ETH_EVENT_INTR_LSC,
2393                                                 &bonded_ethdev->data->port_id,
2394                                                 NULL);
2395                         }
2396                 }
2397                 rte_spinlock_unlock(&internals->lock);
2398         }
2399
2400         if (polling_slave_found)
2401                 /* Set alarm to continue monitoring link status of slave ethdev's */
2402                 rte_eal_alarm_set(internals->link_status_polling_interval_ms * 1000,
2403                                 bond_ethdev_slave_link_status_change_monitor, cb_arg);
2404 }
2405
2406 static int
2407 bond_ethdev_link_update(struct rte_eth_dev *ethdev, int wait_to_complete)
2408 {
2409         int (*link_update)(uint16_t port_id, struct rte_eth_link *eth_link);
2410
2411         struct bond_dev_private *bond_ctx;
2412         struct rte_eth_link slave_link;
2413
2414         bool one_link_update_succeeded;
2415         uint32_t idx;
2416         int ret;
2417
2418         bond_ctx = ethdev->data->dev_private;
2419
2420         ethdev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
2421
2422         if (ethdev->data->dev_started == 0 ||
2423                         bond_ctx->active_slave_count == 0) {
2424                 ethdev->data->dev_link.link_status = ETH_LINK_DOWN;
2425                 return 0;
2426         }
2427
2428         ethdev->data->dev_link.link_status = ETH_LINK_UP;
2429
2430         if (wait_to_complete)
2431                 link_update = rte_eth_link_get;
2432         else
2433                 link_update = rte_eth_link_get_nowait;
2434
2435         switch (bond_ctx->mode) {
2436         case BONDING_MODE_BROADCAST:
2437                 /**
2438                  * Setting link speed to UINT32_MAX to ensure we pick up the
2439                  * value of the first active slave
2440                  */
2441                 ethdev->data->dev_link.link_speed = UINT32_MAX;
2442
2443                 /**
2444                  * link speed is minimum value of all the slaves link speed as
2445                  * packet loss will occur on this slave if transmission at rates
2446                  * greater than this are attempted
2447                  */
2448                 for (idx = 0; idx < bond_ctx->active_slave_count; idx++) {
2449                         ret = link_update(bond_ctx->active_slaves[idx],
2450                                           &slave_link);
2451                         if (ret < 0) {
2452                                 ethdev->data->dev_link.link_speed =
2453                                         ETH_SPEED_NUM_NONE;
2454                                 RTE_BOND_LOG(ERR,
2455                                         "Slave (port %u) link get failed: %s",
2456                                         bond_ctx->active_slaves[idx],
2457                                         rte_strerror(-ret));
2458                                 return 0;
2459                         }
2460
2461                         if (slave_link.link_speed <
2462                                         ethdev->data->dev_link.link_speed)
2463                                 ethdev->data->dev_link.link_speed =
2464                                                 slave_link.link_speed;
2465                 }
2466                 break;
2467         case BONDING_MODE_ACTIVE_BACKUP:
2468                 /* Current primary slave */
2469                 ret = link_update(bond_ctx->current_primary_port, &slave_link);
2470                 if (ret < 0) {
2471                         RTE_BOND_LOG(ERR, "Slave (port %u) link get failed: %s",
2472                                 bond_ctx->current_primary_port,
2473                                 rte_strerror(-ret));
2474                         return 0;
2475                 }
2476
2477                 ethdev->data->dev_link.link_speed = slave_link.link_speed;
2478                 break;
2479         case BONDING_MODE_8023AD:
2480                 ethdev->data->dev_link.link_autoneg =
2481                                 bond_ctx->mode4.slave_link.link_autoneg;
2482                 ethdev->data->dev_link.link_duplex =
2483                                 bond_ctx->mode4.slave_link.link_duplex;
2484                 /* fall through */
2485                 /* to update link speed */
2486         case BONDING_MODE_ROUND_ROBIN:
2487         case BONDING_MODE_BALANCE:
2488         case BONDING_MODE_TLB:
2489         case BONDING_MODE_ALB:
2490         default:
2491                 /**
2492                  * In theses mode the maximum theoretical link speed is the sum
2493                  * of all the slaves
2494                  */
2495                 ethdev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
2496                 one_link_update_succeeded = false;
2497
2498                 for (idx = 0; idx < bond_ctx->active_slave_count; idx++) {
2499                         ret = link_update(bond_ctx->active_slaves[idx],
2500                                         &slave_link);
2501                         if (ret < 0) {
2502                                 RTE_BOND_LOG(ERR,
2503                                         "Slave (port %u) link get failed: %s",
2504                                         bond_ctx->active_slaves[idx],
2505                                         rte_strerror(-ret));
2506                                 continue;
2507                         }
2508
2509                         one_link_update_succeeded = true;
2510                         ethdev->data->dev_link.link_speed +=
2511                                         slave_link.link_speed;
2512                 }
2513
2514                 if (!one_link_update_succeeded) {
2515                         RTE_BOND_LOG(ERR, "All slaves link get failed");
2516                         return 0;
2517                 }
2518         }
2519
2520
2521         return 0;
2522 }
2523
2524
2525 static int
2526 bond_ethdev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
2527 {
2528         struct bond_dev_private *internals = dev->data->dev_private;
2529         struct rte_eth_stats slave_stats;
2530         int i, j;
2531
2532         for (i = 0; i < internals->slave_count; i++) {
2533                 rte_eth_stats_get(internals->slaves[i].port_id, &slave_stats);
2534
2535                 stats->ipackets += slave_stats.ipackets;
2536                 stats->opackets += slave_stats.opackets;
2537                 stats->ibytes += slave_stats.ibytes;
2538                 stats->obytes += slave_stats.obytes;
2539                 stats->imissed += slave_stats.imissed;
2540                 stats->ierrors += slave_stats.ierrors;
2541                 stats->oerrors += slave_stats.oerrors;
2542                 stats->rx_nombuf += slave_stats.rx_nombuf;
2543
2544                 for (j = 0; j < RTE_ETHDEV_QUEUE_STAT_CNTRS; j++) {
2545                         stats->q_ipackets[j] += slave_stats.q_ipackets[j];
2546                         stats->q_opackets[j] += slave_stats.q_opackets[j];
2547                         stats->q_ibytes[j] += slave_stats.q_ibytes[j];
2548                         stats->q_obytes[j] += slave_stats.q_obytes[j];
2549                         stats->q_errors[j] += slave_stats.q_errors[j];
2550                 }
2551
2552         }
2553
2554         return 0;
2555 }
2556
2557 static int
2558 bond_ethdev_stats_reset(struct rte_eth_dev *dev)
2559 {
2560         struct bond_dev_private *internals = dev->data->dev_private;
2561         int i;
2562         int err;
2563         int ret;
2564
2565         for (i = 0, err = 0; i < internals->slave_count; i++) {
2566                 ret = rte_eth_stats_reset(internals->slaves[i].port_id);
2567                 if (ret != 0)
2568                         err = ret;
2569         }
2570
2571         return err;
2572 }
2573
2574 static int
2575 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev)
2576 {
2577         struct bond_dev_private *internals = eth_dev->data->dev_private;
2578         int i;
2579         int ret = 0;
2580         uint16_t port_id;
2581
2582         switch (internals->mode) {
2583         /* Promiscuous mode is propagated to all slaves */
2584         case BONDING_MODE_ROUND_ROBIN:
2585         case BONDING_MODE_BALANCE:
2586         case BONDING_MODE_BROADCAST:
2587         case BONDING_MODE_8023AD: {
2588                 unsigned int slave_ok = 0;
2589
2590                 for (i = 0; i < internals->slave_count; i++) {
2591                         port_id = internals->slaves[i].port_id;
2592
2593                         ret = rte_eth_promiscuous_enable(port_id);
2594                         if (ret != 0)
2595                                 RTE_BOND_LOG(ERR,
2596                                         "Failed to enable promiscuous mode for port %u: %s",
2597                                         port_id, rte_strerror(-ret));
2598                         else
2599                                 slave_ok++;
2600                 }
2601                 /*
2602                  * Report success if operation is successful on at least
2603                  * on one slave. Otherwise return last error code.
2604                  */
2605                 if (slave_ok > 0)
2606                         ret = 0;
2607                 break;
2608         }
2609         /* Promiscuous mode is propagated only to primary slave */
2610         case BONDING_MODE_ACTIVE_BACKUP:
2611         case BONDING_MODE_TLB:
2612         case BONDING_MODE_ALB:
2613         default:
2614                 /* Do not touch promisc when there cannot be primary ports */
2615                 if (internals->slave_count == 0)
2616                         break;
2617                 port_id = internals->current_primary_port;
2618                 ret = rte_eth_promiscuous_enable(port_id);
2619                 if (ret != 0)
2620                         RTE_BOND_LOG(ERR,
2621                                 "Failed to enable promiscuous mode for port %u: %s",
2622                                 port_id, rte_strerror(-ret));
2623         }
2624
2625         return ret;
2626 }
2627
2628 static int
2629 bond_ethdev_promiscuous_disable(struct rte_eth_dev *dev)
2630 {
2631         struct bond_dev_private *internals = dev->data->dev_private;
2632         int i;
2633         int ret = 0;
2634         uint16_t port_id;
2635
2636         switch (internals->mode) {
2637         /* Promiscuous mode is propagated to all slaves */
2638         case BONDING_MODE_ROUND_ROBIN:
2639         case BONDING_MODE_BALANCE:
2640         case BONDING_MODE_BROADCAST:
2641         case BONDING_MODE_8023AD: {
2642                 unsigned int slave_ok = 0;
2643
2644                 for (i = 0; i < internals->slave_count; i++) {
2645                         port_id = internals->slaves[i].port_id;
2646
2647                         if (internals->mode == BONDING_MODE_8023AD &&
2648                             bond_mode_8023ad_ports[port_id].forced_rx_flags ==
2649                                         BOND_8023AD_FORCED_PROMISC) {
2650                                 slave_ok++;
2651                                 continue;
2652                         }
2653                         ret = rte_eth_promiscuous_disable(port_id);
2654                         if (ret != 0)
2655                                 RTE_BOND_LOG(ERR,
2656                                         "Failed to disable promiscuous mode for port %u: %s",
2657                                         port_id, rte_strerror(-ret));
2658                         else
2659                                 slave_ok++;
2660                 }
2661                 /*
2662                  * Report success if operation is successful on at least
2663                  * on one slave. Otherwise return last error code.
2664                  */
2665                 if (slave_ok > 0)
2666                         ret = 0;
2667                 break;
2668         }
2669         /* Promiscuous mode is propagated only to primary slave */
2670         case BONDING_MODE_ACTIVE_BACKUP:
2671         case BONDING_MODE_TLB:
2672         case BONDING_MODE_ALB:
2673         default:
2674                 /* Do not touch promisc when there cannot be primary ports */
2675                 if (internals->slave_count == 0)
2676                         break;
2677                 port_id = internals->current_primary_port;
2678                 ret = rte_eth_promiscuous_disable(port_id);
2679                 if (ret != 0)
2680                         RTE_BOND_LOG(ERR,
2681                                 "Failed to disable promiscuous mode for port %u: %s",
2682                                 port_id, rte_strerror(-ret));
2683         }
2684
2685         return ret;
2686 }
2687
2688 static int
2689 bond_ethdev_allmulticast_enable(struct rte_eth_dev *eth_dev)
2690 {
2691         struct bond_dev_private *internals = eth_dev->data->dev_private;
2692         int i;
2693         int ret = 0;
2694         uint16_t port_id;
2695
2696         switch (internals->mode) {
2697         /* allmulti mode is propagated to all slaves */
2698         case BONDING_MODE_ROUND_ROBIN:
2699         case BONDING_MODE_BALANCE:
2700         case BONDING_MODE_BROADCAST:
2701         case BONDING_MODE_8023AD: {
2702                 unsigned int slave_ok = 0;
2703
2704                 for (i = 0; i < internals->slave_count; i++) {
2705                         port_id = internals->slaves[i].port_id;
2706
2707                         ret = rte_eth_allmulticast_enable(port_id);
2708                         if (ret != 0)
2709                                 RTE_BOND_LOG(ERR,
2710                                         "Failed to enable allmulti mode for port %u: %s",
2711                                         port_id, rte_strerror(-ret));
2712                         else
2713                                 slave_ok++;
2714                 }
2715                 /*
2716                  * Report success if operation is successful on at least
2717                  * on one slave. Otherwise return last error code.
2718                  */
2719                 if (slave_ok > 0)
2720                         ret = 0;
2721                 break;
2722         }
2723         /* allmulti mode is propagated only to primary slave */
2724         case BONDING_MODE_ACTIVE_BACKUP:
2725         case BONDING_MODE_TLB:
2726         case BONDING_MODE_ALB:
2727         default:
2728                 /* Do not touch allmulti when there cannot be primary ports */
2729                 if (internals->slave_count == 0)
2730                         break;
2731                 port_id = internals->current_primary_port;
2732                 ret = rte_eth_allmulticast_enable(port_id);
2733                 if (ret != 0)
2734                         RTE_BOND_LOG(ERR,
2735                                 "Failed to enable allmulti mode for port %u: %s",
2736                                 port_id, rte_strerror(-ret));
2737         }
2738
2739         return ret;
2740 }
2741
2742 static int
2743 bond_ethdev_allmulticast_disable(struct rte_eth_dev *eth_dev)
2744 {
2745         struct bond_dev_private *internals = eth_dev->data->dev_private;
2746         int i;
2747         int ret = 0;
2748         uint16_t port_id;
2749
2750         switch (internals->mode) {
2751         /* allmulti mode is propagated to all slaves */
2752         case BONDING_MODE_ROUND_ROBIN:
2753         case BONDING_MODE_BALANCE:
2754         case BONDING_MODE_BROADCAST:
2755         case BONDING_MODE_8023AD: {
2756                 unsigned int slave_ok = 0;
2757
2758                 for (i = 0; i < internals->slave_count; i++) {
2759                         uint16_t port_id = internals->slaves[i].port_id;
2760
2761                         if (internals->mode == BONDING_MODE_8023AD &&
2762                             bond_mode_8023ad_ports[port_id].forced_rx_flags ==
2763                                         BOND_8023AD_FORCED_ALLMULTI)
2764                                 continue;
2765
2766                         ret = rte_eth_allmulticast_disable(port_id);
2767                         if (ret != 0)
2768                                 RTE_BOND_LOG(ERR,
2769                                         "Failed to disable allmulti mode for port %u: %s",
2770                                         port_id, rte_strerror(-ret));
2771                         else
2772                                 slave_ok++;
2773                 }
2774                 /*
2775                  * Report success if operation is successful on at least
2776                  * on one slave. Otherwise return last error code.
2777                  */
2778                 if (slave_ok > 0)
2779                         ret = 0;
2780                 break;
2781         }
2782         /* allmulti mode is propagated only to primary slave */
2783         case BONDING_MODE_ACTIVE_BACKUP:
2784         case BONDING_MODE_TLB:
2785         case BONDING_MODE_ALB:
2786         default:
2787                 /* Do not touch allmulti when there cannot be primary ports */
2788                 if (internals->slave_count == 0)
2789                         break;
2790                 port_id = internals->current_primary_port;
2791                 ret = rte_eth_allmulticast_disable(port_id);
2792                 if (ret != 0)
2793                         RTE_BOND_LOG(ERR,
2794                                 "Failed to disable allmulti mode for port %u: %s",
2795                                 port_id, rte_strerror(-ret));
2796         }
2797
2798         return ret;
2799 }
2800
2801 static void
2802 bond_ethdev_delayed_lsc_propagation(void *arg)
2803 {
2804         if (arg == NULL)
2805                 return;
2806
2807         rte_eth_dev_callback_process((struct rte_eth_dev *)arg,
2808                         RTE_ETH_EVENT_INTR_LSC, NULL);
2809 }
2810
2811 int
2812 bond_ethdev_lsc_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2813                 void *param, void *ret_param __rte_unused)
2814 {
2815         struct rte_eth_dev *bonded_eth_dev;
2816         struct bond_dev_private *internals;
2817         struct rte_eth_link link;
2818         int rc = -1;
2819         int ret;
2820
2821         uint8_t lsc_flag = 0;
2822         int valid_slave = 0;
2823         uint16_t active_pos;
2824         uint16_t i;
2825
2826         if (type != RTE_ETH_EVENT_INTR_LSC || param == NULL)
2827                 return rc;
2828
2829         bonded_eth_dev = &rte_eth_devices[*(uint16_t *)param];
2830
2831         if (check_for_bonded_ethdev(bonded_eth_dev))
2832                 return rc;
2833
2834         internals = bonded_eth_dev->data->dev_private;
2835
2836         /* If the device isn't started don't handle interrupts */
2837         if (!bonded_eth_dev->data->dev_started)
2838                 return rc;
2839
2840         /* verify that port_id is a valid slave of bonded port */
2841         for (i = 0; i < internals->slave_count; i++) {
2842                 if (internals->slaves[i].port_id == port_id) {
2843                         valid_slave = 1;
2844                         break;
2845                 }
2846         }
2847
2848         if (!valid_slave)
2849                 return rc;
2850
2851         /* Synchronize lsc callback parallel calls either by real link event
2852          * from the slaves PMDs or by the bonding PMD itself.
2853          */
2854         rte_spinlock_lock(&internals->lsc_lock);
2855
2856         /* Search for port in active port list */
2857         active_pos = find_slave_by_id(internals->active_slaves,
2858                         internals->active_slave_count, port_id);
2859
2860         ret = rte_eth_link_get_nowait(port_id, &link);
2861         if (ret < 0)
2862                 RTE_BOND_LOG(ERR, "Slave (port %u) link get failed", port_id);
2863
2864         if (ret == 0 && link.link_status) {
2865                 if (active_pos < internals->active_slave_count)
2866                         goto link_update;
2867
2868                 /* check link state properties if bonded link is up*/
2869                 if (bonded_eth_dev->data->dev_link.link_status == ETH_LINK_UP) {
2870                         if (link_properties_valid(bonded_eth_dev, &link) != 0)
2871                                 RTE_BOND_LOG(ERR, "Invalid link properties "
2872                                              "for slave %d in bonding mode %d",
2873                                              port_id, internals->mode);
2874                 } else {
2875                         /* inherit slave link properties */
2876                         link_properties_set(bonded_eth_dev, &link);
2877                 }
2878
2879                 /* If no active slave ports then set this port to be
2880                  * the primary port.
2881                  */
2882                 if (internals->active_slave_count < 1) {
2883                         /* If first active slave, then change link status */
2884                         bonded_eth_dev->data->dev_link.link_status =
2885                                                                 ETH_LINK_UP;
2886                         internals->current_primary_port = port_id;
2887                         lsc_flag = 1;
2888
2889                         mac_address_slaves_update(bonded_eth_dev);
2890                 }
2891
2892                 activate_slave(bonded_eth_dev, port_id);
2893
2894                 /* If the user has defined the primary port then default to
2895                  * using it.
2896                  */
2897                 if (internals->user_defined_primary_port &&
2898                                 internals->primary_port == port_id)
2899                         bond_ethdev_primary_set(internals, port_id);
2900         } else {
2901                 if (active_pos == internals->active_slave_count)
2902                         goto link_update;
2903
2904                 /* Remove from active slave list */
2905                 deactivate_slave(bonded_eth_dev, port_id);
2906
2907                 if (internals->active_slave_count < 1)
2908                         lsc_flag = 1;
2909
2910                 /* Update primary id, take first active slave from list or if none
2911                  * available set to -1 */
2912                 if (port_id == internals->current_primary_port) {
2913                         if (internals->active_slave_count > 0)
2914                                 bond_ethdev_primary_set(internals,
2915                                                 internals->active_slaves[0]);
2916                         else
2917                                 internals->current_primary_port = internals->primary_port;
2918                         mac_address_slaves_update(bonded_eth_dev);
2919                 }
2920         }
2921
2922 link_update:
2923         /**
2924          * Update bonded device link properties after any change to active
2925          * slaves
2926          */
2927         bond_ethdev_link_update(bonded_eth_dev, 0);
2928
2929         if (lsc_flag) {
2930                 /* Cancel any possible outstanding interrupts if delays are enabled */
2931                 if (internals->link_up_delay_ms > 0 ||
2932                         internals->link_down_delay_ms > 0)
2933                         rte_eal_alarm_cancel(bond_ethdev_delayed_lsc_propagation,
2934                                         bonded_eth_dev);
2935
2936                 if (bonded_eth_dev->data->dev_link.link_status) {
2937                         if (internals->link_up_delay_ms > 0)
2938                                 rte_eal_alarm_set(internals->link_up_delay_ms * 1000,
2939                                                 bond_ethdev_delayed_lsc_propagation,
2940                                                 (void *)bonded_eth_dev);
2941                         else
2942                                 rte_eth_dev_callback_process(bonded_eth_dev,
2943                                                 RTE_ETH_EVENT_INTR_LSC,
2944                                                 NULL);
2945
2946                 } else {
2947                         if (internals->link_down_delay_ms > 0)
2948                                 rte_eal_alarm_set(internals->link_down_delay_ms * 1000,
2949                                                 bond_ethdev_delayed_lsc_propagation,
2950                                                 (void *)bonded_eth_dev);
2951                         else
2952                                 rte_eth_dev_callback_process(bonded_eth_dev,
2953                                                 RTE_ETH_EVENT_INTR_LSC,
2954                                                 NULL);
2955                 }
2956         }
2957
2958         rte_spinlock_unlock(&internals->lsc_lock);
2959
2960         return rc;
2961 }
2962
2963 static int
2964 bond_ethdev_rss_reta_update(struct rte_eth_dev *dev,
2965                 struct rte_eth_rss_reta_entry64 *reta_conf, uint16_t reta_size)
2966 {
2967         unsigned i, j;
2968         int result = 0;
2969         int slave_reta_size;
2970         unsigned reta_count;
2971         struct bond_dev_private *internals = dev->data->dev_private;
2972
2973         if (reta_size != internals->reta_size)
2974                 return -EINVAL;
2975
2976          /* Copy RETA table */
2977         reta_count = (reta_size + RTE_RETA_GROUP_SIZE - 1) /
2978                         RTE_RETA_GROUP_SIZE;
2979
2980         for (i = 0; i < reta_count; i++) {
2981                 internals->reta_conf[i].mask = reta_conf[i].mask;
2982                 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
2983                         if ((reta_conf[i].mask >> j) & 0x01)
2984                                 internals->reta_conf[i].reta[j] = reta_conf[i].reta[j];
2985         }
2986
2987         /* Fill rest of array */
2988         for (; i < RTE_DIM(internals->reta_conf); i += reta_count)
2989                 memcpy(&internals->reta_conf[i], &internals->reta_conf[0],
2990                                 sizeof(internals->reta_conf[0]) * reta_count);
2991
2992         /* Propagate RETA over slaves */
2993         for (i = 0; i < internals->slave_count; i++) {
2994                 slave_reta_size = internals->slaves[i].reta_size;
2995                 result = rte_eth_dev_rss_reta_update(internals->slaves[i].port_id,
2996                                 &internals->reta_conf[0], slave_reta_size);
2997                 if (result < 0)
2998                         return result;
2999         }
3000
3001         return 0;
3002 }
3003
3004 static int
3005 bond_ethdev_rss_reta_query(struct rte_eth_dev *dev,
3006                 struct rte_eth_rss_reta_entry64 *reta_conf, uint16_t reta_size)
3007 {
3008         int i, j;
3009         struct bond_dev_private *internals = dev->data->dev_private;
3010
3011         if (reta_size != internals->reta_size)
3012                 return -EINVAL;
3013
3014          /* Copy RETA table */
3015         for (i = 0; i < reta_size / RTE_RETA_GROUP_SIZE; i++)
3016                 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
3017                         if ((reta_conf[i].mask >> j) & 0x01)
3018                                 reta_conf[i].reta[j] = internals->reta_conf[i].reta[j];
3019
3020         return 0;
3021 }
3022
3023 static int
3024 bond_ethdev_rss_hash_update(struct rte_eth_dev *dev,
3025                 struct rte_eth_rss_conf *rss_conf)
3026 {
3027         int i, result = 0;
3028         struct bond_dev_private *internals = dev->data->dev_private;
3029         struct rte_eth_rss_conf bond_rss_conf;
3030
3031         memcpy(&bond_rss_conf, rss_conf, sizeof(struct rte_eth_rss_conf));
3032
3033         bond_rss_conf.rss_hf &= internals->flow_type_rss_offloads;
3034
3035         if (bond_rss_conf.rss_hf != 0)
3036                 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf = bond_rss_conf.rss_hf;
3037
3038         if (bond_rss_conf.rss_key && bond_rss_conf.rss_key_len <
3039                         sizeof(internals->rss_key)) {
3040                 if (bond_rss_conf.rss_key_len == 0)
3041                         bond_rss_conf.rss_key_len = 40;
3042                 internals->rss_key_len = bond_rss_conf.rss_key_len;
3043                 memcpy(internals->rss_key, bond_rss_conf.rss_key,
3044                                 internals->rss_key_len);
3045         }
3046
3047         for (i = 0; i < internals->slave_count; i++) {
3048                 result = rte_eth_dev_rss_hash_update(internals->slaves[i].port_id,
3049                                 &bond_rss_conf);
3050                 if (result < 0)
3051                         return result;
3052         }
3053
3054         return 0;
3055 }
3056
3057 static int
3058 bond_ethdev_rss_hash_conf_get(struct rte_eth_dev *dev,
3059                 struct rte_eth_rss_conf *rss_conf)
3060 {
3061         struct bond_dev_private *internals = dev->data->dev_private;
3062
3063         rss_conf->rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
3064         rss_conf->rss_key_len = internals->rss_key_len;
3065         if (rss_conf->rss_key)
3066                 memcpy(rss_conf->rss_key, internals->rss_key, internals->rss_key_len);
3067
3068         return 0;
3069 }
3070
3071 static int
3072 bond_ethdev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
3073 {
3074         struct rte_eth_dev *slave_eth_dev;
3075         struct bond_dev_private *internals = dev->data->dev_private;
3076         int ret, i;
3077
3078         rte_spinlock_lock(&internals->lock);
3079
3080         for (i = 0; i < internals->slave_count; i++) {
3081                 slave_eth_dev = &rte_eth_devices[internals->slaves[i].port_id];
3082                 if (*slave_eth_dev->dev_ops->mtu_set == NULL) {
3083                         rte_spinlock_unlock(&internals->lock);
3084                         return -ENOTSUP;
3085                 }
3086         }
3087         for (i = 0; i < internals->slave_count; i++) {
3088                 ret = rte_eth_dev_set_mtu(internals->slaves[i].port_id, mtu);
3089                 if (ret < 0) {
3090                         rte_spinlock_unlock(&internals->lock);
3091                         return ret;
3092                 }
3093         }
3094
3095         rte_spinlock_unlock(&internals->lock);
3096         return 0;
3097 }
3098
3099 static int
3100 bond_ethdev_mac_address_set(struct rte_eth_dev *dev,
3101                         struct rte_ether_addr *addr)
3102 {
3103         if (mac_address_set(dev, addr)) {
3104                 RTE_BOND_LOG(ERR, "Failed to update MAC address");
3105                 return -EINVAL;
3106         }
3107
3108         return 0;
3109 }
3110
3111 static int
3112 bond_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
3113                   const struct rte_flow_ops **ops)
3114 {
3115         *ops = &bond_flow_ops;
3116         return 0;
3117 }
3118
3119 static int
3120 bond_ethdev_mac_addr_add(struct rte_eth_dev *dev,
3121                         struct rte_ether_addr *mac_addr,
3122                         __rte_unused uint32_t index, uint32_t vmdq)
3123 {
3124         struct rte_eth_dev *slave_eth_dev;
3125         struct bond_dev_private *internals = dev->data->dev_private;
3126         int ret, i;
3127
3128         rte_spinlock_lock(&internals->lock);
3129
3130         for (i = 0; i < internals->slave_count; i++) {
3131                 slave_eth_dev = &rte_eth_devices[internals->slaves[i].port_id];
3132                 if (*slave_eth_dev->dev_ops->mac_addr_add == NULL ||
3133                          *slave_eth_dev->dev_ops->mac_addr_remove == NULL) {
3134                         ret = -ENOTSUP;
3135                         goto end;
3136                 }
3137         }
3138
3139         for (i = 0; i < internals->slave_count; i++) {
3140                 ret = rte_eth_dev_mac_addr_add(internals->slaves[i].port_id,
3141                                 mac_addr, vmdq);
3142                 if (ret < 0) {
3143                         /* rollback */
3144                         for (i--; i >= 0; i--)
3145                                 rte_eth_dev_mac_addr_remove(
3146                                         internals->slaves[i].port_id, mac_addr);
3147                         goto end;
3148                 }
3149         }
3150
3151         ret = 0;
3152 end:
3153         rte_spinlock_unlock(&internals->lock);
3154         return ret;
3155 }
3156
3157 static void
3158 bond_ethdev_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
3159 {
3160         struct rte_eth_dev *slave_eth_dev;
3161         struct bond_dev_private *internals = dev->data->dev_private;
3162         int i;
3163
3164         rte_spinlock_lock(&internals->lock);
3165
3166         for (i = 0; i < internals->slave_count; i++) {
3167                 slave_eth_dev = &rte_eth_devices[internals->slaves[i].port_id];
3168                 if (*slave_eth_dev->dev_ops->mac_addr_remove == NULL)
3169                         goto end;
3170         }
3171
3172         struct rte_ether_addr *mac_addr = &dev->data->mac_addrs[index];
3173
3174         for (i = 0; i < internals->slave_count; i++)
3175                 rte_eth_dev_mac_addr_remove(internals->slaves[i].port_id,
3176                                 mac_addr);
3177
3178 end:
3179         rte_spinlock_unlock(&internals->lock);
3180 }
3181
3182 const struct eth_dev_ops default_dev_ops = {
3183         .dev_start            = bond_ethdev_start,
3184         .dev_stop             = bond_ethdev_stop,
3185         .dev_close            = bond_ethdev_close,
3186         .dev_configure        = bond_ethdev_configure,
3187         .dev_infos_get        = bond_ethdev_info,
3188         .vlan_filter_set      = bond_ethdev_vlan_filter_set,
3189         .rx_queue_setup       = bond_ethdev_rx_queue_setup,
3190         .tx_queue_setup       = bond_ethdev_tx_queue_setup,
3191         .rx_queue_release     = bond_ethdev_rx_queue_release,
3192         .tx_queue_release     = bond_ethdev_tx_queue_release,
3193         .link_update          = bond_ethdev_link_update,
3194         .stats_get            = bond_ethdev_stats_get,
3195         .stats_reset          = bond_ethdev_stats_reset,
3196         .promiscuous_enable   = bond_ethdev_promiscuous_enable,
3197         .promiscuous_disable  = bond_ethdev_promiscuous_disable,
3198         .allmulticast_enable  = bond_ethdev_allmulticast_enable,
3199         .allmulticast_disable = bond_ethdev_allmulticast_disable,
3200         .reta_update          = bond_ethdev_rss_reta_update,
3201         .reta_query           = bond_ethdev_rss_reta_query,
3202         .rss_hash_update      = bond_ethdev_rss_hash_update,
3203         .rss_hash_conf_get    = bond_ethdev_rss_hash_conf_get,
3204         .mtu_set              = bond_ethdev_mtu_set,
3205         .mac_addr_set         = bond_ethdev_mac_address_set,
3206         .mac_addr_add         = bond_ethdev_mac_addr_add,
3207         .mac_addr_remove      = bond_ethdev_mac_addr_remove,
3208         .flow_ops_get         = bond_flow_ops_get
3209 };
3210
3211 static int
3212 bond_alloc(struct rte_vdev_device *dev, uint8_t mode)
3213 {
3214         const char *name = rte_vdev_device_name(dev);
3215         uint8_t socket_id = dev->device.numa_node;
3216         struct bond_dev_private *internals = NULL;
3217         struct rte_eth_dev *eth_dev = NULL;
3218         uint32_t vlan_filter_bmp_size;
3219
3220         /* now do all data allocation - for eth_dev structure, dummy pci driver
3221          * and internal (private) data
3222          */
3223
3224         /* reserve an ethdev entry */
3225         eth_dev = rte_eth_vdev_allocate(dev, sizeof(*internals));
3226         if (eth_dev == NULL) {
3227                 RTE_BOND_LOG(ERR, "Unable to allocate rte_eth_dev");
3228                 goto err;
3229         }
3230
3231         internals = eth_dev->data->dev_private;
3232         eth_dev->data->nb_rx_queues = (uint16_t)1;
3233         eth_dev->data->nb_tx_queues = (uint16_t)1;
3234
3235         /* Allocate memory for storing MAC addresses */
3236         eth_dev->data->mac_addrs = rte_zmalloc_socket(name, RTE_ETHER_ADDR_LEN *
3237                         BOND_MAX_MAC_ADDRS, 0, socket_id);
3238         if (eth_dev->data->mac_addrs == NULL) {
3239                 RTE_BOND_LOG(ERR,
3240                              "Failed to allocate %u bytes needed to store MAC addresses",
3241                              RTE_ETHER_ADDR_LEN * BOND_MAX_MAC_ADDRS);
3242                 goto err;
3243         }
3244
3245         eth_dev->dev_ops = &default_dev_ops;
3246         eth_dev->data->dev_flags = RTE_ETH_DEV_INTR_LSC |
3247                                         RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
3248
3249         rte_spinlock_init(&internals->lock);
3250         rte_spinlock_init(&internals->lsc_lock);
3251
3252         internals->port_id = eth_dev->data->port_id;
3253         internals->mode = BONDING_MODE_INVALID;
3254         internals->current_primary_port = RTE_MAX_ETHPORTS + 1;
3255         internals->balance_xmit_policy = BALANCE_XMIT_POLICY_LAYER2;
3256         internals->burst_xmit_hash = burst_xmit_l2_hash;
3257         internals->user_defined_mac = 0;
3258
3259         internals->link_status_polling_enabled = 0;
3260
3261         internals->link_status_polling_interval_ms =
3262                 DEFAULT_POLLING_INTERVAL_10_MS;
3263         internals->link_down_delay_ms = 0;
3264         internals->link_up_delay_ms = 0;
3265
3266         internals->slave_count = 0;
3267         internals->active_slave_count = 0;
3268         internals->rx_offload_capa = 0;
3269         internals->tx_offload_capa = 0;
3270         internals->rx_queue_offload_capa = 0;
3271         internals->tx_queue_offload_capa = 0;
3272         internals->candidate_max_rx_pktlen = 0;
3273         internals->max_rx_pktlen = 0;
3274
3275         /* Initially allow to choose any offload type */
3276         internals->flow_type_rss_offloads = ETH_RSS_PROTO_MASK;
3277
3278         memset(&internals->default_rxconf, 0,
3279                sizeof(internals->default_rxconf));
3280         memset(&internals->default_txconf, 0,
3281                sizeof(internals->default_txconf));
3282
3283         memset(&internals->rx_desc_lim, 0, sizeof(internals->rx_desc_lim));
3284         memset(&internals->tx_desc_lim, 0, sizeof(internals->tx_desc_lim));
3285
3286         memset(internals->active_slaves, 0, sizeof(internals->active_slaves));
3287         memset(internals->slaves, 0, sizeof(internals->slaves));
3288
3289         TAILQ_INIT(&internals->flow_list);
3290         internals->flow_isolated_valid = 0;
3291
3292         /* Set mode 4 default configuration */
3293         bond_mode_8023ad_setup(eth_dev, NULL);
3294         if (bond_ethdev_mode_set(eth_dev, mode)) {
3295                 RTE_BOND_LOG(ERR, "Failed to set bonded device %d mode to %d",
3296                                  eth_dev->data->port_id, mode);
3297                 goto err;
3298         }
3299
3300         vlan_filter_bmp_size =
3301                 rte_bitmap_get_memory_footprint(RTE_ETHER_MAX_VLAN_ID + 1);
3302         internals->vlan_filter_bmpmem = rte_malloc(name, vlan_filter_bmp_size,
3303                                                    RTE_CACHE_LINE_SIZE);
3304         if (internals->vlan_filter_bmpmem == NULL) {
3305                 RTE_BOND_LOG(ERR,
3306                              "Failed to allocate vlan bitmap for bonded device %u",
3307                              eth_dev->data->port_id);
3308                 goto err;
3309         }
3310
3311         internals->vlan_filter_bmp = rte_bitmap_init(RTE_ETHER_MAX_VLAN_ID + 1,
3312                         internals->vlan_filter_bmpmem, vlan_filter_bmp_size);
3313         if (internals->vlan_filter_bmp == NULL) {
3314                 RTE_BOND_LOG(ERR,
3315                              "Failed to init vlan bitmap for bonded device %u",
3316                              eth_dev->data->port_id);
3317                 rte_free(internals->vlan_filter_bmpmem);
3318                 goto err;
3319         }
3320
3321         return eth_dev->data->port_id;
3322
3323 err:
3324         rte_free(internals);
3325         if (eth_dev != NULL)
3326                 eth_dev->data->dev_private = NULL;
3327         rte_eth_dev_release_port(eth_dev);
3328         return -1;
3329 }
3330
3331 static int
3332 bond_probe(struct rte_vdev_device *dev)
3333 {
3334         const char *name;
3335         struct bond_dev_private *internals;
3336         struct rte_kvargs *kvlist;
3337         uint8_t bonding_mode;
3338         int arg_count, port_id;
3339         int socket_id;
3340         uint8_t agg_mode;
3341         struct rte_eth_dev *eth_dev;
3342
3343         if (!dev)
3344                 return -EINVAL;
3345
3346         name = rte_vdev_device_name(dev);
3347         RTE_BOND_LOG(INFO, "Initializing pmd_bond for %s", name);
3348
3349         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
3350                 eth_dev = rte_eth_dev_attach_secondary(name);
3351                 if (!eth_dev) {
3352                         RTE_BOND_LOG(ERR, "Failed to probe %s", name);
3353                         return -1;
3354                 }
3355                 /* TODO: request info from primary to set up Rx and Tx */
3356                 eth_dev->dev_ops = &default_dev_ops;
3357                 eth_dev->device = &dev->device;
3358                 rte_eth_dev_probing_finish(eth_dev);
3359                 return 0;
3360         }
3361
3362         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev),
3363                 pmd_bond_init_valid_arguments);
3364         if (kvlist == NULL)
3365                 return -1;
3366
3367         /* Parse link bonding mode */
3368         if (rte_kvargs_count(kvlist, PMD_BOND_MODE_KVARG) == 1) {
3369                 if (rte_kvargs_process(kvlist, PMD_BOND_MODE_KVARG,
3370                                 &bond_ethdev_parse_slave_mode_kvarg,
3371                                 &bonding_mode) != 0) {
3372                         RTE_BOND_LOG(ERR, "Invalid mode for bonded device %s",
3373                                         name);
3374                         goto parse_error;
3375                 }
3376         } else {
3377                 RTE_BOND_LOG(ERR, "Mode must be specified only once for bonded "
3378                                 "device %s", name);
3379                 goto parse_error;
3380         }
3381
3382         /* Parse socket id to create bonding device on */
3383         arg_count = rte_kvargs_count(kvlist, PMD_BOND_SOCKET_ID_KVARG);
3384         if (arg_count == 1) {
3385                 if (rte_kvargs_process(kvlist, PMD_BOND_SOCKET_ID_KVARG,
3386                                 &bond_ethdev_parse_socket_id_kvarg, &socket_id)
3387                                 != 0) {
3388                         RTE_BOND_LOG(ERR, "Invalid socket Id specified for "
3389                                         "bonded device %s", name);
3390                         goto parse_error;
3391                 }
3392         } else if (arg_count > 1) {
3393                 RTE_BOND_LOG(ERR, "Socket Id can be specified only once for "
3394                                 "bonded device %s", name);
3395                 goto parse_error;
3396         } else {
3397                 socket_id = rte_socket_id();
3398         }
3399
3400         dev->device.numa_node = socket_id;
3401
3402         /* Create link bonding eth device */
3403         port_id = bond_alloc(dev, bonding_mode);
3404         if (port_id < 0) {
3405                 RTE_BOND_LOG(ERR, "Failed to create socket %s in mode %u on "
3406                                 "socket %u.",   name, bonding_mode, socket_id);
3407                 goto parse_error;
3408         }
3409         internals = rte_eth_devices[port_id].data->dev_private;
3410         internals->kvlist = kvlist;
3411
3412         if (rte_kvargs_count(kvlist, PMD_BOND_AGG_MODE_KVARG) == 1) {
3413                 if (rte_kvargs_process(kvlist,
3414                                 PMD_BOND_AGG_MODE_KVARG,
3415                                 &bond_ethdev_parse_slave_agg_mode_kvarg,
3416                                 &agg_mode) != 0) {
3417                         RTE_BOND_LOG(ERR,
3418                                         "Failed to parse agg selection mode for bonded device %s",
3419                                         name);
3420                         goto parse_error;
3421                 }
3422
3423                 if (internals->mode == BONDING_MODE_8023AD)
3424                         internals->mode4.agg_selection = agg_mode;
3425         } else {
3426                 internals->mode4.agg_selection = AGG_STABLE;
3427         }
3428
3429         rte_eth_dev_probing_finish(&rte_eth_devices[port_id]);
3430         RTE_BOND_LOG(INFO, "Create bonded device %s on port %d in mode %u on "
3431                         "socket %u.",   name, port_id, bonding_mode, socket_id);
3432         return 0;
3433
3434 parse_error:
3435         rte_kvargs_free(kvlist);
3436
3437         return -1;
3438 }
3439
3440 static int
3441 bond_remove(struct rte_vdev_device *dev)
3442 {
3443         struct rte_eth_dev *eth_dev;
3444         struct bond_dev_private *internals;
3445         const char *name;
3446         int ret = 0;
3447
3448         if (!dev)
3449                 return -EINVAL;
3450
3451         name = rte_vdev_device_name(dev);
3452         RTE_BOND_LOG(INFO, "Uninitializing pmd_bond for %s", name);
3453
3454         /* find an ethdev entry */
3455         eth_dev = rte_eth_dev_allocated(name);
3456         if (eth_dev == NULL)
3457                 return 0; /* port already released */
3458
3459         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
3460                 return rte_eth_dev_release_port(eth_dev);
3461
3462         RTE_ASSERT(eth_dev->device == &dev->device);
3463
3464         internals = eth_dev->data->dev_private;
3465         if (internals->slave_count != 0)
3466                 return -EBUSY;
3467
3468         if (eth_dev->data->dev_started == 1) {
3469                 ret = bond_ethdev_stop(eth_dev);
3470                 bond_ethdev_close(eth_dev);
3471         }
3472         if (internals->kvlist != NULL)
3473                 rte_kvargs_free(internals->kvlist);
3474         rte_eth_dev_release_port(eth_dev);
3475
3476         return ret;
3477 }
3478
3479 /* this part will resolve the slave portids after all the other pdev and vdev
3480  * have been allocated */
3481 static int
3482 bond_ethdev_configure(struct rte_eth_dev *dev)
3483 {
3484         const char *name = dev->device->name;
3485         struct bond_dev_private *internals = dev->data->dev_private;
3486         struct rte_kvargs *kvlist = internals->kvlist;
3487         int arg_count;
3488         uint16_t port_id = dev - rte_eth_devices;
3489         uint8_t agg_mode;
3490
3491         static const uint8_t default_rss_key[40] = {
3492                 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2, 0x41, 0x67, 0x25, 0x3D,
3493                 0x43, 0xA3, 0x8F, 0xB0, 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
3494                 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C, 0x6A, 0x42, 0xB7, 0x3B,
3495                 0xBE, 0xAC, 0x01, 0xFA
3496         };
3497
3498         unsigned i, j;
3499
3500         /*
3501          * If RSS is enabled, fill table with default values and
3502          * set key to the the value specified in port RSS configuration.
3503          * Fall back to default RSS key if the key is not specified
3504          */
3505         if (dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS) {
3506                 if (dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key != NULL) {
3507                         internals->rss_key_len =
3508                                 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len;
3509                         memcpy(internals->rss_key,
3510                                dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key,
3511                                internals->rss_key_len);
3512                 } else {
3513                         internals->rss_key_len = sizeof(default_rss_key);
3514                         memcpy(internals->rss_key, default_rss_key,
3515                                internals->rss_key_len);
3516                 }
3517
3518                 for (i = 0; i < RTE_DIM(internals->reta_conf); i++) {
3519                         internals->reta_conf[i].mask = ~0LL;
3520                         for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
3521                                 internals->reta_conf[i].reta[j] =
3522                                                 (i * RTE_RETA_GROUP_SIZE + j) %
3523                                                 dev->data->nb_rx_queues;
3524                 }
3525         }
3526
3527         /* set the max_rx_pktlen */
3528         internals->max_rx_pktlen = internals->candidate_max_rx_pktlen;
3529
3530         /*
3531          * if no kvlist, it means that this bonded device has been created
3532          * through the bonding api.
3533          */
3534         if (!kvlist)
3535                 return 0;
3536
3537         /* Parse MAC address for bonded device */
3538         arg_count = rte_kvargs_count(kvlist, PMD_BOND_MAC_ADDR_KVARG);
3539         if (arg_count == 1) {
3540                 struct rte_ether_addr bond_mac;
3541
3542                 if (rte_kvargs_process(kvlist, PMD_BOND_MAC_ADDR_KVARG,
3543                                        &bond_ethdev_parse_bond_mac_addr_kvarg, &bond_mac) < 0) {
3544                         RTE_BOND_LOG(INFO, "Invalid mac address for bonded device %s",
3545                                      name);
3546                         return -1;
3547                 }
3548
3549                 /* Set MAC address */
3550                 if (rte_eth_bond_mac_address_set(port_id, &bond_mac) != 0) {
3551                         RTE_BOND_LOG(ERR,
3552                                      "Failed to set mac address on bonded device %s",
3553                                      name);
3554                         return -1;
3555                 }
3556         } else if (arg_count > 1) {
3557                 RTE_BOND_LOG(ERR,
3558                              "MAC address can be specified only once for bonded device %s",
3559                              name);
3560                 return -1;
3561         }
3562
3563         /* Parse/set balance mode transmit policy */
3564         arg_count = rte_kvargs_count(kvlist, PMD_BOND_XMIT_POLICY_KVARG);
3565         if (arg_count == 1) {
3566                 uint8_t xmit_policy;
3567
3568                 if (rte_kvargs_process(kvlist, PMD_BOND_XMIT_POLICY_KVARG,
3569                                        &bond_ethdev_parse_balance_xmit_policy_kvarg, &xmit_policy) !=
3570                     0) {
3571                         RTE_BOND_LOG(INFO,
3572                                      "Invalid xmit policy specified for bonded device %s",
3573                                      name);
3574                         return -1;
3575                 }
3576
3577                 /* Set balance mode transmit policy*/
3578                 if (rte_eth_bond_xmit_policy_set(port_id, xmit_policy) != 0) {
3579                         RTE_BOND_LOG(ERR,
3580                                      "Failed to set balance xmit policy on bonded device %s",
3581                                      name);
3582                         return -1;
3583                 }
3584         } else if (arg_count > 1) {
3585                 RTE_BOND_LOG(ERR,
3586                              "Transmit policy can be specified only once for bonded device %s",
3587                              name);
3588                 return -1;
3589         }
3590
3591         if (rte_kvargs_count(kvlist, PMD_BOND_AGG_MODE_KVARG) == 1) {
3592                 if (rte_kvargs_process(kvlist,
3593                                        PMD_BOND_AGG_MODE_KVARG,
3594                                        &bond_ethdev_parse_slave_agg_mode_kvarg,
3595                                        &agg_mode) != 0) {
3596                         RTE_BOND_LOG(ERR,
3597                                      "Failed to parse agg selection mode for bonded device %s",
3598                                      name);
3599                 }
3600                 if (internals->mode == BONDING_MODE_8023AD) {
3601                         int ret = rte_eth_bond_8023ad_agg_selection_set(port_id,
3602                                         agg_mode);
3603                         if (ret < 0) {
3604                                 RTE_BOND_LOG(ERR,
3605                                         "Invalid args for agg selection set for bonded device %s",
3606                                         name);
3607                                 return -1;
3608                         }
3609                 }
3610         }
3611
3612         /* Parse/add slave ports to bonded device */
3613         if (rte_kvargs_count(kvlist, PMD_BOND_SLAVE_PORT_KVARG) > 0) {
3614                 struct bond_ethdev_slave_ports slave_ports;
3615                 unsigned i;
3616
3617                 memset(&slave_ports, 0, sizeof(slave_ports));
3618
3619                 if (rte_kvargs_process(kvlist, PMD_BOND_SLAVE_PORT_KVARG,
3620                                        &bond_ethdev_parse_slave_port_kvarg, &slave_ports) != 0) {
3621                         RTE_BOND_LOG(ERR,
3622                                      "Failed to parse slave ports for bonded device %s",
3623                                      name);
3624                         return -1;
3625                 }
3626
3627                 for (i = 0; i < slave_ports.slave_count; i++) {
3628                         if (rte_eth_bond_slave_add(port_id, slave_ports.slaves[i]) != 0) {
3629                                 RTE_BOND_LOG(ERR,
3630                                              "Failed to add port %d as slave to bonded device %s",
3631                                              slave_ports.slaves[i], name);
3632                         }
3633                 }
3634
3635         } else {
3636                 RTE_BOND_LOG(INFO, "No slaves specified for bonded device %s", name);
3637                 return -1;
3638         }
3639
3640         /* Parse/set primary slave port id*/
3641         arg_count = rte_kvargs_count(kvlist, PMD_BOND_PRIMARY_SLAVE_KVARG);
3642         if (arg_count == 1) {
3643                 uint16_t primary_slave_port_id;
3644
3645                 if (rte_kvargs_process(kvlist,
3646                                        PMD_BOND_PRIMARY_SLAVE_KVARG,
3647                                        &bond_ethdev_parse_primary_slave_port_id_kvarg,
3648                                        &primary_slave_port_id) < 0) {
3649                         RTE_BOND_LOG(INFO,
3650                                      "Invalid primary slave port id specified for bonded device %s",
3651                                      name);
3652                         return -1;
3653                 }
3654
3655                 /* Set balance mode transmit policy*/
3656                 if (rte_eth_bond_primary_set(port_id, primary_slave_port_id)
3657                     != 0) {
3658                         RTE_BOND_LOG(ERR,
3659                                      "Failed to set primary slave port %d on bonded device %s",
3660                                      primary_slave_port_id, name);
3661                         return -1;
3662                 }
3663         } else if (arg_count > 1) {
3664                 RTE_BOND_LOG(INFO,
3665                              "Primary slave can be specified only once for bonded device %s",
3666                              name);
3667                 return -1;
3668         }
3669
3670         /* Parse link status monitor polling interval */
3671         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LSC_POLL_PERIOD_KVARG);
3672         if (arg_count == 1) {
3673                 uint32_t lsc_poll_interval_ms;
3674
3675                 if (rte_kvargs_process(kvlist,
3676                                        PMD_BOND_LSC_POLL_PERIOD_KVARG,
3677                                        &bond_ethdev_parse_time_ms_kvarg,
3678                                        &lsc_poll_interval_ms) < 0) {
3679                         RTE_BOND_LOG(INFO,
3680                                      "Invalid lsc polling interval value specified for bonded"
3681                                      " device %s", name);
3682                         return -1;
3683                 }
3684
3685                 if (rte_eth_bond_link_monitoring_set(port_id, lsc_poll_interval_ms)
3686                     != 0) {
3687                         RTE_BOND_LOG(ERR,
3688                                      "Failed to set lsc monitor polling interval (%u ms) on bonded device %s",
3689                                      lsc_poll_interval_ms, name);
3690                         return -1;
3691                 }
3692         } else if (arg_count > 1) {
3693                 RTE_BOND_LOG(INFO,
3694                              "LSC polling interval can be specified only once for bonded"
3695                              " device %s", name);
3696                 return -1;
3697         }
3698
3699         /* Parse link up interrupt propagation delay */
3700         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_UP_PROP_DELAY_KVARG);
3701         if (arg_count == 1) {
3702                 uint32_t link_up_delay_ms;
3703
3704                 if (rte_kvargs_process(kvlist,
3705                                        PMD_BOND_LINK_UP_PROP_DELAY_KVARG,
3706                                        &bond_ethdev_parse_time_ms_kvarg,
3707                                        &link_up_delay_ms) < 0) {
3708                         RTE_BOND_LOG(INFO,
3709                                      "Invalid link up propagation delay value specified for"
3710                                      " bonded device %s", name);
3711                         return -1;
3712                 }
3713
3714                 /* Set balance mode transmit policy*/
3715                 if (rte_eth_bond_link_up_prop_delay_set(port_id, link_up_delay_ms)
3716                     != 0) {
3717                         RTE_BOND_LOG(ERR,
3718                                      "Failed to set link up propagation delay (%u ms) on bonded"
3719                                      " device %s", link_up_delay_ms, name);
3720                         return -1;
3721                 }
3722         } else if (arg_count > 1) {
3723                 RTE_BOND_LOG(INFO,
3724                              "Link up propagation delay can be specified only once for"
3725                              " bonded device %s", name);
3726                 return -1;
3727         }
3728
3729         /* Parse link down interrupt propagation delay */
3730         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG);
3731         if (arg_count == 1) {
3732                 uint32_t link_down_delay_ms;
3733
3734                 if (rte_kvargs_process(kvlist,
3735                                        PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG,
3736                                        &bond_ethdev_parse_time_ms_kvarg,
3737                                        &link_down_delay_ms) < 0) {
3738                         RTE_BOND_LOG(INFO,
3739                                      "Invalid link down propagation delay value specified for"
3740                                      " bonded device %s", name);
3741                         return -1;
3742                 }
3743
3744                 /* Set balance mode transmit policy*/
3745                 if (rte_eth_bond_link_down_prop_delay_set(port_id, link_down_delay_ms)
3746                     != 0) {
3747                         RTE_BOND_LOG(ERR,
3748                                      "Failed to set link down propagation delay (%u ms) on bonded device %s",
3749                                      link_down_delay_ms, name);
3750                         return -1;
3751                 }
3752         } else if (arg_count > 1) {
3753                 RTE_BOND_LOG(INFO,
3754                              "Link down propagation delay can be specified only once for  bonded device %s",
3755                              name);
3756                 return -1;
3757         }
3758
3759         return 0;
3760 }
3761
3762 struct rte_vdev_driver pmd_bond_drv = {
3763         .probe = bond_probe,
3764         .remove = bond_remove,
3765 };
3766
3767 RTE_PMD_REGISTER_VDEV(net_bonding, pmd_bond_drv);
3768 RTE_PMD_REGISTER_ALIAS(net_bonding, eth_bond);
3769
3770 RTE_PMD_REGISTER_PARAM_STRING(net_bonding,
3771         "slave=<ifc> "
3772         "primary=<ifc> "
3773         "mode=[0-6] "
3774         "xmit_policy=[l2 | l23 | l34] "
3775         "agg_mode=[count | stable | bandwidth] "
3776         "socket_id=<int> "
3777         "mac=<mac addr> "
3778         "lsc_poll_period_ms=<int> "
3779         "up_delay=<int> "
3780         "down_delay=<int>");
3781
3782 /* We can't use RTE_LOG_REGISTER_DEFAULT because of the forced name for
3783  * this library, see meson.build.
3784  */
3785 RTE_LOG_REGISTER(bond_logtype, pmd.net.bonding, NOTICE);