lib/librte_pmd_bond/rte_eth_bond_pmd.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33 #include <stdlib.h>
  34 #include <netinet/in.h>
  35
  36 #include <rte_mbuf.h>
  37 #include <rte_malloc.h>
  38 #include <rte_ethdev.h>
  39 #include <rte_tcp.h>
  40 #include <rte_udp.h>
  41 #include <rte_ip.h>
  42 #include <rte_devargs.h>
  43 #include <rte_kvargs.h>
  44 #include <rte_dev.h>
  45 #include <rte_alarm.h>
  46 #include <rte_cycles.h>
  47
  48 #include "rte_eth_bond.h"
  49 #include "rte_eth_bond_private.h"
  50 #include "rte_eth_bond_8023ad_private.h"
  51
  52 #define REORDER_PERIOD_MS 10
  53
  54 #define HASH_L4_PORTS(h) ((h)->src_port ^ (h)->dst_port)
  55
  56 /* Table for statistics in mode 5 TLB */
  57 static uint64_t tlb_last_obytets[RTE_MAX_ETHPORTS];
  58
  59 static uint16_t
  60 bond_ethdev_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
  61 {
  62         struct bond_dev_private *internals;
  63
  64         uint16_t num_rx_slave = 0;
  65         uint16_t num_rx_total = 0;
  66
  67         int i;
  68
  69         /* Cast to structure, containing bonded device's port id and queue id */
  70         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
  71
  72         internals = bd_rx_q->dev_private;
  73
  74
  75         for (i = 0; i < internals->active_slave_count && nb_pkts; i++) {
  76                 /* Offset of pointer to *bufs increases as packets are received
  77                  * from other slaves */
  78                 num_rx_slave = rte_eth_rx_burst(internals->active_slaves[i],
  79                                 bd_rx_q->queue_id, bufs + num_rx_total, nb_pkts);
  80                 if (num_rx_slave) {
  81                         num_rx_total += num_rx_slave;
  82                         nb_pkts -= num_rx_slave;
  83                 }
  84         }
  85
  86         return num_rx_total;
  87 }
  88
  89 static uint16_t
  90 bond_ethdev_rx_burst_active_backup(void *queue, struct rte_mbuf **bufs,
  91                 uint16_t nb_pkts)
  92 {
  93         struct bond_dev_private *internals;
  94
  95         /* Cast to structure, containing bonded device's port id and queue id */
  96         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
  97
  98         internals = bd_rx_q->dev_private;
  99
 100         return rte_eth_rx_burst(internals->current_primary_port,
 101                         bd_rx_q->queue_id, bufs, nb_pkts);
 102 }
 103
 104 static uint16_t
 105 bond_ethdev_rx_burst_8023ad(void *queue, struct rte_mbuf **bufs,
 106                 uint16_t nb_pkts)
 107 {
 108         /* Cast to structure, containing bonded device's port id and queue id */
 109         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
 110         struct bond_dev_private *internals = bd_rx_q->dev_private;
 111         struct ether_addr bond_mac;
 112
 113         struct ether_hdr *hdr;
 114
 115         const uint16_t ether_type_slow_be = rte_be_to_cpu_16(ETHER_TYPE_SLOW);
 116         uint16_t num_rx_total = 0;      /* Total number of received packets */
 117         uint8_t slaves[RTE_MAX_ETHPORTS];
 118         uint8_t slave_count;
 119
 120         uint8_t collecting;  /* current slave collecting status */
 121         const uint8_t promisc = internals->promiscuous_en;
 122         uint8_t i, j, k;
 123
 124         rte_eth_macaddr_get(internals->port_id, &bond_mac);
 125         /* Copy slave list to protect against slave up/down changes during tx
 126          * bursting */
 127         slave_count = internals->active_slave_count;
 128         memcpy(slaves, internals->active_slaves,
 129                         sizeof(internals->active_slaves[0]) * slave_count);
 130
 131         for (i = 0; i < slave_count && num_rx_total < nb_pkts; i++) {
 132                 j = num_rx_total;
 133                 collecting = ACTOR_STATE(&mode_8023ad_ports[slaves[i]], COLLECTING);
 134
 135                 /* Read packets from this slave */
 136                 num_rx_total += rte_eth_rx_burst(slaves[i], bd_rx_q->queue_id,
 137                                 &bufs[num_rx_total], nb_pkts - num_rx_total);
 138
 139                 for (k = j; k < 2 && k < num_rx_total; k++)
 140                         rte_prefetch0(rte_pktmbuf_mtod(bufs[k], void *));
 141
 142                 /* Handle slow protocol packets. */
 143                 while (j < num_rx_total) {
 144                         if (j + 3 < num_rx_total)
 145                                 rte_prefetch0(rte_pktmbuf_mtod(bufs[j + 3], void *));
 146
 147                         hdr = rte_pktmbuf_mtod(bufs[j], struct ether_hdr *);
 148                         /* Remove packet from array if it is slow packet or slave is not
 149                          * in collecting state or bondign interface is not in promiscus
 150                          * mode and packet address does not match. */
 151                         if (unlikely(hdr->ether_type == ether_type_slow_be ||
 152                                 !collecting || (!promisc &&
 153                                         !is_same_ether_addr(&bond_mac, &hdr->d_addr)))) {
 154
 155                                 if (hdr->ether_type == ether_type_slow_be) {
 156                                         bond_mode_8023ad_handle_slow_pkt(internals, slaves[i],
 157                                                 bufs[j]);
 158                                 } else
 159                                         rte_pktmbuf_free(bufs[j]);
 160
 161                                 /* Packet is managed by mode 4 or dropped, shift the array */
 162                                 num_rx_total--;
 163                                 if (j < num_rx_total) {
 164                                         memmove(&bufs[j], &bufs[j + 1], sizeof(bufs[0]) *
 165                                                 (num_rx_total - j));
 166                                 }
 167                         } else
 168                                 j++;
 169                 }
 170         }
 171
 172         return num_rx_total;
 173 }
 174
 175 static uint16_t
 176 bond_ethdev_tx_burst_round_robin(void *queue, struct rte_mbuf **bufs,
 177                 uint16_t nb_pkts)
 178 {
 179         struct bond_dev_private *internals;
 180         struct bond_tx_queue *bd_tx_q;
 181
 182         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_pkts];
 183         uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
 184
 185         uint8_t num_of_slaves;
 186         uint8_t slaves[RTE_MAX_ETHPORTS];
 187
 188         uint16_t num_tx_total = 0, num_tx_slave;
 189
 190         static int slave_idx = 0;
 191         int i, cslave_idx = 0, tx_fail_total = 0;
 192
 193         bd_tx_q = (struct bond_tx_queue *)queue;
 194         internals = bd_tx_q->dev_private;
 195
 196         /* Copy slave list to protect against slave up/down changes during tx
 197          * bursting */
 198         num_of_slaves = internals->active_slave_count;
 199         memcpy(slaves, internals->active_slaves,
 200                         sizeof(internals->active_slaves[0]) * num_of_slaves);
 201
 202         if (num_of_slaves < 1)
 203                 return num_tx_total;
 204
 205         /* Populate slaves mbuf with which packets are to be sent on it  */
 206         for (i = 0; i < nb_pkts; i++) {
 207                 cslave_idx = (slave_idx + i) % num_of_slaves;
 208                 slave_bufs[cslave_idx][(slave_nb_pkts[cslave_idx])++] = bufs[i];
 209         }
 210
 211         /* increment current slave index so the next call to tx burst starts on the
 212          * next slave */
 213         slave_idx = ++cslave_idx;
 214
 215         /* Send packet burst on each slave device */
 216         for (i = 0; i < num_of_slaves; i++) {
 217                 if (slave_nb_pkts[i] > 0) {
 218                         num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
 219                                         slave_bufs[i], slave_nb_pkts[i]);
 220
 221                         /* if tx burst fails move packets to end of bufs */
 222                         if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
 223                                 int tx_fail_slave = slave_nb_pkts[i] - num_tx_slave;
 224
 225                                 tx_fail_total += tx_fail_slave;
 226
 227                                 memcpy(&bufs[nb_pkts - tx_fail_total],
 228                                                 &slave_bufs[i][num_tx_slave],
 229                                                 tx_fail_slave * sizeof(bufs[0]));
 230                         }
 231                         num_tx_total += num_tx_slave;
 232                 }
 233         }
 234
 235         return num_tx_total;
 236 }
 237
 238 static uint16_t
 239 bond_ethdev_tx_burst_active_backup(void *queue,
 240                 struct rte_mbuf **bufs, uint16_t nb_pkts)
 241 {
 242         struct bond_dev_private *internals;
 243         struct bond_tx_queue *bd_tx_q;
 244
 245         bd_tx_q = (struct bond_tx_queue *)queue;
 246         internals = bd_tx_q->dev_private;
 247
 248         if (internals->active_slave_count < 1)
 249                 return 0;
 250
 251         return rte_eth_tx_burst(internals->current_primary_port, bd_tx_q->queue_id,
 252                         bufs, nb_pkts);
 253 }
 254
 255 static inline uint16_t
 256 ether_hash(struct ether_hdr *eth_hdr)
 257 {
 258         uint16_t *word_src_addr = (uint16_t *)eth_hdr->s_addr.addr_bytes;
 259         uint16_t *word_dst_addr = (uint16_t *)eth_hdr->d_addr.addr_bytes;
 260
 261         return (word_src_addr[0] ^ word_dst_addr[0]) ^
 262                         (word_src_addr[1] ^ word_dst_addr[1]) ^
 263                         (word_src_addr[2] ^ word_dst_addr[2]);
 264 }
 265
 266 static inline uint32_t
 267 ipv4_hash(struct ipv4_hdr *ipv4_hdr)
 268 {
 269         return (ipv4_hdr->src_addr ^ ipv4_hdr->dst_addr);
 270 }
 271
 272 static inline uint32_t
 273 ipv6_hash(struct ipv6_hdr *ipv6_hdr)
 274 {
 275         uint32_t *word_src_addr = (uint32_t *)&(ipv6_hdr->src_addr[0]);
 276         uint32_t *word_dst_addr = (uint32_t *)&(ipv6_hdr->dst_addr[0]);
 277
 278         return (word_src_addr[0] ^ word_dst_addr[0]) ^
 279                         (word_src_addr[1] ^ word_dst_addr[1]) ^
 280                         (word_src_addr[2] ^ word_dst_addr[2]) ^
 281                         (word_src_addr[3] ^ word_dst_addr[3]);
 282 }
 283
 284 static inline size_t
 285 get_vlan_offset(struct ether_hdr *eth_hdr)
 286 {
 287         size_t vlan_offset = 0;
 288
 289         /* Calculate VLAN offset */
 290         if (rte_cpu_to_be_16(ETHER_TYPE_VLAN) == eth_hdr->ether_type) {
 291                 struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
 292                 vlan_offset = sizeof(struct vlan_hdr);
 293
 294                 while (rte_cpu_to_be_16(ETHER_TYPE_VLAN) ==
 295                                 vlan_hdr->eth_proto) {
 296                         vlan_hdr = vlan_hdr + 1;
 297                         vlan_offset += sizeof(struct vlan_hdr);
 298                 }
 299         }
 300         return vlan_offset;
 301 }
 302
 303 uint16_t
 304 xmit_l2_hash(const struct rte_mbuf *buf, uint8_t slave_count)
 305 {
 306         struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(buf, struct ether_hdr *);
 307
 308         uint32_t hash = ether_hash(eth_hdr);
 309
 310         return (hash ^= hash >> 8) % slave_count;
 311 }
 312
 313 uint16_t
 314 xmit_l23_hash(const struct rte_mbuf *buf, uint8_t slave_count)
 315 {
 316         struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(buf, struct ether_hdr *);
 317         size_t vlan_offset = get_vlan_offset(eth_hdr);
 318         uint32_t hash, l3hash = 0;
 319
 320         hash = ether_hash(eth_hdr);
 321
 322         if (buf->ol_flags & PKT_RX_IPV4_HDR) {
 323                 struct ipv4_hdr *ipv4_hdr = (struct ipv4_hdr *)
 324                                 ((char *)(eth_hdr + 1) + vlan_offset);
 325                 l3hash = ipv4_hash(ipv4_hdr);
 326
 327         } else if  (buf->ol_flags & PKT_RX_IPV6_HDR) {
 328                 struct ipv6_hdr *ipv6_hdr = (struct ipv6_hdr *)
 329                                 ((char *)(eth_hdr + 1) + vlan_offset);
 330                 l3hash = ipv6_hash(ipv6_hdr);
 331         }
 332
 333         hash = hash ^ l3hash;
 334         hash ^= hash >> 16;
 335         hash ^= hash >> 8;
 336
 337         return hash % slave_count;
 338 }
 339
 340 uint16_t
 341 xmit_l34_hash(const struct rte_mbuf *buf, uint8_t slave_count)
 342 {
 343         struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(buf, struct ether_hdr *);
 344         size_t vlan_offset = get_vlan_offset(eth_hdr);
 345         struct udp_hdr *udp_hdr = NULL;
 346         struct tcp_hdr *tcp_hdr = NULL;
 347         uint32_t hash, l3hash = 0, l4hash = 0;
 348
 349         if (buf->ol_flags & PKT_RX_IPV4_HDR) {
 350                 struct ipv4_hdr *ipv4_hdr = (struct ipv4_hdr *)
 351                                 ((char *)(eth_hdr + 1) + vlan_offset);
 352                 size_t ip_hdr_offset;
 353
 354                 l3hash = ipv4_hash(ipv4_hdr);
 355
 356                 ip_hdr_offset = (ipv4_hdr->version_ihl & IPV4_HDR_IHL_MASK) *
 357                                 IPV4_IHL_MULTIPLIER;
 358
 359                 if (ipv4_hdr->next_proto_id == IPPROTO_TCP) {
 360                         tcp_hdr = (struct tcp_hdr *)((char *)ipv4_hdr +
 361                                         ip_hdr_offset);
 362                         l4hash = HASH_L4_PORTS(tcp_hdr);
 363                 } else if (ipv4_hdr->next_proto_id == IPPROTO_UDP) {
 364                         udp_hdr = (struct udp_hdr *)((char *)ipv4_hdr +
 365                                         ip_hdr_offset);
 366                         l4hash = HASH_L4_PORTS(udp_hdr);
 367                 }
 368         } else if  (buf->ol_flags & PKT_RX_IPV6_HDR) {
 369                 struct ipv6_hdr *ipv6_hdr = (struct ipv6_hdr *)
 370                                 ((char *)(eth_hdr + 1) + vlan_offset);
 371                 l3hash = ipv6_hash(ipv6_hdr);
 372
 373                 if (ipv6_hdr->proto == IPPROTO_TCP) {
 374                         tcp_hdr = (struct tcp_hdr *)(ipv6_hdr + 1);
 375                         l4hash = HASH_L4_PORTS(tcp_hdr);
 376                 } else if (ipv6_hdr->proto == IPPROTO_UDP) {
 377                         udp_hdr = (struct udp_hdr *)(ipv6_hdr + 1);
 378                         l4hash = HASH_L4_PORTS(udp_hdr);
 379                 }
 380         }
 381
 382         hash = l3hash ^ l4hash;
 383         hash ^= hash >> 16;
 384         hash ^= hash >> 8;
 385
 386         return hash % slave_count;
 387 }
 388
 389 struct bwg_slave {
 390         uint64_t bwg_left_int;
 391         uint64_t bwg_left_remainder;
 392         uint8_t slave;
 393 };
 394
 395 static int
 396 bandwidth_cmp(const void *a, const void *b)
 397 {
 398         const struct bwg_slave *bwg_a = a;
 399         const struct bwg_slave *bwg_b = b;
 400         int64_t diff = (int64_t)bwg_b->bwg_left_int - (int64_t)bwg_a->bwg_left_int;
 401         int64_t diff2 = (int64_t)bwg_b->bwg_left_remainder -
 402                         (int64_t)bwg_a->bwg_left_remainder;
 403         if (diff > 0)
 404                 return 1;
 405         else if (diff < 0)
 406                 return -1;
 407         else if (diff2 > 0)
 408                 return 1;
 409         else if (diff2 < 0)
 410                 return -1;
 411         else
 412                 return 0;
 413 }
 414
 415 static void
 416 bandwidth_left(int port_id, uint64_t load, uint8_t update_idx,
 417                 struct bwg_slave *bwg_slave)
 418 {
 419         struct rte_eth_link link_status;
 420
 421         rte_eth_link_get(port_id, &link_status);
 422         uint64_t link_bwg = link_status.link_speed * 1000000ULL / 8;
 423         if (link_bwg == 0)
 424                 return;
 425         link_bwg = (link_bwg * (update_idx+1) * REORDER_PERIOD_MS);
 426         bwg_slave->bwg_left_int = (link_bwg - 1000*load) / link_bwg;
 427         bwg_slave->bwg_left_remainder = (link_bwg - 1000*load) % link_bwg;
 428 }
 429
 430 static void
 431 bond_ethdev_update_tlb_slave_cb(void *arg)
 432 {
 433         struct bond_dev_private *internals = arg;
 434         struct rte_eth_stats slave_stats;
 435         struct bwg_slave bwg_array[RTE_MAX_ETHPORTS];
 436         uint8_t slave_count;
 437         uint64_t tx_bytes;
 438
 439         uint8_t update_stats = 0;
 440         uint8_t i, slave_id;
 441
 442         internals->slave_update_idx++;
 443
 444
 445         if (internals->slave_update_idx >= REORDER_PERIOD_MS)
 446                 update_stats = 1;
 447
 448         for (i = 0; i < internals->active_slave_count; i++) {
 449                 slave_id = internals->active_slaves[i];
 450                 rte_eth_stats_get(slave_id, &slave_stats);
 451                 tx_bytes = slave_stats.obytes - tlb_last_obytets[slave_id];
 452                 bandwidth_left(slave_id, tx_bytes,
 453                                 internals->slave_update_idx, &bwg_array[i]);
 454                 bwg_array[i].slave = slave_id;
 455
 456                 if (update_stats)
 457                         tlb_last_obytets[slave_id] = slave_stats.obytes;
 458         }
 459
 460         if (update_stats == 1)
 461                 internals->slave_update_idx = 0;
 462
 463         slave_count = i;
 464         qsort(bwg_array, slave_count, sizeof(bwg_array[0]), bandwidth_cmp);
 465         for (i = 0; i < slave_count; i++)
 466                 internals->active_slaves[i] = bwg_array[i].slave;
 467
 468         rte_eal_alarm_set(REORDER_PERIOD_MS * 1000, bond_ethdev_update_tlb_slave_cb,
 469                         (struct bond_dev_private *)internals);
 470 }
 471
 472 static uint16_t
 473 bond_ethdev_tx_burst_tlb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 474 {
 475         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
 476         struct bond_dev_private *internals = bd_tx_q->dev_private;
 477
 478         struct rte_eth_dev *primary_port =
 479                         &rte_eth_devices[internals->primary_port];
 480         uint16_t num_tx_total = 0;
 481         uint8_t i, j;
 482
 483         uint8_t num_of_slaves = internals->active_slave_count;
 484         uint8_t slaves[RTE_MAX_ETHPORTS];
 485
 486         struct ether_hdr *ether_hdr;
 487         struct ether_addr primary_slave_addr;
 488         struct ether_addr active_slave_addr;
 489
 490         if (num_of_slaves < 1)
 491                 return num_tx_total;
 492
 493         memcpy(slaves, internals->active_slaves,
 494                                 sizeof(internals->active_slaves[0]) * num_of_slaves);
 495
 496
 497         ether_addr_copy(primary_port->data->mac_addrs, &primary_slave_addr);
 498
 499         if (nb_pkts > 3) {
 500                 for (i = 0; i < 3; i++)
 501                         rte_prefetch0(rte_pktmbuf_mtod(bufs[i], void*));
 502         }
 503
 504         for (i = 0; i < num_of_slaves; i++) {
 505                 ether_addr_copy(&internals->slaves[slaves[i]].persisted_mac_addr,
 506                                 &active_slave_addr);
 507
 508                 for (j = num_tx_total; j < nb_pkts; j++) {
 509                         if (j + 3 < nb_pkts)
 510                                 rte_prefetch0(rte_pktmbuf_mtod(bufs[j+3], void*));
 511
 512                         ether_hdr = rte_pktmbuf_mtod(bufs[j], struct ether_hdr *);
 513                         if (is_same_ether_addr(&ether_hdr->s_addr, &primary_slave_addr))
 514                                 ether_addr_copy(&active_slave_addr, &ether_hdr->s_addr);
 515                 }
 516
 517                 num_tx_total += rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
 518                                 bufs + num_tx_total, nb_pkts - num_tx_total);
 519
 520                 if (num_tx_total == nb_pkts)
 521                         break;
 522         }
 523
 524         return num_tx_total;
 525 }
 526
 527 static uint16_t
 528 bond_ethdev_tx_burst_balance(void *queue, struct rte_mbuf **bufs,
 529                 uint16_t nb_pkts)
 530 {
 531         struct bond_dev_private *internals;
 532         struct bond_tx_queue *bd_tx_q;
 533
 534         uint8_t num_of_slaves;
 535         uint8_t slaves[RTE_MAX_ETHPORTS];
 536
 537         uint16_t num_tx_total = 0, num_tx_slave = 0, tx_fail_total = 0;
 538
 539         int i, op_slave_id;
 540
 541         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_pkts];
 542         uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
 543
 544         bd_tx_q = (struct bond_tx_queue *)queue;
 545         internals = bd_tx_q->dev_private;
 546
 547         /* Copy slave list to protect against slave up/down changes during tx
 548          * bursting */
 549         num_of_slaves = internals->active_slave_count;
 550         memcpy(slaves, internals->active_slaves,
 551                         sizeof(internals->active_slaves[0]) * num_of_slaves);
 552
 553         if (num_of_slaves < 1)
 554                 return num_tx_total;
 555
 556         /* Populate slaves mbuf with the packets which are to be sent on it  */
 557         for (i = 0; i < nb_pkts; i++) {
 558                 /* Select output slave using hash based on xmit policy */
 559                 op_slave_id = internals->xmit_hash(bufs[i], num_of_slaves);
 560
 561                 /* Populate slave mbuf arrays with mbufs for that slave */
 562                 slave_bufs[op_slave_id][slave_nb_pkts[op_slave_id]++] = bufs[i];
 563         }
 564
 565         /* Send packet burst on each slave device */
 566         for (i = 0; i < num_of_slaves; i++) {
 567                 if (slave_nb_pkts[i] > 0) {
 568                         num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
 569                                         slave_bufs[i], slave_nb_pkts[i]);
 570
 571                         /* if tx burst fails move packets to end of bufs */
 572                         if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
 573                                 int slave_tx_fail_count = slave_nb_pkts[i] - num_tx_slave;
 574
 575                                 tx_fail_total += slave_tx_fail_count;
 576                                 memcpy(&bufs[nb_pkts - tx_fail_total],
 577                                                 &slave_bufs[i][num_tx_slave],
 578                                                 slave_tx_fail_count * sizeof(bufs[0]));
 579                         }
 580
 581                         num_tx_total += num_tx_slave;
 582                 }
 583         }
 584
 585         return num_tx_total;
 586 }
 587
 588 static uint16_t
 589 bond_ethdev_tx_burst_8023ad(void *queue, struct rte_mbuf **bufs,
 590                 uint16_t nb_pkts)
 591 {
 592         struct bond_dev_private *internals;
 593         struct bond_tx_queue *bd_tx_q;
 594
 595         uint8_t num_of_slaves;
 596         uint8_t slaves[RTE_MAX_ETHPORTS];
 597          /* positions in slaves, not ID */
 598         uint8_t distributing_offsets[RTE_MAX_ETHPORTS];
 599         uint8_t distributing_count;
 600
 601         uint16_t num_tx_slave, num_tx_total = 0, num_tx_fail_total = 0;
 602         uint16_t i, j, op_slave_idx;
 603         const uint16_t buffs_size = nb_pkts + BOND_MODE_8023AX_SLAVE_TX_PKTS + 1;
 604
 605         /* Allocate additional packets in case 8023AD mode. */
 606         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][buffs_size];
 607         void *slow_pkts[BOND_MODE_8023AX_SLAVE_TX_PKTS] = { NULL };
 608
 609         /* Total amount of packets in slave_bufs */
 610         uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
 611         /* Slow packets placed in each slave */
 612         uint8_t slave_slow_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
 613
 614         bd_tx_q = (struct bond_tx_queue *)queue;
 615         internals = bd_tx_q->dev_private;
 616
 617         /* Copy slave list to protect against slave up/down changes during tx
 618          * bursting */
 619         num_of_slaves = internals->active_slave_count;
 620         if (num_of_slaves < 1)
 621                 return num_tx_total;
 622
 623         memcpy(slaves, internals->active_slaves, sizeof(slaves[0]) * num_of_slaves);
 624
 625         distributing_count = 0;
 626         for (i = 0; i < num_of_slaves; i++) {
 627                 struct port *port = &mode_8023ad_ports[slaves[i]];
 628
 629                 slave_slow_nb_pkts[i] = rte_ring_dequeue_burst(port->tx_ring,
 630                                 slow_pkts, BOND_MODE_8023AX_SLAVE_TX_PKTS);
 631                 slave_nb_pkts[i] = slave_slow_nb_pkts[i];
 632
 633                 for (j = 0; j < slave_slow_nb_pkts[i]; j++)
 634                         slave_bufs[i][j] = slow_pkts[j];
 635
 636                 if (ACTOR_STATE(port, DISTRIBUTING))
 637                         distributing_offsets[distributing_count++] = i;
 638         }
 639
 640         if (likely(distributing_count > 0)) {
 641                 /* Populate slaves mbuf with the packets which are to be sent on it */
 642                 for (i = 0; i < nb_pkts; i++) {
 643                         /* Select output slave using hash based on xmit policy */
 644                         op_slave_idx = internals->xmit_hash(bufs[i], distributing_count);
 645
 646                         /* Populate slave mbuf arrays with mbufs for that slave. Use only
 647                          * slaves that are currently distributing. */
 648                         uint8_t slave_offset = distributing_offsets[op_slave_idx];
 649                         slave_bufs[slave_offset][slave_nb_pkts[slave_offset]] = bufs[i];
 650                         slave_nb_pkts[slave_offset]++;
 651                 }
 652         }
 653
 654         /* Send packet burst on each slave device */
 655         for (i = 0; i < num_of_slaves; i++) {
 656                 if (slave_nb_pkts[i] == 0)
 657                         continue;
 658
 659                 num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
 660                                 slave_bufs[i], slave_nb_pkts[i]);
 661
 662                 /* If tx burst fails drop slow packets */
 663                 for ( ; num_tx_slave < slave_slow_nb_pkts[i]; num_tx_slave++)
 664                         rte_pktmbuf_free(slave_bufs[i][num_tx_slave]);
 665
 666                 num_tx_total += num_tx_slave - slave_slow_nb_pkts[i];
 667                 num_tx_fail_total += slave_nb_pkts[i] - num_tx_slave;
 668
 669                 /* If tx burst fails move packets to end of bufs */
 670                 if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
 671                         uint16_t j = nb_pkts - num_tx_fail_total;
 672                         for ( ; num_tx_slave < slave_nb_pkts[i]; j++, num_tx_slave++)
 673                                 bufs[j] = slave_bufs[i][num_tx_slave];
 674                 }
 675         }
 676
 677         return num_tx_total;
 678 }
 679
 680 #ifdef RTE_MBUF_REFCNT
 681 static uint16_t
 682 bond_ethdev_tx_burst_broadcast(void *queue, struct rte_mbuf **bufs,
 683                 uint16_t nb_pkts)
 684 {
 685         struct bond_dev_private *internals;
 686         struct bond_tx_queue *bd_tx_q;
 687
 688         uint8_t tx_failed_flag = 0, num_of_slaves;
 689         uint8_t slaves[RTE_MAX_ETHPORTS];
 690
 691         uint16_t max_nb_of_tx_pkts = 0;
 692
 693         int slave_tx_total[RTE_MAX_ETHPORTS];
 694         int i, most_successful_tx_slave = -1;
 695
 696         bd_tx_q = (struct bond_tx_queue *)queue;
 697         internals = bd_tx_q->dev_private;
 698
 699         /* Copy slave list to protect against slave up/down changes during tx
 700          * bursting */
 701         num_of_slaves = internals->active_slave_count;
 702         memcpy(slaves, internals->active_slaves,
 703                         sizeof(internals->active_slaves[0]) * num_of_slaves);
 704
 705         if (num_of_slaves < 1)
 706                 return 0;
 707
 708         /* Increment reference count on mbufs */
 709         for (i = 0; i < nb_pkts; i++)
 710                 rte_mbuf_refcnt_update(bufs[i], num_of_slaves - 1);
 711
 712         /* Transmit burst on each active slave */
 713         for (i = 0; i < num_of_slaves; i++) {
 714                 slave_tx_total[i] = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
 715                                         bufs, nb_pkts);
 716
 717                 if (unlikely(slave_tx_total[i] < nb_pkts))
 718                         tx_failed_flag = 1;
 719
 720                 /* record the value and slave index for the slave which transmits the
 721                  * maximum number of packets */
 722                 if (slave_tx_total[i] > max_nb_of_tx_pkts) {
 723                         max_nb_of_tx_pkts = slave_tx_total[i];
 724                         most_successful_tx_slave = i;
 725                 }
 726         }
 727
 728         /* if slaves fail to transmit packets from burst, the calling application
 729          * is not expected to know about multiple references to packets so we must
 730          * handle failures of all packets except those of the most successful slave
 731          */
 732         if (unlikely(tx_failed_flag))
 733                 for (i = 0; i < num_of_slaves; i++)
 734                         if (i != most_successful_tx_slave)
 735                                 while (slave_tx_total[i] < nb_pkts)
 736                                         rte_pktmbuf_free(bufs[slave_tx_total[i]++]);
 737
 738         return max_nb_of_tx_pkts;
 739 }
 740 #endif
 741
 742 void
 743 link_properties_set(struct rte_eth_dev *bonded_eth_dev,
 744                 struct rte_eth_link *slave_dev_link)
 745 {
 746         struct rte_eth_link *bonded_dev_link = &bonded_eth_dev->data->dev_link;
 747         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
 748
 749         if (slave_dev_link->link_status &&
 750                 bonded_eth_dev->data->dev_started) {
 751                 bonded_dev_link->link_duplex = slave_dev_link->link_duplex;
 752                 bonded_dev_link->link_speed = slave_dev_link->link_speed;
 753
 754                 internals->link_props_set = 1;
 755         }
 756 }
 757
 758 void
 759 link_properties_reset(struct rte_eth_dev *bonded_eth_dev)
 760 {
 761         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
 762
 763         memset(&(bonded_eth_dev->data->dev_link), 0,
 764                         sizeof(bonded_eth_dev->data->dev_link));
 765
 766         internals->link_props_set = 0;
 767 }
 768
 769 int
 770 link_properties_valid(struct rte_eth_link *bonded_dev_link,
 771                 struct rte_eth_link *slave_dev_link)
 772 {
 773         if (bonded_dev_link->link_duplex != slave_dev_link->link_duplex ||
 774                 bonded_dev_link->link_speed !=  slave_dev_link->link_speed)
 775                 return -1;
 776
 777         return 0;
 778 }
 779
 780 int
 781 mac_address_get(struct rte_eth_dev *eth_dev, struct ether_addr *dst_mac_addr)
 782 {
 783         struct ether_addr *mac_addr;
 784
 785         if (eth_dev == NULL) {
 786                 RTE_LOG(ERR, PMD, "%s: NULL pointer eth_dev specified\n", __func__);
 787                 return -1;
 788         }
 789
 790         if (dst_mac_addr == NULL) {
 791                 RTE_LOG(ERR, PMD, "%s: NULL pointer MAC specified\n", __func__);
 792                 return -1;
 793         }
 794
 795         mac_addr = eth_dev->data->mac_addrs;
 796
 797         ether_addr_copy(mac_addr, dst_mac_addr);
 798         return 0;
 799 }
 800
 801 int
 802 mac_address_set(struct rte_eth_dev *eth_dev, struct ether_addr *new_mac_addr)
 803 {
 804         struct ether_addr *mac_addr;
 805
 806         if (eth_dev == NULL) {
 807                 RTE_BOND_LOG(ERR, "NULL pointer eth_dev specified");
 808                 return -1;
 809         }
 810
 811         if (new_mac_addr == NULL) {
 812                 RTE_BOND_LOG(ERR, "NULL pointer MAC specified");
 813                 return -1;
 814         }
 815
 816         mac_addr = eth_dev->data->mac_addrs;
 817
 818         /* If new MAC is different to current MAC then update */
 819         if (memcmp(mac_addr, new_mac_addr, sizeof(*mac_addr)) != 0)
 820                 memcpy(mac_addr, new_mac_addr, sizeof(*mac_addr));
 821
 822         return 0;
 823 }
 824
 825 int
 826 mac_address_slaves_update(struct rte_eth_dev *bonded_eth_dev)
 827 {
 828         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
 829         int i;
 830
 831         /* Update slave devices MAC addresses */
 832         if (internals->slave_count < 1)
 833                 return -1;
 834
 835         switch (internals->mode) {
 836         case BONDING_MODE_ROUND_ROBIN:
 837         case BONDING_MODE_BALANCE:
 838 #ifdef RTE_MBUF_REFCNT
 839         case BONDING_MODE_BROADCAST:
 840 #endif
 841                 for (i = 0; i < internals->slave_count; i++) {
 842                         if (mac_address_set(&rte_eth_devices[internals->slaves[i].port_id],
 843                                         bonded_eth_dev->data->mac_addrs)) {
 844                                 RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
 845                                                 internals->slaves[i].port_id);
 846                                 return -1;
 847                         }
 848                 }
 849                 break;
 850         case BONDING_MODE_8023AD:
 851                 bond_mode_8023ad_mac_address_update(bonded_eth_dev);
 852                 break;
 853         case BONDING_MODE_ACTIVE_BACKUP:
 854         case BONDING_MODE_ADAPTIVE_TRANSMIT_LOAD_BALANCING:
 855         default:
 856                 for (i = 0; i < internals->slave_count; i++) {
 857                         if (internals->slaves[i].port_id ==
 858                                         internals->current_primary_port) {
 859                                 if (mac_address_set(&rte_eth_devices[internals->primary_port],
 860                                                 bonded_eth_dev->data->mac_addrs)) {
 861                                         RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
 862                                                         internals->current_primary_port);
 863                                         return -1;
 864                                 }
 865                         } else {
 866                                 if (mac_address_set(
 867                                                 &rte_eth_devices[internals->slaves[i].port_id],
 868                                                 &internals->slaves[i].persisted_mac_addr)) {
 869                                         RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
 870                                                         internals->slaves[i].port_id);
 871                                         return -1;
 872                                 }
 873                         }
 874                 }
 875         }
 876
 877         return 0;
 878 }
 879
 880 int
 881 bond_ethdev_mode_set(struct rte_eth_dev *eth_dev, int mode)
 882 {
 883         struct bond_dev_private *internals;
 884
 885         internals = eth_dev->data->dev_private;
 886
 887         switch (mode) {
 888         case BONDING_MODE_ROUND_ROBIN:
 889                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_round_robin;
 890                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
 891                 break;
 892         case BONDING_MODE_ACTIVE_BACKUP:
 893                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_active_backup;
 894                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup;
 895                 break;
 896         case BONDING_MODE_BALANCE:
 897                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_balance;
 898                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
 899                 break;
 900 #ifdef RTE_MBUF_REFCNT
 901         case BONDING_MODE_BROADCAST:
 902                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_broadcast;
 903                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
 904                 break;
 905 #endif
 906         case BONDING_MODE_8023AD:
 907                 if (bond_mode_8023ad_enable(eth_dev) != 0)
 908                         return -1;
 909
 910                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_8023ad;
 911                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_8023ad;
 912                 RTE_BOND_LOG(WARNING,
 913                                 "Using mode 4, it is necessary to do TX burst and RX burst "
 914                                 "at least every 100ms.");
 915                 break;
 916         case BONDING_MODE_ADAPTIVE_TRANSMIT_LOAD_BALANCING:
 917                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_tlb;
 918                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup;
 919                 break;
 920         default:
 921                 return -1;
 922         }
 923
 924         internals->mode = mode;
 925
 926         return 0;
 927 }
 928
 929 int
 930 slave_configure(struct rte_eth_dev *bonded_eth_dev,
 931                 struct rte_eth_dev *slave_eth_dev)
 932 {
 933         struct bond_rx_queue *bd_rx_q;
 934         struct bond_tx_queue *bd_tx_q;
 935
 936         int errval, q_id;
 937
 938         /* Stop slave */
 939         rte_eth_dev_stop(slave_eth_dev->data->port_id);
 940
 941         /* Enable interrupts on slave device if supported */
 942         if (slave_eth_dev->driver->pci_drv.drv_flags & RTE_PCI_DRV_INTR_LSC)
 943                 slave_eth_dev->data->dev_conf.intr_conf.lsc = 1;
 944
 945         /* Configure device */
 946         errval = rte_eth_dev_configure(slave_eth_dev->data->port_id,
 947                         bonded_eth_dev->data->nb_rx_queues,
 948                         bonded_eth_dev->data->nb_tx_queues,
 949                         &(slave_eth_dev->data->dev_conf));
 950         if (errval != 0) {
 951                 RTE_BOND_LOG(ERR, "Cannot configure slave device: port %u , err (%d)",
 952                                 slave_eth_dev->data->port_id, errval);
 953                 return errval;
 954         }
 955
 956         /* Setup Rx Queues */
 957         for (q_id = 0; q_id < bonded_eth_dev->data->nb_rx_queues; q_id++) {
 958                 bd_rx_q = (struct bond_rx_queue *)bonded_eth_dev->data->rx_queues[q_id];
 959
 960                 errval = rte_eth_rx_queue_setup(slave_eth_dev->data->port_id, q_id,
 961                                 bd_rx_q->nb_rx_desc,
 962                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
 963                                 &(bd_rx_q->rx_conf), bd_rx_q->mb_pool);
 964                 if (errval != 0) {
 965                         RTE_BOND_LOG(ERR,
 966                                         "rte_eth_rx_queue_setup: port=%d queue_id %d, err (%d)",
 967                                         slave_eth_dev->data->port_id, q_id, errval);
 968                         return errval;
 969                 }
 970         }
 971
 972         /* Setup Tx Queues */
 973         for (q_id = 0; q_id < bonded_eth_dev->data->nb_tx_queues; q_id++) {
 974                 bd_tx_q = (struct bond_tx_queue *)bonded_eth_dev->data->tx_queues[q_id];
 975
 976                 errval = rte_eth_tx_queue_setup(slave_eth_dev->data->port_id, q_id,
 977                                 bd_tx_q->nb_tx_desc,
 978                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
 979                                 &bd_tx_q->tx_conf);
 980                 if (errval != 0) {
 981                         RTE_BOND_LOG(ERR,
 982                                         "rte_eth_tx_queue_setup: port=%d queue_id %d, err (%d)",
 983                                         slave_eth_dev->data->port_id, q_id, errval);
 984                         return errval;
 985                 }
 986         }
 987
 988         /* Start device */
 989         errval = rte_eth_dev_start(slave_eth_dev->data->port_id);
 990         if (errval != 0) {
 991                 RTE_BOND_LOG(ERR, "rte_eth_dev_start: port=%u, err (%d)",
 992                                 slave_eth_dev->data->port_id, errval);
 993                 return -1;
 994         }
 995
 996         return 0;
 997 }
 998
 999 void
1000 slave_remove(struct bond_dev_private *internals,
1001                 struct rte_eth_dev *slave_eth_dev)
1002 {
1003         uint8_t i;
1004
1005         for (i = 0; i < internals->slave_count; i++)
1006                 if (internals->slaves[i].port_id ==
1007                                 slave_eth_dev->data->port_id)
1008                         break;
1009
1010         if (i < (internals->slave_count - 1))
1011                 memmove(&internals->slaves[i], &internals->slaves[i + 1],
1012                                 sizeof(internals->slaves[0]) *
1013                                 (internals->slave_count - i - 1));
1014
1015         internals->slave_count--;
1016 }
1017
1018 static void
1019 bond_ethdev_slave_link_status_change_monitor(void *cb_arg);
1020
1021 void
1022 slave_add(struct bond_dev_private *internals,
1023                 struct rte_eth_dev *slave_eth_dev)
1024 {
1025         struct bond_slave_details *slave_details =
1026                         &internals->slaves[internals->slave_count];
1027
1028         slave_details->port_id = slave_eth_dev->data->port_id;
1029         slave_details->last_link_status = 0;
1030
1031         /* If slave device doesn't support interrupts then we need to enabled
1032          * polling to monitor link status */
1033         if (!(slave_eth_dev->pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)) {
1034                 slave_details->link_status_poll_enabled = 1;
1035
1036                 if (!internals->link_status_polling_enabled) {
1037                         internals->link_status_polling_enabled = 1;
1038
1039                         rte_eal_alarm_set(internals->link_status_polling_interval_ms * 1000,
1040                                         bond_ethdev_slave_link_status_change_monitor,
1041                                         (void *)&rte_eth_devices[internals->port_id]);
1042                 }
1043         }
1044
1045         slave_details->link_status_wait_to_complete = 0;
1046         /* clean tlb_last_obytes when adding port for bonding device */
1047         memcpy(&(slave_details->persisted_mac_addr), slave_eth_dev->data->mac_addrs,
1048                         sizeof(struct ether_addr));
1049 }
1050
1051 void
1052 bond_ethdev_primary_set(struct bond_dev_private *internals,
1053                 uint8_t slave_port_id)
1054 {
1055         int i;
1056
1057         if (internals->active_slave_count < 1)
1058                 internals->current_primary_port = slave_port_id;
1059         else
1060                 /* Search bonded device slave ports for new proposed primary port */
1061                 for (i = 0; i < internals->active_slave_count; i++) {
1062                         if (internals->active_slaves[i] == slave_port_id)
1063                                 internals->current_primary_port = slave_port_id;
1064                 }
1065 }
1066
1067 static void
1068 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev);
1069
1070 static int
1071 bond_ethdev_start(struct rte_eth_dev *eth_dev)
1072 {
1073         struct bond_dev_private *internals;
1074         int i;
1075
1076         /* slave eth dev will be started by bonded device */
1077         if (valid_bonded_ethdev(eth_dev)) {
1078                 RTE_BOND_LOG(ERR, "User tried to explicitly start a slave eth_dev (%d)",
1079                                 eth_dev->data->port_id);
1080                 return -1;
1081         }
1082
1083         eth_dev->data->dev_link.link_status = 0;
1084         eth_dev->data->dev_started = 1;
1085
1086         internals = eth_dev->data->dev_private;
1087
1088         if (internals->slave_count == 0) {
1089                 RTE_BOND_LOG(ERR, "Cannot start port since there are no slave devices");
1090                 return -1;
1091         }
1092
1093         if (internals->user_defined_mac == 0) {
1094                 struct ether_addr *new_mac_addr = NULL;
1095
1096                 for (i = 0; i < internals->slave_count; i++)
1097                         if (internals->slaves[i].port_id == internals->primary_port)
1098                                 new_mac_addr = &internals->slaves[i].persisted_mac_addr;
1099
1100                 if (new_mac_addr == NULL)
1101                         return -1;
1102
1103                 if (mac_address_set(eth_dev, new_mac_addr) != 0) {
1104                         RTE_BOND_LOG(ERR, "bonded port (%d) failed to update MAC address",
1105                                         eth_dev->data->port_id);
1106                         return -1;
1107                 }
1108         }
1109
1110         /* Update all slave devices MACs*/
1111         if (mac_address_slaves_update(eth_dev) != 0)
1112                 return -1;
1113
1114         /* If bonded device is configure in promiscuous mode then re-apply config */
1115         if (internals->promiscuous_en)
1116                 bond_ethdev_promiscuous_enable(eth_dev);
1117
1118         /* Reconfigure each slave device if starting bonded device */
1119         for (i = 0; i < internals->slave_count; i++) {
1120                 if (slave_configure(eth_dev,
1121                                 &(rte_eth_devices[internals->slaves[i].port_id])) != 0) {
1122                         RTE_BOND_LOG(ERR,
1123                                         "bonded port (%d) failed to reconfigure slave device (%d)",
1124                                         eth_dev->data->port_id, internals->slaves[i].port_id);
1125                         return -1;
1126                 }
1127         }
1128
1129         if (internals->user_defined_primary_port)
1130                 bond_ethdev_primary_set(internals, internals->primary_port);
1131
1132         if (internals->mode == BONDING_MODE_8023AD)
1133                 bond_mode_8023ad_start(eth_dev);
1134
1135         if (internals->mode == BONDING_MODE_ADAPTIVE_TRANSMIT_LOAD_BALANCING)
1136                 bond_ethdev_update_tlb_slave_cb(internals);
1137
1138         return 0;
1139 }
1140
1141 static void
1142 bond_ethdev_stop(struct rte_eth_dev *eth_dev)
1143 {
1144         struct bond_dev_private *internals = eth_dev->data->dev_private;
1145         uint8_t i;
1146
1147         if (internals->mode == BONDING_MODE_8023AD) {
1148                 struct port *port;
1149                 void *pkt = NULL;
1150
1151                 bond_mode_8023ad_stop(eth_dev);
1152
1153                 /* Discard all messages to/from mode 4 state machines */
1154                 for (i = 0; i < internals->slave_count; i++) {
1155                         port = &mode_8023ad_ports[internals->slaves[i].port_id];
1156
1157                         RTE_VERIFY(port->rx_ring != NULL);
1158                         while (rte_ring_dequeue(port->rx_ring, &pkt) != -ENOENT)
1159                                 rte_pktmbuf_free(pkt);
1160
1161                         RTE_VERIFY(port->tx_ring != NULL);
1162                         while (rte_ring_dequeue(port->tx_ring, &pkt) != -ENOENT)
1163                                 rte_pktmbuf_free(pkt);
1164                 }
1165         }
1166
1167         if (internals->mode == BONDING_MODE_ADAPTIVE_TRANSMIT_LOAD_BALANCING) {
1168                 rte_eal_alarm_cancel(bond_ethdev_update_tlb_slave_cb, internals);
1169         }
1170
1171         internals->active_slave_count = 0;
1172         internals->link_status_polling_enabled = 0;
1173
1174         eth_dev->data->dev_link.link_status = 0;
1175         eth_dev->data->dev_started = 0;
1176 }
1177
1178 static void
1179 bond_ethdev_close(struct rte_eth_dev *dev __rte_unused)
1180 {
1181 }
1182
1183 /* forward declaration */
1184 static int bond_ethdev_configure(struct rte_eth_dev *dev);
1185
1186 static void
1187 bond_ethdev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1188 {
1189         struct bond_dev_private *internals = dev->data->dev_private;
1190
1191         dev_info->driver_name = driver_name;
1192         dev_info->max_mac_addrs = 1;
1193
1194         dev_info->max_rx_pktlen = (uint32_t)2048;
1195
1196         dev_info->max_rx_queues = (uint16_t)128;
1197         dev_info->max_tx_queues = (uint16_t)512;
1198
1199         dev_info->min_rx_bufsize = 0;
1200         dev_info->pci_dev = dev->pci_dev;
1201
1202         dev_info->rx_offload_capa = internals->rx_offload_capa;
1203         dev_info->tx_offload_capa = internals->tx_offload_capa;
1204 }
1205
1206 static int
1207 bond_ethdev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1208                 uint16_t nb_rx_desc, unsigned int socket_id __rte_unused,
1209                 const struct rte_eth_rxconf *rx_conf, struct rte_mempool *mb_pool)
1210 {
1211         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)
1212                         rte_zmalloc_socket(NULL, sizeof(struct bond_rx_queue),
1213                                         0, dev->pci_dev->numa_node);
1214         if (bd_rx_q == NULL)
1215                 return -1;
1216
1217         bd_rx_q->queue_id = rx_queue_id;
1218         bd_rx_q->dev_private = dev->data->dev_private;
1219
1220         bd_rx_q->nb_rx_desc = nb_rx_desc;
1221
1222         memcpy(&(bd_rx_q->rx_conf), rx_conf, sizeof(struct rte_eth_rxconf));
1223         bd_rx_q->mb_pool = mb_pool;
1224
1225         dev->data->rx_queues[rx_queue_id] = bd_rx_q;
1226
1227         return 0;
1228 }
1229
1230 static int
1231 bond_ethdev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1232                 uint16_t nb_tx_desc, unsigned int socket_id __rte_unused,
1233                 const struct rte_eth_txconf *tx_conf)
1234 {
1235         struct bond_tx_queue *bd_tx_q  = (struct bond_tx_queue *)
1236                         rte_zmalloc_socket(NULL, sizeof(struct bond_tx_queue),
1237                                         0, dev->pci_dev->numa_node);
1238
1239         if (bd_tx_q == NULL)
1240                 return -1;
1241
1242         bd_tx_q->queue_id = tx_queue_id;
1243         bd_tx_q->dev_private = dev->data->dev_private;
1244
1245         bd_tx_q->nb_tx_desc = nb_tx_desc;
1246         memcpy(&(bd_tx_q->tx_conf), tx_conf, sizeof(bd_tx_q->tx_conf));
1247
1248         dev->data->tx_queues[tx_queue_id] = bd_tx_q;
1249
1250         return 0;
1251 }
1252
1253 static void
1254 bond_ethdev_rx_queue_release(void *queue)
1255 {
1256         if (queue == NULL)
1257                 return;
1258
1259         rte_free(queue);
1260 }
1261
1262 static void
1263 bond_ethdev_tx_queue_release(void *queue)
1264 {
1265         if (queue == NULL)
1266                 return;
1267
1268         rte_free(queue);
1269 }
1270
1271 static void
1272 bond_ethdev_slave_link_status_change_monitor(void *cb_arg)
1273 {
1274         struct rte_eth_dev *bonded_ethdev, *slave_ethdev;
1275         struct bond_dev_private *internals;
1276
1277         /* Default value for polling slave found is true as we don't want to
1278          * disable the polling thread if we cannot get the lock */
1279         int i, polling_slave_found = 1;
1280
1281         if (cb_arg == NULL)
1282                 return;
1283
1284         bonded_ethdev = (struct rte_eth_dev *)cb_arg;
1285         internals = (struct bond_dev_private *)bonded_ethdev->data->dev_private;
1286
1287         if (!bonded_ethdev->data->dev_started ||
1288                 !internals->link_status_polling_enabled)
1289                 return;
1290
1291         /* If device is currently being configured then don't check slaves link
1292          * status, wait until next period */
1293         if (rte_spinlock_trylock(&internals->lock)) {
1294                 if (internals->slave_count > 0)
1295                         polling_slave_found = 0;
1296
1297                 for (i = 0; i < internals->slave_count; i++) {
1298                         if (!internals->slaves[i].link_status_poll_enabled)
1299                                 continue;
1300
1301                         slave_ethdev = &rte_eth_devices[internals->slaves[i].port_id];
1302                         polling_slave_found = 1;
1303
1304                         /* Update slave link status */
1305                         (*slave_ethdev->dev_ops->link_update)(slave_ethdev,
1306                                         internals->slaves[i].link_status_wait_to_complete);
1307
1308                         /* if link status has changed since last checked then call lsc
1309                          * event callback */
1310                         if (slave_ethdev->data->dev_link.link_status !=
1311                                         internals->slaves[i].last_link_status) {
1312                                 internals->slaves[i].last_link_status =
1313                                                 slave_ethdev->data->dev_link.link_status;
1314
1315                                 bond_ethdev_lsc_event_callback(internals->slaves[i].port_id,
1316                                                 RTE_ETH_EVENT_INTR_LSC,
1317                                                 &bonded_ethdev->data->port_id);
1318                         }
1319                 }
1320                 rte_spinlock_unlock(&internals->lock);
1321         }
1322
1323         if (polling_slave_found)
1324                 /* Set alarm to continue monitoring link status of slave ethdev's */
1325                 rte_eal_alarm_set(internals->link_status_polling_interval_ms * 1000,
1326                                 bond_ethdev_slave_link_status_change_monitor, cb_arg);
1327 }
1328
1329 static int
1330 bond_ethdev_link_update(struct rte_eth_dev *bonded_eth_dev,
1331                 int wait_to_complete)
1332 {
1333         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1334
1335         if (!bonded_eth_dev->data->dev_started ||
1336                 internals->active_slave_count == 0) {
1337                 bonded_eth_dev->data->dev_link.link_status = 0;
1338                 return 0;
1339         } else {
1340                 struct rte_eth_dev *slave_eth_dev;
1341                 int i, link_up = 0;
1342
1343                 for (i = 0; i < internals->active_slave_count; i++) {
1344                         slave_eth_dev = &rte_eth_devices[internals->active_slaves[i]];
1345
1346                         (*slave_eth_dev->dev_ops->link_update)(slave_eth_dev,
1347                                         wait_to_complete);
1348                         if (slave_eth_dev->data->dev_link.link_status == 1) {
1349                                 link_up = 1;
1350                                 break;
1351                         }
1352                 }
1353
1354                 bonded_eth_dev->data->dev_link.link_status = link_up;
1355         }
1356
1357         return 0;
1358 }
1359
1360 static void
1361 bond_ethdev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1362 {
1363         struct bond_dev_private *internals = dev->data->dev_private;
1364         struct rte_eth_stats slave_stats;
1365         int i;
1366
1367         for (i = 0; i < internals->slave_count; i++) {
1368                 rte_eth_stats_get(internals->slaves[i].port_id, &slave_stats);
1369
1370                 stats->ipackets += slave_stats.ipackets;
1371                 stats->opackets += slave_stats.opackets;
1372                 stats->ibytes += slave_stats.ibytes;
1373                 stats->obytes += slave_stats.obytes;
1374                 stats->ierrors += slave_stats.ierrors;
1375                 stats->oerrors += slave_stats.oerrors;
1376                 stats->imcasts += slave_stats.imcasts;
1377                 stats->rx_nombuf += slave_stats.rx_nombuf;
1378                 stats->fdirmatch += slave_stats.fdirmatch;
1379                 stats->fdirmiss += slave_stats.fdirmiss;
1380                 stats->tx_pause_xon += slave_stats.tx_pause_xon;
1381                 stats->rx_pause_xon += slave_stats.rx_pause_xon;
1382                 stats->tx_pause_xoff += slave_stats.tx_pause_xoff;
1383                 stats->rx_pause_xoff += slave_stats.rx_pause_xoff;
1384         }
1385 }
1386
1387 static void
1388 bond_ethdev_stats_reset(struct rte_eth_dev *dev)
1389 {
1390         struct bond_dev_private *internals = dev->data->dev_private;
1391         int i;
1392
1393         for (i = 0; i < internals->slave_count; i++)
1394                 rte_eth_stats_reset(internals->slaves[i].port_id);
1395 }
1396
1397 static void
1398 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev)
1399 {
1400         struct bond_dev_private *internals = eth_dev->data->dev_private;
1401         int i;
1402
1403         internals->promiscuous_en = 1;
1404
1405         switch (internals->mode) {
1406         /* Promiscuous mode is propagated to all slaves */
1407         case BONDING_MODE_ROUND_ROBIN:
1408         case BONDING_MODE_BALANCE:
1409 #ifdef RTE_MBUF_REFCNT
1410         case BONDING_MODE_BROADCAST:
1411 #endif
1412                 for (i = 0; i < internals->slave_count; i++)
1413                         rte_eth_promiscuous_enable(internals->slaves[i].port_id);
1414                 break;
1415         /* In mode4 promiscus mode is managed when slave is added/removed */
1416         case BONDING_MODE_8023AD:
1417                 break;
1418         /* Promiscuous mode is propagated only to primary slave */
1419         case BONDING_MODE_ACTIVE_BACKUP:
1420         case BONDING_MODE_ADAPTIVE_TRANSMIT_LOAD_BALANCING:
1421         default:
1422                 rte_eth_promiscuous_enable(internals->current_primary_port);
1423         }
1424 }
1425
1426 static void
1427 bond_ethdev_promiscuous_disable(struct rte_eth_dev *dev)
1428 {
1429         struct bond_dev_private *internals = dev->data->dev_private;
1430         int i;
1431
1432         internals->promiscuous_en = 0;
1433
1434         switch (internals->mode) {
1435         /* Promiscuous mode is propagated to all slaves */
1436         case BONDING_MODE_ROUND_ROBIN:
1437         case BONDING_MODE_BALANCE:
1438 #ifdef RTE_MBUF_REFCNT
1439         case BONDING_MODE_BROADCAST:
1440 #endif
1441                 for (i = 0; i < internals->slave_count; i++)
1442                         rte_eth_promiscuous_disable(internals->slaves[i].port_id);
1443                 break;
1444         /* In mode4 promiscus mode is set managed when slave is added/removed */
1445         case BONDING_MODE_8023AD:
1446                 break;
1447         /* Promiscuous mode is propagated only to primary slave */
1448         case BONDING_MODE_ACTIVE_BACKUP:
1449         case BONDING_MODE_ADAPTIVE_TRANSMIT_LOAD_BALANCING:
1450         default:
1451                 rte_eth_promiscuous_disable(internals->current_primary_port);
1452         }
1453 }
1454
1455 static void
1456 bond_ethdev_delayed_lsc_propagation(void *arg)
1457 {
1458         if (arg == NULL)
1459                 return;
1460
1461         _rte_eth_dev_callback_process((struct rte_eth_dev *)arg,
1462                         RTE_ETH_EVENT_INTR_LSC);
1463 }
1464
1465 void
1466 bond_ethdev_lsc_event_callback(uint8_t port_id, enum rte_eth_event_type type,
1467                 void *param)
1468 {
1469         struct rte_eth_dev *bonded_eth_dev, *slave_eth_dev;
1470         struct bond_dev_private *internals;
1471         struct rte_eth_link link;
1472
1473         int i, valid_slave = 0;
1474         uint8_t active_pos;
1475         uint8_t lsc_flag = 0;
1476
1477         if (type != RTE_ETH_EVENT_INTR_LSC || param == NULL)
1478                 return;
1479
1480         bonded_eth_dev = &rte_eth_devices[*(uint8_t *)param];
1481         slave_eth_dev = &rte_eth_devices[port_id];
1482
1483         if (valid_bonded_ethdev(bonded_eth_dev))
1484                 return;
1485
1486         internals = bonded_eth_dev->data->dev_private;
1487
1488         /* If the device isn't started don't handle interrupts */
1489         if (!bonded_eth_dev->data->dev_started)
1490                 return;
1491
1492         /* verify that port_id is a valid slave of bonded port */
1493         for (i = 0; i < internals->slave_count; i++) {
1494                 if (internals->slaves[i].port_id == port_id) {
1495                         valid_slave = 1;
1496                         break;
1497                 }
1498         }
1499
1500         if (!valid_slave)
1501                 return;
1502
1503         /* Search for port in active port list */
1504         active_pos = find_slave_by_id(internals->active_slaves,
1505                         internals->active_slave_count, port_id);
1506
1507         rte_eth_link_get_nowait(port_id, &link);
1508         if (link.link_status) {
1509                 if (active_pos < internals->active_slave_count)
1510                         return;
1511
1512                 /* if no active slave ports then set this port to be primary port */
1513                 if (internals->active_slave_count < 1) {
1514                         /* If first active slave, then change link status */
1515                         bonded_eth_dev->data->dev_link.link_status = 1;
1516                         internals->current_primary_port = port_id;
1517                         lsc_flag = 1;
1518
1519                         mac_address_slaves_update(bonded_eth_dev);
1520
1521                         /* Inherit eth dev link properties from first active slave */
1522                         link_properties_set(bonded_eth_dev,
1523                                         &(slave_eth_dev->data->dev_link));
1524                 }
1525
1526                 activate_slave(bonded_eth_dev, port_id);
1527
1528                 /* If user has defined the primary port then default to using it */
1529                 if (internals->user_defined_primary_port &&
1530                                 internals->primary_port == port_id)
1531                         bond_ethdev_primary_set(internals, port_id);
1532         } else {
1533                 if (active_pos == internals->active_slave_count)
1534                         return;
1535
1536                 /* Remove from active slave list */
1537                 deactivate_slave(bonded_eth_dev, port_id);
1538
1539                 /* No active slaves, change link status to down and reset other
1540                  * link properties */
1541                 if (internals->active_slave_count < 1) {
1542                         lsc_flag = 1;
1543                         bonded_eth_dev->data->dev_link.link_status = 0;
1544
1545                         link_properties_reset(bonded_eth_dev);
1546                 }
1547
1548                 /* Update primary id, take first active slave from list or if none
1549                  * available set to -1 */
1550                 if (port_id == internals->current_primary_port) {
1551                         if (internals->active_slave_count > 0)
1552                                 bond_ethdev_primary_set(internals,
1553                                                 internals->active_slaves[0]);
1554                         else
1555                                 internals->current_primary_port = internals->primary_port;
1556                 }
1557         }
1558
1559         if (lsc_flag) {
1560                 /* Cancel any possible outstanding interrupts if delays are enabled */
1561                 if (internals->link_up_delay_ms > 0 ||
1562                         internals->link_down_delay_ms > 0)
1563                         rte_eal_alarm_cancel(bond_ethdev_delayed_lsc_propagation,
1564                                         bonded_eth_dev);
1565
1566                 if (bonded_eth_dev->data->dev_link.link_status) {
1567                         if (internals->link_up_delay_ms > 0)
1568                                 rte_eal_alarm_set(internals->link_up_delay_ms * 1000,
1569                                                 bond_ethdev_delayed_lsc_propagation,
1570                                                 (void *)bonded_eth_dev);
1571                         else
1572                                 _rte_eth_dev_callback_process(bonded_eth_dev,
1573                                                 RTE_ETH_EVENT_INTR_LSC);
1574
1575                 } else {
1576                         if (internals->link_down_delay_ms > 0)
1577                                 rte_eal_alarm_set(internals->link_down_delay_ms * 1000,
1578                                                 bond_ethdev_delayed_lsc_propagation,
1579                                                 (void *)bonded_eth_dev);
1580                         else
1581                                 _rte_eth_dev_callback_process(bonded_eth_dev,
1582                                                 RTE_ETH_EVENT_INTR_LSC);
1583                 }
1584         }
1585 }
1586
1587 struct eth_dev_ops default_dev_ops = {
1588                 .dev_start = bond_ethdev_start,
1589                 .dev_stop = bond_ethdev_stop,
1590                 .dev_close = bond_ethdev_close,
1591                 .dev_configure = bond_ethdev_configure,
1592                 .dev_infos_get = bond_ethdev_info,
1593                 .rx_queue_setup = bond_ethdev_rx_queue_setup,
1594                 .tx_queue_setup = bond_ethdev_tx_queue_setup,
1595                 .rx_queue_release = bond_ethdev_rx_queue_release,
1596                 .tx_queue_release = bond_ethdev_tx_queue_release,
1597                 .link_update = bond_ethdev_link_update,
1598                 .stats_get = bond_ethdev_stats_get,
1599                 .stats_reset = bond_ethdev_stats_reset,
1600                 .promiscuous_enable = bond_ethdev_promiscuous_enable,
1601                 .promiscuous_disable = bond_ethdev_promiscuous_disable
1602 };
1603
1604 static int
1605 bond_init(const char *name, const char *params)
1606 {
1607         struct bond_dev_private *internals;
1608         struct rte_kvargs *kvlist;
1609         uint8_t bonding_mode, socket_id;
1610         int  arg_count, port_id;
1611
1612         RTE_LOG(INFO, EAL, "Initializing pmd_bond for %s\n", name);
1613
1614         kvlist = rte_kvargs_parse(params, pmd_bond_init_valid_arguments);
1615         if (kvlist == NULL)
1616                 return -1;
1617
1618         /* Parse link bonding mode */
1619         if (rte_kvargs_count(kvlist, PMD_BOND_MODE_KVARG) == 1) {
1620                 if (rte_kvargs_process(kvlist, PMD_BOND_MODE_KVARG,
1621                                 &bond_ethdev_parse_slave_mode_kvarg, &bonding_mode) != 0) {
1622                         RTE_LOG(ERR, EAL, "Invalid mode for bonded device %s\n", name);
1623                         return -1;
1624                 }
1625         } else {
1626                 RTE_LOG(ERR, EAL,
1627                                 "Mode must be specified only once for bonded device %s\n",
1628                                 name);
1629                 return -1;
1630         }
1631
1632         /* Parse socket id to create bonding device on */
1633         arg_count = rte_kvargs_count(kvlist, PMD_BOND_SOCKET_ID_KVARG);
1634         if (arg_count == 1) {
1635                 if (rte_kvargs_process(kvlist, PMD_BOND_SOCKET_ID_KVARG,
1636                                 &bond_ethdev_parse_socket_id_kvarg, &socket_id) != 0) {
1637                         RTE_LOG(ERR, EAL,
1638                                         "Invalid socket Id specified for bonded device %s\n",
1639                                         name);
1640                         return -1;
1641                 }
1642         } else if (arg_count > 1) {
1643                 RTE_LOG(ERR, EAL,
1644                                 "Socket Id can be specified only once for bonded device %s\n",
1645                                 name);
1646                 return -1;
1647         } else {
1648                 socket_id = rte_socket_id();
1649         }
1650
1651         /* Create link bonding eth device */
1652         port_id = rte_eth_bond_create(name, bonding_mode, socket_id);
1653         if (port_id < 0) {
1654                 RTE_LOG(ERR, EAL,
1655                                 "Failed to create socket %s in mode %u on socket %u.\n",
1656                                 name, bonding_mode, socket_id);
1657                 return -1;
1658         }
1659         internals = rte_eth_devices[port_id].data->dev_private;
1660         internals->kvlist = kvlist;
1661
1662         RTE_LOG(INFO, EAL,
1663                         "Create bonded device %s on port %d in mode %u on socket %u.\n",
1664                         name, port_id, bonding_mode, socket_id);
1665         return 0;
1666 }
1667
1668 /* this part will resolve the slave portids after all the other pdev and vdev
1669  * have been allocated */
1670 static int
1671 bond_ethdev_configure(struct rte_eth_dev *dev)
1672 {
1673         char *name = dev->data->name;
1674         struct bond_dev_private *internals = dev->data->dev_private;
1675         struct rte_kvargs *kvlist = internals->kvlist;
1676         int arg_count, port_id = dev - rte_eth_devices;
1677
1678         /*
1679          * if no kvlist, it means that this bonded device has been created
1680          * through the bonding api.
1681          */
1682         if (!kvlist)
1683                 return 0;
1684
1685         /* Parse MAC address for bonded device */
1686         arg_count = rte_kvargs_count(kvlist, PMD_BOND_MAC_ADDR_KVARG);
1687         if (arg_count == 1) {
1688                 struct ether_addr bond_mac;
1689
1690                 if (rte_kvargs_process(kvlist, PMD_BOND_MAC_ADDR_KVARG,
1691                                 &bond_ethdev_parse_bond_mac_addr_kvarg, &bond_mac) < 0) {
1692                         RTE_LOG(INFO, EAL, "Invalid mac address for bonded device %s\n",
1693                                         name);
1694                         return -1;
1695                 }
1696
1697                 /* Set MAC address */
1698                 if (rte_eth_bond_mac_address_set(port_id, &bond_mac) != 0) {
1699                         RTE_LOG(ERR, EAL,
1700                                         "Failed to set mac address on bonded device %s\n",
1701                                         name);
1702                         return -1;
1703                 }
1704         } else if (arg_count > 1) {
1705                 RTE_LOG(ERR, EAL,
1706                                 "MAC address can be specified only once for bonded device %s\n",
1707                                 name);
1708                 return -1;
1709         }
1710
1711         /* Parse/set balance mode transmit policy */
1712         arg_count = rte_kvargs_count(kvlist, PMD_BOND_XMIT_POLICY_KVARG);
1713         if (arg_count == 1) {
1714                 uint8_t xmit_policy;
1715
1716                 if (rte_kvargs_process(kvlist, PMD_BOND_XMIT_POLICY_KVARG,
1717                                 &bond_ethdev_parse_balance_xmit_policy_kvarg, &xmit_policy) !=
1718                                                 0) {
1719                         RTE_LOG(INFO, EAL,
1720                                         "Invalid xmit policy specified for bonded device %s\n",
1721                                         name);
1722                         return -1;
1723                 }
1724
1725                 /* Set balance mode transmit policy*/
1726                 if (rte_eth_bond_xmit_policy_set(port_id, xmit_policy) != 0) {
1727                         RTE_LOG(ERR, EAL,
1728                                         "Failed to set balance xmit policy on bonded device %s\n",
1729                                         name);
1730                         return -1;
1731                 }
1732         } else if (arg_count > 1) {
1733                 RTE_LOG(ERR, EAL,
1734                                 "Transmit policy can be specified only once for bonded device"
1735                                 " %s\n", name);
1736                 return -1;
1737         }
1738
1739         /* Parse/add slave ports to bonded device */
1740         if (rte_kvargs_count(kvlist, PMD_BOND_SLAVE_PORT_KVARG) > 0) {
1741                 struct bond_ethdev_slave_ports slave_ports;
1742                 unsigned i;
1743
1744                 memset(&slave_ports, 0, sizeof(slave_ports));
1745
1746                 if (rte_kvargs_process(kvlist, PMD_BOND_SLAVE_PORT_KVARG,
1747                                 &bond_ethdev_parse_slave_port_kvarg, &slave_ports) != 0) {
1748                         RTE_LOG(ERR, EAL,
1749                                         "Failed to parse slave ports for bonded device %s\n",
1750                                         name);
1751                         return -1;
1752                 }
1753
1754                 for (i = 0; i < slave_ports.slave_count; i++) {
1755                         if (rte_eth_bond_slave_add(port_id, slave_ports.slaves[i]) != 0) {
1756                                 RTE_LOG(ERR, EAL,
1757                                                 "Failed to add port %d as slave to bonded device %s\n",
1758                                                 slave_ports.slaves[i], name);
1759                         }
1760                 }
1761
1762         } else {
1763                 RTE_LOG(INFO, EAL, "No slaves specified for bonded device %s\n", name);
1764                 return -1;
1765         }
1766
1767         /* Parse/set primary slave port id*/
1768         arg_count = rte_kvargs_count(kvlist, PMD_BOND_PRIMARY_SLAVE_KVARG);
1769         if (arg_count == 1) {
1770                 uint8_t primary_slave_port_id;
1771
1772                 if (rte_kvargs_process(kvlist,
1773                                 PMD_BOND_PRIMARY_SLAVE_KVARG,
1774                                 &bond_ethdev_parse_primary_slave_port_id_kvarg,
1775                                 &primary_slave_port_id) < 0) {
1776                         RTE_LOG(INFO, EAL,
1777                                         "Invalid primary slave port id specified for bonded device"
1778                                         " %s\n", name);
1779                         return -1;
1780                 }
1781
1782                 /* Set balance mode transmit policy*/
1783                 if (rte_eth_bond_primary_set(port_id, (uint8_t)primary_slave_port_id)
1784                                 != 0) {
1785                         RTE_LOG(ERR, EAL,
1786                                         "Failed to set primary slave port %d on bonded device %s\n",
1787                                         primary_slave_port_id, name);
1788                         return -1;
1789                 }
1790         } else if (arg_count > 1) {
1791                 RTE_LOG(INFO, EAL,
1792                                 "Primary slave can be specified only once for bonded device"
1793                                 " %s\n", name);
1794                 return -1;
1795         }
1796
1797         /* Parse link status monitor polling interval */
1798         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LSC_POLL_PERIOD_KVARG);
1799         if (arg_count == 1) {
1800                 uint32_t lsc_poll_interval_ms;
1801
1802                 if (rte_kvargs_process(kvlist,
1803                                 PMD_BOND_LSC_POLL_PERIOD_KVARG,
1804                                 &bond_ethdev_parse_time_ms_kvarg,
1805                                 &lsc_poll_interval_ms) < 0) {
1806                         RTE_LOG(INFO, EAL,
1807                                         "Invalid lsc polling interval value specified for bonded"
1808                                         " device %s\n", name);
1809                         return -1;
1810                 }
1811
1812                 if (rte_eth_bond_link_monitoring_set(port_id, lsc_poll_interval_ms)
1813                                 != 0) {
1814                         RTE_LOG(ERR, EAL,
1815                                         "Failed to set lsc monitor polling interval (%u ms) on"
1816                                         " bonded device %s\n", lsc_poll_interval_ms, name);
1817                         return -1;
1818                 }
1819         } else if (arg_count > 1) {
1820                 RTE_LOG(INFO, EAL,
1821                                 "LSC polling interval can be specified only once for bonded"
1822                                 " device %s\n", name);
1823                 return -1;
1824         }
1825
1826         /* Parse link up interrupt propagation delay */
1827         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_UP_PROP_DELAY_KVARG);
1828         if (arg_count == 1) {
1829                 uint32_t link_up_delay_ms;
1830
1831                 if (rte_kvargs_process(kvlist,
1832                                 PMD_BOND_LINK_UP_PROP_DELAY_KVARG,
1833                                 &bond_ethdev_parse_time_ms_kvarg,
1834                                 &link_up_delay_ms) < 0) {
1835                         RTE_LOG(INFO, EAL,
1836                                         "Invalid link up propagation delay value specified for"
1837                                         " bonded device %s\n", name);
1838                         return -1;
1839                 }
1840
1841                 /* Set balance mode transmit policy*/
1842                 if (rte_eth_bond_link_up_prop_delay_set(port_id, link_up_delay_ms)
1843                                 != 0) {
1844                         RTE_LOG(ERR, EAL,
1845                                         "Failed to set link up propagation delay (%u ms) on bonded"
1846                                         " device %s\n", link_up_delay_ms, name);
1847                         return -1;
1848                 }
1849         } else if (arg_count > 1) {
1850                 RTE_LOG(INFO, EAL,
1851                                 "Link up propagation delay can be specified only once for"
1852                                 " bonded device %s\n", name);
1853                 return -1;
1854         }
1855
1856         /* Parse link down interrupt propagation delay */
1857         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG);
1858         if (arg_count == 1) {
1859                 uint32_t link_down_delay_ms;
1860
1861                 if (rte_kvargs_process(kvlist,
1862                                 PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG,
1863                                 &bond_ethdev_parse_time_ms_kvarg,
1864                                 &link_down_delay_ms) < 0) {
1865                         RTE_LOG(INFO, EAL,
1866                                         "Invalid link down propagation delay value specified for"
1867                                         " bonded device %s\n", name);
1868                         return -1;
1869                 }
1870
1871                 /* Set balance mode transmit policy*/
1872                 if (rte_eth_bond_link_down_prop_delay_set(port_id, link_down_delay_ms)
1873                                 != 0) {
1874                         RTE_LOG(ERR, EAL,
1875                                         "Failed to set link down propagation delay (%u ms) on"
1876                                         " bonded device %s\n", link_down_delay_ms, name);
1877                         return -1;
1878                 }
1879         } else if (arg_count > 1) {
1880                 RTE_LOG(INFO, EAL,
1881                                 "Link down propagation delay can be specified only once for"
1882                                 " bonded device %s\n", name);
1883                 return -1;
1884         }
1885
1886         return 0;
1887 }
1888
1889 static struct rte_driver bond_drv = {
1890         .name = "eth_bond",
1891         .type = PMD_VDEV,
1892         .init = bond_init,
1893 };
1894
1895 PMD_REGISTER_DRIVER(bond_drv);