4 * Copyright(c) 2010-2017 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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
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.
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.
34 #include <netinet/in.h>
37 #include <rte_malloc.h>
38 #include <rte_ethdev.h>
39 #include <rte_ethdev_vdev.h>
43 #include <rte_ip_frag.h>
44 #include <rte_devargs.h>
45 #include <rte_kvargs.h>
47 #include <rte_alarm.h>
48 #include <rte_cycles.h>
50 #include "rte_eth_bond.h"
51 #include "rte_eth_bond_private.h"
52 #include "rte_eth_bond_8023ad_private.h"
54 #define REORDER_PERIOD_MS 10
55 #define DEFAULT_POLLING_INTERVAL_10_MS (10)
57 #define HASH_L4_PORTS(h) ((h)->src_port ^ (h)->dst_port)
59 /* Table for statistics in mode 5 TLB */
60 static uint64_t tlb_last_obytets[RTE_MAX_ETHPORTS];
63 get_vlan_offset(struct ether_hdr *eth_hdr, uint16_t *proto)
65 size_t vlan_offset = 0;
67 if (rte_cpu_to_be_16(ETHER_TYPE_VLAN) == *proto) {
68 struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
70 vlan_offset = sizeof(struct vlan_hdr);
71 *proto = vlan_hdr->eth_proto;
73 if (rte_cpu_to_be_16(ETHER_TYPE_VLAN) == *proto) {
74 vlan_hdr = vlan_hdr + 1;
75 *proto = vlan_hdr->eth_proto;
76 vlan_offset += sizeof(struct vlan_hdr);
83 bond_ethdev_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
85 struct bond_dev_private *internals;
87 uint16_t num_rx_slave = 0;
88 uint16_t num_rx_total = 0;
92 /* Cast to structure, containing bonded device's port id and queue id */
93 struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
95 internals = bd_rx_q->dev_private;
98 for (i = 0; i < internals->active_slave_count && nb_pkts; i++) {
99 /* Offset of pointer to *bufs increases as packets are received
100 * from other slaves */
101 num_rx_slave = rte_eth_rx_burst(internals->active_slaves[i],
102 bd_rx_q->queue_id, bufs + num_rx_total, nb_pkts);
104 num_rx_total += num_rx_slave;
105 nb_pkts -= num_rx_slave;
113 bond_ethdev_rx_burst_active_backup(void *queue, struct rte_mbuf **bufs,
116 struct bond_dev_private *internals;
118 /* Cast to structure, containing bonded device's port id and queue id */
119 struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
121 internals = bd_rx_q->dev_private;
123 return rte_eth_rx_burst(internals->current_primary_port,
124 bd_rx_q->queue_id, bufs, nb_pkts);
127 static inline uint8_t
128 is_lacp_packets(uint16_t ethertype, uint8_t subtype, uint16_t vlan_tci)
130 const uint16_t ether_type_slow_be = rte_be_to_cpu_16(ETHER_TYPE_SLOW);
132 return !vlan_tci && (ethertype == ether_type_slow_be &&
133 (subtype == SLOW_SUBTYPE_MARKER || subtype == SLOW_SUBTYPE_LACP));
137 bond_ethdev_rx_burst_8023ad(void *queue, struct rte_mbuf **bufs,
140 /* Cast to structure, containing bonded device's port id and queue id */
141 struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
142 struct bond_dev_private *internals = bd_rx_q->dev_private;
143 struct ether_addr bond_mac;
145 struct ether_hdr *hdr;
147 const uint16_t ether_type_slow_be = rte_be_to_cpu_16(ETHER_TYPE_SLOW);
148 uint16_t num_rx_total = 0; /* Total number of received packets */
149 uint8_t slaves[RTE_MAX_ETHPORTS];
150 uint8_t slave_count, idx;
152 uint8_t collecting; /* current slave collecting status */
153 const uint8_t promisc = internals->promiscuous_en;
157 rte_eth_macaddr_get(internals->port_id, &bond_mac);
158 /* Copy slave list to protect against slave up/down changes during tx
160 slave_count = internals->active_slave_count;
161 memcpy(slaves, internals->active_slaves,
162 sizeof(internals->active_slaves[0]) * slave_count);
164 idx = internals->active_slave;
165 if (idx >= slave_count) {
166 internals->active_slave = 0;
169 for (i = 0; i < slave_count && num_rx_total < nb_pkts; i++) {
171 collecting = ACTOR_STATE(&mode_8023ad_ports[slaves[idx]],
174 /* Read packets from this slave */
175 num_rx_total += rte_eth_rx_burst(slaves[idx], bd_rx_q->queue_id,
176 &bufs[num_rx_total], nb_pkts - num_rx_total);
178 for (k = j; k < 2 && k < num_rx_total; k++)
179 rte_prefetch0(rte_pktmbuf_mtod(bufs[k], void *));
181 /* Handle slow protocol packets. */
182 while (j < num_rx_total) {
183 if (j + 3 < num_rx_total)
184 rte_prefetch0(rte_pktmbuf_mtod(bufs[j + 3], void *));
186 hdr = rte_pktmbuf_mtod(bufs[j], struct ether_hdr *);
187 subtype = ((struct slow_protocol_frame *)hdr)->slow_protocol.subtype;
189 /* Remove packet from array if it is slow packet or slave is not
190 * in collecting state or bondign interface is not in promiscus
191 * mode and packet address does not match. */
192 if (unlikely(is_lacp_packets(hdr->ether_type, subtype, bufs[j]->vlan_tci) ||
193 !collecting || (!promisc &&
194 !is_multicast_ether_addr(&hdr->d_addr) &&
195 !is_same_ether_addr(&bond_mac, &hdr->d_addr)))) {
197 if (hdr->ether_type == ether_type_slow_be) {
198 bond_mode_8023ad_handle_slow_pkt(
199 internals, slaves[idx], bufs[j]);
201 rte_pktmbuf_free(bufs[j]);
203 /* Packet is managed by mode 4 or dropped, shift the array */
205 if (j < num_rx_total) {
206 memmove(&bufs[j], &bufs[j + 1], sizeof(bufs[0]) *
212 if (unlikely(++idx == slave_count))
216 internals->active_slave = idx;
220 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
221 uint32_t burstnumberRX;
222 uint32_t burstnumberTX;
224 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
227 arp_op_name(uint16_t arp_op, char *buf)
231 snprintf(buf, sizeof("ARP Request"), "%s", "ARP Request");
234 snprintf(buf, sizeof("ARP Reply"), "%s", "ARP Reply");
236 case ARP_OP_REVREQUEST:
237 snprintf(buf, sizeof("Reverse ARP Request"), "%s",
238 "Reverse ARP Request");
240 case ARP_OP_REVREPLY:
241 snprintf(buf, sizeof("Reverse ARP Reply"), "%s",
242 "Reverse ARP Reply");
244 case ARP_OP_INVREQUEST:
245 snprintf(buf, sizeof("Peer Identify Request"), "%s",
246 "Peer Identify Request");
248 case ARP_OP_INVREPLY:
249 snprintf(buf, sizeof("Peer Identify Reply"), "%s",
250 "Peer Identify Reply");
255 snprintf(buf, sizeof("Unknown"), "%s", "Unknown");
259 #define MaxIPv4String 16
261 ipv4_addr_to_dot(uint32_t be_ipv4_addr, char *buf, uint8_t buf_size)
265 ipv4_addr = rte_be_to_cpu_32(be_ipv4_addr);
266 snprintf(buf, buf_size, "%d.%d.%d.%d", (ipv4_addr >> 24) & 0xFF,
267 (ipv4_addr >> 16) & 0xFF, (ipv4_addr >> 8) & 0xFF,
271 #define MAX_CLIENTS_NUMBER 128
272 uint8_t active_clients;
273 struct client_stats_t {
276 uint32_t ipv4_rx_packets;
277 uint32_t ipv4_tx_packets;
279 struct client_stats_t client_stats[MAX_CLIENTS_NUMBER];
282 update_client_stats(uint32_t addr, uint8_t port, uint32_t *TXorRXindicator)
286 for (; i < MAX_CLIENTS_NUMBER; i++) {
287 if ((client_stats[i].ipv4_addr == addr) && (client_stats[i].port == port)) {
288 /* Just update RX packets number for this client */
289 if (TXorRXindicator == &burstnumberRX)
290 client_stats[i].ipv4_rx_packets++;
292 client_stats[i].ipv4_tx_packets++;
296 /* We have a new client. Insert him to the table, and increment stats */
297 if (TXorRXindicator == &burstnumberRX)
298 client_stats[active_clients].ipv4_rx_packets++;
300 client_stats[active_clients].ipv4_tx_packets++;
301 client_stats[active_clients].ipv4_addr = addr;
302 client_stats[active_clients].port = port;
307 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
308 #define MODE6_DEBUG(info, src_ip, dst_ip, eth_h, arp_op, port, burstnumber) \
309 RTE_LOG(DEBUG, PMD, \
312 "SrcMAC:%02X:%02X:%02X:%02X:%02X:%02X " \
314 "DstMAC:%02X:%02X:%02X:%02X:%02X:%02X " \
320 eth_h->s_addr.addr_bytes[0], \
321 eth_h->s_addr.addr_bytes[1], \
322 eth_h->s_addr.addr_bytes[2], \
323 eth_h->s_addr.addr_bytes[3], \
324 eth_h->s_addr.addr_bytes[4], \
325 eth_h->s_addr.addr_bytes[5], \
327 eth_h->d_addr.addr_bytes[0], \
328 eth_h->d_addr.addr_bytes[1], \
329 eth_h->d_addr.addr_bytes[2], \
330 eth_h->d_addr.addr_bytes[3], \
331 eth_h->d_addr.addr_bytes[4], \
332 eth_h->d_addr.addr_bytes[5], \
339 mode6_debug(const char __attribute__((unused)) *info, struct ether_hdr *eth_h,
340 uint8_t port, uint32_t __attribute__((unused)) *burstnumber)
342 struct ipv4_hdr *ipv4_h;
343 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
344 struct arp_hdr *arp_h;
351 uint16_t ether_type = eth_h->ether_type;
352 uint16_t offset = get_vlan_offset(eth_h, ðer_type);
354 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
355 snprintf(buf, 16, "%s", info);
358 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4)) {
359 ipv4_h = (struct ipv4_hdr *)((char *)(eth_h + 1) + offset);
360 ipv4_addr_to_dot(ipv4_h->src_addr, src_ip, MaxIPv4String);
361 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
362 ipv4_addr_to_dot(ipv4_h->dst_addr, dst_ip, MaxIPv4String);
363 MODE6_DEBUG(buf, src_ip, dst_ip, eth_h, "", port, *burstnumber);
365 update_client_stats(ipv4_h->src_addr, port, burstnumber);
367 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
368 else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_ARP)) {
369 arp_h = (struct arp_hdr *)((char *)(eth_h + 1) + offset);
370 ipv4_addr_to_dot(arp_h->arp_data.arp_sip, src_ip, MaxIPv4String);
371 ipv4_addr_to_dot(arp_h->arp_data.arp_tip, dst_ip, MaxIPv4String);
372 arp_op_name(rte_be_to_cpu_16(arp_h->arp_op), ArpOp);
373 MODE6_DEBUG(buf, src_ip, dst_ip, eth_h, ArpOp, port, *burstnumber);
380 bond_ethdev_rx_burst_alb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
382 struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
383 struct bond_dev_private *internals = bd_tx_q->dev_private;
384 struct ether_hdr *eth_h;
385 uint16_t ether_type, offset;
386 uint16_t nb_recv_pkts;
389 nb_recv_pkts = bond_ethdev_rx_burst(queue, bufs, nb_pkts);
391 for (i = 0; i < nb_recv_pkts; i++) {
392 eth_h = rte_pktmbuf_mtod(bufs[i], struct ether_hdr *);
393 ether_type = eth_h->ether_type;
394 offset = get_vlan_offset(eth_h, ðer_type);
396 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_ARP)) {
397 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
398 mode6_debug("RX ARP:", eth_h, bufs[i]->port, &burstnumberRX);
400 bond_mode_alb_arp_recv(eth_h, offset, internals);
402 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
403 else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
404 mode6_debug("RX IPv4:", eth_h, bufs[i]->port, &burstnumberRX);
412 bond_ethdev_tx_burst_round_robin(void *queue, struct rte_mbuf **bufs,
415 struct bond_dev_private *internals;
416 struct bond_tx_queue *bd_tx_q;
418 struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_pkts];
419 uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
421 uint8_t num_of_slaves;
422 uint8_t slaves[RTE_MAX_ETHPORTS];
424 uint16_t num_tx_total = 0, num_tx_slave;
426 static int slave_idx = 0;
427 int i, cslave_idx = 0, tx_fail_total = 0;
429 bd_tx_q = (struct bond_tx_queue *)queue;
430 internals = bd_tx_q->dev_private;
432 /* Copy slave list to protect against slave up/down changes during tx
434 num_of_slaves = internals->active_slave_count;
435 memcpy(slaves, internals->active_slaves,
436 sizeof(internals->active_slaves[0]) * num_of_slaves);
438 if (num_of_slaves < 1)
441 /* Populate slaves mbuf with which packets are to be sent on it */
442 for (i = 0; i < nb_pkts; i++) {
443 cslave_idx = (slave_idx + i) % num_of_slaves;
444 slave_bufs[cslave_idx][(slave_nb_pkts[cslave_idx])++] = bufs[i];
447 /* increment current slave index so the next call to tx burst starts on the
449 slave_idx = ++cslave_idx;
451 /* Send packet burst on each slave device */
452 for (i = 0; i < num_of_slaves; i++) {
453 if (slave_nb_pkts[i] > 0) {
454 num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
455 slave_bufs[i], slave_nb_pkts[i]);
457 /* if tx burst fails move packets to end of bufs */
458 if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
459 int tx_fail_slave = slave_nb_pkts[i] - num_tx_slave;
461 tx_fail_total += tx_fail_slave;
463 memcpy(&bufs[nb_pkts - tx_fail_total],
464 &slave_bufs[i][num_tx_slave],
465 tx_fail_slave * sizeof(bufs[0]));
467 num_tx_total += num_tx_slave;
475 bond_ethdev_tx_burst_active_backup(void *queue,
476 struct rte_mbuf **bufs, uint16_t nb_pkts)
478 struct bond_dev_private *internals;
479 struct bond_tx_queue *bd_tx_q;
481 bd_tx_q = (struct bond_tx_queue *)queue;
482 internals = bd_tx_q->dev_private;
484 if (internals->active_slave_count < 1)
487 return rte_eth_tx_burst(internals->current_primary_port, bd_tx_q->queue_id,
491 static inline uint16_t
492 ether_hash(struct ether_hdr *eth_hdr)
494 unaligned_uint16_t *word_src_addr =
495 (unaligned_uint16_t *)eth_hdr->s_addr.addr_bytes;
496 unaligned_uint16_t *word_dst_addr =
497 (unaligned_uint16_t *)eth_hdr->d_addr.addr_bytes;
499 return (word_src_addr[0] ^ word_dst_addr[0]) ^
500 (word_src_addr[1] ^ word_dst_addr[1]) ^
501 (word_src_addr[2] ^ word_dst_addr[2]);
504 static inline uint32_t
505 ipv4_hash(struct ipv4_hdr *ipv4_hdr)
507 return ipv4_hdr->src_addr ^ ipv4_hdr->dst_addr;
510 static inline uint32_t
511 ipv6_hash(struct ipv6_hdr *ipv6_hdr)
513 unaligned_uint32_t *word_src_addr =
514 (unaligned_uint32_t *)&(ipv6_hdr->src_addr[0]);
515 unaligned_uint32_t *word_dst_addr =
516 (unaligned_uint32_t *)&(ipv6_hdr->dst_addr[0]);
518 return (word_src_addr[0] ^ word_dst_addr[0]) ^
519 (word_src_addr[1] ^ word_dst_addr[1]) ^
520 (word_src_addr[2] ^ word_dst_addr[2]) ^
521 (word_src_addr[3] ^ word_dst_addr[3]);
525 xmit_l2_hash(const struct rte_mbuf *buf, uint8_t slave_count)
527 struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(buf, struct ether_hdr *);
529 uint32_t hash = ether_hash(eth_hdr);
531 return (hash ^= hash >> 8) % slave_count;
535 xmit_l23_hash(const struct rte_mbuf *buf, uint8_t slave_count)
537 struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(buf, struct ether_hdr *);
538 uint16_t proto = eth_hdr->ether_type;
539 size_t vlan_offset = get_vlan_offset(eth_hdr, &proto);
540 uint32_t hash, l3hash = 0;
542 hash = ether_hash(eth_hdr);
544 if (rte_cpu_to_be_16(ETHER_TYPE_IPv4) == proto) {
545 struct ipv4_hdr *ipv4_hdr = (struct ipv4_hdr *)
546 ((char *)(eth_hdr + 1) + vlan_offset);
547 l3hash = ipv4_hash(ipv4_hdr);
549 } else if (rte_cpu_to_be_16(ETHER_TYPE_IPv6) == proto) {
550 struct ipv6_hdr *ipv6_hdr = (struct ipv6_hdr *)
551 ((char *)(eth_hdr + 1) + vlan_offset);
552 l3hash = ipv6_hash(ipv6_hdr);
555 hash = hash ^ l3hash;
559 return hash % slave_count;
563 xmit_l34_hash(const struct rte_mbuf *buf, uint8_t slave_count)
565 struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(buf, struct ether_hdr *);
566 uint16_t proto = eth_hdr->ether_type;
567 size_t vlan_offset = get_vlan_offset(eth_hdr, &proto);
569 struct udp_hdr *udp_hdr = NULL;
570 struct tcp_hdr *tcp_hdr = NULL;
571 uint32_t hash, l3hash = 0, l4hash = 0;
573 if (rte_cpu_to_be_16(ETHER_TYPE_IPv4) == proto) {
574 struct ipv4_hdr *ipv4_hdr = (struct ipv4_hdr *)
575 ((char *)(eth_hdr + 1) + vlan_offset);
576 size_t ip_hdr_offset;
578 l3hash = ipv4_hash(ipv4_hdr);
580 /* there is no L4 header in fragmented packet */
581 if (likely(rte_ipv4_frag_pkt_is_fragmented(ipv4_hdr) == 0)) {
582 ip_hdr_offset = (ipv4_hdr->version_ihl & IPV4_HDR_IHL_MASK) *
585 if (ipv4_hdr->next_proto_id == IPPROTO_TCP) {
586 tcp_hdr = (struct tcp_hdr *)((char *)ipv4_hdr +
588 l4hash = HASH_L4_PORTS(tcp_hdr);
589 } else if (ipv4_hdr->next_proto_id == IPPROTO_UDP) {
590 udp_hdr = (struct udp_hdr *)((char *)ipv4_hdr +
592 l4hash = HASH_L4_PORTS(udp_hdr);
595 } else if (rte_cpu_to_be_16(ETHER_TYPE_IPv6) == proto) {
596 struct ipv6_hdr *ipv6_hdr = (struct ipv6_hdr *)
597 ((char *)(eth_hdr + 1) + vlan_offset);
598 l3hash = ipv6_hash(ipv6_hdr);
600 if (ipv6_hdr->proto == IPPROTO_TCP) {
601 tcp_hdr = (struct tcp_hdr *)(ipv6_hdr + 1);
602 l4hash = HASH_L4_PORTS(tcp_hdr);
603 } else if (ipv6_hdr->proto == IPPROTO_UDP) {
604 udp_hdr = (struct udp_hdr *)(ipv6_hdr + 1);
605 l4hash = HASH_L4_PORTS(udp_hdr);
609 hash = l3hash ^ l4hash;
613 return hash % slave_count;
617 uint64_t bwg_left_int;
618 uint64_t bwg_left_remainder;
623 bond_tlb_activate_slave(struct bond_dev_private *internals) {
626 for (i = 0; i < internals->active_slave_count; i++) {
627 tlb_last_obytets[internals->active_slaves[i]] = 0;
632 bandwidth_cmp(const void *a, const void *b)
634 const struct bwg_slave *bwg_a = a;
635 const struct bwg_slave *bwg_b = b;
636 int64_t diff = (int64_t)bwg_b->bwg_left_int - (int64_t)bwg_a->bwg_left_int;
637 int64_t diff2 = (int64_t)bwg_b->bwg_left_remainder -
638 (int64_t)bwg_a->bwg_left_remainder;
652 bandwidth_left(uint8_t port_id, uint64_t load, uint8_t update_idx,
653 struct bwg_slave *bwg_slave)
655 struct rte_eth_link link_status;
657 rte_eth_link_get(port_id, &link_status);
658 uint64_t link_bwg = link_status.link_speed * 1000000ULL / 8;
661 link_bwg = link_bwg * (update_idx+1) * REORDER_PERIOD_MS;
662 bwg_slave->bwg_left_int = (link_bwg - 1000*load) / link_bwg;
663 bwg_slave->bwg_left_remainder = (link_bwg - 1000*load) % link_bwg;
667 bond_ethdev_update_tlb_slave_cb(void *arg)
669 struct bond_dev_private *internals = arg;
670 struct rte_eth_stats slave_stats;
671 struct bwg_slave bwg_array[RTE_MAX_ETHPORTS];
675 uint8_t update_stats = 0;
678 internals->slave_update_idx++;
681 if (internals->slave_update_idx >= REORDER_PERIOD_MS)
684 for (i = 0; i < internals->active_slave_count; i++) {
685 slave_id = internals->active_slaves[i];
686 rte_eth_stats_get(slave_id, &slave_stats);
687 tx_bytes = slave_stats.obytes - tlb_last_obytets[slave_id];
688 bandwidth_left(slave_id, tx_bytes,
689 internals->slave_update_idx, &bwg_array[i]);
690 bwg_array[i].slave = slave_id;
693 tlb_last_obytets[slave_id] = slave_stats.obytes;
697 if (update_stats == 1)
698 internals->slave_update_idx = 0;
701 qsort(bwg_array, slave_count, sizeof(bwg_array[0]), bandwidth_cmp);
702 for (i = 0; i < slave_count; i++)
703 internals->tlb_slaves_order[i] = bwg_array[i].slave;
705 rte_eal_alarm_set(REORDER_PERIOD_MS * 1000, bond_ethdev_update_tlb_slave_cb,
706 (struct bond_dev_private *)internals);
710 bond_ethdev_tx_burst_tlb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
712 struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
713 struct bond_dev_private *internals = bd_tx_q->dev_private;
715 struct rte_eth_dev *primary_port =
716 &rte_eth_devices[internals->primary_port];
717 uint16_t num_tx_total = 0;
720 uint8_t num_of_slaves = internals->active_slave_count;
721 uint8_t slaves[RTE_MAX_ETHPORTS];
723 struct ether_hdr *ether_hdr;
724 struct ether_addr primary_slave_addr;
725 struct ether_addr active_slave_addr;
727 if (num_of_slaves < 1)
730 memcpy(slaves, internals->tlb_slaves_order,
731 sizeof(internals->tlb_slaves_order[0]) * num_of_slaves);
734 ether_addr_copy(primary_port->data->mac_addrs, &primary_slave_addr);
737 for (i = 0; i < 3; i++)
738 rte_prefetch0(rte_pktmbuf_mtod(bufs[i], void*));
741 for (i = 0; i < num_of_slaves; i++) {
742 rte_eth_macaddr_get(slaves[i], &active_slave_addr);
743 for (j = num_tx_total; j < nb_pkts; j++) {
745 rte_prefetch0(rte_pktmbuf_mtod(bufs[j+3], void*));
747 ether_hdr = rte_pktmbuf_mtod(bufs[j], struct ether_hdr *);
748 if (is_same_ether_addr(ðer_hdr->s_addr, &primary_slave_addr))
749 ether_addr_copy(&active_slave_addr, ðer_hdr->s_addr);
750 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
751 mode6_debug("TX IPv4:", ether_hdr, slaves[i], &burstnumberTX);
755 num_tx_total += rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
756 bufs + num_tx_total, nb_pkts - num_tx_total);
758 if (num_tx_total == nb_pkts)
766 bond_tlb_disable(struct bond_dev_private *internals)
768 rte_eal_alarm_cancel(bond_ethdev_update_tlb_slave_cb, internals);
772 bond_tlb_enable(struct bond_dev_private *internals)
774 bond_ethdev_update_tlb_slave_cb(internals);
778 bond_ethdev_tx_burst_alb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
780 struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
781 struct bond_dev_private *internals = bd_tx_q->dev_private;
783 struct ether_hdr *eth_h;
784 uint16_t ether_type, offset;
786 struct client_data *client_info;
789 * We create transmit buffers for every slave and one additional to send
790 * through tlb. In worst case every packet will be send on one port.
792 struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS + 1][nb_pkts];
793 uint16_t slave_bufs_pkts[RTE_MAX_ETHPORTS + 1] = { 0 };
796 * We create separate transmit buffers for update packets as they wont be
797 * counted in num_tx_total.
799 struct rte_mbuf *update_bufs[RTE_MAX_ETHPORTS][ALB_HASH_TABLE_SIZE];
800 uint16_t update_bufs_pkts[RTE_MAX_ETHPORTS] = { 0 };
802 struct rte_mbuf *upd_pkt;
805 uint16_t num_send, num_not_send = 0;
806 uint16_t num_tx_total = 0;
811 /* Search tx buffer for ARP packets and forward them to alb */
812 for (i = 0; i < nb_pkts; i++) {
813 eth_h = rte_pktmbuf_mtod(bufs[i], struct ether_hdr *);
814 ether_type = eth_h->ether_type;
815 offset = get_vlan_offset(eth_h, ðer_type);
817 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_ARP)) {
818 slave_idx = bond_mode_alb_arp_xmit(eth_h, offset, internals);
820 /* Change src mac in eth header */
821 rte_eth_macaddr_get(slave_idx, ð_h->s_addr);
823 /* Add packet to slave tx buffer */
824 slave_bufs[slave_idx][slave_bufs_pkts[slave_idx]] = bufs[i];
825 slave_bufs_pkts[slave_idx]++;
827 /* If packet is not ARP, send it with TLB policy */
828 slave_bufs[RTE_MAX_ETHPORTS][slave_bufs_pkts[RTE_MAX_ETHPORTS]] =
830 slave_bufs_pkts[RTE_MAX_ETHPORTS]++;
834 /* Update connected client ARP tables */
835 if (internals->mode6.ntt) {
836 for (i = 0; i < ALB_HASH_TABLE_SIZE; i++) {
837 client_info = &internals->mode6.client_table[i];
839 if (client_info->in_use) {
840 /* Allocate new packet to send ARP update on current slave */
841 upd_pkt = rte_pktmbuf_alloc(internals->mode6.mempool);
842 if (upd_pkt == NULL) {
843 RTE_LOG(ERR, PMD, "Failed to allocate ARP packet from pool\n");
846 pkt_size = sizeof(struct ether_hdr) + sizeof(struct arp_hdr)
847 + client_info->vlan_count * sizeof(struct vlan_hdr);
848 upd_pkt->data_len = pkt_size;
849 upd_pkt->pkt_len = pkt_size;
851 slave_idx = bond_mode_alb_arp_upd(client_info, upd_pkt,
854 /* Add packet to update tx buffer */
855 update_bufs[slave_idx][update_bufs_pkts[slave_idx]] = upd_pkt;
856 update_bufs_pkts[slave_idx]++;
859 internals->mode6.ntt = 0;
862 /* Send ARP packets on proper slaves */
863 for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
864 if (slave_bufs_pkts[i] > 0) {
865 num_send = rte_eth_tx_burst(i, bd_tx_q->queue_id,
866 slave_bufs[i], slave_bufs_pkts[i]);
867 for (j = 0; j < slave_bufs_pkts[i] - num_send; j++) {
868 bufs[nb_pkts - 1 - num_not_send - j] =
869 slave_bufs[i][nb_pkts - 1 - j];
872 num_tx_total += num_send;
873 num_not_send += slave_bufs_pkts[i] - num_send;
875 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
876 /* Print TX stats including update packets */
877 for (j = 0; j < slave_bufs_pkts[i]; j++) {
878 eth_h = rte_pktmbuf_mtod(slave_bufs[i][j], struct ether_hdr *);
879 mode6_debug("TX ARP:", eth_h, i, &burstnumberTX);
885 /* Send update packets on proper slaves */
886 for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
887 if (update_bufs_pkts[i] > 0) {
888 num_send = rte_eth_tx_burst(i, bd_tx_q->queue_id, update_bufs[i],
889 update_bufs_pkts[i]);
890 for (j = num_send; j < update_bufs_pkts[i]; j++) {
891 rte_pktmbuf_free(update_bufs[i][j]);
893 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
894 for (j = 0; j < update_bufs_pkts[i]; j++) {
895 eth_h = rte_pktmbuf_mtod(update_bufs[i][j], struct ether_hdr *);
896 mode6_debug("TX ARPupd:", eth_h, i, &burstnumberTX);
902 /* Send non-ARP packets using tlb policy */
903 if (slave_bufs_pkts[RTE_MAX_ETHPORTS] > 0) {
904 num_send = bond_ethdev_tx_burst_tlb(queue,
905 slave_bufs[RTE_MAX_ETHPORTS],
906 slave_bufs_pkts[RTE_MAX_ETHPORTS]);
908 for (j = 0; j < slave_bufs_pkts[RTE_MAX_ETHPORTS]; j++) {
909 bufs[nb_pkts - 1 - num_not_send - j] =
910 slave_bufs[RTE_MAX_ETHPORTS][nb_pkts - 1 - j];
913 num_tx_total += num_send;
920 bond_ethdev_tx_burst_balance(void *queue, struct rte_mbuf **bufs,
923 struct bond_dev_private *internals;
924 struct bond_tx_queue *bd_tx_q;
926 uint8_t num_of_slaves;
927 uint8_t slaves[RTE_MAX_ETHPORTS];
929 uint16_t num_tx_total = 0, num_tx_slave = 0, tx_fail_total = 0;
933 struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_pkts];
934 uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
936 bd_tx_q = (struct bond_tx_queue *)queue;
937 internals = bd_tx_q->dev_private;
939 /* Copy slave list to protect against slave up/down changes during tx
941 num_of_slaves = internals->active_slave_count;
942 memcpy(slaves, internals->active_slaves,
943 sizeof(internals->active_slaves[0]) * num_of_slaves);
945 if (num_of_slaves < 1)
948 /* Populate slaves mbuf with the packets which are to be sent on it */
949 for (i = 0; i < nb_pkts; i++) {
950 /* Select output slave using hash based on xmit policy */
951 op_slave_id = internals->xmit_hash(bufs[i], num_of_slaves);
953 /* Populate slave mbuf arrays with mbufs for that slave */
954 slave_bufs[op_slave_id][slave_nb_pkts[op_slave_id]++] = bufs[i];
957 /* Send packet burst on each slave device */
958 for (i = 0; i < num_of_slaves; i++) {
959 if (slave_nb_pkts[i] > 0) {
960 num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
961 slave_bufs[i], slave_nb_pkts[i]);
963 /* if tx burst fails move packets to end of bufs */
964 if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
965 int slave_tx_fail_count = slave_nb_pkts[i] - num_tx_slave;
967 tx_fail_total += slave_tx_fail_count;
968 memcpy(&bufs[nb_pkts - tx_fail_total],
969 &slave_bufs[i][num_tx_slave],
970 slave_tx_fail_count * sizeof(bufs[0]));
973 num_tx_total += num_tx_slave;
981 bond_ethdev_tx_burst_8023ad(void *queue, struct rte_mbuf **bufs,
984 struct bond_dev_private *internals;
985 struct bond_tx_queue *bd_tx_q;
987 uint8_t num_of_slaves;
988 uint8_t slaves[RTE_MAX_ETHPORTS];
989 /* positions in slaves, not ID */
990 uint8_t distributing_offsets[RTE_MAX_ETHPORTS];
991 uint8_t distributing_count;
993 uint16_t num_tx_slave, num_tx_total = 0, num_tx_fail_total = 0;
994 uint16_t i, j, op_slave_idx;
995 const uint16_t buffs_size = nb_pkts + BOND_MODE_8023AX_SLAVE_TX_PKTS + 1;
997 /* Allocate additional packets in case 8023AD mode. */
998 struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][buffs_size];
999 void *slow_pkts[BOND_MODE_8023AX_SLAVE_TX_PKTS] = { NULL };
1001 /* Total amount of packets in slave_bufs */
1002 uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
1003 /* Slow packets placed in each slave */
1004 uint8_t slave_slow_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
1006 bd_tx_q = (struct bond_tx_queue *)queue;
1007 internals = bd_tx_q->dev_private;
1009 /* Copy slave list to protect against slave up/down changes during tx
1011 num_of_slaves = internals->active_slave_count;
1012 if (num_of_slaves < 1)
1013 return num_tx_total;
1015 memcpy(slaves, internals->active_slaves, sizeof(slaves[0]) * num_of_slaves);
1017 distributing_count = 0;
1018 for (i = 0; i < num_of_slaves; i++) {
1019 struct port *port = &mode_8023ad_ports[slaves[i]];
1021 slave_slow_nb_pkts[i] = rte_ring_dequeue_burst(port->tx_ring,
1022 slow_pkts, BOND_MODE_8023AX_SLAVE_TX_PKTS,
1024 slave_nb_pkts[i] = slave_slow_nb_pkts[i];
1026 for (j = 0; j < slave_slow_nb_pkts[i]; j++)
1027 slave_bufs[i][j] = slow_pkts[j];
1029 if (ACTOR_STATE(port, DISTRIBUTING))
1030 distributing_offsets[distributing_count++] = i;
1033 if (likely(distributing_count > 0)) {
1034 /* Populate slaves mbuf with the packets which are to be sent on it */
1035 for (i = 0; i < nb_pkts; i++) {
1036 /* Select output slave using hash based on xmit policy */
1037 op_slave_idx = internals->xmit_hash(bufs[i], distributing_count);
1039 /* Populate slave mbuf arrays with mbufs for that slave. Use only
1040 * slaves that are currently distributing. */
1041 uint8_t slave_offset = distributing_offsets[op_slave_idx];
1042 slave_bufs[slave_offset][slave_nb_pkts[slave_offset]] = bufs[i];
1043 slave_nb_pkts[slave_offset]++;
1047 /* Send packet burst on each slave device */
1048 for (i = 0; i < num_of_slaves; i++) {
1049 if (slave_nb_pkts[i] == 0)
1052 num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
1053 slave_bufs[i], slave_nb_pkts[i]);
1055 /* If tx burst fails drop slow packets */
1056 for ( ; num_tx_slave < slave_slow_nb_pkts[i]; num_tx_slave++)
1057 rte_pktmbuf_free(slave_bufs[i][num_tx_slave]);
1059 num_tx_total += num_tx_slave - slave_slow_nb_pkts[i];
1060 num_tx_fail_total += slave_nb_pkts[i] - num_tx_slave;
1062 /* If tx burst fails move packets to end of bufs */
1063 if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
1064 uint16_t j = nb_pkts - num_tx_fail_total;
1065 for ( ; num_tx_slave < slave_nb_pkts[i]; j++, num_tx_slave++)
1066 bufs[j] = slave_bufs[i][num_tx_slave];
1070 return num_tx_total;
1074 bond_ethdev_tx_burst_broadcast(void *queue, struct rte_mbuf **bufs,
1077 struct bond_dev_private *internals;
1078 struct bond_tx_queue *bd_tx_q;
1080 uint8_t tx_failed_flag = 0, num_of_slaves;
1081 uint8_t slaves[RTE_MAX_ETHPORTS];
1083 uint16_t max_nb_of_tx_pkts = 0;
1085 int slave_tx_total[RTE_MAX_ETHPORTS];
1086 int i, most_successful_tx_slave = -1;
1088 bd_tx_q = (struct bond_tx_queue *)queue;
1089 internals = bd_tx_q->dev_private;
1091 /* Copy slave list to protect against slave up/down changes during tx
1093 num_of_slaves = internals->active_slave_count;
1094 memcpy(slaves, internals->active_slaves,
1095 sizeof(internals->active_slaves[0]) * num_of_slaves);
1097 if (num_of_slaves < 1)
1100 /* Increment reference count on mbufs */
1101 for (i = 0; i < nb_pkts; i++)
1102 rte_mbuf_refcnt_update(bufs[i], num_of_slaves - 1);
1104 /* Transmit burst on each active slave */
1105 for (i = 0; i < num_of_slaves; i++) {
1106 slave_tx_total[i] = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
1109 if (unlikely(slave_tx_total[i] < nb_pkts))
1112 /* record the value and slave index for the slave which transmits the
1113 * maximum number of packets */
1114 if (slave_tx_total[i] > max_nb_of_tx_pkts) {
1115 max_nb_of_tx_pkts = slave_tx_total[i];
1116 most_successful_tx_slave = i;
1120 /* if slaves fail to transmit packets from burst, the calling application
1121 * is not expected to know about multiple references to packets so we must
1122 * handle failures of all packets except those of the most successful slave
1124 if (unlikely(tx_failed_flag))
1125 for (i = 0; i < num_of_slaves; i++)
1126 if (i != most_successful_tx_slave)
1127 while (slave_tx_total[i] < nb_pkts)
1128 rte_pktmbuf_free(bufs[slave_tx_total[i]++]);
1130 return max_nb_of_tx_pkts;
1134 link_properties_set(struct rte_eth_dev *bonded_eth_dev,
1135 struct rte_eth_link *slave_dev_link)
1137 struct rte_eth_link *bonded_dev_link = &bonded_eth_dev->data->dev_link;
1138 struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1140 if (slave_dev_link->link_status &&
1141 bonded_eth_dev->data->dev_started) {
1142 bonded_dev_link->link_duplex = slave_dev_link->link_duplex;
1143 bonded_dev_link->link_speed = slave_dev_link->link_speed;
1145 internals->link_props_set = 1;
1150 link_properties_reset(struct rte_eth_dev *bonded_eth_dev)
1152 struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1154 memset(&(bonded_eth_dev->data->dev_link), 0,
1155 sizeof(bonded_eth_dev->data->dev_link));
1157 internals->link_props_set = 0;
1161 link_properties_valid(struct rte_eth_link *bonded_dev_link,
1162 struct rte_eth_link *slave_dev_link)
1164 if (bonded_dev_link->link_duplex != slave_dev_link->link_duplex ||
1165 bonded_dev_link->link_speed != slave_dev_link->link_speed)
1172 mac_address_get(struct rte_eth_dev *eth_dev, struct ether_addr *dst_mac_addr)
1174 struct ether_addr *mac_addr;
1176 if (eth_dev == NULL) {
1177 RTE_LOG(ERR, PMD, "%s: NULL pointer eth_dev specified\n", __func__);
1181 if (dst_mac_addr == NULL) {
1182 RTE_LOG(ERR, PMD, "%s: NULL pointer MAC specified\n", __func__);
1186 mac_addr = eth_dev->data->mac_addrs;
1188 ether_addr_copy(mac_addr, dst_mac_addr);
1193 mac_address_set(struct rte_eth_dev *eth_dev, struct ether_addr *new_mac_addr)
1195 struct ether_addr *mac_addr;
1197 if (eth_dev == NULL) {
1198 RTE_BOND_LOG(ERR, "NULL pointer eth_dev specified");
1202 if (new_mac_addr == NULL) {
1203 RTE_BOND_LOG(ERR, "NULL pointer MAC specified");
1207 mac_addr = eth_dev->data->mac_addrs;
1209 /* If new MAC is different to current MAC then update */
1210 if (memcmp(mac_addr, new_mac_addr, sizeof(*mac_addr)) != 0)
1211 memcpy(mac_addr, new_mac_addr, sizeof(*mac_addr));
1217 mac_address_slaves_update(struct rte_eth_dev *bonded_eth_dev)
1219 struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1222 /* Update slave devices MAC addresses */
1223 if (internals->slave_count < 1)
1226 switch (internals->mode) {
1227 case BONDING_MODE_ROUND_ROBIN:
1228 case BONDING_MODE_BALANCE:
1229 case BONDING_MODE_BROADCAST:
1230 for (i = 0; i < internals->slave_count; i++) {
1231 if (mac_address_set(&rte_eth_devices[internals->slaves[i].port_id],
1232 bonded_eth_dev->data->mac_addrs)) {
1233 RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1234 internals->slaves[i].port_id);
1239 case BONDING_MODE_8023AD:
1240 bond_mode_8023ad_mac_address_update(bonded_eth_dev);
1242 case BONDING_MODE_ACTIVE_BACKUP:
1243 case BONDING_MODE_TLB:
1244 case BONDING_MODE_ALB:
1246 for (i = 0; i < internals->slave_count; i++) {
1247 if (internals->slaves[i].port_id ==
1248 internals->current_primary_port) {
1249 if (mac_address_set(&rte_eth_devices[internals->primary_port],
1250 bonded_eth_dev->data->mac_addrs)) {
1251 RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1252 internals->current_primary_port);
1256 if (mac_address_set(
1257 &rte_eth_devices[internals->slaves[i].port_id],
1258 &internals->slaves[i].persisted_mac_addr)) {
1259 RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1260 internals->slaves[i].port_id);
1271 bond_ethdev_mode_set(struct rte_eth_dev *eth_dev, int mode)
1273 struct bond_dev_private *internals;
1275 internals = eth_dev->data->dev_private;
1278 case BONDING_MODE_ROUND_ROBIN:
1279 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_round_robin;
1280 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1282 case BONDING_MODE_ACTIVE_BACKUP:
1283 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_active_backup;
1284 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup;
1286 case BONDING_MODE_BALANCE:
1287 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_balance;
1288 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1290 case BONDING_MODE_BROADCAST:
1291 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_broadcast;
1292 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1294 case BONDING_MODE_8023AD:
1295 if (bond_mode_8023ad_enable(eth_dev) != 0)
1298 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_8023ad;
1299 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_8023ad;
1300 RTE_LOG(WARNING, PMD,
1301 "Using mode 4, it is necessary to do TX burst and RX burst "
1302 "at least every 100ms.\n");
1304 case BONDING_MODE_TLB:
1305 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_tlb;
1306 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup;
1308 case BONDING_MODE_ALB:
1309 if (bond_mode_alb_enable(eth_dev) != 0)
1312 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_alb;
1313 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_alb;
1319 internals->mode = mode;
1325 slave_configure(struct rte_eth_dev *bonded_eth_dev,
1326 struct rte_eth_dev *slave_eth_dev)
1328 struct bond_rx_queue *bd_rx_q;
1329 struct bond_tx_queue *bd_tx_q;
1335 rte_eth_dev_stop(slave_eth_dev->data->port_id);
1337 /* Enable interrupts on slave device if supported */
1338 if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1339 slave_eth_dev->data->dev_conf.intr_conf.lsc = 1;
1341 /* If RSS is enabled for bonding, try to enable it for slaves */
1342 if (bonded_eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG) {
1343 if (bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len
1345 slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len =
1346 bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len;
1347 slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key =
1348 bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key;
1350 slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
1353 slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf =
1354 bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
1355 slave_eth_dev->data->dev_conf.rxmode.mq_mode =
1356 bonded_eth_dev->data->dev_conf.rxmode.mq_mode;
1359 slave_eth_dev->data->dev_conf.rxmode.hw_vlan_filter =
1360 bonded_eth_dev->data->dev_conf.rxmode.hw_vlan_filter;
1362 /* Configure device */
1363 errval = rte_eth_dev_configure(slave_eth_dev->data->port_id,
1364 bonded_eth_dev->data->nb_rx_queues,
1365 bonded_eth_dev->data->nb_tx_queues,
1366 &(slave_eth_dev->data->dev_conf));
1368 RTE_BOND_LOG(ERR, "Cannot configure slave device: port %u , err (%d)",
1369 slave_eth_dev->data->port_id, errval);
1373 /* Setup Rx Queues */
1374 for (q_id = 0; q_id < bonded_eth_dev->data->nb_rx_queues; q_id++) {
1375 bd_rx_q = (struct bond_rx_queue *)bonded_eth_dev->data->rx_queues[q_id];
1377 errval = rte_eth_rx_queue_setup(slave_eth_dev->data->port_id, q_id,
1378 bd_rx_q->nb_rx_desc,
1379 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1380 &(bd_rx_q->rx_conf), bd_rx_q->mb_pool);
1383 "rte_eth_rx_queue_setup: port=%d queue_id %d, err (%d)",
1384 slave_eth_dev->data->port_id, q_id, errval);
1389 /* Setup Tx Queues */
1390 for (q_id = 0; q_id < bonded_eth_dev->data->nb_tx_queues; q_id++) {
1391 bd_tx_q = (struct bond_tx_queue *)bonded_eth_dev->data->tx_queues[q_id];
1393 errval = rte_eth_tx_queue_setup(slave_eth_dev->data->port_id, q_id,
1394 bd_tx_q->nb_tx_desc,
1395 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1399 "rte_eth_tx_queue_setup: port=%d queue_id %d, err (%d)",
1400 slave_eth_dev->data->port_id, q_id, errval);
1406 errval = rte_eth_dev_start(slave_eth_dev->data->port_id);
1408 RTE_BOND_LOG(ERR, "rte_eth_dev_start: port=%u, err (%d)",
1409 slave_eth_dev->data->port_id, errval);
1413 /* If RSS is enabled for bonding, synchronize RETA */
1414 if (bonded_eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS) {
1416 struct bond_dev_private *internals;
1418 internals = bonded_eth_dev->data->dev_private;
1420 for (i = 0; i < internals->slave_count; i++) {
1421 if (internals->slaves[i].port_id == slave_eth_dev->data->port_id) {
1422 errval = rte_eth_dev_rss_reta_update(
1423 slave_eth_dev->data->port_id,
1424 &internals->reta_conf[0],
1425 internals->slaves[i].reta_size);
1427 RTE_LOG(WARNING, PMD,
1428 "rte_eth_dev_rss_reta_update on slave port %d fails (err %d)."
1429 " RSS Configuration for bonding may be inconsistent.\n",
1430 slave_eth_dev->data->port_id, errval);
1437 /* If lsc interrupt is set, check initial slave's link status */
1438 if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1439 bond_ethdev_lsc_event_callback(slave_eth_dev->data->port_id,
1440 RTE_ETH_EVENT_INTR_LSC, &bonded_eth_dev->data->port_id);
1446 slave_remove(struct bond_dev_private *internals,
1447 struct rte_eth_dev *slave_eth_dev)
1451 for (i = 0; i < internals->slave_count; i++)
1452 if (internals->slaves[i].port_id ==
1453 slave_eth_dev->data->port_id)
1456 if (i < (internals->slave_count - 1))
1457 memmove(&internals->slaves[i], &internals->slaves[i + 1],
1458 sizeof(internals->slaves[0]) *
1459 (internals->slave_count - i - 1));
1461 internals->slave_count--;
1463 /* force reconfiguration of slave interfaces */
1464 _rte_eth_dev_reset(slave_eth_dev);
1468 bond_ethdev_slave_link_status_change_monitor(void *cb_arg);
1471 slave_add(struct bond_dev_private *internals,
1472 struct rte_eth_dev *slave_eth_dev)
1474 struct bond_slave_details *slave_details =
1475 &internals->slaves[internals->slave_count];
1477 slave_details->port_id = slave_eth_dev->data->port_id;
1478 slave_details->last_link_status = 0;
1480 /* Mark slave devices that don't support interrupts so we can
1481 * compensate when we start the bond
1483 if (!(slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1484 slave_details->link_status_poll_enabled = 1;
1487 slave_details->link_status_wait_to_complete = 0;
1488 /* clean tlb_last_obytes when adding port for bonding device */
1489 memcpy(&(slave_details->persisted_mac_addr), slave_eth_dev->data->mac_addrs,
1490 sizeof(struct ether_addr));
1494 bond_ethdev_primary_set(struct bond_dev_private *internals,
1495 uint8_t slave_port_id)
1499 if (internals->active_slave_count < 1)
1500 internals->current_primary_port = slave_port_id;
1502 /* Search bonded device slave ports for new proposed primary port */
1503 for (i = 0; i < internals->active_slave_count; i++) {
1504 if (internals->active_slaves[i] == slave_port_id)
1505 internals->current_primary_port = slave_port_id;
1510 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev);
1513 bond_ethdev_start(struct rte_eth_dev *eth_dev)
1515 struct bond_dev_private *internals;
1518 /* slave eth dev will be started by bonded device */
1519 if (check_for_bonded_ethdev(eth_dev)) {
1520 RTE_BOND_LOG(ERR, "User tried to explicitly start a slave eth_dev (%d)",
1521 eth_dev->data->port_id);
1525 eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
1526 eth_dev->data->dev_started = 1;
1528 internals = eth_dev->data->dev_private;
1530 if (internals->slave_count == 0) {
1531 RTE_BOND_LOG(ERR, "Cannot start port since there are no slave devices");
1535 if (internals->user_defined_mac == 0) {
1536 struct ether_addr *new_mac_addr = NULL;
1538 for (i = 0; i < internals->slave_count; i++)
1539 if (internals->slaves[i].port_id == internals->primary_port)
1540 new_mac_addr = &internals->slaves[i].persisted_mac_addr;
1542 if (new_mac_addr == NULL)
1545 if (mac_address_set(eth_dev, new_mac_addr) != 0) {
1546 RTE_BOND_LOG(ERR, "bonded port (%d) failed to update MAC address",
1547 eth_dev->data->port_id);
1552 /* Update all slave devices MACs*/
1553 if (mac_address_slaves_update(eth_dev) != 0)
1556 /* If bonded device is configure in promiscuous mode then re-apply config */
1557 if (internals->promiscuous_en)
1558 bond_ethdev_promiscuous_enable(eth_dev);
1560 /* Reconfigure each slave device if starting bonded device */
1561 for (i = 0; i < internals->slave_count; i++) {
1562 if (slave_configure(eth_dev,
1563 &(rte_eth_devices[internals->slaves[i].port_id])) != 0) {
1565 "bonded port (%d) failed to reconfigure slave device (%d)",
1566 eth_dev->data->port_id, internals->slaves[i].port_id);
1569 /* We will need to poll for link status if any slave doesn't
1570 * support interrupts
1572 if (internals->slaves[i].link_status_poll_enabled)
1573 internals->link_status_polling_enabled = 1;
1575 /* start polling if needed */
1576 if (internals->link_status_polling_enabled) {
1578 internals->link_status_polling_interval_ms * 1000,
1579 bond_ethdev_slave_link_status_change_monitor,
1580 (void *)&rte_eth_devices[internals->port_id]);
1583 if (internals->user_defined_primary_port)
1584 bond_ethdev_primary_set(internals, internals->primary_port);
1586 if (internals->mode == BONDING_MODE_8023AD)
1587 bond_mode_8023ad_start(eth_dev);
1589 if (internals->mode == BONDING_MODE_TLB ||
1590 internals->mode == BONDING_MODE_ALB)
1591 bond_tlb_enable(internals);
1597 bond_ethdev_free_queues(struct rte_eth_dev *dev)
1601 if (dev->data->rx_queues != NULL) {
1602 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1603 rte_free(dev->data->rx_queues[i]);
1604 dev->data->rx_queues[i] = NULL;
1606 dev->data->nb_rx_queues = 0;
1609 if (dev->data->tx_queues != NULL) {
1610 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1611 rte_free(dev->data->tx_queues[i]);
1612 dev->data->tx_queues[i] = NULL;
1614 dev->data->nb_tx_queues = 0;
1619 bond_ethdev_stop(struct rte_eth_dev *eth_dev)
1621 struct bond_dev_private *internals = eth_dev->data->dev_private;
1624 if (internals->mode == BONDING_MODE_8023AD) {
1628 bond_mode_8023ad_stop(eth_dev);
1630 /* Discard all messages to/from mode 4 state machines */
1631 for (i = 0; i < internals->active_slave_count; i++) {
1632 port = &mode_8023ad_ports[internals->active_slaves[i]];
1634 RTE_ASSERT(port->rx_ring != NULL);
1635 while (rte_ring_dequeue(port->rx_ring, &pkt) != -ENOENT)
1636 rte_pktmbuf_free(pkt);
1638 RTE_ASSERT(port->tx_ring != NULL);
1639 while (rte_ring_dequeue(port->tx_ring, &pkt) != -ENOENT)
1640 rte_pktmbuf_free(pkt);
1644 if (internals->mode == BONDING_MODE_TLB ||
1645 internals->mode == BONDING_MODE_ALB) {
1646 bond_tlb_disable(internals);
1647 for (i = 0; i < internals->active_slave_count; i++)
1648 tlb_last_obytets[internals->active_slaves[i]] = 0;
1651 internals->active_slave_count = 0;
1652 internals->link_status_polling_enabled = 0;
1653 for (i = 0; i < internals->slave_count; i++)
1654 internals->slaves[i].last_link_status = 0;
1656 eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
1657 eth_dev->data->dev_started = 0;
1661 bond_ethdev_close(struct rte_eth_dev *dev)
1663 struct bond_dev_private *internals = dev->data->dev_private;
1664 uint8_t bond_port_id = internals->port_id;
1667 RTE_LOG(INFO, EAL, "Closing bonded device %s\n", dev->data->name);
1668 while (internals->slave_count != skipped) {
1669 uint8_t port_id = internals->slaves[skipped].port_id;
1671 rte_eth_dev_stop(port_id);
1673 if (rte_eth_bond_slave_remove(bond_port_id, port_id) != 0) {
1675 "Failed to remove port %d from bonded device "
1676 "%s\n", port_id, dev->data->name);
1680 bond_ethdev_free_queues(dev);
1681 rte_bitmap_reset(internals->vlan_filter_bmp);
1684 /* forward declaration */
1685 static int bond_ethdev_configure(struct rte_eth_dev *dev);
1688 bond_ethdev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1690 struct bond_dev_private *internals = dev->data->dev_private;
1692 dev_info->max_mac_addrs = 1;
1694 dev_info->max_rx_pktlen = internals->candidate_max_rx_pktlen ?
1695 internals->candidate_max_rx_pktlen : 2048;
1697 dev_info->max_rx_queues = (uint16_t)128;
1698 dev_info->max_tx_queues = (uint16_t)512;
1700 dev_info->min_rx_bufsize = 0;
1702 dev_info->rx_offload_capa = internals->rx_offload_capa;
1703 dev_info->tx_offload_capa = internals->tx_offload_capa;
1704 dev_info->flow_type_rss_offloads = internals->flow_type_rss_offloads;
1706 dev_info->reta_size = internals->reta_size;
1710 bond_ethdev_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1714 struct bond_dev_private *internals = dev->data->dev_private;
1716 /* don't do this while a slave is being added */
1717 rte_spinlock_lock(&internals->lock);
1720 rte_bitmap_set(internals->vlan_filter_bmp, vlan_id);
1722 rte_bitmap_clear(internals->vlan_filter_bmp, vlan_id);
1724 for (i = 0; i < internals->slave_count; i++) {
1725 uint8_t port_id = internals->slaves[i].port_id;
1727 res = rte_eth_dev_vlan_filter(port_id, vlan_id, on);
1729 RTE_LOG(WARNING, PMD,
1730 "Setting VLAN filter on slave port %u not supported.\n",
1734 rte_spinlock_unlock(&internals->lock);
1739 bond_ethdev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1740 uint16_t nb_rx_desc, unsigned int socket_id __rte_unused,
1741 const struct rte_eth_rxconf *rx_conf, struct rte_mempool *mb_pool)
1743 struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)
1744 rte_zmalloc_socket(NULL, sizeof(struct bond_rx_queue),
1745 0, dev->data->numa_node);
1746 if (bd_rx_q == NULL)
1749 bd_rx_q->queue_id = rx_queue_id;
1750 bd_rx_q->dev_private = dev->data->dev_private;
1752 bd_rx_q->nb_rx_desc = nb_rx_desc;
1754 memcpy(&(bd_rx_q->rx_conf), rx_conf, sizeof(struct rte_eth_rxconf));
1755 bd_rx_q->mb_pool = mb_pool;
1757 dev->data->rx_queues[rx_queue_id] = bd_rx_q;
1763 bond_ethdev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1764 uint16_t nb_tx_desc, unsigned int socket_id __rte_unused,
1765 const struct rte_eth_txconf *tx_conf)
1767 struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)
1768 rte_zmalloc_socket(NULL, sizeof(struct bond_tx_queue),
1769 0, dev->data->numa_node);
1771 if (bd_tx_q == NULL)
1774 bd_tx_q->queue_id = tx_queue_id;
1775 bd_tx_q->dev_private = dev->data->dev_private;
1777 bd_tx_q->nb_tx_desc = nb_tx_desc;
1778 memcpy(&(bd_tx_q->tx_conf), tx_conf, sizeof(bd_tx_q->tx_conf));
1780 dev->data->tx_queues[tx_queue_id] = bd_tx_q;
1786 bond_ethdev_rx_queue_release(void *queue)
1795 bond_ethdev_tx_queue_release(void *queue)
1804 bond_ethdev_slave_link_status_change_monitor(void *cb_arg)
1806 struct rte_eth_dev *bonded_ethdev, *slave_ethdev;
1807 struct bond_dev_private *internals;
1809 /* Default value for polling slave found is true as we don't want to
1810 * disable the polling thread if we cannot get the lock */
1811 int i, polling_slave_found = 1;
1816 bonded_ethdev = (struct rte_eth_dev *)cb_arg;
1817 internals = (struct bond_dev_private *)bonded_ethdev->data->dev_private;
1819 if (!bonded_ethdev->data->dev_started ||
1820 !internals->link_status_polling_enabled)
1823 /* If device is currently being configured then don't check slaves link
1824 * status, wait until next period */
1825 if (rte_spinlock_trylock(&internals->lock)) {
1826 if (internals->slave_count > 0)
1827 polling_slave_found = 0;
1829 for (i = 0; i < internals->slave_count; i++) {
1830 if (!internals->slaves[i].link_status_poll_enabled)
1833 slave_ethdev = &rte_eth_devices[internals->slaves[i].port_id];
1834 polling_slave_found = 1;
1836 /* Update slave link status */
1837 (*slave_ethdev->dev_ops->link_update)(slave_ethdev,
1838 internals->slaves[i].link_status_wait_to_complete);
1840 /* if link status has changed since last checked then call lsc
1842 if (slave_ethdev->data->dev_link.link_status !=
1843 internals->slaves[i].last_link_status) {
1844 internals->slaves[i].last_link_status =
1845 slave_ethdev->data->dev_link.link_status;
1847 bond_ethdev_lsc_event_callback(internals->slaves[i].port_id,
1848 RTE_ETH_EVENT_INTR_LSC,
1849 &bonded_ethdev->data->port_id);
1852 rte_spinlock_unlock(&internals->lock);
1855 if (polling_slave_found)
1856 /* Set alarm to continue monitoring link status of slave ethdev's */
1857 rte_eal_alarm_set(internals->link_status_polling_interval_ms * 1000,
1858 bond_ethdev_slave_link_status_change_monitor, cb_arg);
1862 bond_ethdev_link_update(struct rte_eth_dev *bonded_eth_dev,
1863 int wait_to_complete)
1865 struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1867 if (!bonded_eth_dev->data->dev_started ||
1868 internals->active_slave_count == 0) {
1869 bonded_eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
1872 struct rte_eth_dev *slave_eth_dev;
1875 for (i = 0; i < internals->active_slave_count; i++) {
1876 slave_eth_dev = &rte_eth_devices[internals->active_slaves[i]];
1878 (*slave_eth_dev->dev_ops->link_update)(slave_eth_dev,
1880 if (slave_eth_dev->data->dev_link.link_status == ETH_LINK_UP) {
1886 bonded_eth_dev->data->dev_link.link_status = link_up;
1893 bond_ethdev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1895 struct bond_dev_private *internals = dev->data->dev_private;
1896 struct rte_eth_stats slave_stats;
1899 for (i = 0; i < internals->slave_count; i++) {
1900 rte_eth_stats_get(internals->slaves[i].port_id, &slave_stats);
1902 stats->ipackets += slave_stats.ipackets;
1903 stats->opackets += slave_stats.opackets;
1904 stats->ibytes += slave_stats.ibytes;
1905 stats->obytes += slave_stats.obytes;
1906 stats->imissed += slave_stats.imissed;
1907 stats->ierrors += slave_stats.ierrors;
1908 stats->oerrors += slave_stats.oerrors;
1909 stats->rx_nombuf += slave_stats.rx_nombuf;
1911 for (j = 0; j < RTE_ETHDEV_QUEUE_STAT_CNTRS; j++) {
1912 stats->q_ipackets[j] += slave_stats.q_ipackets[j];
1913 stats->q_opackets[j] += slave_stats.q_opackets[j];
1914 stats->q_ibytes[j] += slave_stats.q_ibytes[j];
1915 stats->q_obytes[j] += slave_stats.q_obytes[j];
1916 stats->q_errors[j] += slave_stats.q_errors[j];
1923 bond_ethdev_stats_reset(struct rte_eth_dev *dev)
1925 struct bond_dev_private *internals = dev->data->dev_private;
1928 for (i = 0; i < internals->slave_count; i++)
1929 rte_eth_stats_reset(internals->slaves[i].port_id);
1933 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev)
1935 struct bond_dev_private *internals = eth_dev->data->dev_private;
1938 internals->promiscuous_en = 1;
1940 switch (internals->mode) {
1941 /* Promiscuous mode is propagated to all slaves */
1942 case BONDING_MODE_ROUND_ROBIN:
1943 case BONDING_MODE_BALANCE:
1944 case BONDING_MODE_BROADCAST:
1945 for (i = 0; i < internals->slave_count; i++)
1946 rte_eth_promiscuous_enable(internals->slaves[i].port_id);
1948 /* In mode4 promiscus mode is managed when slave is added/removed */
1949 case BONDING_MODE_8023AD:
1951 /* Promiscuous mode is propagated only to primary slave */
1952 case BONDING_MODE_ACTIVE_BACKUP:
1953 case BONDING_MODE_TLB:
1954 case BONDING_MODE_ALB:
1956 rte_eth_promiscuous_enable(internals->current_primary_port);
1961 bond_ethdev_promiscuous_disable(struct rte_eth_dev *dev)
1963 struct bond_dev_private *internals = dev->data->dev_private;
1966 internals->promiscuous_en = 0;
1968 switch (internals->mode) {
1969 /* Promiscuous mode is propagated to all slaves */
1970 case BONDING_MODE_ROUND_ROBIN:
1971 case BONDING_MODE_BALANCE:
1972 case BONDING_MODE_BROADCAST:
1973 for (i = 0; i < internals->slave_count; i++)
1974 rte_eth_promiscuous_disable(internals->slaves[i].port_id);
1976 /* In mode4 promiscus mode is set managed when slave is added/removed */
1977 case BONDING_MODE_8023AD:
1979 /* Promiscuous mode is propagated only to primary slave */
1980 case BONDING_MODE_ACTIVE_BACKUP:
1981 case BONDING_MODE_TLB:
1982 case BONDING_MODE_ALB:
1984 rte_eth_promiscuous_disable(internals->current_primary_port);
1989 bond_ethdev_delayed_lsc_propagation(void *arg)
1994 _rte_eth_dev_callback_process((struct rte_eth_dev *)arg,
1995 RTE_ETH_EVENT_INTR_LSC, NULL);
1999 bond_ethdev_lsc_event_callback(uint8_t port_id, enum rte_eth_event_type type,
2002 struct rte_eth_dev *bonded_eth_dev, *slave_eth_dev;
2003 struct bond_dev_private *internals;
2004 struct rte_eth_link link;
2006 int i, valid_slave = 0;
2008 uint8_t lsc_flag = 0;
2010 if (type != RTE_ETH_EVENT_INTR_LSC || param == NULL)
2013 bonded_eth_dev = &rte_eth_devices[*(uint8_t *)param];
2014 slave_eth_dev = &rte_eth_devices[port_id];
2016 if (check_for_bonded_ethdev(bonded_eth_dev))
2019 internals = bonded_eth_dev->data->dev_private;
2021 /* If the device isn't started don't handle interrupts */
2022 if (!bonded_eth_dev->data->dev_started)
2025 /* verify that port_id is a valid slave of bonded port */
2026 for (i = 0; i < internals->slave_count; i++) {
2027 if (internals->slaves[i].port_id == port_id) {
2036 /* Search for port in active port list */
2037 active_pos = find_slave_by_id(internals->active_slaves,
2038 internals->active_slave_count, port_id);
2040 rte_eth_link_get_nowait(port_id, &link);
2041 if (link.link_status) {
2042 if (active_pos < internals->active_slave_count)
2045 /* if no active slave ports then set this port to be primary port */
2046 if (internals->active_slave_count < 1) {
2047 /* If first active slave, then change link status */
2048 bonded_eth_dev->data->dev_link.link_status = ETH_LINK_UP;
2049 internals->current_primary_port = port_id;
2052 mac_address_slaves_update(bonded_eth_dev);
2054 /* Inherit eth dev link properties from first active slave */
2055 link_properties_set(bonded_eth_dev,
2056 &(slave_eth_dev->data->dev_link));
2058 if (link_properties_valid(
2059 &bonded_eth_dev->data->dev_link, &link) != 0) {
2060 slave_eth_dev->data->dev_flags &=
2061 (~RTE_ETH_DEV_BONDED_SLAVE);
2063 "port %u invalid speed/duplex\n",
2069 activate_slave(bonded_eth_dev, port_id);
2071 /* If user has defined the primary port then default to using it */
2072 if (internals->user_defined_primary_port &&
2073 internals->primary_port == port_id)
2074 bond_ethdev_primary_set(internals, port_id);
2076 if (active_pos == internals->active_slave_count)
2079 /* Remove from active slave list */
2080 deactivate_slave(bonded_eth_dev, port_id);
2082 /* No active slaves, change link status to down and reset other
2083 * link properties */
2084 if (internals->active_slave_count < 1) {
2086 bonded_eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
2088 link_properties_reset(bonded_eth_dev);
2091 /* Update primary id, take first active slave from list or if none
2092 * available set to -1 */
2093 if (port_id == internals->current_primary_port) {
2094 if (internals->active_slave_count > 0)
2095 bond_ethdev_primary_set(internals,
2096 internals->active_slaves[0]);
2098 internals->current_primary_port = internals->primary_port;
2103 /* Cancel any possible outstanding interrupts if delays are enabled */
2104 if (internals->link_up_delay_ms > 0 ||
2105 internals->link_down_delay_ms > 0)
2106 rte_eal_alarm_cancel(bond_ethdev_delayed_lsc_propagation,
2109 if (bonded_eth_dev->data->dev_link.link_status) {
2110 if (internals->link_up_delay_ms > 0)
2111 rte_eal_alarm_set(internals->link_up_delay_ms * 1000,
2112 bond_ethdev_delayed_lsc_propagation,
2113 (void *)bonded_eth_dev);
2115 _rte_eth_dev_callback_process(bonded_eth_dev,
2116 RTE_ETH_EVENT_INTR_LSC, NULL);
2119 if (internals->link_down_delay_ms > 0)
2120 rte_eal_alarm_set(internals->link_down_delay_ms * 1000,
2121 bond_ethdev_delayed_lsc_propagation,
2122 (void *)bonded_eth_dev);
2124 _rte_eth_dev_callback_process(bonded_eth_dev,
2125 RTE_ETH_EVENT_INTR_LSC, NULL);
2131 bond_ethdev_rss_reta_update(struct rte_eth_dev *dev,
2132 struct rte_eth_rss_reta_entry64 *reta_conf, uint16_t reta_size)
2136 int slave_reta_size;
2137 unsigned reta_count;
2138 struct bond_dev_private *internals = dev->data->dev_private;
2140 if (reta_size != internals->reta_size)
2143 /* Copy RETA table */
2144 reta_count = reta_size / RTE_RETA_GROUP_SIZE;
2146 for (i = 0; i < reta_count; i++) {
2147 internals->reta_conf[i].mask = reta_conf[i].mask;
2148 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
2149 if ((reta_conf[i].mask >> j) & 0x01)
2150 internals->reta_conf[i].reta[j] = reta_conf[i].reta[j];
2153 /* Fill rest of array */
2154 for (; i < RTE_DIM(internals->reta_conf); i += reta_count)
2155 memcpy(&internals->reta_conf[i], &internals->reta_conf[0],
2156 sizeof(internals->reta_conf[0]) * reta_count);
2158 /* Propagate RETA over slaves */
2159 for (i = 0; i < internals->slave_count; i++) {
2160 slave_reta_size = internals->slaves[i].reta_size;
2161 result = rte_eth_dev_rss_reta_update(internals->slaves[i].port_id,
2162 &internals->reta_conf[0], slave_reta_size);
2171 bond_ethdev_rss_reta_query(struct rte_eth_dev *dev,
2172 struct rte_eth_rss_reta_entry64 *reta_conf, uint16_t reta_size)
2175 struct bond_dev_private *internals = dev->data->dev_private;
2177 if (reta_size != internals->reta_size)
2180 /* Copy RETA table */
2181 for (i = 0; i < reta_size / RTE_RETA_GROUP_SIZE; i++)
2182 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
2183 if ((reta_conf[i].mask >> j) & 0x01)
2184 reta_conf[i].reta[j] = internals->reta_conf[i].reta[j];
2190 bond_ethdev_rss_hash_update(struct rte_eth_dev *dev,
2191 struct rte_eth_rss_conf *rss_conf)
2194 struct bond_dev_private *internals = dev->data->dev_private;
2195 struct rte_eth_rss_conf bond_rss_conf;
2197 memcpy(&bond_rss_conf, rss_conf, sizeof(struct rte_eth_rss_conf));
2199 bond_rss_conf.rss_hf &= internals->flow_type_rss_offloads;
2201 if (bond_rss_conf.rss_hf != 0)
2202 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf = bond_rss_conf.rss_hf;
2204 if (bond_rss_conf.rss_key && bond_rss_conf.rss_key_len <
2205 sizeof(internals->rss_key)) {
2206 if (bond_rss_conf.rss_key_len == 0)
2207 bond_rss_conf.rss_key_len = 40;
2208 internals->rss_key_len = bond_rss_conf.rss_key_len;
2209 memcpy(internals->rss_key, bond_rss_conf.rss_key,
2210 internals->rss_key_len);
2213 for (i = 0; i < internals->slave_count; i++) {
2214 result = rte_eth_dev_rss_hash_update(internals->slaves[i].port_id,
2224 bond_ethdev_rss_hash_conf_get(struct rte_eth_dev *dev,
2225 struct rte_eth_rss_conf *rss_conf)
2227 struct bond_dev_private *internals = dev->data->dev_private;
2229 rss_conf->rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
2230 rss_conf->rss_key_len = internals->rss_key_len;
2231 if (rss_conf->rss_key)
2232 memcpy(rss_conf->rss_key, internals->rss_key, internals->rss_key_len);
2237 const struct eth_dev_ops default_dev_ops = {
2238 .dev_start = bond_ethdev_start,
2239 .dev_stop = bond_ethdev_stop,
2240 .dev_close = bond_ethdev_close,
2241 .dev_configure = bond_ethdev_configure,
2242 .dev_infos_get = bond_ethdev_info,
2243 .vlan_filter_set = bond_ethdev_vlan_filter_set,
2244 .rx_queue_setup = bond_ethdev_rx_queue_setup,
2245 .tx_queue_setup = bond_ethdev_tx_queue_setup,
2246 .rx_queue_release = bond_ethdev_rx_queue_release,
2247 .tx_queue_release = bond_ethdev_tx_queue_release,
2248 .link_update = bond_ethdev_link_update,
2249 .stats_get = bond_ethdev_stats_get,
2250 .stats_reset = bond_ethdev_stats_reset,
2251 .promiscuous_enable = bond_ethdev_promiscuous_enable,
2252 .promiscuous_disable = bond_ethdev_promiscuous_disable,
2253 .reta_update = bond_ethdev_rss_reta_update,
2254 .reta_query = bond_ethdev_rss_reta_query,
2255 .rss_hash_update = bond_ethdev_rss_hash_update,
2256 .rss_hash_conf_get = bond_ethdev_rss_hash_conf_get
2260 bond_alloc(struct rte_vdev_device *dev, uint8_t mode)
2262 const char *name = rte_vdev_device_name(dev);
2263 uint8_t socket_id = dev->device.numa_node;
2264 struct bond_dev_private *internals = NULL;
2265 struct rte_eth_dev *eth_dev = NULL;
2266 uint32_t vlan_filter_bmp_size;
2268 /* now do all data allocation - for eth_dev structure, dummy pci driver
2269 * and internal (private) data
2272 if (socket_id >= number_of_sockets()) {
2274 "Invalid socket id specified to create bonded device on.");
2278 /* reserve an ethdev entry */
2279 eth_dev = rte_eth_vdev_allocate(dev, sizeof(*internals));
2280 if (eth_dev == NULL) {
2281 RTE_BOND_LOG(ERR, "Unable to allocate rte_eth_dev");
2285 internals = eth_dev->data->dev_private;
2286 eth_dev->data->nb_rx_queues = (uint16_t)1;
2287 eth_dev->data->nb_tx_queues = (uint16_t)1;
2289 eth_dev->data->mac_addrs = rte_zmalloc_socket(name, ETHER_ADDR_LEN, 0,
2291 if (eth_dev->data->mac_addrs == NULL) {
2292 RTE_BOND_LOG(ERR, "Unable to malloc mac_addrs");
2296 eth_dev->dev_ops = &default_dev_ops;
2297 eth_dev->data->dev_flags = RTE_ETH_DEV_INTR_LSC |
2298 RTE_ETH_DEV_DETACHABLE;
2300 rte_spinlock_init(&internals->lock);
2302 internals->port_id = eth_dev->data->port_id;
2303 internals->mode = BONDING_MODE_INVALID;
2304 internals->current_primary_port = RTE_MAX_ETHPORTS + 1;
2305 internals->balance_xmit_policy = BALANCE_XMIT_POLICY_LAYER2;
2306 internals->xmit_hash = xmit_l2_hash;
2307 internals->user_defined_mac = 0;
2308 internals->link_props_set = 0;
2310 internals->link_status_polling_enabled = 0;
2312 internals->link_status_polling_interval_ms =
2313 DEFAULT_POLLING_INTERVAL_10_MS;
2314 internals->link_down_delay_ms = 0;
2315 internals->link_up_delay_ms = 0;
2317 internals->slave_count = 0;
2318 internals->active_slave_count = 0;
2319 internals->rx_offload_capa = 0;
2320 internals->tx_offload_capa = 0;
2321 internals->candidate_max_rx_pktlen = 0;
2322 internals->max_rx_pktlen = 0;
2324 /* Initially allow to choose any offload type */
2325 internals->flow_type_rss_offloads = ETH_RSS_PROTO_MASK;
2327 memset(internals->active_slaves, 0, sizeof(internals->active_slaves));
2328 memset(internals->slaves, 0, sizeof(internals->slaves));
2330 /* Set mode 4 default configuration */
2331 bond_mode_8023ad_setup(eth_dev, NULL);
2332 if (bond_ethdev_mode_set(eth_dev, mode)) {
2333 RTE_BOND_LOG(ERR, "Failed to set bonded device %d mode too %d",
2334 eth_dev->data->port_id, mode);
2338 vlan_filter_bmp_size =
2339 rte_bitmap_get_memory_footprint(ETHER_MAX_VLAN_ID + 1);
2340 internals->vlan_filter_bmpmem = rte_malloc(name, vlan_filter_bmp_size,
2341 RTE_CACHE_LINE_SIZE);
2342 if (internals->vlan_filter_bmpmem == NULL) {
2344 "Failed to allocate vlan bitmap for bonded device %u\n",
2345 eth_dev->data->port_id);
2349 internals->vlan_filter_bmp = rte_bitmap_init(ETHER_MAX_VLAN_ID + 1,
2350 internals->vlan_filter_bmpmem, vlan_filter_bmp_size);
2351 if (internals->vlan_filter_bmp == NULL) {
2353 "Failed to init vlan bitmap for bonded device %u\n",
2354 eth_dev->data->port_id);
2355 rte_free(internals->vlan_filter_bmpmem);
2359 return eth_dev->data->port_id;
2362 rte_free(internals);
2363 if (eth_dev != NULL) {
2364 rte_free(eth_dev->data->mac_addrs);
2365 rte_eth_dev_release_port(eth_dev);
2371 bond_probe(struct rte_vdev_device *dev)
2374 struct bond_dev_private *internals;
2375 struct rte_kvargs *kvlist;
2376 uint8_t bonding_mode, socket_id;
2377 int arg_count, port_id;
2382 name = rte_vdev_device_name(dev);
2383 RTE_LOG(INFO, EAL, "Initializing pmd_bond for %s\n", name);
2385 kvlist = rte_kvargs_parse(rte_vdev_device_args(dev),
2386 pmd_bond_init_valid_arguments);
2390 /* Parse link bonding mode */
2391 if (rte_kvargs_count(kvlist, PMD_BOND_MODE_KVARG) == 1) {
2392 if (rte_kvargs_process(kvlist, PMD_BOND_MODE_KVARG,
2393 &bond_ethdev_parse_slave_mode_kvarg,
2394 &bonding_mode) != 0) {
2395 RTE_LOG(ERR, EAL, "Invalid mode for bonded device %s\n",
2400 RTE_LOG(ERR, EAL, "Mode must be specified only once for bonded "
2401 "device %s\n", name);
2405 /* Parse socket id to create bonding device on */
2406 arg_count = rte_kvargs_count(kvlist, PMD_BOND_SOCKET_ID_KVARG);
2407 if (arg_count == 1) {
2408 if (rte_kvargs_process(kvlist, PMD_BOND_SOCKET_ID_KVARG,
2409 &bond_ethdev_parse_socket_id_kvarg, &socket_id)
2411 RTE_LOG(ERR, EAL, "Invalid socket Id specified for "
2412 "bonded device %s\n", name);
2415 } else if (arg_count > 1) {
2416 RTE_LOG(ERR, EAL, "Socket Id can be specified only once for "
2417 "bonded device %s\n", name);
2420 socket_id = rte_socket_id();
2423 dev->device.numa_node = socket_id;
2425 /* Create link bonding eth device */
2426 port_id = bond_alloc(dev, bonding_mode);
2428 RTE_LOG(ERR, EAL, "Failed to create socket %s in mode %u on "
2429 "socket %u.\n", name, bonding_mode, socket_id);
2432 internals = rte_eth_devices[port_id].data->dev_private;
2433 internals->kvlist = kvlist;
2435 RTE_LOG(INFO, EAL, "Create bonded device %s on port %d in mode %u on "
2436 "socket %u.\n", name, port_id, bonding_mode, socket_id);
2440 rte_kvargs_free(kvlist);
2446 bond_remove(struct rte_vdev_device *dev)
2448 struct rte_eth_dev *eth_dev;
2449 struct bond_dev_private *internals;
2455 name = rte_vdev_device_name(dev);
2456 RTE_LOG(INFO, EAL, "Uninitializing pmd_bond for %s\n", name);
2458 /* now free all data allocation - for eth_dev structure,
2459 * dummy pci driver and internal (private) data
2462 /* find an ethdev entry */
2463 eth_dev = rte_eth_dev_allocated(name);
2464 if (eth_dev == NULL)
2467 RTE_ASSERT(eth_dev->device == &dev->device);
2469 internals = eth_dev->data->dev_private;
2470 if (internals->slave_count != 0)
2473 if (eth_dev->data->dev_started == 1) {
2474 bond_ethdev_stop(eth_dev);
2475 bond_ethdev_close(eth_dev);
2478 eth_dev->dev_ops = NULL;
2479 eth_dev->rx_pkt_burst = NULL;
2480 eth_dev->tx_pkt_burst = NULL;
2482 internals = eth_dev->data->dev_private;
2483 rte_bitmap_free(internals->vlan_filter_bmp);
2484 rte_free(internals->vlan_filter_bmpmem);
2485 rte_free(eth_dev->data->dev_private);
2486 rte_free(eth_dev->data->mac_addrs);
2488 rte_eth_dev_release_port(eth_dev);
2493 /* this part will resolve the slave portids after all the other pdev and vdev
2494 * have been allocated */
2496 bond_ethdev_configure(struct rte_eth_dev *dev)
2498 char *name = dev->data->name;
2499 struct bond_dev_private *internals = dev->data->dev_private;
2500 struct rte_kvargs *kvlist = internals->kvlist;
2502 uint8_t port_id = dev - rte_eth_devices;
2504 static const uint8_t default_rss_key[40] = {
2505 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2, 0x41, 0x67, 0x25, 0x3D,
2506 0x43, 0xA3, 0x8F, 0xB0, 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
2507 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C, 0x6A, 0x42, 0xB7, 0x3B,
2508 0xBE, 0xAC, 0x01, 0xFA
2513 /* If RSS is enabled, fill table and key with default values */
2514 if (dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS) {
2515 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key = internals->rss_key;
2516 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len = 0;
2517 memcpy(internals->rss_key, default_rss_key, 40);
2519 for (i = 0; i < RTE_DIM(internals->reta_conf); i++) {
2520 internals->reta_conf[i].mask = ~0LL;
2521 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
2522 internals->reta_conf[i].reta[j] = j % dev->data->nb_rx_queues;
2526 /* set the max_rx_pktlen */
2527 internals->max_rx_pktlen = internals->candidate_max_rx_pktlen;
2530 * if no kvlist, it means that this bonded device has been created
2531 * through the bonding api.
2536 /* Parse MAC address for bonded device */
2537 arg_count = rte_kvargs_count(kvlist, PMD_BOND_MAC_ADDR_KVARG);
2538 if (arg_count == 1) {
2539 struct ether_addr bond_mac;
2541 if (rte_kvargs_process(kvlist, PMD_BOND_MAC_ADDR_KVARG,
2542 &bond_ethdev_parse_bond_mac_addr_kvarg, &bond_mac) < 0) {
2543 RTE_LOG(INFO, EAL, "Invalid mac address for bonded device %s\n",
2548 /* Set MAC address */
2549 if (rte_eth_bond_mac_address_set(port_id, &bond_mac) != 0) {
2551 "Failed to set mac address on bonded device %s\n",
2555 } else if (arg_count > 1) {
2557 "MAC address can be specified only once for bonded device %s\n",
2562 /* Parse/set balance mode transmit policy */
2563 arg_count = rte_kvargs_count(kvlist, PMD_BOND_XMIT_POLICY_KVARG);
2564 if (arg_count == 1) {
2565 uint8_t xmit_policy;
2567 if (rte_kvargs_process(kvlist, PMD_BOND_XMIT_POLICY_KVARG,
2568 &bond_ethdev_parse_balance_xmit_policy_kvarg, &xmit_policy) !=
2571 "Invalid xmit policy specified for bonded device %s\n",
2576 /* Set balance mode transmit policy*/
2577 if (rte_eth_bond_xmit_policy_set(port_id, xmit_policy) != 0) {
2579 "Failed to set balance xmit policy on bonded device %s\n",
2583 } else if (arg_count > 1) {
2585 "Transmit policy can be specified only once for bonded device"
2590 /* Parse/add slave ports to bonded device */
2591 if (rte_kvargs_count(kvlist, PMD_BOND_SLAVE_PORT_KVARG) > 0) {
2592 struct bond_ethdev_slave_ports slave_ports;
2595 memset(&slave_ports, 0, sizeof(slave_ports));
2597 if (rte_kvargs_process(kvlist, PMD_BOND_SLAVE_PORT_KVARG,
2598 &bond_ethdev_parse_slave_port_kvarg, &slave_ports) != 0) {
2600 "Failed to parse slave ports for bonded device %s\n",
2605 for (i = 0; i < slave_ports.slave_count; i++) {
2606 if (rte_eth_bond_slave_add(port_id, slave_ports.slaves[i]) != 0) {
2608 "Failed to add port %d as slave to bonded device %s\n",
2609 slave_ports.slaves[i], name);
2614 RTE_LOG(INFO, EAL, "No slaves specified for bonded device %s\n", name);
2618 /* Parse/set primary slave port id*/
2619 arg_count = rte_kvargs_count(kvlist, PMD_BOND_PRIMARY_SLAVE_KVARG);
2620 if (arg_count == 1) {
2621 uint8_t primary_slave_port_id;
2623 if (rte_kvargs_process(kvlist,
2624 PMD_BOND_PRIMARY_SLAVE_KVARG,
2625 &bond_ethdev_parse_primary_slave_port_id_kvarg,
2626 &primary_slave_port_id) < 0) {
2628 "Invalid primary slave port id specified for bonded device"
2633 /* Set balance mode transmit policy*/
2634 if (rte_eth_bond_primary_set(port_id, (uint8_t)primary_slave_port_id)
2637 "Failed to set primary slave port %d on bonded device %s\n",
2638 primary_slave_port_id, name);
2641 } else if (arg_count > 1) {
2643 "Primary slave can be specified only once for bonded device"
2648 /* Parse link status monitor polling interval */
2649 arg_count = rte_kvargs_count(kvlist, PMD_BOND_LSC_POLL_PERIOD_KVARG);
2650 if (arg_count == 1) {
2651 uint32_t lsc_poll_interval_ms;
2653 if (rte_kvargs_process(kvlist,
2654 PMD_BOND_LSC_POLL_PERIOD_KVARG,
2655 &bond_ethdev_parse_time_ms_kvarg,
2656 &lsc_poll_interval_ms) < 0) {
2658 "Invalid lsc polling interval value specified for bonded"
2659 " device %s\n", name);
2663 if (rte_eth_bond_link_monitoring_set(port_id, lsc_poll_interval_ms)
2666 "Failed to set lsc monitor polling interval (%u ms) on"
2667 " bonded device %s\n", lsc_poll_interval_ms, name);
2670 } else if (arg_count > 1) {
2672 "LSC polling interval can be specified only once for bonded"
2673 " device %s\n", name);
2677 /* Parse link up interrupt propagation delay */
2678 arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_UP_PROP_DELAY_KVARG);
2679 if (arg_count == 1) {
2680 uint32_t link_up_delay_ms;
2682 if (rte_kvargs_process(kvlist,
2683 PMD_BOND_LINK_UP_PROP_DELAY_KVARG,
2684 &bond_ethdev_parse_time_ms_kvarg,
2685 &link_up_delay_ms) < 0) {
2687 "Invalid link up propagation delay value specified for"
2688 " bonded device %s\n", name);
2692 /* Set balance mode transmit policy*/
2693 if (rte_eth_bond_link_up_prop_delay_set(port_id, link_up_delay_ms)
2696 "Failed to set link up propagation delay (%u ms) on bonded"
2697 " device %s\n", link_up_delay_ms, name);
2700 } else if (arg_count > 1) {
2702 "Link up propagation delay can be specified only once for"
2703 " bonded device %s\n", name);
2707 /* Parse link down interrupt propagation delay */
2708 arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG);
2709 if (arg_count == 1) {
2710 uint32_t link_down_delay_ms;
2712 if (rte_kvargs_process(kvlist,
2713 PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG,
2714 &bond_ethdev_parse_time_ms_kvarg,
2715 &link_down_delay_ms) < 0) {
2717 "Invalid link down propagation delay value specified for"
2718 " bonded device %s\n", name);
2722 /* Set balance mode transmit policy*/
2723 if (rte_eth_bond_link_down_prop_delay_set(port_id, link_down_delay_ms)
2726 "Failed to set link down propagation delay (%u ms) on"
2727 " bonded device %s\n", link_down_delay_ms, name);
2730 } else if (arg_count > 1) {
2732 "Link down propagation delay can be specified only once for"
2733 " bonded device %s\n", name);
2740 struct rte_vdev_driver pmd_bond_drv = {
2741 .probe = bond_probe,
2742 .remove = bond_remove,
2745 RTE_PMD_REGISTER_VDEV(net_bonding, pmd_bond_drv);
2746 RTE_PMD_REGISTER_ALIAS(net_bonding, eth_bond);
2748 RTE_PMD_REGISTER_PARAM_STRING(net_bonding,
2752 "xmit_policy=[l2 | l23 | l34] "
2755 "lsc_poll_period_ms=<int> "
2757 "down_delay=<int>");