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, struct rte_mbuf *mbuf)
130 const uint16_t ether_type_slow_be = rte_be_to_cpu_16(ETHER_TYPE_SLOW);
132 return !((mbuf->ol_flags & PKT_RX_VLAN) ? mbuf->vlan_tci : 0) &&
133 (ethertype == ether_type_slow_be &&
134 (subtype == SLOW_SUBTYPE_MARKER || subtype == SLOW_SUBTYPE_LACP));
137 /*****************************************************************************
138 * Flow director's setup for mode 4 optimization
141 static struct rte_flow_item_eth flow_item_eth_type_8023ad = {
142 .dst.addr_bytes = { 0 },
143 .src.addr_bytes = { 0 },
144 .type = RTE_BE16(ETHER_TYPE_SLOW),
147 static struct rte_flow_item_eth flow_item_eth_mask_type_8023ad = {
148 .dst.addr_bytes = { 0 },
149 .src.addr_bytes = { 0 },
153 static struct rte_flow_item flow_item_8023ad[] = {
155 .type = RTE_FLOW_ITEM_TYPE_ETH,
156 .spec = &flow_item_eth_type_8023ad,
158 .mask = &flow_item_eth_mask_type_8023ad,
161 .type = RTE_FLOW_ITEM_TYPE_END,
168 const struct rte_flow_attr flow_attr_8023ad = {
177 bond_ethdev_8023ad_flow_verify(struct rte_eth_dev *bond_dev,
178 uint16_t slave_port) {
179 struct rte_eth_dev_info slave_info;
180 struct rte_flow_error error;
181 struct bond_dev_private *internals = (struct bond_dev_private *)
182 (bond_dev->data->dev_private);
184 const struct rte_flow_action_queue lacp_queue_conf = {
188 const struct rte_flow_action actions[] = {
190 .type = RTE_FLOW_ACTION_TYPE_QUEUE,
191 .conf = &lacp_queue_conf
194 .type = RTE_FLOW_ACTION_TYPE_END,
198 int ret = rte_flow_validate(slave_port, &flow_attr_8023ad,
199 flow_item_8023ad, actions, &error);
201 RTE_BOND_LOG(ERR, "%s: %s (slave_port=%d queue_id=%d)",
202 __func__, error.message, slave_port,
203 internals->mode4.dedicated_queues.rx_qid);
207 rte_eth_dev_info_get(slave_port, &slave_info);
208 if (slave_info.max_rx_queues < bond_dev->data->nb_rx_queues ||
209 slave_info.max_tx_queues < bond_dev->data->nb_tx_queues) {
211 "%s: Slave %d capabilities doesn't allow to allocate additional queues",
212 __func__, slave_port);
220 bond_8023ad_slow_pkt_hw_filter_supported(uint16_t port_id) {
221 struct rte_eth_dev *bond_dev = &rte_eth_devices[port_id];
222 struct bond_dev_private *internals = (struct bond_dev_private *)
223 (bond_dev->data->dev_private);
224 struct rte_eth_dev_info bond_info;
227 /* Verify if all slaves in bonding supports flow director and */
228 if (internals->slave_count > 0) {
229 rte_eth_dev_info_get(bond_dev->data->port_id, &bond_info);
231 internals->mode4.dedicated_queues.rx_qid = bond_info.nb_rx_queues;
232 internals->mode4.dedicated_queues.tx_qid = bond_info.nb_tx_queues;
234 for (idx = 0; idx < internals->slave_count; idx++) {
235 if (bond_ethdev_8023ad_flow_verify(bond_dev,
236 internals->slaves[idx].port_id) != 0)
245 bond_ethdev_8023ad_flow_set(struct rte_eth_dev *bond_dev, uint16_t slave_port) {
247 struct rte_flow_error error;
248 struct bond_dev_private *internals = (struct bond_dev_private *)
249 (bond_dev->data->dev_private);
251 struct rte_flow_action_queue lacp_queue_conf = {
252 .index = internals->mode4.dedicated_queues.rx_qid,
255 const struct rte_flow_action actions[] = {
257 .type = RTE_FLOW_ACTION_TYPE_QUEUE,
258 .conf = &lacp_queue_conf
261 .type = RTE_FLOW_ACTION_TYPE_END,
265 internals->mode4.dedicated_queues.flow[slave_port] = rte_flow_create(slave_port,
266 &flow_attr_8023ad, flow_item_8023ad, actions, &error);
267 if (internals->mode4.dedicated_queues.flow[slave_port] == NULL) {
268 RTE_BOND_LOG(ERR, "bond_ethdev_8023ad_flow_set: %s "
269 "(slave_port=%d queue_id=%d)",
270 error.message, slave_port,
271 internals->mode4.dedicated_queues.rx_qid);
279 bond_ethdev_rx_burst_8023ad_fast_queue(void *queue, struct rte_mbuf **bufs,
282 struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
283 struct bond_dev_private *internals = bd_rx_q->dev_private;
284 uint16_t num_rx_total = 0; /* Total number of received packets */
285 uint16_t slaves[RTE_MAX_ETHPORTS];
286 uint16_t slave_count;
290 /* Copy slave list to protect against slave up/down changes during tx
292 slave_count = internals->active_slave_count;
293 memcpy(slaves, internals->active_slaves,
294 sizeof(internals->active_slaves[0]) * slave_count);
296 for (i = 0, idx = internals->active_slave;
297 i < slave_count && num_rx_total < nb_pkts; i++, idx++) {
298 idx = idx % slave_count;
300 /* Read packets from this slave */
301 num_rx_total += rte_eth_rx_burst(slaves[idx], bd_rx_q->queue_id,
302 &bufs[num_rx_total], nb_pkts - num_rx_total);
305 internals->active_slave = idx;
311 bond_ethdev_tx_burst_8023ad_fast_queue(void *queue, struct rte_mbuf **bufs,
314 struct bond_dev_private *internals;
315 struct bond_tx_queue *bd_tx_q;
317 uint16_t num_of_slaves;
318 uint16_t slaves[RTE_MAX_ETHPORTS];
319 /* positions in slaves, not ID */
320 uint8_t distributing_offsets[RTE_MAX_ETHPORTS];
321 uint8_t distributing_count;
323 uint16_t num_tx_slave, num_tx_total = 0, num_tx_fail_total = 0;
324 uint16_t i, op_slave_idx;
326 struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_pkts];
328 /* Total amount of packets in slave_bufs */
329 uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
330 /* Slow packets placed in each slave */
332 if (unlikely(nb_pkts == 0))
335 bd_tx_q = (struct bond_tx_queue *)queue;
336 internals = bd_tx_q->dev_private;
338 /* Copy slave list to protect against slave up/down changes during tx
340 num_of_slaves = internals->active_slave_count;
341 if (num_of_slaves < 1)
344 memcpy(slaves, internals->active_slaves, sizeof(slaves[0]) *
347 distributing_count = 0;
348 for (i = 0; i < num_of_slaves; i++) {
349 struct port *port = &mode_8023ad_ports[slaves[i]];
350 if (ACTOR_STATE(port, DISTRIBUTING))
351 distributing_offsets[distributing_count++] = i;
354 if (likely(distributing_count > 0)) {
355 /* Populate slaves mbuf with the packets which are to be sent */
356 for (i = 0; i < nb_pkts; i++) {
357 /* Select output slave using hash based on xmit policy */
358 op_slave_idx = internals->xmit_hash(bufs[i],
361 /* Populate slave mbuf arrays with mbufs for that slave.
362 * Use only slaves that are currently distributing.
364 uint8_t slave_offset =
365 distributing_offsets[op_slave_idx];
366 slave_bufs[slave_offset][slave_nb_pkts[slave_offset]] =
368 slave_nb_pkts[slave_offset]++;
372 /* Send packet burst on each slave device */
373 for (i = 0; i < num_of_slaves; i++) {
374 if (slave_nb_pkts[i] == 0)
377 num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
378 slave_bufs[i], slave_nb_pkts[i]);
380 num_tx_total += num_tx_slave;
381 num_tx_fail_total += slave_nb_pkts[i] - num_tx_slave;
383 /* If tx burst fails move packets to end of bufs */
384 if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
385 uint16_t j = nb_pkts - num_tx_fail_total;
386 for ( ; num_tx_slave < slave_nb_pkts[i]; j++,
388 bufs[j] = slave_bufs[i][num_tx_slave];
397 bond_ethdev_rx_burst_8023ad(void *queue, struct rte_mbuf **bufs,
400 /* Cast to structure, containing bonded device's port id and queue id */
401 struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
402 struct bond_dev_private *internals = bd_rx_q->dev_private;
403 struct ether_addr bond_mac;
405 struct ether_hdr *hdr;
407 const uint16_t ether_type_slow_be = rte_be_to_cpu_16(ETHER_TYPE_SLOW);
408 uint16_t num_rx_total = 0; /* Total number of received packets */
409 uint16_t slaves[RTE_MAX_ETHPORTS];
410 uint16_t slave_count, idx;
412 uint8_t collecting; /* current slave collecting status */
413 const uint8_t promisc = internals->promiscuous_en;
417 rte_eth_macaddr_get(internals->port_id, &bond_mac);
418 /* Copy slave list to protect against slave up/down changes during tx
420 slave_count = internals->active_slave_count;
421 memcpy(slaves, internals->active_slaves,
422 sizeof(internals->active_slaves[0]) * slave_count);
424 idx = internals->active_slave;
425 if (idx >= slave_count) {
426 internals->active_slave = 0;
429 for (i = 0; i < slave_count && num_rx_total < nb_pkts; i++) {
431 collecting = ACTOR_STATE(&mode_8023ad_ports[slaves[idx]],
434 /* Read packets from this slave */
435 num_rx_total += rte_eth_rx_burst(slaves[idx], bd_rx_q->queue_id,
436 &bufs[num_rx_total], nb_pkts - num_rx_total);
438 for (k = j; k < 2 && k < num_rx_total; k++)
439 rte_prefetch0(rte_pktmbuf_mtod(bufs[k], void *));
441 /* Handle slow protocol packets. */
442 while (j < num_rx_total) {
444 /* If packet is not pure L2 and is known, skip it */
445 if ((bufs[j]->packet_type & ~RTE_PTYPE_L2_ETHER) != 0) {
450 if (j + 3 < num_rx_total)
451 rte_prefetch0(rte_pktmbuf_mtod(bufs[j + 3], void *));
453 hdr = rte_pktmbuf_mtod(bufs[j], struct ether_hdr *);
454 subtype = ((struct slow_protocol_frame *)hdr)->slow_protocol.subtype;
456 /* Remove packet from array if it is slow packet or slave is not
457 * in collecting state or bonding interface is not in promiscuous
458 * mode and packet address does not match. */
459 if (unlikely(is_lacp_packets(hdr->ether_type, subtype, bufs[j]) ||
460 !collecting || (!promisc &&
461 !is_multicast_ether_addr(&hdr->d_addr) &&
462 !is_same_ether_addr(&bond_mac, &hdr->d_addr)))) {
464 if (hdr->ether_type == ether_type_slow_be) {
465 bond_mode_8023ad_handle_slow_pkt(
466 internals, slaves[idx], bufs[j]);
468 rte_pktmbuf_free(bufs[j]);
470 /* Packet is managed by mode 4 or dropped, shift the array */
472 if (j < num_rx_total) {
473 memmove(&bufs[j], &bufs[j + 1], sizeof(bufs[0]) *
479 if (unlikely(++idx == slave_count))
483 internals->active_slave = idx;
487 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
488 uint32_t burstnumberRX;
489 uint32_t burstnumberTX;
491 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
494 arp_op_name(uint16_t arp_op, char *buf)
498 snprintf(buf, sizeof("ARP Request"), "%s", "ARP Request");
501 snprintf(buf, sizeof("ARP Reply"), "%s", "ARP Reply");
503 case ARP_OP_REVREQUEST:
504 snprintf(buf, sizeof("Reverse ARP Request"), "%s",
505 "Reverse ARP Request");
507 case ARP_OP_REVREPLY:
508 snprintf(buf, sizeof("Reverse ARP Reply"), "%s",
509 "Reverse ARP Reply");
511 case ARP_OP_INVREQUEST:
512 snprintf(buf, sizeof("Peer Identify Request"), "%s",
513 "Peer Identify Request");
515 case ARP_OP_INVREPLY:
516 snprintf(buf, sizeof("Peer Identify Reply"), "%s",
517 "Peer Identify Reply");
522 snprintf(buf, sizeof("Unknown"), "%s", "Unknown");
526 #define MaxIPv4String 16
528 ipv4_addr_to_dot(uint32_t be_ipv4_addr, char *buf, uint8_t buf_size)
532 ipv4_addr = rte_be_to_cpu_32(be_ipv4_addr);
533 snprintf(buf, buf_size, "%d.%d.%d.%d", (ipv4_addr >> 24) & 0xFF,
534 (ipv4_addr >> 16) & 0xFF, (ipv4_addr >> 8) & 0xFF,
538 #define MAX_CLIENTS_NUMBER 128
539 uint8_t active_clients;
540 struct client_stats_t {
543 uint32_t ipv4_rx_packets;
544 uint32_t ipv4_tx_packets;
546 struct client_stats_t client_stats[MAX_CLIENTS_NUMBER];
549 update_client_stats(uint32_t addr, uint16_t port, uint32_t *TXorRXindicator)
553 for (; i < MAX_CLIENTS_NUMBER; i++) {
554 if ((client_stats[i].ipv4_addr == addr) && (client_stats[i].port == port)) {
555 /* Just update RX packets number for this client */
556 if (TXorRXindicator == &burstnumberRX)
557 client_stats[i].ipv4_rx_packets++;
559 client_stats[i].ipv4_tx_packets++;
563 /* We have a new client. Insert him to the table, and increment stats */
564 if (TXorRXindicator == &burstnumberRX)
565 client_stats[active_clients].ipv4_rx_packets++;
567 client_stats[active_clients].ipv4_tx_packets++;
568 client_stats[active_clients].ipv4_addr = addr;
569 client_stats[active_clients].port = port;
574 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
575 #define MODE6_DEBUG(info, src_ip, dst_ip, eth_h, arp_op, port, burstnumber) \
576 RTE_LOG(DEBUG, PMD, \
579 "SrcMAC:%02X:%02X:%02X:%02X:%02X:%02X " \
581 "DstMAC:%02X:%02X:%02X:%02X:%02X:%02X " \
587 eth_h->s_addr.addr_bytes[0], \
588 eth_h->s_addr.addr_bytes[1], \
589 eth_h->s_addr.addr_bytes[2], \
590 eth_h->s_addr.addr_bytes[3], \
591 eth_h->s_addr.addr_bytes[4], \
592 eth_h->s_addr.addr_bytes[5], \
594 eth_h->d_addr.addr_bytes[0], \
595 eth_h->d_addr.addr_bytes[1], \
596 eth_h->d_addr.addr_bytes[2], \
597 eth_h->d_addr.addr_bytes[3], \
598 eth_h->d_addr.addr_bytes[4], \
599 eth_h->d_addr.addr_bytes[5], \
606 mode6_debug(const char __attribute__((unused)) *info, struct ether_hdr *eth_h,
607 uint16_t port, uint32_t __attribute__((unused)) *burstnumber)
609 struct ipv4_hdr *ipv4_h;
610 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
611 struct arp_hdr *arp_h;
618 uint16_t ether_type = eth_h->ether_type;
619 uint16_t offset = get_vlan_offset(eth_h, ðer_type);
621 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
622 snprintf(buf, 16, "%s", info);
625 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4)) {
626 ipv4_h = (struct ipv4_hdr *)((char *)(eth_h + 1) + offset);
627 ipv4_addr_to_dot(ipv4_h->src_addr, src_ip, MaxIPv4String);
628 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
629 ipv4_addr_to_dot(ipv4_h->dst_addr, dst_ip, MaxIPv4String);
630 MODE6_DEBUG(buf, src_ip, dst_ip, eth_h, "", port, *burstnumber);
632 update_client_stats(ipv4_h->src_addr, port, burstnumber);
634 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
635 else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_ARP)) {
636 arp_h = (struct arp_hdr *)((char *)(eth_h + 1) + offset);
637 ipv4_addr_to_dot(arp_h->arp_data.arp_sip, src_ip, MaxIPv4String);
638 ipv4_addr_to_dot(arp_h->arp_data.arp_tip, dst_ip, MaxIPv4String);
639 arp_op_name(rte_be_to_cpu_16(arp_h->arp_op), ArpOp);
640 MODE6_DEBUG(buf, src_ip, dst_ip, eth_h, ArpOp, port, *burstnumber);
647 bond_ethdev_rx_burst_alb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
649 struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
650 struct bond_dev_private *internals = bd_tx_q->dev_private;
651 struct ether_hdr *eth_h;
652 uint16_t ether_type, offset;
653 uint16_t nb_recv_pkts;
656 nb_recv_pkts = bond_ethdev_rx_burst(queue, bufs, nb_pkts);
658 for (i = 0; i < nb_recv_pkts; i++) {
659 eth_h = rte_pktmbuf_mtod(bufs[i], struct ether_hdr *);
660 ether_type = eth_h->ether_type;
661 offset = get_vlan_offset(eth_h, ðer_type);
663 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_ARP)) {
664 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
665 mode6_debug("RX ARP:", eth_h, bufs[i]->port, &burstnumberRX);
667 bond_mode_alb_arp_recv(eth_h, offset, internals);
669 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
670 else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
671 mode6_debug("RX IPv4:", eth_h, bufs[i]->port, &burstnumberRX);
679 bond_ethdev_tx_burst_round_robin(void *queue, struct rte_mbuf **bufs,
682 struct bond_dev_private *internals;
683 struct bond_tx_queue *bd_tx_q;
685 struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_pkts];
686 uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
688 uint16_t num_of_slaves;
689 uint16_t slaves[RTE_MAX_ETHPORTS];
691 uint16_t num_tx_total = 0, num_tx_slave;
693 static int slave_idx = 0;
694 int i, cslave_idx = 0, tx_fail_total = 0;
696 bd_tx_q = (struct bond_tx_queue *)queue;
697 internals = bd_tx_q->dev_private;
699 /* Copy slave list to protect against slave up/down changes during tx
701 num_of_slaves = internals->active_slave_count;
702 memcpy(slaves, internals->active_slaves,
703 sizeof(internals->active_slaves[0]) * num_of_slaves);
705 if (num_of_slaves < 1)
708 /* Populate slaves mbuf with which packets are to be sent on it */
709 for (i = 0; i < nb_pkts; i++) {
710 cslave_idx = (slave_idx + i) % num_of_slaves;
711 slave_bufs[cslave_idx][(slave_nb_pkts[cslave_idx])++] = bufs[i];
714 /* increment current slave index so the next call to tx burst starts on the
716 slave_idx = ++cslave_idx;
718 /* Send packet burst on each slave device */
719 for (i = 0; i < num_of_slaves; i++) {
720 if (slave_nb_pkts[i] > 0) {
721 num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
722 slave_bufs[i], slave_nb_pkts[i]);
724 /* if tx burst fails move packets to end of bufs */
725 if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
726 int tx_fail_slave = slave_nb_pkts[i] - num_tx_slave;
728 tx_fail_total += tx_fail_slave;
730 memcpy(&bufs[nb_pkts - tx_fail_total],
731 &slave_bufs[i][num_tx_slave],
732 tx_fail_slave * sizeof(bufs[0]));
734 num_tx_total += num_tx_slave;
742 bond_ethdev_tx_burst_active_backup(void *queue,
743 struct rte_mbuf **bufs, uint16_t nb_pkts)
745 struct bond_dev_private *internals;
746 struct bond_tx_queue *bd_tx_q;
748 bd_tx_q = (struct bond_tx_queue *)queue;
749 internals = bd_tx_q->dev_private;
751 if (internals->active_slave_count < 1)
754 return rte_eth_tx_burst(internals->current_primary_port, bd_tx_q->queue_id,
758 static inline uint16_t
759 ether_hash(struct ether_hdr *eth_hdr)
761 unaligned_uint16_t *word_src_addr =
762 (unaligned_uint16_t *)eth_hdr->s_addr.addr_bytes;
763 unaligned_uint16_t *word_dst_addr =
764 (unaligned_uint16_t *)eth_hdr->d_addr.addr_bytes;
766 return (word_src_addr[0] ^ word_dst_addr[0]) ^
767 (word_src_addr[1] ^ word_dst_addr[1]) ^
768 (word_src_addr[2] ^ word_dst_addr[2]);
771 static inline uint32_t
772 ipv4_hash(struct ipv4_hdr *ipv4_hdr)
774 return ipv4_hdr->src_addr ^ ipv4_hdr->dst_addr;
777 static inline uint32_t
778 ipv6_hash(struct ipv6_hdr *ipv6_hdr)
780 unaligned_uint32_t *word_src_addr =
781 (unaligned_uint32_t *)&(ipv6_hdr->src_addr[0]);
782 unaligned_uint32_t *word_dst_addr =
783 (unaligned_uint32_t *)&(ipv6_hdr->dst_addr[0]);
785 return (word_src_addr[0] ^ word_dst_addr[0]) ^
786 (word_src_addr[1] ^ word_dst_addr[1]) ^
787 (word_src_addr[2] ^ word_dst_addr[2]) ^
788 (word_src_addr[3] ^ word_dst_addr[3]);
792 xmit_l2_hash(const struct rte_mbuf *buf, uint8_t slave_count)
794 struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(buf, struct ether_hdr *);
796 uint32_t hash = ether_hash(eth_hdr);
798 return (hash ^= hash >> 8) % slave_count;
802 xmit_l23_hash(const struct rte_mbuf *buf, uint8_t slave_count)
804 struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(buf, struct ether_hdr *);
805 uint16_t proto = eth_hdr->ether_type;
806 size_t vlan_offset = get_vlan_offset(eth_hdr, &proto);
807 uint32_t hash, l3hash = 0;
809 hash = ether_hash(eth_hdr);
811 if (rte_cpu_to_be_16(ETHER_TYPE_IPv4) == proto) {
812 struct ipv4_hdr *ipv4_hdr = (struct ipv4_hdr *)
813 ((char *)(eth_hdr + 1) + vlan_offset);
814 l3hash = ipv4_hash(ipv4_hdr);
816 } else if (rte_cpu_to_be_16(ETHER_TYPE_IPv6) == proto) {
817 struct ipv6_hdr *ipv6_hdr = (struct ipv6_hdr *)
818 ((char *)(eth_hdr + 1) + vlan_offset);
819 l3hash = ipv6_hash(ipv6_hdr);
822 hash = hash ^ l3hash;
826 return hash % slave_count;
830 xmit_l34_hash(const struct rte_mbuf *buf, uint8_t slave_count)
832 struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(buf, struct ether_hdr *);
833 uint16_t proto = eth_hdr->ether_type;
834 size_t vlan_offset = get_vlan_offset(eth_hdr, &proto);
836 struct udp_hdr *udp_hdr = NULL;
837 struct tcp_hdr *tcp_hdr = NULL;
838 uint32_t hash, l3hash = 0, l4hash = 0;
840 if (rte_cpu_to_be_16(ETHER_TYPE_IPv4) == proto) {
841 struct ipv4_hdr *ipv4_hdr = (struct ipv4_hdr *)
842 ((char *)(eth_hdr + 1) + vlan_offset);
843 size_t ip_hdr_offset;
845 l3hash = ipv4_hash(ipv4_hdr);
847 /* there is no L4 header in fragmented packet */
848 if (likely(rte_ipv4_frag_pkt_is_fragmented(ipv4_hdr) == 0)) {
849 ip_hdr_offset = (ipv4_hdr->version_ihl & IPV4_HDR_IHL_MASK) *
852 if (ipv4_hdr->next_proto_id == IPPROTO_TCP) {
853 tcp_hdr = (struct tcp_hdr *)((char *)ipv4_hdr +
855 l4hash = HASH_L4_PORTS(tcp_hdr);
856 } else if (ipv4_hdr->next_proto_id == IPPROTO_UDP) {
857 udp_hdr = (struct udp_hdr *)((char *)ipv4_hdr +
859 l4hash = HASH_L4_PORTS(udp_hdr);
862 } else if (rte_cpu_to_be_16(ETHER_TYPE_IPv6) == proto) {
863 struct ipv6_hdr *ipv6_hdr = (struct ipv6_hdr *)
864 ((char *)(eth_hdr + 1) + vlan_offset);
865 l3hash = ipv6_hash(ipv6_hdr);
867 if (ipv6_hdr->proto == IPPROTO_TCP) {
868 tcp_hdr = (struct tcp_hdr *)(ipv6_hdr + 1);
869 l4hash = HASH_L4_PORTS(tcp_hdr);
870 } else if (ipv6_hdr->proto == IPPROTO_UDP) {
871 udp_hdr = (struct udp_hdr *)(ipv6_hdr + 1);
872 l4hash = HASH_L4_PORTS(udp_hdr);
876 hash = l3hash ^ l4hash;
880 return hash % slave_count;
884 uint64_t bwg_left_int;
885 uint64_t bwg_left_remainder;
890 bond_tlb_activate_slave(struct bond_dev_private *internals) {
893 for (i = 0; i < internals->active_slave_count; i++) {
894 tlb_last_obytets[internals->active_slaves[i]] = 0;
899 bandwidth_cmp(const void *a, const void *b)
901 const struct bwg_slave *bwg_a = a;
902 const struct bwg_slave *bwg_b = b;
903 int64_t diff = (int64_t)bwg_b->bwg_left_int - (int64_t)bwg_a->bwg_left_int;
904 int64_t diff2 = (int64_t)bwg_b->bwg_left_remainder -
905 (int64_t)bwg_a->bwg_left_remainder;
919 bandwidth_left(uint16_t port_id, uint64_t load, uint8_t update_idx,
920 struct bwg_slave *bwg_slave)
922 struct rte_eth_link link_status;
924 rte_eth_link_get_nowait(port_id, &link_status);
925 uint64_t link_bwg = link_status.link_speed * 1000000ULL / 8;
928 link_bwg = link_bwg * (update_idx+1) * REORDER_PERIOD_MS;
929 bwg_slave->bwg_left_int = (link_bwg - 1000*load) / link_bwg;
930 bwg_slave->bwg_left_remainder = (link_bwg - 1000*load) % link_bwg;
934 bond_ethdev_update_tlb_slave_cb(void *arg)
936 struct bond_dev_private *internals = arg;
937 struct rte_eth_stats slave_stats;
938 struct bwg_slave bwg_array[RTE_MAX_ETHPORTS];
942 uint8_t update_stats = 0;
945 internals->slave_update_idx++;
948 if (internals->slave_update_idx >= REORDER_PERIOD_MS)
951 for (i = 0; i < internals->active_slave_count; i++) {
952 slave_id = internals->active_slaves[i];
953 rte_eth_stats_get(slave_id, &slave_stats);
954 tx_bytes = slave_stats.obytes - tlb_last_obytets[slave_id];
955 bandwidth_left(slave_id, tx_bytes,
956 internals->slave_update_idx, &bwg_array[i]);
957 bwg_array[i].slave = slave_id;
960 tlb_last_obytets[slave_id] = slave_stats.obytes;
964 if (update_stats == 1)
965 internals->slave_update_idx = 0;
968 qsort(bwg_array, slave_count, sizeof(bwg_array[0]), bandwidth_cmp);
969 for (i = 0; i < slave_count; i++)
970 internals->tlb_slaves_order[i] = bwg_array[i].slave;
972 rte_eal_alarm_set(REORDER_PERIOD_MS * 1000, bond_ethdev_update_tlb_slave_cb,
973 (struct bond_dev_private *)internals);
977 bond_ethdev_tx_burst_tlb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
979 struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
980 struct bond_dev_private *internals = bd_tx_q->dev_private;
982 struct rte_eth_dev *primary_port =
983 &rte_eth_devices[internals->primary_port];
984 uint16_t num_tx_total = 0;
987 uint16_t num_of_slaves = internals->active_slave_count;
988 uint16_t slaves[RTE_MAX_ETHPORTS];
990 struct ether_hdr *ether_hdr;
991 struct ether_addr primary_slave_addr;
992 struct ether_addr active_slave_addr;
994 if (num_of_slaves < 1)
997 memcpy(slaves, internals->tlb_slaves_order,
998 sizeof(internals->tlb_slaves_order[0]) * num_of_slaves);
1001 ether_addr_copy(primary_port->data->mac_addrs, &primary_slave_addr);
1004 for (i = 0; i < 3; i++)
1005 rte_prefetch0(rte_pktmbuf_mtod(bufs[i], void*));
1008 for (i = 0; i < num_of_slaves; i++) {
1009 rte_eth_macaddr_get(slaves[i], &active_slave_addr);
1010 for (j = num_tx_total; j < nb_pkts; j++) {
1011 if (j + 3 < nb_pkts)
1012 rte_prefetch0(rte_pktmbuf_mtod(bufs[j+3], void*));
1014 ether_hdr = rte_pktmbuf_mtod(bufs[j], struct ether_hdr *);
1015 if (is_same_ether_addr(ðer_hdr->s_addr, &primary_slave_addr))
1016 ether_addr_copy(&active_slave_addr, ðer_hdr->s_addr);
1017 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
1018 mode6_debug("TX IPv4:", ether_hdr, slaves[i], &burstnumberTX);
1022 num_tx_total += rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
1023 bufs + num_tx_total, nb_pkts - num_tx_total);
1025 if (num_tx_total == nb_pkts)
1029 return num_tx_total;
1033 bond_tlb_disable(struct bond_dev_private *internals)
1035 rte_eal_alarm_cancel(bond_ethdev_update_tlb_slave_cb, internals);
1039 bond_tlb_enable(struct bond_dev_private *internals)
1041 bond_ethdev_update_tlb_slave_cb(internals);
1045 bond_ethdev_tx_burst_alb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
1047 struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
1048 struct bond_dev_private *internals = bd_tx_q->dev_private;
1050 struct ether_hdr *eth_h;
1051 uint16_t ether_type, offset;
1053 struct client_data *client_info;
1056 * We create transmit buffers for every slave and one additional to send
1057 * through tlb. In worst case every packet will be send on one port.
1059 struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS + 1][nb_pkts];
1060 uint16_t slave_bufs_pkts[RTE_MAX_ETHPORTS + 1] = { 0 };
1063 * We create separate transmit buffers for update packets as they won't
1064 * be counted in num_tx_total.
1066 struct rte_mbuf *update_bufs[RTE_MAX_ETHPORTS][ALB_HASH_TABLE_SIZE];
1067 uint16_t update_bufs_pkts[RTE_MAX_ETHPORTS] = { 0 };
1069 struct rte_mbuf *upd_pkt;
1072 uint16_t num_send, num_not_send = 0;
1073 uint16_t num_tx_total = 0;
1078 /* Search tx buffer for ARP packets and forward them to alb */
1079 for (i = 0; i < nb_pkts; i++) {
1080 eth_h = rte_pktmbuf_mtod(bufs[i], struct ether_hdr *);
1081 ether_type = eth_h->ether_type;
1082 offset = get_vlan_offset(eth_h, ðer_type);
1084 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_ARP)) {
1085 slave_idx = bond_mode_alb_arp_xmit(eth_h, offset, internals);
1087 /* Change src mac in eth header */
1088 rte_eth_macaddr_get(slave_idx, ð_h->s_addr);
1090 /* Add packet to slave tx buffer */
1091 slave_bufs[slave_idx][slave_bufs_pkts[slave_idx]] = bufs[i];
1092 slave_bufs_pkts[slave_idx]++;
1094 /* If packet is not ARP, send it with TLB policy */
1095 slave_bufs[RTE_MAX_ETHPORTS][slave_bufs_pkts[RTE_MAX_ETHPORTS]] =
1097 slave_bufs_pkts[RTE_MAX_ETHPORTS]++;
1101 /* Update connected client ARP tables */
1102 if (internals->mode6.ntt) {
1103 for (i = 0; i < ALB_HASH_TABLE_SIZE; i++) {
1104 client_info = &internals->mode6.client_table[i];
1106 if (client_info->in_use) {
1107 /* Allocate new packet to send ARP update on current slave */
1108 upd_pkt = rte_pktmbuf_alloc(internals->mode6.mempool);
1109 if (upd_pkt == NULL) {
1110 RTE_LOG(ERR, PMD, "Failed to allocate ARP packet from pool\n");
1113 pkt_size = sizeof(struct ether_hdr) + sizeof(struct arp_hdr)
1114 + client_info->vlan_count * sizeof(struct vlan_hdr);
1115 upd_pkt->data_len = pkt_size;
1116 upd_pkt->pkt_len = pkt_size;
1118 slave_idx = bond_mode_alb_arp_upd(client_info, upd_pkt,
1121 /* Add packet to update tx buffer */
1122 update_bufs[slave_idx][update_bufs_pkts[slave_idx]] = upd_pkt;
1123 update_bufs_pkts[slave_idx]++;
1126 internals->mode6.ntt = 0;
1129 /* Send ARP packets on proper slaves */
1130 for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
1131 if (slave_bufs_pkts[i] > 0) {
1132 num_send = rte_eth_tx_burst(i, bd_tx_q->queue_id,
1133 slave_bufs[i], slave_bufs_pkts[i]);
1134 for (j = 0; j < slave_bufs_pkts[i] - num_send; j++) {
1135 bufs[nb_pkts - 1 - num_not_send - j] =
1136 slave_bufs[i][nb_pkts - 1 - j];
1139 num_tx_total += num_send;
1140 num_not_send += slave_bufs_pkts[i] - num_send;
1142 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
1143 /* Print TX stats including update packets */
1144 for (j = 0; j < slave_bufs_pkts[i]; j++) {
1145 eth_h = rte_pktmbuf_mtod(slave_bufs[i][j], struct ether_hdr *);
1146 mode6_debug("TX ARP:", eth_h, i, &burstnumberTX);
1152 /* Send update packets on proper slaves */
1153 for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
1154 if (update_bufs_pkts[i] > 0) {
1155 num_send = rte_eth_tx_burst(i, bd_tx_q->queue_id, update_bufs[i],
1156 update_bufs_pkts[i]);
1157 for (j = num_send; j < update_bufs_pkts[i]; j++) {
1158 rte_pktmbuf_free(update_bufs[i][j]);
1160 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
1161 for (j = 0; j < update_bufs_pkts[i]; j++) {
1162 eth_h = rte_pktmbuf_mtod(update_bufs[i][j], struct ether_hdr *);
1163 mode6_debug("TX ARPupd:", eth_h, i, &burstnumberTX);
1169 /* Send non-ARP packets using tlb policy */
1170 if (slave_bufs_pkts[RTE_MAX_ETHPORTS] > 0) {
1171 num_send = bond_ethdev_tx_burst_tlb(queue,
1172 slave_bufs[RTE_MAX_ETHPORTS],
1173 slave_bufs_pkts[RTE_MAX_ETHPORTS]);
1175 for (j = 0; j < slave_bufs_pkts[RTE_MAX_ETHPORTS]; j++) {
1176 bufs[nb_pkts - 1 - num_not_send - j] =
1177 slave_bufs[RTE_MAX_ETHPORTS][nb_pkts - 1 - j];
1180 num_tx_total += num_send;
1183 return num_tx_total;
1187 bond_ethdev_tx_burst_balance(void *queue, struct rte_mbuf **bufs,
1190 struct bond_dev_private *internals;
1191 struct bond_tx_queue *bd_tx_q;
1193 uint16_t num_of_slaves;
1194 uint16_t slaves[RTE_MAX_ETHPORTS];
1196 uint16_t num_tx_total = 0, num_tx_slave = 0, tx_fail_total = 0;
1200 struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_pkts];
1201 uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
1203 bd_tx_q = (struct bond_tx_queue *)queue;
1204 internals = bd_tx_q->dev_private;
1206 /* Copy slave list to protect against slave up/down changes during tx
1208 num_of_slaves = internals->active_slave_count;
1209 memcpy(slaves, internals->active_slaves,
1210 sizeof(internals->active_slaves[0]) * num_of_slaves);
1212 if (num_of_slaves < 1)
1213 return num_tx_total;
1215 /* Populate slaves mbuf with the packets which are to be sent on it */
1216 for (i = 0; i < nb_pkts; i++) {
1217 /* Select output slave using hash based on xmit policy */
1218 op_slave_id = internals->xmit_hash(bufs[i], num_of_slaves);
1220 /* Populate slave mbuf arrays with mbufs for that slave */
1221 slave_bufs[op_slave_id][slave_nb_pkts[op_slave_id]++] = bufs[i];
1224 /* Send packet burst on each slave device */
1225 for (i = 0; i < num_of_slaves; i++) {
1226 if (slave_nb_pkts[i] > 0) {
1227 num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
1228 slave_bufs[i], slave_nb_pkts[i]);
1230 /* if tx burst fails move packets to end of bufs */
1231 if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
1232 int slave_tx_fail_count = slave_nb_pkts[i] - num_tx_slave;
1234 tx_fail_total += slave_tx_fail_count;
1235 memcpy(&bufs[nb_pkts - tx_fail_total],
1236 &slave_bufs[i][num_tx_slave],
1237 slave_tx_fail_count * sizeof(bufs[0]));
1240 num_tx_total += num_tx_slave;
1244 return num_tx_total;
1248 bond_ethdev_tx_burst_8023ad(void *queue, struct rte_mbuf **bufs,
1251 struct bond_dev_private *internals;
1252 struct bond_tx_queue *bd_tx_q;
1254 uint16_t num_of_slaves;
1255 uint16_t slaves[RTE_MAX_ETHPORTS];
1256 /* positions in slaves, not ID */
1257 uint8_t distributing_offsets[RTE_MAX_ETHPORTS];
1258 uint8_t distributing_count;
1260 uint16_t num_tx_slave, num_tx_total = 0, num_tx_fail_total = 0;
1261 uint16_t i, j, op_slave_idx;
1262 const uint16_t buffs_size = nb_pkts + BOND_MODE_8023AX_SLAVE_TX_PKTS + 1;
1264 /* Allocate additional packets in case 8023AD mode. */
1265 struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][buffs_size];
1266 void *slow_pkts[BOND_MODE_8023AX_SLAVE_TX_PKTS] = { NULL };
1268 /* Total amount of packets in slave_bufs */
1269 uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
1270 /* Slow packets placed in each slave */
1271 uint8_t slave_slow_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
1273 bd_tx_q = (struct bond_tx_queue *)queue;
1274 internals = bd_tx_q->dev_private;
1276 /* Copy slave list to protect against slave up/down changes during tx
1278 num_of_slaves = internals->active_slave_count;
1279 if (num_of_slaves < 1)
1280 return num_tx_total;
1282 memcpy(slaves, internals->active_slaves, sizeof(slaves[0]) * num_of_slaves);
1284 distributing_count = 0;
1285 for (i = 0; i < num_of_slaves; i++) {
1286 struct port *port = &mode_8023ad_ports[slaves[i]];
1288 slave_slow_nb_pkts[i] = rte_ring_dequeue_burst(port->tx_ring,
1289 slow_pkts, BOND_MODE_8023AX_SLAVE_TX_PKTS,
1291 slave_nb_pkts[i] = slave_slow_nb_pkts[i];
1293 for (j = 0; j < slave_slow_nb_pkts[i]; j++)
1294 slave_bufs[i][j] = slow_pkts[j];
1296 if (ACTOR_STATE(port, DISTRIBUTING))
1297 distributing_offsets[distributing_count++] = i;
1300 if (likely(distributing_count > 0)) {
1301 /* Populate slaves mbuf with the packets which are to be sent on it */
1302 for (i = 0; i < nb_pkts; i++) {
1303 /* Select output slave using hash based on xmit policy */
1304 op_slave_idx = internals->xmit_hash(bufs[i], distributing_count);
1306 /* Populate slave mbuf arrays with mbufs for that slave. Use only
1307 * slaves that are currently distributing. */
1308 uint8_t slave_offset = distributing_offsets[op_slave_idx];
1309 slave_bufs[slave_offset][slave_nb_pkts[slave_offset]] = bufs[i];
1310 slave_nb_pkts[slave_offset]++;
1314 /* Send packet burst on each slave device */
1315 for (i = 0; i < num_of_slaves; i++) {
1316 if (slave_nb_pkts[i] == 0)
1319 num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
1320 slave_bufs[i], slave_nb_pkts[i]);
1322 /* If tx burst fails drop slow packets */
1323 for ( ; num_tx_slave < slave_slow_nb_pkts[i]; num_tx_slave++)
1324 rte_pktmbuf_free(slave_bufs[i][num_tx_slave]);
1326 num_tx_total += num_tx_slave - slave_slow_nb_pkts[i];
1327 num_tx_fail_total += slave_nb_pkts[i] - num_tx_slave;
1329 /* If tx burst fails move packets to end of bufs */
1330 if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
1331 uint16_t j = nb_pkts - num_tx_fail_total;
1332 for ( ; num_tx_slave < slave_nb_pkts[i]; j++, num_tx_slave++)
1333 bufs[j] = slave_bufs[i][num_tx_slave];
1337 return num_tx_total;
1341 bond_ethdev_tx_burst_broadcast(void *queue, struct rte_mbuf **bufs,
1344 struct bond_dev_private *internals;
1345 struct bond_tx_queue *bd_tx_q;
1347 uint8_t tx_failed_flag = 0, num_of_slaves;
1348 uint16_t slaves[RTE_MAX_ETHPORTS];
1350 uint16_t max_nb_of_tx_pkts = 0;
1352 int slave_tx_total[RTE_MAX_ETHPORTS];
1353 int i, most_successful_tx_slave = -1;
1355 bd_tx_q = (struct bond_tx_queue *)queue;
1356 internals = bd_tx_q->dev_private;
1358 /* Copy slave list to protect against slave up/down changes during tx
1360 num_of_slaves = internals->active_slave_count;
1361 memcpy(slaves, internals->active_slaves,
1362 sizeof(internals->active_slaves[0]) * num_of_slaves);
1364 if (num_of_slaves < 1)
1367 /* Increment reference count on mbufs */
1368 for (i = 0; i < nb_pkts; i++)
1369 rte_mbuf_refcnt_update(bufs[i], num_of_slaves - 1);
1371 /* Transmit burst on each active slave */
1372 for (i = 0; i < num_of_slaves; i++) {
1373 slave_tx_total[i] = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
1376 if (unlikely(slave_tx_total[i] < nb_pkts))
1379 /* record the value and slave index for the slave which transmits the
1380 * maximum number of packets */
1381 if (slave_tx_total[i] > max_nb_of_tx_pkts) {
1382 max_nb_of_tx_pkts = slave_tx_total[i];
1383 most_successful_tx_slave = i;
1387 /* if slaves fail to transmit packets from burst, the calling application
1388 * is not expected to know about multiple references to packets so we must
1389 * handle failures of all packets except those of the most successful slave
1391 if (unlikely(tx_failed_flag))
1392 for (i = 0; i < num_of_slaves; i++)
1393 if (i != most_successful_tx_slave)
1394 while (slave_tx_total[i] < nb_pkts)
1395 rte_pktmbuf_free(bufs[slave_tx_total[i]++]);
1397 return max_nb_of_tx_pkts;
1401 link_properties_set(struct rte_eth_dev *ethdev, struct rte_eth_link *slave_link)
1403 struct bond_dev_private *bond_ctx = ethdev->data->dev_private;
1405 if (bond_ctx->mode == BONDING_MODE_8023AD) {
1407 * If in mode 4 then save the link properties of the first
1408 * slave, all subsequent slaves must match these properties
1410 struct rte_eth_link *bond_link = &bond_ctx->mode4.slave_link;
1412 bond_link->link_autoneg = slave_link->link_autoneg;
1413 bond_link->link_duplex = slave_link->link_duplex;
1414 bond_link->link_speed = slave_link->link_speed;
1417 * In any other mode the link properties are set to default
1418 * values of AUTONEG/DUPLEX
1420 ethdev->data->dev_link.link_autoneg = ETH_LINK_AUTONEG;
1421 ethdev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
1426 link_properties_valid(struct rte_eth_dev *ethdev,
1427 struct rte_eth_link *slave_link)
1429 struct bond_dev_private *bond_ctx = ethdev->data->dev_private;
1431 if (bond_ctx->mode == BONDING_MODE_8023AD) {
1432 struct rte_eth_link *bond_link = &bond_ctx->mode4.slave_link;
1434 if (bond_link->link_duplex != slave_link->link_duplex ||
1435 bond_link->link_autoneg != slave_link->link_autoneg ||
1436 bond_link->link_speed != slave_link->link_speed)
1444 mac_address_get(struct rte_eth_dev *eth_dev, struct ether_addr *dst_mac_addr)
1446 struct ether_addr *mac_addr;
1448 if (eth_dev == NULL) {
1449 RTE_LOG(ERR, PMD, "%s: NULL pointer eth_dev specified\n", __func__);
1453 if (dst_mac_addr == NULL) {
1454 RTE_LOG(ERR, PMD, "%s: NULL pointer MAC specified\n", __func__);
1458 mac_addr = eth_dev->data->mac_addrs;
1460 ether_addr_copy(mac_addr, dst_mac_addr);
1465 mac_address_set(struct rte_eth_dev *eth_dev, struct ether_addr *new_mac_addr)
1467 struct ether_addr *mac_addr;
1469 if (eth_dev == NULL) {
1470 RTE_BOND_LOG(ERR, "NULL pointer eth_dev specified");
1474 if (new_mac_addr == NULL) {
1475 RTE_BOND_LOG(ERR, "NULL pointer MAC specified");
1479 mac_addr = eth_dev->data->mac_addrs;
1481 /* If new MAC is different to current MAC then update */
1482 if (memcmp(mac_addr, new_mac_addr, sizeof(*mac_addr)) != 0)
1483 memcpy(mac_addr, new_mac_addr, sizeof(*mac_addr));
1489 mac_address_slaves_update(struct rte_eth_dev *bonded_eth_dev)
1491 struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1494 /* Update slave devices MAC addresses */
1495 if (internals->slave_count < 1)
1498 switch (internals->mode) {
1499 case BONDING_MODE_ROUND_ROBIN:
1500 case BONDING_MODE_BALANCE:
1501 case BONDING_MODE_BROADCAST:
1502 for (i = 0; i < internals->slave_count; i++) {
1503 if (mac_address_set(&rte_eth_devices[internals->slaves[i].port_id],
1504 bonded_eth_dev->data->mac_addrs)) {
1505 RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1506 internals->slaves[i].port_id);
1511 case BONDING_MODE_8023AD:
1512 bond_mode_8023ad_mac_address_update(bonded_eth_dev);
1514 case BONDING_MODE_ACTIVE_BACKUP:
1515 case BONDING_MODE_TLB:
1516 case BONDING_MODE_ALB:
1518 for (i = 0; i < internals->slave_count; i++) {
1519 if (internals->slaves[i].port_id ==
1520 internals->current_primary_port) {
1521 if (mac_address_set(&rte_eth_devices[internals->primary_port],
1522 bonded_eth_dev->data->mac_addrs)) {
1523 RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1524 internals->current_primary_port);
1528 if (mac_address_set(
1529 &rte_eth_devices[internals->slaves[i].port_id],
1530 &internals->slaves[i].persisted_mac_addr)) {
1531 RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1532 internals->slaves[i].port_id);
1543 bond_ethdev_mode_set(struct rte_eth_dev *eth_dev, int mode)
1545 struct bond_dev_private *internals;
1547 internals = eth_dev->data->dev_private;
1550 case BONDING_MODE_ROUND_ROBIN:
1551 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_round_robin;
1552 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1554 case BONDING_MODE_ACTIVE_BACKUP:
1555 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_active_backup;
1556 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup;
1558 case BONDING_MODE_BALANCE:
1559 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_balance;
1560 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1562 case BONDING_MODE_BROADCAST:
1563 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_broadcast;
1564 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1566 case BONDING_MODE_8023AD:
1567 if (bond_mode_8023ad_enable(eth_dev) != 0)
1570 if (internals->mode4.dedicated_queues.enabled == 0) {
1571 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_8023ad;
1572 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_8023ad;
1573 RTE_LOG(WARNING, PMD,
1574 "Using mode 4, it is necessary to do TX burst "
1575 "and RX burst at least every 100ms.\n");
1577 /* Use flow director's optimization */
1578 eth_dev->rx_pkt_burst =
1579 bond_ethdev_rx_burst_8023ad_fast_queue;
1580 eth_dev->tx_pkt_burst =
1581 bond_ethdev_tx_burst_8023ad_fast_queue;
1584 case BONDING_MODE_TLB:
1585 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_tlb;
1586 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup;
1588 case BONDING_MODE_ALB:
1589 if (bond_mode_alb_enable(eth_dev) != 0)
1592 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_alb;
1593 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_alb;
1599 internals->mode = mode;
1606 slave_configure_slow_queue(struct rte_eth_dev *bonded_eth_dev,
1607 struct rte_eth_dev *slave_eth_dev)
1610 struct bond_dev_private *internals = (struct bond_dev_private *)
1611 bonded_eth_dev->data->dev_private;
1612 struct port *port = &mode_8023ad_ports[slave_eth_dev->data->port_id];
1614 if (port->slow_pool == NULL) {
1616 int slave_id = slave_eth_dev->data->port_id;
1618 snprintf(mem_name, RTE_DIM(mem_name), "slave_port%u_slow_pool",
1620 port->slow_pool = rte_pktmbuf_pool_create(mem_name, 8191,
1621 250, 0, RTE_MBUF_DEFAULT_BUF_SIZE,
1622 slave_eth_dev->data->numa_node);
1624 /* Any memory allocation failure in initialization is critical because
1625 * resources can't be free, so reinitialization is impossible. */
1626 if (port->slow_pool == NULL) {
1627 rte_panic("Slave %u: Failed to create memory pool '%s': %s\n",
1628 slave_id, mem_name, rte_strerror(rte_errno));
1632 if (internals->mode4.dedicated_queues.enabled == 1) {
1633 /* Configure slow Rx queue */
1635 errval = rte_eth_rx_queue_setup(slave_eth_dev->data->port_id,
1636 internals->mode4.dedicated_queues.rx_qid, 128,
1637 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1638 NULL, port->slow_pool);
1641 "rte_eth_rx_queue_setup: port=%d queue_id %d, err (%d)",
1642 slave_eth_dev->data->port_id,
1643 internals->mode4.dedicated_queues.rx_qid,
1648 errval = rte_eth_tx_queue_setup(slave_eth_dev->data->port_id,
1649 internals->mode4.dedicated_queues.tx_qid, 512,
1650 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1654 "rte_eth_tx_queue_setup: port=%d queue_id %d, err (%d)",
1655 slave_eth_dev->data->port_id,
1656 internals->mode4.dedicated_queues.tx_qid,
1665 slave_configure(struct rte_eth_dev *bonded_eth_dev,
1666 struct rte_eth_dev *slave_eth_dev)
1668 struct bond_rx_queue *bd_rx_q;
1669 struct bond_tx_queue *bd_tx_q;
1670 uint16_t nb_rx_queues;
1671 uint16_t nb_tx_queues;
1675 struct rte_flow_error flow_error;
1677 struct bond_dev_private *internals = (struct bond_dev_private *)
1678 bonded_eth_dev->data->dev_private;
1681 rte_eth_dev_stop(slave_eth_dev->data->port_id);
1683 /* Enable interrupts on slave device if supported */
1684 if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1685 slave_eth_dev->data->dev_conf.intr_conf.lsc = 1;
1687 /* If RSS is enabled for bonding, try to enable it for slaves */
1688 if (bonded_eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG) {
1689 if (bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len
1691 slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len =
1692 bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len;
1693 slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key =
1694 bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key;
1696 slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
1699 slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf =
1700 bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
1701 slave_eth_dev->data->dev_conf.rxmode.mq_mode =
1702 bonded_eth_dev->data->dev_conf.rxmode.mq_mode;
1705 slave_eth_dev->data->dev_conf.rxmode.hw_vlan_filter =
1706 bonded_eth_dev->data->dev_conf.rxmode.hw_vlan_filter;
1708 nb_rx_queues = bonded_eth_dev->data->nb_rx_queues;
1709 nb_tx_queues = bonded_eth_dev->data->nb_tx_queues;
1711 if (internals->mode == BONDING_MODE_8023AD) {
1712 if (internals->mode4.dedicated_queues.enabled == 1) {
1718 /* Configure device */
1719 errval = rte_eth_dev_configure(slave_eth_dev->data->port_id,
1720 nb_rx_queues, nb_tx_queues,
1721 &(slave_eth_dev->data->dev_conf));
1723 RTE_BOND_LOG(ERR, "Cannot configure slave device: port %u , err (%d)",
1724 slave_eth_dev->data->port_id, errval);
1728 /* Setup Rx Queues */
1729 for (q_id = 0; q_id < bonded_eth_dev->data->nb_rx_queues; q_id++) {
1730 bd_rx_q = (struct bond_rx_queue *)bonded_eth_dev->data->rx_queues[q_id];
1732 errval = rte_eth_rx_queue_setup(slave_eth_dev->data->port_id, q_id,
1733 bd_rx_q->nb_rx_desc,
1734 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1735 &(bd_rx_q->rx_conf), bd_rx_q->mb_pool);
1738 "rte_eth_rx_queue_setup: port=%d queue_id %d, err (%d)",
1739 slave_eth_dev->data->port_id, q_id, errval);
1744 /* Setup Tx Queues */
1745 for (q_id = 0; q_id < bonded_eth_dev->data->nb_tx_queues; q_id++) {
1746 bd_tx_q = (struct bond_tx_queue *)bonded_eth_dev->data->tx_queues[q_id];
1748 errval = rte_eth_tx_queue_setup(slave_eth_dev->data->port_id, q_id,
1749 bd_tx_q->nb_tx_desc,
1750 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1754 "rte_eth_tx_queue_setup: port=%d queue_id %d, err (%d)",
1755 slave_eth_dev->data->port_id, q_id, errval);
1760 if (internals->mode == BONDING_MODE_8023AD &&
1761 internals->mode4.dedicated_queues.enabled == 1) {
1762 if (slave_configure_slow_queue(bonded_eth_dev, slave_eth_dev)
1766 if (bond_ethdev_8023ad_flow_verify(bonded_eth_dev,
1767 slave_eth_dev->data->port_id) != 0) {
1769 "rte_eth_tx_queue_setup: port=%d queue_id %d, err (%d)",
1770 slave_eth_dev->data->port_id, q_id, errval);
1774 if (internals->mode4.dedicated_queues.flow[slave_eth_dev->data->port_id] != NULL)
1775 rte_flow_destroy(slave_eth_dev->data->port_id,
1776 internals->mode4.dedicated_queues.flow[slave_eth_dev->data->port_id],
1779 bond_ethdev_8023ad_flow_set(bonded_eth_dev,
1780 slave_eth_dev->data->port_id);
1784 errval = rte_eth_dev_start(slave_eth_dev->data->port_id);
1786 RTE_BOND_LOG(ERR, "rte_eth_dev_start: port=%u, err (%d)",
1787 slave_eth_dev->data->port_id, errval);
1791 /* If RSS is enabled for bonding, synchronize RETA */
1792 if (bonded_eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS) {
1794 struct bond_dev_private *internals;
1796 internals = bonded_eth_dev->data->dev_private;
1798 for (i = 0; i < internals->slave_count; i++) {
1799 if (internals->slaves[i].port_id == slave_eth_dev->data->port_id) {
1800 errval = rte_eth_dev_rss_reta_update(
1801 slave_eth_dev->data->port_id,
1802 &internals->reta_conf[0],
1803 internals->slaves[i].reta_size);
1805 RTE_LOG(WARNING, PMD,
1806 "rte_eth_dev_rss_reta_update on slave port %d fails (err %d)."
1807 " RSS Configuration for bonding may be inconsistent.\n",
1808 slave_eth_dev->data->port_id, errval);
1815 /* If lsc interrupt is set, check initial slave's link status */
1816 if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
1817 slave_eth_dev->dev_ops->link_update(slave_eth_dev, 0);
1818 bond_ethdev_lsc_event_callback(slave_eth_dev->data->port_id,
1819 RTE_ETH_EVENT_INTR_LSC, &bonded_eth_dev->data->port_id,
1827 slave_remove(struct bond_dev_private *internals,
1828 struct rte_eth_dev *slave_eth_dev)
1832 for (i = 0; i < internals->slave_count; i++)
1833 if (internals->slaves[i].port_id ==
1834 slave_eth_dev->data->port_id)
1837 if (i < (internals->slave_count - 1))
1838 memmove(&internals->slaves[i], &internals->slaves[i + 1],
1839 sizeof(internals->slaves[0]) *
1840 (internals->slave_count - i - 1));
1842 internals->slave_count--;
1844 /* force reconfiguration of slave interfaces */
1845 _rte_eth_dev_reset(slave_eth_dev);
1849 bond_ethdev_slave_link_status_change_monitor(void *cb_arg);
1852 slave_add(struct bond_dev_private *internals,
1853 struct rte_eth_dev *slave_eth_dev)
1855 struct bond_slave_details *slave_details =
1856 &internals->slaves[internals->slave_count];
1858 slave_details->port_id = slave_eth_dev->data->port_id;
1859 slave_details->last_link_status = 0;
1861 /* Mark slave devices that don't support interrupts so we can
1862 * compensate when we start the bond
1864 if (!(slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1865 slave_details->link_status_poll_enabled = 1;
1868 slave_details->link_status_wait_to_complete = 0;
1869 /* clean tlb_last_obytes when adding port for bonding device */
1870 memcpy(&(slave_details->persisted_mac_addr), slave_eth_dev->data->mac_addrs,
1871 sizeof(struct ether_addr));
1875 bond_ethdev_primary_set(struct bond_dev_private *internals,
1876 uint16_t slave_port_id)
1880 if (internals->active_slave_count < 1)
1881 internals->current_primary_port = slave_port_id;
1883 /* Search bonded device slave ports for new proposed primary port */
1884 for (i = 0; i < internals->active_slave_count; i++) {
1885 if (internals->active_slaves[i] == slave_port_id)
1886 internals->current_primary_port = slave_port_id;
1891 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev);
1894 bond_ethdev_start(struct rte_eth_dev *eth_dev)
1896 struct bond_dev_private *internals;
1899 /* slave eth dev will be started by bonded device */
1900 if (check_for_bonded_ethdev(eth_dev)) {
1901 RTE_BOND_LOG(ERR, "User tried to explicitly start a slave eth_dev (%d)",
1902 eth_dev->data->port_id);
1906 eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
1907 eth_dev->data->dev_started = 1;
1909 internals = eth_dev->data->dev_private;
1911 if (internals->slave_count == 0) {
1912 RTE_BOND_LOG(ERR, "Cannot start port since there are no slave devices");
1916 if (internals->user_defined_mac == 0) {
1917 struct ether_addr *new_mac_addr = NULL;
1919 for (i = 0; i < internals->slave_count; i++)
1920 if (internals->slaves[i].port_id == internals->primary_port)
1921 new_mac_addr = &internals->slaves[i].persisted_mac_addr;
1923 if (new_mac_addr == NULL)
1926 if (mac_address_set(eth_dev, new_mac_addr) != 0) {
1927 RTE_BOND_LOG(ERR, "bonded port (%d) failed to update MAC address",
1928 eth_dev->data->port_id);
1933 /* Update all slave devices MACs*/
1934 if (mac_address_slaves_update(eth_dev) != 0)
1937 /* If bonded device is configure in promiscuous mode then re-apply config */
1938 if (internals->promiscuous_en)
1939 bond_ethdev_promiscuous_enable(eth_dev);
1941 if (internals->mode == BONDING_MODE_8023AD) {
1942 if (internals->mode4.dedicated_queues.enabled == 1) {
1943 internals->mode4.dedicated_queues.rx_qid =
1944 eth_dev->data->nb_rx_queues;
1945 internals->mode4.dedicated_queues.tx_qid =
1946 eth_dev->data->nb_tx_queues;
1951 /* Reconfigure each slave device if starting bonded device */
1952 for (i = 0; i < internals->slave_count; i++) {
1953 struct rte_eth_dev *slave_ethdev =
1954 &(rte_eth_devices[internals->slaves[i].port_id]);
1955 if (slave_configure(eth_dev, slave_ethdev) != 0) {
1957 "bonded port (%d) failed to reconfigure slave device (%d)",
1958 eth_dev->data->port_id,
1959 internals->slaves[i].port_id);
1962 /* We will need to poll for link status if any slave doesn't
1963 * support interrupts
1965 if (internals->slaves[i].link_status_poll_enabled)
1966 internals->link_status_polling_enabled = 1;
1968 /* start polling if needed */
1969 if (internals->link_status_polling_enabled) {
1971 internals->link_status_polling_interval_ms * 1000,
1972 bond_ethdev_slave_link_status_change_monitor,
1973 (void *)&rte_eth_devices[internals->port_id]);
1976 if (internals->user_defined_primary_port)
1977 bond_ethdev_primary_set(internals, internals->primary_port);
1979 if (internals->mode == BONDING_MODE_8023AD)
1980 bond_mode_8023ad_start(eth_dev);
1982 if (internals->mode == BONDING_MODE_TLB ||
1983 internals->mode == BONDING_MODE_ALB)
1984 bond_tlb_enable(internals);
1990 bond_ethdev_free_queues(struct rte_eth_dev *dev)
1994 if (dev->data->rx_queues != NULL) {
1995 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1996 rte_free(dev->data->rx_queues[i]);
1997 dev->data->rx_queues[i] = NULL;
1999 dev->data->nb_rx_queues = 0;
2002 if (dev->data->tx_queues != NULL) {
2003 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2004 rte_free(dev->data->tx_queues[i]);
2005 dev->data->tx_queues[i] = NULL;
2007 dev->data->nb_tx_queues = 0;
2012 bond_ethdev_stop(struct rte_eth_dev *eth_dev)
2014 struct bond_dev_private *internals = eth_dev->data->dev_private;
2017 if (internals->mode == BONDING_MODE_8023AD) {
2021 bond_mode_8023ad_stop(eth_dev);
2023 /* Discard all messages to/from mode 4 state machines */
2024 for (i = 0; i < internals->active_slave_count; i++) {
2025 port = &mode_8023ad_ports[internals->active_slaves[i]];
2027 RTE_ASSERT(port->rx_ring != NULL);
2028 while (rte_ring_dequeue(port->rx_ring, &pkt) != -ENOENT)
2029 rte_pktmbuf_free(pkt);
2031 RTE_ASSERT(port->tx_ring != NULL);
2032 while (rte_ring_dequeue(port->tx_ring, &pkt) != -ENOENT)
2033 rte_pktmbuf_free(pkt);
2037 if (internals->mode == BONDING_MODE_TLB ||
2038 internals->mode == BONDING_MODE_ALB) {
2039 bond_tlb_disable(internals);
2040 for (i = 0; i < internals->active_slave_count; i++)
2041 tlb_last_obytets[internals->active_slaves[i]] = 0;
2044 internals->active_slave_count = 0;
2045 internals->link_status_polling_enabled = 0;
2046 for (i = 0; i < internals->slave_count; i++)
2047 internals->slaves[i].last_link_status = 0;
2049 eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
2050 eth_dev->data->dev_started = 0;
2054 bond_ethdev_close(struct rte_eth_dev *dev)
2056 struct bond_dev_private *internals = dev->data->dev_private;
2057 uint8_t bond_port_id = internals->port_id;
2060 RTE_LOG(INFO, EAL, "Closing bonded device %s\n", dev->device->name);
2061 while (internals->slave_count != skipped) {
2062 uint16_t port_id = internals->slaves[skipped].port_id;
2064 rte_eth_dev_stop(port_id);
2066 if (rte_eth_bond_slave_remove(bond_port_id, port_id) != 0) {
2068 "Failed to remove port %d from bonded device "
2069 "%s\n", port_id, dev->device->name);
2073 bond_ethdev_free_queues(dev);
2074 rte_bitmap_reset(internals->vlan_filter_bmp);
2077 /* forward declaration */
2078 static int bond_ethdev_configure(struct rte_eth_dev *dev);
2081 bond_ethdev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2083 struct bond_dev_private *internals = dev->data->dev_private;
2085 uint16_t max_nb_rx_queues = UINT16_MAX;
2086 uint16_t max_nb_tx_queues = UINT16_MAX;
2088 dev_info->max_mac_addrs = 1;
2090 dev_info->max_rx_pktlen = internals->candidate_max_rx_pktlen ?
2091 internals->candidate_max_rx_pktlen :
2092 ETHER_MAX_JUMBO_FRAME_LEN;
2094 /* Max number of tx/rx queues that the bonded device can support is the
2095 * minimum values of the bonded slaves, as all slaves must be capable
2096 * of supporting the same number of tx/rx queues.
2098 if (internals->slave_count > 0) {
2099 struct rte_eth_dev_info slave_info;
2102 for (idx = 0; idx < internals->slave_count; idx++) {
2103 rte_eth_dev_info_get(internals->slaves[idx].port_id,
2106 if (slave_info.max_rx_queues < max_nb_rx_queues)
2107 max_nb_rx_queues = slave_info.max_rx_queues;
2109 if (slave_info.max_tx_queues < max_nb_tx_queues)
2110 max_nb_tx_queues = slave_info.max_tx_queues;
2114 dev_info->max_rx_queues = max_nb_rx_queues;
2115 dev_info->max_tx_queues = max_nb_tx_queues;
2118 * If dedicated hw queues enabled for link bonding device in LACP mode
2119 * then we need to reduce the maximum number of data path queues by 1.
2121 if (internals->mode == BONDING_MODE_8023AD &&
2122 internals->mode4.dedicated_queues.enabled == 1) {
2123 dev_info->max_rx_queues--;
2124 dev_info->max_tx_queues--;
2127 dev_info->min_rx_bufsize = 0;
2129 dev_info->rx_offload_capa = internals->rx_offload_capa;
2130 dev_info->tx_offload_capa = internals->tx_offload_capa;
2131 dev_info->flow_type_rss_offloads = internals->flow_type_rss_offloads;
2133 dev_info->reta_size = internals->reta_size;
2137 bond_ethdev_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
2141 struct bond_dev_private *internals = dev->data->dev_private;
2143 /* don't do this while a slave is being added */
2144 rte_spinlock_lock(&internals->lock);
2147 rte_bitmap_set(internals->vlan_filter_bmp, vlan_id);
2149 rte_bitmap_clear(internals->vlan_filter_bmp, vlan_id);
2151 for (i = 0; i < internals->slave_count; i++) {
2152 uint16_t port_id = internals->slaves[i].port_id;
2154 res = rte_eth_dev_vlan_filter(port_id, vlan_id, on);
2156 RTE_LOG(WARNING, PMD,
2157 "Setting VLAN filter on slave port %u not supported.\n",
2161 rte_spinlock_unlock(&internals->lock);
2166 bond_ethdev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
2167 uint16_t nb_rx_desc, unsigned int socket_id __rte_unused,
2168 const struct rte_eth_rxconf *rx_conf, struct rte_mempool *mb_pool)
2170 struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)
2171 rte_zmalloc_socket(NULL, sizeof(struct bond_rx_queue),
2172 0, dev->data->numa_node);
2173 if (bd_rx_q == NULL)
2176 bd_rx_q->queue_id = rx_queue_id;
2177 bd_rx_q->dev_private = dev->data->dev_private;
2179 bd_rx_q->nb_rx_desc = nb_rx_desc;
2181 memcpy(&(bd_rx_q->rx_conf), rx_conf, sizeof(struct rte_eth_rxconf));
2182 bd_rx_q->mb_pool = mb_pool;
2184 dev->data->rx_queues[rx_queue_id] = bd_rx_q;
2190 bond_ethdev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
2191 uint16_t nb_tx_desc, unsigned int socket_id __rte_unused,
2192 const struct rte_eth_txconf *tx_conf)
2194 struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)
2195 rte_zmalloc_socket(NULL, sizeof(struct bond_tx_queue),
2196 0, dev->data->numa_node);
2198 if (bd_tx_q == NULL)
2201 bd_tx_q->queue_id = tx_queue_id;
2202 bd_tx_q->dev_private = dev->data->dev_private;
2204 bd_tx_q->nb_tx_desc = nb_tx_desc;
2205 memcpy(&(bd_tx_q->tx_conf), tx_conf, sizeof(bd_tx_q->tx_conf));
2207 dev->data->tx_queues[tx_queue_id] = bd_tx_q;
2213 bond_ethdev_rx_queue_release(void *queue)
2222 bond_ethdev_tx_queue_release(void *queue)
2231 bond_ethdev_slave_link_status_change_monitor(void *cb_arg)
2233 struct rte_eth_dev *bonded_ethdev, *slave_ethdev;
2234 struct bond_dev_private *internals;
2236 /* Default value for polling slave found is true as we don't want to
2237 * disable the polling thread if we cannot get the lock */
2238 int i, polling_slave_found = 1;
2243 bonded_ethdev = (struct rte_eth_dev *)cb_arg;
2244 internals = (struct bond_dev_private *)bonded_ethdev->data->dev_private;
2246 if (!bonded_ethdev->data->dev_started ||
2247 !internals->link_status_polling_enabled)
2250 /* If device is currently being configured then don't check slaves link
2251 * status, wait until next period */
2252 if (rte_spinlock_trylock(&internals->lock)) {
2253 if (internals->slave_count > 0)
2254 polling_slave_found = 0;
2256 for (i = 0; i < internals->slave_count; i++) {
2257 if (!internals->slaves[i].link_status_poll_enabled)
2260 slave_ethdev = &rte_eth_devices[internals->slaves[i].port_id];
2261 polling_slave_found = 1;
2263 /* Update slave link status */
2264 (*slave_ethdev->dev_ops->link_update)(slave_ethdev,
2265 internals->slaves[i].link_status_wait_to_complete);
2267 /* if link status has changed since last checked then call lsc
2269 if (slave_ethdev->data->dev_link.link_status !=
2270 internals->slaves[i].last_link_status) {
2271 internals->slaves[i].last_link_status =
2272 slave_ethdev->data->dev_link.link_status;
2274 bond_ethdev_lsc_event_callback(internals->slaves[i].port_id,
2275 RTE_ETH_EVENT_INTR_LSC,
2276 &bonded_ethdev->data->port_id,
2280 rte_spinlock_unlock(&internals->lock);
2283 if (polling_slave_found)
2284 /* Set alarm to continue monitoring link status of slave ethdev's */
2285 rte_eal_alarm_set(internals->link_status_polling_interval_ms * 1000,
2286 bond_ethdev_slave_link_status_change_monitor, cb_arg);
2290 bond_ethdev_link_update(struct rte_eth_dev *ethdev, int wait_to_complete)
2292 void (*link_update)(uint16_t port_id, struct rte_eth_link *eth_link);
2294 struct bond_dev_private *bond_ctx;
2295 struct rte_eth_link slave_link;
2299 bond_ctx = ethdev->data->dev_private;
2301 ethdev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
2303 if (ethdev->data->dev_started == 0 ||
2304 bond_ctx->active_slave_count == 0) {
2305 ethdev->data->dev_link.link_status = ETH_LINK_DOWN;
2309 ethdev->data->dev_link.link_status = ETH_LINK_UP;
2311 if (wait_to_complete)
2312 link_update = rte_eth_link_get;
2314 link_update = rte_eth_link_get_nowait;
2316 switch (bond_ctx->mode) {
2317 case BONDING_MODE_BROADCAST:
2319 * Setting link speed to UINT32_MAX to ensure we pick up the
2320 * value of the first active slave
2322 ethdev->data->dev_link.link_speed = UINT32_MAX;
2325 * link speed is minimum value of all the slaves link speed as
2326 * packet loss will occur on this slave if transmission at rates
2327 * greater than this are attempted
2329 for (idx = 1; idx < bond_ctx->active_slave_count; idx++) {
2330 link_update(bond_ctx->active_slaves[0], &slave_link);
2332 if (slave_link.link_speed <
2333 ethdev->data->dev_link.link_speed)
2334 ethdev->data->dev_link.link_speed =
2335 slave_link.link_speed;
2338 case BONDING_MODE_ACTIVE_BACKUP:
2339 /* Current primary slave */
2340 link_update(bond_ctx->current_primary_port, &slave_link);
2342 ethdev->data->dev_link.link_speed = slave_link.link_speed;
2344 case BONDING_MODE_8023AD:
2345 ethdev->data->dev_link.link_autoneg =
2346 bond_ctx->mode4.slave_link.link_autoneg;
2347 ethdev->data->dev_link.link_duplex =
2348 bond_ctx->mode4.slave_link.link_duplex;
2349 /* fall through to update link speed */
2350 case BONDING_MODE_ROUND_ROBIN:
2351 case BONDING_MODE_BALANCE:
2352 case BONDING_MODE_TLB:
2353 case BONDING_MODE_ALB:
2356 * In theses mode the maximum theoretical link speed is the sum
2359 ethdev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
2361 for (idx = 0; idx < bond_ctx->active_slave_count; idx++) {
2362 link_update(bond_ctx->active_slaves[idx], &slave_link);
2364 ethdev->data->dev_link.link_speed +=
2365 slave_link.link_speed;
2375 bond_ethdev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
2377 struct bond_dev_private *internals = dev->data->dev_private;
2378 struct rte_eth_stats slave_stats;
2381 for (i = 0; i < internals->slave_count; i++) {
2382 rte_eth_stats_get(internals->slaves[i].port_id, &slave_stats);
2384 stats->ipackets += slave_stats.ipackets;
2385 stats->opackets += slave_stats.opackets;
2386 stats->ibytes += slave_stats.ibytes;
2387 stats->obytes += slave_stats.obytes;
2388 stats->imissed += slave_stats.imissed;
2389 stats->ierrors += slave_stats.ierrors;
2390 stats->oerrors += slave_stats.oerrors;
2391 stats->rx_nombuf += slave_stats.rx_nombuf;
2393 for (j = 0; j < RTE_ETHDEV_QUEUE_STAT_CNTRS; j++) {
2394 stats->q_ipackets[j] += slave_stats.q_ipackets[j];
2395 stats->q_opackets[j] += slave_stats.q_opackets[j];
2396 stats->q_ibytes[j] += slave_stats.q_ibytes[j];
2397 stats->q_obytes[j] += slave_stats.q_obytes[j];
2398 stats->q_errors[j] += slave_stats.q_errors[j];
2407 bond_ethdev_stats_reset(struct rte_eth_dev *dev)
2409 struct bond_dev_private *internals = dev->data->dev_private;
2412 for (i = 0; i < internals->slave_count; i++)
2413 rte_eth_stats_reset(internals->slaves[i].port_id);
2417 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev)
2419 struct bond_dev_private *internals = eth_dev->data->dev_private;
2422 internals->promiscuous_en = 1;
2424 switch (internals->mode) {
2425 /* Promiscuous mode is propagated to all slaves */
2426 case BONDING_MODE_ROUND_ROBIN:
2427 case BONDING_MODE_BALANCE:
2428 case BONDING_MODE_BROADCAST:
2429 for (i = 0; i < internals->slave_count; i++)
2430 rte_eth_promiscuous_enable(internals->slaves[i].port_id);
2432 /* In mode4 promiscus mode is managed when slave is added/removed */
2433 case BONDING_MODE_8023AD:
2435 /* Promiscuous mode is propagated only to primary slave */
2436 case BONDING_MODE_ACTIVE_BACKUP:
2437 case BONDING_MODE_TLB:
2438 case BONDING_MODE_ALB:
2440 rte_eth_promiscuous_enable(internals->current_primary_port);
2445 bond_ethdev_promiscuous_disable(struct rte_eth_dev *dev)
2447 struct bond_dev_private *internals = dev->data->dev_private;
2450 internals->promiscuous_en = 0;
2452 switch (internals->mode) {
2453 /* Promiscuous mode is propagated to all slaves */
2454 case BONDING_MODE_ROUND_ROBIN:
2455 case BONDING_MODE_BALANCE:
2456 case BONDING_MODE_BROADCAST:
2457 for (i = 0; i < internals->slave_count; i++)
2458 rte_eth_promiscuous_disable(internals->slaves[i].port_id);
2460 /* In mode4 promiscus mode is set managed when slave is added/removed */
2461 case BONDING_MODE_8023AD:
2463 /* Promiscuous mode is propagated only to primary slave */
2464 case BONDING_MODE_ACTIVE_BACKUP:
2465 case BONDING_MODE_TLB:
2466 case BONDING_MODE_ALB:
2468 rte_eth_promiscuous_disable(internals->current_primary_port);
2473 bond_ethdev_delayed_lsc_propagation(void *arg)
2478 _rte_eth_dev_callback_process((struct rte_eth_dev *)arg,
2479 RTE_ETH_EVENT_INTR_LSC, NULL, NULL);
2483 bond_ethdev_lsc_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2484 void *param, void *ret_param __rte_unused)
2486 struct rte_eth_dev *bonded_eth_dev;
2487 struct bond_dev_private *internals;
2488 struct rte_eth_link link;
2491 int i, valid_slave = 0;
2493 uint8_t lsc_flag = 0;
2495 if (type != RTE_ETH_EVENT_INTR_LSC || param == NULL)
2498 bonded_eth_dev = &rte_eth_devices[*(uint8_t *)param];
2500 if (check_for_bonded_ethdev(bonded_eth_dev))
2503 internals = bonded_eth_dev->data->dev_private;
2505 /* If the device isn't started don't handle interrupts */
2506 if (!bonded_eth_dev->data->dev_started)
2509 /* verify that port_id is a valid slave of bonded port */
2510 for (i = 0; i < internals->slave_count; i++) {
2511 if (internals->slaves[i].port_id == port_id) {
2520 /* Search for port in active port list */
2521 active_pos = find_slave_by_id(internals->active_slaves,
2522 internals->active_slave_count, port_id);
2524 rte_eth_link_get_nowait(port_id, &link);
2525 if (link.link_status) {
2526 if (active_pos < internals->active_slave_count)
2529 /* if no active slave ports then set this port to be primary port */
2530 if (internals->active_slave_count < 1) {
2531 /* If first active slave, then change link status */
2532 bonded_eth_dev->data->dev_link.link_status = ETH_LINK_UP;
2533 internals->current_primary_port = port_id;
2536 mac_address_slaves_update(bonded_eth_dev);
2539 activate_slave(bonded_eth_dev, port_id);
2541 /* If user has defined the primary port then default to using it */
2542 if (internals->user_defined_primary_port &&
2543 internals->primary_port == port_id)
2544 bond_ethdev_primary_set(internals, port_id);
2546 if (active_pos == internals->active_slave_count)
2549 /* Remove from active slave list */
2550 deactivate_slave(bonded_eth_dev, port_id);
2552 if (internals->active_slave_count < 1)
2555 /* Update primary id, take first active slave from list or if none
2556 * available set to -1 */
2557 if (port_id == internals->current_primary_port) {
2558 if (internals->active_slave_count > 0)
2559 bond_ethdev_primary_set(internals,
2560 internals->active_slaves[0]);
2562 internals->current_primary_port = internals->primary_port;
2567 * Update bonded device link properties after any change to active
2570 bond_ethdev_link_update(bonded_eth_dev, 0);
2573 /* Cancel any possible outstanding interrupts if delays are enabled */
2574 if (internals->link_up_delay_ms > 0 ||
2575 internals->link_down_delay_ms > 0)
2576 rte_eal_alarm_cancel(bond_ethdev_delayed_lsc_propagation,
2579 if (bonded_eth_dev->data->dev_link.link_status) {
2580 if (internals->link_up_delay_ms > 0)
2581 rte_eal_alarm_set(internals->link_up_delay_ms * 1000,
2582 bond_ethdev_delayed_lsc_propagation,
2583 (void *)bonded_eth_dev);
2585 _rte_eth_dev_callback_process(bonded_eth_dev,
2586 RTE_ETH_EVENT_INTR_LSC,
2590 if (internals->link_down_delay_ms > 0)
2591 rte_eal_alarm_set(internals->link_down_delay_ms * 1000,
2592 bond_ethdev_delayed_lsc_propagation,
2593 (void *)bonded_eth_dev);
2595 _rte_eth_dev_callback_process(bonded_eth_dev,
2596 RTE_ETH_EVENT_INTR_LSC,
2604 bond_ethdev_rss_reta_update(struct rte_eth_dev *dev,
2605 struct rte_eth_rss_reta_entry64 *reta_conf, uint16_t reta_size)
2609 int slave_reta_size;
2610 unsigned reta_count;
2611 struct bond_dev_private *internals = dev->data->dev_private;
2613 if (reta_size != internals->reta_size)
2616 /* Copy RETA table */
2617 reta_count = reta_size / RTE_RETA_GROUP_SIZE;
2619 for (i = 0; i < reta_count; i++) {
2620 internals->reta_conf[i].mask = reta_conf[i].mask;
2621 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
2622 if ((reta_conf[i].mask >> j) & 0x01)
2623 internals->reta_conf[i].reta[j] = reta_conf[i].reta[j];
2626 /* Fill rest of array */
2627 for (; i < RTE_DIM(internals->reta_conf); i += reta_count)
2628 memcpy(&internals->reta_conf[i], &internals->reta_conf[0],
2629 sizeof(internals->reta_conf[0]) * reta_count);
2631 /* Propagate RETA over slaves */
2632 for (i = 0; i < internals->slave_count; i++) {
2633 slave_reta_size = internals->slaves[i].reta_size;
2634 result = rte_eth_dev_rss_reta_update(internals->slaves[i].port_id,
2635 &internals->reta_conf[0], slave_reta_size);
2644 bond_ethdev_rss_reta_query(struct rte_eth_dev *dev,
2645 struct rte_eth_rss_reta_entry64 *reta_conf, uint16_t reta_size)
2648 struct bond_dev_private *internals = dev->data->dev_private;
2650 if (reta_size != internals->reta_size)
2653 /* Copy RETA table */
2654 for (i = 0; i < reta_size / RTE_RETA_GROUP_SIZE; i++)
2655 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
2656 if ((reta_conf[i].mask >> j) & 0x01)
2657 reta_conf[i].reta[j] = internals->reta_conf[i].reta[j];
2663 bond_ethdev_rss_hash_update(struct rte_eth_dev *dev,
2664 struct rte_eth_rss_conf *rss_conf)
2667 struct bond_dev_private *internals = dev->data->dev_private;
2668 struct rte_eth_rss_conf bond_rss_conf;
2670 memcpy(&bond_rss_conf, rss_conf, sizeof(struct rte_eth_rss_conf));
2672 bond_rss_conf.rss_hf &= internals->flow_type_rss_offloads;
2674 if (bond_rss_conf.rss_hf != 0)
2675 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf = bond_rss_conf.rss_hf;
2677 if (bond_rss_conf.rss_key && bond_rss_conf.rss_key_len <
2678 sizeof(internals->rss_key)) {
2679 if (bond_rss_conf.rss_key_len == 0)
2680 bond_rss_conf.rss_key_len = 40;
2681 internals->rss_key_len = bond_rss_conf.rss_key_len;
2682 memcpy(internals->rss_key, bond_rss_conf.rss_key,
2683 internals->rss_key_len);
2686 for (i = 0; i < internals->slave_count; i++) {
2687 result = rte_eth_dev_rss_hash_update(internals->slaves[i].port_id,
2697 bond_ethdev_rss_hash_conf_get(struct rte_eth_dev *dev,
2698 struct rte_eth_rss_conf *rss_conf)
2700 struct bond_dev_private *internals = dev->data->dev_private;
2702 rss_conf->rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
2703 rss_conf->rss_key_len = internals->rss_key_len;
2704 if (rss_conf->rss_key)
2705 memcpy(rss_conf->rss_key, internals->rss_key, internals->rss_key_len);
2710 const struct eth_dev_ops default_dev_ops = {
2711 .dev_start = bond_ethdev_start,
2712 .dev_stop = bond_ethdev_stop,
2713 .dev_close = bond_ethdev_close,
2714 .dev_configure = bond_ethdev_configure,
2715 .dev_infos_get = bond_ethdev_info,
2716 .vlan_filter_set = bond_ethdev_vlan_filter_set,
2717 .rx_queue_setup = bond_ethdev_rx_queue_setup,
2718 .tx_queue_setup = bond_ethdev_tx_queue_setup,
2719 .rx_queue_release = bond_ethdev_rx_queue_release,
2720 .tx_queue_release = bond_ethdev_tx_queue_release,
2721 .link_update = bond_ethdev_link_update,
2722 .stats_get = bond_ethdev_stats_get,
2723 .stats_reset = bond_ethdev_stats_reset,
2724 .promiscuous_enable = bond_ethdev_promiscuous_enable,
2725 .promiscuous_disable = bond_ethdev_promiscuous_disable,
2726 .reta_update = bond_ethdev_rss_reta_update,
2727 .reta_query = bond_ethdev_rss_reta_query,
2728 .rss_hash_update = bond_ethdev_rss_hash_update,
2729 .rss_hash_conf_get = bond_ethdev_rss_hash_conf_get
2733 bond_alloc(struct rte_vdev_device *dev, uint8_t mode)
2735 const char *name = rte_vdev_device_name(dev);
2736 uint8_t socket_id = dev->device.numa_node;
2737 struct bond_dev_private *internals = NULL;
2738 struct rte_eth_dev *eth_dev = NULL;
2739 uint32_t vlan_filter_bmp_size;
2741 /* now do all data allocation - for eth_dev structure, dummy pci driver
2742 * and internal (private) data
2745 /* reserve an ethdev entry */
2746 eth_dev = rte_eth_vdev_allocate(dev, sizeof(*internals));
2747 if (eth_dev == NULL) {
2748 RTE_BOND_LOG(ERR, "Unable to allocate rte_eth_dev");
2752 internals = eth_dev->data->dev_private;
2753 eth_dev->data->nb_rx_queues = (uint16_t)1;
2754 eth_dev->data->nb_tx_queues = (uint16_t)1;
2756 eth_dev->data->mac_addrs = rte_zmalloc_socket(name, ETHER_ADDR_LEN, 0,
2758 if (eth_dev->data->mac_addrs == NULL) {
2759 RTE_BOND_LOG(ERR, "Unable to malloc mac_addrs");
2763 eth_dev->dev_ops = &default_dev_ops;
2764 eth_dev->data->dev_flags = RTE_ETH_DEV_INTR_LSC;
2766 rte_spinlock_init(&internals->lock);
2768 internals->port_id = eth_dev->data->port_id;
2769 internals->mode = BONDING_MODE_INVALID;
2770 internals->current_primary_port = RTE_MAX_ETHPORTS + 1;
2771 internals->balance_xmit_policy = BALANCE_XMIT_POLICY_LAYER2;
2772 internals->xmit_hash = xmit_l2_hash;
2773 internals->user_defined_mac = 0;
2775 internals->link_status_polling_enabled = 0;
2777 internals->link_status_polling_interval_ms =
2778 DEFAULT_POLLING_INTERVAL_10_MS;
2779 internals->link_down_delay_ms = 0;
2780 internals->link_up_delay_ms = 0;
2782 internals->slave_count = 0;
2783 internals->active_slave_count = 0;
2784 internals->rx_offload_capa = 0;
2785 internals->tx_offload_capa = 0;
2786 internals->candidate_max_rx_pktlen = 0;
2787 internals->max_rx_pktlen = 0;
2789 /* Initially allow to choose any offload type */
2790 internals->flow_type_rss_offloads = ETH_RSS_PROTO_MASK;
2792 memset(internals->active_slaves, 0, sizeof(internals->active_slaves));
2793 memset(internals->slaves, 0, sizeof(internals->slaves));
2795 /* Set mode 4 default configuration */
2796 bond_mode_8023ad_setup(eth_dev, NULL);
2797 if (bond_ethdev_mode_set(eth_dev, mode)) {
2798 RTE_BOND_LOG(ERR, "Failed to set bonded device %d mode too %d",
2799 eth_dev->data->port_id, mode);
2803 vlan_filter_bmp_size =
2804 rte_bitmap_get_memory_footprint(ETHER_MAX_VLAN_ID + 1);
2805 internals->vlan_filter_bmpmem = rte_malloc(name, vlan_filter_bmp_size,
2806 RTE_CACHE_LINE_SIZE);
2807 if (internals->vlan_filter_bmpmem == NULL) {
2809 "Failed to allocate vlan bitmap for bonded device %u\n",
2810 eth_dev->data->port_id);
2814 internals->vlan_filter_bmp = rte_bitmap_init(ETHER_MAX_VLAN_ID + 1,
2815 internals->vlan_filter_bmpmem, vlan_filter_bmp_size);
2816 if (internals->vlan_filter_bmp == NULL) {
2818 "Failed to init vlan bitmap for bonded device %u\n",
2819 eth_dev->data->port_id);
2820 rte_free(internals->vlan_filter_bmpmem);
2824 return eth_dev->data->port_id;
2827 rte_free(internals);
2828 if (eth_dev != NULL) {
2829 rte_free(eth_dev->data->mac_addrs);
2830 rte_eth_dev_release_port(eth_dev);
2836 bond_probe(struct rte_vdev_device *dev)
2839 struct bond_dev_private *internals;
2840 struct rte_kvargs *kvlist;
2841 uint8_t bonding_mode, socket_id/*, agg_mode*/;
2842 int arg_count, port_id;
2847 name = rte_vdev_device_name(dev);
2848 RTE_LOG(INFO, EAL, "Initializing pmd_bond for %s\n", name);
2850 kvlist = rte_kvargs_parse(rte_vdev_device_args(dev),
2851 pmd_bond_init_valid_arguments);
2855 /* Parse link bonding mode */
2856 if (rte_kvargs_count(kvlist, PMD_BOND_MODE_KVARG) == 1) {
2857 if (rte_kvargs_process(kvlist, PMD_BOND_MODE_KVARG,
2858 &bond_ethdev_parse_slave_mode_kvarg,
2859 &bonding_mode) != 0) {
2860 RTE_LOG(ERR, EAL, "Invalid mode for bonded device %s\n",
2865 RTE_LOG(ERR, EAL, "Mode must be specified only once for bonded "
2866 "device %s\n", name);
2870 /* Parse socket id to create bonding device on */
2871 arg_count = rte_kvargs_count(kvlist, PMD_BOND_SOCKET_ID_KVARG);
2872 if (arg_count == 1) {
2873 if (rte_kvargs_process(kvlist, PMD_BOND_SOCKET_ID_KVARG,
2874 &bond_ethdev_parse_socket_id_kvarg, &socket_id)
2876 RTE_LOG(ERR, EAL, "Invalid socket Id specified for "
2877 "bonded device %s\n", name);
2880 } else if (arg_count > 1) {
2881 RTE_LOG(ERR, EAL, "Socket Id can be specified only once for "
2882 "bonded device %s\n", name);
2885 socket_id = rte_socket_id();
2888 dev->device.numa_node = socket_id;
2890 /* Create link bonding eth device */
2891 port_id = bond_alloc(dev, bonding_mode);
2893 RTE_LOG(ERR, EAL, "Failed to create socket %s in mode %u on "
2894 "socket %u.\n", name, bonding_mode, socket_id);
2897 internals = rte_eth_devices[port_id].data->dev_private;
2898 internals->kvlist = kvlist;
2900 RTE_LOG(INFO, EAL, "Create bonded device %s on port %d in mode %u on "
2901 "socket %u.\n", name, port_id, bonding_mode, socket_id);
2905 rte_kvargs_free(kvlist);
2911 bond_remove(struct rte_vdev_device *dev)
2913 struct rte_eth_dev *eth_dev;
2914 struct bond_dev_private *internals;
2920 name = rte_vdev_device_name(dev);
2921 RTE_LOG(INFO, EAL, "Uninitializing pmd_bond for %s\n", name);
2923 /* now free all data allocation - for eth_dev structure,
2924 * dummy pci driver and internal (private) data
2927 /* find an ethdev entry */
2928 eth_dev = rte_eth_dev_allocated(name);
2929 if (eth_dev == NULL)
2932 RTE_ASSERT(eth_dev->device == &dev->device);
2934 internals = eth_dev->data->dev_private;
2935 if (internals->slave_count != 0)
2938 if (eth_dev->data->dev_started == 1) {
2939 bond_ethdev_stop(eth_dev);
2940 bond_ethdev_close(eth_dev);
2943 eth_dev->dev_ops = NULL;
2944 eth_dev->rx_pkt_burst = NULL;
2945 eth_dev->tx_pkt_burst = NULL;
2947 internals = eth_dev->data->dev_private;
2948 rte_bitmap_free(internals->vlan_filter_bmp);
2949 rte_free(internals->vlan_filter_bmpmem);
2950 rte_free(eth_dev->data->dev_private);
2951 rte_free(eth_dev->data->mac_addrs);
2953 rte_eth_dev_release_port(eth_dev);
2958 /* this part will resolve the slave portids after all the other pdev and vdev
2959 * have been allocated */
2961 bond_ethdev_configure(struct rte_eth_dev *dev)
2963 const char *name = dev->device->name;
2964 struct bond_dev_private *internals = dev->data->dev_private;
2965 struct rte_kvargs *kvlist = internals->kvlist;
2967 uint16_t port_id = dev - rte_eth_devices;
2970 static const uint8_t default_rss_key[40] = {
2971 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2, 0x41, 0x67, 0x25, 0x3D,
2972 0x43, 0xA3, 0x8F, 0xB0, 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
2973 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C, 0x6A, 0x42, 0xB7, 0x3B,
2974 0xBE, 0xAC, 0x01, 0xFA
2979 /* If RSS is enabled, fill table and key with default values */
2980 if (dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS) {
2981 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key = internals->rss_key;
2982 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len = 0;
2983 memcpy(internals->rss_key, default_rss_key, 40);
2985 for (i = 0; i < RTE_DIM(internals->reta_conf); i++) {
2986 internals->reta_conf[i].mask = ~0LL;
2987 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
2988 internals->reta_conf[i].reta[j] = j % dev->data->nb_rx_queues;
2992 /* set the max_rx_pktlen */
2993 internals->max_rx_pktlen = internals->candidate_max_rx_pktlen;
2996 * if no kvlist, it means that this bonded device has been created
2997 * through the bonding api.
3002 /* Parse MAC address for bonded device */
3003 arg_count = rte_kvargs_count(kvlist, PMD_BOND_MAC_ADDR_KVARG);
3004 if (arg_count == 1) {
3005 struct ether_addr bond_mac;
3007 if (rte_kvargs_process(kvlist, PMD_BOND_MAC_ADDR_KVARG,
3008 &bond_ethdev_parse_bond_mac_addr_kvarg, &bond_mac) < 0) {
3009 RTE_LOG(INFO, EAL, "Invalid mac address for bonded device %s\n",
3014 /* Set MAC address */
3015 if (rte_eth_bond_mac_address_set(port_id, &bond_mac) != 0) {
3017 "Failed to set mac address on bonded device %s\n",
3021 } else if (arg_count > 1) {
3023 "MAC address can be specified only once for bonded device %s\n",
3028 /* Parse/set balance mode transmit policy */
3029 arg_count = rte_kvargs_count(kvlist, PMD_BOND_XMIT_POLICY_KVARG);
3030 if (arg_count == 1) {
3031 uint8_t xmit_policy;
3033 if (rte_kvargs_process(kvlist, PMD_BOND_XMIT_POLICY_KVARG,
3034 &bond_ethdev_parse_balance_xmit_policy_kvarg, &xmit_policy) !=
3037 "Invalid xmit policy specified for bonded device %s\n",
3042 /* Set balance mode transmit policy*/
3043 if (rte_eth_bond_xmit_policy_set(port_id, xmit_policy) != 0) {
3045 "Failed to set balance xmit policy on bonded device %s\n",
3049 } else if (arg_count > 1) {
3051 "Transmit policy can be specified only once for bonded device"
3056 if (rte_kvargs_count(kvlist, PMD_BOND_AGG_MODE_KVARG) == 1) {
3057 if (rte_kvargs_process(kvlist,
3058 PMD_BOND_AGG_MODE_KVARG,
3059 &bond_ethdev_parse_slave_agg_mode_kvarg,
3062 "Failed to parse agg selection mode for bonded device %s\n",
3065 if (internals->mode == BONDING_MODE_8023AD)
3067 rte_eth_bond_8023ad_agg_selection_set(port_id,
3071 /* Parse/add slave ports to bonded device */
3072 if (rte_kvargs_count(kvlist, PMD_BOND_SLAVE_PORT_KVARG) > 0) {
3073 struct bond_ethdev_slave_ports slave_ports;
3076 memset(&slave_ports, 0, sizeof(slave_ports));
3078 if (rte_kvargs_process(kvlist, PMD_BOND_SLAVE_PORT_KVARG,
3079 &bond_ethdev_parse_slave_port_kvarg, &slave_ports) != 0) {
3081 "Failed to parse slave ports for bonded device %s\n",
3086 for (i = 0; i < slave_ports.slave_count; i++) {
3087 if (rte_eth_bond_slave_add(port_id, slave_ports.slaves[i]) != 0) {
3089 "Failed to add port %d as slave to bonded device %s\n",
3090 slave_ports.slaves[i], name);
3095 RTE_LOG(INFO, EAL, "No slaves specified for bonded device %s\n", name);
3099 /* Parse/set primary slave port id*/
3100 arg_count = rte_kvargs_count(kvlist, PMD_BOND_PRIMARY_SLAVE_KVARG);
3101 if (arg_count == 1) {
3102 uint16_t primary_slave_port_id;
3104 if (rte_kvargs_process(kvlist,
3105 PMD_BOND_PRIMARY_SLAVE_KVARG,
3106 &bond_ethdev_parse_primary_slave_port_id_kvarg,
3107 &primary_slave_port_id) < 0) {
3109 "Invalid primary slave port id specified for bonded device"
3114 /* Set balance mode transmit policy*/
3115 if (rte_eth_bond_primary_set(port_id, primary_slave_port_id)
3118 "Failed to set primary slave port %d on bonded device %s\n",
3119 primary_slave_port_id, name);
3122 } else if (arg_count > 1) {
3124 "Primary slave can be specified only once for bonded device"
3129 /* Parse link status monitor polling interval */
3130 arg_count = rte_kvargs_count(kvlist, PMD_BOND_LSC_POLL_PERIOD_KVARG);
3131 if (arg_count == 1) {
3132 uint32_t lsc_poll_interval_ms;
3134 if (rte_kvargs_process(kvlist,
3135 PMD_BOND_LSC_POLL_PERIOD_KVARG,
3136 &bond_ethdev_parse_time_ms_kvarg,
3137 &lsc_poll_interval_ms) < 0) {
3139 "Invalid lsc polling interval value specified for bonded"
3140 " device %s\n", name);
3144 if (rte_eth_bond_link_monitoring_set(port_id, lsc_poll_interval_ms)
3147 "Failed to set lsc monitor polling interval (%u ms) on"
3148 " bonded device %s\n", lsc_poll_interval_ms, name);
3151 } else if (arg_count > 1) {
3153 "LSC polling interval can be specified only once for bonded"
3154 " device %s\n", name);
3158 /* Parse link up interrupt propagation delay */
3159 arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_UP_PROP_DELAY_KVARG);
3160 if (arg_count == 1) {
3161 uint32_t link_up_delay_ms;
3163 if (rte_kvargs_process(kvlist,
3164 PMD_BOND_LINK_UP_PROP_DELAY_KVARG,
3165 &bond_ethdev_parse_time_ms_kvarg,
3166 &link_up_delay_ms) < 0) {
3168 "Invalid link up propagation delay value specified for"
3169 " bonded device %s\n", name);
3173 /* Set balance mode transmit policy*/
3174 if (rte_eth_bond_link_up_prop_delay_set(port_id, link_up_delay_ms)
3177 "Failed to set link up propagation delay (%u ms) on bonded"
3178 " device %s\n", link_up_delay_ms, name);
3181 } else if (arg_count > 1) {
3183 "Link up propagation delay can be specified only once for"
3184 " bonded device %s\n", name);
3188 /* Parse link down interrupt propagation delay */
3189 arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG);
3190 if (arg_count == 1) {
3191 uint32_t link_down_delay_ms;
3193 if (rte_kvargs_process(kvlist,
3194 PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG,
3195 &bond_ethdev_parse_time_ms_kvarg,
3196 &link_down_delay_ms) < 0) {
3198 "Invalid link down propagation delay value specified for"
3199 " bonded device %s\n", name);
3203 /* Set balance mode transmit policy*/
3204 if (rte_eth_bond_link_down_prop_delay_set(port_id, link_down_delay_ms)
3207 "Failed to set link down propagation delay (%u ms) on"
3208 " bonded device %s\n", link_down_delay_ms, name);
3211 } else if (arg_count > 1) {
3213 "Link down propagation delay can be specified only once for"
3214 " bonded device %s\n", name);
3221 struct rte_vdev_driver pmd_bond_drv = {
3222 .probe = bond_probe,
3223 .remove = bond_remove,
3226 RTE_PMD_REGISTER_VDEV(net_bonding, pmd_bond_drv);
3227 RTE_PMD_REGISTER_ALIAS(net_bonding, eth_bond);
3229 RTE_PMD_REGISTER_PARAM_STRING(net_bonding,
3233 "xmit_policy=[l2 | l23 | l34] "
3234 "agg_mode=[count | stable | bandwidth] "
3237 "lsc_poll_period_ms=<int> "
3239 "down_delay=<int>");