1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2017 Intel Corporation
5 #include <netinet/in.h>
8 #include <rte_malloc.h>
9 #include <rte_ethdev_driver.h>
10 #include <rte_ethdev_vdev.h>
14 #include <rte_ip_frag.h>
15 #include <rte_devargs.h>
16 #include <rte_kvargs.h>
17 #include <rte_bus_vdev.h>
18 #include <rte_alarm.h>
19 #include <rte_cycles.h>
20 #include <rte_string_fns.h>
22 #include "rte_eth_bond.h"
23 #include "rte_eth_bond_private.h"
24 #include "rte_eth_bond_8023ad_private.h"
26 #define REORDER_PERIOD_MS 10
27 #define DEFAULT_POLLING_INTERVAL_10_MS (10)
29 #define HASH_L4_PORTS(h) ((h)->src_port ^ (h)->dst_port)
31 /* Table for statistics in mode 5 TLB */
32 static uint64_t tlb_last_obytets[RTE_MAX_ETHPORTS];
35 get_vlan_offset(struct ether_hdr *eth_hdr, uint16_t *proto)
37 size_t vlan_offset = 0;
39 if (rte_cpu_to_be_16(ETHER_TYPE_VLAN) == *proto) {
40 struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
42 vlan_offset = sizeof(struct vlan_hdr);
43 *proto = vlan_hdr->eth_proto;
45 if (rte_cpu_to_be_16(ETHER_TYPE_VLAN) == *proto) {
46 vlan_hdr = vlan_hdr + 1;
47 *proto = vlan_hdr->eth_proto;
48 vlan_offset += sizeof(struct vlan_hdr);
55 bond_ethdev_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
57 struct bond_dev_private *internals;
59 uint16_t num_rx_slave = 0;
60 uint16_t num_rx_total = 0;
64 /* Cast to structure, containing bonded device's port id and queue id */
65 struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
67 internals = bd_rx_q->dev_private;
70 for (i = 0; i < internals->active_slave_count && nb_pkts; i++) {
71 /* Offset of pointer to *bufs increases as packets are received
72 * from other slaves */
73 num_rx_slave = rte_eth_rx_burst(internals->active_slaves[i],
74 bd_rx_q->queue_id, bufs + num_rx_total, nb_pkts);
76 num_rx_total += num_rx_slave;
77 nb_pkts -= num_rx_slave;
85 bond_ethdev_rx_burst_active_backup(void *queue, struct rte_mbuf **bufs,
88 struct bond_dev_private *internals;
90 /* Cast to structure, containing bonded device's port id and queue id */
91 struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
93 internals = bd_rx_q->dev_private;
95 return rte_eth_rx_burst(internals->current_primary_port,
96 bd_rx_q->queue_id, bufs, nb_pkts);
100 is_lacp_packets(uint16_t ethertype, uint8_t subtype, struct rte_mbuf *mbuf)
102 const uint16_t ether_type_slow_be = rte_be_to_cpu_16(ETHER_TYPE_SLOW);
104 return !((mbuf->ol_flags & PKT_RX_VLAN) ? mbuf->vlan_tci : 0) &&
105 (ethertype == ether_type_slow_be &&
106 (subtype == SLOW_SUBTYPE_MARKER || subtype == SLOW_SUBTYPE_LACP));
109 /*****************************************************************************
110 * Flow director's setup for mode 4 optimization
113 static struct rte_flow_item_eth flow_item_eth_type_8023ad = {
114 .dst.addr_bytes = { 0 },
115 .src.addr_bytes = { 0 },
116 .type = RTE_BE16(ETHER_TYPE_SLOW),
119 static struct rte_flow_item_eth flow_item_eth_mask_type_8023ad = {
120 .dst.addr_bytes = { 0 },
121 .src.addr_bytes = { 0 },
125 static struct rte_flow_item flow_item_8023ad[] = {
127 .type = RTE_FLOW_ITEM_TYPE_ETH,
128 .spec = &flow_item_eth_type_8023ad,
130 .mask = &flow_item_eth_mask_type_8023ad,
133 .type = RTE_FLOW_ITEM_TYPE_END,
140 const struct rte_flow_attr flow_attr_8023ad = {
149 bond_ethdev_8023ad_flow_verify(struct rte_eth_dev *bond_dev,
150 uint16_t slave_port) {
151 struct rte_eth_dev_info slave_info;
152 struct rte_flow_error error;
153 struct bond_dev_private *internals = (struct bond_dev_private *)
154 (bond_dev->data->dev_private);
156 const struct rte_flow_action_queue lacp_queue_conf = {
160 const struct rte_flow_action actions[] = {
162 .type = RTE_FLOW_ACTION_TYPE_QUEUE,
163 .conf = &lacp_queue_conf
166 .type = RTE_FLOW_ACTION_TYPE_END,
170 int ret = rte_flow_validate(slave_port, &flow_attr_8023ad,
171 flow_item_8023ad, actions, &error);
173 RTE_BOND_LOG(ERR, "%s: %s (slave_port=%d queue_id=%d)",
174 __func__, error.message, slave_port,
175 internals->mode4.dedicated_queues.rx_qid);
179 rte_eth_dev_info_get(slave_port, &slave_info);
180 if (slave_info.max_rx_queues < bond_dev->data->nb_rx_queues ||
181 slave_info.max_tx_queues < bond_dev->data->nb_tx_queues) {
183 "%s: Slave %d capabilities doesn't allow to allocate additional queues",
184 __func__, slave_port);
192 bond_8023ad_slow_pkt_hw_filter_supported(uint16_t port_id) {
193 struct rte_eth_dev *bond_dev = &rte_eth_devices[port_id];
194 struct bond_dev_private *internals = (struct bond_dev_private *)
195 (bond_dev->data->dev_private);
196 struct rte_eth_dev_info bond_info;
199 /* Verify if all slaves in bonding supports flow director and */
200 if (internals->slave_count > 0) {
201 rte_eth_dev_info_get(bond_dev->data->port_id, &bond_info);
203 internals->mode4.dedicated_queues.rx_qid = bond_info.nb_rx_queues;
204 internals->mode4.dedicated_queues.tx_qid = bond_info.nb_tx_queues;
206 for (idx = 0; idx < internals->slave_count; idx++) {
207 if (bond_ethdev_8023ad_flow_verify(bond_dev,
208 internals->slaves[idx].port_id) != 0)
217 bond_ethdev_8023ad_flow_set(struct rte_eth_dev *bond_dev, uint16_t slave_port) {
219 struct rte_flow_error error;
220 struct bond_dev_private *internals = (struct bond_dev_private *)
221 (bond_dev->data->dev_private);
223 struct rte_flow_action_queue lacp_queue_conf = {
224 .index = internals->mode4.dedicated_queues.rx_qid,
227 const struct rte_flow_action actions[] = {
229 .type = RTE_FLOW_ACTION_TYPE_QUEUE,
230 .conf = &lacp_queue_conf
233 .type = RTE_FLOW_ACTION_TYPE_END,
237 internals->mode4.dedicated_queues.flow[slave_port] = rte_flow_create(slave_port,
238 &flow_attr_8023ad, flow_item_8023ad, actions, &error);
239 if (internals->mode4.dedicated_queues.flow[slave_port] == NULL) {
240 RTE_BOND_LOG(ERR, "bond_ethdev_8023ad_flow_set: %s "
241 "(slave_port=%d queue_id=%d)",
242 error.message, slave_port,
243 internals->mode4.dedicated_queues.rx_qid);
251 bond_ethdev_rx_burst_8023ad_fast_queue(void *queue, struct rte_mbuf **bufs,
254 struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
255 struct bond_dev_private *internals = bd_rx_q->dev_private;
256 uint16_t num_rx_total = 0; /* Total number of received packets */
257 uint16_t slaves[RTE_MAX_ETHPORTS];
258 uint16_t slave_count;
262 /* Copy slave list to protect against slave up/down changes during tx
264 slave_count = internals->active_slave_count;
265 memcpy(slaves, internals->active_slaves,
266 sizeof(internals->active_slaves[0]) * slave_count);
268 for (i = 0, idx = internals->active_slave;
269 i < slave_count && num_rx_total < nb_pkts; i++, idx++) {
270 idx = idx % slave_count;
272 /* Read packets from this slave */
273 num_rx_total += rte_eth_rx_burst(slaves[idx], bd_rx_q->queue_id,
274 &bufs[num_rx_total], nb_pkts - num_rx_total);
277 internals->active_slave = idx;
283 bond_ethdev_tx_burst_8023ad_fast_queue(void *queue, struct rte_mbuf **bufs,
286 struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
287 struct bond_dev_private *internals = bd_tx_q->dev_private;
289 uint16_t slave_port_ids[RTE_MAX_ETHPORTS];
290 uint16_t slave_count;
292 uint16_t dist_slave_port_ids[RTE_MAX_ETHPORTS];
293 uint16_t dist_slave_count;
295 /* 2-D array to sort mbufs for transmission on each slave into */
296 struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_bufs];
297 /* Number of mbufs for transmission on each slave */
298 uint16_t slave_nb_bufs[RTE_MAX_ETHPORTS] = { 0 };
299 /* Mapping array generated by hash function to map mbufs to slaves */
300 uint16_t bufs_slave_port_idxs[RTE_MAX_ETHPORTS] = { 0 };
302 uint16_t slave_tx_count, slave_tx_fail_count[RTE_MAX_ETHPORTS] = { 0 };
303 uint16_t total_tx_count = 0, total_tx_fail_count = 0;
307 if (unlikely(nb_bufs == 0))
310 /* Copy slave list to protect against slave up/down changes during tx
312 slave_count = internals->active_slave_count;
313 if (unlikely(slave_count < 1))
316 memcpy(slave_port_ids, internals->active_slaves,
317 sizeof(slave_port_ids[0]) * slave_count);
320 dist_slave_count = 0;
321 for (i = 0; i < slave_count; i++) {
322 struct port *port = &mode_8023ad_ports[slave_port_ids[i]];
324 if (ACTOR_STATE(port, DISTRIBUTING))
325 dist_slave_port_ids[dist_slave_count++] =
329 if (unlikely(dist_slave_count < 1))
333 * Populate slaves mbuf with the packets which are to be sent on it
334 * selecting output slave using hash based on xmit policy
336 internals->burst_xmit_hash(bufs, nb_bufs, dist_slave_count,
337 bufs_slave_port_idxs);
339 for (i = 0; i < nb_bufs; i++) {
340 /* Populate slave mbuf arrays with mbufs for that slave. */
341 uint8_t slave_idx = bufs_slave_port_idxs[i];
343 slave_bufs[slave_idx][slave_nb_bufs[slave_idx]++] = bufs[i];
347 /* Send packet burst on each slave device */
348 for (i = 0; i < dist_slave_count; i++) {
349 if (slave_nb_bufs[i] == 0)
352 slave_tx_count = rte_eth_tx_burst(dist_slave_port_ids[i],
353 bd_tx_q->queue_id, slave_bufs[i],
356 total_tx_count += slave_tx_count;
358 /* If tx burst fails move packets to end of bufs */
359 if (unlikely(slave_tx_count < slave_nb_bufs[i])) {
360 slave_tx_fail_count[i] = slave_nb_bufs[i] -
362 total_tx_fail_count += slave_tx_fail_count[i];
365 * Shift bufs to beginning of array to allow reordering
368 for (j = 0; j < slave_tx_fail_count[i]; j++) {
370 slave_bufs[i][(slave_tx_count - 1) + j];
376 * If there are tx burst failures we move packets to end of bufs to
377 * preserve expected PMD behaviour of all failed transmitted being
378 * at the end of the input mbuf array
380 if (unlikely(total_tx_fail_count > 0)) {
381 int bufs_idx = nb_bufs - total_tx_fail_count - 1;
383 for (i = 0; i < slave_count; i++) {
384 if (slave_tx_fail_count[i] > 0) {
385 for (j = 0; j < slave_tx_fail_count[i]; j++)
386 bufs[bufs_idx++] = slave_bufs[i][j];
391 return total_tx_count;
396 bond_ethdev_rx_burst_8023ad(void *queue, struct rte_mbuf **bufs,
399 /* Cast to structure, containing bonded device's port id and queue id */
400 struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
401 struct bond_dev_private *internals = bd_rx_q->dev_private;
402 struct ether_addr bond_mac;
404 struct ether_hdr *hdr;
406 const uint16_t ether_type_slow_be = rte_be_to_cpu_16(ETHER_TYPE_SLOW);
407 uint16_t num_rx_total = 0; /* Total number of received packets */
408 uint16_t slaves[RTE_MAX_ETHPORTS];
409 uint16_t slave_count, idx;
411 uint8_t collecting; /* current slave collecting status */
412 const uint8_t promisc = internals->promiscuous_en;
416 rte_eth_macaddr_get(internals->port_id, &bond_mac);
417 /* Copy slave list to protect against slave up/down changes during tx
419 slave_count = internals->active_slave_count;
420 memcpy(slaves, internals->active_slaves,
421 sizeof(internals->active_slaves[0]) * slave_count);
423 idx = internals->active_slave;
424 if (idx >= slave_count) {
425 internals->active_slave = 0;
428 for (i = 0; i < slave_count && num_rx_total < nb_pkts; i++) {
430 collecting = ACTOR_STATE(&mode_8023ad_ports[slaves[idx]],
433 /* Read packets from this slave */
434 num_rx_total += rte_eth_rx_burst(slaves[idx], bd_rx_q->queue_id,
435 &bufs[num_rx_total], nb_pkts - num_rx_total);
437 for (k = j; k < 2 && k < num_rx_total; k++)
438 rte_prefetch0(rte_pktmbuf_mtod(bufs[k], void *));
440 /* Handle slow protocol packets. */
441 while (j < num_rx_total) {
443 /* If packet is not pure L2 and is known, skip it */
444 if ((bufs[j]->packet_type & ~RTE_PTYPE_L2_ETHER) != 0) {
449 if (j + 3 < num_rx_total)
450 rte_prefetch0(rte_pktmbuf_mtod(bufs[j + 3], void *));
452 hdr = rte_pktmbuf_mtod(bufs[j], struct ether_hdr *);
453 subtype = ((struct slow_protocol_frame *)hdr)->slow_protocol.subtype;
455 /* Remove packet from array if it is slow packet or slave is not
456 * in collecting state or bonding interface is not in promiscuous
457 * mode and packet address does not match. */
458 if (unlikely(is_lacp_packets(hdr->ether_type, subtype, bufs[j]) ||
459 !collecting || (!promisc &&
460 !is_multicast_ether_addr(&hdr->d_addr) &&
461 !is_same_ether_addr(&bond_mac, &hdr->d_addr)))) {
463 if (hdr->ether_type == ether_type_slow_be) {
464 bond_mode_8023ad_handle_slow_pkt(
465 internals, slaves[idx], bufs[j]);
467 rte_pktmbuf_free(bufs[j]);
469 /* Packet is managed by mode 4 or dropped, shift the array */
471 if (j < num_rx_total) {
472 memmove(&bufs[j], &bufs[j + 1], sizeof(bufs[0]) *
478 if (unlikely(++idx == slave_count))
482 internals->active_slave = idx;
486 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
487 uint32_t burstnumberRX;
488 uint32_t burstnumberTX;
490 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
493 arp_op_name(uint16_t arp_op, char *buf)
497 snprintf(buf, sizeof("ARP Request"), "%s", "ARP Request");
500 snprintf(buf, sizeof("ARP Reply"), "%s", "ARP Reply");
502 case ARP_OP_REVREQUEST:
503 snprintf(buf, sizeof("Reverse ARP Request"), "%s",
504 "Reverse ARP Request");
506 case ARP_OP_REVREPLY:
507 snprintf(buf, sizeof("Reverse ARP Reply"), "%s",
508 "Reverse ARP Reply");
510 case ARP_OP_INVREQUEST:
511 snprintf(buf, sizeof("Peer Identify Request"), "%s",
512 "Peer Identify Request");
514 case ARP_OP_INVREPLY:
515 snprintf(buf, sizeof("Peer Identify Reply"), "%s",
516 "Peer Identify Reply");
521 snprintf(buf, sizeof("Unknown"), "%s", "Unknown");
525 #define MaxIPv4String 16
527 ipv4_addr_to_dot(uint32_t be_ipv4_addr, char *buf, uint8_t buf_size)
531 ipv4_addr = rte_be_to_cpu_32(be_ipv4_addr);
532 snprintf(buf, buf_size, "%d.%d.%d.%d", (ipv4_addr >> 24) & 0xFF,
533 (ipv4_addr >> 16) & 0xFF, (ipv4_addr >> 8) & 0xFF,
537 #define MAX_CLIENTS_NUMBER 128
538 uint8_t active_clients;
539 struct client_stats_t {
542 uint32_t ipv4_rx_packets;
543 uint32_t ipv4_tx_packets;
545 struct client_stats_t client_stats[MAX_CLIENTS_NUMBER];
548 update_client_stats(uint32_t addr, uint16_t port, uint32_t *TXorRXindicator)
552 for (; i < MAX_CLIENTS_NUMBER; i++) {
553 if ((client_stats[i].ipv4_addr == addr) && (client_stats[i].port == port)) {
554 /* Just update RX packets number for this client */
555 if (TXorRXindicator == &burstnumberRX)
556 client_stats[i].ipv4_rx_packets++;
558 client_stats[i].ipv4_tx_packets++;
562 /* We have a new client. Insert him to the table, and increment stats */
563 if (TXorRXindicator == &burstnumberRX)
564 client_stats[active_clients].ipv4_rx_packets++;
566 client_stats[active_clients].ipv4_tx_packets++;
567 client_stats[active_clients].ipv4_addr = addr;
568 client_stats[active_clients].port = port;
573 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
574 #define MODE6_DEBUG(info, src_ip, dst_ip, eth_h, arp_op, port, burstnumber) \
575 RTE_LOG(DEBUG, PMD, \
578 "SrcMAC:%02X:%02X:%02X:%02X:%02X:%02X " \
580 "DstMAC:%02X:%02X:%02X:%02X:%02X:%02X " \
586 eth_h->s_addr.addr_bytes[0], \
587 eth_h->s_addr.addr_bytes[1], \
588 eth_h->s_addr.addr_bytes[2], \
589 eth_h->s_addr.addr_bytes[3], \
590 eth_h->s_addr.addr_bytes[4], \
591 eth_h->s_addr.addr_bytes[5], \
593 eth_h->d_addr.addr_bytes[0], \
594 eth_h->d_addr.addr_bytes[1], \
595 eth_h->d_addr.addr_bytes[2], \
596 eth_h->d_addr.addr_bytes[3], \
597 eth_h->d_addr.addr_bytes[4], \
598 eth_h->d_addr.addr_bytes[5], \
605 mode6_debug(const char __attribute__((unused)) *info, struct ether_hdr *eth_h,
606 uint16_t port, uint32_t __attribute__((unused)) *burstnumber)
608 struct ipv4_hdr *ipv4_h;
609 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
610 struct arp_hdr *arp_h;
617 uint16_t ether_type = eth_h->ether_type;
618 uint16_t offset = get_vlan_offset(eth_h, ðer_type);
620 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
621 strlcpy(buf, info, 16);
624 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4)) {
625 ipv4_h = (struct ipv4_hdr *)((char *)(eth_h + 1) + offset);
626 ipv4_addr_to_dot(ipv4_h->src_addr, src_ip, MaxIPv4String);
627 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
628 ipv4_addr_to_dot(ipv4_h->dst_addr, dst_ip, MaxIPv4String);
629 MODE6_DEBUG(buf, src_ip, dst_ip, eth_h, "", port, *burstnumber);
631 update_client_stats(ipv4_h->src_addr, port, burstnumber);
633 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
634 else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_ARP)) {
635 arp_h = (struct arp_hdr *)((char *)(eth_h + 1) + offset);
636 ipv4_addr_to_dot(arp_h->arp_data.arp_sip, src_ip, MaxIPv4String);
637 ipv4_addr_to_dot(arp_h->arp_data.arp_tip, dst_ip, MaxIPv4String);
638 arp_op_name(rte_be_to_cpu_16(arp_h->arp_op), ArpOp);
639 MODE6_DEBUG(buf, src_ip, dst_ip, eth_h, ArpOp, port, *burstnumber);
646 bond_ethdev_rx_burst_alb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
648 struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
649 struct bond_dev_private *internals = bd_tx_q->dev_private;
650 struct ether_hdr *eth_h;
651 uint16_t ether_type, offset;
652 uint16_t nb_recv_pkts;
655 nb_recv_pkts = bond_ethdev_rx_burst(queue, bufs, nb_pkts);
657 for (i = 0; i < nb_recv_pkts; i++) {
658 eth_h = rte_pktmbuf_mtod(bufs[i], struct ether_hdr *);
659 ether_type = eth_h->ether_type;
660 offset = get_vlan_offset(eth_h, ðer_type);
662 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_ARP)) {
663 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
664 mode6_debug("RX ARP:", eth_h, bufs[i]->port, &burstnumberRX);
666 bond_mode_alb_arp_recv(eth_h, offset, internals);
668 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
669 else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
670 mode6_debug("RX IPv4:", eth_h, bufs[i]->port, &burstnumberRX);
678 bond_ethdev_tx_burst_round_robin(void *queue, struct rte_mbuf **bufs,
681 struct bond_dev_private *internals;
682 struct bond_tx_queue *bd_tx_q;
684 struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_pkts];
685 uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
687 uint16_t num_of_slaves;
688 uint16_t slaves[RTE_MAX_ETHPORTS];
690 uint16_t num_tx_total = 0, num_tx_slave;
692 static int slave_idx = 0;
693 int i, cslave_idx = 0, tx_fail_total = 0;
695 bd_tx_q = (struct bond_tx_queue *)queue;
696 internals = bd_tx_q->dev_private;
698 /* Copy slave list to protect against slave up/down changes during tx
700 num_of_slaves = internals->active_slave_count;
701 memcpy(slaves, internals->active_slaves,
702 sizeof(internals->active_slaves[0]) * num_of_slaves);
704 if (num_of_slaves < 1)
707 /* Populate slaves mbuf with which packets are to be sent on it */
708 for (i = 0; i < nb_pkts; i++) {
709 cslave_idx = (slave_idx + i) % num_of_slaves;
710 slave_bufs[cslave_idx][(slave_nb_pkts[cslave_idx])++] = bufs[i];
713 /* increment current slave index so the next call to tx burst starts on the
715 slave_idx = ++cslave_idx;
717 /* Send packet burst on each slave device */
718 for (i = 0; i < num_of_slaves; i++) {
719 if (slave_nb_pkts[i] > 0) {
720 num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
721 slave_bufs[i], slave_nb_pkts[i]);
723 /* if tx burst fails move packets to end of bufs */
724 if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
725 int tx_fail_slave = slave_nb_pkts[i] - num_tx_slave;
727 tx_fail_total += tx_fail_slave;
729 memcpy(&bufs[nb_pkts - tx_fail_total],
730 &slave_bufs[i][num_tx_slave],
731 tx_fail_slave * sizeof(bufs[0]));
733 num_tx_total += num_tx_slave;
741 bond_ethdev_tx_burst_active_backup(void *queue,
742 struct rte_mbuf **bufs, uint16_t nb_pkts)
744 struct bond_dev_private *internals;
745 struct bond_tx_queue *bd_tx_q;
747 bd_tx_q = (struct bond_tx_queue *)queue;
748 internals = bd_tx_q->dev_private;
750 if (internals->active_slave_count < 1)
753 return rte_eth_tx_burst(internals->current_primary_port, bd_tx_q->queue_id,
757 static inline uint16_t
758 ether_hash(struct ether_hdr *eth_hdr)
760 unaligned_uint16_t *word_src_addr =
761 (unaligned_uint16_t *)eth_hdr->s_addr.addr_bytes;
762 unaligned_uint16_t *word_dst_addr =
763 (unaligned_uint16_t *)eth_hdr->d_addr.addr_bytes;
765 return (word_src_addr[0] ^ word_dst_addr[0]) ^
766 (word_src_addr[1] ^ word_dst_addr[1]) ^
767 (word_src_addr[2] ^ word_dst_addr[2]);
770 static inline uint32_t
771 ipv4_hash(struct ipv4_hdr *ipv4_hdr)
773 return ipv4_hdr->src_addr ^ ipv4_hdr->dst_addr;
776 static inline uint32_t
777 ipv6_hash(struct ipv6_hdr *ipv6_hdr)
779 unaligned_uint32_t *word_src_addr =
780 (unaligned_uint32_t *)&(ipv6_hdr->src_addr[0]);
781 unaligned_uint32_t *word_dst_addr =
782 (unaligned_uint32_t *)&(ipv6_hdr->dst_addr[0]);
784 return (word_src_addr[0] ^ word_dst_addr[0]) ^
785 (word_src_addr[1] ^ word_dst_addr[1]) ^
786 (word_src_addr[2] ^ word_dst_addr[2]) ^
787 (word_src_addr[3] ^ word_dst_addr[3]);
792 burst_xmit_l2_hash(struct rte_mbuf **buf, uint16_t nb_pkts,
793 uint8_t slave_count, uint16_t *slaves)
795 struct ether_hdr *eth_hdr;
799 for (i = 0; i < nb_pkts; i++) {
800 eth_hdr = rte_pktmbuf_mtod(buf[i], struct ether_hdr *);
802 hash = ether_hash(eth_hdr);
804 slaves[i] = (hash ^= hash >> 8) % slave_count;
809 burst_xmit_l23_hash(struct rte_mbuf **buf, uint16_t nb_pkts,
810 uint8_t slave_count, uint16_t *slaves)
813 struct ether_hdr *eth_hdr;
816 uint32_t hash, l3hash;
818 for (i = 0; i < nb_pkts; i++) {
819 eth_hdr = rte_pktmbuf_mtod(buf[i], struct ether_hdr *);
822 proto = eth_hdr->ether_type;
823 hash = ether_hash(eth_hdr);
825 vlan_offset = get_vlan_offset(eth_hdr, &proto);
827 if (rte_cpu_to_be_16(ETHER_TYPE_IPv4) == proto) {
828 struct ipv4_hdr *ipv4_hdr = (struct ipv4_hdr *)
829 ((char *)(eth_hdr + 1) + vlan_offset);
830 l3hash = ipv4_hash(ipv4_hdr);
832 } else if (rte_cpu_to_be_16(ETHER_TYPE_IPv6) == proto) {
833 struct ipv6_hdr *ipv6_hdr = (struct ipv6_hdr *)
834 ((char *)(eth_hdr + 1) + vlan_offset);
835 l3hash = ipv6_hash(ipv6_hdr);
838 hash = hash ^ l3hash;
842 slaves[i] = hash % slave_count;
847 burst_xmit_l34_hash(struct rte_mbuf **buf, uint16_t nb_pkts,
848 uint8_t slave_count, uint16_t *slaves)
850 struct ether_hdr *eth_hdr;
855 struct udp_hdr *udp_hdr;
856 struct tcp_hdr *tcp_hdr;
857 uint32_t hash, l3hash, l4hash;
859 for (i = 0; i < nb_pkts; i++) {
860 eth_hdr = rte_pktmbuf_mtod(buf[i], struct ether_hdr *);
861 proto = eth_hdr->ether_type;
862 vlan_offset = get_vlan_offset(eth_hdr, &proto);
866 if (rte_cpu_to_be_16(ETHER_TYPE_IPv4) == proto) {
867 struct ipv4_hdr *ipv4_hdr = (struct ipv4_hdr *)
868 ((char *)(eth_hdr + 1) + vlan_offset);
869 size_t ip_hdr_offset;
871 l3hash = ipv4_hash(ipv4_hdr);
873 /* there is no L4 header in fragmented packet */
874 if (likely(rte_ipv4_frag_pkt_is_fragmented(ipv4_hdr)
876 ip_hdr_offset = (ipv4_hdr->version_ihl
877 & IPV4_HDR_IHL_MASK) *
880 if (ipv4_hdr->next_proto_id == IPPROTO_TCP) {
881 tcp_hdr = (struct tcp_hdr *)
884 l4hash = HASH_L4_PORTS(tcp_hdr);
885 } else if (ipv4_hdr->next_proto_id ==
887 udp_hdr = (struct udp_hdr *)
890 l4hash = HASH_L4_PORTS(udp_hdr);
893 } else if (rte_cpu_to_be_16(ETHER_TYPE_IPv6) == proto) {
894 struct ipv6_hdr *ipv6_hdr = (struct ipv6_hdr *)
895 ((char *)(eth_hdr + 1) + vlan_offset);
896 l3hash = ipv6_hash(ipv6_hdr);
898 if (ipv6_hdr->proto == IPPROTO_TCP) {
899 tcp_hdr = (struct tcp_hdr *)(ipv6_hdr + 1);
900 l4hash = HASH_L4_PORTS(tcp_hdr);
901 } else if (ipv6_hdr->proto == IPPROTO_UDP) {
902 udp_hdr = (struct udp_hdr *)(ipv6_hdr + 1);
903 l4hash = HASH_L4_PORTS(udp_hdr);
907 hash = l3hash ^ l4hash;
911 slaves[i] = hash % slave_count;
916 uint64_t bwg_left_int;
917 uint64_t bwg_left_remainder;
922 bond_tlb_activate_slave(struct bond_dev_private *internals) {
925 for (i = 0; i < internals->active_slave_count; i++) {
926 tlb_last_obytets[internals->active_slaves[i]] = 0;
931 bandwidth_cmp(const void *a, const void *b)
933 const struct bwg_slave *bwg_a = a;
934 const struct bwg_slave *bwg_b = b;
935 int64_t diff = (int64_t)bwg_b->bwg_left_int - (int64_t)bwg_a->bwg_left_int;
936 int64_t diff2 = (int64_t)bwg_b->bwg_left_remainder -
937 (int64_t)bwg_a->bwg_left_remainder;
951 bandwidth_left(uint16_t port_id, uint64_t load, uint8_t update_idx,
952 struct bwg_slave *bwg_slave)
954 struct rte_eth_link link_status;
956 rte_eth_link_get_nowait(port_id, &link_status);
957 uint64_t link_bwg = link_status.link_speed * 1000000ULL / 8;
960 link_bwg = link_bwg * (update_idx+1) * REORDER_PERIOD_MS;
961 bwg_slave->bwg_left_int = (link_bwg - 1000*load) / link_bwg;
962 bwg_slave->bwg_left_remainder = (link_bwg - 1000*load) % link_bwg;
966 bond_ethdev_update_tlb_slave_cb(void *arg)
968 struct bond_dev_private *internals = arg;
969 struct rte_eth_stats slave_stats;
970 struct bwg_slave bwg_array[RTE_MAX_ETHPORTS];
974 uint8_t update_stats = 0;
977 internals->slave_update_idx++;
980 if (internals->slave_update_idx >= REORDER_PERIOD_MS)
983 for (i = 0; i < internals->active_slave_count; i++) {
984 slave_id = internals->active_slaves[i];
985 rte_eth_stats_get(slave_id, &slave_stats);
986 tx_bytes = slave_stats.obytes - tlb_last_obytets[slave_id];
987 bandwidth_left(slave_id, tx_bytes,
988 internals->slave_update_idx, &bwg_array[i]);
989 bwg_array[i].slave = slave_id;
992 tlb_last_obytets[slave_id] = slave_stats.obytes;
996 if (update_stats == 1)
997 internals->slave_update_idx = 0;
1000 qsort(bwg_array, slave_count, sizeof(bwg_array[0]), bandwidth_cmp);
1001 for (i = 0; i < slave_count; i++)
1002 internals->tlb_slaves_order[i] = bwg_array[i].slave;
1004 rte_eal_alarm_set(REORDER_PERIOD_MS * 1000, bond_ethdev_update_tlb_slave_cb,
1005 (struct bond_dev_private *)internals);
1009 bond_ethdev_tx_burst_tlb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
1011 struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
1012 struct bond_dev_private *internals = bd_tx_q->dev_private;
1014 struct rte_eth_dev *primary_port =
1015 &rte_eth_devices[internals->primary_port];
1016 uint16_t num_tx_total = 0;
1019 uint16_t num_of_slaves = internals->active_slave_count;
1020 uint16_t slaves[RTE_MAX_ETHPORTS];
1022 struct ether_hdr *ether_hdr;
1023 struct ether_addr primary_slave_addr;
1024 struct ether_addr active_slave_addr;
1026 if (num_of_slaves < 1)
1027 return num_tx_total;
1029 memcpy(slaves, internals->tlb_slaves_order,
1030 sizeof(internals->tlb_slaves_order[0]) * num_of_slaves);
1033 ether_addr_copy(primary_port->data->mac_addrs, &primary_slave_addr);
1036 for (i = 0; i < 3; i++)
1037 rte_prefetch0(rte_pktmbuf_mtod(bufs[i], void*));
1040 for (i = 0; i < num_of_slaves; i++) {
1041 rte_eth_macaddr_get(slaves[i], &active_slave_addr);
1042 for (j = num_tx_total; j < nb_pkts; j++) {
1043 if (j + 3 < nb_pkts)
1044 rte_prefetch0(rte_pktmbuf_mtod(bufs[j+3], void*));
1046 ether_hdr = rte_pktmbuf_mtod(bufs[j], struct ether_hdr *);
1047 if (is_same_ether_addr(ðer_hdr->s_addr, &primary_slave_addr))
1048 ether_addr_copy(&active_slave_addr, ðer_hdr->s_addr);
1049 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
1050 mode6_debug("TX IPv4:", ether_hdr, slaves[i], &burstnumberTX);
1054 num_tx_total += rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
1055 bufs + num_tx_total, nb_pkts - num_tx_total);
1057 if (num_tx_total == nb_pkts)
1061 return num_tx_total;
1065 bond_tlb_disable(struct bond_dev_private *internals)
1067 rte_eal_alarm_cancel(bond_ethdev_update_tlb_slave_cb, internals);
1071 bond_tlb_enable(struct bond_dev_private *internals)
1073 bond_ethdev_update_tlb_slave_cb(internals);
1077 bond_ethdev_tx_burst_alb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
1079 struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
1080 struct bond_dev_private *internals = bd_tx_q->dev_private;
1082 struct ether_hdr *eth_h;
1083 uint16_t ether_type, offset;
1085 struct client_data *client_info;
1088 * We create transmit buffers for every slave and one additional to send
1089 * through tlb. In worst case every packet will be send on one port.
1091 struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS + 1][nb_pkts];
1092 uint16_t slave_bufs_pkts[RTE_MAX_ETHPORTS + 1] = { 0 };
1095 * We create separate transmit buffers for update packets as they won't
1096 * be counted in num_tx_total.
1098 struct rte_mbuf *update_bufs[RTE_MAX_ETHPORTS][ALB_HASH_TABLE_SIZE];
1099 uint16_t update_bufs_pkts[RTE_MAX_ETHPORTS] = { 0 };
1101 struct rte_mbuf *upd_pkt;
1104 uint16_t num_send, num_not_send = 0;
1105 uint16_t num_tx_total = 0;
1110 /* Search tx buffer for ARP packets and forward them to alb */
1111 for (i = 0; i < nb_pkts; i++) {
1112 eth_h = rte_pktmbuf_mtod(bufs[i], struct ether_hdr *);
1113 ether_type = eth_h->ether_type;
1114 offset = get_vlan_offset(eth_h, ðer_type);
1116 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_ARP)) {
1117 slave_idx = bond_mode_alb_arp_xmit(eth_h, offset, internals);
1119 /* Change src mac in eth header */
1120 rte_eth_macaddr_get(slave_idx, ð_h->s_addr);
1122 /* Add packet to slave tx buffer */
1123 slave_bufs[slave_idx][slave_bufs_pkts[slave_idx]] = bufs[i];
1124 slave_bufs_pkts[slave_idx]++;
1126 /* If packet is not ARP, send it with TLB policy */
1127 slave_bufs[RTE_MAX_ETHPORTS][slave_bufs_pkts[RTE_MAX_ETHPORTS]] =
1129 slave_bufs_pkts[RTE_MAX_ETHPORTS]++;
1133 /* Update connected client ARP tables */
1134 if (internals->mode6.ntt) {
1135 for (i = 0; i < ALB_HASH_TABLE_SIZE; i++) {
1136 client_info = &internals->mode6.client_table[i];
1138 if (client_info->in_use) {
1139 /* Allocate new packet to send ARP update on current slave */
1140 upd_pkt = rte_pktmbuf_alloc(internals->mode6.mempool);
1141 if (upd_pkt == NULL) {
1142 RTE_LOG(ERR, PMD, "Failed to allocate ARP packet from pool\n");
1145 pkt_size = sizeof(struct ether_hdr) + sizeof(struct arp_hdr)
1146 + client_info->vlan_count * sizeof(struct vlan_hdr);
1147 upd_pkt->data_len = pkt_size;
1148 upd_pkt->pkt_len = pkt_size;
1150 slave_idx = bond_mode_alb_arp_upd(client_info, upd_pkt,
1153 /* Add packet to update tx buffer */
1154 update_bufs[slave_idx][update_bufs_pkts[slave_idx]] = upd_pkt;
1155 update_bufs_pkts[slave_idx]++;
1158 internals->mode6.ntt = 0;
1161 /* Send ARP packets on proper slaves */
1162 for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
1163 if (slave_bufs_pkts[i] > 0) {
1164 num_send = rte_eth_tx_burst(i, bd_tx_q->queue_id,
1165 slave_bufs[i], slave_bufs_pkts[i]);
1166 for (j = 0; j < slave_bufs_pkts[i] - num_send; j++) {
1167 bufs[nb_pkts - 1 - num_not_send - j] =
1168 slave_bufs[i][nb_pkts - 1 - j];
1171 num_tx_total += num_send;
1172 num_not_send += slave_bufs_pkts[i] - num_send;
1174 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
1175 /* Print TX stats including update packets */
1176 for (j = 0; j < slave_bufs_pkts[i]; j++) {
1177 eth_h = rte_pktmbuf_mtod(slave_bufs[i][j], struct ether_hdr *);
1178 mode6_debug("TX ARP:", eth_h, i, &burstnumberTX);
1184 /* Send update packets on proper slaves */
1185 for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
1186 if (update_bufs_pkts[i] > 0) {
1187 num_send = rte_eth_tx_burst(i, bd_tx_q->queue_id, update_bufs[i],
1188 update_bufs_pkts[i]);
1189 for (j = num_send; j < update_bufs_pkts[i]; j++) {
1190 rte_pktmbuf_free(update_bufs[i][j]);
1192 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
1193 for (j = 0; j < update_bufs_pkts[i]; j++) {
1194 eth_h = rte_pktmbuf_mtod(update_bufs[i][j], struct ether_hdr *);
1195 mode6_debug("TX ARPupd:", eth_h, i, &burstnumberTX);
1201 /* Send non-ARP packets using tlb policy */
1202 if (slave_bufs_pkts[RTE_MAX_ETHPORTS] > 0) {
1203 num_send = bond_ethdev_tx_burst_tlb(queue,
1204 slave_bufs[RTE_MAX_ETHPORTS],
1205 slave_bufs_pkts[RTE_MAX_ETHPORTS]);
1207 for (j = 0; j < slave_bufs_pkts[RTE_MAX_ETHPORTS]; j++) {
1208 bufs[nb_pkts - 1 - num_not_send - j] =
1209 slave_bufs[RTE_MAX_ETHPORTS][nb_pkts - 1 - j];
1212 num_tx_total += num_send;
1215 return num_tx_total;
1219 bond_ethdev_tx_burst_balance(void *queue, struct rte_mbuf **bufs,
1222 struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
1223 struct bond_dev_private *internals = bd_tx_q->dev_private;
1225 uint16_t slave_port_ids[RTE_MAX_ETHPORTS];
1226 uint16_t slave_count;
1228 /* Array to sort mbufs for transmission on each slave into */
1229 struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_bufs];
1230 /* Number of mbufs for transmission on each slave */
1231 uint16_t slave_nb_bufs[RTE_MAX_ETHPORTS] = { 0 };
1232 /* Mapping array generated by hash function to map mbufs to slaves */
1233 uint16_t bufs_slave_port_idxs[nb_bufs];
1235 uint16_t slave_tx_count, slave_tx_fail_count[RTE_MAX_ETHPORTS] = { 0 };
1236 uint16_t total_tx_count = 0, total_tx_fail_count = 0;
1240 if (unlikely(nb_bufs == 0))
1243 /* Copy slave list to protect against slave up/down changes during tx
1245 slave_count = internals->active_slave_count;
1246 if (unlikely(slave_count < 1))
1249 memcpy(slave_port_ids, internals->active_slaves,
1250 sizeof(slave_port_ids[0]) * slave_count);
1253 * Populate slaves mbuf with the packets which are to be sent on it
1254 * selecting output slave using hash based on xmit policy
1256 internals->burst_xmit_hash(bufs, nb_bufs, slave_count,
1257 bufs_slave_port_idxs);
1259 for (i = 0; i < nb_bufs; i++) {
1260 /* Populate slave mbuf arrays with mbufs for that slave. */
1261 uint8_t slave_idx = bufs_slave_port_idxs[i];
1263 slave_bufs[slave_idx][slave_nb_bufs[slave_idx]++] = bufs[i];
1266 /* Send packet burst on each slave device */
1267 for (i = 0; i < slave_count; i++) {
1268 if (slave_nb_bufs[i] == 0)
1271 slave_tx_count = rte_eth_tx_burst(slave_port_ids[i],
1272 bd_tx_q->queue_id, slave_bufs[i],
1275 total_tx_count += slave_tx_count;
1277 /* If tx burst fails move packets to end of bufs */
1278 if (unlikely(slave_tx_count < slave_nb_bufs[i])) {
1279 slave_tx_fail_count[i] = slave_nb_bufs[i] -
1281 total_tx_fail_count += slave_tx_fail_count[i];
1284 * Shift bufs to beginning of array to allow reordering
1287 for (j = 0; j < slave_tx_fail_count[i]; j++) {
1289 slave_bufs[i][(slave_tx_count - 1) + j];
1295 * If there are tx burst failures we move packets to end of bufs to
1296 * preserve expected PMD behaviour of all failed transmitted being
1297 * at the end of the input mbuf array
1299 if (unlikely(total_tx_fail_count > 0)) {
1300 int bufs_idx = nb_bufs - total_tx_fail_count - 1;
1302 for (i = 0; i < slave_count; i++) {
1303 if (slave_tx_fail_count[i] > 0) {
1304 for (j = 0; j < slave_tx_fail_count[i]; j++)
1305 bufs[bufs_idx++] = slave_bufs[i][j];
1310 return total_tx_count;
1314 bond_ethdev_tx_burst_8023ad(void *queue, struct rte_mbuf **bufs,
1317 struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
1318 struct bond_dev_private *internals = bd_tx_q->dev_private;
1320 uint16_t slave_port_ids[RTE_MAX_ETHPORTS];
1321 uint16_t slave_count;
1323 uint16_t dist_slave_port_ids[RTE_MAX_ETHPORTS];
1324 uint16_t dist_slave_count;
1326 /* 2-D array to sort mbufs for transmission on each slave into */
1327 struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_bufs];
1328 /* Number of mbufs for transmission on each slave */
1329 uint16_t slave_nb_bufs[RTE_MAX_ETHPORTS] = { 0 };
1330 /* Mapping array generated by hash function to map mbufs to slaves */
1331 uint16_t bufs_slave_port_idxs[RTE_MAX_ETHPORTS] = { 0 };
1333 uint16_t slave_tx_count, slave_tx_fail_count[RTE_MAX_ETHPORTS] = { 0 };
1334 uint16_t total_tx_count = 0, total_tx_fail_count = 0;
1338 if (unlikely(nb_bufs == 0))
1341 /* Copy slave list to protect against slave up/down changes during tx
1343 slave_count = internals->active_slave_count;
1344 if (unlikely(slave_count < 1))
1347 memcpy(slave_port_ids, internals->active_slaves,
1348 sizeof(slave_port_ids[0]) * slave_count);
1350 dist_slave_count = 0;
1351 for (i = 0; i < slave_count; i++) {
1352 struct port *port = &mode_8023ad_ports[slave_port_ids[i]];
1354 if (ACTOR_STATE(port, DISTRIBUTING))
1355 dist_slave_port_ids[dist_slave_count++] =
1359 if (likely(dist_slave_count > 1)) {
1362 * Populate slaves mbuf with the packets which are to be sent
1363 * on it, selecting output slave using hash based on xmit policy
1365 internals->burst_xmit_hash(bufs, nb_bufs, dist_slave_count,
1366 bufs_slave_port_idxs);
1368 for (i = 0; i < nb_bufs; i++) {
1370 * Populate slave mbuf arrays with mbufs for that
1373 uint8_t slave_idx = bufs_slave_port_idxs[i];
1375 slave_bufs[slave_idx][slave_nb_bufs[slave_idx]++] =
1380 /* Send packet burst on each slave device */
1381 for (i = 0; i < dist_slave_count; i++) {
1382 if (slave_nb_bufs[i] == 0)
1385 slave_tx_count = rte_eth_tx_burst(
1386 dist_slave_port_ids[i],
1387 bd_tx_q->queue_id, slave_bufs[i],
1390 total_tx_count += slave_tx_count;
1392 /* If tx burst fails move packets to end of bufs */
1393 if (unlikely(slave_tx_count < slave_nb_bufs[i])) {
1394 slave_tx_fail_count[i] = slave_nb_bufs[i] -
1396 total_tx_fail_count += slave_tx_fail_count[i];
1399 * Shift bufs to beginning of array to allow
1402 for (j = 0; j < slave_tx_fail_count[i]; j++)
1405 [(slave_tx_count - 1)
1411 * If there are tx burst failures we move packets to end of
1412 * bufs to preserve expected PMD behaviour of all failed
1413 * transmitted being at the end of the input mbuf array
1415 if (unlikely(total_tx_fail_count > 0)) {
1416 int bufs_idx = nb_bufs - total_tx_fail_count - 1;
1418 for (i = 0; i < slave_count; i++) {
1419 if (slave_tx_fail_count[i] > 0) {
1421 j < slave_tx_fail_count[i];
1431 /* Check for LACP control packets and send if available */
1432 for (i = 0; i < slave_count; i++) {
1433 struct port *port = &mode_8023ad_ports[slave_port_ids[i]];
1434 struct rte_mbuf *ctrl_pkt = NULL;
1436 if (likely(rte_ring_empty(port->tx_ring)))
1439 if (rte_ring_dequeue(port->tx_ring,
1440 (void **)&ctrl_pkt) != -ENOENT) {
1441 slave_tx_count = rte_eth_tx_burst(slave_port_ids[i],
1442 bd_tx_q->queue_id, &ctrl_pkt, 1);
1444 * re-enqueue LAG control plane packets to buffering
1445 * ring if transmission fails so the packet isn't lost.
1447 if (slave_tx_count != 1)
1448 rte_ring_enqueue(port->tx_ring, ctrl_pkt);
1452 return total_tx_count;
1456 bond_ethdev_tx_burst_broadcast(void *queue, struct rte_mbuf **bufs,
1459 struct bond_dev_private *internals;
1460 struct bond_tx_queue *bd_tx_q;
1462 uint8_t tx_failed_flag = 0, num_of_slaves;
1463 uint16_t slaves[RTE_MAX_ETHPORTS];
1465 uint16_t max_nb_of_tx_pkts = 0;
1467 int slave_tx_total[RTE_MAX_ETHPORTS];
1468 int i, most_successful_tx_slave = -1;
1470 bd_tx_q = (struct bond_tx_queue *)queue;
1471 internals = bd_tx_q->dev_private;
1473 /* Copy slave list to protect against slave up/down changes during tx
1475 num_of_slaves = internals->active_slave_count;
1476 memcpy(slaves, internals->active_slaves,
1477 sizeof(internals->active_slaves[0]) * num_of_slaves);
1479 if (num_of_slaves < 1)
1482 /* Increment reference count on mbufs */
1483 for (i = 0; i < nb_pkts; i++)
1484 rte_mbuf_refcnt_update(bufs[i], num_of_slaves - 1);
1486 /* Transmit burst on each active slave */
1487 for (i = 0; i < num_of_slaves; i++) {
1488 slave_tx_total[i] = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
1491 if (unlikely(slave_tx_total[i] < nb_pkts))
1494 /* record the value and slave index for the slave which transmits the
1495 * maximum number of packets */
1496 if (slave_tx_total[i] > max_nb_of_tx_pkts) {
1497 max_nb_of_tx_pkts = slave_tx_total[i];
1498 most_successful_tx_slave = i;
1502 /* if slaves fail to transmit packets from burst, the calling application
1503 * is not expected to know about multiple references to packets so we must
1504 * handle failures of all packets except those of the most successful slave
1506 if (unlikely(tx_failed_flag))
1507 for (i = 0; i < num_of_slaves; i++)
1508 if (i != most_successful_tx_slave)
1509 while (slave_tx_total[i] < nb_pkts)
1510 rte_pktmbuf_free(bufs[slave_tx_total[i]++]);
1512 return max_nb_of_tx_pkts;
1516 link_properties_set(struct rte_eth_dev *ethdev, struct rte_eth_link *slave_link)
1518 struct bond_dev_private *bond_ctx = ethdev->data->dev_private;
1520 if (bond_ctx->mode == BONDING_MODE_8023AD) {
1522 * If in mode 4 then save the link properties of the first
1523 * slave, all subsequent slaves must match these properties
1525 struct rte_eth_link *bond_link = &bond_ctx->mode4.slave_link;
1527 bond_link->link_autoneg = slave_link->link_autoneg;
1528 bond_link->link_duplex = slave_link->link_duplex;
1529 bond_link->link_speed = slave_link->link_speed;
1532 * In any other mode the link properties are set to default
1533 * values of AUTONEG/DUPLEX
1535 ethdev->data->dev_link.link_autoneg = ETH_LINK_AUTONEG;
1536 ethdev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
1541 link_properties_valid(struct rte_eth_dev *ethdev,
1542 struct rte_eth_link *slave_link)
1544 struct bond_dev_private *bond_ctx = ethdev->data->dev_private;
1546 if (bond_ctx->mode == BONDING_MODE_8023AD) {
1547 struct rte_eth_link *bond_link = &bond_ctx->mode4.slave_link;
1549 if (bond_link->link_duplex != slave_link->link_duplex ||
1550 bond_link->link_autoneg != slave_link->link_autoneg ||
1551 bond_link->link_speed != slave_link->link_speed)
1559 mac_address_get(struct rte_eth_dev *eth_dev, struct ether_addr *dst_mac_addr)
1561 struct ether_addr *mac_addr;
1563 if (eth_dev == NULL) {
1564 RTE_LOG(ERR, PMD, "%s: NULL pointer eth_dev specified\n", __func__);
1568 if (dst_mac_addr == NULL) {
1569 RTE_LOG(ERR, PMD, "%s: NULL pointer MAC specified\n", __func__);
1573 mac_addr = eth_dev->data->mac_addrs;
1575 ether_addr_copy(mac_addr, dst_mac_addr);
1580 mac_address_set(struct rte_eth_dev *eth_dev, struct ether_addr *new_mac_addr)
1582 struct ether_addr *mac_addr;
1584 if (eth_dev == NULL) {
1585 RTE_BOND_LOG(ERR, "NULL pointer eth_dev specified");
1589 if (new_mac_addr == NULL) {
1590 RTE_BOND_LOG(ERR, "NULL pointer MAC specified");
1594 mac_addr = eth_dev->data->mac_addrs;
1596 /* If new MAC is different to current MAC then update */
1597 if (memcmp(mac_addr, new_mac_addr, sizeof(*mac_addr)) != 0)
1598 memcpy(mac_addr, new_mac_addr, sizeof(*mac_addr));
1604 mac_address_slaves_update(struct rte_eth_dev *bonded_eth_dev)
1606 struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1609 /* Update slave devices MAC addresses */
1610 if (internals->slave_count < 1)
1613 switch (internals->mode) {
1614 case BONDING_MODE_ROUND_ROBIN:
1615 case BONDING_MODE_BALANCE:
1616 case BONDING_MODE_BROADCAST:
1617 for (i = 0; i < internals->slave_count; i++) {
1618 if (rte_eth_dev_default_mac_addr_set(
1619 internals->slaves[i].port_id,
1620 bonded_eth_dev->data->mac_addrs)) {
1621 RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1622 internals->slaves[i].port_id);
1627 case BONDING_MODE_8023AD:
1628 bond_mode_8023ad_mac_address_update(bonded_eth_dev);
1630 case BONDING_MODE_ACTIVE_BACKUP:
1631 case BONDING_MODE_TLB:
1632 case BONDING_MODE_ALB:
1634 for (i = 0; i < internals->slave_count; i++) {
1635 if (internals->slaves[i].port_id ==
1636 internals->current_primary_port) {
1637 if (rte_eth_dev_default_mac_addr_set(
1638 internals->primary_port,
1639 bonded_eth_dev->data->mac_addrs)) {
1640 RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1641 internals->current_primary_port);
1645 if (rte_eth_dev_default_mac_addr_set(
1646 internals->slaves[i].port_id,
1647 &internals->slaves[i].persisted_mac_addr)) {
1648 RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1649 internals->slaves[i].port_id);
1660 bond_ethdev_mode_set(struct rte_eth_dev *eth_dev, int mode)
1662 struct bond_dev_private *internals;
1664 internals = eth_dev->data->dev_private;
1667 case BONDING_MODE_ROUND_ROBIN:
1668 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_round_robin;
1669 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1671 case BONDING_MODE_ACTIVE_BACKUP:
1672 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_active_backup;
1673 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup;
1675 case BONDING_MODE_BALANCE:
1676 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_balance;
1677 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1679 case BONDING_MODE_BROADCAST:
1680 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_broadcast;
1681 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1683 case BONDING_MODE_8023AD:
1684 if (bond_mode_8023ad_enable(eth_dev) != 0)
1687 if (internals->mode4.dedicated_queues.enabled == 0) {
1688 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_8023ad;
1689 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_8023ad;
1690 RTE_LOG(WARNING, PMD,
1691 "Using mode 4, it is necessary to do TX burst "
1692 "and RX burst at least every 100ms.\n");
1694 /* Use flow director's optimization */
1695 eth_dev->rx_pkt_burst =
1696 bond_ethdev_rx_burst_8023ad_fast_queue;
1697 eth_dev->tx_pkt_burst =
1698 bond_ethdev_tx_burst_8023ad_fast_queue;
1701 case BONDING_MODE_TLB:
1702 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_tlb;
1703 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup;
1705 case BONDING_MODE_ALB:
1706 if (bond_mode_alb_enable(eth_dev) != 0)
1709 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_alb;
1710 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_alb;
1716 internals->mode = mode;
1723 slave_configure_slow_queue(struct rte_eth_dev *bonded_eth_dev,
1724 struct rte_eth_dev *slave_eth_dev)
1727 struct bond_dev_private *internals = (struct bond_dev_private *)
1728 bonded_eth_dev->data->dev_private;
1729 struct port *port = &mode_8023ad_ports[slave_eth_dev->data->port_id];
1731 if (port->slow_pool == NULL) {
1733 int slave_id = slave_eth_dev->data->port_id;
1735 snprintf(mem_name, RTE_DIM(mem_name), "slave_port%u_slow_pool",
1737 port->slow_pool = rte_pktmbuf_pool_create(mem_name, 8191,
1738 250, 0, RTE_MBUF_DEFAULT_BUF_SIZE,
1739 slave_eth_dev->data->numa_node);
1741 /* Any memory allocation failure in initialization is critical because
1742 * resources can't be free, so reinitialization is impossible. */
1743 if (port->slow_pool == NULL) {
1744 rte_panic("Slave %u: Failed to create memory pool '%s': %s\n",
1745 slave_id, mem_name, rte_strerror(rte_errno));
1749 if (internals->mode4.dedicated_queues.enabled == 1) {
1750 /* Configure slow Rx queue */
1752 errval = rte_eth_rx_queue_setup(slave_eth_dev->data->port_id,
1753 internals->mode4.dedicated_queues.rx_qid, 128,
1754 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1755 NULL, port->slow_pool);
1758 "rte_eth_rx_queue_setup: port=%d queue_id %d, err (%d)",
1759 slave_eth_dev->data->port_id,
1760 internals->mode4.dedicated_queues.rx_qid,
1765 errval = rte_eth_tx_queue_setup(slave_eth_dev->data->port_id,
1766 internals->mode4.dedicated_queues.tx_qid, 512,
1767 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1771 "rte_eth_tx_queue_setup: port=%d queue_id %d, err (%d)",
1772 slave_eth_dev->data->port_id,
1773 internals->mode4.dedicated_queues.tx_qid,
1782 slave_configure(struct rte_eth_dev *bonded_eth_dev,
1783 struct rte_eth_dev *slave_eth_dev)
1785 struct bond_rx_queue *bd_rx_q;
1786 struct bond_tx_queue *bd_tx_q;
1787 uint16_t nb_rx_queues;
1788 uint16_t nb_tx_queues;
1792 struct rte_flow_error flow_error;
1794 struct bond_dev_private *internals = (struct bond_dev_private *)
1795 bonded_eth_dev->data->dev_private;
1798 rte_eth_dev_stop(slave_eth_dev->data->port_id);
1800 /* Enable interrupts on slave device if supported */
1801 if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1802 slave_eth_dev->data->dev_conf.intr_conf.lsc = 1;
1804 /* If RSS is enabled for bonding, try to enable it for slaves */
1805 if (bonded_eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG) {
1806 if (bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len
1808 slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len =
1809 bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len;
1810 slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key =
1811 bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key;
1813 slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
1816 slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf =
1817 bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
1818 slave_eth_dev->data->dev_conf.rxmode.mq_mode =
1819 bonded_eth_dev->data->dev_conf.rxmode.mq_mode;
1822 slave_eth_dev->data->dev_conf.rxmode.hw_vlan_filter =
1823 bonded_eth_dev->data->dev_conf.rxmode.hw_vlan_filter;
1825 nb_rx_queues = bonded_eth_dev->data->nb_rx_queues;
1826 nb_tx_queues = bonded_eth_dev->data->nb_tx_queues;
1828 if (internals->mode == BONDING_MODE_8023AD) {
1829 if (internals->mode4.dedicated_queues.enabled == 1) {
1835 errval = rte_eth_dev_set_mtu(slave_eth_dev->data->port_id,
1836 bonded_eth_dev->data->mtu);
1837 if (errval != 0 && errval != -ENOTSUP) {
1838 RTE_BOND_LOG(ERR, "rte_eth_dev_set_mtu: port %u, err (%d)",
1839 slave_eth_dev->data->port_id, errval);
1843 /* Configure device */
1844 errval = rte_eth_dev_configure(slave_eth_dev->data->port_id,
1845 nb_rx_queues, nb_tx_queues,
1846 &(slave_eth_dev->data->dev_conf));
1848 RTE_BOND_LOG(ERR, "Cannot configure slave device: port %u , err (%d)",
1849 slave_eth_dev->data->port_id, errval);
1853 /* Setup Rx Queues */
1854 for (q_id = 0; q_id < bonded_eth_dev->data->nb_rx_queues; q_id++) {
1855 bd_rx_q = (struct bond_rx_queue *)bonded_eth_dev->data->rx_queues[q_id];
1857 errval = rte_eth_rx_queue_setup(slave_eth_dev->data->port_id, q_id,
1858 bd_rx_q->nb_rx_desc,
1859 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1860 &(bd_rx_q->rx_conf), bd_rx_q->mb_pool);
1863 "rte_eth_rx_queue_setup: port=%d queue_id %d, err (%d)",
1864 slave_eth_dev->data->port_id, q_id, errval);
1869 /* Setup Tx Queues */
1870 for (q_id = 0; q_id < bonded_eth_dev->data->nb_tx_queues; q_id++) {
1871 bd_tx_q = (struct bond_tx_queue *)bonded_eth_dev->data->tx_queues[q_id];
1873 errval = rte_eth_tx_queue_setup(slave_eth_dev->data->port_id, q_id,
1874 bd_tx_q->nb_tx_desc,
1875 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1879 "rte_eth_tx_queue_setup: port=%d queue_id %d, err (%d)",
1880 slave_eth_dev->data->port_id, q_id, errval);
1885 if (internals->mode == BONDING_MODE_8023AD &&
1886 internals->mode4.dedicated_queues.enabled == 1) {
1887 if (slave_configure_slow_queue(bonded_eth_dev, slave_eth_dev)
1891 if (bond_ethdev_8023ad_flow_verify(bonded_eth_dev,
1892 slave_eth_dev->data->port_id) != 0) {
1894 "rte_eth_tx_queue_setup: port=%d queue_id %d, err (%d)",
1895 slave_eth_dev->data->port_id, q_id, errval);
1899 if (internals->mode4.dedicated_queues.flow[slave_eth_dev->data->port_id] != NULL)
1900 rte_flow_destroy(slave_eth_dev->data->port_id,
1901 internals->mode4.dedicated_queues.flow[slave_eth_dev->data->port_id],
1904 bond_ethdev_8023ad_flow_set(bonded_eth_dev,
1905 slave_eth_dev->data->port_id);
1909 errval = rte_eth_dev_start(slave_eth_dev->data->port_id);
1911 RTE_BOND_LOG(ERR, "rte_eth_dev_start: port=%u, err (%d)",
1912 slave_eth_dev->data->port_id, errval);
1916 /* If RSS is enabled for bonding, synchronize RETA */
1917 if (bonded_eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS) {
1919 struct bond_dev_private *internals;
1921 internals = bonded_eth_dev->data->dev_private;
1923 for (i = 0; i < internals->slave_count; i++) {
1924 if (internals->slaves[i].port_id == slave_eth_dev->data->port_id) {
1925 errval = rte_eth_dev_rss_reta_update(
1926 slave_eth_dev->data->port_id,
1927 &internals->reta_conf[0],
1928 internals->slaves[i].reta_size);
1930 RTE_LOG(WARNING, PMD,
1931 "rte_eth_dev_rss_reta_update on slave port %d fails (err %d)."
1932 " RSS Configuration for bonding may be inconsistent.\n",
1933 slave_eth_dev->data->port_id, errval);
1940 /* If lsc interrupt is set, check initial slave's link status */
1941 if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
1942 slave_eth_dev->dev_ops->link_update(slave_eth_dev, 0);
1943 bond_ethdev_lsc_event_callback(slave_eth_dev->data->port_id,
1944 RTE_ETH_EVENT_INTR_LSC, &bonded_eth_dev->data->port_id,
1952 slave_remove(struct bond_dev_private *internals,
1953 struct rte_eth_dev *slave_eth_dev)
1957 for (i = 0; i < internals->slave_count; i++)
1958 if (internals->slaves[i].port_id ==
1959 slave_eth_dev->data->port_id)
1962 if (i < (internals->slave_count - 1))
1963 memmove(&internals->slaves[i], &internals->slaves[i + 1],
1964 sizeof(internals->slaves[0]) *
1965 (internals->slave_count - i - 1));
1967 internals->slave_count--;
1969 /* force reconfiguration of slave interfaces */
1970 _rte_eth_dev_reset(slave_eth_dev);
1974 bond_ethdev_slave_link_status_change_monitor(void *cb_arg);
1977 slave_add(struct bond_dev_private *internals,
1978 struct rte_eth_dev *slave_eth_dev)
1980 struct bond_slave_details *slave_details =
1981 &internals->slaves[internals->slave_count];
1983 slave_details->port_id = slave_eth_dev->data->port_id;
1984 slave_details->last_link_status = 0;
1986 /* Mark slave devices that don't support interrupts so we can
1987 * compensate when we start the bond
1989 if (!(slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1990 slave_details->link_status_poll_enabled = 1;
1993 slave_details->link_status_wait_to_complete = 0;
1994 /* clean tlb_last_obytes when adding port for bonding device */
1995 memcpy(&(slave_details->persisted_mac_addr), slave_eth_dev->data->mac_addrs,
1996 sizeof(struct ether_addr));
2000 bond_ethdev_primary_set(struct bond_dev_private *internals,
2001 uint16_t slave_port_id)
2005 if (internals->active_slave_count < 1)
2006 internals->current_primary_port = slave_port_id;
2008 /* Search bonded device slave ports for new proposed primary port */
2009 for (i = 0; i < internals->active_slave_count; i++) {
2010 if (internals->active_slaves[i] == slave_port_id)
2011 internals->current_primary_port = slave_port_id;
2016 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev);
2019 bond_ethdev_start(struct rte_eth_dev *eth_dev)
2021 struct bond_dev_private *internals;
2024 /* slave eth dev will be started by bonded device */
2025 if (check_for_bonded_ethdev(eth_dev)) {
2026 RTE_BOND_LOG(ERR, "User tried to explicitly start a slave eth_dev (%d)",
2027 eth_dev->data->port_id);
2031 eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
2032 eth_dev->data->dev_started = 1;
2034 internals = eth_dev->data->dev_private;
2036 if (internals->slave_count == 0) {
2037 RTE_BOND_LOG(ERR, "Cannot start port since there are no slave devices");
2041 if (internals->user_defined_mac == 0) {
2042 struct ether_addr *new_mac_addr = NULL;
2044 for (i = 0; i < internals->slave_count; i++)
2045 if (internals->slaves[i].port_id == internals->primary_port)
2046 new_mac_addr = &internals->slaves[i].persisted_mac_addr;
2048 if (new_mac_addr == NULL)
2051 if (mac_address_set(eth_dev, new_mac_addr) != 0) {
2052 RTE_BOND_LOG(ERR, "bonded port (%d) failed to update MAC address",
2053 eth_dev->data->port_id);
2058 /* Update all slave devices MACs*/
2059 if (mac_address_slaves_update(eth_dev) != 0)
2062 /* If bonded device is configure in promiscuous mode then re-apply config */
2063 if (internals->promiscuous_en)
2064 bond_ethdev_promiscuous_enable(eth_dev);
2066 if (internals->mode == BONDING_MODE_8023AD) {
2067 if (internals->mode4.dedicated_queues.enabled == 1) {
2068 internals->mode4.dedicated_queues.rx_qid =
2069 eth_dev->data->nb_rx_queues;
2070 internals->mode4.dedicated_queues.tx_qid =
2071 eth_dev->data->nb_tx_queues;
2076 /* Reconfigure each slave device if starting bonded device */
2077 for (i = 0; i < internals->slave_count; i++) {
2078 struct rte_eth_dev *slave_ethdev =
2079 &(rte_eth_devices[internals->slaves[i].port_id]);
2080 if (slave_configure(eth_dev, slave_ethdev) != 0) {
2082 "bonded port (%d) failed to reconfigure slave device (%d)",
2083 eth_dev->data->port_id,
2084 internals->slaves[i].port_id);
2087 /* We will need to poll for link status if any slave doesn't
2088 * support interrupts
2090 if (internals->slaves[i].link_status_poll_enabled)
2091 internals->link_status_polling_enabled = 1;
2093 /* start polling if needed */
2094 if (internals->link_status_polling_enabled) {
2096 internals->link_status_polling_interval_ms * 1000,
2097 bond_ethdev_slave_link_status_change_monitor,
2098 (void *)&rte_eth_devices[internals->port_id]);
2101 if (internals->user_defined_primary_port)
2102 bond_ethdev_primary_set(internals, internals->primary_port);
2104 if (internals->mode == BONDING_MODE_8023AD)
2105 bond_mode_8023ad_start(eth_dev);
2107 if (internals->mode == BONDING_MODE_TLB ||
2108 internals->mode == BONDING_MODE_ALB)
2109 bond_tlb_enable(internals);
2115 bond_ethdev_free_queues(struct rte_eth_dev *dev)
2119 if (dev->data->rx_queues != NULL) {
2120 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2121 rte_free(dev->data->rx_queues[i]);
2122 dev->data->rx_queues[i] = NULL;
2124 dev->data->nb_rx_queues = 0;
2127 if (dev->data->tx_queues != NULL) {
2128 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2129 rte_free(dev->data->tx_queues[i]);
2130 dev->data->tx_queues[i] = NULL;
2132 dev->data->nb_tx_queues = 0;
2137 bond_ethdev_stop(struct rte_eth_dev *eth_dev)
2139 struct bond_dev_private *internals = eth_dev->data->dev_private;
2142 if (internals->mode == BONDING_MODE_8023AD) {
2146 bond_mode_8023ad_stop(eth_dev);
2148 /* Discard all messages to/from mode 4 state machines */
2149 for (i = 0; i < internals->active_slave_count; i++) {
2150 port = &mode_8023ad_ports[internals->active_slaves[i]];
2152 RTE_ASSERT(port->rx_ring != NULL);
2153 while (rte_ring_dequeue(port->rx_ring, &pkt) != -ENOENT)
2154 rte_pktmbuf_free(pkt);
2156 RTE_ASSERT(port->tx_ring != NULL);
2157 while (rte_ring_dequeue(port->tx_ring, &pkt) != -ENOENT)
2158 rte_pktmbuf_free(pkt);
2162 if (internals->mode == BONDING_MODE_TLB ||
2163 internals->mode == BONDING_MODE_ALB) {
2164 bond_tlb_disable(internals);
2165 for (i = 0; i < internals->active_slave_count; i++)
2166 tlb_last_obytets[internals->active_slaves[i]] = 0;
2169 internals->active_slave_count = 0;
2170 internals->link_status_polling_enabled = 0;
2171 for (i = 0; i < internals->slave_count; i++)
2172 internals->slaves[i].last_link_status = 0;
2174 eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
2175 eth_dev->data->dev_started = 0;
2179 bond_ethdev_close(struct rte_eth_dev *dev)
2181 struct bond_dev_private *internals = dev->data->dev_private;
2182 uint8_t bond_port_id = internals->port_id;
2185 RTE_LOG(INFO, EAL, "Closing bonded device %s\n", dev->device->name);
2186 while (internals->slave_count != skipped) {
2187 uint16_t port_id = internals->slaves[skipped].port_id;
2189 rte_eth_dev_stop(port_id);
2191 if (rte_eth_bond_slave_remove(bond_port_id, port_id) != 0) {
2193 "Failed to remove port %d from bonded device "
2194 "%s\n", port_id, dev->device->name);
2198 bond_ethdev_free_queues(dev);
2199 rte_bitmap_reset(internals->vlan_filter_bmp);
2202 /* forward declaration */
2203 static int bond_ethdev_configure(struct rte_eth_dev *dev);
2206 bond_ethdev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2208 struct bond_dev_private *internals = dev->data->dev_private;
2210 uint16_t max_nb_rx_queues = UINT16_MAX;
2211 uint16_t max_nb_tx_queues = UINT16_MAX;
2213 dev_info->max_mac_addrs = 1;
2215 dev_info->max_rx_pktlen = internals->candidate_max_rx_pktlen ?
2216 internals->candidate_max_rx_pktlen :
2217 ETHER_MAX_JUMBO_FRAME_LEN;
2219 /* Max number of tx/rx queues that the bonded device can support is the
2220 * minimum values of the bonded slaves, as all slaves must be capable
2221 * of supporting the same number of tx/rx queues.
2223 if (internals->slave_count > 0) {
2224 struct rte_eth_dev_info slave_info;
2227 for (idx = 0; idx < internals->slave_count; idx++) {
2228 rte_eth_dev_info_get(internals->slaves[idx].port_id,
2231 if (slave_info.max_rx_queues < max_nb_rx_queues)
2232 max_nb_rx_queues = slave_info.max_rx_queues;
2234 if (slave_info.max_tx_queues < max_nb_tx_queues)
2235 max_nb_tx_queues = slave_info.max_tx_queues;
2239 dev_info->max_rx_queues = max_nb_rx_queues;
2240 dev_info->max_tx_queues = max_nb_tx_queues;
2243 * If dedicated hw queues enabled for link bonding device in LACP mode
2244 * then we need to reduce the maximum number of data path queues by 1.
2246 if (internals->mode == BONDING_MODE_8023AD &&
2247 internals->mode4.dedicated_queues.enabled == 1) {
2248 dev_info->max_rx_queues--;
2249 dev_info->max_tx_queues--;
2252 dev_info->min_rx_bufsize = 0;
2254 dev_info->rx_offload_capa = internals->rx_offload_capa;
2255 dev_info->tx_offload_capa = internals->tx_offload_capa;
2256 dev_info->flow_type_rss_offloads = internals->flow_type_rss_offloads;
2258 dev_info->reta_size = internals->reta_size;
2262 bond_ethdev_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
2266 struct bond_dev_private *internals = dev->data->dev_private;
2268 /* don't do this while a slave is being added */
2269 rte_spinlock_lock(&internals->lock);
2272 rte_bitmap_set(internals->vlan_filter_bmp, vlan_id);
2274 rte_bitmap_clear(internals->vlan_filter_bmp, vlan_id);
2276 for (i = 0; i < internals->slave_count; i++) {
2277 uint16_t port_id = internals->slaves[i].port_id;
2279 res = rte_eth_dev_vlan_filter(port_id, vlan_id, on);
2281 RTE_LOG(WARNING, PMD,
2282 "Setting VLAN filter on slave port %u not supported.\n",
2286 rte_spinlock_unlock(&internals->lock);
2291 bond_ethdev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
2292 uint16_t nb_rx_desc, unsigned int socket_id __rte_unused,
2293 const struct rte_eth_rxconf *rx_conf, struct rte_mempool *mb_pool)
2295 struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)
2296 rte_zmalloc_socket(NULL, sizeof(struct bond_rx_queue),
2297 0, dev->data->numa_node);
2298 if (bd_rx_q == NULL)
2301 bd_rx_q->queue_id = rx_queue_id;
2302 bd_rx_q->dev_private = dev->data->dev_private;
2304 bd_rx_q->nb_rx_desc = nb_rx_desc;
2306 memcpy(&(bd_rx_q->rx_conf), rx_conf, sizeof(struct rte_eth_rxconf));
2307 bd_rx_q->mb_pool = mb_pool;
2309 dev->data->rx_queues[rx_queue_id] = bd_rx_q;
2315 bond_ethdev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
2316 uint16_t nb_tx_desc, unsigned int socket_id __rte_unused,
2317 const struct rte_eth_txconf *tx_conf)
2319 struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)
2320 rte_zmalloc_socket(NULL, sizeof(struct bond_tx_queue),
2321 0, dev->data->numa_node);
2323 if (bd_tx_q == NULL)
2326 bd_tx_q->queue_id = tx_queue_id;
2327 bd_tx_q->dev_private = dev->data->dev_private;
2329 bd_tx_q->nb_tx_desc = nb_tx_desc;
2330 memcpy(&(bd_tx_q->tx_conf), tx_conf, sizeof(bd_tx_q->tx_conf));
2332 dev->data->tx_queues[tx_queue_id] = bd_tx_q;
2338 bond_ethdev_rx_queue_release(void *queue)
2347 bond_ethdev_tx_queue_release(void *queue)
2356 bond_ethdev_slave_link_status_change_monitor(void *cb_arg)
2358 struct rte_eth_dev *bonded_ethdev, *slave_ethdev;
2359 struct bond_dev_private *internals;
2361 /* Default value for polling slave found is true as we don't want to
2362 * disable the polling thread if we cannot get the lock */
2363 int i, polling_slave_found = 1;
2368 bonded_ethdev = (struct rte_eth_dev *)cb_arg;
2369 internals = (struct bond_dev_private *)bonded_ethdev->data->dev_private;
2371 if (!bonded_ethdev->data->dev_started ||
2372 !internals->link_status_polling_enabled)
2375 /* If device is currently being configured then don't check slaves link
2376 * status, wait until next period */
2377 if (rte_spinlock_trylock(&internals->lock)) {
2378 if (internals->slave_count > 0)
2379 polling_slave_found = 0;
2381 for (i = 0; i < internals->slave_count; i++) {
2382 if (!internals->slaves[i].link_status_poll_enabled)
2385 slave_ethdev = &rte_eth_devices[internals->slaves[i].port_id];
2386 polling_slave_found = 1;
2388 /* Update slave link status */
2389 (*slave_ethdev->dev_ops->link_update)(slave_ethdev,
2390 internals->slaves[i].link_status_wait_to_complete);
2392 /* if link status has changed since last checked then call lsc
2394 if (slave_ethdev->data->dev_link.link_status !=
2395 internals->slaves[i].last_link_status) {
2396 internals->slaves[i].last_link_status =
2397 slave_ethdev->data->dev_link.link_status;
2399 bond_ethdev_lsc_event_callback(internals->slaves[i].port_id,
2400 RTE_ETH_EVENT_INTR_LSC,
2401 &bonded_ethdev->data->port_id,
2405 rte_spinlock_unlock(&internals->lock);
2408 if (polling_slave_found)
2409 /* Set alarm to continue monitoring link status of slave ethdev's */
2410 rte_eal_alarm_set(internals->link_status_polling_interval_ms * 1000,
2411 bond_ethdev_slave_link_status_change_monitor, cb_arg);
2415 bond_ethdev_link_update(struct rte_eth_dev *ethdev, int wait_to_complete)
2417 void (*link_update)(uint16_t port_id, struct rte_eth_link *eth_link);
2419 struct bond_dev_private *bond_ctx;
2420 struct rte_eth_link slave_link;
2424 bond_ctx = ethdev->data->dev_private;
2426 ethdev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
2428 if (ethdev->data->dev_started == 0 ||
2429 bond_ctx->active_slave_count == 0) {
2430 ethdev->data->dev_link.link_status = ETH_LINK_DOWN;
2434 ethdev->data->dev_link.link_status = ETH_LINK_UP;
2436 if (wait_to_complete)
2437 link_update = rte_eth_link_get;
2439 link_update = rte_eth_link_get_nowait;
2441 switch (bond_ctx->mode) {
2442 case BONDING_MODE_BROADCAST:
2444 * Setting link speed to UINT32_MAX to ensure we pick up the
2445 * value of the first active slave
2447 ethdev->data->dev_link.link_speed = UINT32_MAX;
2450 * link speed is minimum value of all the slaves link speed as
2451 * packet loss will occur on this slave if transmission at rates
2452 * greater than this are attempted
2454 for (idx = 1; idx < bond_ctx->active_slave_count; idx++) {
2455 link_update(bond_ctx->active_slaves[0], &slave_link);
2457 if (slave_link.link_speed <
2458 ethdev->data->dev_link.link_speed)
2459 ethdev->data->dev_link.link_speed =
2460 slave_link.link_speed;
2463 case BONDING_MODE_ACTIVE_BACKUP:
2464 /* Current primary slave */
2465 link_update(bond_ctx->current_primary_port, &slave_link);
2467 ethdev->data->dev_link.link_speed = slave_link.link_speed;
2469 case BONDING_MODE_8023AD:
2470 ethdev->data->dev_link.link_autoneg =
2471 bond_ctx->mode4.slave_link.link_autoneg;
2472 ethdev->data->dev_link.link_duplex =
2473 bond_ctx->mode4.slave_link.link_duplex;
2474 /* fall through to update link speed */
2475 case BONDING_MODE_ROUND_ROBIN:
2476 case BONDING_MODE_BALANCE:
2477 case BONDING_MODE_TLB:
2478 case BONDING_MODE_ALB:
2481 * In theses mode the maximum theoretical link speed is the sum
2484 ethdev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
2486 for (idx = 0; idx < bond_ctx->active_slave_count; idx++) {
2487 link_update(bond_ctx->active_slaves[idx], &slave_link);
2489 ethdev->data->dev_link.link_speed +=
2490 slave_link.link_speed;
2500 bond_ethdev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
2502 struct bond_dev_private *internals = dev->data->dev_private;
2503 struct rte_eth_stats slave_stats;
2506 for (i = 0; i < internals->slave_count; i++) {
2507 rte_eth_stats_get(internals->slaves[i].port_id, &slave_stats);
2509 stats->ipackets += slave_stats.ipackets;
2510 stats->opackets += slave_stats.opackets;
2511 stats->ibytes += slave_stats.ibytes;
2512 stats->obytes += slave_stats.obytes;
2513 stats->imissed += slave_stats.imissed;
2514 stats->ierrors += slave_stats.ierrors;
2515 stats->oerrors += slave_stats.oerrors;
2516 stats->rx_nombuf += slave_stats.rx_nombuf;
2518 for (j = 0; j < RTE_ETHDEV_QUEUE_STAT_CNTRS; j++) {
2519 stats->q_ipackets[j] += slave_stats.q_ipackets[j];
2520 stats->q_opackets[j] += slave_stats.q_opackets[j];
2521 stats->q_ibytes[j] += slave_stats.q_ibytes[j];
2522 stats->q_obytes[j] += slave_stats.q_obytes[j];
2523 stats->q_errors[j] += slave_stats.q_errors[j];
2532 bond_ethdev_stats_reset(struct rte_eth_dev *dev)
2534 struct bond_dev_private *internals = dev->data->dev_private;
2537 for (i = 0; i < internals->slave_count; i++)
2538 rte_eth_stats_reset(internals->slaves[i].port_id);
2542 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev)
2544 struct bond_dev_private *internals = eth_dev->data->dev_private;
2547 internals->promiscuous_en = 1;
2549 switch (internals->mode) {
2550 /* Promiscuous mode is propagated to all slaves */
2551 case BONDING_MODE_ROUND_ROBIN:
2552 case BONDING_MODE_BALANCE:
2553 case BONDING_MODE_BROADCAST:
2554 for (i = 0; i < internals->slave_count; i++)
2555 rte_eth_promiscuous_enable(internals->slaves[i].port_id);
2557 /* In mode4 promiscus mode is managed when slave is added/removed */
2558 case BONDING_MODE_8023AD:
2560 /* Promiscuous mode is propagated only to primary slave */
2561 case BONDING_MODE_ACTIVE_BACKUP:
2562 case BONDING_MODE_TLB:
2563 case BONDING_MODE_ALB:
2565 rte_eth_promiscuous_enable(internals->current_primary_port);
2570 bond_ethdev_promiscuous_disable(struct rte_eth_dev *dev)
2572 struct bond_dev_private *internals = dev->data->dev_private;
2575 internals->promiscuous_en = 0;
2577 switch (internals->mode) {
2578 /* Promiscuous mode is propagated to all slaves */
2579 case BONDING_MODE_ROUND_ROBIN:
2580 case BONDING_MODE_BALANCE:
2581 case BONDING_MODE_BROADCAST:
2582 for (i = 0; i < internals->slave_count; i++)
2583 rte_eth_promiscuous_disable(internals->slaves[i].port_id);
2585 /* In mode4 promiscus mode is set managed when slave is added/removed */
2586 case BONDING_MODE_8023AD:
2588 /* Promiscuous mode is propagated only to primary slave */
2589 case BONDING_MODE_ACTIVE_BACKUP:
2590 case BONDING_MODE_TLB:
2591 case BONDING_MODE_ALB:
2593 rte_eth_promiscuous_disable(internals->current_primary_port);
2598 bond_ethdev_delayed_lsc_propagation(void *arg)
2603 _rte_eth_dev_callback_process((struct rte_eth_dev *)arg,
2604 RTE_ETH_EVENT_INTR_LSC, NULL);
2608 bond_ethdev_lsc_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2609 void *param, void *ret_param __rte_unused)
2611 struct rte_eth_dev *bonded_eth_dev;
2612 struct bond_dev_private *internals;
2613 struct rte_eth_link link;
2616 int i, valid_slave = 0;
2618 uint8_t lsc_flag = 0;
2620 if (type != RTE_ETH_EVENT_INTR_LSC || param == NULL)
2623 bonded_eth_dev = &rte_eth_devices[*(uint8_t *)param];
2625 if (check_for_bonded_ethdev(bonded_eth_dev))
2628 internals = bonded_eth_dev->data->dev_private;
2630 /* If the device isn't started don't handle interrupts */
2631 if (!bonded_eth_dev->data->dev_started)
2634 /* verify that port_id is a valid slave of bonded port */
2635 for (i = 0; i < internals->slave_count; i++) {
2636 if (internals->slaves[i].port_id == port_id) {
2645 /* Search for port in active port list */
2646 active_pos = find_slave_by_id(internals->active_slaves,
2647 internals->active_slave_count, port_id);
2649 rte_eth_link_get_nowait(port_id, &link);
2650 if (link.link_status) {
2651 if (active_pos < internals->active_slave_count)
2654 /* if no active slave ports then set this port to be primary port */
2655 if (internals->active_slave_count < 1) {
2656 /* If first active slave, then change link status */
2657 bonded_eth_dev->data->dev_link.link_status = ETH_LINK_UP;
2658 internals->current_primary_port = port_id;
2661 mac_address_slaves_update(bonded_eth_dev);
2664 activate_slave(bonded_eth_dev, port_id);
2666 /* If user has defined the primary port then default to using it */
2667 if (internals->user_defined_primary_port &&
2668 internals->primary_port == port_id)
2669 bond_ethdev_primary_set(internals, port_id);
2671 if (active_pos == internals->active_slave_count)
2674 /* Remove from active slave list */
2675 deactivate_slave(bonded_eth_dev, port_id);
2677 if (internals->active_slave_count < 1)
2680 /* Update primary id, take first active slave from list or if none
2681 * available set to -1 */
2682 if (port_id == internals->current_primary_port) {
2683 if (internals->active_slave_count > 0)
2684 bond_ethdev_primary_set(internals,
2685 internals->active_slaves[0]);
2687 internals->current_primary_port = internals->primary_port;
2692 * Update bonded device link properties after any change to active
2695 bond_ethdev_link_update(bonded_eth_dev, 0);
2698 /* Cancel any possible outstanding interrupts if delays are enabled */
2699 if (internals->link_up_delay_ms > 0 ||
2700 internals->link_down_delay_ms > 0)
2701 rte_eal_alarm_cancel(bond_ethdev_delayed_lsc_propagation,
2704 if (bonded_eth_dev->data->dev_link.link_status) {
2705 if (internals->link_up_delay_ms > 0)
2706 rte_eal_alarm_set(internals->link_up_delay_ms * 1000,
2707 bond_ethdev_delayed_lsc_propagation,
2708 (void *)bonded_eth_dev);
2710 _rte_eth_dev_callback_process(bonded_eth_dev,
2711 RTE_ETH_EVENT_INTR_LSC,
2715 if (internals->link_down_delay_ms > 0)
2716 rte_eal_alarm_set(internals->link_down_delay_ms * 1000,
2717 bond_ethdev_delayed_lsc_propagation,
2718 (void *)bonded_eth_dev);
2720 _rte_eth_dev_callback_process(bonded_eth_dev,
2721 RTE_ETH_EVENT_INTR_LSC,
2729 bond_ethdev_rss_reta_update(struct rte_eth_dev *dev,
2730 struct rte_eth_rss_reta_entry64 *reta_conf, uint16_t reta_size)
2734 int slave_reta_size;
2735 unsigned reta_count;
2736 struct bond_dev_private *internals = dev->data->dev_private;
2738 if (reta_size != internals->reta_size)
2741 /* Copy RETA table */
2742 reta_count = reta_size / RTE_RETA_GROUP_SIZE;
2744 for (i = 0; i < reta_count; i++) {
2745 internals->reta_conf[i].mask = reta_conf[i].mask;
2746 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
2747 if ((reta_conf[i].mask >> j) & 0x01)
2748 internals->reta_conf[i].reta[j] = reta_conf[i].reta[j];
2751 /* Fill rest of array */
2752 for (; i < RTE_DIM(internals->reta_conf); i += reta_count)
2753 memcpy(&internals->reta_conf[i], &internals->reta_conf[0],
2754 sizeof(internals->reta_conf[0]) * reta_count);
2756 /* Propagate RETA over slaves */
2757 for (i = 0; i < internals->slave_count; i++) {
2758 slave_reta_size = internals->slaves[i].reta_size;
2759 result = rte_eth_dev_rss_reta_update(internals->slaves[i].port_id,
2760 &internals->reta_conf[0], slave_reta_size);
2769 bond_ethdev_rss_reta_query(struct rte_eth_dev *dev,
2770 struct rte_eth_rss_reta_entry64 *reta_conf, uint16_t reta_size)
2773 struct bond_dev_private *internals = dev->data->dev_private;
2775 if (reta_size != internals->reta_size)
2778 /* Copy RETA table */
2779 for (i = 0; i < reta_size / RTE_RETA_GROUP_SIZE; i++)
2780 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
2781 if ((reta_conf[i].mask >> j) & 0x01)
2782 reta_conf[i].reta[j] = internals->reta_conf[i].reta[j];
2788 bond_ethdev_rss_hash_update(struct rte_eth_dev *dev,
2789 struct rte_eth_rss_conf *rss_conf)
2792 struct bond_dev_private *internals = dev->data->dev_private;
2793 struct rte_eth_rss_conf bond_rss_conf;
2795 memcpy(&bond_rss_conf, rss_conf, sizeof(struct rte_eth_rss_conf));
2797 bond_rss_conf.rss_hf &= internals->flow_type_rss_offloads;
2799 if (bond_rss_conf.rss_hf != 0)
2800 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf = bond_rss_conf.rss_hf;
2802 if (bond_rss_conf.rss_key && bond_rss_conf.rss_key_len <
2803 sizeof(internals->rss_key)) {
2804 if (bond_rss_conf.rss_key_len == 0)
2805 bond_rss_conf.rss_key_len = 40;
2806 internals->rss_key_len = bond_rss_conf.rss_key_len;
2807 memcpy(internals->rss_key, bond_rss_conf.rss_key,
2808 internals->rss_key_len);
2811 for (i = 0; i < internals->slave_count; i++) {
2812 result = rte_eth_dev_rss_hash_update(internals->slaves[i].port_id,
2822 bond_ethdev_rss_hash_conf_get(struct rte_eth_dev *dev,
2823 struct rte_eth_rss_conf *rss_conf)
2825 struct bond_dev_private *internals = dev->data->dev_private;
2827 rss_conf->rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
2828 rss_conf->rss_key_len = internals->rss_key_len;
2829 if (rss_conf->rss_key)
2830 memcpy(rss_conf->rss_key, internals->rss_key, internals->rss_key_len);
2836 bond_ethdev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
2838 struct rte_eth_dev *slave_eth_dev;
2839 struct bond_dev_private *internals = dev->data->dev_private;
2842 rte_spinlock_lock(&internals->lock);
2844 for (i = 0; i < internals->slave_count; i++) {
2845 slave_eth_dev = &rte_eth_devices[internals->slaves[i].port_id];
2846 if (*slave_eth_dev->dev_ops->mtu_set == NULL) {
2847 rte_spinlock_unlock(&internals->lock);
2851 for (i = 0; i < internals->slave_count; i++) {
2852 ret = rte_eth_dev_set_mtu(internals->slaves[i].port_id, mtu);
2854 rte_spinlock_unlock(&internals->lock);
2859 rte_spinlock_unlock(&internals->lock);
2864 bond_ethdev_mac_address_set(struct rte_eth_dev *dev, struct ether_addr *addr)
2866 if (mac_address_set(dev, addr))
2867 RTE_BOND_LOG(ERR, "Failed to update MAC address");
2870 const struct eth_dev_ops default_dev_ops = {
2871 .dev_start = bond_ethdev_start,
2872 .dev_stop = bond_ethdev_stop,
2873 .dev_close = bond_ethdev_close,
2874 .dev_configure = bond_ethdev_configure,
2875 .dev_infos_get = bond_ethdev_info,
2876 .vlan_filter_set = bond_ethdev_vlan_filter_set,
2877 .rx_queue_setup = bond_ethdev_rx_queue_setup,
2878 .tx_queue_setup = bond_ethdev_tx_queue_setup,
2879 .rx_queue_release = bond_ethdev_rx_queue_release,
2880 .tx_queue_release = bond_ethdev_tx_queue_release,
2881 .link_update = bond_ethdev_link_update,
2882 .stats_get = bond_ethdev_stats_get,
2883 .stats_reset = bond_ethdev_stats_reset,
2884 .promiscuous_enable = bond_ethdev_promiscuous_enable,
2885 .promiscuous_disable = bond_ethdev_promiscuous_disable,
2886 .reta_update = bond_ethdev_rss_reta_update,
2887 .reta_query = bond_ethdev_rss_reta_query,
2888 .rss_hash_update = bond_ethdev_rss_hash_update,
2889 .rss_hash_conf_get = bond_ethdev_rss_hash_conf_get,
2890 .mtu_set = bond_ethdev_mtu_set,
2891 .mac_addr_set = bond_ethdev_mac_address_set
2895 bond_alloc(struct rte_vdev_device *dev, uint8_t mode)
2897 const char *name = rte_vdev_device_name(dev);
2898 uint8_t socket_id = dev->device.numa_node;
2899 struct bond_dev_private *internals = NULL;
2900 struct rte_eth_dev *eth_dev = NULL;
2901 uint32_t vlan_filter_bmp_size;
2903 /* now do all data allocation - for eth_dev structure, dummy pci driver
2904 * and internal (private) data
2907 /* reserve an ethdev entry */
2908 eth_dev = rte_eth_vdev_allocate(dev, sizeof(*internals));
2909 if (eth_dev == NULL) {
2910 RTE_BOND_LOG(ERR, "Unable to allocate rte_eth_dev");
2914 internals = eth_dev->data->dev_private;
2915 eth_dev->data->nb_rx_queues = (uint16_t)1;
2916 eth_dev->data->nb_tx_queues = (uint16_t)1;
2918 eth_dev->data->mac_addrs = rte_zmalloc_socket(name, ETHER_ADDR_LEN, 0,
2920 if (eth_dev->data->mac_addrs == NULL) {
2921 RTE_BOND_LOG(ERR, "Unable to malloc mac_addrs");
2925 eth_dev->dev_ops = &default_dev_ops;
2926 eth_dev->data->dev_flags = RTE_ETH_DEV_INTR_LSC;
2928 rte_spinlock_init(&internals->lock);
2930 internals->port_id = eth_dev->data->port_id;
2931 internals->mode = BONDING_MODE_INVALID;
2932 internals->current_primary_port = RTE_MAX_ETHPORTS + 1;
2933 internals->balance_xmit_policy = BALANCE_XMIT_POLICY_LAYER2;
2934 internals->burst_xmit_hash = burst_xmit_l2_hash;
2935 internals->user_defined_mac = 0;
2937 internals->link_status_polling_enabled = 0;
2939 internals->link_status_polling_interval_ms =
2940 DEFAULT_POLLING_INTERVAL_10_MS;
2941 internals->link_down_delay_ms = 0;
2942 internals->link_up_delay_ms = 0;
2944 internals->slave_count = 0;
2945 internals->active_slave_count = 0;
2946 internals->rx_offload_capa = 0;
2947 internals->tx_offload_capa = 0;
2948 internals->candidate_max_rx_pktlen = 0;
2949 internals->max_rx_pktlen = 0;
2951 /* Initially allow to choose any offload type */
2952 internals->flow_type_rss_offloads = ETH_RSS_PROTO_MASK;
2954 memset(internals->active_slaves, 0, sizeof(internals->active_slaves));
2955 memset(internals->slaves, 0, sizeof(internals->slaves));
2957 /* Set mode 4 default configuration */
2958 bond_mode_8023ad_setup(eth_dev, NULL);
2959 if (bond_ethdev_mode_set(eth_dev, mode)) {
2960 RTE_BOND_LOG(ERR, "Failed to set bonded device %d mode too %d",
2961 eth_dev->data->port_id, mode);
2965 vlan_filter_bmp_size =
2966 rte_bitmap_get_memory_footprint(ETHER_MAX_VLAN_ID + 1);
2967 internals->vlan_filter_bmpmem = rte_malloc(name, vlan_filter_bmp_size,
2968 RTE_CACHE_LINE_SIZE);
2969 if (internals->vlan_filter_bmpmem == NULL) {
2971 "Failed to allocate vlan bitmap for bonded device %u\n",
2972 eth_dev->data->port_id);
2976 internals->vlan_filter_bmp = rte_bitmap_init(ETHER_MAX_VLAN_ID + 1,
2977 internals->vlan_filter_bmpmem, vlan_filter_bmp_size);
2978 if (internals->vlan_filter_bmp == NULL) {
2980 "Failed to init vlan bitmap for bonded device %u\n",
2981 eth_dev->data->port_id);
2982 rte_free(internals->vlan_filter_bmpmem);
2986 return eth_dev->data->port_id;
2989 rte_free(internals);
2990 if (eth_dev != NULL) {
2991 rte_free(eth_dev->data->mac_addrs);
2992 rte_eth_dev_release_port(eth_dev);
2998 bond_probe(struct rte_vdev_device *dev)
3001 struct bond_dev_private *internals;
3002 struct rte_kvargs *kvlist;
3003 uint8_t bonding_mode, socket_id/*, agg_mode*/;
3004 int arg_count, port_id;
3010 name = rte_vdev_device_name(dev);
3011 RTE_LOG(INFO, EAL, "Initializing pmd_bond for %s\n", name);
3013 kvlist = rte_kvargs_parse(rte_vdev_device_args(dev),
3014 pmd_bond_init_valid_arguments);
3018 /* Parse link bonding mode */
3019 if (rte_kvargs_count(kvlist, PMD_BOND_MODE_KVARG) == 1) {
3020 if (rte_kvargs_process(kvlist, PMD_BOND_MODE_KVARG,
3021 &bond_ethdev_parse_slave_mode_kvarg,
3022 &bonding_mode) != 0) {
3023 RTE_LOG(ERR, EAL, "Invalid mode for bonded device %s\n",
3028 RTE_LOG(ERR, EAL, "Mode must be specified only once for bonded "
3029 "device %s\n", name);
3033 /* Parse socket id to create bonding device on */
3034 arg_count = rte_kvargs_count(kvlist, PMD_BOND_SOCKET_ID_KVARG);
3035 if (arg_count == 1) {
3036 if (rte_kvargs_process(kvlist, PMD_BOND_SOCKET_ID_KVARG,
3037 &bond_ethdev_parse_socket_id_kvarg, &socket_id)
3039 RTE_LOG(ERR, EAL, "Invalid socket Id specified for "
3040 "bonded device %s\n", name);
3043 } else if (arg_count > 1) {
3044 RTE_LOG(ERR, EAL, "Socket Id can be specified only once for "
3045 "bonded device %s\n", name);
3048 socket_id = rte_socket_id();
3051 dev->device.numa_node = socket_id;
3053 /* Create link bonding eth device */
3054 port_id = bond_alloc(dev, bonding_mode);
3056 RTE_LOG(ERR, EAL, "Failed to create socket %s in mode %u on "
3057 "socket %u.\n", name, bonding_mode, socket_id);
3060 internals = rte_eth_devices[port_id].data->dev_private;
3061 internals->kvlist = kvlist;
3064 if (rte_kvargs_count(kvlist, PMD_BOND_AGG_MODE_KVARG) == 1) {
3065 if (rte_kvargs_process(kvlist,
3066 PMD_BOND_AGG_MODE_KVARG,
3067 &bond_ethdev_parse_slave_agg_mode_kvarg,
3070 "Failed to parse agg selection mode for bonded device %s\n",
3075 if (internals->mode == BONDING_MODE_8023AD)
3076 rte_eth_bond_8023ad_agg_selection_set(port_id,
3079 rte_eth_bond_8023ad_agg_selection_set(port_id, AGG_STABLE);
3082 RTE_LOG(INFO, EAL, "Create bonded device %s on port %d in mode %u on "
3083 "socket %u.\n", name, port_id, bonding_mode, socket_id);
3087 rte_kvargs_free(kvlist);
3093 bond_remove(struct rte_vdev_device *dev)
3095 struct rte_eth_dev *eth_dev;
3096 struct bond_dev_private *internals;
3102 name = rte_vdev_device_name(dev);
3103 RTE_LOG(INFO, EAL, "Uninitializing pmd_bond for %s\n", name);
3105 /* now free all data allocation - for eth_dev structure,
3106 * dummy pci driver and internal (private) data
3109 /* find an ethdev entry */
3110 eth_dev = rte_eth_dev_allocated(name);
3111 if (eth_dev == NULL)
3114 RTE_ASSERT(eth_dev->device == &dev->device);
3116 internals = eth_dev->data->dev_private;
3117 if (internals->slave_count != 0)
3120 if (eth_dev->data->dev_started == 1) {
3121 bond_ethdev_stop(eth_dev);
3122 bond_ethdev_close(eth_dev);
3125 eth_dev->dev_ops = NULL;
3126 eth_dev->rx_pkt_burst = NULL;
3127 eth_dev->tx_pkt_burst = NULL;
3129 internals = eth_dev->data->dev_private;
3130 rte_bitmap_free(internals->vlan_filter_bmp);
3131 rte_free(internals->vlan_filter_bmpmem);
3132 rte_free(eth_dev->data->dev_private);
3133 rte_free(eth_dev->data->mac_addrs);
3135 rte_eth_dev_release_port(eth_dev);
3140 /* this part will resolve the slave portids after all the other pdev and vdev
3141 * have been allocated */
3143 bond_ethdev_configure(struct rte_eth_dev *dev)
3145 const char *name = dev->device->name;
3146 struct bond_dev_private *internals = dev->data->dev_private;
3147 struct rte_kvargs *kvlist = internals->kvlist;
3149 uint16_t port_id = dev - rte_eth_devices;
3152 static const uint8_t default_rss_key[40] = {
3153 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2, 0x41, 0x67, 0x25, 0x3D,
3154 0x43, 0xA3, 0x8F, 0xB0, 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
3155 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C, 0x6A, 0x42, 0xB7, 0x3B,
3156 0xBE, 0xAC, 0x01, 0xFA
3161 /* If RSS is enabled, fill table and key with default values */
3162 if (dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS) {
3163 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key = internals->rss_key;
3164 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len = 0;
3165 memcpy(internals->rss_key, default_rss_key, 40);
3167 for (i = 0; i < RTE_DIM(internals->reta_conf); i++) {
3168 internals->reta_conf[i].mask = ~0LL;
3169 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
3170 internals->reta_conf[i].reta[j] = j % dev->data->nb_rx_queues;
3174 /* set the max_rx_pktlen */
3175 internals->max_rx_pktlen = internals->candidate_max_rx_pktlen;
3178 * if no kvlist, it means that this bonded device has been created
3179 * through the bonding api.
3184 /* Parse MAC address for bonded device */
3185 arg_count = rte_kvargs_count(kvlist, PMD_BOND_MAC_ADDR_KVARG);
3186 if (arg_count == 1) {
3187 struct ether_addr bond_mac;
3189 if (rte_kvargs_process(kvlist, PMD_BOND_MAC_ADDR_KVARG,
3190 &bond_ethdev_parse_bond_mac_addr_kvarg, &bond_mac) < 0) {
3191 RTE_LOG(INFO, EAL, "Invalid mac address for bonded device %s\n",
3196 /* Set MAC address */
3197 if (rte_eth_bond_mac_address_set(port_id, &bond_mac) != 0) {
3199 "Failed to set mac address on bonded device %s\n",
3203 } else if (arg_count > 1) {
3205 "MAC address can be specified only once for bonded device %s\n",
3210 /* Parse/set balance mode transmit policy */
3211 arg_count = rte_kvargs_count(kvlist, PMD_BOND_XMIT_POLICY_KVARG);
3212 if (arg_count == 1) {
3213 uint8_t xmit_policy;
3215 if (rte_kvargs_process(kvlist, PMD_BOND_XMIT_POLICY_KVARG,
3216 &bond_ethdev_parse_balance_xmit_policy_kvarg, &xmit_policy) !=
3219 "Invalid xmit policy specified for bonded device %s\n",
3224 /* Set balance mode transmit policy*/
3225 if (rte_eth_bond_xmit_policy_set(port_id, xmit_policy) != 0) {
3227 "Failed to set balance xmit policy on bonded device %s\n",
3231 } else if (arg_count > 1) {
3233 "Transmit policy can be specified only once for bonded device"
3238 if (rte_kvargs_count(kvlist, PMD_BOND_AGG_MODE_KVARG) == 1) {
3239 if (rte_kvargs_process(kvlist,
3240 PMD_BOND_AGG_MODE_KVARG,
3241 &bond_ethdev_parse_slave_agg_mode_kvarg,
3244 "Failed to parse agg selection mode for bonded device %s\n",
3247 if (internals->mode == BONDING_MODE_8023AD)
3248 rte_eth_bond_8023ad_agg_selection_set(port_id,
3252 /* Parse/add slave ports to bonded device */
3253 if (rte_kvargs_count(kvlist, PMD_BOND_SLAVE_PORT_KVARG) > 0) {
3254 struct bond_ethdev_slave_ports slave_ports;
3257 memset(&slave_ports, 0, sizeof(slave_ports));
3259 if (rte_kvargs_process(kvlist, PMD_BOND_SLAVE_PORT_KVARG,
3260 &bond_ethdev_parse_slave_port_kvarg, &slave_ports) != 0) {
3262 "Failed to parse slave ports for bonded device %s\n",
3267 for (i = 0; i < slave_ports.slave_count; i++) {
3268 if (rte_eth_bond_slave_add(port_id, slave_ports.slaves[i]) != 0) {
3270 "Failed to add port %d as slave to bonded device %s\n",
3271 slave_ports.slaves[i], name);
3276 RTE_LOG(INFO, EAL, "No slaves specified for bonded device %s\n", name);
3280 /* Parse/set primary slave port id*/
3281 arg_count = rte_kvargs_count(kvlist, PMD_BOND_PRIMARY_SLAVE_KVARG);
3282 if (arg_count == 1) {
3283 uint16_t primary_slave_port_id;
3285 if (rte_kvargs_process(kvlist,
3286 PMD_BOND_PRIMARY_SLAVE_KVARG,
3287 &bond_ethdev_parse_primary_slave_port_id_kvarg,
3288 &primary_slave_port_id) < 0) {
3290 "Invalid primary slave port id specified for bonded device"
3295 /* Set balance mode transmit policy*/
3296 if (rte_eth_bond_primary_set(port_id, primary_slave_port_id)
3299 "Failed to set primary slave port %d on bonded device %s\n",
3300 primary_slave_port_id, name);
3303 } else if (arg_count > 1) {
3305 "Primary slave can be specified only once for bonded device"
3310 /* Parse link status monitor polling interval */
3311 arg_count = rte_kvargs_count(kvlist, PMD_BOND_LSC_POLL_PERIOD_KVARG);
3312 if (arg_count == 1) {
3313 uint32_t lsc_poll_interval_ms;
3315 if (rte_kvargs_process(kvlist,
3316 PMD_BOND_LSC_POLL_PERIOD_KVARG,
3317 &bond_ethdev_parse_time_ms_kvarg,
3318 &lsc_poll_interval_ms) < 0) {
3320 "Invalid lsc polling interval value specified for bonded"
3321 " device %s\n", name);
3325 if (rte_eth_bond_link_monitoring_set(port_id, lsc_poll_interval_ms)
3328 "Failed to set lsc monitor polling interval (%u ms) on"
3329 " bonded device %s\n", lsc_poll_interval_ms, name);
3332 } else if (arg_count > 1) {
3334 "LSC polling interval can be specified only once for bonded"
3335 " device %s\n", name);
3339 /* Parse link up interrupt propagation delay */
3340 arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_UP_PROP_DELAY_KVARG);
3341 if (arg_count == 1) {
3342 uint32_t link_up_delay_ms;
3344 if (rte_kvargs_process(kvlist,
3345 PMD_BOND_LINK_UP_PROP_DELAY_KVARG,
3346 &bond_ethdev_parse_time_ms_kvarg,
3347 &link_up_delay_ms) < 0) {
3349 "Invalid link up propagation delay value specified for"
3350 " bonded device %s\n", name);
3354 /* Set balance mode transmit policy*/
3355 if (rte_eth_bond_link_up_prop_delay_set(port_id, link_up_delay_ms)
3358 "Failed to set link up propagation delay (%u ms) on bonded"
3359 " device %s\n", link_up_delay_ms, name);
3362 } else if (arg_count > 1) {
3364 "Link up propagation delay can be specified only once for"
3365 " bonded device %s\n", name);
3369 /* Parse link down interrupt propagation delay */
3370 arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG);
3371 if (arg_count == 1) {
3372 uint32_t link_down_delay_ms;
3374 if (rte_kvargs_process(kvlist,
3375 PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG,
3376 &bond_ethdev_parse_time_ms_kvarg,
3377 &link_down_delay_ms) < 0) {
3379 "Invalid link down propagation delay value specified for"
3380 " bonded device %s\n", name);
3384 /* Set balance mode transmit policy*/
3385 if (rte_eth_bond_link_down_prop_delay_set(port_id, link_down_delay_ms)
3388 "Failed to set link down propagation delay (%u ms) on"
3389 " bonded device %s\n", link_down_delay_ms, name);
3392 } else if (arg_count > 1) {
3394 "Link down propagation delay can be specified only once for"
3395 " bonded device %s\n", name);
3402 struct rte_vdev_driver pmd_bond_drv = {
3403 .probe = bond_probe,
3404 .remove = bond_remove,
3407 RTE_PMD_REGISTER_VDEV(net_bonding, pmd_bond_drv);
3408 RTE_PMD_REGISTER_ALIAS(net_bonding, eth_bond);
3410 RTE_PMD_REGISTER_PARAM_STRING(net_bonding,
3414 "xmit_policy=[l2 | l23 | l34] "
3415 "agg_mode=[count | stable | bandwidth] "
3418 "lsc_poll_period_ms=<int> "
3420 "down_delay=<int>");