net/bonding: fix leak on remove
[dpdk.git] / drivers / net / bonding / rte_eth_bond_pmd.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2017 Intel Corporation
3  */
4 #include <stdlib.h>
5 #include <stdbool.h>
6 #include <netinet/in.h>
7
8 #include <rte_mbuf.h>
9 #include <rte_malloc.h>
10 #include <ethdev_driver.h>
11 #include <ethdev_vdev.h>
12 #include <rte_tcp.h>
13 #include <rte_udp.h>
14 #include <rte_ip.h>
15 #include <rte_ip_frag.h>
16 #include <rte_devargs.h>
17 #include <rte_kvargs.h>
18 #include <rte_bus_vdev.h>
19 #include <rte_alarm.h>
20 #include <rte_cycles.h>
21 #include <rte_string_fns.h>
22
23 #include "rte_eth_bond.h"
24 #include "eth_bond_private.h"
25 #include "eth_bond_8023ad_private.h"
26
27 #define REORDER_PERIOD_MS 10
28 #define DEFAULT_POLLING_INTERVAL_10_MS (10)
29 #define BOND_MAX_MAC_ADDRS 16
30
31 #define HASH_L4_PORTS(h) ((h)->src_port ^ (h)->dst_port)
32
33 /* Table for statistics in mode 5 TLB */
34 static uint64_t tlb_last_obytets[RTE_MAX_ETHPORTS];
35
36 static inline size_t
37 get_vlan_offset(struct rte_ether_hdr *eth_hdr, uint16_t *proto)
38 {
39         size_t vlan_offset = 0;
40
41         if (rte_cpu_to_be_16(RTE_ETHER_TYPE_VLAN) == *proto ||
42                 rte_cpu_to_be_16(RTE_ETHER_TYPE_QINQ) == *proto) {
43                 struct rte_vlan_hdr *vlan_hdr =
44                         (struct rte_vlan_hdr *)(eth_hdr + 1);
45
46                 vlan_offset = sizeof(struct rte_vlan_hdr);
47                 *proto = vlan_hdr->eth_proto;
48
49                 if (rte_cpu_to_be_16(RTE_ETHER_TYPE_VLAN) == *proto) {
50                         vlan_hdr = vlan_hdr + 1;
51                         *proto = vlan_hdr->eth_proto;
52                         vlan_offset += sizeof(struct rte_vlan_hdr);
53                 }
54         }
55         return vlan_offset;
56 }
57
58 static uint16_t
59 bond_ethdev_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
60 {
61         struct bond_dev_private *internals;
62
63         uint16_t num_rx_total = 0;
64         uint16_t slave_count;
65         uint16_t active_slave;
66         int i;
67
68         /* Cast to structure, containing bonded device's port id and queue id */
69         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
70         internals = bd_rx_q->dev_private;
71         slave_count = internals->active_slave_count;
72         active_slave = bd_rx_q->active_slave;
73
74         for (i = 0; i < slave_count && nb_pkts; i++) {
75                 uint16_t num_rx_slave;
76
77                 /* Offset of pointer to *bufs increases as packets are received
78                  * from other slaves */
79                 num_rx_slave =
80                         rte_eth_rx_burst(internals->active_slaves[active_slave],
81                                          bd_rx_q->queue_id,
82                                          bufs + num_rx_total, nb_pkts);
83                 num_rx_total += num_rx_slave;
84                 nb_pkts -= num_rx_slave;
85                 if (++active_slave == slave_count)
86                         active_slave = 0;
87         }
88
89         if (++bd_rx_q->active_slave >= slave_count)
90                 bd_rx_q->active_slave = 0;
91         return num_rx_total;
92 }
93
94 static uint16_t
95 bond_ethdev_rx_burst_active_backup(void *queue, struct rte_mbuf **bufs,
96                 uint16_t nb_pkts)
97 {
98         struct bond_dev_private *internals;
99
100         /* Cast to structure, containing bonded device's port id and queue id */
101         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
102
103         internals = bd_rx_q->dev_private;
104
105         return rte_eth_rx_burst(internals->current_primary_port,
106                         bd_rx_q->queue_id, bufs, nb_pkts);
107 }
108
109 static inline uint8_t
110 is_lacp_packets(uint16_t ethertype, uint8_t subtype, struct rte_mbuf *mbuf)
111 {
112         const uint16_t ether_type_slow_be =
113                 rte_be_to_cpu_16(RTE_ETHER_TYPE_SLOW);
114
115         return !((mbuf->ol_flags & PKT_RX_VLAN) ? mbuf->vlan_tci : 0) &&
116                 (ethertype == ether_type_slow_be &&
117                 (subtype == SLOW_SUBTYPE_MARKER || subtype == SLOW_SUBTYPE_LACP));
118 }
119
120 /*****************************************************************************
121  * Flow director's setup for mode 4 optimization
122  */
123
124 static struct rte_flow_item_eth flow_item_eth_type_8023ad = {
125         .dst.addr_bytes = { 0 },
126         .src.addr_bytes = { 0 },
127         .type = RTE_BE16(RTE_ETHER_TYPE_SLOW),
128 };
129
130 static struct rte_flow_item_eth flow_item_eth_mask_type_8023ad = {
131         .dst.addr_bytes = { 0 },
132         .src.addr_bytes = { 0 },
133         .type = 0xFFFF,
134 };
135
136 static struct rte_flow_item flow_item_8023ad[] = {
137         {
138                 .type = RTE_FLOW_ITEM_TYPE_ETH,
139                 .spec = &flow_item_eth_type_8023ad,
140                 .last = NULL,
141                 .mask = &flow_item_eth_mask_type_8023ad,
142         },
143         {
144                 .type = RTE_FLOW_ITEM_TYPE_END,
145                 .spec = NULL,
146                 .last = NULL,
147                 .mask = NULL,
148         }
149 };
150
151 const struct rte_flow_attr flow_attr_8023ad = {
152         .group = 0,
153         .priority = 0,
154         .ingress = 1,
155         .egress = 0,
156         .reserved = 0,
157 };
158
159 int
160 bond_ethdev_8023ad_flow_verify(struct rte_eth_dev *bond_dev,
161                 uint16_t slave_port) {
162         struct rte_eth_dev_info slave_info;
163         struct rte_flow_error error;
164         struct bond_dev_private *internals = bond_dev->data->dev_private;
165
166         const struct rte_flow_action_queue lacp_queue_conf = {
167                 .index = 0,
168         };
169
170         const struct rte_flow_action actions[] = {
171                 {
172                         .type = RTE_FLOW_ACTION_TYPE_QUEUE,
173                         .conf = &lacp_queue_conf
174                 },
175                 {
176                         .type = RTE_FLOW_ACTION_TYPE_END,
177                 }
178         };
179
180         int ret = rte_flow_validate(slave_port, &flow_attr_8023ad,
181                         flow_item_8023ad, actions, &error);
182         if (ret < 0) {
183                 RTE_BOND_LOG(ERR, "%s: %s (slave_port=%d queue_id=%d)",
184                                 __func__, error.message, slave_port,
185                                 internals->mode4.dedicated_queues.rx_qid);
186                 return -1;
187         }
188
189         ret = rte_eth_dev_info_get(slave_port, &slave_info);
190         if (ret != 0) {
191                 RTE_BOND_LOG(ERR,
192                         "%s: Error during getting device (port %u) info: %s\n",
193                         __func__, slave_port, strerror(-ret));
194
195                 return ret;
196         }
197
198         if (slave_info.max_rx_queues < bond_dev->data->nb_rx_queues ||
199                         slave_info.max_tx_queues < bond_dev->data->nb_tx_queues) {
200                 RTE_BOND_LOG(ERR,
201                         "%s: Slave %d capabilities doesn't allow to allocate additional queues",
202                         __func__, slave_port);
203                 return -1;
204         }
205
206         return 0;
207 }
208
209 int
210 bond_8023ad_slow_pkt_hw_filter_supported(uint16_t port_id) {
211         struct rte_eth_dev *bond_dev = &rte_eth_devices[port_id];
212         struct bond_dev_private *internals = bond_dev->data->dev_private;
213         struct rte_eth_dev_info bond_info;
214         uint16_t idx;
215         int ret;
216
217         /* Verify if all slaves in bonding supports flow director and */
218         if (internals->slave_count > 0) {
219                 ret = rte_eth_dev_info_get(bond_dev->data->port_id, &bond_info);
220                 if (ret != 0) {
221                         RTE_BOND_LOG(ERR,
222                                 "%s: Error during getting device (port %u) info: %s\n",
223                                 __func__, bond_dev->data->port_id,
224                                 strerror(-ret));
225
226                         return ret;
227                 }
228
229                 internals->mode4.dedicated_queues.rx_qid = bond_info.nb_rx_queues;
230                 internals->mode4.dedicated_queues.tx_qid = bond_info.nb_tx_queues;
231
232                 for (idx = 0; idx < internals->slave_count; idx++) {
233                         if (bond_ethdev_8023ad_flow_verify(bond_dev,
234                                         internals->slaves[idx].port_id) != 0)
235                                 return -1;
236                 }
237         }
238
239         return 0;
240 }
241
242 int
243 bond_ethdev_8023ad_flow_set(struct rte_eth_dev *bond_dev, uint16_t slave_port) {
244
245         struct rte_flow_error error;
246         struct bond_dev_private *internals = bond_dev->data->dev_private;
247         struct rte_flow_action_queue lacp_queue_conf = {
248                 .index = internals->mode4.dedicated_queues.rx_qid,
249         };
250
251         const struct rte_flow_action actions[] = {
252                 {
253                         .type = RTE_FLOW_ACTION_TYPE_QUEUE,
254                         .conf = &lacp_queue_conf
255                 },
256                 {
257                         .type = RTE_FLOW_ACTION_TYPE_END,
258                 }
259         };
260
261         internals->mode4.dedicated_queues.flow[slave_port] = rte_flow_create(slave_port,
262                         &flow_attr_8023ad, flow_item_8023ad, actions, &error);
263         if (internals->mode4.dedicated_queues.flow[slave_port] == NULL) {
264                 RTE_BOND_LOG(ERR, "bond_ethdev_8023ad_flow_set: %s "
265                                 "(slave_port=%d queue_id=%d)",
266                                 error.message, slave_port,
267                                 internals->mode4.dedicated_queues.rx_qid);
268                 return -1;
269         }
270
271         return 0;
272 }
273
274 static inline uint16_t
275 rx_burst_8023ad(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts,
276                 bool dedicated_rxq)
277 {
278         /* Cast to structure, containing bonded device's port id and queue id */
279         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
280         struct bond_dev_private *internals = bd_rx_q->dev_private;
281         struct rte_eth_dev *bonded_eth_dev =
282                                         &rte_eth_devices[internals->port_id];
283         struct rte_ether_addr *bond_mac = bonded_eth_dev->data->mac_addrs;
284         struct rte_ether_hdr *hdr;
285
286         const uint16_t ether_type_slow_be =
287                 rte_be_to_cpu_16(RTE_ETHER_TYPE_SLOW);
288         uint16_t num_rx_total = 0;      /* Total number of received packets */
289         uint16_t slaves[RTE_MAX_ETHPORTS];
290         uint16_t slave_count, idx;
291
292         uint8_t collecting;  /* current slave collecting status */
293         const uint8_t promisc = rte_eth_promiscuous_get(internals->port_id);
294         const uint8_t allmulti = rte_eth_allmulticast_get(internals->port_id);
295         uint8_t subtype;
296         uint16_t i;
297         uint16_t j;
298         uint16_t k;
299
300         /* Copy slave list to protect against slave up/down changes during tx
301          * bursting */
302         slave_count = internals->active_slave_count;
303         memcpy(slaves, internals->active_slaves,
304                         sizeof(internals->active_slaves[0]) * slave_count);
305
306         idx = bd_rx_q->active_slave;
307         if (idx >= slave_count) {
308                 bd_rx_q->active_slave = 0;
309                 idx = 0;
310         }
311         for (i = 0; i < slave_count && num_rx_total < nb_pkts; i++) {
312                 j = num_rx_total;
313                 collecting = ACTOR_STATE(&bond_mode_8023ad_ports[slaves[idx]],
314                                          COLLECTING);
315
316                 /* Read packets from this slave */
317                 num_rx_total += rte_eth_rx_burst(slaves[idx], bd_rx_q->queue_id,
318                                 &bufs[num_rx_total], nb_pkts - num_rx_total);
319
320                 for (k = j; k < 2 && k < num_rx_total; k++)
321                         rte_prefetch0(rte_pktmbuf_mtod(bufs[k], void *));
322
323                 /* Handle slow protocol packets. */
324                 while (j < num_rx_total) {
325                         if (j + 3 < num_rx_total)
326                                 rte_prefetch0(rte_pktmbuf_mtod(bufs[j + 3], void *));
327
328                         hdr = rte_pktmbuf_mtod(bufs[j], struct rte_ether_hdr *);
329                         subtype = ((struct slow_protocol_frame *)hdr)->slow_protocol.subtype;
330
331                         /* Remove packet from array if:
332                          * - it is slow packet but no dedicated rxq is present,
333                          * - slave is not in collecting state,
334                          * - bonding interface is not in promiscuous mode:
335                          *   - packet is unicast and address does not match,
336                          *   - packet is multicast and bonding interface
337                          *     is not in allmulti,
338                          */
339                         if (unlikely(
340                                 (!dedicated_rxq &&
341                                  is_lacp_packets(hdr->ether_type, subtype,
342                                                  bufs[j])) ||
343                                 !collecting ||
344                                 (!promisc &&
345                                  ((rte_is_unicast_ether_addr(&hdr->d_addr) &&
346                                    !rte_is_same_ether_addr(bond_mac,
347                                                        &hdr->d_addr)) ||
348                                   (!allmulti &&
349                                    rte_is_multicast_ether_addr(&hdr->d_addr)))))) {
350
351                                 if (hdr->ether_type == ether_type_slow_be) {
352                                         bond_mode_8023ad_handle_slow_pkt(
353                                             internals, slaves[idx], bufs[j]);
354                                 } else
355                                         rte_pktmbuf_free(bufs[j]);
356
357                                 /* Packet is managed by mode 4 or dropped, shift the array */
358                                 num_rx_total--;
359                                 if (j < num_rx_total) {
360                                         memmove(&bufs[j], &bufs[j + 1], sizeof(bufs[0]) *
361                                                 (num_rx_total - j));
362                                 }
363                         } else
364                                 j++;
365                 }
366                 if (unlikely(++idx == slave_count))
367                         idx = 0;
368         }
369
370         if (++bd_rx_q->active_slave >= slave_count)
371                 bd_rx_q->active_slave = 0;
372
373         return num_rx_total;
374 }
375
376 static uint16_t
377 bond_ethdev_rx_burst_8023ad(void *queue, struct rte_mbuf **bufs,
378                 uint16_t nb_pkts)
379 {
380         return rx_burst_8023ad(queue, bufs, nb_pkts, false);
381 }
382
383 static uint16_t
384 bond_ethdev_rx_burst_8023ad_fast_queue(void *queue, struct rte_mbuf **bufs,
385                 uint16_t nb_pkts)
386 {
387         return rx_burst_8023ad(queue, bufs, nb_pkts, true);
388 }
389
390 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
391 uint32_t burstnumberRX;
392 uint32_t burstnumberTX;
393
394 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
395
396 static void
397 arp_op_name(uint16_t arp_op, char *buf, size_t buf_len)
398 {
399         switch (arp_op) {
400         case RTE_ARP_OP_REQUEST:
401                 strlcpy(buf, "ARP Request", buf_len);
402                 return;
403         case RTE_ARP_OP_REPLY:
404                 strlcpy(buf, "ARP Reply", buf_len);
405                 return;
406         case RTE_ARP_OP_REVREQUEST:
407                 strlcpy(buf, "Reverse ARP Request", buf_len);
408                 return;
409         case RTE_ARP_OP_REVREPLY:
410                 strlcpy(buf, "Reverse ARP Reply", buf_len);
411                 return;
412         case RTE_ARP_OP_INVREQUEST:
413                 strlcpy(buf, "Peer Identify Request", buf_len);
414                 return;
415         case RTE_ARP_OP_INVREPLY:
416                 strlcpy(buf, "Peer Identify Reply", buf_len);
417                 return;
418         default:
419                 break;
420         }
421         strlcpy(buf, "Unknown", buf_len);
422         return;
423 }
424 #endif
425 #define MaxIPv4String   16
426 static void
427 ipv4_addr_to_dot(uint32_t be_ipv4_addr, char *buf, uint8_t buf_size)
428 {
429         uint32_t ipv4_addr;
430
431         ipv4_addr = rte_be_to_cpu_32(be_ipv4_addr);
432         snprintf(buf, buf_size, "%d.%d.%d.%d", (ipv4_addr >> 24) & 0xFF,
433                 (ipv4_addr >> 16) & 0xFF, (ipv4_addr >> 8) & 0xFF,
434                 ipv4_addr & 0xFF);
435 }
436
437 #define MAX_CLIENTS_NUMBER      128
438 uint8_t active_clients;
439 struct client_stats_t {
440         uint16_t port;
441         uint32_t ipv4_addr;
442         uint32_t ipv4_rx_packets;
443         uint32_t ipv4_tx_packets;
444 };
445 struct client_stats_t client_stats[MAX_CLIENTS_NUMBER];
446
447 static void
448 update_client_stats(uint32_t addr, uint16_t port, uint32_t *TXorRXindicator)
449 {
450         int i = 0;
451
452         for (; i < MAX_CLIENTS_NUMBER; i++)     {
453                 if ((client_stats[i].ipv4_addr == addr) && (client_stats[i].port == port))      {
454                         /* Just update RX packets number for this client */
455                         if (TXorRXindicator == &burstnumberRX)
456                                 client_stats[i].ipv4_rx_packets++;
457                         else
458                                 client_stats[i].ipv4_tx_packets++;
459                         return;
460                 }
461         }
462         /* We have a new client. Insert him to the table, and increment stats */
463         if (TXorRXindicator == &burstnumberRX)
464                 client_stats[active_clients].ipv4_rx_packets++;
465         else
466                 client_stats[active_clients].ipv4_tx_packets++;
467         client_stats[active_clients].ipv4_addr = addr;
468         client_stats[active_clients].port = port;
469         active_clients++;
470
471 }
472
473 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
474 #define MODE6_DEBUG(info, src_ip, dst_ip, eth_h, arp_op, port, burstnumber) \
475         rte_log(RTE_LOG_DEBUG, bond_logtype,                            \
476                 "%s port:%d SrcMAC:%02X:%02X:%02X:%02X:%02X:%02X SrcIP:%s " \
477                 "DstMAC:%02X:%02X:%02X:%02X:%02X:%02X DstIP:%s %s %d\n", \
478                 info,                                                   \
479                 port,                                                   \
480                 eth_h->s_addr.addr_bytes[0], eth_h->s_addr.addr_bytes[1], \
481                 eth_h->s_addr.addr_bytes[2], eth_h->s_addr.addr_bytes[3], \
482                 eth_h->s_addr.addr_bytes[4], eth_h->s_addr.addr_bytes[5], \
483                 src_ip,                                                 \
484                 eth_h->d_addr.addr_bytes[0], eth_h->d_addr.addr_bytes[1], \
485                 eth_h->d_addr.addr_bytes[2], eth_h->d_addr.addr_bytes[3], \
486                 eth_h->d_addr.addr_bytes[4], eth_h->d_addr.addr_bytes[5], \
487                 dst_ip,                                                 \
488                 arp_op, ++burstnumber)
489 #endif
490
491 static void
492 mode6_debug(const char __rte_unused *info,
493         struct rte_ether_hdr *eth_h, uint16_t port,
494         uint32_t __rte_unused *burstnumber)
495 {
496         struct rte_ipv4_hdr *ipv4_h;
497 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
498         struct rte_arp_hdr *arp_h;
499         char dst_ip[16];
500         char ArpOp[24];
501         char buf[16];
502 #endif
503         char src_ip[16];
504
505         uint16_t ether_type = eth_h->ether_type;
506         uint16_t offset = get_vlan_offset(eth_h, &ether_type);
507
508 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
509         strlcpy(buf, info, 16);
510 #endif
511
512         if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
513                 ipv4_h = (struct rte_ipv4_hdr *)((char *)(eth_h + 1) + offset);
514                 ipv4_addr_to_dot(ipv4_h->src_addr, src_ip, MaxIPv4String);
515 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
516                 ipv4_addr_to_dot(ipv4_h->dst_addr, dst_ip, MaxIPv4String);
517                 MODE6_DEBUG(buf, src_ip, dst_ip, eth_h, "", port, *burstnumber);
518 #endif
519                 update_client_stats(ipv4_h->src_addr, port, burstnumber);
520         }
521 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
522         else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_ARP)) {
523                 arp_h = (struct rte_arp_hdr *)((char *)(eth_h + 1) + offset);
524                 ipv4_addr_to_dot(arp_h->arp_data.arp_sip, src_ip, MaxIPv4String);
525                 ipv4_addr_to_dot(arp_h->arp_data.arp_tip, dst_ip, MaxIPv4String);
526                 arp_op_name(rte_be_to_cpu_16(arp_h->arp_opcode),
527                                 ArpOp, sizeof(ArpOp));
528                 MODE6_DEBUG(buf, src_ip, dst_ip, eth_h, ArpOp, port, *burstnumber);
529         }
530 #endif
531 }
532 #endif
533
534 static uint16_t
535 bond_ethdev_rx_burst_alb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
536 {
537         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
538         struct bond_dev_private *internals = bd_rx_q->dev_private;
539         struct rte_ether_hdr *eth_h;
540         uint16_t ether_type, offset;
541         uint16_t nb_recv_pkts;
542         int i;
543
544         nb_recv_pkts = bond_ethdev_rx_burst(queue, bufs, nb_pkts);
545
546         for (i = 0; i < nb_recv_pkts; i++) {
547                 eth_h = rte_pktmbuf_mtod(bufs[i], struct rte_ether_hdr *);
548                 ether_type = eth_h->ether_type;
549                 offset = get_vlan_offset(eth_h, &ether_type);
550
551                 if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_ARP)) {
552 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
553                         mode6_debug("RX ARP:", eth_h, bufs[i]->port, &burstnumberRX);
554 #endif
555                         bond_mode_alb_arp_recv(eth_h, offset, internals);
556                 }
557 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
558                 else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
559                         mode6_debug("RX IPv4:", eth_h, bufs[i]->port, &burstnumberRX);
560 #endif
561         }
562
563         return nb_recv_pkts;
564 }
565
566 static uint16_t
567 bond_ethdev_tx_burst_round_robin(void *queue, struct rte_mbuf **bufs,
568                 uint16_t nb_pkts)
569 {
570         struct bond_dev_private *internals;
571         struct bond_tx_queue *bd_tx_q;
572
573         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_pkts];
574         uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
575
576         uint16_t num_of_slaves;
577         uint16_t slaves[RTE_MAX_ETHPORTS];
578
579         uint16_t num_tx_total = 0, num_tx_slave;
580
581         static int slave_idx = 0;
582         int i, cslave_idx = 0, tx_fail_total = 0;
583
584         bd_tx_q = (struct bond_tx_queue *)queue;
585         internals = bd_tx_q->dev_private;
586
587         /* Copy slave list to protect against slave up/down changes during tx
588          * bursting */
589         num_of_slaves = internals->active_slave_count;
590         memcpy(slaves, internals->active_slaves,
591                         sizeof(internals->active_slaves[0]) * num_of_slaves);
592
593         if (num_of_slaves < 1)
594                 return num_tx_total;
595
596         /* Populate slaves mbuf with which packets are to be sent on it  */
597         for (i = 0; i < nb_pkts; i++) {
598                 cslave_idx = (slave_idx + i) % num_of_slaves;
599                 slave_bufs[cslave_idx][(slave_nb_pkts[cslave_idx])++] = bufs[i];
600         }
601
602         /* increment current slave index so the next call to tx burst starts on the
603          * next slave */
604         slave_idx = ++cslave_idx;
605
606         /* Send packet burst on each slave device */
607         for (i = 0; i < num_of_slaves; i++) {
608                 if (slave_nb_pkts[i] > 0) {
609                         num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
610                                         slave_bufs[i], slave_nb_pkts[i]);
611
612                         /* if tx burst fails move packets to end of bufs */
613                         if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
614                                 int tx_fail_slave = slave_nb_pkts[i] - num_tx_slave;
615
616                                 tx_fail_total += tx_fail_slave;
617
618                                 memcpy(&bufs[nb_pkts - tx_fail_total],
619                                        &slave_bufs[i][num_tx_slave],
620                                        tx_fail_slave * sizeof(bufs[0]));
621                         }
622                         num_tx_total += num_tx_slave;
623                 }
624         }
625
626         return num_tx_total;
627 }
628
629 static uint16_t
630 bond_ethdev_tx_burst_active_backup(void *queue,
631                 struct rte_mbuf **bufs, uint16_t nb_pkts)
632 {
633         struct bond_dev_private *internals;
634         struct bond_tx_queue *bd_tx_q;
635
636         bd_tx_q = (struct bond_tx_queue *)queue;
637         internals = bd_tx_q->dev_private;
638
639         if (internals->active_slave_count < 1)
640                 return 0;
641
642         return rte_eth_tx_burst(internals->current_primary_port, bd_tx_q->queue_id,
643                         bufs, nb_pkts);
644 }
645
646 static inline uint16_t
647 ether_hash(struct rte_ether_hdr *eth_hdr)
648 {
649         unaligned_uint16_t *word_src_addr =
650                 (unaligned_uint16_t *)eth_hdr->s_addr.addr_bytes;
651         unaligned_uint16_t *word_dst_addr =
652                 (unaligned_uint16_t *)eth_hdr->d_addr.addr_bytes;
653
654         return (word_src_addr[0] ^ word_dst_addr[0]) ^
655                         (word_src_addr[1] ^ word_dst_addr[1]) ^
656                         (word_src_addr[2] ^ word_dst_addr[2]);
657 }
658
659 static inline uint32_t
660 ipv4_hash(struct rte_ipv4_hdr *ipv4_hdr)
661 {
662         return ipv4_hdr->src_addr ^ ipv4_hdr->dst_addr;
663 }
664
665 static inline uint32_t
666 ipv6_hash(struct rte_ipv6_hdr *ipv6_hdr)
667 {
668         unaligned_uint32_t *word_src_addr =
669                 (unaligned_uint32_t *)&(ipv6_hdr->src_addr[0]);
670         unaligned_uint32_t *word_dst_addr =
671                 (unaligned_uint32_t *)&(ipv6_hdr->dst_addr[0]);
672
673         return (word_src_addr[0] ^ word_dst_addr[0]) ^
674                         (word_src_addr[1] ^ word_dst_addr[1]) ^
675                         (word_src_addr[2] ^ word_dst_addr[2]) ^
676                         (word_src_addr[3] ^ word_dst_addr[3]);
677 }
678
679
680 void
681 burst_xmit_l2_hash(struct rte_mbuf **buf, uint16_t nb_pkts,
682                 uint16_t slave_count, uint16_t *slaves)
683 {
684         struct rte_ether_hdr *eth_hdr;
685         uint32_t hash;
686         int i;
687
688         for (i = 0; i < nb_pkts; i++) {
689                 eth_hdr = rte_pktmbuf_mtod(buf[i], struct rte_ether_hdr *);
690
691                 hash = ether_hash(eth_hdr);
692
693                 slaves[i] = (hash ^= hash >> 8) % slave_count;
694         }
695 }
696
697 void
698 burst_xmit_l23_hash(struct rte_mbuf **buf, uint16_t nb_pkts,
699                 uint16_t slave_count, uint16_t *slaves)
700 {
701         uint16_t i;
702         struct rte_ether_hdr *eth_hdr;
703         uint16_t proto;
704         size_t vlan_offset;
705         uint32_t hash, l3hash;
706
707         for (i = 0; i < nb_pkts; i++) {
708                 eth_hdr = rte_pktmbuf_mtod(buf[i], struct rte_ether_hdr *);
709                 l3hash = 0;
710
711                 proto = eth_hdr->ether_type;
712                 hash = ether_hash(eth_hdr);
713
714                 vlan_offset = get_vlan_offset(eth_hdr, &proto);
715
716                 if (rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) == proto) {
717                         struct rte_ipv4_hdr *ipv4_hdr = (struct rte_ipv4_hdr *)
718                                         ((char *)(eth_hdr + 1) + vlan_offset);
719                         l3hash = ipv4_hash(ipv4_hdr);
720
721                 } else if (rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) == proto) {
722                         struct rte_ipv6_hdr *ipv6_hdr = (struct rte_ipv6_hdr *)
723                                         ((char *)(eth_hdr + 1) + vlan_offset);
724                         l3hash = ipv6_hash(ipv6_hdr);
725                 }
726
727                 hash = hash ^ l3hash;
728                 hash ^= hash >> 16;
729                 hash ^= hash >> 8;
730
731                 slaves[i] = hash % slave_count;
732         }
733 }
734
735 void
736 burst_xmit_l34_hash(struct rte_mbuf **buf, uint16_t nb_pkts,
737                 uint16_t slave_count, uint16_t *slaves)
738 {
739         struct rte_ether_hdr *eth_hdr;
740         uint16_t proto;
741         size_t vlan_offset;
742         int i;
743
744         struct rte_udp_hdr *udp_hdr;
745         struct rte_tcp_hdr *tcp_hdr;
746         uint32_t hash, l3hash, l4hash;
747
748         for (i = 0; i < nb_pkts; i++) {
749                 eth_hdr = rte_pktmbuf_mtod(buf[i], struct rte_ether_hdr *);
750                 size_t pkt_end = (size_t)eth_hdr + rte_pktmbuf_data_len(buf[i]);
751                 proto = eth_hdr->ether_type;
752                 vlan_offset = get_vlan_offset(eth_hdr, &proto);
753                 l3hash = 0;
754                 l4hash = 0;
755
756                 if (rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) == proto) {
757                         struct rte_ipv4_hdr *ipv4_hdr = (struct rte_ipv4_hdr *)
758                                         ((char *)(eth_hdr + 1) + vlan_offset);
759                         size_t ip_hdr_offset;
760
761                         l3hash = ipv4_hash(ipv4_hdr);
762
763                         /* there is no L4 header in fragmented packet */
764                         if (likely(rte_ipv4_frag_pkt_is_fragmented(ipv4_hdr)
765                                                                 == 0)) {
766                                 ip_hdr_offset = (ipv4_hdr->version_ihl
767                                         & RTE_IPV4_HDR_IHL_MASK) *
768                                         RTE_IPV4_IHL_MULTIPLIER;
769
770                                 if (ipv4_hdr->next_proto_id == IPPROTO_TCP) {
771                                         tcp_hdr = (struct rte_tcp_hdr *)
772                                                 ((char *)ipv4_hdr +
773                                                         ip_hdr_offset);
774                                         if ((size_t)tcp_hdr + sizeof(*tcp_hdr)
775                                                         < pkt_end)
776                                                 l4hash = HASH_L4_PORTS(tcp_hdr);
777                                 } else if (ipv4_hdr->next_proto_id ==
778                                                                 IPPROTO_UDP) {
779                                         udp_hdr = (struct rte_udp_hdr *)
780                                                 ((char *)ipv4_hdr +
781                                                         ip_hdr_offset);
782                                         if ((size_t)udp_hdr + sizeof(*udp_hdr)
783                                                         < pkt_end)
784                                                 l4hash = HASH_L4_PORTS(udp_hdr);
785                                 }
786                         }
787                 } else if  (rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) == proto) {
788                         struct rte_ipv6_hdr *ipv6_hdr = (struct rte_ipv6_hdr *)
789                                         ((char *)(eth_hdr + 1) + vlan_offset);
790                         l3hash = ipv6_hash(ipv6_hdr);
791
792                         if (ipv6_hdr->proto == IPPROTO_TCP) {
793                                 tcp_hdr = (struct rte_tcp_hdr *)(ipv6_hdr + 1);
794                                 l4hash = HASH_L4_PORTS(tcp_hdr);
795                         } else if (ipv6_hdr->proto == IPPROTO_UDP) {
796                                 udp_hdr = (struct rte_udp_hdr *)(ipv6_hdr + 1);
797                                 l4hash = HASH_L4_PORTS(udp_hdr);
798                         }
799                 }
800
801                 hash = l3hash ^ l4hash;
802                 hash ^= hash >> 16;
803                 hash ^= hash >> 8;
804
805                 slaves[i] = hash % slave_count;
806         }
807 }
808
809 struct bwg_slave {
810         uint64_t bwg_left_int;
811         uint64_t bwg_left_remainder;
812         uint16_t slave;
813 };
814
815 void
816 bond_tlb_activate_slave(struct bond_dev_private *internals) {
817         int i;
818
819         for (i = 0; i < internals->active_slave_count; i++) {
820                 tlb_last_obytets[internals->active_slaves[i]] = 0;
821         }
822 }
823
824 static int
825 bandwidth_cmp(const void *a, const void *b)
826 {
827         const struct bwg_slave *bwg_a = a;
828         const struct bwg_slave *bwg_b = b;
829         int64_t diff = (int64_t)bwg_b->bwg_left_int - (int64_t)bwg_a->bwg_left_int;
830         int64_t diff2 = (int64_t)bwg_b->bwg_left_remainder -
831                         (int64_t)bwg_a->bwg_left_remainder;
832         if (diff > 0)
833                 return 1;
834         else if (diff < 0)
835                 return -1;
836         else if (diff2 > 0)
837                 return 1;
838         else if (diff2 < 0)
839                 return -1;
840         else
841                 return 0;
842 }
843
844 static void
845 bandwidth_left(uint16_t port_id, uint64_t load, uint8_t update_idx,
846                 struct bwg_slave *bwg_slave)
847 {
848         struct rte_eth_link link_status;
849         int ret;
850
851         ret = rte_eth_link_get_nowait(port_id, &link_status);
852         if (ret < 0) {
853                 RTE_BOND_LOG(ERR, "Slave (port %u) link get failed: %s",
854                              port_id, rte_strerror(-ret));
855                 return;
856         }
857         uint64_t link_bwg = link_status.link_speed * 1000000ULL / 8;
858         if (link_bwg == 0)
859                 return;
860         link_bwg = link_bwg * (update_idx+1) * REORDER_PERIOD_MS;
861         bwg_slave->bwg_left_int = (link_bwg - 1000*load) / link_bwg;
862         bwg_slave->bwg_left_remainder = (link_bwg - 1000*load) % link_bwg;
863 }
864
865 static void
866 bond_ethdev_update_tlb_slave_cb(void *arg)
867 {
868         struct bond_dev_private *internals = arg;
869         struct rte_eth_stats slave_stats;
870         struct bwg_slave bwg_array[RTE_MAX_ETHPORTS];
871         uint16_t slave_count;
872         uint64_t tx_bytes;
873
874         uint8_t update_stats = 0;
875         uint16_t slave_id;
876         uint16_t i;
877
878         internals->slave_update_idx++;
879
880
881         if (internals->slave_update_idx >= REORDER_PERIOD_MS)
882                 update_stats = 1;
883
884         for (i = 0; i < internals->active_slave_count; i++) {
885                 slave_id = internals->active_slaves[i];
886                 rte_eth_stats_get(slave_id, &slave_stats);
887                 tx_bytes = slave_stats.obytes - tlb_last_obytets[slave_id];
888                 bandwidth_left(slave_id, tx_bytes,
889                                 internals->slave_update_idx, &bwg_array[i]);
890                 bwg_array[i].slave = slave_id;
891
892                 if (update_stats) {
893                         tlb_last_obytets[slave_id] = slave_stats.obytes;
894                 }
895         }
896
897         if (update_stats == 1)
898                 internals->slave_update_idx = 0;
899
900         slave_count = i;
901         qsort(bwg_array, slave_count, sizeof(bwg_array[0]), bandwidth_cmp);
902         for (i = 0; i < slave_count; i++)
903                 internals->tlb_slaves_order[i] = bwg_array[i].slave;
904
905         rte_eal_alarm_set(REORDER_PERIOD_MS * 1000, bond_ethdev_update_tlb_slave_cb,
906                         (struct bond_dev_private *)internals);
907 }
908
909 static uint16_t
910 bond_ethdev_tx_burst_tlb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
911 {
912         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
913         struct bond_dev_private *internals = bd_tx_q->dev_private;
914
915         struct rte_eth_dev *primary_port =
916                         &rte_eth_devices[internals->primary_port];
917         uint16_t num_tx_total = 0;
918         uint16_t i, j;
919
920         uint16_t num_of_slaves = internals->active_slave_count;
921         uint16_t slaves[RTE_MAX_ETHPORTS];
922
923         struct rte_ether_hdr *ether_hdr;
924         struct rte_ether_addr primary_slave_addr;
925         struct rte_ether_addr active_slave_addr;
926
927         if (num_of_slaves < 1)
928                 return num_tx_total;
929
930         memcpy(slaves, internals->tlb_slaves_order,
931                                 sizeof(internals->tlb_slaves_order[0]) * num_of_slaves);
932
933
934         rte_ether_addr_copy(primary_port->data->mac_addrs, &primary_slave_addr);
935
936         if (nb_pkts > 3) {
937                 for (i = 0; i < 3; i++)
938                         rte_prefetch0(rte_pktmbuf_mtod(bufs[i], void*));
939         }
940
941         for (i = 0; i < num_of_slaves; i++) {
942                 rte_eth_macaddr_get(slaves[i], &active_slave_addr);
943                 for (j = num_tx_total; j < nb_pkts; j++) {
944                         if (j + 3 < nb_pkts)
945                                 rte_prefetch0(rte_pktmbuf_mtod(bufs[j+3], void*));
946
947                         ether_hdr = rte_pktmbuf_mtod(bufs[j],
948                                                 struct rte_ether_hdr *);
949                         if (rte_is_same_ether_addr(&ether_hdr->s_addr,
950                                                         &primary_slave_addr))
951                                 rte_ether_addr_copy(&active_slave_addr,
952                                                 &ether_hdr->s_addr);
953 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
954                                         mode6_debug("TX IPv4:", ether_hdr, slaves[i], &burstnumberTX);
955 #endif
956                 }
957
958                 num_tx_total += rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
959                                 bufs + num_tx_total, nb_pkts - num_tx_total);
960
961                 if (num_tx_total == nb_pkts)
962                         break;
963         }
964
965         return num_tx_total;
966 }
967
968 void
969 bond_tlb_disable(struct bond_dev_private *internals)
970 {
971         rte_eal_alarm_cancel(bond_ethdev_update_tlb_slave_cb, internals);
972 }
973
974 void
975 bond_tlb_enable(struct bond_dev_private *internals)
976 {
977         bond_ethdev_update_tlb_slave_cb(internals);
978 }
979
980 static uint16_t
981 bond_ethdev_tx_burst_alb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
982 {
983         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
984         struct bond_dev_private *internals = bd_tx_q->dev_private;
985
986         struct rte_ether_hdr *eth_h;
987         uint16_t ether_type, offset;
988
989         struct client_data *client_info;
990
991         /*
992          * We create transmit buffers for every slave and one additional to send
993          * through tlb. In worst case every packet will be send on one port.
994          */
995         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS + 1][nb_pkts];
996         uint16_t slave_bufs_pkts[RTE_MAX_ETHPORTS + 1] = { 0 };
997
998         /*
999          * We create separate transmit buffers for update packets as they won't
1000          * be counted in num_tx_total.
1001          */
1002         struct rte_mbuf *update_bufs[RTE_MAX_ETHPORTS][ALB_HASH_TABLE_SIZE];
1003         uint16_t update_bufs_pkts[RTE_MAX_ETHPORTS] = { 0 };
1004
1005         struct rte_mbuf *upd_pkt;
1006         size_t pkt_size;
1007
1008         uint16_t num_send, num_not_send = 0;
1009         uint16_t num_tx_total = 0;
1010         uint16_t slave_idx;
1011
1012         int i, j;
1013
1014         /* Search tx buffer for ARP packets and forward them to alb */
1015         for (i = 0; i < nb_pkts; i++) {
1016                 eth_h = rte_pktmbuf_mtod(bufs[i], struct rte_ether_hdr *);
1017                 ether_type = eth_h->ether_type;
1018                 offset = get_vlan_offset(eth_h, &ether_type);
1019
1020                 if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_ARP)) {
1021                         slave_idx = bond_mode_alb_arp_xmit(eth_h, offset, internals);
1022
1023                         /* Change src mac in eth header */
1024                         rte_eth_macaddr_get(slave_idx, &eth_h->s_addr);
1025
1026                         /* Add packet to slave tx buffer */
1027                         slave_bufs[slave_idx][slave_bufs_pkts[slave_idx]] = bufs[i];
1028                         slave_bufs_pkts[slave_idx]++;
1029                 } else {
1030                         /* If packet is not ARP, send it with TLB policy */
1031                         slave_bufs[RTE_MAX_ETHPORTS][slave_bufs_pkts[RTE_MAX_ETHPORTS]] =
1032                                         bufs[i];
1033                         slave_bufs_pkts[RTE_MAX_ETHPORTS]++;
1034                 }
1035         }
1036
1037         /* Update connected client ARP tables */
1038         if (internals->mode6.ntt) {
1039                 for (i = 0; i < ALB_HASH_TABLE_SIZE; i++) {
1040                         client_info = &internals->mode6.client_table[i];
1041
1042                         if (client_info->in_use) {
1043                                 /* Allocate new packet to send ARP update on current slave */
1044                                 upd_pkt = rte_pktmbuf_alloc(internals->mode6.mempool);
1045                                 if (upd_pkt == NULL) {
1046                                         RTE_BOND_LOG(ERR,
1047                                                      "Failed to allocate ARP packet from pool");
1048                                         continue;
1049                                 }
1050                                 pkt_size = sizeof(struct rte_ether_hdr) +
1051                                         sizeof(struct rte_arp_hdr) +
1052                                         client_info->vlan_count *
1053                                         sizeof(struct rte_vlan_hdr);
1054                                 upd_pkt->data_len = pkt_size;
1055                                 upd_pkt->pkt_len = pkt_size;
1056
1057                                 slave_idx = bond_mode_alb_arp_upd(client_info, upd_pkt,
1058                                                 internals);
1059
1060                                 /* Add packet to update tx buffer */
1061                                 update_bufs[slave_idx][update_bufs_pkts[slave_idx]] = upd_pkt;
1062                                 update_bufs_pkts[slave_idx]++;
1063                         }
1064                 }
1065                 internals->mode6.ntt = 0;
1066         }
1067
1068         /* Send ARP packets on proper slaves */
1069         for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
1070                 if (slave_bufs_pkts[i] > 0) {
1071                         num_send = rte_eth_tx_burst(i, bd_tx_q->queue_id,
1072                                         slave_bufs[i], slave_bufs_pkts[i]);
1073                         for (j = 0; j < slave_bufs_pkts[i] - num_send; j++) {
1074                                 bufs[nb_pkts - 1 - num_not_send - j] =
1075                                                 slave_bufs[i][nb_pkts - 1 - j];
1076                         }
1077
1078                         num_tx_total += num_send;
1079                         num_not_send += slave_bufs_pkts[i] - num_send;
1080
1081 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
1082         /* Print TX stats including update packets */
1083                         for (j = 0; j < slave_bufs_pkts[i]; j++) {
1084                                 eth_h = rte_pktmbuf_mtod(slave_bufs[i][j],
1085                                                         struct rte_ether_hdr *);
1086                                 mode6_debug("TX ARP:", eth_h, i, &burstnumberTX);
1087                         }
1088 #endif
1089                 }
1090         }
1091
1092         /* Send update packets on proper slaves */
1093         for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
1094                 if (update_bufs_pkts[i] > 0) {
1095                         num_send = rte_eth_tx_burst(i, bd_tx_q->queue_id, update_bufs[i],
1096                                         update_bufs_pkts[i]);
1097                         for (j = num_send; j < update_bufs_pkts[i]; j++) {
1098                                 rte_pktmbuf_free(update_bufs[i][j]);
1099                         }
1100 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
1101                         for (j = 0; j < update_bufs_pkts[i]; j++) {
1102                                 eth_h = rte_pktmbuf_mtod(update_bufs[i][j],
1103                                                         struct rte_ether_hdr *);
1104                                 mode6_debug("TX ARPupd:", eth_h, i, &burstnumberTX);
1105                         }
1106 #endif
1107                 }
1108         }
1109
1110         /* Send non-ARP packets using tlb policy */
1111         if (slave_bufs_pkts[RTE_MAX_ETHPORTS] > 0) {
1112                 num_send = bond_ethdev_tx_burst_tlb(queue,
1113                                 slave_bufs[RTE_MAX_ETHPORTS],
1114                                 slave_bufs_pkts[RTE_MAX_ETHPORTS]);
1115
1116                 for (j = 0; j < slave_bufs_pkts[RTE_MAX_ETHPORTS]; j++) {
1117                         bufs[nb_pkts - 1 - num_not_send - j] =
1118                                         slave_bufs[RTE_MAX_ETHPORTS][nb_pkts - 1 - j];
1119                 }
1120
1121                 num_tx_total += num_send;
1122         }
1123
1124         return num_tx_total;
1125 }
1126
1127 static inline uint16_t
1128 tx_burst_balance(void *queue, struct rte_mbuf **bufs, uint16_t nb_bufs,
1129                  uint16_t *slave_port_ids, uint16_t slave_count)
1130 {
1131         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
1132         struct bond_dev_private *internals = bd_tx_q->dev_private;
1133
1134         /* Array to sort mbufs for transmission on each slave into */
1135         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_bufs];
1136         /* Number of mbufs for transmission on each slave */
1137         uint16_t slave_nb_bufs[RTE_MAX_ETHPORTS] = { 0 };
1138         /* Mapping array generated by hash function to map mbufs to slaves */
1139         uint16_t bufs_slave_port_idxs[nb_bufs];
1140
1141         uint16_t slave_tx_count;
1142         uint16_t total_tx_count = 0, total_tx_fail_count = 0;
1143
1144         uint16_t i;
1145
1146         /*
1147          * Populate slaves mbuf with the packets which are to be sent on it
1148          * selecting output slave using hash based on xmit policy
1149          */
1150         internals->burst_xmit_hash(bufs, nb_bufs, slave_count,
1151                         bufs_slave_port_idxs);
1152
1153         for (i = 0; i < nb_bufs; i++) {
1154                 /* Populate slave mbuf arrays with mbufs for that slave. */
1155                 uint16_t slave_idx = bufs_slave_port_idxs[i];
1156
1157                 slave_bufs[slave_idx][slave_nb_bufs[slave_idx]++] = bufs[i];
1158         }
1159
1160         /* Send packet burst on each slave device */
1161         for (i = 0; i < slave_count; i++) {
1162                 if (slave_nb_bufs[i] == 0)
1163                         continue;
1164
1165                 slave_tx_count = rte_eth_tx_burst(slave_port_ids[i],
1166                                 bd_tx_q->queue_id, slave_bufs[i],
1167                                 slave_nb_bufs[i]);
1168
1169                 total_tx_count += slave_tx_count;
1170
1171                 /* If tx burst fails move packets to end of bufs */
1172                 if (unlikely(slave_tx_count < slave_nb_bufs[i])) {
1173                         int slave_tx_fail_count = slave_nb_bufs[i] -
1174                                         slave_tx_count;
1175                         total_tx_fail_count += slave_tx_fail_count;
1176                         memcpy(&bufs[nb_bufs - total_tx_fail_count],
1177                                &slave_bufs[i][slave_tx_count],
1178                                slave_tx_fail_count * sizeof(bufs[0]));
1179                 }
1180         }
1181
1182         return total_tx_count;
1183 }
1184
1185 static uint16_t
1186 bond_ethdev_tx_burst_balance(void *queue, struct rte_mbuf **bufs,
1187                 uint16_t nb_bufs)
1188 {
1189         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
1190         struct bond_dev_private *internals = bd_tx_q->dev_private;
1191
1192         uint16_t slave_port_ids[RTE_MAX_ETHPORTS];
1193         uint16_t slave_count;
1194
1195         if (unlikely(nb_bufs == 0))
1196                 return 0;
1197
1198         /* Copy slave list to protect against slave up/down changes during tx
1199          * bursting
1200          */
1201         slave_count = internals->active_slave_count;
1202         if (unlikely(slave_count < 1))
1203                 return 0;
1204
1205         memcpy(slave_port_ids, internals->active_slaves,
1206                         sizeof(slave_port_ids[0]) * slave_count);
1207         return tx_burst_balance(queue, bufs, nb_bufs, slave_port_ids,
1208                                 slave_count);
1209 }
1210
1211 static inline uint16_t
1212 tx_burst_8023ad(void *queue, struct rte_mbuf **bufs, uint16_t nb_bufs,
1213                 bool dedicated_txq)
1214 {
1215         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
1216         struct bond_dev_private *internals = bd_tx_q->dev_private;
1217
1218         uint16_t slave_port_ids[RTE_MAX_ETHPORTS];
1219         uint16_t slave_count;
1220
1221         uint16_t dist_slave_port_ids[RTE_MAX_ETHPORTS];
1222         uint16_t dist_slave_count;
1223
1224         uint16_t slave_tx_count;
1225
1226         uint16_t i;
1227
1228         /* Copy slave list to protect against slave up/down changes during tx
1229          * bursting */
1230         slave_count = internals->active_slave_count;
1231         if (unlikely(slave_count < 1))
1232                 return 0;
1233
1234         memcpy(slave_port_ids, internals->active_slaves,
1235                         sizeof(slave_port_ids[0]) * slave_count);
1236
1237         if (dedicated_txq)
1238                 goto skip_tx_ring;
1239
1240         /* Check for LACP control packets and send if available */
1241         for (i = 0; i < slave_count; i++) {
1242                 struct port *port = &bond_mode_8023ad_ports[slave_port_ids[i]];
1243                 struct rte_mbuf *ctrl_pkt = NULL;
1244
1245                 if (likely(rte_ring_empty(port->tx_ring)))
1246                         continue;
1247
1248                 if (rte_ring_dequeue(port->tx_ring,
1249                                      (void **)&ctrl_pkt) != -ENOENT) {
1250                         slave_tx_count = rte_eth_tx_burst(slave_port_ids[i],
1251                                         bd_tx_q->queue_id, &ctrl_pkt, 1);
1252                         /*
1253                          * re-enqueue LAG control plane packets to buffering
1254                          * ring if transmission fails so the packet isn't lost.
1255                          */
1256                         if (slave_tx_count != 1)
1257                                 rte_ring_enqueue(port->tx_ring, ctrl_pkt);
1258                 }
1259         }
1260
1261 skip_tx_ring:
1262         if (unlikely(nb_bufs == 0))
1263                 return 0;
1264
1265         dist_slave_count = 0;
1266         for (i = 0; i < slave_count; i++) {
1267                 struct port *port = &bond_mode_8023ad_ports[slave_port_ids[i]];
1268
1269                 if (ACTOR_STATE(port, DISTRIBUTING))
1270                         dist_slave_port_ids[dist_slave_count++] =
1271                                         slave_port_ids[i];
1272         }
1273
1274         if (unlikely(dist_slave_count < 1))
1275                 return 0;
1276
1277         return tx_burst_balance(queue, bufs, nb_bufs, dist_slave_port_ids,
1278                                 dist_slave_count);
1279 }
1280
1281 static uint16_t
1282 bond_ethdev_tx_burst_8023ad(void *queue, struct rte_mbuf **bufs,
1283                 uint16_t nb_bufs)
1284 {
1285         return tx_burst_8023ad(queue, bufs, nb_bufs, false);
1286 }
1287
1288 static uint16_t
1289 bond_ethdev_tx_burst_8023ad_fast_queue(void *queue, struct rte_mbuf **bufs,
1290                 uint16_t nb_bufs)
1291 {
1292         return tx_burst_8023ad(queue, bufs, nb_bufs, true);
1293 }
1294
1295 static uint16_t
1296 bond_ethdev_tx_burst_broadcast(void *queue, struct rte_mbuf **bufs,
1297                 uint16_t nb_pkts)
1298 {
1299         struct bond_dev_private *internals;
1300         struct bond_tx_queue *bd_tx_q;
1301
1302         uint16_t slaves[RTE_MAX_ETHPORTS];
1303         uint8_t tx_failed_flag = 0;
1304         uint16_t num_of_slaves;
1305
1306         uint16_t max_nb_of_tx_pkts = 0;
1307
1308         int slave_tx_total[RTE_MAX_ETHPORTS];
1309         int i, most_successful_tx_slave = -1;
1310
1311         bd_tx_q = (struct bond_tx_queue *)queue;
1312         internals = bd_tx_q->dev_private;
1313
1314         /* Copy slave list to protect against slave up/down changes during tx
1315          * bursting */
1316         num_of_slaves = internals->active_slave_count;
1317         memcpy(slaves, internals->active_slaves,
1318                         sizeof(internals->active_slaves[0]) * num_of_slaves);
1319
1320         if (num_of_slaves < 1)
1321                 return 0;
1322
1323         /* Increment reference count on mbufs */
1324         for (i = 0; i < nb_pkts; i++)
1325                 rte_mbuf_refcnt_update(bufs[i], num_of_slaves - 1);
1326
1327         /* Transmit burst on each active slave */
1328         for (i = 0; i < num_of_slaves; i++) {
1329                 slave_tx_total[i] = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
1330                                         bufs, nb_pkts);
1331
1332                 if (unlikely(slave_tx_total[i] < nb_pkts))
1333                         tx_failed_flag = 1;
1334
1335                 /* record the value and slave index for the slave which transmits the
1336                  * maximum number of packets */
1337                 if (slave_tx_total[i] > max_nb_of_tx_pkts) {
1338                         max_nb_of_tx_pkts = slave_tx_total[i];
1339                         most_successful_tx_slave = i;
1340                 }
1341         }
1342
1343         /* if slaves fail to transmit packets from burst, the calling application
1344          * is not expected to know about multiple references to packets so we must
1345          * handle failures of all packets except those of the most successful slave
1346          */
1347         if (unlikely(tx_failed_flag))
1348                 for (i = 0; i < num_of_slaves; i++)
1349                         if (i != most_successful_tx_slave)
1350                                 while (slave_tx_total[i] < nb_pkts)
1351                                         rte_pktmbuf_free(bufs[slave_tx_total[i]++]);
1352
1353         return max_nb_of_tx_pkts;
1354 }
1355
1356 static void
1357 link_properties_set(struct rte_eth_dev *ethdev, struct rte_eth_link *slave_link)
1358 {
1359         struct bond_dev_private *bond_ctx = ethdev->data->dev_private;
1360
1361         if (bond_ctx->mode == BONDING_MODE_8023AD) {
1362                 /**
1363                  * If in mode 4 then save the link properties of the first
1364                  * slave, all subsequent slaves must match these properties
1365                  */
1366                 struct rte_eth_link *bond_link = &bond_ctx->mode4.slave_link;
1367
1368                 bond_link->link_autoneg = slave_link->link_autoneg;
1369                 bond_link->link_duplex = slave_link->link_duplex;
1370                 bond_link->link_speed = slave_link->link_speed;
1371         } else {
1372                 /**
1373                  * In any other mode the link properties are set to default
1374                  * values of AUTONEG/DUPLEX
1375                  */
1376                 ethdev->data->dev_link.link_autoneg = ETH_LINK_AUTONEG;
1377                 ethdev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
1378         }
1379 }
1380
1381 static int
1382 link_properties_valid(struct rte_eth_dev *ethdev,
1383                 struct rte_eth_link *slave_link)
1384 {
1385         struct bond_dev_private *bond_ctx = ethdev->data->dev_private;
1386
1387         if (bond_ctx->mode == BONDING_MODE_8023AD) {
1388                 struct rte_eth_link *bond_link = &bond_ctx->mode4.slave_link;
1389
1390                 if (bond_link->link_duplex != slave_link->link_duplex ||
1391                         bond_link->link_autoneg != slave_link->link_autoneg ||
1392                         bond_link->link_speed != slave_link->link_speed)
1393                         return -1;
1394         }
1395
1396         return 0;
1397 }
1398
1399 int
1400 mac_address_get(struct rte_eth_dev *eth_dev,
1401                 struct rte_ether_addr *dst_mac_addr)
1402 {
1403         struct rte_ether_addr *mac_addr;
1404
1405         if (eth_dev == NULL) {
1406                 RTE_BOND_LOG(ERR, "NULL pointer eth_dev specified");
1407                 return -1;
1408         }
1409
1410         if (dst_mac_addr == NULL) {
1411                 RTE_BOND_LOG(ERR, "NULL pointer MAC specified");
1412                 return -1;
1413         }
1414
1415         mac_addr = eth_dev->data->mac_addrs;
1416
1417         rte_ether_addr_copy(mac_addr, dst_mac_addr);
1418         return 0;
1419 }
1420
1421 int
1422 mac_address_set(struct rte_eth_dev *eth_dev,
1423                 struct rte_ether_addr *new_mac_addr)
1424 {
1425         struct rte_ether_addr *mac_addr;
1426
1427         if (eth_dev == NULL) {
1428                 RTE_BOND_LOG(ERR, "NULL pointer eth_dev specified");
1429                 return -1;
1430         }
1431
1432         if (new_mac_addr == NULL) {
1433                 RTE_BOND_LOG(ERR, "NULL pointer MAC specified");
1434                 return -1;
1435         }
1436
1437         mac_addr = eth_dev->data->mac_addrs;
1438
1439         /* If new MAC is different to current MAC then update */
1440         if (memcmp(mac_addr, new_mac_addr, sizeof(*mac_addr)) != 0)
1441                 memcpy(mac_addr, new_mac_addr, sizeof(*mac_addr));
1442
1443         return 0;
1444 }
1445
1446 static const struct rte_ether_addr null_mac_addr;
1447
1448 /*
1449  * Add additional MAC addresses to the slave
1450  */
1451 int
1452 slave_add_mac_addresses(struct rte_eth_dev *bonded_eth_dev,
1453                 uint16_t slave_port_id)
1454 {
1455         int i, ret;
1456         struct rte_ether_addr *mac_addr;
1457
1458         for (i = 1; i < BOND_MAX_MAC_ADDRS; i++) {
1459                 mac_addr = &bonded_eth_dev->data->mac_addrs[i];
1460                 if (rte_is_same_ether_addr(mac_addr, &null_mac_addr))
1461                         break;
1462
1463                 ret = rte_eth_dev_mac_addr_add(slave_port_id, mac_addr, 0);
1464                 if (ret < 0) {
1465                         /* rollback */
1466                         for (i--; i > 0; i--)
1467                                 rte_eth_dev_mac_addr_remove(slave_port_id,
1468                                         &bonded_eth_dev->data->mac_addrs[i]);
1469                         return ret;
1470                 }
1471         }
1472
1473         return 0;
1474 }
1475
1476 /*
1477  * Remove additional MAC addresses from the slave
1478  */
1479 int
1480 slave_remove_mac_addresses(struct rte_eth_dev *bonded_eth_dev,
1481                 uint16_t slave_port_id)
1482 {
1483         int i, rc, ret;
1484         struct rte_ether_addr *mac_addr;
1485
1486         rc = 0;
1487         for (i = 1; i < BOND_MAX_MAC_ADDRS; i++) {
1488                 mac_addr = &bonded_eth_dev->data->mac_addrs[i];
1489                 if (rte_is_same_ether_addr(mac_addr, &null_mac_addr))
1490                         break;
1491
1492                 ret = rte_eth_dev_mac_addr_remove(slave_port_id, mac_addr);
1493                 /* save only the first error */
1494                 if (ret < 0 && rc == 0)
1495                         rc = ret;
1496         }
1497
1498         return rc;
1499 }
1500
1501 int
1502 mac_address_slaves_update(struct rte_eth_dev *bonded_eth_dev)
1503 {
1504         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1505         bool set;
1506         int i;
1507
1508         /* Update slave devices MAC addresses */
1509         if (internals->slave_count < 1)
1510                 return -1;
1511
1512         switch (internals->mode) {
1513         case BONDING_MODE_ROUND_ROBIN:
1514         case BONDING_MODE_BALANCE:
1515         case BONDING_MODE_BROADCAST:
1516                 for (i = 0; i < internals->slave_count; i++) {
1517                         if (rte_eth_dev_default_mac_addr_set(
1518                                         internals->slaves[i].port_id,
1519                                         bonded_eth_dev->data->mac_addrs)) {
1520                                 RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1521                                                 internals->slaves[i].port_id);
1522                                 return -1;
1523                         }
1524                 }
1525                 break;
1526         case BONDING_MODE_8023AD:
1527                 bond_mode_8023ad_mac_address_update(bonded_eth_dev);
1528                 break;
1529         case BONDING_MODE_ACTIVE_BACKUP:
1530         case BONDING_MODE_TLB:
1531         case BONDING_MODE_ALB:
1532         default:
1533                 set = true;
1534                 for (i = 0; i < internals->slave_count; i++) {
1535                         if (internals->slaves[i].port_id ==
1536                                         internals->current_primary_port) {
1537                                 if (rte_eth_dev_default_mac_addr_set(
1538                                                 internals->current_primary_port,
1539                                                 bonded_eth_dev->data->mac_addrs)) {
1540                                         RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1541                                                         internals->current_primary_port);
1542                                         set = false;
1543                                 }
1544                         } else {
1545                                 if (rte_eth_dev_default_mac_addr_set(
1546                                                 internals->slaves[i].port_id,
1547                                                 &internals->slaves[i].persisted_mac_addr)) {
1548                                         RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1549                                                         internals->slaves[i].port_id);
1550                                 }
1551                         }
1552                 }
1553                 if (!set)
1554                         return -1;
1555         }
1556
1557         return 0;
1558 }
1559
1560 int
1561 bond_ethdev_mode_set(struct rte_eth_dev *eth_dev, int mode)
1562 {
1563         struct bond_dev_private *internals;
1564
1565         internals = eth_dev->data->dev_private;
1566
1567         switch (mode) {
1568         case BONDING_MODE_ROUND_ROBIN:
1569                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_round_robin;
1570                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1571                 break;
1572         case BONDING_MODE_ACTIVE_BACKUP:
1573                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_active_backup;
1574                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup;
1575                 break;
1576         case BONDING_MODE_BALANCE:
1577                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_balance;
1578                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1579                 break;
1580         case BONDING_MODE_BROADCAST:
1581                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_broadcast;
1582                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1583                 break;
1584         case BONDING_MODE_8023AD:
1585                 if (bond_mode_8023ad_enable(eth_dev) != 0)
1586                         return -1;
1587
1588                 if (internals->mode4.dedicated_queues.enabled == 0) {
1589                         eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_8023ad;
1590                         eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_8023ad;
1591                         RTE_BOND_LOG(WARNING,
1592                                 "Using mode 4, it is necessary to do TX burst "
1593                                 "and RX burst at least every 100ms.");
1594                 } else {
1595                         /* Use flow director's optimization */
1596                         eth_dev->rx_pkt_burst =
1597                                         bond_ethdev_rx_burst_8023ad_fast_queue;
1598                         eth_dev->tx_pkt_burst =
1599                                         bond_ethdev_tx_burst_8023ad_fast_queue;
1600                 }
1601                 break;
1602         case BONDING_MODE_TLB:
1603                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_tlb;
1604                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup;
1605                 break;
1606         case BONDING_MODE_ALB:
1607                 if (bond_mode_alb_enable(eth_dev) != 0)
1608                         return -1;
1609
1610                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_alb;
1611                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_alb;
1612                 break;
1613         default:
1614                 return -1;
1615         }
1616
1617         internals->mode = mode;
1618
1619         return 0;
1620 }
1621
1622
1623 static int
1624 slave_configure_slow_queue(struct rte_eth_dev *bonded_eth_dev,
1625                 struct rte_eth_dev *slave_eth_dev)
1626 {
1627         int errval = 0;
1628         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1629         struct port *port = &bond_mode_8023ad_ports[slave_eth_dev->data->port_id];
1630
1631         if (port->slow_pool == NULL) {
1632                 char mem_name[256];
1633                 int slave_id = slave_eth_dev->data->port_id;
1634
1635                 snprintf(mem_name, RTE_DIM(mem_name), "slave_port%u_slow_pool",
1636                                 slave_id);
1637                 port->slow_pool = rte_pktmbuf_pool_create(mem_name, 8191,
1638                         250, 0, RTE_MBUF_DEFAULT_BUF_SIZE,
1639                         slave_eth_dev->data->numa_node);
1640
1641                 /* Any memory allocation failure in initialization is critical because
1642                  * resources can't be free, so reinitialization is impossible. */
1643                 if (port->slow_pool == NULL) {
1644                         rte_panic("Slave %u: Failed to create memory pool '%s': %s\n",
1645                                 slave_id, mem_name, rte_strerror(rte_errno));
1646                 }
1647         }
1648
1649         if (internals->mode4.dedicated_queues.enabled == 1) {
1650                 /* Configure slow Rx queue */
1651
1652                 errval = rte_eth_rx_queue_setup(slave_eth_dev->data->port_id,
1653                                 internals->mode4.dedicated_queues.rx_qid, 128,
1654                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1655                                 NULL, port->slow_pool);
1656                 if (errval != 0) {
1657                         RTE_BOND_LOG(ERR,
1658                                         "rte_eth_rx_queue_setup: port=%d queue_id %d, err (%d)",
1659                                         slave_eth_dev->data->port_id,
1660                                         internals->mode4.dedicated_queues.rx_qid,
1661                                         errval);
1662                         return errval;
1663                 }
1664
1665                 errval = rte_eth_tx_queue_setup(slave_eth_dev->data->port_id,
1666                                 internals->mode4.dedicated_queues.tx_qid, 512,
1667                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1668                                 NULL);
1669                 if (errval != 0) {
1670                         RTE_BOND_LOG(ERR,
1671                                 "rte_eth_tx_queue_setup: port=%d queue_id %d, err (%d)",
1672                                 slave_eth_dev->data->port_id,
1673                                 internals->mode4.dedicated_queues.tx_qid,
1674                                 errval);
1675                         return errval;
1676                 }
1677         }
1678         return 0;
1679 }
1680
1681 int
1682 slave_configure(struct rte_eth_dev *bonded_eth_dev,
1683                 struct rte_eth_dev *slave_eth_dev)
1684 {
1685         struct bond_rx_queue *bd_rx_q;
1686         struct bond_tx_queue *bd_tx_q;
1687         uint16_t nb_rx_queues;
1688         uint16_t nb_tx_queues;
1689
1690         int errval;
1691         uint16_t q_id;
1692         struct rte_flow_error flow_error;
1693
1694         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1695
1696         /* Stop slave */
1697         errval = rte_eth_dev_stop(slave_eth_dev->data->port_id);
1698         if (errval != 0)
1699                 RTE_BOND_LOG(ERR, "rte_eth_dev_stop: port %u, err (%d)",
1700                              slave_eth_dev->data->port_id, errval);
1701
1702         /* Enable interrupts on slave device if supported */
1703         if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1704                 slave_eth_dev->data->dev_conf.intr_conf.lsc = 1;
1705
1706         /* If RSS is enabled for bonding, try to enable it for slaves  */
1707         if (bonded_eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG) {
1708                 if (internals->rss_key_len != 0) {
1709                         slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len =
1710                                         internals->rss_key_len;
1711                         slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key =
1712                                         internals->rss_key;
1713                 } else {
1714                         slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
1715                 }
1716
1717                 slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf =
1718                                 bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
1719                 slave_eth_dev->data->dev_conf.rxmode.mq_mode =
1720                                 bonded_eth_dev->data->dev_conf.rxmode.mq_mode;
1721         }
1722
1723         if (bonded_eth_dev->data->dev_conf.rxmode.offloads &
1724                         DEV_RX_OFFLOAD_VLAN_FILTER)
1725                 slave_eth_dev->data->dev_conf.rxmode.offloads |=
1726                                 DEV_RX_OFFLOAD_VLAN_FILTER;
1727         else
1728                 slave_eth_dev->data->dev_conf.rxmode.offloads &=
1729                                 ~DEV_RX_OFFLOAD_VLAN_FILTER;
1730
1731         slave_eth_dev->data->dev_conf.rxmode.max_rx_pkt_len =
1732                         bonded_eth_dev->data->dev_conf.rxmode.max_rx_pkt_len;
1733
1734         if (bonded_eth_dev->data->dev_conf.rxmode.offloads &
1735                         DEV_RX_OFFLOAD_JUMBO_FRAME)
1736                 slave_eth_dev->data->dev_conf.rxmode.offloads |=
1737                                 DEV_RX_OFFLOAD_JUMBO_FRAME;
1738         else
1739                 slave_eth_dev->data->dev_conf.rxmode.offloads &=
1740                                 ~DEV_RX_OFFLOAD_JUMBO_FRAME;
1741
1742         nb_rx_queues = bonded_eth_dev->data->nb_rx_queues;
1743         nb_tx_queues = bonded_eth_dev->data->nb_tx_queues;
1744
1745         if (internals->mode == BONDING_MODE_8023AD) {
1746                 if (internals->mode4.dedicated_queues.enabled == 1) {
1747                         nb_rx_queues++;
1748                         nb_tx_queues++;
1749                 }
1750         }
1751
1752         errval = rte_eth_dev_set_mtu(slave_eth_dev->data->port_id,
1753                                      bonded_eth_dev->data->mtu);
1754         if (errval != 0 && errval != -ENOTSUP) {
1755                 RTE_BOND_LOG(ERR, "rte_eth_dev_set_mtu: port %u, err (%d)",
1756                                 slave_eth_dev->data->port_id, errval);
1757                 return errval;
1758         }
1759
1760         /* Configure device */
1761         errval = rte_eth_dev_configure(slave_eth_dev->data->port_id,
1762                         nb_rx_queues, nb_tx_queues,
1763                         &(slave_eth_dev->data->dev_conf));
1764         if (errval != 0) {
1765                 RTE_BOND_LOG(ERR, "Cannot configure slave device: port %u, err (%d)",
1766                                 slave_eth_dev->data->port_id, errval);
1767                 return errval;
1768         }
1769
1770         /* Setup Rx Queues */
1771         for (q_id = 0; q_id < bonded_eth_dev->data->nb_rx_queues; q_id++) {
1772                 bd_rx_q = (struct bond_rx_queue *)bonded_eth_dev->data->rx_queues[q_id];
1773
1774                 errval = rte_eth_rx_queue_setup(slave_eth_dev->data->port_id, q_id,
1775                                 bd_rx_q->nb_rx_desc,
1776                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1777                                 &(bd_rx_q->rx_conf), bd_rx_q->mb_pool);
1778                 if (errval != 0) {
1779                         RTE_BOND_LOG(ERR,
1780                                         "rte_eth_rx_queue_setup: port=%d queue_id %d, err (%d)",
1781                                         slave_eth_dev->data->port_id, q_id, errval);
1782                         return errval;
1783                 }
1784         }
1785
1786         /* Setup Tx Queues */
1787         for (q_id = 0; q_id < bonded_eth_dev->data->nb_tx_queues; q_id++) {
1788                 bd_tx_q = (struct bond_tx_queue *)bonded_eth_dev->data->tx_queues[q_id];
1789
1790                 errval = rte_eth_tx_queue_setup(slave_eth_dev->data->port_id, q_id,
1791                                 bd_tx_q->nb_tx_desc,
1792                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1793                                 &bd_tx_q->tx_conf);
1794                 if (errval != 0) {
1795                         RTE_BOND_LOG(ERR,
1796                                 "rte_eth_tx_queue_setup: port=%d queue_id %d, err (%d)",
1797                                 slave_eth_dev->data->port_id, q_id, errval);
1798                         return errval;
1799                 }
1800         }
1801
1802         if (internals->mode == BONDING_MODE_8023AD &&
1803                         internals->mode4.dedicated_queues.enabled == 1) {
1804                 if (slave_configure_slow_queue(bonded_eth_dev, slave_eth_dev)
1805                                 != 0)
1806                         return errval;
1807
1808                 if (bond_ethdev_8023ad_flow_verify(bonded_eth_dev,
1809                                 slave_eth_dev->data->port_id) != 0) {
1810                         RTE_BOND_LOG(ERR,
1811                                 "rte_eth_tx_queue_setup: port=%d queue_id %d, err (%d)",
1812                                 slave_eth_dev->data->port_id, q_id, errval);
1813                         return -1;
1814                 }
1815
1816                 if (internals->mode4.dedicated_queues.flow[slave_eth_dev->data->port_id] != NULL)
1817                         rte_flow_destroy(slave_eth_dev->data->port_id,
1818                                         internals->mode4.dedicated_queues.flow[slave_eth_dev->data->port_id],
1819                                         &flow_error);
1820
1821                 bond_ethdev_8023ad_flow_set(bonded_eth_dev,
1822                                 slave_eth_dev->data->port_id);
1823         }
1824
1825         /* Start device */
1826         errval = rte_eth_dev_start(slave_eth_dev->data->port_id);
1827         if (errval != 0) {
1828                 RTE_BOND_LOG(ERR, "rte_eth_dev_start: port=%u, err (%d)",
1829                                 slave_eth_dev->data->port_id, errval);
1830                 return -1;
1831         }
1832
1833         /* If RSS is enabled for bonding, synchronize RETA */
1834         if (bonded_eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS) {
1835                 int i;
1836                 struct bond_dev_private *internals;
1837
1838                 internals = bonded_eth_dev->data->dev_private;
1839
1840                 for (i = 0; i < internals->slave_count; i++) {
1841                         if (internals->slaves[i].port_id == slave_eth_dev->data->port_id) {
1842                                 errval = rte_eth_dev_rss_reta_update(
1843                                                 slave_eth_dev->data->port_id,
1844                                                 &internals->reta_conf[0],
1845                                                 internals->slaves[i].reta_size);
1846                                 if (errval != 0) {
1847                                         RTE_BOND_LOG(WARNING,
1848                                                      "rte_eth_dev_rss_reta_update on slave port %d fails (err %d)."
1849                                                      " RSS Configuration for bonding may be inconsistent.",
1850                                                      slave_eth_dev->data->port_id, errval);
1851                                 }
1852                                 break;
1853                         }
1854                 }
1855         }
1856
1857         /* If lsc interrupt is set, check initial slave's link status */
1858         if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
1859                 slave_eth_dev->dev_ops->link_update(slave_eth_dev, 0);
1860                 bond_ethdev_lsc_event_callback(slave_eth_dev->data->port_id,
1861                         RTE_ETH_EVENT_INTR_LSC, &bonded_eth_dev->data->port_id,
1862                         NULL);
1863         }
1864
1865         return 0;
1866 }
1867
1868 void
1869 slave_remove(struct bond_dev_private *internals,
1870                 struct rte_eth_dev *slave_eth_dev)
1871 {
1872         uint16_t i;
1873
1874         for (i = 0; i < internals->slave_count; i++)
1875                 if (internals->slaves[i].port_id ==
1876                                 slave_eth_dev->data->port_id)
1877                         break;
1878
1879         if (i < (internals->slave_count - 1)) {
1880                 struct rte_flow *flow;
1881
1882                 memmove(&internals->slaves[i], &internals->slaves[i + 1],
1883                                 sizeof(internals->slaves[0]) *
1884                                 (internals->slave_count - i - 1));
1885                 TAILQ_FOREACH(flow, &internals->flow_list, next) {
1886                         memmove(&flow->flows[i], &flow->flows[i + 1],
1887                                 sizeof(flow->flows[0]) *
1888                                 (internals->slave_count - i - 1));
1889                         flow->flows[internals->slave_count - 1] = NULL;
1890                 }
1891         }
1892
1893         internals->slave_count--;
1894
1895         /* force reconfiguration of slave interfaces */
1896         rte_eth_dev_internal_reset(slave_eth_dev);
1897 }
1898
1899 static void
1900 bond_ethdev_slave_link_status_change_monitor(void *cb_arg);
1901
1902 void
1903 slave_add(struct bond_dev_private *internals,
1904                 struct rte_eth_dev *slave_eth_dev)
1905 {
1906         struct bond_slave_details *slave_details =
1907                         &internals->slaves[internals->slave_count];
1908
1909         slave_details->port_id = slave_eth_dev->data->port_id;
1910         slave_details->last_link_status = 0;
1911
1912         /* Mark slave devices that don't support interrupts so we can
1913          * compensate when we start the bond
1914          */
1915         if (!(slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1916                 slave_details->link_status_poll_enabled = 1;
1917         }
1918
1919         slave_details->link_status_wait_to_complete = 0;
1920         /* clean tlb_last_obytes when adding port for bonding device */
1921         memcpy(&(slave_details->persisted_mac_addr), slave_eth_dev->data->mac_addrs,
1922                         sizeof(struct rte_ether_addr));
1923 }
1924
1925 void
1926 bond_ethdev_primary_set(struct bond_dev_private *internals,
1927                 uint16_t slave_port_id)
1928 {
1929         int i;
1930
1931         if (internals->active_slave_count < 1)
1932                 internals->current_primary_port = slave_port_id;
1933         else
1934                 /* Search bonded device slave ports for new proposed primary port */
1935                 for (i = 0; i < internals->active_slave_count; i++) {
1936                         if (internals->active_slaves[i] == slave_port_id)
1937                                 internals->current_primary_port = slave_port_id;
1938                 }
1939 }
1940
1941 static int
1942 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev);
1943
1944 static int
1945 bond_ethdev_start(struct rte_eth_dev *eth_dev)
1946 {
1947         struct bond_dev_private *internals;
1948         int i;
1949
1950         /* slave eth dev will be started by bonded device */
1951         if (check_for_bonded_ethdev(eth_dev)) {
1952                 RTE_BOND_LOG(ERR, "User tried to explicitly start a slave eth_dev (%d)",
1953                                 eth_dev->data->port_id);
1954                 return -1;
1955         }
1956
1957         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
1958         eth_dev->data->dev_started = 1;
1959
1960         internals = eth_dev->data->dev_private;
1961
1962         if (internals->slave_count == 0) {
1963                 RTE_BOND_LOG(ERR, "Cannot start port since there are no slave devices");
1964                 goto out_err;
1965         }
1966
1967         if (internals->user_defined_mac == 0) {
1968                 struct rte_ether_addr *new_mac_addr = NULL;
1969
1970                 for (i = 0; i < internals->slave_count; i++)
1971                         if (internals->slaves[i].port_id == internals->primary_port)
1972                                 new_mac_addr = &internals->slaves[i].persisted_mac_addr;
1973
1974                 if (new_mac_addr == NULL)
1975                         goto out_err;
1976
1977                 if (mac_address_set(eth_dev, new_mac_addr) != 0) {
1978                         RTE_BOND_LOG(ERR, "bonded port (%d) failed to update MAC address",
1979                                         eth_dev->data->port_id);
1980                         goto out_err;
1981                 }
1982         }
1983
1984         if (internals->mode == BONDING_MODE_8023AD) {
1985                 if (internals->mode4.dedicated_queues.enabled == 1) {
1986                         internals->mode4.dedicated_queues.rx_qid =
1987                                         eth_dev->data->nb_rx_queues;
1988                         internals->mode4.dedicated_queues.tx_qid =
1989                                         eth_dev->data->nb_tx_queues;
1990                 }
1991         }
1992
1993
1994         /* Reconfigure each slave device if starting bonded device */
1995         for (i = 0; i < internals->slave_count; i++) {
1996                 struct rte_eth_dev *slave_ethdev =
1997                                 &(rte_eth_devices[internals->slaves[i].port_id]);
1998                 if (slave_configure(eth_dev, slave_ethdev) != 0) {
1999                         RTE_BOND_LOG(ERR,
2000                                 "bonded port (%d) failed to reconfigure slave device (%d)",
2001                                 eth_dev->data->port_id,
2002                                 internals->slaves[i].port_id);
2003                         goto out_err;
2004                 }
2005                 /* We will need to poll for link status if any slave doesn't
2006                  * support interrupts
2007                  */
2008                 if (internals->slaves[i].link_status_poll_enabled)
2009                         internals->link_status_polling_enabled = 1;
2010         }
2011
2012         /* start polling if needed */
2013         if (internals->link_status_polling_enabled) {
2014                 rte_eal_alarm_set(
2015                         internals->link_status_polling_interval_ms * 1000,
2016                         bond_ethdev_slave_link_status_change_monitor,
2017                         (void *)&rte_eth_devices[internals->port_id]);
2018         }
2019
2020         /* Update all slave devices MACs*/
2021         if (mac_address_slaves_update(eth_dev) != 0)
2022                 goto out_err;
2023
2024         if (internals->user_defined_primary_port)
2025                 bond_ethdev_primary_set(internals, internals->primary_port);
2026
2027         if (internals->mode == BONDING_MODE_8023AD)
2028                 bond_mode_8023ad_start(eth_dev);
2029
2030         if (internals->mode == BONDING_MODE_TLB ||
2031                         internals->mode == BONDING_MODE_ALB)
2032                 bond_tlb_enable(internals);
2033
2034         return 0;
2035
2036 out_err:
2037         eth_dev->data->dev_started = 0;
2038         return -1;
2039 }
2040
2041 static void
2042 bond_ethdev_free_queues(struct rte_eth_dev *dev)
2043 {
2044         uint16_t i;
2045
2046         if (dev->data->rx_queues != NULL) {
2047                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2048                         rte_free(dev->data->rx_queues[i]);
2049                         dev->data->rx_queues[i] = NULL;
2050                 }
2051                 dev->data->nb_rx_queues = 0;
2052         }
2053
2054         if (dev->data->tx_queues != NULL) {
2055                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2056                         rte_free(dev->data->tx_queues[i]);
2057                         dev->data->tx_queues[i] = NULL;
2058                 }
2059                 dev->data->nb_tx_queues = 0;
2060         }
2061 }
2062
2063 int
2064 bond_ethdev_stop(struct rte_eth_dev *eth_dev)
2065 {
2066         struct bond_dev_private *internals = eth_dev->data->dev_private;
2067         uint16_t i;
2068         int ret;
2069
2070         if (internals->mode == BONDING_MODE_8023AD) {
2071                 struct port *port;
2072                 void *pkt = NULL;
2073
2074                 bond_mode_8023ad_stop(eth_dev);
2075
2076                 /* Discard all messages to/from mode 4 state machines */
2077                 for (i = 0; i < internals->active_slave_count; i++) {
2078                         port = &bond_mode_8023ad_ports[internals->active_slaves[i]];
2079
2080                         RTE_ASSERT(port->rx_ring != NULL);
2081                         while (rte_ring_dequeue(port->rx_ring, &pkt) != -ENOENT)
2082                                 rte_pktmbuf_free(pkt);
2083
2084                         RTE_ASSERT(port->tx_ring != NULL);
2085                         while (rte_ring_dequeue(port->tx_ring, &pkt) != -ENOENT)
2086                                 rte_pktmbuf_free(pkt);
2087                 }
2088         }
2089
2090         if (internals->mode == BONDING_MODE_TLB ||
2091                         internals->mode == BONDING_MODE_ALB) {
2092                 bond_tlb_disable(internals);
2093                 for (i = 0; i < internals->active_slave_count; i++)
2094                         tlb_last_obytets[internals->active_slaves[i]] = 0;
2095         }
2096
2097         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
2098         eth_dev->data->dev_started = 0;
2099
2100         internals->link_status_polling_enabled = 0;
2101         for (i = 0; i < internals->slave_count; i++) {
2102                 uint16_t slave_id = internals->slaves[i].port_id;
2103                 if (find_slave_by_id(internals->active_slaves,
2104                                 internals->active_slave_count, slave_id) !=
2105                                                 internals->active_slave_count) {
2106                         internals->slaves[i].last_link_status = 0;
2107                         ret = rte_eth_dev_stop(slave_id);
2108                         if (ret != 0) {
2109                                 RTE_BOND_LOG(ERR, "Failed to stop device on port %u",
2110                                              slave_id);
2111                                 return ret;
2112                         }
2113                         deactivate_slave(eth_dev, slave_id);
2114                 }
2115         }
2116
2117         return 0;
2118 }
2119
2120 int
2121 bond_ethdev_close(struct rte_eth_dev *dev)
2122 {
2123         struct bond_dev_private *internals = dev->data->dev_private;
2124         uint16_t bond_port_id = internals->port_id;
2125         int skipped = 0;
2126         struct rte_flow_error ferror;
2127
2128         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2129                 return 0;
2130
2131         RTE_BOND_LOG(INFO, "Closing bonded device %s", dev->device->name);
2132         while (internals->slave_count != skipped) {
2133                 uint16_t port_id = internals->slaves[skipped].port_id;
2134
2135                 if (rte_eth_dev_stop(port_id) != 0) {
2136                         RTE_BOND_LOG(ERR, "Failed to stop device on port %u",
2137                                      port_id);
2138                         skipped++;
2139                 }
2140
2141                 if (rte_eth_bond_slave_remove(bond_port_id, port_id) != 0) {
2142                         RTE_BOND_LOG(ERR,
2143                                      "Failed to remove port %d from bonded device %s",
2144                                      port_id, dev->device->name);
2145                         skipped++;
2146                 }
2147         }
2148         bond_flow_ops.flush(dev, &ferror);
2149         bond_ethdev_free_queues(dev);
2150         rte_bitmap_reset(internals->vlan_filter_bmp);
2151         rte_bitmap_free(internals->vlan_filter_bmp);
2152         rte_free(internals->vlan_filter_bmpmem);
2153
2154         /* Try to release mempool used in mode6. If the bond
2155          * device is not mode6, free the NULL is not problem.
2156          */
2157         rte_mempool_free(internals->mode6.mempool);
2158
2159         return 0;
2160 }
2161
2162 /* forward declaration */
2163 static int bond_ethdev_configure(struct rte_eth_dev *dev);
2164
2165 static int
2166 bond_ethdev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2167 {
2168         struct bond_dev_private *internals = dev->data->dev_private;
2169         struct bond_slave_details slave;
2170         int ret;
2171
2172         uint16_t max_nb_rx_queues = UINT16_MAX;
2173         uint16_t max_nb_tx_queues = UINT16_MAX;
2174         uint16_t max_rx_desc_lim = UINT16_MAX;
2175         uint16_t max_tx_desc_lim = UINT16_MAX;
2176
2177         dev_info->max_mac_addrs = BOND_MAX_MAC_ADDRS;
2178
2179         dev_info->max_rx_pktlen = internals->candidate_max_rx_pktlen ?
2180                         internals->candidate_max_rx_pktlen :
2181                         RTE_ETHER_MAX_JUMBO_FRAME_LEN;
2182
2183         /* Max number of tx/rx queues that the bonded device can support is the
2184          * minimum values of the bonded slaves, as all slaves must be capable
2185          * of supporting the same number of tx/rx queues.
2186          */
2187         if (internals->slave_count > 0) {
2188                 struct rte_eth_dev_info slave_info;
2189                 uint16_t idx;
2190
2191                 for (idx = 0; idx < internals->slave_count; idx++) {
2192                         slave = internals->slaves[idx];
2193                         ret = rte_eth_dev_info_get(slave.port_id, &slave_info);
2194                         if (ret != 0) {
2195                                 RTE_BOND_LOG(ERR,
2196                                         "%s: Error during getting device (port %u) info: %s\n",
2197                                         __func__,
2198                                         slave.port_id,
2199                                         strerror(-ret));
2200
2201                                 return ret;
2202                         }
2203
2204                         if (slave_info.max_rx_queues < max_nb_rx_queues)
2205                                 max_nb_rx_queues = slave_info.max_rx_queues;
2206
2207                         if (slave_info.max_tx_queues < max_nb_tx_queues)
2208                                 max_nb_tx_queues = slave_info.max_tx_queues;
2209
2210                         if (slave_info.rx_desc_lim.nb_max < max_rx_desc_lim)
2211                                 max_rx_desc_lim = slave_info.rx_desc_lim.nb_max;
2212
2213                         if (slave_info.tx_desc_lim.nb_max < max_tx_desc_lim)
2214                                 max_tx_desc_lim = slave_info.tx_desc_lim.nb_max;
2215                 }
2216         }
2217
2218         dev_info->max_rx_queues = max_nb_rx_queues;
2219         dev_info->max_tx_queues = max_nb_tx_queues;
2220
2221         memcpy(&dev_info->default_rxconf, &internals->default_rxconf,
2222                sizeof(dev_info->default_rxconf));
2223         memcpy(&dev_info->default_txconf, &internals->default_txconf,
2224                sizeof(dev_info->default_txconf));
2225
2226         dev_info->rx_desc_lim.nb_max = max_rx_desc_lim;
2227         dev_info->tx_desc_lim.nb_max = max_tx_desc_lim;
2228
2229         /**
2230          * If dedicated hw queues enabled for link bonding device in LACP mode
2231          * then we need to reduce the maximum number of data path queues by 1.
2232          */
2233         if (internals->mode == BONDING_MODE_8023AD &&
2234                 internals->mode4.dedicated_queues.enabled == 1) {
2235                 dev_info->max_rx_queues--;
2236                 dev_info->max_tx_queues--;
2237         }
2238
2239         dev_info->min_rx_bufsize = 0;
2240
2241         dev_info->rx_offload_capa = internals->rx_offload_capa;
2242         dev_info->tx_offload_capa = internals->tx_offload_capa;
2243         dev_info->rx_queue_offload_capa = internals->rx_queue_offload_capa;
2244         dev_info->tx_queue_offload_capa = internals->tx_queue_offload_capa;
2245         dev_info->flow_type_rss_offloads = internals->flow_type_rss_offloads;
2246
2247         dev_info->reta_size = internals->reta_size;
2248
2249         return 0;
2250 }
2251
2252 static int
2253 bond_ethdev_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
2254 {
2255         int res;
2256         uint16_t i;
2257         struct bond_dev_private *internals = dev->data->dev_private;
2258
2259         /* don't do this while a slave is being added */
2260         rte_spinlock_lock(&internals->lock);
2261
2262         if (on)
2263                 rte_bitmap_set(internals->vlan_filter_bmp, vlan_id);
2264         else
2265                 rte_bitmap_clear(internals->vlan_filter_bmp, vlan_id);
2266
2267         for (i = 0; i < internals->slave_count; i++) {
2268                 uint16_t port_id = internals->slaves[i].port_id;
2269
2270                 res = rte_eth_dev_vlan_filter(port_id, vlan_id, on);
2271                 if (res == ENOTSUP)
2272                         RTE_BOND_LOG(WARNING,
2273                                      "Setting VLAN filter on slave port %u not supported.",
2274                                      port_id);
2275         }
2276
2277         rte_spinlock_unlock(&internals->lock);
2278         return 0;
2279 }
2280
2281 static int
2282 bond_ethdev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
2283                 uint16_t nb_rx_desc, unsigned int socket_id __rte_unused,
2284                 const struct rte_eth_rxconf *rx_conf, struct rte_mempool *mb_pool)
2285 {
2286         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)
2287                         rte_zmalloc_socket(NULL, sizeof(struct bond_rx_queue),
2288                                         0, dev->data->numa_node);
2289         if (bd_rx_q == NULL)
2290                 return -1;
2291
2292         bd_rx_q->queue_id = rx_queue_id;
2293         bd_rx_q->dev_private = dev->data->dev_private;
2294
2295         bd_rx_q->nb_rx_desc = nb_rx_desc;
2296
2297         memcpy(&(bd_rx_q->rx_conf), rx_conf, sizeof(struct rte_eth_rxconf));
2298         bd_rx_q->mb_pool = mb_pool;
2299
2300         dev->data->rx_queues[rx_queue_id] = bd_rx_q;
2301
2302         return 0;
2303 }
2304
2305 static int
2306 bond_ethdev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
2307                 uint16_t nb_tx_desc, unsigned int socket_id __rte_unused,
2308                 const struct rte_eth_txconf *tx_conf)
2309 {
2310         struct bond_tx_queue *bd_tx_q  = (struct bond_tx_queue *)
2311                         rte_zmalloc_socket(NULL, sizeof(struct bond_tx_queue),
2312                                         0, dev->data->numa_node);
2313
2314         if (bd_tx_q == NULL)
2315                 return -1;
2316
2317         bd_tx_q->queue_id = tx_queue_id;
2318         bd_tx_q->dev_private = dev->data->dev_private;
2319
2320         bd_tx_q->nb_tx_desc = nb_tx_desc;
2321         memcpy(&(bd_tx_q->tx_conf), tx_conf, sizeof(bd_tx_q->tx_conf));
2322
2323         dev->data->tx_queues[tx_queue_id] = bd_tx_q;
2324
2325         return 0;
2326 }
2327
2328 static void
2329 bond_ethdev_rx_queue_release(void *queue)
2330 {
2331         if (queue == NULL)
2332                 return;
2333
2334         rte_free(queue);
2335 }
2336
2337 static void
2338 bond_ethdev_tx_queue_release(void *queue)
2339 {
2340         if (queue == NULL)
2341                 return;
2342
2343         rte_free(queue);
2344 }
2345
2346 static void
2347 bond_ethdev_slave_link_status_change_monitor(void *cb_arg)
2348 {
2349         struct rte_eth_dev *bonded_ethdev, *slave_ethdev;
2350         struct bond_dev_private *internals;
2351
2352         /* Default value for polling slave found is true as we don't want to
2353          * disable the polling thread if we cannot get the lock */
2354         int i, polling_slave_found = 1;
2355
2356         if (cb_arg == NULL)
2357                 return;
2358
2359         bonded_ethdev = cb_arg;
2360         internals = bonded_ethdev->data->dev_private;
2361
2362         if (!bonded_ethdev->data->dev_started ||
2363                 !internals->link_status_polling_enabled)
2364                 return;
2365
2366         /* If device is currently being configured then don't check slaves link
2367          * status, wait until next period */
2368         if (rte_spinlock_trylock(&internals->lock)) {
2369                 if (internals->slave_count > 0)
2370                         polling_slave_found = 0;
2371
2372                 for (i = 0; i < internals->slave_count; i++) {
2373                         if (!internals->slaves[i].link_status_poll_enabled)
2374                                 continue;
2375
2376                         slave_ethdev = &rte_eth_devices[internals->slaves[i].port_id];
2377                         polling_slave_found = 1;
2378
2379                         /* Update slave link status */
2380                         (*slave_ethdev->dev_ops->link_update)(slave_ethdev,
2381                                         internals->slaves[i].link_status_wait_to_complete);
2382
2383                         /* if link status has changed since last checked then call lsc
2384                          * event callback */
2385                         if (slave_ethdev->data->dev_link.link_status !=
2386                                         internals->slaves[i].last_link_status) {
2387                                 internals->slaves[i].last_link_status =
2388                                                 slave_ethdev->data->dev_link.link_status;
2389
2390                                 bond_ethdev_lsc_event_callback(internals->slaves[i].port_id,
2391                                                 RTE_ETH_EVENT_INTR_LSC,
2392                                                 &bonded_ethdev->data->port_id,
2393                                                 NULL);
2394                         }
2395                 }
2396                 rte_spinlock_unlock(&internals->lock);
2397         }
2398
2399         if (polling_slave_found)
2400                 /* Set alarm to continue monitoring link status of slave ethdev's */
2401                 rte_eal_alarm_set(internals->link_status_polling_interval_ms * 1000,
2402                                 bond_ethdev_slave_link_status_change_monitor, cb_arg);
2403 }
2404
2405 static int
2406 bond_ethdev_link_update(struct rte_eth_dev *ethdev, int wait_to_complete)
2407 {
2408         int (*link_update)(uint16_t port_id, struct rte_eth_link *eth_link);
2409
2410         struct bond_dev_private *bond_ctx;
2411         struct rte_eth_link slave_link;
2412
2413         bool one_link_update_succeeded;
2414         uint32_t idx;
2415         int ret;
2416
2417         bond_ctx = ethdev->data->dev_private;
2418
2419         ethdev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
2420
2421         if (ethdev->data->dev_started == 0 ||
2422                         bond_ctx->active_slave_count == 0) {
2423                 ethdev->data->dev_link.link_status = ETH_LINK_DOWN;
2424                 return 0;
2425         }
2426
2427         ethdev->data->dev_link.link_status = ETH_LINK_UP;
2428
2429         if (wait_to_complete)
2430                 link_update = rte_eth_link_get;
2431         else
2432                 link_update = rte_eth_link_get_nowait;
2433
2434         switch (bond_ctx->mode) {
2435         case BONDING_MODE_BROADCAST:
2436                 /**
2437                  * Setting link speed to UINT32_MAX to ensure we pick up the
2438                  * value of the first active slave
2439                  */
2440                 ethdev->data->dev_link.link_speed = UINT32_MAX;
2441
2442                 /**
2443                  * link speed is minimum value of all the slaves link speed as
2444                  * packet loss will occur on this slave if transmission at rates
2445                  * greater than this are attempted
2446                  */
2447                 for (idx = 0; idx < bond_ctx->active_slave_count; idx++) {
2448                         ret = link_update(bond_ctx->active_slaves[idx],
2449                                           &slave_link);
2450                         if (ret < 0) {
2451                                 ethdev->data->dev_link.link_speed =
2452                                         ETH_SPEED_NUM_NONE;
2453                                 RTE_BOND_LOG(ERR,
2454                                         "Slave (port %u) link get failed: %s",
2455                                         bond_ctx->active_slaves[idx],
2456                                         rte_strerror(-ret));
2457                                 return 0;
2458                         }
2459
2460                         if (slave_link.link_speed <
2461                                         ethdev->data->dev_link.link_speed)
2462                                 ethdev->data->dev_link.link_speed =
2463                                                 slave_link.link_speed;
2464                 }
2465                 break;
2466         case BONDING_MODE_ACTIVE_BACKUP:
2467                 /* Current primary slave */
2468                 ret = link_update(bond_ctx->current_primary_port, &slave_link);
2469                 if (ret < 0) {
2470                         RTE_BOND_LOG(ERR, "Slave (port %u) link get failed: %s",
2471                                 bond_ctx->current_primary_port,
2472                                 rte_strerror(-ret));
2473                         return 0;
2474                 }
2475
2476                 ethdev->data->dev_link.link_speed = slave_link.link_speed;
2477                 break;
2478         case BONDING_MODE_8023AD:
2479                 ethdev->data->dev_link.link_autoneg =
2480                                 bond_ctx->mode4.slave_link.link_autoneg;
2481                 ethdev->data->dev_link.link_duplex =
2482                                 bond_ctx->mode4.slave_link.link_duplex;
2483                 /* fall through */
2484                 /* to update link speed */
2485         case BONDING_MODE_ROUND_ROBIN:
2486         case BONDING_MODE_BALANCE:
2487         case BONDING_MODE_TLB:
2488         case BONDING_MODE_ALB:
2489         default:
2490                 /**
2491                  * In theses mode the maximum theoretical link speed is the sum
2492                  * of all the slaves
2493                  */
2494                 ethdev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
2495                 one_link_update_succeeded = false;
2496
2497                 for (idx = 0; idx < bond_ctx->active_slave_count; idx++) {
2498                         ret = link_update(bond_ctx->active_slaves[idx],
2499                                         &slave_link);
2500                         if (ret < 0) {
2501                                 RTE_BOND_LOG(ERR,
2502                                         "Slave (port %u) link get failed: %s",
2503                                         bond_ctx->active_slaves[idx],
2504                                         rte_strerror(-ret));
2505                                 continue;
2506                         }
2507
2508                         one_link_update_succeeded = true;
2509                         ethdev->data->dev_link.link_speed +=
2510                                         slave_link.link_speed;
2511                 }
2512
2513                 if (!one_link_update_succeeded) {
2514                         RTE_BOND_LOG(ERR, "All slaves link get failed");
2515                         return 0;
2516                 }
2517         }
2518
2519
2520         return 0;
2521 }
2522
2523
2524 static int
2525 bond_ethdev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
2526 {
2527         struct bond_dev_private *internals = dev->data->dev_private;
2528         struct rte_eth_stats slave_stats;
2529         int i, j;
2530
2531         for (i = 0; i < internals->slave_count; i++) {
2532                 rte_eth_stats_get(internals->slaves[i].port_id, &slave_stats);
2533
2534                 stats->ipackets += slave_stats.ipackets;
2535                 stats->opackets += slave_stats.opackets;
2536                 stats->ibytes += slave_stats.ibytes;
2537                 stats->obytes += slave_stats.obytes;
2538                 stats->imissed += slave_stats.imissed;
2539                 stats->ierrors += slave_stats.ierrors;
2540                 stats->oerrors += slave_stats.oerrors;
2541                 stats->rx_nombuf += slave_stats.rx_nombuf;
2542
2543                 for (j = 0; j < RTE_ETHDEV_QUEUE_STAT_CNTRS; j++) {
2544                         stats->q_ipackets[j] += slave_stats.q_ipackets[j];
2545                         stats->q_opackets[j] += slave_stats.q_opackets[j];
2546                         stats->q_ibytes[j] += slave_stats.q_ibytes[j];
2547                         stats->q_obytes[j] += slave_stats.q_obytes[j];
2548                         stats->q_errors[j] += slave_stats.q_errors[j];
2549                 }
2550
2551         }
2552
2553         return 0;
2554 }
2555
2556 static int
2557 bond_ethdev_stats_reset(struct rte_eth_dev *dev)
2558 {
2559         struct bond_dev_private *internals = dev->data->dev_private;
2560         int i;
2561         int err;
2562         int ret;
2563
2564         for (i = 0, err = 0; i < internals->slave_count; i++) {
2565                 ret = rte_eth_stats_reset(internals->slaves[i].port_id);
2566                 if (ret != 0)
2567                         err = ret;
2568         }
2569
2570         return err;
2571 }
2572
2573 static int
2574 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev)
2575 {
2576         struct bond_dev_private *internals = eth_dev->data->dev_private;
2577         int i;
2578         int ret = 0;
2579         uint16_t port_id;
2580
2581         switch (internals->mode) {
2582         /* Promiscuous mode is propagated to all slaves */
2583         case BONDING_MODE_ROUND_ROBIN:
2584         case BONDING_MODE_BALANCE:
2585         case BONDING_MODE_BROADCAST:
2586         case BONDING_MODE_8023AD: {
2587                 unsigned int slave_ok = 0;
2588
2589                 for (i = 0; i < internals->slave_count; i++) {
2590                         port_id = internals->slaves[i].port_id;
2591
2592                         ret = rte_eth_promiscuous_enable(port_id);
2593                         if (ret != 0)
2594                                 RTE_BOND_LOG(ERR,
2595                                         "Failed to enable promiscuous mode for port %u: %s",
2596                                         port_id, rte_strerror(-ret));
2597                         else
2598                                 slave_ok++;
2599                 }
2600                 /*
2601                  * Report success if operation is successful on at least
2602                  * on one slave. Otherwise return last error code.
2603                  */
2604                 if (slave_ok > 0)
2605                         ret = 0;
2606                 break;
2607         }
2608         /* Promiscuous mode is propagated only to primary slave */
2609         case BONDING_MODE_ACTIVE_BACKUP:
2610         case BONDING_MODE_TLB:
2611         case BONDING_MODE_ALB:
2612         default:
2613                 /* Do not touch promisc when there cannot be primary ports */
2614                 if (internals->slave_count == 0)
2615                         break;
2616                 port_id = internals->current_primary_port;
2617                 ret = rte_eth_promiscuous_enable(port_id);
2618                 if (ret != 0)
2619                         RTE_BOND_LOG(ERR,
2620                                 "Failed to enable promiscuous mode for port %u: %s",
2621                                 port_id, rte_strerror(-ret));
2622         }
2623
2624         return ret;
2625 }
2626
2627 static int
2628 bond_ethdev_promiscuous_disable(struct rte_eth_dev *dev)
2629 {
2630         struct bond_dev_private *internals = dev->data->dev_private;
2631         int i;
2632         int ret = 0;
2633         uint16_t port_id;
2634
2635         switch (internals->mode) {
2636         /* Promiscuous mode is propagated to all slaves */
2637         case BONDING_MODE_ROUND_ROBIN:
2638         case BONDING_MODE_BALANCE:
2639         case BONDING_MODE_BROADCAST:
2640         case BONDING_MODE_8023AD: {
2641                 unsigned int slave_ok = 0;
2642
2643                 for (i = 0; i < internals->slave_count; i++) {
2644                         port_id = internals->slaves[i].port_id;
2645
2646                         if (internals->mode == BONDING_MODE_8023AD &&
2647                             bond_mode_8023ad_ports[port_id].forced_rx_flags ==
2648                                         BOND_8023AD_FORCED_PROMISC) {
2649                                 slave_ok++;
2650                                 continue;
2651                         }
2652                         ret = rte_eth_promiscuous_disable(port_id);
2653                         if (ret != 0)
2654                                 RTE_BOND_LOG(ERR,
2655                                         "Failed to disable promiscuous mode for port %u: %s",
2656                                         port_id, rte_strerror(-ret));
2657                         else
2658                                 slave_ok++;
2659                 }
2660                 /*
2661                  * Report success if operation is successful on at least
2662                  * on one slave. Otherwise return last error code.
2663                  */
2664                 if (slave_ok > 0)
2665                         ret = 0;
2666                 break;
2667         }
2668         /* Promiscuous mode is propagated only to primary slave */
2669         case BONDING_MODE_ACTIVE_BACKUP:
2670         case BONDING_MODE_TLB:
2671         case BONDING_MODE_ALB:
2672         default:
2673                 /* Do not touch promisc when there cannot be primary ports */
2674                 if (internals->slave_count == 0)
2675                         break;
2676                 port_id = internals->current_primary_port;
2677                 ret = rte_eth_promiscuous_disable(port_id);
2678                 if (ret != 0)
2679                         RTE_BOND_LOG(ERR,
2680                                 "Failed to disable promiscuous mode for port %u: %s",
2681                                 port_id, rte_strerror(-ret));
2682         }
2683
2684         return ret;
2685 }
2686
2687 static int
2688 bond_ethdev_allmulticast_enable(struct rte_eth_dev *eth_dev)
2689 {
2690         struct bond_dev_private *internals = eth_dev->data->dev_private;
2691         int i;
2692         int ret = 0;
2693         uint16_t port_id;
2694
2695         switch (internals->mode) {
2696         /* allmulti mode is propagated to all slaves */
2697         case BONDING_MODE_ROUND_ROBIN:
2698         case BONDING_MODE_BALANCE:
2699         case BONDING_MODE_BROADCAST:
2700         case BONDING_MODE_8023AD: {
2701                 unsigned int slave_ok = 0;
2702
2703                 for (i = 0; i < internals->slave_count; i++) {
2704                         port_id = internals->slaves[i].port_id;
2705
2706                         ret = rte_eth_allmulticast_enable(port_id);
2707                         if (ret != 0)
2708                                 RTE_BOND_LOG(ERR,
2709                                         "Failed to enable allmulti mode for port %u: %s",
2710                                         port_id, rte_strerror(-ret));
2711                         else
2712                                 slave_ok++;
2713                 }
2714                 /*
2715                  * Report success if operation is successful on at least
2716                  * on one slave. Otherwise return last error code.
2717                  */
2718                 if (slave_ok > 0)
2719                         ret = 0;
2720                 break;
2721         }
2722         /* allmulti mode is propagated only to primary slave */
2723         case BONDING_MODE_ACTIVE_BACKUP:
2724         case BONDING_MODE_TLB:
2725         case BONDING_MODE_ALB:
2726         default:
2727                 /* Do not touch allmulti when there cannot be primary ports */
2728                 if (internals->slave_count == 0)
2729                         break;
2730                 port_id = internals->current_primary_port;
2731                 ret = rte_eth_allmulticast_enable(port_id);
2732                 if (ret != 0)
2733                         RTE_BOND_LOG(ERR,
2734                                 "Failed to enable allmulti mode for port %u: %s",
2735                                 port_id, rte_strerror(-ret));
2736         }
2737
2738         return ret;
2739 }
2740
2741 static int
2742 bond_ethdev_allmulticast_disable(struct rte_eth_dev *eth_dev)
2743 {
2744         struct bond_dev_private *internals = eth_dev->data->dev_private;
2745         int i;
2746         int ret = 0;
2747         uint16_t port_id;
2748
2749         switch (internals->mode) {
2750         /* allmulti mode is propagated to all slaves */
2751         case BONDING_MODE_ROUND_ROBIN:
2752         case BONDING_MODE_BALANCE:
2753         case BONDING_MODE_BROADCAST:
2754         case BONDING_MODE_8023AD: {
2755                 unsigned int slave_ok = 0;
2756
2757                 for (i = 0; i < internals->slave_count; i++) {
2758                         uint16_t port_id = internals->slaves[i].port_id;
2759
2760                         if (internals->mode == BONDING_MODE_8023AD &&
2761                             bond_mode_8023ad_ports[port_id].forced_rx_flags ==
2762                                         BOND_8023AD_FORCED_ALLMULTI)
2763                                 continue;
2764
2765                         ret = rte_eth_allmulticast_disable(port_id);
2766                         if (ret != 0)
2767                                 RTE_BOND_LOG(ERR,
2768                                         "Failed to disable allmulti mode for port %u: %s",
2769                                         port_id, rte_strerror(-ret));
2770                         else
2771                                 slave_ok++;
2772                 }
2773                 /*
2774                  * Report success if operation is successful on at least
2775                  * on one slave. Otherwise return last error code.
2776                  */
2777                 if (slave_ok > 0)
2778                         ret = 0;
2779                 break;
2780         }
2781         /* allmulti mode is propagated only to primary slave */
2782         case BONDING_MODE_ACTIVE_BACKUP:
2783         case BONDING_MODE_TLB:
2784         case BONDING_MODE_ALB:
2785         default:
2786                 /* Do not touch allmulti when there cannot be primary ports */
2787                 if (internals->slave_count == 0)
2788                         break;
2789                 port_id = internals->current_primary_port;
2790                 ret = rte_eth_allmulticast_disable(port_id);
2791                 if (ret != 0)
2792                         RTE_BOND_LOG(ERR,
2793                                 "Failed to disable allmulti mode for port %u: %s",
2794                                 port_id, rte_strerror(-ret));
2795         }
2796
2797         return ret;
2798 }
2799
2800 static void
2801 bond_ethdev_delayed_lsc_propagation(void *arg)
2802 {
2803         if (arg == NULL)
2804                 return;
2805
2806         rte_eth_dev_callback_process((struct rte_eth_dev *)arg,
2807                         RTE_ETH_EVENT_INTR_LSC, NULL);
2808 }
2809
2810 int
2811 bond_ethdev_lsc_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2812                 void *param, void *ret_param __rte_unused)
2813 {
2814         struct rte_eth_dev *bonded_eth_dev;
2815         struct bond_dev_private *internals;
2816         struct rte_eth_link link;
2817         int rc = -1;
2818         int ret;
2819
2820         uint8_t lsc_flag = 0;
2821         int valid_slave = 0;
2822         uint16_t active_pos;
2823         uint16_t i;
2824
2825         if (type != RTE_ETH_EVENT_INTR_LSC || param == NULL)
2826                 return rc;
2827
2828         bonded_eth_dev = &rte_eth_devices[*(uint16_t *)param];
2829
2830         if (check_for_bonded_ethdev(bonded_eth_dev))
2831                 return rc;
2832
2833         internals = bonded_eth_dev->data->dev_private;
2834
2835         /* If the device isn't started don't handle interrupts */
2836         if (!bonded_eth_dev->data->dev_started)
2837                 return rc;
2838
2839         /* verify that port_id is a valid slave of bonded port */
2840         for (i = 0; i < internals->slave_count; i++) {
2841                 if (internals->slaves[i].port_id == port_id) {
2842                         valid_slave = 1;
2843                         break;
2844                 }
2845         }
2846
2847         if (!valid_slave)
2848                 return rc;
2849
2850         /* Synchronize lsc callback parallel calls either by real link event
2851          * from the slaves PMDs or by the bonding PMD itself.
2852          */
2853         rte_spinlock_lock(&internals->lsc_lock);
2854
2855         /* Search for port in active port list */
2856         active_pos = find_slave_by_id(internals->active_slaves,
2857                         internals->active_slave_count, port_id);
2858
2859         ret = rte_eth_link_get_nowait(port_id, &link);
2860         if (ret < 0)
2861                 RTE_BOND_LOG(ERR, "Slave (port %u) link get failed", port_id);
2862
2863         if (ret == 0 && link.link_status) {
2864                 if (active_pos < internals->active_slave_count)
2865                         goto link_update;
2866
2867                 /* check link state properties if bonded link is up*/
2868                 if (bonded_eth_dev->data->dev_link.link_status == ETH_LINK_UP) {
2869                         if (link_properties_valid(bonded_eth_dev, &link) != 0)
2870                                 RTE_BOND_LOG(ERR, "Invalid link properties "
2871                                              "for slave %d in bonding mode %d",
2872                                              port_id, internals->mode);
2873                 } else {
2874                         /* inherit slave link properties */
2875                         link_properties_set(bonded_eth_dev, &link);
2876                 }
2877
2878                 /* If no active slave ports then set this port to be
2879                  * the primary port.
2880                  */
2881                 if (internals->active_slave_count < 1) {
2882                         /* If first active slave, then change link status */
2883                         bonded_eth_dev->data->dev_link.link_status =
2884                                                                 ETH_LINK_UP;
2885                         internals->current_primary_port = port_id;
2886                         lsc_flag = 1;
2887
2888                         mac_address_slaves_update(bonded_eth_dev);
2889                 }
2890
2891                 activate_slave(bonded_eth_dev, port_id);
2892
2893                 /* If the user has defined the primary port then default to
2894                  * using it.
2895                  */
2896                 if (internals->user_defined_primary_port &&
2897                                 internals->primary_port == port_id)
2898                         bond_ethdev_primary_set(internals, port_id);
2899         } else {
2900                 if (active_pos == internals->active_slave_count)
2901                         goto link_update;
2902
2903                 /* Remove from active slave list */
2904                 deactivate_slave(bonded_eth_dev, port_id);
2905
2906                 if (internals->active_slave_count < 1)
2907                         lsc_flag = 1;
2908
2909                 /* Update primary id, take first active slave from list or if none
2910                  * available set to -1 */
2911                 if (port_id == internals->current_primary_port) {
2912                         if (internals->active_slave_count > 0)
2913                                 bond_ethdev_primary_set(internals,
2914                                                 internals->active_slaves[0]);
2915                         else
2916                                 internals->current_primary_port = internals->primary_port;
2917                         mac_address_slaves_update(bonded_eth_dev);
2918                 }
2919         }
2920
2921 link_update:
2922         /**
2923          * Update bonded device link properties after any change to active
2924          * slaves
2925          */
2926         bond_ethdev_link_update(bonded_eth_dev, 0);
2927
2928         if (lsc_flag) {
2929                 /* Cancel any possible outstanding interrupts if delays are enabled */
2930                 if (internals->link_up_delay_ms > 0 ||
2931                         internals->link_down_delay_ms > 0)
2932                         rte_eal_alarm_cancel(bond_ethdev_delayed_lsc_propagation,
2933                                         bonded_eth_dev);
2934
2935                 if (bonded_eth_dev->data->dev_link.link_status) {
2936                         if (internals->link_up_delay_ms > 0)
2937                                 rte_eal_alarm_set(internals->link_up_delay_ms * 1000,
2938                                                 bond_ethdev_delayed_lsc_propagation,
2939                                                 (void *)bonded_eth_dev);
2940                         else
2941                                 rte_eth_dev_callback_process(bonded_eth_dev,
2942                                                 RTE_ETH_EVENT_INTR_LSC,
2943                                                 NULL);
2944
2945                 } else {
2946                         if (internals->link_down_delay_ms > 0)
2947                                 rte_eal_alarm_set(internals->link_down_delay_ms * 1000,
2948                                                 bond_ethdev_delayed_lsc_propagation,
2949                                                 (void *)bonded_eth_dev);
2950                         else
2951                                 rte_eth_dev_callback_process(bonded_eth_dev,
2952                                                 RTE_ETH_EVENT_INTR_LSC,
2953                                                 NULL);
2954                 }
2955         }
2956
2957         rte_spinlock_unlock(&internals->lsc_lock);
2958
2959         return rc;
2960 }
2961
2962 static int
2963 bond_ethdev_rss_reta_update(struct rte_eth_dev *dev,
2964                 struct rte_eth_rss_reta_entry64 *reta_conf, uint16_t reta_size)
2965 {
2966         unsigned i, j;
2967         int result = 0;
2968         int slave_reta_size;
2969         unsigned reta_count;
2970         struct bond_dev_private *internals = dev->data->dev_private;
2971
2972         if (reta_size != internals->reta_size)
2973                 return -EINVAL;
2974
2975          /* Copy RETA table */
2976         reta_count = (reta_size + RTE_RETA_GROUP_SIZE - 1) /
2977                         RTE_RETA_GROUP_SIZE;
2978
2979         for (i = 0; i < reta_count; i++) {
2980                 internals->reta_conf[i].mask = reta_conf[i].mask;
2981                 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
2982                         if ((reta_conf[i].mask >> j) & 0x01)
2983                                 internals->reta_conf[i].reta[j] = reta_conf[i].reta[j];
2984         }
2985
2986         /* Fill rest of array */
2987         for (; i < RTE_DIM(internals->reta_conf); i += reta_count)
2988                 memcpy(&internals->reta_conf[i], &internals->reta_conf[0],
2989                                 sizeof(internals->reta_conf[0]) * reta_count);
2990
2991         /* Propagate RETA over slaves */
2992         for (i = 0; i < internals->slave_count; i++) {
2993                 slave_reta_size = internals->slaves[i].reta_size;
2994                 result = rte_eth_dev_rss_reta_update(internals->slaves[i].port_id,
2995                                 &internals->reta_conf[0], slave_reta_size);
2996                 if (result < 0)
2997                         return result;
2998         }
2999
3000         return 0;
3001 }
3002
3003 static int
3004 bond_ethdev_rss_reta_query(struct rte_eth_dev *dev,
3005                 struct rte_eth_rss_reta_entry64 *reta_conf, uint16_t reta_size)
3006 {
3007         int i, j;
3008         struct bond_dev_private *internals = dev->data->dev_private;
3009
3010         if (reta_size != internals->reta_size)
3011                 return -EINVAL;
3012
3013          /* Copy RETA table */
3014         for (i = 0; i < reta_size / RTE_RETA_GROUP_SIZE; i++)
3015                 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
3016                         if ((reta_conf[i].mask >> j) & 0x01)
3017                                 reta_conf[i].reta[j] = internals->reta_conf[i].reta[j];
3018
3019         return 0;
3020 }
3021
3022 static int
3023 bond_ethdev_rss_hash_update(struct rte_eth_dev *dev,
3024                 struct rte_eth_rss_conf *rss_conf)
3025 {
3026         int i, result = 0;
3027         struct bond_dev_private *internals = dev->data->dev_private;
3028         struct rte_eth_rss_conf bond_rss_conf;
3029
3030         memcpy(&bond_rss_conf, rss_conf, sizeof(struct rte_eth_rss_conf));
3031
3032         bond_rss_conf.rss_hf &= internals->flow_type_rss_offloads;
3033
3034         if (bond_rss_conf.rss_hf != 0)
3035                 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf = bond_rss_conf.rss_hf;
3036
3037         if (bond_rss_conf.rss_key && bond_rss_conf.rss_key_len <
3038                         sizeof(internals->rss_key)) {
3039                 if (bond_rss_conf.rss_key_len == 0)
3040                         bond_rss_conf.rss_key_len = 40;
3041                 internals->rss_key_len = bond_rss_conf.rss_key_len;
3042                 memcpy(internals->rss_key, bond_rss_conf.rss_key,
3043                                 internals->rss_key_len);
3044         }
3045
3046         for (i = 0; i < internals->slave_count; i++) {
3047                 result = rte_eth_dev_rss_hash_update(internals->slaves[i].port_id,
3048                                 &bond_rss_conf);
3049                 if (result < 0)
3050                         return result;
3051         }
3052
3053         return 0;
3054 }
3055
3056 static int
3057 bond_ethdev_rss_hash_conf_get(struct rte_eth_dev *dev,
3058                 struct rte_eth_rss_conf *rss_conf)
3059 {
3060         struct bond_dev_private *internals = dev->data->dev_private;
3061
3062         rss_conf->rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
3063         rss_conf->rss_key_len = internals->rss_key_len;
3064         if (rss_conf->rss_key)
3065                 memcpy(rss_conf->rss_key, internals->rss_key, internals->rss_key_len);
3066
3067         return 0;
3068 }
3069
3070 static int
3071 bond_ethdev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
3072 {
3073         struct rte_eth_dev *slave_eth_dev;
3074         struct bond_dev_private *internals = dev->data->dev_private;
3075         int ret, i;
3076
3077         rte_spinlock_lock(&internals->lock);
3078
3079         for (i = 0; i < internals->slave_count; i++) {
3080                 slave_eth_dev = &rte_eth_devices[internals->slaves[i].port_id];
3081                 if (*slave_eth_dev->dev_ops->mtu_set == NULL) {
3082                         rte_spinlock_unlock(&internals->lock);
3083                         return -ENOTSUP;
3084                 }
3085         }
3086         for (i = 0; i < internals->slave_count; i++) {
3087                 ret = rte_eth_dev_set_mtu(internals->slaves[i].port_id, mtu);
3088                 if (ret < 0) {
3089                         rte_spinlock_unlock(&internals->lock);
3090                         return ret;
3091                 }
3092         }
3093
3094         rte_spinlock_unlock(&internals->lock);
3095         return 0;
3096 }
3097
3098 static int
3099 bond_ethdev_mac_address_set(struct rte_eth_dev *dev,
3100                         struct rte_ether_addr *addr)
3101 {
3102         if (mac_address_set(dev, addr)) {
3103                 RTE_BOND_LOG(ERR, "Failed to update MAC address");
3104                 return -EINVAL;
3105         }
3106
3107         return 0;
3108 }
3109
3110 static int
3111 bond_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
3112                   const struct rte_flow_ops **ops)
3113 {
3114         *ops = &bond_flow_ops;
3115         return 0;
3116 }
3117
3118 static int
3119 bond_ethdev_mac_addr_add(struct rte_eth_dev *dev,
3120                         struct rte_ether_addr *mac_addr,
3121                         __rte_unused uint32_t index, uint32_t vmdq)
3122 {
3123         struct rte_eth_dev *slave_eth_dev;
3124         struct bond_dev_private *internals = dev->data->dev_private;
3125         int ret, i;
3126
3127         rte_spinlock_lock(&internals->lock);
3128
3129         for (i = 0; i < internals->slave_count; i++) {
3130                 slave_eth_dev = &rte_eth_devices[internals->slaves[i].port_id];
3131                 if (*slave_eth_dev->dev_ops->mac_addr_add == NULL ||
3132                          *slave_eth_dev->dev_ops->mac_addr_remove == NULL) {
3133                         ret = -ENOTSUP;
3134                         goto end;
3135                 }
3136         }
3137
3138         for (i = 0; i < internals->slave_count; i++) {
3139                 ret = rte_eth_dev_mac_addr_add(internals->slaves[i].port_id,
3140                                 mac_addr, vmdq);
3141                 if (ret < 0) {
3142                         /* rollback */
3143                         for (i--; i >= 0; i--)
3144                                 rte_eth_dev_mac_addr_remove(
3145                                         internals->slaves[i].port_id, mac_addr);
3146                         goto end;
3147                 }
3148         }
3149
3150         ret = 0;
3151 end:
3152         rte_spinlock_unlock(&internals->lock);
3153         return ret;
3154 }
3155
3156 static void
3157 bond_ethdev_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
3158 {
3159         struct rte_eth_dev *slave_eth_dev;
3160         struct bond_dev_private *internals = dev->data->dev_private;
3161         int i;
3162
3163         rte_spinlock_lock(&internals->lock);
3164
3165         for (i = 0; i < internals->slave_count; i++) {
3166                 slave_eth_dev = &rte_eth_devices[internals->slaves[i].port_id];
3167                 if (*slave_eth_dev->dev_ops->mac_addr_remove == NULL)
3168                         goto end;
3169         }
3170
3171         struct rte_ether_addr *mac_addr = &dev->data->mac_addrs[index];
3172
3173         for (i = 0; i < internals->slave_count; i++)
3174                 rte_eth_dev_mac_addr_remove(internals->slaves[i].port_id,
3175                                 mac_addr);
3176
3177 end:
3178         rte_spinlock_unlock(&internals->lock);
3179 }
3180
3181 const struct eth_dev_ops default_dev_ops = {
3182         .dev_start            = bond_ethdev_start,
3183         .dev_stop             = bond_ethdev_stop,
3184         .dev_close            = bond_ethdev_close,
3185         .dev_configure        = bond_ethdev_configure,
3186         .dev_infos_get        = bond_ethdev_info,
3187         .vlan_filter_set      = bond_ethdev_vlan_filter_set,
3188         .rx_queue_setup       = bond_ethdev_rx_queue_setup,
3189         .tx_queue_setup       = bond_ethdev_tx_queue_setup,
3190         .rx_queue_release     = bond_ethdev_rx_queue_release,
3191         .tx_queue_release     = bond_ethdev_tx_queue_release,
3192         .link_update          = bond_ethdev_link_update,
3193         .stats_get            = bond_ethdev_stats_get,
3194         .stats_reset          = bond_ethdev_stats_reset,
3195         .promiscuous_enable   = bond_ethdev_promiscuous_enable,
3196         .promiscuous_disable  = bond_ethdev_promiscuous_disable,
3197         .allmulticast_enable  = bond_ethdev_allmulticast_enable,
3198         .allmulticast_disable = bond_ethdev_allmulticast_disable,
3199         .reta_update          = bond_ethdev_rss_reta_update,
3200         .reta_query           = bond_ethdev_rss_reta_query,
3201         .rss_hash_update      = bond_ethdev_rss_hash_update,
3202         .rss_hash_conf_get    = bond_ethdev_rss_hash_conf_get,
3203         .mtu_set              = bond_ethdev_mtu_set,
3204         .mac_addr_set         = bond_ethdev_mac_address_set,
3205         .mac_addr_add         = bond_ethdev_mac_addr_add,
3206         .mac_addr_remove      = bond_ethdev_mac_addr_remove,
3207         .flow_ops_get         = bond_flow_ops_get
3208 };
3209
3210 static int
3211 bond_alloc(struct rte_vdev_device *dev, uint8_t mode)
3212 {
3213         const char *name = rte_vdev_device_name(dev);
3214         uint8_t socket_id = dev->device.numa_node;
3215         struct bond_dev_private *internals = NULL;
3216         struct rte_eth_dev *eth_dev = NULL;
3217         uint32_t vlan_filter_bmp_size;
3218
3219         /* now do all data allocation - for eth_dev structure, dummy pci driver
3220          * and internal (private) data
3221          */
3222
3223         /* reserve an ethdev entry */
3224         eth_dev = rte_eth_vdev_allocate(dev, sizeof(*internals));
3225         if (eth_dev == NULL) {
3226                 RTE_BOND_LOG(ERR, "Unable to allocate rte_eth_dev");
3227                 goto err;
3228         }
3229
3230         internals = eth_dev->data->dev_private;
3231         eth_dev->data->nb_rx_queues = (uint16_t)1;
3232         eth_dev->data->nb_tx_queues = (uint16_t)1;
3233
3234         /* Allocate memory for storing MAC addresses */
3235         eth_dev->data->mac_addrs = rte_zmalloc_socket(name, RTE_ETHER_ADDR_LEN *
3236                         BOND_MAX_MAC_ADDRS, 0, socket_id);
3237         if (eth_dev->data->mac_addrs == NULL) {
3238                 RTE_BOND_LOG(ERR,
3239                              "Failed to allocate %u bytes needed to store MAC addresses",
3240                              RTE_ETHER_ADDR_LEN * BOND_MAX_MAC_ADDRS);
3241                 goto err;
3242         }
3243
3244         eth_dev->dev_ops = &default_dev_ops;
3245         eth_dev->data->dev_flags = RTE_ETH_DEV_INTR_LSC |
3246                                         RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
3247
3248         rte_spinlock_init(&internals->lock);
3249         rte_spinlock_init(&internals->lsc_lock);
3250
3251         internals->port_id = eth_dev->data->port_id;
3252         internals->mode = BONDING_MODE_INVALID;
3253         internals->current_primary_port = RTE_MAX_ETHPORTS + 1;
3254         internals->balance_xmit_policy = BALANCE_XMIT_POLICY_LAYER2;
3255         internals->burst_xmit_hash = burst_xmit_l2_hash;
3256         internals->user_defined_mac = 0;
3257
3258         internals->link_status_polling_enabled = 0;
3259
3260         internals->link_status_polling_interval_ms =
3261                 DEFAULT_POLLING_INTERVAL_10_MS;
3262         internals->link_down_delay_ms = 0;
3263         internals->link_up_delay_ms = 0;
3264
3265         internals->slave_count = 0;
3266         internals->active_slave_count = 0;
3267         internals->rx_offload_capa = 0;
3268         internals->tx_offload_capa = 0;
3269         internals->rx_queue_offload_capa = 0;
3270         internals->tx_queue_offload_capa = 0;
3271         internals->candidate_max_rx_pktlen = 0;
3272         internals->max_rx_pktlen = 0;
3273
3274         /* Initially allow to choose any offload type */
3275         internals->flow_type_rss_offloads = ETH_RSS_PROTO_MASK;
3276
3277         memset(&internals->default_rxconf, 0,
3278                sizeof(internals->default_rxconf));
3279         memset(&internals->default_txconf, 0,
3280                sizeof(internals->default_txconf));
3281
3282         memset(&internals->rx_desc_lim, 0, sizeof(internals->rx_desc_lim));
3283         memset(&internals->tx_desc_lim, 0, sizeof(internals->tx_desc_lim));
3284
3285         memset(internals->active_slaves, 0, sizeof(internals->active_slaves));
3286         memset(internals->slaves, 0, sizeof(internals->slaves));
3287
3288         TAILQ_INIT(&internals->flow_list);
3289         internals->flow_isolated_valid = 0;
3290
3291         /* Set mode 4 default configuration */
3292         bond_mode_8023ad_setup(eth_dev, NULL);
3293         if (bond_ethdev_mode_set(eth_dev, mode)) {
3294                 RTE_BOND_LOG(ERR, "Failed to set bonded device %d mode to %d",
3295                                  eth_dev->data->port_id, mode);
3296                 goto err;
3297         }
3298
3299         vlan_filter_bmp_size =
3300                 rte_bitmap_get_memory_footprint(RTE_ETHER_MAX_VLAN_ID + 1);
3301         internals->vlan_filter_bmpmem = rte_malloc(name, vlan_filter_bmp_size,
3302                                                    RTE_CACHE_LINE_SIZE);
3303         if (internals->vlan_filter_bmpmem == NULL) {
3304                 RTE_BOND_LOG(ERR,
3305                              "Failed to allocate vlan bitmap for bonded device %u",
3306                              eth_dev->data->port_id);
3307                 goto err;
3308         }
3309
3310         internals->vlan_filter_bmp = rte_bitmap_init(RTE_ETHER_MAX_VLAN_ID + 1,
3311                         internals->vlan_filter_bmpmem, vlan_filter_bmp_size);
3312         if (internals->vlan_filter_bmp == NULL) {
3313                 RTE_BOND_LOG(ERR,
3314                              "Failed to init vlan bitmap for bonded device %u",
3315                              eth_dev->data->port_id);
3316                 rte_free(internals->vlan_filter_bmpmem);
3317                 goto err;
3318         }
3319
3320         return eth_dev->data->port_id;
3321
3322 err:
3323         rte_free(internals);
3324         if (eth_dev != NULL)
3325                 eth_dev->data->dev_private = NULL;
3326         rte_eth_dev_release_port(eth_dev);
3327         return -1;
3328 }
3329
3330 static int
3331 bond_probe(struct rte_vdev_device *dev)
3332 {
3333         const char *name;
3334         struct bond_dev_private *internals;
3335         struct rte_kvargs *kvlist;
3336         uint8_t bonding_mode, socket_id/*, agg_mode*/;
3337         int  arg_count, port_id;
3338         uint8_t agg_mode;
3339         struct rte_eth_dev *eth_dev;
3340
3341         if (!dev)
3342                 return -EINVAL;
3343
3344         name = rte_vdev_device_name(dev);
3345         RTE_BOND_LOG(INFO, "Initializing pmd_bond for %s", name);
3346
3347         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
3348                 eth_dev = rte_eth_dev_attach_secondary(name);
3349                 if (!eth_dev) {
3350                         RTE_BOND_LOG(ERR, "Failed to probe %s", name);
3351                         return -1;
3352                 }
3353                 /* TODO: request info from primary to set up Rx and Tx */
3354                 eth_dev->dev_ops = &default_dev_ops;
3355                 eth_dev->device = &dev->device;
3356                 rte_eth_dev_probing_finish(eth_dev);
3357                 return 0;
3358         }
3359
3360         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev),
3361                 pmd_bond_init_valid_arguments);
3362         if (kvlist == NULL)
3363                 return -1;
3364
3365         /* Parse link bonding mode */
3366         if (rte_kvargs_count(kvlist, PMD_BOND_MODE_KVARG) == 1) {
3367                 if (rte_kvargs_process(kvlist, PMD_BOND_MODE_KVARG,
3368                                 &bond_ethdev_parse_slave_mode_kvarg,
3369                                 &bonding_mode) != 0) {
3370                         RTE_BOND_LOG(ERR, "Invalid mode for bonded device %s",
3371                                         name);
3372                         goto parse_error;
3373                 }
3374         } else {
3375                 RTE_BOND_LOG(ERR, "Mode must be specified only once for bonded "
3376                                 "device %s", name);
3377                 goto parse_error;
3378         }
3379
3380         /* Parse socket id to create bonding device on */
3381         arg_count = rte_kvargs_count(kvlist, PMD_BOND_SOCKET_ID_KVARG);
3382         if (arg_count == 1) {
3383                 if (rte_kvargs_process(kvlist, PMD_BOND_SOCKET_ID_KVARG,
3384                                 &bond_ethdev_parse_socket_id_kvarg, &socket_id)
3385                                 != 0) {
3386                         RTE_BOND_LOG(ERR, "Invalid socket Id specified for "
3387                                         "bonded device %s", name);
3388                         goto parse_error;
3389                 }
3390         } else if (arg_count > 1) {
3391                 RTE_BOND_LOG(ERR, "Socket Id can be specified only once for "
3392                                 "bonded device %s", name);
3393                 goto parse_error;
3394         } else {
3395                 socket_id = rte_socket_id();
3396         }
3397
3398         dev->device.numa_node = socket_id;
3399
3400         /* Create link bonding eth device */
3401         port_id = bond_alloc(dev, bonding_mode);
3402         if (port_id < 0) {
3403                 RTE_BOND_LOG(ERR, "Failed to create socket %s in mode %u on "
3404                                 "socket %u.",   name, bonding_mode, socket_id);
3405                 goto parse_error;
3406         }
3407         internals = rte_eth_devices[port_id].data->dev_private;
3408         internals->kvlist = kvlist;
3409
3410         if (rte_kvargs_count(kvlist, PMD_BOND_AGG_MODE_KVARG) == 1) {
3411                 if (rte_kvargs_process(kvlist,
3412                                 PMD_BOND_AGG_MODE_KVARG,
3413                                 &bond_ethdev_parse_slave_agg_mode_kvarg,
3414                                 &agg_mode) != 0) {
3415                         RTE_BOND_LOG(ERR,
3416                                         "Failed to parse agg selection mode for bonded device %s",
3417                                         name);
3418                         goto parse_error;
3419                 }
3420
3421                 if (internals->mode == BONDING_MODE_8023AD)
3422                         internals->mode4.agg_selection = agg_mode;
3423         } else {
3424                 internals->mode4.agg_selection = AGG_STABLE;
3425         }
3426
3427         rte_eth_dev_probing_finish(&rte_eth_devices[port_id]);
3428         RTE_BOND_LOG(INFO, "Create bonded device %s on port %d in mode %u on "
3429                         "socket %u.",   name, port_id, bonding_mode, socket_id);
3430         return 0;
3431
3432 parse_error:
3433         rte_kvargs_free(kvlist);
3434
3435         return -1;
3436 }
3437
3438 static int
3439 bond_remove(struct rte_vdev_device *dev)
3440 {
3441         struct rte_eth_dev *eth_dev;
3442         struct bond_dev_private *internals;
3443         const char *name;
3444         int ret = 0;
3445
3446         if (!dev)
3447                 return -EINVAL;
3448
3449         name = rte_vdev_device_name(dev);
3450         RTE_BOND_LOG(INFO, "Uninitializing pmd_bond for %s", name);
3451
3452         /* find an ethdev entry */
3453         eth_dev = rte_eth_dev_allocated(name);
3454         if (eth_dev == NULL)
3455                 return 0; /* port already released */
3456
3457         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
3458                 return rte_eth_dev_release_port(eth_dev);
3459
3460         RTE_ASSERT(eth_dev->device == &dev->device);
3461
3462         internals = eth_dev->data->dev_private;
3463         if (internals->slave_count != 0)
3464                 return -EBUSY;
3465
3466         if (eth_dev->data->dev_started == 1) {
3467                 ret = bond_ethdev_stop(eth_dev);
3468                 bond_ethdev_close(eth_dev);
3469         }
3470         if (internals->kvlist != NULL)
3471                 rte_kvargs_free(internals->kvlist);
3472         rte_eth_dev_release_port(eth_dev);
3473
3474         return ret;
3475 }
3476
3477 /* this part will resolve the slave portids after all the other pdev and vdev
3478  * have been allocated */
3479 static int
3480 bond_ethdev_configure(struct rte_eth_dev *dev)
3481 {
3482         const char *name = dev->device->name;
3483         struct bond_dev_private *internals = dev->data->dev_private;
3484         struct rte_kvargs *kvlist = internals->kvlist;
3485         int arg_count;
3486         uint16_t port_id = dev - rte_eth_devices;
3487         uint8_t agg_mode;
3488
3489         static const uint8_t default_rss_key[40] = {
3490                 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2, 0x41, 0x67, 0x25, 0x3D,
3491                 0x43, 0xA3, 0x8F, 0xB0, 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
3492                 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C, 0x6A, 0x42, 0xB7, 0x3B,
3493                 0xBE, 0xAC, 0x01, 0xFA
3494         };
3495
3496         unsigned i, j;
3497
3498         /*
3499          * If RSS is enabled, fill table with default values and
3500          * set key to the the value specified in port RSS configuration.
3501          * Fall back to default RSS key if the key is not specified
3502          */
3503         if (dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS) {
3504                 if (dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key != NULL) {
3505                         internals->rss_key_len =
3506                                 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len;
3507                         memcpy(internals->rss_key,
3508                                dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key,
3509                                internals->rss_key_len);
3510                 } else {
3511                         internals->rss_key_len = sizeof(default_rss_key);
3512                         memcpy(internals->rss_key, default_rss_key,
3513                                internals->rss_key_len);
3514                 }
3515
3516                 for (i = 0; i < RTE_DIM(internals->reta_conf); i++) {
3517                         internals->reta_conf[i].mask = ~0LL;
3518                         for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
3519                                 internals->reta_conf[i].reta[j] =
3520                                                 (i * RTE_RETA_GROUP_SIZE + j) %
3521                                                 dev->data->nb_rx_queues;
3522                 }
3523         }
3524
3525         /* set the max_rx_pktlen */
3526         internals->max_rx_pktlen = internals->candidate_max_rx_pktlen;
3527
3528         /*
3529          * if no kvlist, it means that this bonded device has been created
3530          * through the bonding api.
3531          */
3532         if (!kvlist)
3533                 return 0;
3534
3535         /* Parse MAC address for bonded device */
3536         arg_count = rte_kvargs_count(kvlist, PMD_BOND_MAC_ADDR_KVARG);
3537         if (arg_count == 1) {
3538                 struct rte_ether_addr bond_mac;
3539
3540                 if (rte_kvargs_process(kvlist, PMD_BOND_MAC_ADDR_KVARG,
3541                                        &bond_ethdev_parse_bond_mac_addr_kvarg, &bond_mac) < 0) {
3542                         RTE_BOND_LOG(INFO, "Invalid mac address for bonded device %s",
3543                                      name);
3544                         return -1;
3545                 }
3546
3547                 /* Set MAC address */
3548                 if (rte_eth_bond_mac_address_set(port_id, &bond_mac) != 0) {
3549                         RTE_BOND_LOG(ERR,
3550                                      "Failed to set mac address on bonded device %s",
3551                                      name);
3552                         return -1;
3553                 }
3554         } else if (arg_count > 1) {
3555                 RTE_BOND_LOG(ERR,
3556                              "MAC address can be specified only once for bonded device %s",
3557                              name);
3558                 return -1;
3559         }
3560
3561         /* Parse/set balance mode transmit policy */
3562         arg_count = rte_kvargs_count(kvlist, PMD_BOND_XMIT_POLICY_KVARG);
3563         if (arg_count == 1) {
3564                 uint8_t xmit_policy;
3565
3566                 if (rte_kvargs_process(kvlist, PMD_BOND_XMIT_POLICY_KVARG,
3567                                        &bond_ethdev_parse_balance_xmit_policy_kvarg, &xmit_policy) !=
3568                     0) {
3569                         RTE_BOND_LOG(INFO,
3570                                      "Invalid xmit policy specified for bonded device %s",
3571                                      name);
3572                         return -1;
3573                 }
3574
3575                 /* Set balance mode transmit policy*/
3576                 if (rte_eth_bond_xmit_policy_set(port_id, xmit_policy) != 0) {
3577                         RTE_BOND_LOG(ERR,
3578                                      "Failed to set balance xmit policy on bonded device %s",
3579                                      name);
3580                         return -1;
3581                 }
3582         } else if (arg_count > 1) {
3583                 RTE_BOND_LOG(ERR,
3584                              "Transmit policy can be specified only once for bonded device %s",
3585                              name);
3586                 return -1;
3587         }
3588
3589         if (rte_kvargs_count(kvlist, PMD_BOND_AGG_MODE_KVARG) == 1) {
3590                 if (rte_kvargs_process(kvlist,
3591                                        PMD_BOND_AGG_MODE_KVARG,
3592                                        &bond_ethdev_parse_slave_agg_mode_kvarg,
3593                                        &agg_mode) != 0) {
3594                         RTE_BOND_LOG(ERR,
3595                                      "Failed to parse agg selection mode for bonded device %s",
3596                                      name);
3597                 }
3598                 if (internals->mode == BONDING_MODE_8023AD) {
3599                         int ret = rte_eth_bond_8023ad_agg_selection_set(port_id,
3600                                         agg_mode);
3601                         if (ret < 0) {
3602                                 RTE_BOND_LOG(ERR,
3603                                         "Invalid args for agg selection set for bonded device %s",
3604                                         name);
3605                                 return -1;
3606                         }
3607                 }
3608         }
3609
3610         /* Parse/add slave ports to bonded device */
3611         if (rte_kvargs_count(kvlist, PMD_BOND_SLAVE_PORT_KVARG) > 0) {
3612                 struct bond_ethdev_slave_ports slave_ports;
3613                 unsigned i;
3614
3615                 memset(&slave_ports, 0, sizeof(slave_ports));
3616
3617                 if (rte_kvargs_process(kvlist, PMD_BOND_SLAVE_PORT_KVARG,
3618                                        &bond_ethdev_parse_slave_port_kvarg, &slave_ports) != 0) {
3619                         RTE_BOND_LOG(ERR,
3620                                      "Failed to parse slave ports for bonded device %s",
3621                                      name);
3622                         return -1;
3623                 }
3624
3625                 for (i = 0; i < slave_ports.slave_count; i++) {
3626                         if (rte_eth_bond_slave_add(port_id, slave_ports.slaves[i]) != 0) {
3627                                 RTE_BOND_LOG(ERR,
3628                                              "Failed to add port %d as slave to bonded device %s",
3629                                              slave_ports.slaves[i], name);
3630                         }
3631                 }
3632
3633         } else {
3634                 RTE_BOND_LOG(INFO, "No slaves specified for bonded device %s", name);
3635                 return -1;
3636         }
3637
3638         /* Parse/set primary slave port id*/
3639         arg_count = rte_kvargs_count(kvlist, PMD_BOND_PRIMARY_SLAVE_KVARG);
3640         if (arg_count == 1) {
3641                 uint16_t primary_slave_port_id;
3642
3643                 if (rte_kvargs_process(kvlist,
3644                                        PMD_BOND_PRIMARY_SLAVE_KVARG,
3645                                        &bond_ethdev_parse_primary_slave_port_id_kvarg,
3646                                        &primary_slave_port_id) < 0) {
3647                         RTE_BOND_LOG(INFO,
3648                                      "Invalid primary slave port id specified for bonded device %s",
3649                                      name);
3650                         return -1;
3651                 }
3652
3653                 /* Set balance mode transmit policy*/
3654                 if (rte_eth_bond_primary_set(port_id, primary_slave_port_id)
3655                     != 0) {
3656                         RTE_BOND_LOG(ERR,
3657                                      "Failed to set primary slave port %d on bonded device %s",
3658                                      primary_slave_port_id, name);
3659                         return -1;
3660                 }
3661         } else if (arg_count > 1) {
3662                 RTE_BOND_LOG(INFO,
3663                              "Primary slave can be specified only once for bonded device %s",
3664                              name);
3665                 return -1;
3666         }
3667
3668         /* Parse link status monitor polling interval */
3669         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LSC_POLL_PERIOD_KVARG);
3670         if (arg_count == 1) {
3671                 uint32_t lsc_poll_interval_ms;
3672
3673                 if (rte_kvargs_process(kvlist,
3674                                        PMD_BOND_LSC_POLL_PERIOD_KVARG,
3675                                        &bond_ethdev_parse_time_ms_kvarg,
3676                                        &lsc_poll_interval_ms) < 0) {
3677                         RTE_BOND_LOG(INFO,
3678                                      "Invalid lsc polling interval value specified for bonded"
3679                                      " device %s", name);
3680                         return -1;
3681                 }
3682
3683                 if (rte_eth_bond_link_monitoring_set(port_id, lsc_poll_interval_ms)
3684                     != 0) {
3685                         RTE_BOND_LOG(ERR,
3686                                      "Failed to set lsc monitor polling interval (%u ms) on bonded device %s",
3687                                      lsc_poll_interval_ms, name);
3688                         return -1;
3689                 }
3690         } else if (arg_count > 1) {
3691                 RTE_BOND_LOG(INFO,
3692                              "LSC polling interval can be specified only once for bonded"
3693                              " device %s", name);
3694                 return -1;
3695         }
3696
3697         /* Parse link up interrupt propagation delay */
3698         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_UP_PROP_DELAY_KVARG);
3699         if (arg_count == 1) {
3700                 uint32_t link_up_delay_ms;
3701
3702                 if (rte_kvargs_process(kvlist,
3703                                        PMD_BOND_LINK_UP_PROP_DELAY_KVARG,
3704                                        &bond_ethdev_parse_time_ms_kvarg,
3705                                        &link_up_delay_ms) < 0) {
3706                         RTE_BOND_LOG(INFO,
3707                                      "Invalid link up propagation delay value specified for"
3708                                      " bonded device %s", name);
3709                         return -1;
3710                 }
3711
3712                 /* Set balance mode transmit policy*/
3713                 if (rte_eth_bond_link_up_prop_delay_set(port_id, link_up_delay_ms)
3714                     != 0) {
3715                         RTE_BOND_LOG(ERR,
3716                                      "Failed to set link up propagation delay (%u ms) on bonded"
3717                                      " device %s", link_up_delay_ms, name);
3718                         return -1;
3719                 }
3720         } else if (arg_count > 1) {
3721                 RTE_BOND_LOG(INFO,
3722                              "Link up propagation delay can be specified only once for"
3723                              " bonded device %s", name);
3724                 return -1;
3725         }
3726
3727         /* Parse link down interrupt propagation delay */
3728         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG);
3729         if (arg_count == 1) {
3730                 uint32_t link_down_delay_ms;
3731
3732                 if (rte_kvargs_process(kvlist,
3733                                        PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG,
3734                                        &bond_ethdev_parse_time_ms_kvarg,
3735                                        &link_down_delay_ms) < 0) {
3736                         RTE_BOND_LOG(INFO,
3737                                      "Invalid link down propagation delay value specified for"
3738                                      " bonded device %s", name);
3739                         return -1;
3740                 }
3741
3742                 /* Set balance mode transmit policy*/
3743                 if (rte_eth_bond_link_down_prop_delay_set(port_id, link_down_delay_ms)
3744                     != 0) {
3745                         RTE_BOND_LOG(ERR,
3746                                      "Failed to set link down propagation delay (%u ms) on bonded device %s",
3747                                      link_down_delay_ms, name);
3748                         return -1;
3749                 }
3750         } else if (arg_count > 1) {
3751                 RTE_BOND_LOG(INFO,
3752                              "Link down propagation delay can be specified only once for  bonded device %s",
3753                              name);
3754                 return -1;
3755         }
3756
3757         return 0;
3758 }
3759
3760 struct rte_vdev_driver pmd_bond_drv = {
3761         .probe = bond_probe,
3762         .remove = bond_remove,
3763 };
3764
3765 RTE_PMD_REGISTER_VDEV(net_bonding, pmd_bond_drv);
3766 RTE_PMD_REGISTER_ALIAS(net_bonding, eth_bond);
3767
3768 RTE_PMD_REGISTER_PARAM_STRING(net_bonding,
3769         "slave=<ifc> "
3770         "primary=<ifc> "
3771         "mode=[0-6] "
3772         "xmit_policy=[l2 | l23 | l34] "
3773         "agg_mode=[count | stable | bandwidth] "
3774         "socket_id=<int> "
3775         "mac=<mac addr> "
3776         "lsc_poll_period_ms=<int> "
3777         "up_delay=<int> "
3778         "down_delay=<int>");
3779
3780 RTE_LOG_REGISTER(bond_logtype, pmd.net.bonding, NOTICE);