xref: /dpdk/drivers/net/bonding/rte_eth_bond_api.c (revision 2a7bb4fdf61e9edfb7adbaecb50e728b82da9e23)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2017 Intel Corporation
3  */
4 
5 #include <string.h>
6 
7 #include <rte_mbuf.h>
8 #include <rte_malloc.h>
9 #include <rte_ethdev_driver.h>
10 #include <rte_tcp.h>
11 #include <rte_bus_vdev.h>
12 #include <rte_kvargs.h>
13 
14 #include "rte_eth_bond.h"
15 #include "rte_eth_bond_private.h"
16 #include "rte_eth_bond_8023ad_private.h"
17 
18 int
19 check_for_bonded_ethdev(const struct rte_eth_dev *eth_dev)
20 {
21 	/* Check valid pointer */
22 	if (eth_dev == NULL ||
23 		eth_dev->device == NULL ||
24 		eth_dev->device->driver == NULL ||
25 		eth_dev->device->driver->name == NULL)
26 		return -1;
27 
28 	/* return 0 if driver name matches */
29 	return eth_dev->device->driver->name != pmd_bond_drv.driver.name;
30 }
31 
32 int
33 valid_bonded_port_id(uint16_t port_id)
34 {
35 	RTE_ETH_VALID_PORTID_OR_ERR_RET(port_id, -1);
36 	return check_for_bonded_ethdev(&rte_eth_devices[port_id]);
37 }
38 
39 int
40 check_for_master_bonded_ethdev(const struct rte_eth_dev *eth_dev)
41 {
42 	int i;
43 	struct bond_dev_private *internals;
44 
45 	if (check_for_bonded_ethdev(eth_dev) != 0)
46 		return 0;
47 
48 	internals = eth_dev->data->dev_private;
49 
50 	/* Check if any of slave devices is a bonded device */
51 	for (i = 0; i < internals->slave_count; i++)
52 		if (valid_bonded_port_id(internals->slaves[i].port_id) == 0)
53 			return 1;
54 
55 	return 0;
56 }
57 
58 int
59 valid_slave_port_id(uint16_t port_id, uint8_t mode)
60 {
61 	RTE_ETH_VALID_PORTID_OR_ERR_RET(port_id, -1);
62 
63 	/* Verify that port_id refers to a non bonded port */
64 	if (check_for_bonded_ethdev(&rte_eth_devices[port_id]) == 0 &&
65 			mode == BONDING_MODE_8023AD) {
66 		RTE_BOND_LOG(ERR, "Cannot add slave to bonded device in 802.3ad"
67 				" mode as slave is also a bonded device, only "
68 				"physical devices can be support in this mode.");
69 		return -1;
70 	}
71 
72 	return 0;
73 }
74 
75 void
76 activate_slave(struct rte_eth_dev *eth_dev, uint16_t port_id)
77 {
78 	struct bond_dev_private *internals = eth_dev->data->dev_private;
79 	uint8_t active_count = internals->active_slave_count;
80 
81 	if (internals->mode == BONDING_MODE_8023AD)
82 		bond_mode_8023ad_activate_slave(eth_dev, port_id);
83 
84 	if (internals->mode == BONDING_MODE_TLB
85 			|| internals->mode == BONDING_MODE_ALB) {
86 
87 		internals->tlb_slaves_order[active_count] = port_id;
88 	}
89 
90 	RTE_ASSERT(internals->active_slave_count <
91 			(RTE_DIM(internals->active_slaves) - 1));
92 
93 	internals->active_slaves[internals->active_slave_count] = port_id;
94 	internals->active_slave_count++;
95 
96 	if (internals->mode == BONDING_MODE_TLB)
97 		bond_tlb_activate_slave(internals);
98 	if (internals->mode == BONDING_MODE_ALB)
99 		bond_mode_alb_client_list_upd(eth_dev);
100 }
101 
102 void
103 deactivate_slave(struct rte_eth_dev *eth_dev, uint16_t port_id)
104 {
105 	uint16_t slave_pos;
106 	struct bond_dev_private *internals = eth_dev->data->dev_private;
107 	uint16_t active_count = internals->active_slave_count;
108 
109 	if (internals->mode == BONDING_MODE_8023AD) {
110 		bond_mode_8023ad_stop(eth_dev);
111 		bond_mode_8023ad_deactivate_slave(eth_dev, port_id);
112 	} else if (internals->mode == BONDING_MODE_TLB
113 			|| internals->mode == BONDING_MODE_ALB)
114 		bond_tlb_disable(internals);
115 
116 	slave_pos = find_slave_by_id(internals->active_slaves, active_count,
117 			port_id);
118 
119 	/* If slave was not at the end of the list
120 	 * shift active slaves up active array list */
121 	if (slave_pos < active_count) {
122 		active_count--;
123 		memmove(internals->active_slaves + slave_pos,
124 				internals->active_slaves + slave_pos + 1,
125 				(active_count - slave_pos) *
126 					sizeof(internals->active_slaves[0]));
127 	}
128 
129 	RTE_ASSERT(active_count < RTE_DIM(internals->active_slaves));
130 	internals->active_slave_count = active_count;
131 
132 	if (eth_dev->data->dev_started) {
133 		if (internals->mode == BONDING_MODE_8023AD) {
134 			bond_mode_8023ad_start(eth_dev);
135 		} else if (internals->mode == BONDING_MODE_TLB) {
136 			bond_tlb_enable(internals);
137 		} else if (internals->mode == BONDING_MODE_ALB) {
138 			bond_tlb_enable(internals);
139 			bond_mode_alb_client_list_upd(eth_dev);
140 		}
141 	}
142 }
143 
144 int
145 rte_eth_bond_create(const char *name, uint8_t mode, uint8_t socket_id)
146 {
147 	struct bond_dev_private *internals;
148 	char devargs[52];
149 	uint16_t port_id;
150 	int ret;
151 
152 	if (name == NULL) {
153 		RTE_BOND_LOG(ERR, "Invalid name specified");
154 		return -EINVAL;
155 	}
156 
157 	ret = snprintf(devargs, sizeof(devargs),
158 		"driver=net_bonding,mode=%d,socket_id=%d", mode, socket_id);
159 	if (ret < 0 || ret >= (int)sizeof(devargs))
160 		return -ENOMEM;
161 
162 	ret = rte_vdev_init(name, devargs);
163 	if (ret)
164 		return -ENOMEM;
165 
166 	ret = rte_eth_dev_get_port_by_name(name, &port_id);
167 	RTE_ASSERT(!ret);
168 
169 	/*
170 	 * To make bond_ethdev_configure() happy we need to free the
171 	 * internals->kvlist here.
172 	 *
173 	 * Also see comment in bond_ethdev_configure().
174 	 */
175 	internals = rte_eth_devices[port_id].data->dev_private;
176 	rte_kvargs_free(internals->kvlist);
177 	internals->kvlist = NULL;
178 
179 	return port_id;
180 }
181 
182 int
183 rte_eth_bond_free(const char *name)
184 {
185 	return rte_vdev_uninit(name);
186 }
187 
188 static int
189 slave_vlan_filter_set(uint16_t bonded_port_id, uint16_t slave_port_id)
190 {
191 	struct rte_eth_dev *bonded_eth_dev;
192 	struct bond_dev_private *internals;
193 	int found;
194 	int res = 0;
195 	uint64_t slab = 0;
196 	uint32_t pos = 0;
197 	uint16_t first;
198 
199 	bonded_eth_dev = &rte_eth_devices[bonded_port_id];
200 	if ((bonded_eth_dev->data->dev_conf.rxmode.offloads &
201 			DEV_RX_OFFLOAD_VLAN_FILTER) == 0)
202 		return 0;
203 
204 	internals = bonded_eth_dev->data->dev_private;
205 	found = rte_bitmap_scan(internals->vlan_filter_bmp, &pos, &slab);
206 	first = pos;
207 
208 	if (!found)
209 		return 0;
210 
211 	do {
212 		uint32_t i;
213 		uint64_t mask;
214 
215 		for (i = 0, mask = 1;
216 		     i < RTE_BITMAP_SLAB_BIT_SIZE;
217 		     i ++, mask <<= 1) {
218 			if (unlikely(slab & mask)) {
219 				uint16_t vlan_id = pos + i;
220 
221 				res = rte_eth_dev_vlan_filter(slave_port_id,
222 							      vlan_id, 1);
223 			}
224 		}
225 		found = rte_bitmap_scan(internals->vlan_filter_bmp,
226 					&pos, &slab);
227 	} while (found && first != pos && res == 0);
228 
229 	return res;
230 }
231 
232 static int
233 slave_rte_flow_prepare(uint16_t slave_id, struct bond_dev_private *internals)
234 {
235 	struct rte_flow *flow;
236 	struct rte_flow_error ferror;
237 	uint16_t slave_port_id = internals->slaves[slave_id].port_id;
238 
239 	if (internals->flow_isolated_valid != 0) {
240 		rte_eth_dev_stop(slave_port_id);
241 		if (rte_flow_isolate(slave_port_id, internals->flow_isolated,
242 		    &ferror)) {
243 			RTE_BOND_LOG(ERR, "rte_flow_isolate failed for slave"
244 				     " %d: %s", slave_id, ferror.message ?
245 				     ferror.message : "(no stated reason)");
246 			return -1;
247 		}
248 	}
249 	TAILQ_FOREACH(flow, &internals->flow_list, next) {
250 		flow->flows[slave_id] = rte_flow_create(slave_port_id,
251 							flow->rule.attr,
252 							flow->rule.pattern,
253 							flow->rule.actions,
254 							&ferror);
255 		if (flow->flows[slave_id] == NULL) {
256 			RTE_BOND_LOG(ERR, "Cannot create flow for slave"
257 				     " %d: %s", slave_id,
258 				     ferror.message ? ferror.message :
259 				     "(no stated reason)");
260 			/* Destroy successful bond flows from the slave */
261 			TAILQ_FOREACH(flow, &internals->flow_list, next) {
262 				if (flow->flows[slave_id] != NULL) {
263 					rte_flow_destroy(slave_port_id,
264 							 flow->flows[slave_id],
265 							 &ferror);
266 					flow->flows[slave_id] = NULL;
267 				}
268 			}
269 			return -1;
270 		}
271 	}
272 	return 0;
273 }
274 
275 static void
276 eth_bond_slave_inherit_dev_info_rx_first(struct bond_dev_private *internals,
277 					 const struct rte_eth_dev_info *di)
278 {
279 	struct rte_eth_rxconf *rxconf_i = &internals->default_rxconf;
280 
281 	internals->reta_size = di->reta_size;
282 
283 	/* Inherit Rx offload capabilities from the first slave device */
284 	internals->rx_offload_capa = di->rx_offload_capa;
285 	internals->rx_queue_offload_capa = di->rx_queue_offload_capa;
286 	internals->flow_type_rss_offloads = di->flow_type_rss_offloads;
287 
288 	/* Inherit maximum Rx packet size from the first slave device */
289 	internals->candidate_max_rx_pktlen = di->max_rx_pktlen;
290 
291 	/* Inherit default Rx queue settings from the first slave device */
292 	memcpy(rxconf_i, &di->default_rxconf, sizeof(*rxconf_i));
293 
294 	/*
295 	 * Turn off descriptor prefetch and writeback by default for all
296 	 * slave devices. Applications may tweak this setting if need be.
297 	 */
298 	rxconf_i->rx_thresh.pthresh = 0;
299 	rxconf_i->rx_thresh.hthresh = 0;
300 	rxconf_i->rx_thresh.wthresh = 0;
301 
302 	/* Setting this to zero should effectively enable default values */
303 	rxconf_i->rx_free_thresh = 0;
304 
305 	/* Disable deferred start by default for all slave devices */
306 	rxconf_i->rx_deferred_start = 0;
307 }
308 
309 static void
310 eth_bond_slave_inherit_dev_info_tx_first(struct bond_dev_private *internals,
311 					 const struct rte_eth_dev_info *di)
312 {
313 	struct rte_eth_txconf *txconf_i = &internals->default_txconf;
314 
315 	/* Inherit Tx offload capabilities from the first slave device */
316 	internals->tx_offload_capa = di->tx_offload_capa;
317 	internals->tx_queue_offload_capa = di->tx_queue_offload_capa;
318 
319 	/* Inherit default Tx queue settings from the first slave device */
320 	memcpy(txconf_i, &di->default_txconf, sizeof(*txconf_i));
321 
322 	/*
323 	 * Turn off descriptor prefetch and writeback by default for all
324 	 * slave devices. Applications may tweak this setting if need be.
325 	 */
326 	txconf_i->tx_thresh.pthresh = 0;
327 	txconf_i->tx_thresh.hthresh = 0;
328 	txconf_i->tx_thresh.wthresh = 0;
329 
330 	/*
331 	 * Setting these parameters to zero assumes that default
332 	 * values will be configured implicitly by slave devices.
333 	 */
334 	txconf_i->tx_free_thresh = 0;
335 	txconf_i->tx_rs_thresh = 0;
336 
337 	/* Disable deferred start by default for all slave devices */
338 	txconf_i->tx_deferred_start = 0;
339 }
340 
341 static void
342 eth_bond_slave_inherit_dev_info_rx_next(struct bond_dev_private *internals,
343 					const struct rte_eth_dev_info *di)
344 {
345 	struct rte_eth_rxconf *rxconf_i = &internals->default_rxconf;
346 	const struct rte_eth_rxconf *rxconf = &di->default_rxconf;
347 
348 	internals->rx_offload_capa &= di->rx_offload_capa;
349 	internals->rx_queue_offload_capa &= di->rx_queue_offload_capa;
350 	internals->flow_type_rss_offloads &= di->flow_type_rss_offloads;
351 
352 	/*
353 	 * If at least one slave device suggests enabling this
354 	 * setting by default, enable it for all slave devices
355 	 * since disabling it may not be necessarily supported.
356 	 */
357 	if (rxconf->rx_drop_en == 1)
358 		rxconf_i->rx_drop_en = 1;
359 
360 	/*
361 	 * Adding a new slave device may cause some of previously inherited
362 	 * offloads to be withdrawn from the internal rx_queue_offload_capa
363 	 * value. Thus, the new internal value of default Rx queue offloads
364 	 * has to be masked by rx_queue_offload_capa to make sure that only
365 	 * commonly supported offloads are preserved from both the previous
366 	 * value and the value being inhereted from the new slave device.
367 	 */
368 	rxconf_i->offloads = (rxconf_i->offloads | rxconf->offloads) &
369 			     internals->rx_queue_offload_capa;
370 
371 	/*
372 	 * RETA size is GCD of all slaves RETA sizes, so, if all sizes will be
373 	 * the power of 2, the lower one is GCD
374 	 */
375 	if (internals->reta_size > di->reta_size)
376 		internals->reta_size = di->reta_size;
377 
378 	if (!internals->max_rx_pktlen &&
379 	    di->max_rx_pktlen < internals->candidate_max_rx_pktlen)
380 		internals->candidate_max_rx_pktlen = di->max_rx_pktlen;
381 }
382 
383 static void
384 eth_bond_slave_inherit_dev_info_tx_next(struct bond_dev_private *internals,
385 					const struct rte_eth_dev_info *di)
386 {
387 	struct rte_eth_txconf *txconf_i = &internals->default_txconf;
388 	const struct rte_eth_txconf *txconf = &di->default_txconf;
389 
390 	internals->tx_offload_capa &= di->tx_offload_capa;
391 	internals->tx_queue_offload_capa &= di->tx_queue_offload_capa;
392 
393 	/*
394 	 * Adding a new slave device may cause some of previously inherited
395 	 * offloads to be withdrawn from the internal tx_queue_offload_capa
396 	 * value. Thus, the new internal value of default Tx queue offloads
397 	 * has to be masked by tx_queue_offload_capa to make sure that only
398 	 * commonly supported offloads are preserved from both the previous
399 	 * value and the value being inhereted from the new slave device.
400 	 */
401 	txconf_i->offloads = (txconf_i->offloads | txconf->offloads) &
402 			     internals->tx_queue_offload_capa;
403 }
404 
405 static void
406 eth_bond_slave_inherit_desc_lim_first(struct rte_eth_desc_lim *bond_desc_lim,
407 		const struct rte_eth_desc_lim *slave_desc_lim)
408 {
409 	memcpy(bond_desc_lim, slave_desc_lim, sizeof(*bond_desc_lim));
410 }
411 
412 static int
413 eth_bond_slave_inherit_desc_lim_next(struct rte_eth_desc_lim *bond_desc_lim,
414 		const struct rte_eth_desc_lim *slave_desc_lim)
415 {
416 	bond_desc_lim->nb_max = RTE_MIN(bond_desc_lim->nb_max,
417 					slave_desc_lim->nb_max);
418 	bond_desc_lim->nb_min = RTE_MAX(bond_desc_lim->nb_min,
419 					slave_desc_lim->nb_min);
420 	bond_desc_lim->nb_align = RTE_MAX(bond_desc_lim->nb_align,
421 					  slave_desc_lim->nb_align);
422 
423 	if (bond_desc_lim->nb_min > bond_desc_lim->nb_max ||
424 	    bond_desc_lim->nb_align > bond_desc_lim->nb_max) {
425 		RTE_BOND_LOG(ERR, "Failed to inherit descriptor limits");
426 		return -EINVAL;
427 	}
428 
429 	/* Treat maximum number of segments equal to 0 as unspecified */
430 	if (slave_desc_lim->nb_seg_max != 0 &&
431 	    (bond_desc_lim->nb_seg_max == 0 ||
432 	     slave_desc_lim->nb_seg_max < bond_desc_lim->nb_seg_max))
433 		bond_desc_lim->nb_seg_max = slave_desc_lim->nb_seg_max;
434 	if (slave_desc_lim->nb_mtu_seg_max != 0 &&
435 	    (bond_desc_lim->nb_mtu_seg_max == 0 ||
436 	     slave_desc_lim->nb_mtu_seg_max < bond_desc_lim->nb_mtu_seg_max))
437 		bond_desc_lim->nb_mtu_seg_max = slave_desc_lim->nb_mtu_seg_max;
438 
439 	return 0;
440 }
441 
442 static int
443 __eth_bond_slave_add_lock_free(uint16_t bonded_port_id, uint16_t slave_port_id)
444 {
445 	struct rte_eth_dev *bonded_eth_dev, *slave_eth_dev;
446 	struct bond_dev_private *internals;
447 	struct rte_eth_link link_props;
448 	struct rte_eth_dev_info dev_info;
449 
450 	bonded_eth_dev = &rte_eth_devices[bonded_port_id];
451 	internals = bonded_eth_dev->data->dev_private;
452 
453 	if (valid_slave_port_id(slave_port_id, internals->mode) != 0)
454 		return -1;
455 
456 	slave_eth_dev = &rte_eth_devices[slave_port_id];
457 	if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_BONDED_SLAVE) {
458 		RTE_BOND_LOG(ERR, "Slave device is already a slave of a bonded device");
459 		return -1;
460 	}
461 
462 	rte_eth_dev_info_get(slave_port_id, &dev_info);
463 	if (dev_info.max_rx_pktlen < internals->max_rx_pktlen) {
464 		RTE_BOND_LOG(ERR, "Slave (port %u) max_rx_pktlen too small",
465 			     slave_port_id);
466 		return -1;
467 	}
468 
469 	slave_add(internals, slave_eth_dev);
470 
471 	/* We need to store slaves reta_size to be able to synchronize RETA for all
472 	 * slave devices even if its sizes are different.
473 	 */
474 	internals->slaves[internals->slave_count].reta_size = dev_info.reta_size;
475 
476 	if (internals->slave_count < 1) {
477 		/* if MAC is not user defined then use MAC of first slave add to
478 		 * bonded device */
479 		if (!internals->user_defined_mac) {
480 			if (mac_address_set(bonded_eth_dev,
481 					    slave_eth_dev->data->mac_addrs)) {
482 				RTE_BOND_LOG(ERR, "Failed to set MAC address");
483 				return -1;
484 			}
485 		}
486 
487 		/* Inherit eth dev link properties from first slave */
488 		link_properties_set(bonded_eth_dev,
489 				&(slave_eth_dev->data->dev_link));
490 
491 		/* Make primary slave */
492 		internals->primary_port = slave_port_id;
493 		internals->current_primary_port = slave_port_id;
494 
495 		/* Inherit queues settings from first slave */
496 		internals->nb_rx_queues = slave_eth_dev->data->nb_rx_queues;
497 		internals->nb_tx_queues = slave_eth_dev->data->nb_tx_queues;
498 
499 		eth_bond_slave_inherit_dev_info_rx_first(internals, &dev_info);
500 		eth_bond_slave_inherit_dev_info_tx_first(internals, &dev_info);
501 
502 		eth_bond_slave_inherit_desc_lim_first(&internals->rx_desc_lim,
503 						      &dev_info.rx_desc_lim);
504 		eth_bond_slave_inherit_desc_lim_first(&internals->tx_desc_lim,
505 						      &dev_info.tx_desc_lim);
506 	} else {
507 		int ret;
508 
509 		eth_bond_slave_inherit_dev_info_rx_next(internals, &dev_info);
510 		eth_bond_slave_inherit_dev_info_tx_next(internals, &dev_info);
511 
512 		ret = eth_bond_slave_inherit_desc_lim_next(
513 				&internals->rx_desc_lim, &dev_info.rx_desc_lim);
514 		if (ret != 0)
515 			return ret;
516 
517 		ret = eth_bond_slave_inherit_desc_lim_next(
518 				&internals->tx_desc_lim, &dev_info.tx_desc_lim);
519 		if (ret != 0)
520 			return ret;
521 	}
522 
523 	bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf &=
524 			internals->flow_type_rss_offloads;
525 
526 	if (slave_rte_flow_prepare(internals->slave_count, internals) != 0) {
527 		RTE_BOND_LOG(ERR, "Failed to prepare new slave flows: port=%d",
528 			     slave_port_id);
529 		return -1;
530 	}
531 
532 	/* Add additional MAC addresses to the slave */
533 	if (slave_add_mac_addresses(bonded_eth_dev, slave_port_id) != 0) {
534 		RTE_BOND_LOG(ERR, "Failed to add mac address(es) to slave %hu",
535 				slave_port_id);
536 		return -1;
537 	}
538 
539 	internals->slave_count++;
540 
541 	if (bonded_eth_dev->data->dev_started) {
542 		if (slave_configure(bonded_eth_dev, slave_eth_dev) != 0) {
543 			internals->slave_count--;
544 			RTE_BOND_LOG(ERR, "rte_bond_slaves_configure: port=%d",
545 					slave_port_id);
546 			return -1;
547 		}
548 	}
549 
550 	/* Add slave details to bonded device */
551 	slave_eth_dev->data->dev_flags |= RTE_ETH_DEV_BONDED_SLAVE;
552 
553 	/* Update all slave devices MACs */
554 	mac_address_slaves_update(bonded_eth_dev);
555 
556 	/* Register link status change callback with bonded device pointer as
557 	 * argument*/
558 	rte_eth_dev_callback_register(slave_port_id, RTE_ETH_EVENT_INTR_LSC,
559 			bond_ethdev_lsc_event_callback, &bonded_eth_dev->data->port_id);
560 
561 	/* If bonded device is started then we can add the slave to our active
562 	 * slave array */
563 	if (bonded_eth_dev->data->dev_started) {
564 		rte_eth_link_get_nowait(slave_port_id, &link_props);
565 
566 		 if (link_props.link_status == ETH_LINK_UP) {
567 			if (internals->active_slave_count == 0 &&
568 			    !internals->user_defined_primary_port)
569 				bond_ethdev_primary_set(internals,
570 							slave_port_id);
571 		}
572 	}
573 
574 	slave_vlan_filter_set(bonded_port_id, slave_port_id);
575 
576 	return 0;
577 
578 }
579 
580 int
581 rte_eth_bond_slave_add(uint16_t bonded_port_id, uint16_t slave_port_id)
582 {
583 	struct rte_eth_dev *bonded_eth_dev;
584 	struct bond_dev_private *internals;
585 
586 	int retval;
587 
588 	/* Verify that port id's are valid bonded and slave ports */
589 	if (valid_bonded_port_id(bonded_port_id) != 0)
590 		return -1;
591 
592 	bonded_eth_dev = &rte_eth_devices[bonded_port_id];
593 	internals = bonded_eth_dev->data->dev_private;
594 
595 	rte_spinlock_lock(&internals->lock);
596 
597 	retval = __eth_bond_slave_add_lock_free(bonded_port_id, slave_port_id);
598 
599 	rte_spinlock_unlock(&internals->lock);
600 
601 	return retval;
602 }
603 
604 static int
605 __eth_bond_slave_remove_lock_free(uint16_t bonded_port_id,
606 				   uint16_t slave_port_id)
607 {
608 	struct rte_eth_dev *bonded_eth_dev;
609 	struct bond_dev_private *internals;
610 	struct rte_eth_dev *slave_eth_dev;
611 	struct rte_flow_error flow_error;
612 	struct rte_flow *flow;
613 	int i, slave_idx;
614 
615 	bonded_eth_dev = &rte_eth_devices[bonded_port_id];
616 	internals = bonded_eth_dev->data->dev_private;
617 
618 	if (valid_slave_port_id(slave_port_id, internals->mode) < 0)
619 		return -1;
620 
621 	/* first remove from active slave list */
622 	slave_idx = find_slave_by_id(internals->active_slaves,
623 		internals->active_slave_count, slave_port_id);
624 
625 	if (slave_idx < internals->active_slave_count)
626 		deactivate_slave(bonded_eth_dev, slave_port_id);
627 
628 	slave_idx = -1;
629 	/* now find in slave list */
630 	for (i = 0; i < internals->slave_count; i++)
631 		if (internals->slaves[i].port_id == slave_port_id) {
632 			slave_idx = i;
633 			break;
634 		}
635 
636 	if (slave_idx < 0) {
637 		RTE_BOND_LOG(ERR, "Couldn't find slave in port list, slave count %d",
638 				internals->slave_count);
639 		return -1;
640 	}
641 
642 	/* Un-register link status change callback with bonded device pointer as
643 	 * argument*/
644 	rte_eth_dev_callback_unregister(slave_port_id, RTE_ETH_EVENT_INTR_LSC,
645 			bond_ethdev_lsc_event_callback,
646 			&rte_eth_devices[bonded_port_id].data->port_id);
647 
648 	/* Restore original MAC address of slave device */
649 	rte_eth_dev_default_mac_addr_set(slave_port_id,
650 			&(internals->slaves[slave_idx].persisted_mac_addr));
651 
652 	/* remove additional MAC addresses from the slave */
653 	slave_remove_mac_addresses(bonded_eth_dev, slave_port_id);
654 
655 	/*
656 	 * Remove bond device flows from slave device.
657 	 * Note: don't restore flow isolate mode.
658 	 */
659 	TAILQ_FOREACH(flow, &internals->flow_list, next) {
660 		if (flow->flows[slave_idx] != NULL) {
661 			rte_flow_destroy(slave_port_id, flow->flows[slave_idx],
662 					 &flow_error);
663 			flow->flows[slave_idx] = NULL;
664 		}
665 	}
666 
667 	slave_eth_dev = &rte_eth_devices[slave_port_id];
668 	slave_remove(internals, slave_eth_dev);
669 	slave_eth_dev->data->dev_flags &= (~RTE_ETH_DEV_BONDED_SLAVE);
670 
671 	/*  first slave in the active list will be the primary by default,
672 	 *  otherwise use first device in list */
673 	if (internals->current_primary_port == slave_port_id) {
674 		if (internals->active_slave_count > 0)
675 			internals->current_primary_port = internals->active_slaves[0];
676 		else if (internals->slave_count > 0)
677 			internals->current_primary_port = internals->slaves[0].port_id;
678 		else
679 			internals->primary_port = 0;
680 	}
681 
682 	if (internals->active_slave_count < 1) {
683 		/* if no slaves are any longer attached to bonded device and MAC is not
684 		 * user defined then clear MAC of bonded device as it will be reset
685 		 * when a new slave is added */
686 		if (internals->slave_count < 1 && !internals->user_defined_mac)
687 			memset(rte_eth_devices[bonded_port_id].data->mac_addrs, 0,
688 					sizeof(*(rte_eth_devices[bonded_port_id].data->mac_addrs)));
689 	}
690 	if (internals->slave_count == 0) {
691 		internals->rx_offload_capa = 0;
692 		internals->tx_offload_capa = 0;
693 		internals->rx_queue_offload_capa = 0;
694 		internals->tx_queue_offload_capa = 0;
695 		internals->flow_type_rss_offloads = ETH_RSS_PROTO_MASK;
696 		internals->reta_size = 0;
697 		internals->candidate_max_rx_pktlen = 0;
698 		internals->max_rx_pktlen = 0;
699 	}
700 	return 0;
701 }
702 
703 int
704 rte_eth_bond_slave_remove(uint16_t bonded_port_id, uint16_t slave_port_id)
705 {
706 	struct rte_eth_dev *bonded_eth_dev;
707 	struct bond_dev_private *internals;
708 	int retval;
709 
710 	if (valid_bonded_port_id(bonded_port_id) != 0)
711 		return -1;
712 
713 	bonded_eth_dev = &rte_eth_devices[bonded_port_id];
714 	internals = bonded_eth_dev->data->dev_private;
715 
716 	rte_spinlock_lock(&internals->lock);
717 
718 	retval = __eth_bond_slave_remove_lock_free(bonded_port_id, slave_port_id);
719 
720 	rte_spinlock_unlock(&internals->lock);
721 
722 	return retval;
723 }
724 
725 int
726 rte_eth_bond_mode_set(uint16_t bonded_port_id, uint8_t mode)
727 {
728 	struct rte_eth_dev *bonded_eth_dev;
729 
730 	if (valid_bonded_port_id(bonded_port_id) != 0)
731 		return -1;
732 
733 	bonded_eth_dev = &rte_eth_devices[bonded_port_id];
734 
735 	if (check_for_master_bonded_ethdev(bonded_eth_dev) != 0 &&
736 			mode == BONDING_MODE_8023AD)
737 		return -1;
738 
739 	return bond_ethdev_mode_set(bonded_eth_dev, mode);
740 }
741 
742 int
743 rte_eth_bond_mode_get(uint16_t bonded_port_id)
744 {
745 	struct bond_dev_private *internals;
746 
747 	if (valid_bonded_port_id(bonded_port_id) != 0)
748 		return -1;
749 
750 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
751 
752 	return internals->mode;
753 }
754 
755 int
756 rte_eth_bond_primary_set(uint16_t bonded_port_id, uint16_t slave_port_id)
757 {
758 	struct bond_dev_private *internals;
759 
760 	if (valid_bonded_port_id(bonded_port_id) != 0)
761 		return -1;
762 
763 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
764 
765 	if (valid_slave_port_id(slave_port_id, internals->mode) != 0)
766 		return -1;
767 
768 	internals->user_defined_primary_port = 1;
769 	internals->primary_port = slave_port_id;
770 
771 	bond_ethdev_primary_set(internals, slave_port_id);
772 
773 	return 0;
774 }
775 
776 int
777 rte_eth_bond_primary_get(uint16_t bonded_port_id)
778 {
779 	struct bond_dev_private *internals;
780 
781 	if (valid_bonded_port_id(bonded_port_id) != 0)
782 		return -1;
783 
784 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
785 
786 	if (internals->slave_count < 1)
787 		return -1;
788 
789 	return internals->current_primary_port;
790 }
791 
792 int
793 rte_eth_bond_slaves_get(uint16_t bonded_port_id, uint16_t slaves[],
794 			uint16_t len)
795 {
796 	struct bond_dev_private *internals;
797 	uint8_t i;
798 
799 	if (valid_bonded_port_id(bonded_port_id) != 0)
800 		return -1;
801 
802 	if (slaves == NULL)
803 		return -1;
804 
805 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
806 
807 	if (internals->slave_count > len)
808 		return -1;
809 
810 	for (i = 0; i < internals->slave_count; i++)
811 		slaves[i] = internals->slaves[i].port_id;
812 
813 	return internals->slave_count;
814 }
815 
816 int
817 rte_eth_bond_active_slaves_get(uint16_t bonded_port_id, uint16_t slaves[],
818 		uint16_t len)
819 {
820 	struct bond_dev_private *internals;
821 
822 	if (valid_bonded_port_id(bonded_port_id) != 0)
823 		return -1;
824 
825 	if (slaves == NULL)
826 		return -1;
827 
828 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
829 
830 	if (internals->active_slave_count > len)
831 		return -1;
832 
833 	memcpy(slaves, internals->active_slaves,
834 	internals->active_slave_count * sizeof(internals->active_slaves[0]));
835 
836 	return internals->active_slave_count;
837 }
838 
839 int
840 rte_eth_bond_mac_address_set(uint16_t bonded_port_id,
841 		struct ether_addr *mac_addr)
842 {
843 	struct rte_eth_dev *bonded_eth_dev;
844 	struct bond_dev_private *internals;
845 
846 	if (valid_bonded_port_id(bonded_port_id) != 0)
847 		return -1;
848 
849 	bonded_eth_dev = &rte_eth_devices[bonded_port_id];
850 	internals = bonded_eth_dev->data->dev_private;
851 
852 	/* Set MAC Address of Bonded Device */
853 	if (mac_address_set(bonded_eth_dev, mac_addr))
854 		return -1;
855 
856 	internals->user_defined_mac = 1;
857 
858 	/* Update all slave devices MACs*/
859 	if (internals->slave_count > 0)
860 		return mac_address_slaves_update(bonded_eth_dev);
861 
862 	return 0;
863 }
864 
865 int
866 rte_eth_bond_mac_address_reset(uint16_t bonded_port_id)
867 {
868 	struct rte_eth_dev *bonded_eth_dev;
869 	struct bond_dev_private *internals;
870 
871 	if (valid_bonded_port_id(bonded_port_id) != 0)
872 		return -1;
873 
874 	bonded_eth_dev = &rte_eth_devices[bonded_port_id];
875 	internals = bonded_eth_dev->data->dev_private;
876 
877 	internals->user_defined_mac = 0;
878 
879 	if (internals->slave_count > 0) {
880 		int slave_port;
881 		/* Get the primary slave location based on the primary port
882 		 * number as, while slave_add(), we will keep the primary
883 		 * slave based on slave_count,but not based on the primary port.
884 		 */
885 		for (slave_port = 0; slave_port < internals->slave_count;
886 		     slave_port++) {
887 			if (internals->slaves[slave_port].port_id ==
888 			    internals->primary_port)
889 				break;
890 		}
891 
892 		/* Set MAC Address of Bonded Device */
893 		if (mac_address_set(bonded_eth_dev,
894 			&internals->slaves[slave_port].persisted_mac_addr)
895 				!= 0) {
896 			RTE_BOND_LOG(ERR, "Failed to set MAC address on bonded device");
897 			return -1;
898 		}
899 		/* Update all slave devices MAC addresses */
900 		return mac_address_slaves_update(bonded_eth_dev);
901 	}
902 	/* No need to update anything as no slaves present */
903 	return 0;
904 }
905 
906 int
907 rte_eth_bond_xmit_policy_set(uint16_t bonded_port_id, uint8_t policy)
908 {
909 	struct bond_dev_private *internals;
910 
911 	if (valid_bonded_port_id(bonded_port_id) != 0)
912 		return -1;
913 
914 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
915 
916 	switch (policy) {
917 	case BALANCE_XMIT_POLICY_LAYER2:
918 		internals->balance_xmit_policy = policy;
919 		internals->burst_xmit_hash = burst_xmit_l2_hash;
920 		break;
921 	case BALANCE_XMIT_POLICY_LAYER23:
922 		internals->balance_xmit_policy = policy;
923 		internals->burst_xmit_hash = burst_xmit_l23_hash;
924 		break;
925 	case BALANCE_XMIT_POLICY_LAYER34:
926 		internals->balance_xmit_policy = policy;
927 		internals->burst_xmit_hash = burst_xmit_l34_hash;
928 		break;
929 
930 	default:
931 		return -1;
932 	}
933 	return 0;
934 }
935 
936 int
937 rte_eth_bond_xmit_policy_get(uint16_t bonded_port_id)
938 {
939 	struct bond_dev_private *internals;
940 
941 	if (valid_bonded_port_id(bonded_port_id) != 0)
942 		return -1;
943 
944 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
945 
946 	return internals->balance_xmit_policy;
947 }
948 
949 int
950 rte_eth_bond_link_monitoring_set(uint16_t bonded_port_id, uint32_t internal_ms)
951 {
952 	struct bond_dev_private *internals;
953 
954 	if (valid_bonded_port_id(bonded_port_id) != 0)
955 		return -1;
956 
957 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
958 	internals->link_status_polling_interval_ms = internal_ms;
959 
960 	return 0;
961 }
962 
963 int
964 rte_eth_bond_link_monitoring_get(uint16_t bonded_port_id)
965 {
966 	struct bond_dev_private *internals;
967 
968 	if (valid_bonded_port_id(bonded_port_id) != 0)
969 		return -1;
970 
971 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
972 
973 	return internals->link_status_polling_interval_ms;
974 }
975 
976 int
977 rte_eth_bond_link_down_prop_delay_set(uint16_t bonded_port_id,
978 				       uint32_t delay_ms)
979 
980 {
981 	struct bond_dev_private *internals;
982 
983 	if (valid_bonded_port_id(bonded_port_id) != 0)
984 		return -1;
985 
986 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
987 	internals->link_down_delay_ms = delay_ms;
988 
989 	return 0;
990 }
991 
992 int
993 rte_eth_bond_link_down_prop_delay_get(uint16_t bonded_port_id)
994 {
995 	struct bond_dev_private *internals;
996 
997 	if (valid_bonded_port_id(bonded_port_id) != 0)
998 		return -1;
999 
1000 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
1001 
1002 	return internals->link_down_delay_ms;
1003 }
1004 
1005 int
1006 rte_eth_bond_link_up_prop_delay_set(uint16_t bonded_port_id, uint32_t delay_ms)
1007 
1008 {
1009 	struct bond_dev_private *internals;
1010 
1011 	if (valid_bonded_port_id(bonded_port_id) != 0)
1012 		return -1;
1013 
1014 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
1015 	internals->link_up_delay_ms = delay_ms;
1016 
1017 	return 0;
1018 }
1019 
1020 int
1021 rte_eth_bond_link_up_prop_delay_get(uint16_t bonded_port_id)
1022 {
1023 	struct bond_dev_private *internals;
1024 
1025 	if (valid_bonded_port_id(bonded_port_id) != 0)
1026 		return -1;
1027 
1028 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
1029 
1030 	return internals->link_up_delay_ms;
1031 }
1032