xref: /dpdk/drivers/net/bonding/rte_eth_bond_api.c (revision 1cde1b9a9b4dbf31cb5e5ccdfc5da3cb079f43a2)
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 	uint16_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 	/* Resetting active_slave when reaches to max
133 	 * no of slaves in active list
134 	 */
135 	if (internals->active_slave >= active_count)
136 		internals->active_slave = 0;
137 
138 	if (eth_dev->data->dev_started) {
139 		if (internals->mode == BONDING_MODE_8023AD) {
140 			bond_mode_8023ad_start(eth_dev);
141 		} else if (internals->mode == BONDING_MODE_TLB) {
142 			bond_tlb_enable(internals);
143 		} else if (internals->mode == BONDING_MODE_ALB) {
144 			bond_tlb_enable(internals);
145 			bond_mode_alb_client_list_upd(eth_dev);
146 		}
147 	}
148 }
149 
150 int
151 rte_eth_bond_create(const char *name, uint8_t mode, uint8_t socket_id)
152 {
153 	struct bond_dev_private *internals;
154 	char devargs[52];
155 	uint16_t port_id;
156 	int ret;
157 
158 	if (name == NULL) {
159 		RTE_BOND_LOG(ERR, "Invalid name specified");
160 		return -EINVAL;
161 	}
162 
163 	ret = snprintf(devargs, sizeof(devargs),
164 		"driver=net_bonding,mode=%d,socket_id=%d", mode, socket_id);
165 	if (ret < 0 || ret >= (int)sizeof(devargs))
166 		return -ENOMEM;
167 
168 	ret = rte_vdev_init(name, devargs);
169 	if (ret)
170 		return -ENOMEM;
171 
172 	ret = rte_eth_dev_get_port_by_name(name, &port_id);
173 	RTE_ASSERT(!ret);
174 
175 	/*
176 	 * To make bond_ethdev_configure() happy we need to free the
177 	 * internals->kvlist here.
178 	 *
179 	 * Also see comment in bond_ethdev_configure().
180 	 */
181 	internals = rte_eth_devices[port_id].data->dev_private;
182 	rte_kvargs_free(internals->kvlist);
183 	internals->kvlist = NULL;
184 
185 	return port_id;
186 }
187 
188 int
189 rte_eth_bond_free(const char *name)
190 {
191 	return rte_vdev_uninit(name);
192 }
193 
194 static int
195 slave_vlan_filter_set(uint16_t bonded_port_id, uint16_t slave_port_id)
196 {
197 	struct rte_eth_dev *bonded_eth_dev;
198 	struct bond_dev_private *internals;
199 	int found;
200 	int res = 0;
201 	uint64_t slab = 0;
202 	uint32_t pos = 0;
203 	uint16_t first;
204 
205 	bonded_eth_dev = &rte_eth_devices[bonded_port_id];
206 	if ((bonded_eth_dev->data->dev_conf.rxmode.offloads &
207 			DEV_RX_OFFLOAD_VLAN_FILTER) == 0)
208 		return 0;
209 
210 	internals = bonded_eth_dev->data->dev_private;
211 	found = rte_bitmap_scan(internals->vlan_filter_bmp, &pos, &slab);
212 	first = pos;
213 
214 	if (!found)
215 		return 0;
216 
217 	do {
218 		uint32_t i;
219 		uint64_t mask;
220 
221 		for (i = 0, mask = 1;
222 		     i < RTE_BITMAP_SLAB_BIT_SIZE;
223 		     i ++, mask <<= 1) {
224 			if (unlikely(slab & mask)) {
225 				uint16_t vlan_id = pos + i;
226 
227 				res = rte_eth_dev_vlan_filter(slave_port_id,
228 							      vlan_id, 1);
229 			}
230 		}
231 		found = rte_bitmap_scan(internals->vlan_filter_bmp,
232 					&pos, &slab);
233 	} while (found && first != pos && res == 0);
234 
235 	return res;
236 }
237 
238 static int
239 slave_rte_flow_prepare(uint16_t slave_id, struct bond_dev_private *internals)
240 {
241 	struct rte_flow *flow;
242 	struct rte_flow_error ferror;
243 	uint16_t slave_port_id = internals->slaves[slave_id].port_id;
244 
245 	if (internals->flow_isolated_valid != 0) {
246 		rte_eth_dev_stop(slave_port_id);
247 		if (rte_flow_isolate(slave_port_id, internals->flow_isolated,
248 		    &ferror)) {
249 			RTE_BOND_LOG(ERR, "rte_flow_isolate failed for slave"
250 				     " %d: %s", slave_id, ferror.message ?
251 				     ferror.message : "(no stated reason)");
252 			return -1;
253 		}
254 	}
255 	TAILQ_FOREACH(flow, &internals->flow_list, next) {
256 		flow->flows[slave_id] = rte_flow_create(slave_port_id,
257 							flow->rule.attr,
258 							flow->rule.pattern,
259 							flow->rule.actions,
260 							&ferror);
261 		if (flow->flows[slave_id] == NULL) {
262 			RTE_BOND_LOG(ERR, "Cannot create flow for slave"
263 				     " %d: %s", slave_id,
264 				     ferror.message ? ferror.message :
265 				     "(no stated reason)");
266 			/* Destroy successful bond flows from the slave */
267 			TAILQ_FOREACH(flow, &internals->flow_list, next) {
268 				if (flow->flows[slave_id] != NULL) {
269 					rte_flow_destroy(slave_port_id,
270 							 flow->flows[slave_id],
271 							 &ferror);
272 					flow->flows[slave_id] = NULL;
273 				}
274 			}
275 			return -1;
276 		}
277 	}
278 	return 0;
279 }
280 
281 static void
282 eth_bond_slave_inherit_dev_info_rx_first(struct bond_dev_private *internals,
283 					 const struct rte_eth_dev_info *di)
284 {
285 	struct rte_eth_rxconf *rxconf_i = &internals->default_rxconf;
286 
287 	internals->reta_size = di->reta_size;
288 
289 	/* Inherit Rx offload capabilities from the first slave device */
290 	internals->rx_offload_capa = di->rx_offload_capa;
291 	internals->rx_queue_offload_capa = di->rx_queue_offload_capa;
292 	internals->flow_type_rss_offloads = di->flow_type_rss_offloads;
293 
294 	/* Inherit maximum Rx packet size from the first slave device */
295 	internals->candidate_max_rx_pktlen = di->max_rx_pktlen;
296 
297 	/* Inherit default Rx queue settings from the first slave device */
298 	memcpy(rxconf_i, &di->default_rxconf, sizeof(*rxconf_i));
299 
300 	/*
301 	 * Turn off descriptor prefetch and writeback by default for all
302 	 * slave devices. Applications may tweak this setting if need be.
303 	 */
304 	rxconf_i->rx_thresh.pthresh = 0;
305 	rxconf_i->rx_thresh.hthresh = 0;
306 	rxconf_i->rx_thresh.wthresh = 0;
307 
308 	/* Setting this to zero should effectively enable default values */
309 	rxconf_i->rx_free_thresh = 0;
310 
311 	/* Disable deferred start by default for all slave devices */
312 	rxconf_i->rx_deferred_start = 0;
313 }
314 
315 static void
316 eth_bond_slave_inherit_dev_info_tx_first(struct bond_dev_private *internals,
317 					 const struct rte_eth_dev_info *di)
318 {
319 	struct rte_eth_txconf *txconf_i = &internals->default_txconf;
320 
321 	/* Inherit Tx offload capabilities from the first slave device */
322 	internals->tx_offload_capa = di->tx_offload_capa;
323 	internals->tx_queue_offload_capa = di->tx_queue_offload_capa;
324 
325 	/* Inherit default Tx queue settings from the first slave device */
326 	memcpy(txconf_i, &di->default_txconf, sizeof(*txconf_i));
327 
328 	/*
329 	 * Turn off descriptor prefetch and writeback by default for all
330 	 * slave devices. Applications may tweak this setting if need be.
331 	 */
332 	txconf_i->tx_thresh.pthresh = 0;
333 	txconf_i->tx_thresh.hthresh = 0;
334 	txconf_i->tx_thresh.wthresh = 0;
335 
336 	/*
337 	 * Setting these parameters to zero assumes that default
338 	 * values will be configured implicitly by slave devices.
339 	 */
340 	txconf_i->tx_free_thresh = 0;
341 	txconf_i->tx_rs_thresh = 0;
342 
343 	/* Disable deferred start by default for all slave devices */
344 	txconf_i->tx_deferred_start = 0;
345 }
346 
347 static void
348 eth_bond_slave_inherit_dev_info_rx_next(struct bond_dev_private *internals,
349 					const struct rte_eth_dev_info *di)
350 {
351 	struct rte_eth_rxconf *rxconf_i = &internals->default_rxconf;
352 	const struct rte_eth_rxconf *rxconf = &di->default_rxconf;
353 
354 	internals->rx_offload_capa &= di->rx_offload_capa;
355 	internals->rx_queue_offload_capa &= di->rx_queue_offload_capa;
356 	internals->flow_type_rss_offloads &= di->flow_type_rss_offloads;
357 
358 	/*
359 	 * If at least one slave device suggests enabling this
360 	 * setting by default, enable it for all slave devices
361 	 * since disabling it may not be necessarily supported.
362 	 */
363 	if (rxconf->rx_drop_en == 1)
364 		rxconf_i->rx_drop_en = 1;
365 
366 	/*
367 	 * Adding a new slave device may cause some of previously inherited
368 	 * offloads to be withdrawn from the internal rx_queue_offload_capa
369 	 * value. Thus, the new internal value of default Rx queue offloads
370 	 * has to be masked by rx_queue_offload_capa to make sure that only
371 	 * commonly supported offloads are preserved from both the previous
372 	 * value and the value being inhereted from the new slave device.
373 	 */
374 	rxconf_i->offloads = (rxconf_i->offloads | rxconf->offloads) &
375 			     internals->rx_queue_offload_capa;
376 
377 	/*
378 	 * RETA size is GCD of all slaves RETA sizes, so, if all sizes will be
379 	 * the power of 2, the lower one is GCD
380 	 */
381 	if (internals->reta_size > di->reta_size)
382 		internals->reta_size = di->reta_size;
383 
384 	if (!internals->max_rx_pktlen &&
385 	    di->max_rx_pktlen < internals->candidate_max_rx_pktlen)
386 		internals->candidate_max_rx_pktlen = di->max_rx_pktlen;
387 }
388 
389 static void
390 eth_bond_slave_inherit_dev_info_tx_next(struct bond_dev_private *internals,
391 					const struct rte_eth_dev_info *di)
392 {
393 	struct rte_eth_txconf *txconf_i = &internals->default_txconf;
394 	const struct rte_eth_txconf *txconf = &di->default_txconf;
395 
396 	internals->tx_offload_capa &= di->tx_offload_capa;
397 	internals->tx_queue_offload_capa &= di->tx_queue_offload_capa;
398 
399 	/*
400 	 * Adding a new slave device may cause some of previously inherited
401 	 * offloads to be withdrawn from the internal tx_queue_offload_capa
402 	 * value. Thus, the new internal value of default Tx queue offloads
403 	 * has to be masked by tx_queue_offload_capa to make sure that only
404 	 * commonly supported offloads are preserved from both the previous
405 	 * value and the value being inhereted from the new slave device.
406 	 */
407 	txconf_i->offloads = (txconf_i->offloads | txconf->offloads) &
408 			     internals->tx_queue_offload_capa;
409 }
410 
411 static void
412 eth_bond_slave_inherit_desc_lim_first(struct rte_eth_desc_lim *bond_desc_lim,
413 		const struct rte_eth_desc_lim *slave_desc_lim)
414 {
415 	memcpy(bond_desc_lim, slave_desc_lim, sizeof(*bond_desc_lim));
416 }
417 
418 static int
419 eth_bond_slave_inherit_desc_lim_next(struct rte_eth_desc_lim *bond_desc_lim,
420 		const struct rte_eth_desc_lim *slave_desc_lim)
421 {
422 	bond_desc_lim->nb_max = RTE_MIN(bond_desc_lim->nb_max,
423 					slave_desc_lim->nb_max);
424 	bond_desc_lim->nb_min = RTE_MAX(bond_desc_lim->nb_min,
425 					slave_desc_lim->nb_min);
426 	bond_desc_lim->nb_align = RTE_MAX(bond_desc_lim->nb_align,
427 					  slave_desc_lim->nb_align);
428 
429 	if (bond_desc_lim->nb_min > bond_desc_lim->nb_max ||
430 	    bond_desc_lim->nb_align > bond_desc_lim->nb_max) {
431 		RTE_BOND_LOG(ERR, "Failed to inherit descriptor limits");
432 		return -EINVAL;
433 	}
434 
435 	/* Treat maximum number of segments equal to 0 as unspecified */
436 	if (slave_desc_lim->nb_seg_max != 0 &&
437 	    (bond_desc_lim->nb_seg_max == 0 ||
438 	     slave_desc_lim->nb_seg_max < bond_desc_lim->nb_seg_max))
439 		bond_desc_lim->nb_seg_max = slave_desc_lim->nb_seg_max;
440 	if (slave_desc_lim->nb_mtu_seg_max != 0 &&
441 	    (bond_desc_lim->nb_mtu_seg_max == 0 ||
442 	     slave_desc_lim->nb_mtu_seg_max < bond_desc_lim->nb_mtu_seg_max))
443 		bond_desc_lim->nb_mtu_seg_max = slave_desc_lim->nb_mtu_seg_max;
444 
445 	return 0;
446 }
447 
448 static int
449 __eth_bond_slave_add_lock_free(uint16_t bonded_port_id, uint16_t slave_port_id)
450 {
451 	struct rte_eth_dev *bonded_eth_dev, *slave_eth_dev;
452 	struct bond_dev_private *internals;
453 	struct rte_eth_link link_props;
454 	struct rte_eth_dev_info dev_info;
455 	int ret;
456 
457 	bonded_eth_dev = &rte_eth_devices[bonded_port_id];
458 	internals = bonded_eth_dev->data->dev_private;
459 
460 	if (valid_slave_port_id(slave_port_id, internals->mode) != 0)
461 		return -1;
462 
463 	slave_eth_dev = &rte_eth_devices[slave_port_id];
464 	if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_BONDED_SLAVE) {
465 		RTE_BOND_LOG(ERR, "Slave device is already a slave of a bonded device");
466 		return -1;
467 	}
468 
469 	ret = rte_eth_dev_info_get(slave_port_id, &dev_info);
470 	if (ret != 0) {
471 		RTE_BOND_LOG(ERR,
472 			"%s: Error during getting device (port %u) info: %s\n",
473 			__func__, slave_port_id, strerror(-ret));
474 
475 		return ret;
476 	}
477 	if (dev_info.max_rx_pktlen < internals->max_rx_pktlen) {
478 		RTE_BOND_LOG(ERR, "Slave (port %u) max_rx_pktlen too small",
479 			     slave_port_id);
480 		return -1;
481 	}
482 
483 	slave_add(internals, slave_eth_dev);
484 
485 	/* We need to store slaves reta_size to be able to synchronize RETA for all
486 	 * slave devices even if its sizes are different.
487 	 */
488 	internals->slaves[internals->slave_count].reta_size = dev_info.reta_size;
489 
490 	if (internals->slave_count < 1) {
491 		/* if MAC is not user defined then use MAC of first slave add to
492 		 * bonded device */
493 		if (!internals->user_defined_mac) {
494 			if (mac_address_set(bonded_eth_dev,
495 					    slave_eth_dev->data->mac_addrs)) {
496 				RTE_BOND_LOG(ERR, "Failed to set MAC address");
497 				return -1;
498 			}
499 		}
500 
501 		/* Make primary slave */
502 		internals->primary_port = slave_port_id;
503 		internals->current_primary_port = slave_port_id;
504 
505 		/* Inherit queues settings from first slave */
506 		internals->nb_rx_queues = slave_eth_dev->data->nb_rx_queues;
507 		internals->nb_tx_queues = slave_eth_dev->data->nb_tx_queues;
508 
509 		eth_bond_slave_inherit_dev_info_rx_first(internals, &dev_info);
510 		eth_bond_slave_inherit_dev_info_tx_first(internals, &dev_info);
511 
512 		eth_bond_slave_inherit_desc_lim_first(&internals->rx_desc_lim,
513 						      &dev_info.rx_desc_lim);
514 		eth_bond_slave_inherit_desc_lim_first(&internals->tx_desc_lim,
515 						      &dev_info.tx_desc_lim);
516 	} else {
517 		int ret;
518 
519 		eth_bond_slave_inherit_dev_info_rx_next(internals, &dev_info);
520 		eth_bond_slave_inherit_dev_info_tx_next(internals, &dev_info);
521 
522 		ret = eth_bond_slave_inherit_desc_lim_next(
523 				&internals->rx_desc_lim, &dev_info.rx_desc_lim);
524 		if (ret != 0)
525 			return ret;
526 
527 		ret = eth_bond_slave_inherit_desc_lim_next(
528 				&internals->tx_desc_lim, &dev_info.tx_desc_lim);
529 		if (ret != 0)
530 			return ret;
531 	}
532 
533 	bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf &=
534 			internals->flow_type_rss_offloads;
535 
536 	if (slave_rte_flow_prepare(internals->slave_count, internals) != 0) {
537 		RTE_BOND_LOG(ERR, "Failed to prepare new slave flows: port=%d",
538 			     slave_port_id);
539 		return -1;
540 	}
541 
542 	/* Add additional MAC addresses to the slave */
543 	if (slave_add_mac_addresses(bonded_eth_dev, slave_port_id) != 0) {
544 		RTE_BOND_LOG(ERR, "Failed to add mac address(es) to slave %hu",
545 				slave_port_id);
546 		return -1;
547 	}
548 
549 	internals->slave_count++;
550 
551 	if (bonded_eth_dev->data->dev_started) {
552 		if (slave_configure(bonded_eth_dev, slave_eth_dev) != 0) {
553 			internals->slave_count--;
554 			RTE_BOND_LOG(ERR, "rte_bond_slaves_configure: port=%d",
555 					slave_port_id);
556 			return -1;
557 		}
558 	}
559 
560 	/* Update all slave devices MACs */
561 	mac_address_slaves_update(bonded_eth_dev);
562 
563 	/* Register link status change callback with bonded device pointer as
564 	 * argument*/
565 	rte_eth_dev_callback_register(slave_port_id, RTE_ETH_EVENT_INTR_LSC,
566 			bond_ethdev_lsc_event_callback, &bonded_eth_dev->data->port_id);
567 
568 	/* If bonded device is started then we can add the slave to our active
569 	 * slave array */
570 	if (bonded_eth_dev->data->dev_started) {
571 		ret = rte_eth_link_get_nowait(slave_port_id, &link_props);
572 		if (ret < 0) {
573 			rte_eth_dev_callback_unregister(slave_port_id,
574 					RTE_ETH_EVENT_INTR_LSC,
575 					bond_ethdev_lsc_event_callback,
576 					&bonded_eth_dev->data->port_id);
577 			internals->slave_count--;
578 			RTE_BOND_LOG(ERR,
579 				"Slave (port %u) link get failed: %s\n",
580 				slave_port_id, rte_strerror(-ret));
581 			return -1;
582 		}
583 
584 		 if (link_props.link_status == ETH_LINK_UP) {
585 			if (internals->active_slave_count == 0 &&
586 			    !internals->user_defined_primary_port)
587 				bond_ethdev_primary_set(internals,
588 							slave_port_id);
589 		}
590 	}
591 
592 	/* Add slave details to bonded device */
593 	slave_eth_dev->data->dev_flags |= RTE_ETH_DEV_BONDED_SLAVE;
594 
595 	slave_vlan_filter_set(bonded_port_id, slave_port_id);
596 
597 	return 0;
598 
599 }
600 
601 int
602 rte_eth_bond_slave_add(uint16_t bonded_port_id, uint16_t slave_port_id)
603 {
604 	struct rte_eth_dev *bonded_eth_dev;
605 	struct bond_dev_private *internals;
606 
607 	int retval;
608 
609 	/* Verify that port id's are valid bonded and slave ports */
610 	if (valid_bonded_port_id(bonded_port_id) != 0)
611 		return -1;
612 
613 	bonded_eth_dev = &rte_eth_devices[bonded_port_id];
614 	internals = bonded_eth_dev->data->dev_private;
615 
616 	rte_spinlock_lock(&internals->lock);
617 
618 	retval = __eth_bond_slave_add_lock_free(bonded_port_id, slave_port_id);
619 
620 	rte_spinlock_unlock(&internals->lock);
621 
622 	return retval;
623 }
624 
625 static int
626 __eth_bond_slave_remove_lock_free(uint16_t bonded_port_id,
627 				   uint16_t slave_port_id)
628 {
629 	struct rte_eth_dev *bonded_eth_dev;
630 	struct bond_dev_private *internals;
631 	struct rte_eth_dev *slave_eth_dev;
632 	struct rte_flow_error flow_error;
633 	struct rte_flow *flow;
634 	int i, slave_idx;
635 
636 	bonded_eth_dev = &rte_eth_devices[bonded_port_id];
637 	internals = bonded_eth_dev->data->dev_private;
638 
639 	if (valid_slave_port_id(slave_port_id, internals->mode) < 0)
640 		return -1;
641 
642 	/* first remove from active slave list */
643 	slave_idx = find_slave_by_id(internals->active_slaves,
644 		internals->active_slave_count, slave_port_id);
645 
646 	if (slave_idx < internals->active_slave_count)
647 		deactivate_slave(bonded_eth_dev, slave_port_id);
648 
649 	slave_idx = -1;
650 	/* now find in slave list */
651 	for (i = 0; i < internals->slave_count; i++)
652 		if (internals->slaves[i].port_id == slave_port_id) {
653 			slave_idx = i;
654 			break;
655 		}
656 
657 	if (slave_idx < 0) {
658 		RTE_BOND_LOG(ERR, "Couldn't find slave in port list, slave count %d",
659 				internals->slave_count);
660 		return -1;
661 	}
662 
663 	/* Un-register link status change callback with bonded device pointer as
664 	 * argument*/
665 	rte_eth_dev_callback_unregister(slave_port_id, RTE_ETH_EVENT_INTR_LSC,
666 			bond_ethdev_lsc_event_callback,
667 			&rte_eth_devices[bonded_port_id].data->port_id);
668 
669 	/* Restore original MAC address of slave device */
670 	rte_eth_dev_default_mac_addr_set(slave_port_id,
671 			&(internals->slaves[slave_idx].persisted_mac_addr));
672 
673 	/* remove additional MAC addresses from the slave */
674 	slave_remove_mac_addresses(bonded_eth_dev, slave_port_id);
675 
676 	/*
677 	 * Remove bond device flows from slave device.
678 	 * Note: don't restore flow isolate mode.
679 	 */
680 	TAILQ_FOREACH(flow, &internals->flow_list, next) {
681 		if (flow->flows[slave_idx] != NULL) {
682 			rte_flow_destroy(slave_port_id, flow->flows[slave_idx],
683 					 &flow_error);
684 			flow->flows[slave_idx] = NULL;
685 		}
686 	}
687 
688 	slave_eth_dev = &rte_eth_devices[slave_port_id];
689 	slave_remove(internals, slave_eth_dev);
690 	slave_eth_dev->data->dev_flags &= (~RTE_ETH_DEV_BONDED_SLAVE);
691 
692 	/*  first slave in the active list will be the primary by default,
693 	 *  otherwise use first device in list */
694 	if (internals->current_primary_port == slave_port_id) {
695 		if (internals->active_slave_count > 0)
696 			internals->current_primary_port = internals->active_slaves[0];
697 		else if (internals->slave_count > 0)
698 			internals->current_primary_port = internals->slaves[0].port_id;
699 		else
700 			internals->primary_port = 0;
701 	}
702 
703 	if (internals->active_slave_count < 1) {
704 		/* if no slaves are any longer attached to bonded device and MAC is not
705 		 * user defined then clear MAC of bonded device as it will be reset
706 		 * when a new slave is added */
707 		if (internals->slave_count < 1 && !internals->user_defined_mac)
708 			memset(rte_eth_devices[bonded_port_id].data->mac_addrs, 0,
709 					sizeof(*(rte_eth_devices[bonded_port_id].data->mac_addrs)));
710 	}
711 	if (internals->slave_count == 0) {
712 		internals->rx_offload_capa = 0;
713 		internals->tx_offload_capa = 0;
714 		internals->rx_queue_offload_capa = 0;
715 		internals->tx_queue_offload_capa = 0;
716 		internals->flow_type_rss_offloads = ETH_RSS_PROTO_MASK;
717 		internals->reta_size = 0;
718 		internals->candidate_max_rx_pktlen = 0;
719 		internals->max_rx_pktlen = 0;
720 	}
721 	return 0;
722 }
723 
724 int
725 rte_eth_bond_slave_remove(uint16_t bonded_port_id, uint16_t slave_port_id)
726 {
727 	struct rte_eth_dev *bonded_eth_dev;
728 	struct bond_dev_private *internals;
729 	int retval;
730 
731 	if (valid_bonded_port_id(bonded_port_id) != 0)
732 		return -1;
733 
734 	bonded_eth_dev = &rte_eth_devices[bonded_port_id];
735 	internals = bonded_eth_dev->data->dev_private;
736 
737 	rte_spinlock_lock(&internals->lock);
738 
739 	retval = __eth_bond_slave_remove_lock_free(bonded_port_id, slave_port_id);
740 
741 	rte_spinlock_unlock(&internals->lock);
742 
743 	return retval;
744 }
745 
746 int
747 rte_eth_bond_mode_set(uint16_t bonded_port_id, uint8_t mode)
748 {
749 	struct rte_eth_dev *bonded_eth_dev;
750 
751 	if (valid_bonded_port_id(bonded_port_id) != 0)
752 		return -1;
753 
754 	bonded_eth_dev = &rte_eth_devices[bonded_port_id];
755 
756 	if (check_for_master_bonded_ethdev(bonded_eth_dev) != 0 &&
757 			mode == BONDING_MODE_8023AD)
758 		return -1;
759 
760 	return bond_ethdev_mode_set(bonded_eth_dev, mode);
761 }
762 
763 int
764 rte_eth_bond_mode_get(uint16_t bonded_port_id)
765 {
766 	struct bond_dev_private *internals;
767 
768 	if (valid_bonded_port_id(bonded_port_id) != 0)
769 		return -1;
770 
771 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
772 
773 	return internals->mode;
774 }
775 
776 int
777 rte_eth_bond_primary_set(uint16_t bonded_port_id, uint16_t slave_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 (valid_slave_port_id(slave_port_id, internals->mode) != 0)
787 		return -1;
788 
789 	internals->user_defined_primary_port = 1;
790 	internals->primary_port = slave_port_id;
791 
792 	bond_ethdev_primary_set(internals, slave_port_id);
793 
794 	return 0;
795 }
796 
797 int
798 rte_eth_bond_primary_get(uint16_t bonded_port_id)
799 {
800 	struct bond_dev_private *internals;
801 
802 	if (valid_bonded_port_id(bonded_port_id) != 0)
803 		return -1;
804 
805 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
806 
807 	if (internals->slave_count < 1)
808 		return -1;
809 
810 	return internals->current_primary_port;
811 }
812 
813 int
814 rte_eth_bond_slaves_get(uint16_t bonded_port_id, uint16_t slaves[],
815 			uint16_t len)
816 {
817 	struct bond_dev_private *internals;
818 	uint16_t i;
819 
820 	if (valid_bonded_port_id(bonded_port_id) != 0)
821 		return -1;
822 
823 	if (slaves == NULL)
824 		return -1;
825 
826 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
827 
828 	if (internals->slave_count > len)
829 		return -1;
830 
831 	for (i = 0; i < internals->slave_count; i++)
832 		slaves[i] = internals->slaves[i].port_id;
833 
834 	return internals->slave_count;
835 }
836 
837 int
838 rte_eth_bond_active_slaves_get(uint16_t bonded_port_id, uint16_t slaves[],
839 		uint16_t len)
840 {
841 	struct bond_dev_private *internals;
842 
843 	if (valid_bonded_port_id(bonded_port_id) != 0)
844 		return -1;
845 
846 	if (slaves == NULL)
847 		return -1;
848 
849 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
850 
851 	if (internals->active_slave_count > len)
852 		return -1;
853 
854 	memcpy(slaves, internals->active_slaves,
855 	internals->active_slave_count * sizeof(internals->active_slaves[0]));
856 
857 	return internals->active_slave_count;
858 }
859 
860 int
861 rte_eth_bond_mac_address_set(uint16_t bonded_port_id,
862 		struct rte_ether_addr *mac_addr)
863 {
864 	struct rte_eth_dev *bonded_eth_dev;
865 	struct bond_dev_private *internals;
866 
867 	if (valid_bonded_port_id(bonded_port_id) != 0)
868 		return -1;
869 
870 	bonded_eth_dev = &rte_eth_devices[bonded_port_id];
871 	internals = bonded_eth_dev->data->dev_private;
872 
873 	/* Set MAC Address of Bonded Device */
874 	if (mac_address_set(bonded_eth_dev, mac_addr))
875 		return -1;
876 
877 	internals->user_defined_mac = 1;
878 
879 	/* Update all slave devices MACs*/
880 	if (internals->slave_count > 0)
881 		return mac_address_slaves_update(bonded_eth_dev);
882 
883 	return 0;
884 }
885 
886 int
887 rte_eth_bond_mac_address_reset(uint16_t bonded_port_id)
888 {
889 	struct rte_eth_dev *bonded_eth_dev;
890 	struct bond_dev_private *internals;
891 
892 	if (valid_bonded_port_id(bonded_port_id) != 0)
893 		return -1;
894 
895 	bonded_eth_dev = &rte_eth_devices[bonded_port_id];
896 	internals = bonded_eth_dev->data->dev_private;
897 
898 	internals->user_defined_mac = 0;
899 
900 	if (internals->slave_count > 0) {
901 		int slave_port;
902 		/* Get the primary slave location based on the primary port
903 		 * number as, while slave_add(), we will keep the primary
904 		 * slave based on slave_count,but not based on the primary port.
905 		 */
906 		for (slave_port = 0; slave_port < internals->slave_count;
907 		     slave_port++) {
908 			if (internals->slaves[slave_port].port_id ==
909 			    internals->primary_port)
910 				break;
911 		}
912 
913 		/* Set MAC Address of Bonded Device */
914 		if (mac_address_set(bonded_eth_dev,
915 			&internals->slaves[slave_port].persisted_mac_addr)
916 				!= 0) {
917 			RTE_BOND_LOG(ERR, "Failed to set MAC address on bonded device");
918 			return -1;
919 		}
920 		/* Update all slave devices MAC addresses */
921 		return mac_address_slaves_update(bonded_eth_dev);
922 	}
923 	/* No need to update anything as no slaves present */
924 	return 0;
925 }
926 
927 int
928 rte_eth_bond_xmit_policy_set(uint16_t bonded_port_id, uint8_t policy)
929 {
930 	struct bond_dev_private *internals;
931 
932 	if (valid_bonded_port_id(bonded_port_id) != 0)
933 		return -1;
934 
935 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
936 
937 	switch (policy) {
938 	case BALANCE_XMIT_POLICY_LAYER2:
939 		internals->balance_xmit_policy = policy;
940 		internals->burst_xmit_hash = burst_xmit_l2_hash;
941 		break;
942 	case BALANCE_XMIT_POLICY_LAYER23:
943 		internals->balance_xmit_policy = policy;
944 		internals->burst_xmit_hash = burst_xmit_l23_hash;
945 		break;
946 	case BALANCE_XMIT_POLICY_LAYER34:
947 		internals->balance_xmit_policy = policy;
948 		internals->burst_xmit_hash = burst_xmit_l34_hash;
949 		break;
950 
951 	default:
952 		return -1;
953 	}
954 	return 0;
955 }
956 
957 int
958 rte_eth_bond_xmit_policy_get(uint16_t bonded_port_id)
959 {
960 	struct bond_dev_private *internals;
961 
962 	if (valid_bonded_port_id(bonded_port_id) != 0)
963 		return -1;
964 
965 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
966 
967 	return internals->balance_xmit_policy;
968 }
969 
970 int
971 rte_eth_bond_link_monitoring_set(uint16_t bonded_port_id, uint32_t internal_ms)
972 {
973 	struct bond_dev_private *internals;
974 
975 	if (valid_bonded_port_id(bonded_port_id) != 0)
976 		return -1;
977 
978 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
979 	internals->link_status_polling_interval_ms = internal_ms;
980 
981 	return 0;
982 }
983 
984 int
985 rte_eth_bond_link_monitoring_get(uint16_t bonded_port_id)
986 {
987 	struct bond_dev_private *internals;
988 
989 	if (valid_bonded_port_id(bonded_port_id) != 0)
990 		return -1;
991 
992 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
993 
994 	return internals->link_status_polling_interval_ms;
995 }
996 
997 int
998 rte_eth_bond_link_down_prop_delay_set(uint16_t bonded_port_id,
999 				       uint32_t delay_ms)
1000 
1001 {
1002 	struct bond_dev_private *internals;
1003 
1004 	if (valid_bonded_port_id(bonded_port_id) != 0)
1005 		return -1;
1006 
1007 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
1008 	internals->link_down_delay_ms = delay_ms;
1009 
1010 	return 0;
1011 }
1012 
1013 int
1014 rte_eth_bond_link_down_prop_delay_get(uint16_t bonded_port_id)
1015 {
1016 	struct bond_dev_private *internals;
1017 
1018 	if (valid_bonded_port_id(bonded_port_id) != 0)
1019 		return -1;
1020 
1021 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
1022 
1023 	return internals->link_down_delay_ms;
1024 }
1025 
1026 int
1027 rte_eth_bond_link_up_prop_delay_set(uint16_t bonded_port_id, uint32_t delay_ms)
1028 
1029 {
1030 	struct bond_dev_private *internals;
1031 
1032 	if (valid_bonded_port_id(bonded_port_id) != 0)
1033 		return -1;
1034 
1035 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
1036 	internals->link_up_delay_ms = delay_ms;
1037 
1038 	return 0;
1039 }
1040 
1041 int
1042 rte_eth_bond_link_up_prop_delay_get(uint16_t bonded_port_id)
1043 {
1044 	struct bond_dev_private *internals;
1045 
1046 	if (valid_bonded_port_id(bonded_port_id) != 0)
1047 		return -1;
1048 
1049 	internals = rte_eth_devices[bonded_port_id].data->dev_private;
1050 
1051 	return internals->link_up_delay_ms;
1052 }
1053