xref: /dpdk/drivers/net/failsafe/failsafe_ops.c (revision f8dbaebbf1c9efcbb2e2354b341ed62175466a57)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2017 6WIND S.A.
3  * Copyright 2017 Mellanox Technologies, Ltd
4  */
5 
6 #include <stdbool.h>
7 #include <stdint.h>
8 #include <unistd.h>
9 
10 #include <rte_debug.h>
11 #include <rte_atomic.h>
12 #include <ethdev_driver.h>
13 #include <rte_malloc.h>
14 #include <rte_flow.h>
15 #include <rte_cycles.h>
16 #include <rte_ethdev.h>
17 #include <rte_string_fns.h>
18 
19 #include "failsafe_private.h"
20 
21 static int
22 fs_dev_configure(struct rte_eth_dev *dev)
23 {
24 	struct sub_device *sdev;
25 	uint8_t i;
26 	int ret;
27 
28 	fs_lock(dev, 0);
29 	FOREACH_SUBDEV(sdev, i, dev) {
30 		int rmv_interrupt = 0;
31 		int lsc_interrupt = 0;
32 		int lsc_enabled;
33 
34 		if (sdev->state != DEV_PROBED &&
35 		    !(PRIV(dev)->alarm_lock == 0 && sdev->state == DEV_ACTIVE))
36 			continue;
37 
38 		rmv_interrupt = ETH(sdev)->data->dev_flags &
39 				RTE_ETH_DEV_INTR_RMV;
40 		if (rmv_interrupt) {
41 			DEBUG("Enabling RMV interrupts for sub_device %d", i);
42 			dev->data->dev_conf.intr_conf.rmv = 1;
43 		} else {
44 			DEBUG("sub_device %d does not support RMV event", i);
45 		}
46 		lsc_enabled = dev->data->dev_conf.intr_conf.lsc;
47 		lsc_interrupt = lsc_enabled &&
48 				(ETH(sdev)->data->dev_flags &
49 				 RTE_ETH_DEV_INTR_LSC);
50 		if (lsc_interrupt) {
51 			DEBUG("Enabling LSC interrupts for sub_device %d", i);
52 			dev->data->dev_conf.intr_conf.lsc = 1;
53 		} else if (lsc_enabled && !lsc_interrupt) {
54 			DEBUG("Disabling LSC interrupts for sub_device %d", i);
55 			dev->data->dev_conf.intr_conf.lsc = 0;
56 		}
57 		DEBUG("Configuring sub-device %d", i);
58 		ret = rte_eth_dev_configure(PORT_ID(sdev),
59 					dev->data->nb_rx_queues,
60 					dev->data->nb_tx_queues,
61 					&dev->data->dev_conf);
62 		if (ret) {
63 			if (!fs_err(sdev, ret))
64 				continue;
65 			ERROR("Could not configure sub_device %d", i);
66 			fs_unlock(dev, 0);
67 			return ret;
68 		}
69 		if (rmv_interrupt && sdev->rmv_callback == 0) {
70 			ret = rte_eth_dev_callback_register(PORT_ID(sdev),
71 					RTE_ETH_EVENT_INTR_RMV,
72 					failsafe_eth_rmv_event_callback,
73 					sdev);
74 			if (ret)
75 				WARN("Failed to register RMV callback for sub_device %d",
76 				     SUB_ID(sdev));
77 			else
78 				sdev->rmv_callback = 1;
79 		}
80 		dev->data->dev_conf.intr_conf.rmv = 0;
81 		if (lsc_interrupt && sdev->lsc_callback == 0) {
82 			ret = rte_eth_dev_callback_register(PORT_ID(sdev),
83 						RTE_ETH_EVENT_INTR_LSC,
84 						failsafe_eth_lsc_event_callback,
85 						dev);
86 			if (ret)
87 				WARN("Failed to register LSC callback for sub_device %d",
88 				     SUB_ID(sdev));
89 			else
90 				sdev->lsc_callback = 1;
91 		}
92 		dev->data->dev_conf.intr_conf.lsc = lsc_enabled;
93 		sdev->state = DEV_ACTIVE;
94 	}
95 	if (PRIV(dev)->state < DEV_ACTIVE)
96 		PRIV(dev)->state = DEV_ACTIVE;
97 	fs_unlock(dev, 0);
98 	return 0;
99 }
100 
101 static void
102 fs_set_queues_state_start(struct rte_eth_dev *dev)
103 {
104 	struct rxq *rxq;
105 	struct txq *txq;
106 	uint16_t i;
107 
108 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
109 		rxq = dev->data->rx_queues[i];
110 		if (rxq != NULL && !rxq->info.conf.rx_deferred_start)
111 			dev->data->rx_queue_state[i] =
112 						RTE_ETH_QUEUE_STATE_STARTED;
113 	}
114 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
115 		txq = dev->data->tx_queues[i];
116 		if (txq != NULL && !txq->info.conf.tx_deferred_start)
117 			dev->data->tx_queue_state[i] =
118 						RTE_ETH_QUEUE_STATE_STARTED;
119 	}
120 }
121 
122 static int
123 fs_dev_start(struct rte_eth_dev *dev)
124 {
125 	struct sub_device *sdev;
126 	uint8_t i;
127 	int ret;
128 
129 	fs_lock(dev, 0);
130 	ret = failsafe_rx_intr_install(dev);
131 	if (ret) {
132 		fs_unlock(dev, 0);
133 		return ret;
134 	}
135 	FOREACH_SUBDEV(sdev, i, dev) {
136 		if (sdev->state != DEV_ACTIVE)
137 			continue;
138 		DEBUG("Starting sub_device %d", i);
139 		ret = rte_eth_dev_start(PORT_ID(sdev));
140 		if (ret) {
141 			if (!fs_err(sdev, ret))
142 				continue;
143 			fs_unlock(dev, 0);
144 			return ret;
145 		}
146 		ret = failsafe_rx_intr_install_subdevice(sdev);
147 		if (ret) {
148 			if (!fs_err(sdev, ret))
149 				continue;
150 			if (fs_err(sdev, rte_eth_dev_stop(PORT_ID(sdev))) < 0)
151 				ERROR("Failed to stop sub-device %u",
152 				      SUB_ID(sdev));
153 			fs_unlock(dev, 0);
154 			return ret;
155 		}
156 		sdev->state = DEV_STARTED;
157 	}
158 	if (PRIV(dev)->state < DEV_STARTED) {
159 		PRIV(dev)->state = DEV_STARTED;
160 		fs_set_queues_state_start(dev);
161 	}
162 	fs_switch_dev(dev, NULL);
163 	fs_unlock(dev, 0);
164 	return 0;
165 }
166 
167 static void
168 fs_set_queues_state_stop(struct rte_eth_dev *dev)
169 {
170 	uint16_t i;
171 
172 	for (i = 0; i < dev->data->nb_rx_queues; i++)
173 		if (dev->data->rx_queues[i] != NULL)
174 			dev->data->rx_queue_state[i] =
175 						RTE_ETH_QUEUE_STATE_STOPPED;
176 	for (i = 0; i < dev->data->nb_tx_queues; i++)
177 		if (dev->data->tx_queues[i] != NULL)
178 			dev->data->tx_queue_state[i] =
179 						RTE_ETH_QUEUE_STATE_STOPPED;
180 }
181 
182 static int
183 fs_dev_stop(struct rte_eth_dev *dev)
184 {
185 	struct sub_device *sdev;
186 	uint8_t i;
187 	int ret;
188 
189 	fs_lock(dev, 0);
190 	PRIV(dev)->state = DEV_STARTED - 1;
191 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_STARTED) {
192 		ret = rte_eth_dev_stop(PORT_ID(sdev));
193 		if (fs_err(sdev, ret) < 0) {
194 			ERROR("Failed to stop device %u",
195 			      PORT_ID(sdev));
196 			PRIV(dev)->state = DEV_STARTED + 1;
197 			fs_unlock(dev, 0);
198 			return ret;
199 		}
200 		failsafe_rx_intr_uninstall_subdevice(sdev);
201 		sdev->state = DEV_STARTED - 1;
202 	}
203 	failsafe_rx_intr_uninstall(dev);
204 	fs_set_queues_state_stop(dev);
205 	fs_unlock(dev, 0);
206 
207 	return 0;
208 }
209 
210 static int
211 fs_dev_set_link_up(struct rte_eth_dev *dev)
212 {
213 	struct sub_device *sdev;
214 	uint8_t i;
215 	int ret;
216 
217 	fs_lock(dev, 0);
218 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
219 		DEBUG("Calling rte_eth_dev_set_link_up on sub_device %d", i);
220 		ret = rte_eth_dev_set_link_up(PORT_ID(sdev));
221 		if ((ret = fs_err(sdev, ret))) {
222 			ERROR("Operation rte_eth_dev_set_link_up failed for sub_device %d"
223 			      " with error %d", i, ret);
224 			fs_unlock(dev, 0);
225 			return ret;
226 		}
227 	}
228 	fs_unlock(dev, 0);
229 	return 0;
230 }
231 
232 static int
233 fs_dev_set_link_down(struct rte_eth_dev *dev)
234 {
235 	struct sub_device *sdev;
236 	uint8_t i;
237 	int ret;
238 
239 	fs_lock(dev, 0);
240 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
241 		DEBUG("Calling rte_eth_dev_set_link_down on sub_device %d", i);
242 		ret = rte_eth_dev_set_link_down(PORT_ID(sdev));
243 		if ((ret = fs_err(sdev, ret))) {
244 			ERROR("Operation rte_eth_dev_set_link_down failed for sub_device %d"
245 			      " with error %d", i, ret);
246 			fs_unlock(dev, 0);
247 			return ret;
248 		}
249 	}
250 	fs_unlock(dev, 0);
251 	return 0;
252 }
253 
254 static int
255 fs_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
256 {
257 	struct sub_device *sdev;
258 	uint8_t i;
259 	int ret;
260 	int err = 0;
261 	bool failure = true;
262 
263 	fs_lock(dev, 0);
264 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
265 		uint16_t port_id = ETH(sdev)->data->port_id;
266 
267 		ret = rte_eth_dev_rx_queue_stop(port_id, rx_queue_id);
268 		ret = fs_err(sdev, ret);
269 		if (ret) {
270 			ERROR("Rx queue stop failed for subdevice %d", i);
271 			err = ret;
272 		} else {
273 			failure = false;
274 		}
275 	}
276 	dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
277 	fs_unlock(dev, 0);
278 	/* Return 0 in case of at least one successful queue stop */
279 	return (failure) ? err : 0;
280 }
281 
282 static int
283 fs_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
284 {
285 	struct sub_device *sdev;
286 	uint8_t i;
287 	int ret;
288 
289 	fs_lock(dev, 0);
290 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
291 		uint16_t port_id = ETH(sdev)->data->port_id;
292 
293 		ret = rte_eth_dev_rx_queue_start(port_id, rx_queue_id);
294 		ret = fs_err(sdev, ret);
295 		if (ret) {
296 			ERROR("Rx queue start failed for subdevice %d", i);
297 			fs_rx_queue_stop(dev, rx_queue_id);
298 			fs_unlock(dev, 0);
299 			return ret;
300 		}
301 	}
302 	dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
303 	fs_unlock(dev, 0);
304 	return 0;
305 }
306 
307 static int
308 fs_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
309 {
310 	struct sub_device *sdev;
311 	uint8_t i;
312 	int ret;
313 	int err = 0;
314 	bool failure = true;
315 
316 	fs_lock(dev, 0);
317 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
318 		uint16_t port_id = ETH(sdev)->data->port_id;
319 
320 		ret = rte_eth_dev_tx_queue_stop(port_id, tx_queue_id);
321 		ret = fs_err(sdev, ret);
322 		if (ret) {
323 			ERROR("Tx queue stop failed for subdevice %d", i);
324 			err = ret;
325 		} else {
326 			failure = false;
327 		}
328 	}
329 	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
330 	fs_unlock(dev, 0);
331 	/* Return 0 in case of at least one successful queue stop */
332 	return (failure) ? err : 0;
333 }
334 
335 static int
336 fs_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
337 {
338 	struct sub_device *sdev;
339 	uint8_t i;
340 	int ret;
341 
342 	fs_lock(dev, 0);
343 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
344 		uint16_t port_id = ETH(sdev)->data->port_id;
345 
346 		ret = rte_eth_dev_tx_queue_start(port_id, tx_queue_id);
347 		ret = fs_err(sdev, ret);
348 		if (ret) {
349 			ERROR("Tx queue start failed for subdevice %d", i);
350 			fs_tx_queue_stop(dev, tx_queue_id);
351 			fs_unlock(dev, 0);
352 			return ret;
353 		}
354 	}
355 	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
356 	fs_unlock(dev, 0);
357 	return 0;
358 }
359 
360 static void
361 fs_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
362 {
363 	struct sub_device *sdev;
364 	uint8_t i;
365 	struct rxq *rxq = dev->data->rx_queues[qid];
366 
367 	if (rxq == NULL)
368 		return;
369 	fs_lock(dev, 0);
370 	if (rxq->event_fd >= 0)
371 		close(rxq->event_fd);
372 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
373 		if (ETH(sdev)->data->rx_queues != NULL &&
374 		    ETH(sdev)->data->rx_queues[rxq->qid] != NULL)
375 			SUBOPS(sdev, rx_queue_release)(ETH(sdev), rxq->qid);
376 	}
377 	dev->data->rx_queues[rxq->qid] = NULL;
378 	rte_free(rxq);
379 	fs_unlock(dev, 0);
380 }
381 
382 static int
383 fs_rx_queue_setup(struct rte_eth_dev *dev,
384 		uint16_t rx_queue_id,
385 		uint16_t nb_rx_desc,
386 		unsigned int socket_id,
387 		const struct rte_eth_rxconf *rx_conf,
388 		struct rte_mempool *mb_pool)
389 {
390 	/*
391 	 * FIXME: Add a proper interface in rte_eal_interrupts for
392 	 * allocating eventfd as an interrupt vector.
393 	 * For the time being, fake as if we are using MSIX interrupts,
394 	 * this will cause rte_intr_efd_enable to allocate an eventfd for us.
395 	 */
396 	struct rte_intr_handle *intr_handle;
397 	struct sub_device *sdev;
398 	struct rxq *rxq;
399 	uint8_t i;
400 	int ret;
401 
402 	intr_handle = rte_intr_instance_alloc(RTE_INTR_INSTANCE_F_PRIVATE);
403 	if (intr_handle == NULL)
404 		return -ENOMEM;
405 
406 	if (rte_intr_type_set(intr_handle, RTE_INTR_HANDLE_VFIO_MSIX))
407 		return -rte_errno;
408 
409 	if (rte_intr_efds_index_set(intr_handle, 0, -1))
410 		return -rte_errno;
411 
412 	fs_lock(dev, 0);
413 	if (rx_conf->rx_deferred_start) {
414 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
415 			if (SUBOPS(sdev, rx_queue_start) == NULL) {
416 				ERROR("Rx queue deferred start is not "
417 					"supported for subdevice %d", i);
418 				fs_unlock(dev, 0);
419 				return -EINVAL;
420 			}
421 		}
422 	}
423 	rxq = dev->data->rx_queues[rx_queue_id];
424 	if (rxq != NULL) {
425 		fs_rx_queue_release(dev, rx_queue_id);
426 		dev->data->rx_queues[rx_queue_id] = NULL;
427 	}
428 	rxq = rte_zmalloc(NULL,
429 			  sizeof(*rxq) +
430 			  sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
431 			  RTE_CACHE_LINE_SIZE);
432 	if (rxq == NULL) {
433 		fs_unlock(dev, 0);
434 		return -ENOMEM;
435 	}
436 	FOREACH_SUBDEV(sdev, i, dev)
437 		rte_atomic64_init(&rxq->refcnt[i]);
438 	rxq->qid = rx_queue_id;
439 	rxq->socket_id = socket_id;
440 	rxq->info.mp = mb_pool;
441 	rxq->info.conf = *rx_conf;
442 	rxq->info.nb_desc = nb_rx_desc;
443 	rxq->priv = PRIV(dev);
444 	rxq->sdev = PRIV(dev)->subs;
445 	ret = rte_intr_efd_enable(intr_handle, 1);
446 	if (ret < 0) {
447 		fs_unlock(dev, 0);
448 		return ret;
449 	}
450 	rxq->event_fd = rte_intr_efds_index_get(intr_handle, 0);
451 	dev->data->rx_queues[rx_queue_id] = rxq;
452 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
453 		ret = rte_eth_rx_queue_setup(PORT_ID(sdev),
454 				rx_queue_id,
455 				nb_rx_desc, socket_id,
456 				rx_conf, mb_pool);
457 		if ((ret = fs_err(sdev, ret))) {
458 			ERROR("RX queue setup failed for sub_device %d", i);
459 			goto free_rxq;
460 		}
461 	}
462 	fs_unlock(dev, 0);
463 	return 0;
464 free_rxq:
465 	fs_rx_queue_release(dev, rx_queue_id);
466 	fs_unlock(dev, 0);
467 	return ret;
468 }
469 
470 static int
471 fs_rx_intr_enable(struct rte_eth_dev *dev, uint16_t idx)
472 {
473 	struct rxq *rxq;
474 	struct sub_device *sdev;
475 	uint8_t i;
476 	int ret;
477 	int rc = 0;
478 
479 	fs_lock(dev, 0);
480 	if (idx >= dev->data->nb_rx_queues) {
481 		rc = -EINVAL;
482 		goto unlock;
483 	}
484 	rxq = dev->data->rx_queues[idx];
485 	if (rxq == NULL || rxq->event_fd <= 0) {
486 		rc = -EINVAL;
487 		goto unlock;
488 	}
489 	/* Fail if proxy service is nor running. */
490 	if (PRIV(dev)->rxp.sstate != SS_RUNNING) {
491 		ERROR("failsafe interrupt services are not running");
492 		rc = -EAGAIN;
493 		goto unlock;
494 	}
495 	rxq->enable_events = 1;
496 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
497 		ret = rte_eth_dev_rx_intr_enable(PORT_ID(sdev), idx);
498 		ret = fs_err(sdev, ret);
499 		if (ret)
500 			rc = ret;
501 	}
502 unlock:
503 	fs_unlock(dev, 0);
504 	if (rc)
505 		rte_errno = -rc;
506 	return rc;
507 }
508 
509 static int
510 fs_rx_intr_disable(struct rte_eth_dev *dev, uint16_t idx)
511 {
512 	struct rxq *rxq;
513 	struct sub_device *sdev;
514 	uint64_t u64;
515 	uint8_t i;
516 	int rc = 0;
517 	int ret;
518 
519 	fs_lock(dev, 0);
520 	if (idx >= dev->data->nb_rx_queues) {
521 		rc = -EINVAL;
522 		goto unlock;
523 	}
524 	rxq = dev->data->rx_queues[idx];
525 	if (rxq == NULL || rxq->event_fd <= 0) {
526 		rc = -EINVAL;
527 		goto unlock;
528 	}
529 	rxq->enable_events = 0;
530 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
531 		ret = rte_eth_dev_rx_intr_disable(PORT_ID(sdev), idx);
532 		ret = fs_err(sdev, ret);
533 		if (ret)
534 			rc = ret;
535 	}
536 	/* Clear pending events */
537 	while (read(rxq->event_fd, &u64, sizeof(uint64_t)) >  0)
538 		;
539 unlock:
540 	fs_unlock(dev, 0);
541 	if (rc)
542 		rte_errno = -rc;
543 	return rc;
544 }
545 
546 static void
547 fs_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
548 {
549 	struct sub_device *sdev;
550 	uint8_t i;
551 	struct txq *txq = dev->data->tx_queues[qid];
552 
553 	if (txq == NULL)
554 		return;
555 	fs_lock(dev, 0);
556 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
557 		if (ETH(sdev)->data->tx_queues != NULL &&
558 		    ETH(sdev)->data->tx_queues[txq->qid] != NULL)
559 			SUBOPS(sdev, tx_queue_release)(ETH(sdev), txq->qid);
560 	}
561 	dev->data->tx_queues[txq->qid] = NULL;
562 	rte_free(txq);
563 	fs_unlock(dev, 0);
564 }
565 
566 static int
567 fs_tx_queue_setup(struct rte_eth_dev *dev,
568 		uint16_t tx_queue_id,
569 		uint16_t nb_tx_desc,
570 		unsigned int socket_id,
571 		const struct rte_eth_txconf *tx_conf)
572 {
573 	struct sub_device *sdev;
574 	struct txq *txq;
575 	uint8_t i;
576 	int ret;
577 
578 	fs_lock(dev, 0);
579 	if (tx_conf->tx_deferred_start) {
580 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
581 			if (SUBOPS(sdev, tx_queue_start) == NULL) {
582 				ERROR("Tx queue deferred start is not "
583 					"supported for subdevice %d", i);
584 				fs_unlock(dev, 0);
585 				return -EINVAL;
586 			}
587 		}
588 	}
589 	txq = dev->data->tx_queues[tx_queue_id];
590 	if (txq != NULL) {
591 		fs_tx_queue_release(dev, tx_queue_id);
592 		dev->data->tx_queues[tx_queue_id] = NULL;
593 	}
594 	txq = rte_zmalloc("ethdev TX queue",
595 			  sizeof(*txq) +
596 			  sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
597 			  RTE_CACHE_LINE_SIZE);
598 	if (txq == NULL) {
599 		fs_unlock(dev, 0);
600 		return -ENOMEM;
601 	}
602 	FOREACH_SUBDEV(sdev, i, dev)
603 		rte_atomic64_init(&txq->refcnt[i]);
604 	txq->qid = tx_queue_id;
605 	txq->socket_id = socket_id;
606 	txq->info.conf = *tx_conf;
607 	txq->info.nb_desc = nb_tx_desc;
608 	txq->priv = PRIV(dev);
609 	dev->data->tx_queues[tx_queue_id] = txq;
610 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
611 		ret = rte_eth_tx_queue_setup(PORT_ID(sdev),
612 				tx_queue_id,
613 				nb_tx_desc, socket_id,
614 				tx_conf);
615 		if ((ret = fs_err(sdev, ret))) {
616 			ERROR("TX queue setup failed for sub_device %d", i);
617 			goto free_txq;
618 		}
619 	}
620 	fs_unlock(dev, 0);
621 	return 0;
622 free_txq:
623 	fs_tx_queue_release(dev, tx_queue_id);
624 	fs_unlock(dev, 0);
625 	return ret;
626 }
627 
628 static void
629 fs_dev_free_queues(struct rte_eth_dev *dev)
630 {
631 	uint16_t i;
632 
633 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
634 		fs_rx_queue_release(dev, i);
635 		dev->data->rx_queues[i] = NULL;
636 	}
637 	dev->data->nb_rx_queues = 0;
638 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
639 		fs_tx_queue_release(dev, i);
640 		dev->data->tx_queues[i] = NULL;
641 	}
642 	dev->data->nb_tx_queues = 0;
643 }
644 
645 int
646 failsafe_eth_dev_close(struct rte_eth_dev *dev)
647 {
648 	struct sub_device *sdev;
649 	uint8_t i;
650 	int err, ret = 0;
651 
652 	fs_lock(dev, 0);
653 	failsafe_hotplug_alarm_cancel(dev);
654 	if (PRIV(dev)->state == DEV_STARTED) {
655 		ret = dev->dev_ops->dev_stop(dev);
656 		if (ret != 0) {
657 			fs_unlock(dev, 0);
658 			return ret;
659 		}
660 	}
661 	PRIV(dev)->state = DEV_ACTIVE - 1;
662 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
663 		DEBUG("Closing sub_device %d", i);
664 		failsafe_eth_dev_unregister_callbacks(sdev);
665 		err = rte_eth_dev_close(PORT_ID(sdev));
666 		if (err) {
667 			ret = ret ? ret : err;
668 			ERROR("Error while closing sub-device %u",
669 					PORT_ID(sdev));
670 		}
671 		sdev->state = DEV_ACTIVE - 1;
672 	}
673 	rte_eth_dev_callback_unregister(RTE_ETH_ALL, RTE_ETH_EVENT_NEW,
674 					failsafe_eth_new_event_callback, dev);
675 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
676 		fs_unlock(dev, 0);
677 		return ret;
678 	}
679 	fs_dev_free_queues(dev);
680 	err = failsafe_eal_uninit(dev);
681 	if (err) {
682 		ret = ret ? ret : err;
683 		ERROR("Error while uninitializing sub-EAL");
684 	}
685 	failsafe_args_free(dev);
686 	rte_free(PRIV(dev)->subs);
687 	rte_free(PRIV(dev)->mcast_addrs);
688 	/* mac_addrs must not be freed alone because part of dev_private */
689 	dev->data->mac_addrs = NULL;
690 	fs_unlock(dev, 0);
691 	err = pthread_mutex_destroy(&PRIV(dev)->hotplug_mutex);
692 	if (err) {
693 		ret = ret ? ret : err;
694 		ERROR("Error while destroying hotplug mutex");
695 	}
696 	return ret;
697 }
698 
699 static int
700 fs_promiscuous_enable(struct rte_eth_dev *dev)
701 {
702 	struct sub_device *sdev;
703 	uint8_t i;
704 	int ret = 0;
705 
706 	fs_lock(dev, 0);
707 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
708 		ret = rte_eth_promiscuous_enable(PORT_ID(sdev));
709 		ret = fs_err(sdev, ret);
710 		if (ret != 0) {
711 			ERROR("Promiscuous mode enable failed for subdevice %d",
712 				PORT_ID(sdev));
713 			break;
714 		}
715 	}
716 	if (ret != 0) {
717 		/* Rollback in the case of failure */
718 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
719 			ret = rte_eth_promiscuous_disable(PORT_ID(sdev));
720 			ret = fs_err(sdev, ret);
721 			if (ret != 0)
722 				ERROR("Promiscuous mode disable during rollback failed for subdevice %d",
723 					PORT_ID(sdev));
724 		}
725 	}
726 	fs_unlock(dev, 0);
727 
728 	return ret;
729 }
730 
731 static int
732 fs_promiscuous_disable(struct rte_eth_dev *dev)
733 {
734 	struct sub_device *sdev;
735 	uint8_t i;
736 	int ret = 0;
737 
738 	fs_lock(dev, 0);
739 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
740 		ret = rte_eth_promiscuous_disable(PORT_ID(sdev));
741 		ret = fs_err(sdev, ret);
742 		if (ret != 0) {
743 			ERROR("Promiscuous mode disable failed for subdevice %d",
744 				PORT_ID(sdev));
745 			break;
746 		}
747 	}
748 	if (ret != 0) {
749 		/* Rollback in the case of failure */
750 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
751 			ret = rte_eth_promiscuous_enable(PORT_ID(sdev));
752 			ret = fs_err(sdev, ret);
753 			if (ret != 0)
754 				ERROR("Promiscuous mode enable during rollback failed for subdevice %d",
755 					PORT_ID(sdev));
756 		}
757 	}
758 	fs_unlock(dev, 0);
759 
760 	return ret;
761 }
762 
763 static int
764 fs_allmulticast_enable(struct rte_eth_dev *dev)
765 {
766 	struct sub_device *sdev;
767 	uint8_t i;
768 	int ret = 0;
769 
770 	fs_lock(dev, 0);
771 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
772 		ret = rte_eth_allmulticast_enable(PORT_ID(sdev));
773 		ret = fs_err(sdev, ret);
774 		if (ret != 0) {
775 			ERROR("All-multicast mode enable failed for subdevice %d",
776 				PORT_ID(sdev));
777 			break;
778 		}
779 	}
780 	if (ret != 0) {
781 		/* Rollback in the case of failure */
782 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
783 			ret = rte_eth_allmulticast_disable(PORT_ID(sdev));
784 			ret = fs_err(sdev, ret);
785 			if (ret != 0)
786 				ERROR("All-multicast mode disable during rollback failed for subdevice %d",
787 					PORT_ID(sdev));
788 		}
789 	}
790 	fs_unlock(dev, 0);
791 
792 	return ret;
793 }
794 
795 static int
796 fs_allmulticast_disable(struct rte_eth_dev *dev)
797 {
798 	struct sub_device *sdev;
799 	uint8_t i;
800 	int ret = 0;
801 
802 	fs_lock(dev, 0);
803 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
804 		ret = rte_eth_allmulticast_disable(PORT_ID(sdev));
805 		ret = fs_err(sdev, ret);
806 		if (ret != 0) {
807 			ERROR("All-multicast mode disable failed for subdevice %d",
808 				PORT_ID(sdev));
809 			break;
810 		}
811 	}
812 	if (ret != 0) {
813 		/* Rollback in the case of failure */
814 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
815 			ret = rte_eth_allmulticast_enable(PORT_ID(sdev));
816 			ret = fs_err(sdev, ret);
817 			if (ret != 0)
818 				ERROR("All-multicast mode enable during rollback failed for subdevice %d",
819 					PORT_ID(sdev));
820 		}
821 	}
822 	fs_unlock(dev, 0);
823 
824 	return ret;
825 }
826 
827 static int
828 fs_link_update(struct rte_eth_dev *dev,
829 		int wait_to_complete)
830 {
831 	struct sub_device *sdev;
832 	uint8_t i;
833 	int ret;
834 
835 	fs_lock(dev, 0);
836 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
837 		DEBUG("Calling link_update on sub_device %d", i);
838 		ret = (SUBOPS(sdev, link_update))(ETH(sdev), wait_to_complete);
839 		if (ret && ret != -1 && sdev->remove == 0 &&
840 		    rte_eth_dev_is_removed(PORT_ID(sdev)) == 0) {
841 			ERROR("Link update failed for sub_device %d with error %d",
842 			      i, ret);
843 			fs_unlock(dev, 0);
844 			return ret;
845 		}
846 	}
847 	if (TX_SUBDEV(dev)) {
848 		struct rte_eth_link *l1;
849 		struct rte_eth_link *l2;
850 
851 		l1 = &dev->data->dev_link;
852 		l2 = &ETH(TX_SUBDEV(dev))->data->dev_link;
853 		if (memcmp(l1, l2, sizeof(*l1))) {
854 			*l1 = *l2;
855 			fs_unlock(dev, 0);
856 			return 0;
857 		}
858 	}
859 	fs_unlock(dev, 0);
860 	return -1;
861 }
862 
863 static int
864 fs_stats_get(struct rte_eth_dev *dev,
865 	     struct rte_eth_stats *stats)
866 {
867 	struct rte_eth_stats backup;
868 	struct sub_device *sdev;
869 	uint8_t i;
870 	int ret;
871 
872 	fs_lock(dev, 0);
873 	rte_memcpy(stats, &PRIV(dev)->stats_accumulator, sizeof(*stats));
874 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
875 		struct rte_eth_stats *snapshot = &sdev->stats_snapshot.stats;
876 		uint64_t *timestamp = &sdev->stats_snapshot.timestamp;
877 
878 		rte_memcpy(&backup, snapshot, sizeof(backup));
879 		ret = rte_eth_stats_get(PORT_ID(sdev), snapshot);
880 		if (ret) {
881 			if (!fs_err(sdev, ret)) {
882 				rte_memcpy(snapshot, &backup, sizeof(backup));
883 				goto inc;
884 			}
885 			ERROR("Operation rte_eth_stats_get failed for sub_device %d with error %d",
886 				  i, ret);
887 			*timestamp = 0;
888 			fs_unlock(dev, 0);
889 			return ret;
890 		}
891 		*timestamp = rte_rdtsc();
892 inc:
893 		failsafe_stats_increment(stats, snapshot);
894 	}
895 	fs_unlock(dev, 0);
896 	return 0;
897 }
898 
899 static int
900 fs_stats_reset(struct rte_eth_dev *dev)
901 {
902 	struct sub_device *sdev;
903 	uint8_t i;
904 	int ret;
905 
906 	fs_lock(dev, 0);
907 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
908 		ret = rte_eth_stats_reset(PORT_ID(sdev));
909 		if (ret) {
910 			if (!fs_err(sdev, ret))
911 				continue;
912 
913 			ERROR("Operation rte_eth_stats_reset failed for sub_device %d with error %d",
914 			      i, ret);
915 			fs_unlock(dev, 0);
916 			return ret;
917 		}
918 		memset(&sdev->stats_snapshot, 0, sizeof(struct rte_eth_stats));
919 	}
920 	memset(&PRIV(dev)->stats_accumulator, 0, sizeof(struct rte_eth_stats));
921 	fs_unlock(dev, 0);
922 
923 	return 0;
924 }
925 
926 static int
927 __fs_xstats_count(struct rte_eth_dev *dev)
928 {
929 	struct sub_device *sdev;
930 	int count = 0;
931 	uint8_t i;
932 	int ret;
933 
934 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
935 		ret = rte_eth_xstats_get_names(PORT_ID(sdev), NULL, 0);
936 		if (ret < 0)
937 			return ret;
938 		count += ret;
939 	}
940 
941 	return count;
942 }
943 
944 static int
945 __fs_xstats_get_names(struct rte_eth_dev *dev,
946 		    struct rte_eth_xstat_name *xstats_names,
947 		    unsigned int limit)
948 {
949 	struct sub_device *sdev;
950 	unsigned int count = 0;
951 	uint8_t i;
952 
953 	/* Caller only cares about count */
954 	if (!xstats_names)
955 		return  __fs_xstats_count(dev);
956 
957 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
958 		struct rte_eth_xstat_name *sub_names = xstats_names + count;
959 		int j, r;
960 
961 		if (count >= limit)
962 			break;
963 
964 		r = rte_eth_xstats_get_names(PORT_ID(sdev),
965 					     sub_names, limit - count);
966 		if (r < 0)
967 			return r;
968 
969 		/* add subN_ prefix to names */
970 		for (j = 0; j < r; j++) {
971 			char *xname = sub_names[j].name;
972 			char tmp[RTE_ETH_XSTATS_NAME_SIZE];
973 
974 			if ((xname[0] == 't' || xname[0] == 'r') &&
975 			    xname[1] == 'x' && xname[2] == '_')
976 				snprintf(tmp, sizeof(tmp), "%.3ssub%u_%s",
977 					 xname, i, xname + 3);
978 			else
979 				snprintf(tmp, sizeof(tmp), "sub%u_%s",
980 					 i, xname);
981 
982 			strlcpy(xname, tmp, RTE_ETH_XSTATS_NAME_SIZE);
983 		}
984 		count += r;
985 	}
986 	return count;
987 }
988 
989 static int
990 fs_xstats_get_names(struct rte_eth_dev *dev,
991 		    struct rte_eth_xstat_name *xstats_names,
992 		    unsigned int limit)
993 {
994 	int ret;
995 
996 	fs_lock(dev, 0);
997 	ret = __fs_xstats_get_names(dev, xstats_names, limit);
998 	fs_unlock(dev, 0);
999 	return ret;
1000 }
1001 
1002 static int
1003 __fs_xstats_get(struct rte_eth_dev *dev,
1004 	      struct rte_eth_xstat *xstats,
1005 	      unsigned int n)
1006 {
1007 	unsigned int count = 0;
1008 	struct sub_device *sdev;
1009 	uint8_t i;
1010 	int j, ret;
1011 
1012 	ret = __fs_xstats_count(dev);
1013 	/*
1014 	 * if error
1015 	 * or caller did not give enough space
1016 	 * or just querying
1017 	 */
1018 	if (ret < 0 || ret > (int)n || xstats == NULL)
1019 		return ret;
1020 
1021 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1022 		ret = rte_eth_xstats_get(PORT_ID(sdev), xstats, n);
1023 		if (ret < 0)
1024 			return ret;
1025 
1026 		if (ret > (int)n)
1027 			return n + count;
1028 
1029 		/* add offset to id's from sub-device */
1030 		for (j = 0; j < ret; j++)
1031 			xstats[j].id += count;
1032 
1033 		xstats += ret;
1034 		n -= ret;
1035 		count += ret;
1036 	}
1037 
1038 	return count;
1039 }
1040 
1041 static int
1042 fs_xstats_get(struct rte_eth_dev *dev,
1043 	      struct rte_eth_xstat *xstats,
1044 	      unsigned int n)
1045 {
1046 	int ret;
1047 
1048 	fs_lock(dev, 0);
1049 	ret = __fs_xstats_get(dev, xstats, n);
1050 	fs_unlock(dev, 0);
1051 
1052 	return ret;
1053 }
1054 
1055 
1056 static int
1057 fs_xstats_reset(struct rte_eth_dev *dev)
1058 {
1059 	struct sub_device *sdev;
1060 	uint8_t i;
1061 	int r = 0;
1062 
1063 	fs_lock(dev, 0);
1064 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1065 		r = rte_eth_xstats_reset(PORT_ID(sdev));
1066 		if (r < 0)
1067 			break;
1068 	}
1069 	fs_unlock(dev, 0);
1070 
1071 	return r;
1072 }
1073 
1074 static void
1075 fs_dev_merge_desc_lim(struct rte_eth_desc_lim *to,
1076 		      const struct rte_eth_desc_lim *from)
1077 {
1078 	to->nb_max = RTE_MIN(to->nb_max, from->nb_max);
1079 	to->nb_min = RTE_MAX(to->nb_min, from->nb_min);
1080 	to->nb_align = RTE_MAX(to->nb_align, from->nb_align);
1081 
1082 	to->nb_seg_max = RTE_MIN(to->nb_seg_max, from->nb_seg_max);
1083 	to->nb_mtu_seg_max = RTE_MIN(to->nb_mtu_seg_max, from->nb_mtu_seg_max);
1084 }
1085 
1086 /*
1087  * Merge the information from sub-devices.
1088  *
1089  * The reported values must be the common subset of all sub devices
1090  */
1091 static void
1092 fs_dev_merge_info(struct rte_eth_dev_info *info,
1093 		  const struct rte_eth_dev_info *sinfo)
1094 {
1095 	info->min_mtu = RTE_MAX(info->min_mtu, sinfo->min_mtu);
1096 	info->max_mtu = RTE_MIN(info->max_mtu, sinfo->max_mtu);
1097 	info->max_rx_pktlen = RTE_MIN(info->max_rx_pktlen, sinfo->max_rx_pktlen);
1098 	info->max_rx_queues = RTE_MIN(info->max_rx_queues, sinfo->max_rx_queues);
1099 	info->max_tx_queues = RTE_MIN(info->max_tx_queues, sinfo->max_tx_queues);
1100 	info->max_mac_addrs = RTE_MIN(info->max_mac_addrs, sinfo->max_mac_addrs);
1101 	info->max_hash_mac_addrs = RTE_MIN(info->max_hash_mac_addrs,
1102 					sinfo->max_hash_mac_addrs);
1103 	info->max_vmdq_pools = RTE_MIN(info->max_vmdq_pools, sinfo->max_vmdq_pools);
1104 	info->max_vfs = RTE_MIN(info->max_vfs, sinfo->max_vfs);
1105 
1106 	fs_dev_merge_desc_lim(&info->rx_desc_lim, &sinfo->rx_desc_lim);
1107 	fs_dev_merge_desc_lim(&info->tx_desc_lim, &sinfo->tx_desc_lim);
1108 
1109 	info->rx_offload_capa &= sinfo->rx_offload_capa;
1110 	info->tx_offload_capa &= sinfo->tx_offload_capa;
1111 	info->rx_queue_offload_capa &= sinfo->rx_queue_offload_capa;
1112 	info->tx_queue_offload_capa &= sinfo->tx_queue_offload_capa;
1113 	info->flow_type_rss_offloads &= sinfo->flow_type_rss_offloads;
1114 
1115 	/*
1116 	 * RETA size is a GCD of RETA sizes indicated by sub-devices.
1117 	 * Each of these sizes is a power of 2, so use the lower one.
1118 	 */
1119 	info->reta_size = RTE_MIN(info->reta_size, sinfo->reta_size);
1120 
1121 	info->hash_key_size = RTE_MIN(info->hash_key_size,
1122 				      sinfo->hash_key_size);
1123 }
1124 
1125 /**
1126  * Fail-safe dev_infos_get rules:
1127  *
1128  * No sub_device:
1129  *   Numerables:
1130  *      Use the maximum possible values for any field, so as not
1131  *      to impede any further configuration effort.
1132  *   Capabilities:
1133  *      Limits capabilities to those that are understood by the
1134  *      fail-safe PMD. This understanding stems from the fail-safe
1135  *      being capable of verifying that the related capability is
1136  *      expressed within the device configuration (struct rte_eth_conf).
1137  *
1138  * At least one probed sub_device:
1139  *   Numerables:
1140  *      Uses values from the active probed sub_device
1141  *      The rationale here is that if any sub_device is less capable
1142  *      (for example concerning the number of queues) than the active
1143  *      sub_device, then its subsequent configuration will fail.
1144  *      It is impossible to foresee this failure when the failing sub_device
1145  *      is supposed to be plugged-in later on, so the configuration process
1146  *      is the single point of failure and error reporting.
1147  *   Capabilities:
1148  *      Uses a logical AND of RX capabilities among
1149  *      all sub_devices and the default capabilities.
1150  *      Uses a logical AND of TX capabilities among
1151  *      the active probed sub_device and the default capabilities.
1152  *      Uses a logical AND of device capabilities among
1153  *      all sub_devices and the default capabilities.
1154  *
1155  */
1156 static int
1157 fs_dev_infos_get(struct rte_eth_dev *dev,
1158 		  struct rte_eth_dev_info *infos)
1159 {
1160 	struct sub_device *sdev;
1161 	uint8_t i;
1162 	int ret;
1163 
1164 	/* Use maximum upper bounds by default */
1165 	infos->min_mtu = RTE_ETHER_MIN_MTU;
1166 	infos->max_mtu = UINT16_MAX;
1167 	infos->max_rx_pktlen = UINT32_MAX;
1168 	infos->max_rx_queues = RTE_MAX_QUEUES_PER_PORT;
1169 	infos->max_tx_queues = RTE_MAX_QUEUES_PER_PORT;
1170 	infos->max_mac_addrs = FAILSAFE_MAX_ETHADDR;
1171 	infos->max_hash_mac_addrs = UINT32_MAX;
1172 	infos->max_vfs = UINT16_MAX;
1173 	infos->max_vmdq_pools = UINT16_MAX;
1174 	infos->reta_size = UINT16_MAX;
1175 	infos->hash_key_size = UINT8_MAX;
1176 
1177 	/*
1178 	 * Set of capabilities that can be verified upon
1179 	 * configuring a sub-device.
1180 	 */
1181 	infos->rx_offload_capa =
1182 		RTE_ETH_RX_OFFLOAD_VLAN_STRIP |
1183 		RTE_ETH_RX_OFFLOAD_IPV4_CKSUM |
1184 		RTE_ETH_RX_OFFLOAD_UDP_CKSUM |
1185 		RTE_ETH_RX_OFFLOAD_TCP_CKSUM |
1186 		RTE_ETH_RX_OFFLOAD_TCP_LRO |
1187 		RTE_ETH_RX_OFFLOAD_QINQ_STRIP |
1188 		RTE_ETH_RX_OFFLOAD_OUTER_IPV4_CKSUM |
1189 		RTE_ETH_RX_OFFLOAD_MACSEC_STRIP |
1190 		RTE_ETH_RX_OFFLOAD_HEADER_SPLIT |
1191 		RTE_ETH_RX_OFFLOAD_VLAN_FILTER |
1192 		RTE_ETH_RX_OFFLOAD_VLAN_EXTEND |
1193 		RTE_ETH_RX_OFFLOAD_SCATTER |
1194 		RTE_ETH_RX_OFFLOAD_TIMESTAMP |
1195 		RTE_ETH_RX_OFFLOAD_SECURITY |
1196 		RTE_ETH_RX_OFFLOAD_RSS_HASH;
1197 
1198 	infos->rx_queue_offload_capa =
1199 		RTE_ETH_RX_OFFLOAD_VLAN_STRIP |
1200 		RTE_ETH_RX_OFFLOAD_IPV4_CKSUM |
1201 		RTE_ETH_RX_OFFLOAD_UDP_CKSUM |
1202 		RTE_ETH_RX_OFFLOAD_TCP_CKSUM |
1203 		RTE_ETH_RX_OFFLOAD_TCP_LRO |
1204 		RTE_ETH_RX_OFFLOAD_QINQ_STRIP |
1205 		RTE_ETH_RX_OFFLOAD_OUTER_IPV4_CKSUM |
1206 		RTE_ETH_RX_OFFLOAD_MACSEC_STRIP |
1207 		RTE_ETH_RX_OFFLOAD_HEADER_SPLIT |
1208 		RTE_ETH_RX_OFFLOAD_VLAN_FILTER |
1209 		RTE_ETH_RX_OFFLOAD_VLAN_EXTEND |
1210 		RTE_ETH_RX_OFFLOAD_SCATTER |
1211 		RTE_ETH_RX_OFFLOAD_TIMESTAMP |
1212 		RTE_ETH_RX_OFFLOAD_SECURITY |
1213 		RTE_ETH_RX_OFFLOAD_RSS_HASH;
1214 
1215 	infos->tx_offload_capa =
1216 		RTE_ETH_TX_OFFLOAD_MULTI_SEGS |
1217 		RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE |
1218 		RTE_ETH_TX_OFFLOAD_IPV4_CKSUM |
1219 		RTE_ETH_TX_OFFLOAD_UDP_CKSUM |
1220 		RTE_ETH_TX_OFFLOAD_TCP_CKSUM |
1221 		RTE_ETH_TX_OFFLOAD_TCP_TSO;
1222 
1223 	infos->flow_type_rss_offloads =
1224 		RTE_ETH_RSS_IP |
1225 		RTE_ETH_RSS_UDP |
1226 		RTE_ETH_RSS_TCP;
1227 	infos->dev_capa =
1228 		RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
1229 		RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
1230 	infos->dev_capa &= ~RTE_ETH_DEV_CAPA_FLOW_RULE_KEEP;
1231 
1232 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
1233 		struct rte_eth_dev_info sub_info;
1234 
1235 		ret = rte_eth_dev_info_get(PORT_ID(sdev), &sub_info);
1236 		ret = fs_err(sdev, ret);
1237 		if (ret != 0)
1238 			return ret;
1239 
1240 		fs_dev_merge_info(infos, &sub_info);
1241 	}
1242 
1243 	return 0;
1244 }
1245 
1246 static const uint32_t *
1247 fs_dev_supported_ptypes_get(struct rte_eth_dev *dev)
1248 {
1249 	struct sub_device *sdev;
1250 	struct rte_eth_dev *edev;
1251 	const uint32_t *ret;
1252 
1253 	fs_lock(dev, 0);
1254 	sdev = TX_SUBDEV(dev);
1255 	if (sdev == NULL) {
1256 		ret = NULL;
1257 		goto unlock;
1258 	}
1259 	edev = ETH(sdev);
1260 	/* ENOTSUP: counts as no supported ptypes */
1261 	if (SUBOPS(sdev, dev_supported_ptypes_get) == NULL) {
1262 		ret = NULL;
1263 		goto unlock;
1264 	}
1265 	/*
1266 	 * The API does not permit to do a clean AND of all ptypes,
1267 	 * It is also incomplete by design and we do not really care
1268 	 * to have a best possible value in this context.
1269 	 * We just return the ptypes of the device of highest
1270 	 * priority, usually the PREFERRED device.
1271 	 */
1272 	ret = SUBOPS(sdev, dev_supported_ptypes_get)(edev);
1273 unlock:
1274 	fs_unlock(dev, 0);
1275 	return ret;
1276 }
1277 
1278 static int
1279 fs_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1280 {
1281 	struct sub_device *sdev;
1282 	uint8_t i;
1283 	int ret;
1284 
1285 	fs_lock(dev, 0);
1286 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1287 		DEBUG("Calling rte_eth_dev_set_mtu on sub_device %d", i);
1288 		ret = rte_eth_dev_set_mtu(PORT_ID(sdev), mtu);
1289 		if ((ret = fs_err(sdev, ret))) {
1290 			ERROR("Operation rte_eth_dev_set_mtu failed for sub_device %d with error %d",
1291 			      i, ret);
1292 			fs_unlock(dev, 0);
1293 			return ret;
1294 		}
1295 	}
1296 	fs_unlock(dev, 0);
1297 	return 0;
1298 }
1299 
1300 static int
1301 fs_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1302 {
1303 	struct sub_device *sdev;
1304 	uint8_t i;
1305 	int ret;
1306 
1307 	fs_lock(dev, 0);
1308 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1309 		DEBUG("Calling rte_eth_dev_vlan_filter on sub_device %d", i);
1310 		ret = rte_eth_dev_vlan_filter(PORT_ID(sdev), vlan_id, on);
1311 		if ((ret = fs_err(sdev, ret))) {
1312 			ERROR("Operation rte_eth_dev_vlan_filter failed for sub_device %d"
1313 			      " with error %d", i, ret);
1314 			fs_unlock(dev, 0);
1315 			return ret;
1316 		}
1317 	}
1318 	fs_unlock(dev, 0);
1319 	return 0;
1320 }
1321 
1322 static int
1323 fs_flow_ctrl_get(struct rte_eth_dev *dev,
1324 		struct rte_eth_fc_conf *fc_conf)
1325 {
1326 	struct sub_device *sdev;
1327 	int ret;
1328 
1329 	fs_lock(dev, 0);
1330 	sdev = TX_SUBDEV(dev);
1331 	if (sdev == NULL) {
1332 		ret = 0;
1333 		goto unlock;
1334 	}
1335 	if (SUBOPS(sdev, flow_ctrl_get) == NULL) {
1336 		ret = -ENOTSUP;
1337 		goto unlock;
1338 	}
1339 	ret = SUBOPS(sdev, flow_ctrl_get)(ETH(sdev), fc_conf);
1340 unlock:
1341 	fs_unlock(dev, 0);
1342 	return ret;
1343 }
1344 
1345 static int
1346 fs_flow_ctrl_set(struct rte_eth_dev *dev,
1347 		struct rte_eth_fc_conf *fc_conf)
1348 {
1349 	struct sub_device *sdev;
1350 	uint8_t i;
1351 	int ret;
1352 
1353 	fs_lock(dev, 0);
1354 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1355 		DEBUG("Calling rte_eth_dev_flow_ctrl_set on sub_device %d", i);
1356 		ret = rte_eth_dev_flow_ctrl_set(PORT_ID(sdev), fc_conf);
1357 		if ((ret = fs_err(sdev, ret))) {
1358 			ERROR("Operation rte_eth_dev_flow_ctrl_set failed for sub_device %d"
1359 			      " with error %d", i, ret);
1360 			fs_unlock(dev, 0);
1361 			return ret;
1362 		}
1363 	}
1364 	fs_unlock(dev, 0);
1365 	return 0;
1366 }
1367 
1368 static void
1369 fs_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1370 {
1371 	struct sub_device *sdev;
1372 	uint8_t i;
1373 
1374 	fs_lock(dev, 0);
1375 	/* No check: already done within the rte_eth_dev_mac_addr_remove
1376 	 * call for the fail-safe device.
1377 	 */
1378 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
1379 		rte_eth_dev_mac_addr_remove(PORT_ID(sdev),
1380 				&dev->data->mac_addrs[index]);
1381 	PRIV(dev)->mac_addr_pool[index] = 0;
1382 	fs_unlock(dev, 0);
1383 }
1384 
1385 static int
1386 fs_mac_addr_add(struct rte_eth_dev *dev,
1387 		struct rte_ether_addr *mac_addr,
1388 		uint32_t index,
1389 		uint32_t vmdq)
1390 {
1391 	struct sub_device *sdev;
1392 	int ret;
1393 	uint8_t i;
1394 
1395 	RTE_ASSERT(index < FAILSAFE_MAX_ETHADDR);
1396 	fs_lock(dev, 0);
1397 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1398 		ret = rte_eth_dev_mac_addr_add(PORT_ID(sdev), mac_addr, vmdq);
1399 		if ((ret = fs_err(sdev, ret))) {
1400 			ERROR("Operation rte_eth_dev_mac_addr_add failed for sub_device %"
1401 			      PRIu8 " with error %d", i, ret);
1402 			fs_unlock(dev, 0);
1403 			return ret;
1404 		}
1405 	}
1406 	if (index >= PRIV(dev)->nb_mac_addr) {
1407 		DEBUG("Growing mac_addrs array");
1408 		PRIV(dev)->nb_mac_addr = index;
1409 	}
1410 	PRIV(dev)->mac_addr_pool[index] = vmdq;
1411 	fs_unlock(dev, 0);
1412 	return 0;
1413 }
1414 
1415 static int
1416 fs_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1417 {
1418 	struct sub_device *sdev;
1419 	uint8_t i;
1420 	int ret;
1421 
1422 	fs_lock(dev, 0);
1423 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1424 		ret = rte_eth_dev_default_mac_addr_set(PORT_ID(sdev), mac_addr);
1425 		ret = fs_err(sdev, ret);
1426 		if (ret) {
1427 			ERROR("Operation rte_eth_dev_mac_addr_set failed for sub_device %d with error %d",
1428 				i, ret);
1429 			fs_unlock(dev, 0);
1430 			return ret;
1431 		}
1432 	}
1433 	fs_unlock(dev, 0);
1434 
1435 	return 0;
1436 }
1437 
1438 static int
1439 fs_set_mc_addr_list(struct rte_eth_dev *dev,
1440 		    struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1441 {
1442 	struct sub_device *sdev;
1443 	uint8_t i;
1444 	int ret;
1445 	void *mcast_addrs;
1446 
1447 	fs_lock(dev, 0);
1448 
1449 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1450 		ret = rte_eth_dev_set_mc_addr_list(PORT_ID(sdev),
1451 						   mc_addr_set, nb_mc_addr);
1452 		if (ret != 0) {
1453 			ERROR("Operation rte_eth_dev_set_mc_addr_list failed for sub_device %d with error %d",
1454 			      i, ret);
1455 			goto rollback;
1456 		}
1457 	}
1458 
1459 	mcast_addrs = rte_realloc(PRIV(dev)->mcast_addrs,
1460 		nb_mc_addr * sizeof(PRIV(dev)->mcast_addrs[0]), 0);
1461 	if (mcast_addrs == NULL && nb_mc_addr > 0) {
1462 		ret = -ENOMEM;
1463 		goto rollback;
1464 	}
1465 	rte_memcpy(mcast_addrs, mc_addr_set,
1466 		   nb_mc_addr * sizeof(PRIV(dev)->mcast_addrs[0]));
1467 	PRIV(dev)->nb_mcast_addr = nb_mc_addr;
1468 	PRIV(dev)->mcast_addrs = mcast_addrs;
1469 
1470 	fs_unlock(dev, 0);
1471 	return 0;
1472 
1473 rollback:
1474 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1475 		int rc = rte_eth_dev_set_mc_addr_list(PORT_ID(sdev),
1476 			PRIV(dev)->mcast_addrs,	PRIV(dev)->nb_mcast_addr);
1477 		if (rc != 0) {
1478 			ERROR("Multicast MAC address list rollback for sub_device %d failed with error %d",
1479 			      i, rc);
1480 		}
1481 	}
1482 
1483 	fs_unlock(dev, 0);
1484 	return ret;
1485 }
1486 
1487 static int
1488 fs_rss_hash_update(struct rte_eth_dev *dev,
1489 			struct rte_eth_rss_conf *rss_conf)
1490 {
1491 	struct sub_device *sdev;
1492 	uint8_t i;
1493 	int ret;
1494 
1495 	fs_lock(dev, 0);
1496 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1497 		ret = rte_eth_dev_rss_hash_update(PORT_ID(sdev), rss_conf);
1498 		ret = fs_err(sdev, ret);
1499 		if (ret) {
1500 			ERROR("Operation rte_eth_dev_rss_hash_update"
1501 				" failed for sub_device %d with error %d",
1502 				i, ret);
1503 			fs_unlock(dev, 0);
1504 			return ret;
1505 		}
1506 	}
1507 	fs_unlock(dev, 0);
1508 
1509 	return 0;
1510 }
1511 
1512 static int
1513 fs_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
1514 		const struct rte_flow_ops **ops)
1515 {
1516 	*ops = &fs_flow_ops;
1517 	return 0;
1518 }
1519 
1520 const struct eth_dev_ops failsafe_ops = {
1521 	.dev_configure = fs_dev_configure,
1522 	.dev_start = fs_dev_start,
1523 	.dev_stop = fs_dev_stop,
1524 	.dev_set_link_down = fs_dev_set_link_down,
1525 	.dev_set_link_up = fs_dev_set_link_up,
1526 	.dev_close = failsafe_eth_dev_close,
1527 	.promiscuous_enable = fs_promiscuous_enable,
1528 	.promiscuous_disable = fs_promiscuous_disable,
1529 	.allmulticast_enable = fs_allmulticast_enable,
1530 	.allmulticast_disable = fs_allmulticast_disable,
1531 	.link_update = fs_link_update,
1532 	.stats_get = fs_stats_get,
1533 	.stats_reset = fs_stats_reset,
1534 	.xstats_get = fs_xstats_get,
1535 	.xstats_get_names = fs_xstats_get_names,
1536 	.xstats_reset = fs_xstats_reset,
1537 	.dev_infos_get = fs_dev_infos_get,
1538 	.dev_supported_ptypes_get = fs_dev_supported_ptypes_get,
1539 	.mtu_set = fs_mtu_set,
1540 	.vlan_filter_set = fs_vlan_filter_set,
1541 	.rx_queue_start = fs_rx_queue_start,
1542 	.rx_queue_stop = fs_rx_queue_stop,
1543 	.tx_queue_start = fs_tx_queue_start,
1544 	.tx_queue_stop = fs_tx_queue_stop,
1545 	.rx_queue_setup = fs_rx_queue_setup,
1546 	.tx_queue_setup = fs_tx_queue_setup,
1547 	.rx_queue_release = fs_rx_queue_release,
1548 	.tx_queue_release = fs_tx_queue_release,
1549 	.rx_queue_intr_enable = fs_rx_intr_enable,
1550 	.rx_queue_intr_disable = fs_rx_intr_disable,
1551 	.flow_ctrl_get = fs_flow_ctrl_get,
1552 	.flow_ctrl_set = fs_flow_ctrl_set,
1553 	.mac_addr_remove = fs_mac_addr_remove,
1554 	.mac_addr_add = fs_mac_addr_add,
1555 	.mac_addr_set = fs_mac_addr_set,
1556 	.set_mc_addr_list = fs_set_mc_addr_list,
1557 	.rss_hash_update = fs_rss_hash_update,
1558 	.flow_ops_get = fs_flow_ops_get,
1559 };
1560