xref: /dpdk/drivers/net/failsafe/failsafe_ops.c (revision 7adf992fb9bf7162a7edc45b50d10fbb1d57824d)
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 <rte_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 			rte_eth_dev_stop(PORT_ID(sdev));
151 			fs_unlock(dev, 0);
152 			return ret;
153 		}
154 		sdev->state = DEV_STARTED;
155 	}
156 	if (PRIV(dev)->state < DEV_STARTED) {
157 		PRIV(dev)->state = DEV_STARTED;
158 		fs_set_queues_state_start(dev);
159 	}
160 	fs_switch_dev(dev, NULL);
161 	fs_unlock(dev, 0);
162 	return 0;
163 }
164 
165 static void
166 fs_set_queues_state_stop(struct rte_eth_dev *dev)
167 {
168 	uint16_t i;
169 
170 	for (i = 0; i < dev->data->nb_rx_queues; i++)
171 		if (dev->data->rx_queues[i] != NULL)
172 			dev->data->rx_queue_state[i] =
173 						RTE_ETH_QUEUE_STATE_STOPPED;
174 	for (i = 0; i < dev->data->nb_tx_queues; i++)
175 		if (dev->data->tx_queues[i] != NULL)
176 			dev->data->tx_queue_state[i] =
177 						RTE_ETH_QUEUE_STATE_STOPPED;
178 }
179 
180 static void
181 fs_dev_stop(struct rte_eth_dev *dev)
182 {
183 	struct sub_device *sdev;
184 	uint8_t i;
185 
186 	fs_lock(dev, 0);
187 	PRIV(dev)->state = DEV_STARTED - 1;
188 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_STARTED) {
189 		rte_eth_dev_stop(PORT_ID(sdev));
190 		failsafe_rx_intr_uninstall_subdevice(sdev);
191 		sdev->state = DEV_STARTED - 1;
192 	}
193 	failsafe_rx_intr_uninstall(dev);
194 	fs_set_queues_state_stop(dev);
195 	fs_unlock(dev, 0);
196 }
197 
198 static int
199 fs_dev_set_link_up(struct rte_eth_dev *dev)
200 {
201 	struct sub_device *sdev;
202 	uint8_t i;
203 	int ret;
204 
205 	fs_lock(dev, 0);
206 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
207 		DEBUG("Calling rte_eth_dev_set_link_up on sub_device %d", i);
208 		ret = rte_eth_dev_set_link_up(PORT_ID(sdev));
209 		if ((ret = fs_err(sdev, ret))) {
210 			ERROR("Operation rte_eth_dev_set_link_up failed for sub_device %d"
211 			      " with error %d", i, ret);
212 			fs_unlock(dev, 0);
213 			return ret;
214 		}
215 	}
216 	fs_unlock(dev, 0);
217 	return 0;
218 }
219 
220 static int
221 fs_dev_set_link_down(struct rte_eth_dev *dev)
222 {
223 	struct sub_device *sdev;
224 	uint8_t i;
225 	int ret;
226 
227 	fs_lock(dev, 0);
228 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
229 		DEBUG("Calling rte_eth_dev_set_link_down on sub_device %d", i);
230 		ret = rte_eth_dev_set_link_down(PORT_ID(sdev));
231 		if ((ret = fs_err(sdev, ret))) {
232 			ERROR("Operation rte_eth_dev_set_link_down failed for sub_device %d"
233 			      " with error %d", i, ret);
234 			fs_unlock(dev, 0);
235 			return ret;
236 		}
237 	}
238 	fs_unlock(dev, 0);
239 	return 0;
240 }
241 
242 static void fs_dev_free_queues(struct rte_eth_dev *dev);
243 static void
244 fs_dev_close(struct rte_eth_dev *dev)
245 {
246 	struct sub_device *sdev;
247 	uint8_t i;
248 
249 	fs_lock(dev, 0);
250 	failsafe_hotplug_alarm_cancel(dev);
251 	if (PRIV(dev)->state == DEV_STARTED)
252 		dev->dev_ops->dev_stop(dev);
253 	PRIV(dev)->state = DEV_ACTIVE - 1;
254 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
255 		DEBUG("Closing sub_device %d", i);
256 		failsafe_eth_dev_unregister_callbacks(sdev);
257 		rte_eth_dev_close(PORT_ID(sdev));
258 		sdev->state = DEV_ACTIVE - 1;
259 	}
260 	fs_dev_free_queues(dev);
261 	fs_unlock(dev, 0);
262 }
263 
264 static int
265 fs_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
266 {
267 	struct sub_device *sdev;
268 	uint8_t i;
269 	int ret;
270 	int err = 0;
271 	bool failure = true;
272 
273 	fs_lock(dev, 0);
274 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
275 		uint16_t port_id = ETH(sdev)->data->port_id;
276 
277 		ret = rte_eth_dev_rx_queue_stop(port_id, rx_queue_id);
278 		ret = fs_err(sdev, ret);
279 		if (ret) {
280 			ERROR("Rx queue stop failed for subdevice %d", i);
281 			err = ret;
282 		} else {
283 			failure = false;
284 		}
285 	}
286 	dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
287 	fs_unlock(dev, 0);
288 	/* Return 0 in case of at least one successful queue stop */
289 	return (failure) ? err : 0;
290 }
291 
292 static int
293 fs_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
294 {
295 	struct sub_device *sdev;
296 	uint8_t i;
297 	int ret;
298 
299 	fs_lock(dev, 0);
300 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
301 		uint16_t port_id = ETH(sdev)->data->port_id;
302 
303 		ret = rte_eth_dev_rx_queue_start(port_id, rx_queue_id);
304 		ret = fs_err(sdev, ret);
305 		if (ret) {
306 			ERROR("Rx queue start failed for subdevice %d", i);
307 			fs_rx_queue_stop(dev, rx_queue_id);
308 			fs_unlock(dev, 0);
309 			return ret;
310 		}
311 	}
312 	dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
313 	fs_unlock(dev, 0);
314 	return 0;
315 }
316 
317 static int
318 fs_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
319 {
320 	struct sub_device *sdev;
321 	uint8_t i;
322 	int ret;
323 	int err = 0;
324 	bool failure = true;
325 
326 	fs_lock(dev, 0);
327 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
328 		uint16_t port_id = ETH(sdev)->data->port_id;
329 
330 		ret = rte_eth_dev_tx_queue_stop(port_id, tx_queue_id);
331 		ret = fs_err(sdev, ret);
332 		if (ret) {
333 			ERROR("Tx queue stop failed for subdevice %d", i);
334 			err = ret;
335 		} else {
336 			failure = false;
337 		}
338 	}
339 	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
340 	fs_unlock(dev, 0);
341 	/* Return 0 in case of at least one successful queue stop */
342 	return (failure) ? err : 0;
343 }
344 
345 static int
346 fs_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
347 {
348 	struct sub_device *sdev;
349 	uint8_t i;
350 	int ret;
351 
352 	fs_lock(dev, 0);
353 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
354 		uint16_t port_id = ETH(sdev)->data->port_id;
355 
356 		ret = rte_eth_dev_tx_queue_start(port_id, tx_queue_id);
357 		ret = fs_err(sdev, ret);
358 		if (ret) {
359 			ERROR("Tx queue start failed for subdevice %d", i);
360 			fs_tx_queue_stop(dev, tx_queue_id);
361 			fs_unlock(dev, 0);
362 			return ret;
363 		}
364 	}
365 	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
366 	fs_unlock(dev, 0);
367 	return 0;
368 }
369 
370 static void
371 fs_rx_queue_release(void *queue)
372 {
373 	struct rte_eth_dev *dev;
374 	struct sub_device *sdev;
375 	uint8_t i;
376 	struct rxq *rxq;
377 
378 	if (queue == NULL)
379 		return;
380 	rxq = queue;
381 	dev = &rte_eth_devices[rxq->priv->data->port_id];
382 	fs_lock(dev, 0);
383 	if (rxq->event_fd > 0)
384 		close(rxq->event_fd);
385 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
386 		if (ETH(sdev)->data->rx_queues != NULL &&
387 		    ETH(sdev)->data->rx_queues[rxq->qid] != NULL) {
388 			SUBOPS(sdev, rx_queue_release)
389 				(ETH(sdev)->data->rx_queues[rxq->qid]);
390 		}
391 	}
392 	dev->data->rx_queues[rxq->qid] = NULL;
393 	rte_free(rxq);
394 	fs_unlock(dev, 0);
395 }
396 
397 static int
398 fs_rx_queue_setup(struct rte_eth_dev *dev,
399 		uint16_t rx_queue_id,
400 		uint16_t nb_rx_desc,
401 		unsigned int socket_id,
402 		const struct rte_eth_rxconf *rx_conf,
403 		struct rte_mempool *mb_pool)
404 {
405 	/*
406 	 * FIXME: Add a proper interface in rte_eal_interrupts for
407 	 * allocating eventfd as an interrupt vector.
408 	 * For the time being, fake as if we are using MSIX interrupts,
409 	 * this will cause rte_intr_efd_enable to allocate an eventfd for us.
410 	 */
411 	struct rte_intr_handle intr_handle = {
412 		.type = RTE_INTR_HANDLE_VFIO_MSIX,
413 		.efds = { -1, },
414 	};
415 	struct sub_device *sdev;
416 	struct rxq *rxq;
417 	uint8_t i;
418 	int ret;
419 
420 	fs_lock(dev, 0);
421 	if (rx_conf->rx_deferred_start) {
422 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
423 			if (SUBOPS(sdev, rx_queue_start) == NULL) {
424 				ERROR("Rx queue deferred start is not "
425 					"supported for subdevice %d", i);
426 				fs_unlock(dev, 0);
427 				return -EINVAL;
428 			}
429 		}
430 	}
431 	rxq = dev->data->rx_queues[rx_queue_id];
432 	if (rxq != NULL) {
433 		fs_rx_queue_release(rxq);
434 		dev->data->rx_queues[rx_queue_id] = NULL;
435 	}
436 	rxq = rte_zmalloc(NULL,
437 			  sizeof(*rxq) +
438 			  sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
439 			  RTE_CACHE_LINE_SIZE);
440 	if (rxq == NULL) {
441 		fs_unlock(dev, 0);
442 		return -ENOMEM;
443 	}
444 	FOREACH_SUBDEV(sdev, i, dev)
445 		rte_atomic64_init(&rxq->refcnt[i]);
446 	rxq->qid = rx_queue_id;
447 	rxq->socket_id = socket_id;
448 	rxq->info.mp = mb_pool;
449 	rxq->info.conf = *rx_conf;
450 	rxq->info.nb_desc = nb_rx_desc;
451 	rxq->priv = PRIV(dev);
452 	rxq->sdev = PRIV(dev)->subs;
453 	ret = rte_intr_efd_enable(&intr_handle, 1);
454 	if (ret < 0) {
455 		fs_unlock(dev, 0);
456 		return ret;
457 	}
458 	rxq->event_fd = intr_handle.efds[0];
459 	dev->data->rx_queues[rx_queue_id] = rxq;
460 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
461 		ret = rte_eth_rx_queue_setup(PORT_ID(sdev),
462 				rx_queue_id,
463 				nb_rx_desc, socket_id,
464 				rx_conf, mb_pool);
465 		if ((ret = fs_err(sdev, ret))) {
466 			ERROR("RX queue setup failed for sub_device %d", i);
467 			goto free_rxq;
468 		}
469 	}
470 	fs_unlock(dev, 0);
471 	return 0;
472 free_rxq:
473 	fs_rx_queue_release(rxq);
474 	fs_unlock(dev, 0);
475 	return ret;
476 }
477 
478 static int
479 fs_rx_intr_enable(struct rte_eth_dev *dev, uint16_t idx)
480 {
481 	struct rxq *rxq;
482 	struct sub_device *sdev;
483 	uint8_t i;
484 	int ret;
485 	int rc = 0;
486 
487 	fs_lock(dev, 0);
488 	if (idx >= dev->data->nb_rx_queues) {
489 		rc = -EINVAL;
490 		goto unlock;
491 	}
492 	rxq = dev->data->rx_queues[idx];
493 	if (rxq == NULL || rxq->event_fd <= 0) {
494 		rc = -EINVAL;
495 		goto unlock;
496 	}
497 	/* Fail if proxy service is nor running. */
498 	if (PRIV(dev)->rxp.sstate != SS_RUNNING) {
499 		ERROR("failsafe interrupt services are not running");
500 		rc = -EAGAIN;
501 		goto unlock;
502 	}
503 	rxq->enable_events = 1;
504 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
505 		ret = rte_eth_dev_rx_intr_enable(PORT_ID(sdev), idx);
506 		ret = fs_err(sdev, ret);
507 		if (ret)
508 			rc = ret;
509 	}
510 unlock:
511 	fs_unlock(dev, 0);
512 	if (rc)
513 		rte_errno = -rc;
514 	return rc;
515 }
516 
517 static int
518 fs_rx_intr_disable(struct rte_eth_dev *dev, uint16_t idx)
519 {
520 	struct rxq *rxq;
521 	struct sub_device *sdev;
522 	uint64_t u64;
523 	uint8_t i;
524 	int rc = 0;
525 	int ret;
526 
527 	fs_lock(dev, 0);
528 	if (idx >= dev->data->nb_rx_queues) {
529 		rc = -EINVAL;
530 		goto unlock;
531 	}
532 	rxq = dev->data->rx_queues[idx];
533 	if (rxq == NULL || rxq->event_fd <= 0) {
534 		rc = -EINVAL;
535 		goto unlock;
536 	}
537 	rxq->enable_events = 0;
538 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
539 		ret = rte_eth_dev_rx_intr_disable(PORT_ID(sdev), idx);
540 		ret = fs_err(sdev, ret);
541 		if (ret)
542 			rc = ret;
543 	}
544 	/* Clear pending events */
545 	while (read(rxq->event_fd, &u64, sizeof(uint64_t)) >  0)
546 		;
547 unlock:
548 	fs_unlock(dev, 0);
549 	if (rc)
550 		rte_errno = -rc;
551 	return rc;
552 }
553 
554 static void
555 fs_tx_queue_release(void *queue)
556 {
557 	struct rte_eth_dev *dev;
558 	struct sub_device *sdev;
559 	uint8_t i;
560 	struct txq *txq;
561 
562 	if (queue == NULL)
563 		return;
564 	txq = queue;
565 	dev = &rte_eth_devices[txq->priv->data->port_id];
566 	fs_lock(dev, 0);
567 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
568 		if (ETH(sdev)->data->tx_queues != NULL &&
569 		    ETH(sdev)->data->tx_queues[txq->qid] != NULL) {
570 			SUBOPS(sdev, tx_queue_release)
571 				(ETH(sdev)->data->tx_queues[txq->qid]);
572 		}
573 	}
574 	dev->data->tx_queues[txq->qid] = NULL;
575 	rte_free(txq);
576 	fs_unlock(dev, 0);
577 }
578 
579 static int
580 fs_tx_queue_setup(struct rte_eth_dev *dev,
581 		uint16_t tx_queue_id,
582 		uint16_t nb_tx_desc,
583 		unsigned int socket_id,
584 		const struct rte_eth_txconf *tx_conf)
585 {
586 	struct sub_device *sdev;
587 	struct txq *txq;
588 	uint8_t i;
589 	int ret;
590 
591 	fs_lock(dev, 0);
592 	if (tx_conf->tx_deferred_start) {
593 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
594 			if (SUBOPS(sdev, tx_queue_start) == NULL) {
595 				ERROR("Tx queue deferred start is not "
596 					"supported for subdevice %d", i);
597 				fs_unlock(dev, 0);
598 				return -EINVAL;
599 			}
600 		}
601 	}
602 	txq = dev->data->tx_queues[tx_queue_id];
603 	if (txq != NULL) {
604 		fs_tx_queue_release(txq);
605 		dev->data->tx_queues[tx_queue_id] = NULL;
606 	}
607 	txq = rte_zmalloc("ethdev TX queue",
608 			  sizeof(*txq) +
609 			  sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
610 			  RTE_CACHE_LINE_SIZE);
611 	if (txq == NULL) {
612 		fs_unlock(dev, 0);
613 		return -ENOMEM;
614 	}
615 	FOREACH_SUBDEV(sdev, i, dev)
616 		rte_atomic64_init(&txq->refcnt[i]);
617 	txq->qid = tx_queue_id;
618 	txq->socket_id = socket_id;
619 	txq->info.conf = *tx_conf;
620 	txq->info.nb_desc = nb_tx_desc;
621 	txq->priv = PRIV(dev);
622 	dev->data->tx_queues[tx_queue_id] = txq;
623 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
624 		ret = rte_eth_tx_queue_setup(PORT_ID(sdev),
625 				tx_queue_id,
626 				nb_tx_desc, socket_id,
627 				tx_conf);
628 		if ((ret = fs_err(sdev, ret))) {
629 			ERROR("TX queue setup failed for sub_device %d", i);
630 			goto free_txq;
631 		}
632 	}
633 	fs_unlock(dev, 0);
634 	return 0;
635 free_txq:
636 	fs_tx_queue_release(txq);
637 	fs_unlock(dev, 0);
638 	return ret;
639 }
640 
641 static void
642 fs_dev_free_queues(struct rte_eth_dev *dev)
643 {
644 	uint16_t i;
645 
646 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
647 		fs_rx_queue_release(dev->data->rx_queues[i]);
648 		dev->data->rx_queues[i] = NULL;
649 	}
650 	dev->data->nb_rx_queues = 0;
651 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
652 		fs_tx_queue_release(dev->data->tx_queues[i]);
653 		dev->data->tx_queues[i] = NULL;
654 	}
655 	dev->data->nb_tx_queues = 0;
656 }
657 
658 static int
659 fs_promiscuous_enable(struct rte_eth_dev *dev)
660 {
661 	struct sub_device *sdev;
662 	uint8_t i;
663 	int ret = 0;
664 
665 	fs_lock(dev, 0);
666 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
667 		ret = rte_eth_promiscuous_enable(PORT_ID(sdev));
668 		ret = fs_err(sdev, ret);
669 		if (ret != 0) {
670 			ERROR("Promiscuous mode enable failed for subdevice %d",
671 				PORT_ID(sdev));
672 			break;
673 		}
674 	}
675 	if (ret != 0) {
676 		/* Rollback in the case of failure */
677 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
678 			ret = rte_eth_promiscuous_disable(PORT_ID(sdev));
679 			ret = fs_err(sdev, ret);
680 			if (ret != 0)
681 				ERROR("Promiscuous mode disable during rollback failed for subdevice %d",
682 					PORT_ID(sdev));
683 		}
684 	}
685 	fs_unlock(dev, 0);
686 
687 	return ret;
688 }
689 
690 static int
691 fs_promiscuous_disable(struct rte_eth_dev *dev)
692 {
693 	struct sub_device *sdev;
694 	uint8_t i;
695 	int ret = 0;
696 
697 	fs_lock(dev, 0);
698 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
699 		ret = rte_eth_promiscuous_disable(PORT_ID(sdev));
700 		ret = fs_err(sdev, ret);
701 		if (ret != 0) {
702 			ERROR("Promiscuous mode disable failed for subdevice %d",
703 				PORT_ID(sdev));
704 			break;
705 		}
706 	}
707 	if (ret != 0) {
708 		/* Rollback in the case of failure */
709 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
710 			ret = rte_eth_promiscuous_enable(PORT_ID(sdev));
711 			ret = fs_err(sdev, ret);
712 			if (ret != 0)
713 				ERROR("Promiscuous mode enable during rollback failed for subdevice %d",
714 					PORT_ID(sdev));
715 		}
716 	}
717 	fs_unlock(dev, 0);
718 
719 	return ret;
720 }
721 
722 static int
723 fs_allmulticast_enable(struct rte_eth_dev *dev)
724 {
725 	struct sub_device *sdev;
726 	uint8_t i;
727 	int ret = 0;
728 
729 	fs_lock(dev, 0);
730 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
731 		ret = rte_eth_allmulticast_enable(PORT_ID(sdev));
732 		ret = fs_err(sdev, ret);
733 		if (ret != 0) {
734 			ERROR("All-multicast mode enable failed for subdevice %d",
735 				PORT_ID(sdev));
736 			break;
737 		}
738 	}
739 	if (ret != 0) {
740 		/* Rollback in the case of failure */
741 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
742 			ret = rte_eth_allmulticast_disable(PORT_ID(sdev));
743 			ret = fs_err(sdev, ret);
744 			if (ret != 0)
745 				ERROR("All-multicast mode disable during rollback failed for subdevice %d",
746 					PORT_ID(sdev));
747 		}
748 	}
749 	fs_unlock(dev, 0);
750 
751 	return ret;
752 }
753 
754 static int
755 fs_allmulticast_disable(struct rte_eth_dev *dev)
756 {
757 	struct sub_device *sdev;
758 	uint8_t i;
759 	int ret = 0;
760 
761 	fs_lock(dev, 0);
762 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
763 		ret = rte_eth_allmulticast_disable(PORT_ID(sdev));
764 		ret = fs_err(sdev, ret);
765 		if (ret != 0) {
766 			ERROR("All-multicast mode disable failed for subdevice %d",
767 				PORT_ID(sdev));
768 			break;
769 		}
770 	}
771 	if (ret != 0) {
772 		/* Rollback in the case of failure */
773 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
774 			ret = rte_eth_allmulticast_enable(PORT_ID(sdev));
775 			ret = fs_err(sdev, ret);
776 			if (ret != 0)
777 				ERROR("All-multicast mode enable during rollback failed for subdevice %d",
778 					PORT_ID(sdev));
779 		}
780 	}
781 	fs_unlock(dev, 0);
782 
783 	return ret;
784 }
785 
786 static int
787 fs_link_update(struct rte_eth_dev *dev,
788 		int wait_to_complete)
789 {
790 	struct sub_device *sdev;
791 	uint8_t i;
792 	int ret;
793 
794 	fs_lock(dev, 0);
795 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
796 		DEBUG("Calling link_update on sub_device %d", i);
797 		ret = (SUBOPS(sdev, link_update))(ETH(sdev), wait_to_complete);
798 		if (ret && ret != -1 && sdev->remove == 0 &&
799 		    rte_eth_dev_is_removed(PORT_ID(sdev)) == 0) {
800 			ERROR("Link update failed for sub_device %d with error %d",
801 			      i, ret);
802 			fs_unlock(dev, 0);
803 			return ret;
804 		}
805 	}
806 	if (TX_SUBDEV(dev)) {
807 		struct rte_eth_link *l1;
808 		struct rte_eth_link *l2;
809 
810 		l1 = &dev->data->dev_link;
811 		l2 = &ETH(TX_SUBDEV(dev))->data->dev_link;
812 		if (memcmp(l1, l2, sizeof(*l1))) {
813 			*l1 = *l2;
814 			fs_unlock(dev, 0);
815 			return 0;
816 		}
817 	}
818 	fs_unlock(dev, 0);
819 	return -1;
820 }
821 
822 static int
823 fs_stats_get(struct rte_eth_dev *dev,
824 	     struct rte_eth_stats *stats)
825 {
826 	struct rte_eth_stats backup;
827 	struct sub_device *sdev;
828 	uint8_t i;
829 	int ret;
830 
831 	fs_lock(dev, 0);
832 	rte_memcpy(stats, &PRIV(dev)->stats_accumulator, sizeof(*stats));
833 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
834 		struct rte_eth_stats *snapshot = &sdev->stats_snapshot.stats;
835 		uint64_t *timestamp = &sdev->stats_snapshot.timestamp;
836 
837 		rte_memcpy(&backup, snapshot, sizeof(backup));
838 		ret = rte_eth_stats_get(PORT_ID(sdev), snapshot);
839 		if (ret) {
840 			if (!fs_err(sdev, ret)) {
841 				rte_memcpy(snapshot, &backup, sizeof(backup));
842 				goto inc;
843 			}
844 			ERROR("Operation rte_eth_stats_get failed for sub_device %d with error %d",
845 				  i, ret);
846 			*timestamp = 0;
847 			fs_unlock(dev, 0);
848 			return ret;
849 		}
850 		*timestamp = rte_rdtsc();
851 inc:
852 		failsafe_stats_increment(stats, snapshot);
853 	}
854 	fs_unlock(dev, 0);
855 	return 0;
856 }
857 
858 static int
859 fs_stats_reset(struct rte_eth_dev *dev)
860 {
861 	struct sub_device *sdev;
862 	uint8_t i;
863 	int ret;
864 
865 	fs_lock(dev, 0);
866 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
867 		ret = rte_eth_stats_reset(PORT_ID(sdev));
868 		if (ret) {
869 			if (!fs_err(sdev, ret))
870 				continue;
871 
872 			ERROR("Operation rte_eth_stats_reset failed for sub_device %d with error %d",
873 			      i, ret);
874 			fs_unlock(dev, 0);
875 			return ret;
876 		}
877 		memset(&sdev->stats_snapshot, 0, sizeof(struct rte_eth_stats));
878 	}
879 	memset(&PRIV(dev)->stats_accumulator, 0, sizeof(struct rte_eth_stats));
880 	fs_unlock(dev, 0);
881 
882 	return 0;
883 }
884 
885 static int
886 __fs_xstats_count(struct rte_eth_dev *dev)
887 {
888 	struct sub_device *sdev;
889 	int count = 0;
890 	uint8_t i;
891 	int ret;
892 
893 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
894 		ret = rte_eth_xstats_get_names(PORT_ID(sdev), NULL, 0);
895 		if (ret < 0)
896 			return ret;
897 		count += ret;
898 	}
899 
900 	return count;
901 }
902 
903 static int
904 __fs_xstats_get_names(struct rte_eth_dev *dev,
905 		    struct rte_eth_xstat_name *xstats_names,
906 		    unsigned int limit)
907 {
908 	struct sub_device *sdev;
909 	unsigned int count = 0;
910 	uint8_t i;
911 
912 	/* Caller only cares about count */
913 	if (!xstats_names)
914 		return  __fs_xstats_count(dev);
915 
916 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
917 		struct rte_eth_xstat_name *sub_names = xstats_names + count;
918 		int j, r;
919 
920 		if (count >= limit)
921 			break;
922 
923 		r = rte_eth_xstats_get_names(PORT_ID(sdev),
924 					     sub_names, limit - count);
925 		if (r < 0)
926 			return r;
927 
928 		/* add subN_ prefix to names */
929 		for (j = 0; j < r; j++) {
930 			char *xname = sub_names[j].name;
931 			char tmp[RTE_ETH_XSTATS_NAME_SIZE];
932 
933 			if ((xname[0] == 't' || xname[0] == 'r') &&
934 			    xname[1] == 'x' && xname[2] == '_')
935 				snprintf(tmp, sizeof(tmp), "%.3ssub%u_%s",
936 					 xname, i, xname + 3);
937 			else
938 				snprintf(tmp, sizeof(tmp), "sub%u_%s",
939 					 i, xname);
940 
941 			strlcpy(xname, tmp, RTE_ETH_XSTATS_NAME_SIZE);
942 		}
943 		count += r;
944 	}
945 	return count;
946 }
947 
948 static int
949 fs_xstats_get_names(struct rte_eth_dev *dev,
950 		    struct rte_eth_xstat_name *xstats_names,
951 		    unsigned int limit)
952 {
953 	int ret;
954 
955 	fs_lock(dev, 0);
956 	ret = __fs_xstats_get_names(dev, xstats_names, limit);
957 	fs_unlock(dev, 0);
958 	return ret;
959 }
960 
961 static int
962 __fs_xstats_get(struct rte_eth_dev *dev,
963 	      struct rte_eth_xstat *xstats,
964 	      unsigned int n)
965 {
966 	unsigned int count = 0;
967 	struct sub_device *sdev;
968 	uint8_t i;
969 	int j, ret;
970 
971 	ret = __fs_xstats_count(dev);
972 	/*
973 	 * if error
974 	 * or caller did not give enough space
975 	 * or just querying
976 	 */
977 	if (ret < 0 || ret > (int)n || xstats == NULL)
978 		return ret;
979 
980 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
981 		ret = rte_eth_xstats_get(PORT_ID(sdev), xstats, n);
982 		if (ret < 0)
983 			return ret;
984 
985 		if (ret > (int)n)
986 			return n + count;
987 
988 		/* add offset to id's from sub-device */
989 		for (j = 0; j < ret; j++)
990 			xstats[j].id += count;
991 
992 		xstats += ret;
993 		n -= ret;
994 		count += ret;
995 	}
996 
997 	return count;
998 }
999 
1000 static int
1001 fs_xstats_get(struct rte_eth_dev *dev,
1002 	      struct rte_eth_xstat *xstats,
1003 	      unsigned int n)
1004 {
1005 	int ret;
1006 
1007 	fs_lock(dev, 0);
1008 	ret = __fs_xstats_get(dev, xstats, n);
1009 	fs_unlock(dev, 0);
1010 
1011 	return ret;
1012 }
1013 
1014 
1015 static int
1016 fs_xstats_reset(struct rte_eth_dev *dev)
1017 {
1018 	struct sub_device *sdev;
1019 	uint8_t i;
1020 	int r = 0;
1021 
1022 	fs_lock(dev, 0);
1023 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1024 		r = rte_eth_xstats_reset(PORT_ID(sdev));
1025 		if (r < 0)
1026 			break;
1027 	}
1028 	fs_unlock(dev, 0);
1029 
1030 	return r;
1031 }
1032 
1033 static void
1034 fs_dev_merge_desc_lim(struct rte_eth_desc_lim *to,
1035 		      const struct rte_eth_desc_lim *from)
1036 {
1037 	to->nb_max = RTE_MIN(to->nb_max, from->nb_max);
1038 	to->nb_min = RTE_MAX(to->nb_min, from->nb_min);
1039 	to->nb_align = RTE_MAX(to->nb_align, from->nb_align);
1040 
1041 	to->nb_seg_max = RTE_MIN(to->nb_seg_max, from->nb_seg_max);
1042 	to->nb_mtu_seg_max = RTE_MIN(to->nb_mtu_seg_max, from->nb_mtu_seg_max);
1043 }
1044 
1045 /*
1046  * Merge the information from sub-devices.
1047  *
1048  * The reported values must be the common subset of all sub devices
1049  */
1050 static void
1051 fs_dev_merge_info(struct rte_eth_dev_info *info,
1052 		  const struct rte_eth_dev_info *sinfo)
1053 {
1054 	info->max_rx_pktlen = RTE_MIN(info->max_rx_pktlen, sinfo->max_rx_pktlen);
1055 	info->max_rx_queues = RTE_MIN(info->max_rx_queues, sinfo->max_rx_queues);
1056 	info->max_tx_queues = RTE_MIN(info->max_tx_queues, sinfo->max_tx_queues);
1057 	info->max_mac_addrs = RTE_MIN(info->max_mac_addrs, sinfo->max_mac_addrs);
1058 	info->max_hash_mac_addrs = RTE_MIN(info->max_hash_mac_addrs,
1059 					sinfo->max_hash_mac_addrs);
1060 	info->max_vmdq_pools = RTE_MIN(info->max_vmdq_pools, sinfo->max_vmdq_pools);
1061 	info->max_vfs = RTE_MIN(info->max_vfs, sinfo->max_vfs);
1062 
1063 	fs_dev_merge_desc_lim(&info->rx_desc_lim, &sinfo->rx_desc_lim);
1064 	fs_dev_merge_desc_lim(&info->tx_desc_lim, &sinfo->tx_desc_lim);
1065 
1066 	info->rx_offload_capa &= sinfo->rx_offload_capa;
1067 	info->tx_offload_capa &= sinfo->tx_offload_capa;
1068 	info->rx_queue_offload_capa &= sinfo->rx_queue_offload_capa;
1069 	info->tx_queue_offload_capa &= sinfo->tx_queue_offload_capa;
1070 	info->flow_type_rss_offloads &= sinfo->flow_type_rss_offloads;
1071 	info->hash_key_size = RTE_MIN(info->hash_key_size,
1072 				      sinfo->hash_key_size);
1073 }
1074 
1075 /**
1076  * Fail-safe dev_infos_get rules:
1077  *
1078  * No sub_device:
1079  *   Numerables:
1080  *      Use the maximum possible values for any field, so as not
1081  *      to impede any further configuration effort.
1082  *   Capabilities:
1083  *      Limits capabilities to those that are understood by the
1084  *      fail-safe PMD. This understanding stems from the fail-safe
1085  *      being capable of verifying that the related capability is
1086  *      expressed within the device configuration (struct rte_eth_conf).
1087  *
1088  * At least one probed sub_device:
1089  *   Numerables:
1090  *      Uses values from the active probed sub_device
1091  *      The rationale here is that if any sub_device is less capable
1092  *      (for example concerning the number of queues) than the active
1093  *      sub_device, then its subsequent configuration will fail.
1094  *      It is impossible to foresee this failure when the failing sub_device
1095  *      is supposed to be plugged-in later on, so the configuration process
1096  *      is the single point of failure and error reporting.
1097  *   Capabilities:
1098  *      Uses a logical AND of RX capabilities among
1099  *      all sub_devices and the default capabilities.
1100  *      Uses a logical AND of TX capabilities among
1101  *      the active probed sub_device and the default capabilities.
1102  *      Uses a logical AND of device capabilities among
1103  *      all sub_devices and the default capabilities.
1104  *
1105  */
1106 static int
1107 fs_dev_infos_get(struct rte_eth_dev *dev,
1108 		  struct rte_eth_dev_info *infos)
1109 {
1110 	struct sub_device *sdev;
1111 	uint8_t i;
1112 	int ret;
1113 
1114 	/* Use maximum upper bounds by default */
1115 	infos->max_rx_pktlen = UINT32_MAX;
1116 	infos->max_rx_queues = RTE_MAX_QUEUES_PER_PORT;
1117 	infos->max_tx_queues = RTE_MAX_QUEUES_PER_PORT;
1118 	infos->max_mac_addrs = FAILSAFE_MAX_ETHADDR;
1119 	infos->max_hash_mac_addrs = UINT32_MAX;
1120 	infos->max_vfs = UINT16_MAX;
1121 	infos->max_vmdq_pools = UINT16_MAX;
1122 	infos->hash_key_size = UINT8_MAX;
1123 
1124 	/*
1125 	 * Set of capabilities that can be verified upon
1126 	 * configuring a sub-device.
1127 	 */
1128 	infos->rx_offload_capa =
1129 		DEV_RX_OFFLOAD_VLAN_STRIP |
1130 		DEV_RX_OFFLOAD_IPV4_CKSUM |
1131 		DEV_RX_OFFLOAD_UDP_CKSUM |
1132 		DEV_RX_OFFLOAD_TCP_CKSUM |
1133 		DEV_RX_OFFLOAD_TCP_LRO |
1134 		DEV_RX_OFFLOAD_QINQ_STRIP |
1135 		DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM |
1136 		DEV_RX_OFFLOAD_MACSEC_STRIP |
1137 		DEV_RX_OFFLOAD_HEADER_SPLIT |
1138 		DEV_RX_OFFLOAD_VLAN_FILTER |
1139 		DEV_RX_OFFLOAD_VLAN_EXTEND |
1140 		DEV_RX_OFFLOAD_JUMBO_FRAME |
1141 		DEV_RX_OFFLOAD_SCATTER |
1142 		DEV_RX_OFFLOAD_TIMESTAMP |
1143 		DEV_RX_OFFLOAD_SECURITY;
1144 
1145 	infos->rx_queue_offload_capa =
1146 		DEV_RX_OFFLOAD_VLAN_STRIP |
1147 		DEV_RX_OFFLOAD_IPV4_CKSUM |
1148 		DEV_RX_OFFLOAD_UDP_CKSUM |
1149 		DEV_RX_OFFLOAD_TCP_CKSUM |
1150 		DEV_RX_OFFLOAD_TCP_LRO |
1151 		DEV_RX_OFFLOAD_QINQ_STRIP |
1152 		DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM |
1153 		DEV_RX_OFFLOAD_MACSEC_STRIP |
1154 		DEV_RX_OFFLOAD_HEADER_SPLIT |
1155 		DEV_RX_OFFLOAD_VLAN_FILTER |
1156 		DEV_RX_OFFLOAD_VLAN_EXTEND |
1157 		DEV_RX_OFFLOAD_JUMBO_FRAME |
1158 		DEV_RX_OFFLOAD_SCATTER |
1159 		DEV_RX_OFFLOAD_TIMESTAMP |
1160 		DEV_RX_OFFLOAD_SECURITY;
1161 
1162 	infos->tx_offload_capa =
1163 		DEV_TX_OFFLOAD_MULTI_SEGS |
1164 		DEV_TX_OFFLOAD_MBUF_FAST_FREE |
1165 		DEV_TX_OFFLOAD_IPV4_CKSUM |
1166 		DEV_TX_OFFLOAD_UDP_CKSUM |
1167 		DEV_TX_OFFLOAD_TCP_CKSUM |
1168 		DEV_TX_OFFLOAD_TCP_TSO;
1169 
1170 	infos->flow_type_rss_offloads =
1171 		ETH_RSS_IP |
1172 		ETH_RSS_UDP |
1173 		ETH_RSS_TCP;
1174 	infos->dev_capa =
1175 		RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
1176 		RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
1177 
1178 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
1179 		struct rte_eth_dev_info sub_info;
1180 
1181 		ret = rte_eth_dev_info_get(PORT_ID(sdev), &sub_info);
1182 		ret = fs_err(sdev, ret);
1183 		if (ret != 0)
1184 			return ret;
1185 
1186 		fs_dev_merge_info(infos, &sub_info);
1187 	}
1188 
1189 	return 0;
1190 }
1191 
1192 static const uint32_t *
1193 fs_dev_supported_ptypes_get(struct rte_eth_dev *dev)
1194 {
1195 	struct sub_device *sdev;
1196 	struct rte_eth_dev *edev;
1197 	const uint32_t *ret;
1198 
1199 	fs_lock(dev, 0);
1200 	sdev = TX_SUBDEV(dev);
1201 	if (sdev == NULL) {
1202 		ret = NULL;
1203 		goto unlock;
1204 	}
1205 	edev = ETH(sdev);
1206 	/* ENOTSUP: counts as no supported ptypes */
1207 	if (SUBOPS(sdev, dev_supported_ptypes_get) == NULL) {
1208 		ret = NULL;
1209 		goto unlock;
1210 	}
1211 	/*
1212 	 * The API does not permit to do a clean AND of all ptypes,
1213 	 * It is also incomplete by design and we do not really care
1214 	 * to have a best possible value in this context.
1215 	 * We just return the ptypes of the device of highest
1216 	 * priority, usually the PREFERRED device.
1217 	 */
1218 	ret = SUBOPS(sdev, dev_supported_ptypes_get)(edev);
1219 unlock:
1220 	fs_unlock(dev, 0);
1221 	return ret;
1222 }
1223 
1224 static int
1225 fs_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1226 {
1227 	struct sub_device *sdev;
1228 	uint8_t i;
1229 	int ret;
1230 
1231 	fs_lock(dev, 0);
1232 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1233 		DEBUG("Calling rte_eth_dev_set_mtu on sub_device %d", i);
1234 		ret = rte_eth_dev_set_mtu(PORT_ID(sdev), mtu);
1235 		if ((ret = fs_err(sdev, ret))) {
1236 			ERROR("Operation rte_eth_dev_set_mtu failed for sub_device %d with error %d",
1237 			      i, ret);
1238 			fs_unlock(dev, 0);
1239 			return ret;
1240 		}
1241 	}
1242 	fs_unlock(dev, 0);
1243 	return 0;
1244 }
1245 
1246 static int
1247 fs_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1248 {
1249 	struct sub_device *sdev;
1250 	uint8_t i;
1251 	int ret;
1252 
1253 	fs_lock(dev, 0);
1254 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1255 		DEBUG("Calling rte_eth_dev_vlan_filter on sub_device %d", i);
1256 		ret = rte_eth_dev_vlan_filter(PORT_ID(sdev), vlan_id, on);
1257 		if ((ret = fs_err(sdev, ret))) {
1258 			ERROR("Operation rte_eth_dev_vlan_filter failed for sub_device %d"
1259 			      " with error %d", i, ret);
1260 			fs_unlock(dev, 0);
1261 			return ret;
1262 		}
1263 	}
1264 	fs_unlock(dev, 0);
1265 	return 0;
1266 }
1267 
1268 static int
1269 fs_flow_ctrl_get(struct rte_eth_dev *dev,
1270 		struct rte_eth_fc_conf *fc_conf)
1271 {
1272 	struct sub_device *sdev;
1273 	int ret;
1274 
1275 	fs_lock(dev, 0);
1276 	sdev = TX_SUBDEV(dev);
1277 	if (sdev == NULL) {
1278 		ret = 0;
1279 		goto unlock;
1280 	}
1281 	if (SUBOPS(sdev, flow_ctrl_get) == NULL) {
1282 		ret = -ENOTSUP;
1283 		goto unlock;
1284 	}
1285 	ret = SUBOPS(sdev, flow_ctrl_get)(ETH(sdev), fc_conf);
1286 unlock:
1287 	fs_unlock(dev, 0);
1288 	return ret;
1289 }
1290 
1291 static int
1292 fs_flow_ctrl_set(struct rte_eth_dev *dev,
1293 		struct rte_eth_fc_conf *fc_conf)
1294 {
1295 	struct sub_device *sdev;
1296 	uint8_t i;
1297 	int ret;
1298 
1299 	fs_lock(dev, 0);
1300 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1301 		DEBUG("Calling rte_eth_dev_flow_ctrl_set on sub_device %d", i);
1302 		ret = rte_eth_dev_flow_ctrl_set(PORT_ID(sdev), fc_conf);
1303 		if ((ret = fs_err(sdev, ret))) {
1304 			ERROR("Operation rte_eth_dev_flow_ctrl_set failed for sub_device %d"
1305 			      " with error %d", i, ret);
1306 			fs_unlock(dev, 0);
1307 			return ret;
1308 		}
1309 	}
1310 	fs_unlock(dev, 0);
1311 	return 0;
1312 }
1313 
1314 static void
1315 fs_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1316 {
1317 	struct sub_device *sdev;
1318 	uint8_t i;
1319 
1320 	fs_lock(dev, 0);
1321 	/* No check: already done within the rte_eth_dev_mac_addr_remove
1322 	 * call for the fail-safe device.
1323 	 */
1324 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
1325 		rte_eth_dev_mac_addr_remove(PORT_ID(sdev),
1326 				&dev->data->mac_addrs[index]);
1327 	PRIV(dev)->mac_addr_pool[index] = 0;
1328 	fs_unlock(dev, 0);
1329 }
1330 
1331 static int
1332 fs_mac_addr_add(struct rte_eth_dev *dev,
1333 		struct rte_ether_addr *mac_addr,
1334 		uint32_t index,
1335 		uint32_t vmdq)
1336 {
1337 	struct sub_device *sdev;
1338 	int ret;
1339 	uint8_t i;
1340 
1341 	RTE_ASSERT(index < FAILSAFE_MAX_ETHADDR);
1342 	fs_lock(dev, 0);
1343 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1344 		ret = rte_eth_dev_mac_addr_add(PORT_ID(sdev), mac_addr, vmdq);
1345 		if ((ret = fs_err(sdev, ret))) {
1346 			ERROR("Operation rte_eth_dev_mac_addr_add failed for sub_device %"
1347 			      PRIu8 " with error %d", i, ret);
1348 			fs_unlock(dev, 0);
1349 			return ret;
1350 		}
1351 	}
1352 	if (index >= PRIV(dev)->nb_mac_addr) {
1353 		DEBUG("Growing mac_addrs array");
1354 		PRIV(dev)->nb_mac_addr = index;
1355 	}
1356 	PRIV(dev)->mac_addr_pool[index] = vmdq;
1357 	fs_unlock(dev, 0);
1358 	return 0;
1359 }
1360 
1361 static int
1362 fs_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1363 {
1364 	struct sub_device *sdev;
1365 	uint8_t i;
1366 	int ret;
1367 
1368 	fs_lock(dev, 0);
1369 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1370 		ret = rte_eth_dev_default_mac_addr_set(PORT_ID(sdev), mac_addr);
1371 		ret = fs_err(sdev, ret);
1372 		if (ret) {
1373 			ERROR("Operation rte_eth_dev_mac_addr_set failed for sub_device %d with error %d",
1374 				i, ret);
1375 			fs_unlock(dev, 0);
1376 			return ret;
1377 		}
1378 	}
1379 	fs_unlock(dev, 0);
1380 
1381 	return 0;
1382 }
1383 
1384 static int
1385 fs_set_mc_addr_list(struct rte_eth_dev *dev,
1386 		    struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1387 {
1388 	struct sub_device *sdev;
1389 	uint8_t i;
1390 	int ret;
1391 	void *mcast_addrs;
1392 
1393 	fs_lock(dev, 0);
1394 
1395 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1396 		ret = rte_eth_dev_set_mc_addr_list(PORT_ID(sdev),
1397 						   mc_addr_set, nb_mc_addr);
1398 		if (ret != 0) {
1399 			ERROR("Operation rte_eth_dev_set_mc_addr_list failed for sub_device %d with error %d",
1400 			      i, ret);
1401 			goto rollback;
1402 		}
1403 	}
1404 
1405 	mcast_addrs = rte_realloc(PRIV(dev)->mcast_addrs,
1406 		nb_mc_addr * sizeof(PRIV(dev)->mcast_addrs[0]), 0);
1407 	if (mcast_addrs == NULL && nb_mc_addr > 0) {
1408 		ret = -ENOMEM;
1409 		goto rollback;
1410 	}
1411 	rte_memcpy(mcast_addrs, mc_addr_set,
1412 		   nb_mc_addr * sizeof(PRIV(dev)->mcast_addrs[0]));
1413 	PRIV(dev)->nb_mcast_addr = nb_mc_addr;
1414 	PRIV(dev)->mcast_addrs = mcast_addrs;
1415 
1416 	fs_unlock(dev, 0);
1417 	return 0;
1418 
1419 rollback:
1420 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1421 		int rc = rte_eth_dev_set_mc_addr_list(PORT_ID(sdev),
1422 			PRIV(dev)->mcast_addrs,	PRIV(dev)->nb_mcast_addr);
1423 		if (rc != 0) {
1424 			ERROR("Multicast MAC address list rollback for sub_device %d failed with error %d",
1425 			      i, rc);
1426 		}
1427 	}
1428 
1429 	fs_unlock(dev, 0);
1430 	return ret;
1431 }
1432 
1433 static int
1434 fs_rss_hash_update(struct rte_eth_dev *dev,
1435 			struct rte_eth_rss_conf *rss_conf)
1436 {
1437 	struct sub_device *sdev;
1438 	uint8_t i;
1439 	int ret;
1440 
1441 	fs_lock(dev, 0);
1442 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1443 		ret = rte_eth_dev_rss_hash_update(PORT_ID(sdev), rss_conf);
1444 		ret = fs_err(sdev, ret);
1445 		if (ret) {
1446 			ERROR("Operation rte_eth_dev_rss_hash_update"
1447 				" failed for sub_device %d with error %d",
1448 				i, ret);
1449 			fs_unlock(dev, 0);
1450 			return ret;
1451 		}
1452 	}
1453 	fs_unlock(dev, 0);
1454 
1455 	return 0;
1456 }
1457 
1458 static int
1459 fs_filter_ctrl(struct rte_eth_dev *dev __rte_unused,
1460 		enum rte_filter_type type,
1461 		enum rte_filter_op op,
1462 		void *arg)
1463 {
1464 	if (type == RTE_ETH_FILTER_GENERIC &&
1465 	    op == RTE_ETH_FILTER_GET) {
1466 		*(const void **)arg = &fs_flow_ops;
1467 		return 0;
1468 	}
1469 	return -ENOTSUP;
1470 }
1471 
1472 const struct eth_dev_ops failsafe_ops = {
1473 	.dev_configure = fs_dev_configure,
1474 	.dev_start = fs_dev_start,
1475 	.dev_stop = fs_dev_stop,
1476 	.dev_set_link_down = fs_dev_set_link_down,
1477 	.dev_set_link_up = fs_dev_set_link_up,
1478 	.dev_close = fs_dev_close,
1479 	.promiscuous_enable = fs_promiscuous_enable,
1480 	.promiscuous_disable = fs_promiscuous_disable,
1481 	.allmulticast_enable = fs_allmulticast_enable,
1482 	.allmulticast_disable = fs_allmulticast_disable,
1483 	.link_update = fs_link_update,
1484 	.stats_get = fs_stats_get,
1485 	.stats_reset = fs_stats_reset,
1486 	.xstats_get = fs_xstats_get,
1487 	.xstats_get_names = fs_xstats_get_names,
1488 	.xstats_reset = fs_xstats_reset,
1489 	.dev_infos_get = fs_dev_infos_get,
1490 	.dev_supported_ptypes_get = fs_dev_supported_ptypes_get,
1491 	.mtu_set = fs_mtu_set,
1492 	.vlan_filter_set = fs_vlan_filter_set,
1493 	.rx_queue_start = fs_rx_queue_start,
1494 	.rx_queue_stop = fs_rx_queue_stop,
1495 	.tx_queue_start = fs_tx_queue_start,
1496 	.tx_queue_stop = fs_tx_queue_stop,
1497 	.rx_queue_setup = fs_rx_queue_setup,
1498 	.tx_queue_setup = fs_tx_queue_setup,
1499 	.rx_queue_release = fs_rx_queue_release,
1500 	.tx_queue_release = fs_tx_queue_release,
1501 	.rx_queue_intr_enable = fs_rx_intr_enable,
1502 	.rx_queue_intr_disable = fs_rx_intr_disable,
1503 	.flow_ctrl_get = fs_flow_ctrl_get,
1504 	.flow_ctrl_set = fs_flow_ctrl_set,
1505 	.mac_addr_remove = fs_mac_addr_remove,
1506 	.mac_addr_add = fs_mac_addr_add,
1507 	.mac_addr_set = fs_mac_addr_set,
1508 	.set_mc_addr_list = fs_set_mc_addr_list,
1509 	.rss_hash_update = fs_rss_hash_update,
1510 	.filter_ctrl = fs_filter_ctrl,
1511 };
1512