xref: /dpdk/drivers/net/failsafe/failsafe_ops.c (revision 084d0cdb572f87ec6a52d32f2f7890fd337ddfba)
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 #ifdef RTE_EXEC_ENV_LINUX
10 #include <sys/eventfd.h>
11 #endif
12 
13 #include <rte_debug.h>
14 #include <rte_atomic.h>
15 #include <ethdev_driver.h>
16 #include <rte_malloc.h>
17 #include <rte_flow.h>
18 #include <rte_cycles.h>
19 #include <rte_ethdev.h>
20 #include <rte_string_fns.h>
21 
22 #include "failsafe_private.h"
23 
24 static int
25 fs_dev_configure(struct rte_eth_dev *dev)
26 {
27 	struct sub_device *sdev;
28 	uint8_t i;
29 	int ret;
30 
31 	ret = fs_lock(dev, 0);
32 	if (ret != 0)
33 		return ret;
34 	FOREACH_SUBDEV(sdev, i, dev) {
35 		int rmv_interrupt = 0;
36 		int lsc_interrupt = 0;
37 		int lsc_enabled;
38 
39 		if (sdev->state != DEV_PROBED &&
40 		    !(PRIV(dev)->alarm_lock == 0 && sdev->state == DEV_ACTIVE))
41 			continue;
42 
43 		rmv_interrupt = ETH(sdev)->data->dev_flags &
44 				RTE_ETH_DEV_INTR_RMV;
45 		if (rmv_interrupt) {
46 			DEBUG("Enabling RMV interrupts for sub_device %d", i);
47 			dev->data->dev_conf.intr_conf.rmv = 1;
48 		} else {
49 			DEBUG("sub_device %d does not support RMV event", i);
50 		}
51 		lsc_enabled = dev->data->dev_conf.intr_conf.lsc;
52 		lsc_interrupt = lsc_enabled &&
53 				(ETH(sdev)->data->dev_flags &
54 				 RTE_ETH_DEV_INTR_LSC);
55 		if (lsc_interrupt) {
56 			DEBUG("Enabling LSC interrupts for sub_device %d", i);
57 			dev->data->dev_conf.intr_conf.lsc = 1;
58 		} else if (lsc_enabled && !lsc_interrupt) {
59 			DEBUG("Disabling LSC interrupts for sub_device %d", i);
60 			dev->data->dev_conf.intr_conf.lsc = 0;
61 		}
62 		DEBUG("Configuring sub-device %d", i);
63 		ret = rte_eth_dev_configure(PORT_ID(sdev),
64 					dev->data->nb_rx_queues,
65 					dev->data->nb_tx_queues,
66 					&dev->data->dev_conf);
67 		if (ret) {
68 			if (!fs_err(sdev, ret))
69 				continue;
70 			ERROR("Could not configure sub_device %d", i);
71 			fs_unlock(dev, 0);
72 			return ret;
73 		}
74 		if (rmv_interrupt && sdev->rmv_callback == 0) {
75 			ret = rte_eth_dev_callback_register(PORT_ID(sdev),
76 					RTE_ETH_EVENT_INTR_RMV,
77 					failsafe_eth_rmv_event_callback,
78 					sdev);
79 			if (ret)
80 				WARN("Failed to register RMV callback for sub_device %d",
81 				     SUB_ID(sdev));
82 			else
83 				sdev->rmv_callback = 1;
84 		}
85 		dev->data->dev_conf.intr_conf.rmv = 0;
86 		if (lsc_interrupt && sdev->lsc_callback == 0) {
87 			ret = rte_eth_dev_callback_register(PORT_ID(sdev),
88 						RTE_ETH_EVENT_INTR_LSC,
89 						failsafe_eth_lsc_event_callback,
90 						dev);
91 			if (ret)
92 				WARN("Failed to register LSC callback for sub_device %d",
93 				     SUB_ID(sdev));
94 			else
95 				sdev->lsc_callback = 1;
96 		}
97 		dev->data->dev_conf.intr_conf.lsc = lsc_enabled;
98 		sdev->state = DEV_ACTIVE;
99 	}
100 	if (PRIV(dev)->state < DEV_ACTIVE)
101 		PRIV(dev)->state = DEV_ACTIVE;
102 	fs_unlock(dev, 0);
103 	return 0;
104 }
105 
106 static void
107 fs_set_queues_state_start(struct rte_eth_dev *dev)
108 {
109 	struct rxq *rxq;
110 	struct txq *txq;
111 	uint16_t i;
112 
113 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
114 		__rte_assume(i < RTE_MAX_QUEUES_PER_PORT);
115 		rxq = dev->data->rx_queues[i];
116 		if (rxq != NULL && !rxq->info.conf.rx_deferred_start)
117 			dev->data->rx_queue_state[i] =
118 						RTE_ETH_QUEUE_STATE_STARTED;
119 	}
120 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
121 		__rte_assume(i < RTE_MAX_QUEUES_PER_PORT);
122 		txq = dev->data->tx_queues[i];
123 		if (txq != NULL && !txq->info.conf.tx_deferred_start)
124 			dev->data->tx_queue_state[i] =
125 						RTE_ETH_QUEUE_STATE_STARTED;
126 	}
127 }
128 
129 static int
130 fs_dev_start(struct rte_eth_dev *dev)
131 {
132 	struct sub_device *sdev;
133 	uint8_t i;
134 	int ret;
135 
136 	ret = fs_lock(dev, 0);
137 	if (ret != 0)
138 		return ret;
139 	ret = failsafe_rx_intr_install(dev);
140 	if (ret) {
141 		fs_unlock(dev, 0);
142 		return ret;
143 	}
144 	FOREACH_SUBDEV(sdev, i, dev) {
145 		if (sdev->state != DEV_ACTIVE)
146 			continue;
147 		DEBUG("Starting sub_device %d", i);
148 		ret = rte_eth_dev_start(PORT_ID(sdev));
149 		if (ret) {
150 			if (!fs_err(sdev, ret))
151 				continue;
152 			fs_unlock(dev, 0);
153 			return ret;
154 		}
155 		ret = failsafe_rx_intr_install_subdevice(sdev);
156 		if (ret) {
157 			if (!fs_err(sdev, ret))
158 				continue;
159 			if (fs_err(sdev, rte_eth_dev_stop(PORT_ID(sdev))) < 0)
160 				ERROR("Failed to stop sub-device %u",
161 				      SUB_ID(sdev));
162 			fs_unlock(dev, 0);
163 			return ret;
164 		}
165 		sdev->state = DEV_STARTED;
166 	}
167 	if (PRIV(dev)->state < DEV_STARTED) {
168 		PRIV(dev)->state = DEV_STARTED;
169 		fs_set_queues_state_start(dev);
170 	}
171 	fs_switch_dev(dev, NULL);
172 	fs_unlock(dev, 0);
173 	return 0;
174 }
175 
176 static void
177 fs_set_queues_state_stop(struct rte_eth_dev *dev)
178 {
179 	uint16_t i;
180 
181 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
182 		__rte_assume(i < RTE_MAX_QUEUES_PER_PORT);
183 		if (dev->data->rx_queues[i] != NULL)
184 			dev->data->rx_queue_state[i] =
185 						RTE_ETH_QUEUE_STATE_STOPPED;
186 	}
187 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
188 		__rte_assume(i < RTE_MAX_QUEUES_PER_PORT);
189 		if (dev->data->tx_queues[i] != NULL)
190 			dev->data->tx_queue_state[i] =
191 						RTE_ETH_QUEUE_STATE_STOPPED;
192 	}
193 }
194 
195 static int
196 fs_dev_stop(struct rte_eth_dev *dev)
197 {
198 	struct sub_device *sdev;
199 	uint8_t i;
200 	int ret;
201 
202 	ret = fs_lock(dev, 0);
203 	if (ret != 0)
204 		return ret;
205 	PRIV(dev)->state = DEV_STARTED - 1;
206 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_STARTED) {
207 		ret = rte_eth_dev_stop(PORT_ID(sdev));
208 		if (fs_err(sdev, ret) < 0) {
209 			ERROR("Failed to stop device %u",
210 			      PORT_ID(sdev));
211 			PRIV(dev)->state = DEV_STARTED + 1;
212 			fs_unlock(dev, 0);
213 			return ret;
214 		}
215 		failsafe_rx_intr_uninstall_subdevice(sdev);
216 		sdev->state = DEV_STARTED - 1;
217 	}
218 	failsafe_rx_intr_uninstall(dev);
219 	fs_set_queues_state_stop(dev);
220 	fs_unlock(dev, 0);
221 
222 	return 0;
223 }
224 
225 static int
226 fs_dev_set_link_up(struct rte_eth_dev *dev)
227 {
228 	struct sub_device *sdev;
229 	uint8_t i;
230 	int ret;
231 
232 	ret = fs_lock(dev, 0);
233 	if (ret != 0)
234 		return ret;
235 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
236 		DEBUG("Calling rte_eth_dev_set_link_up on sub_device %d", i);
237 		ret = rte_eth_dev_set_link_up(PORT_ID(sdev));
238 		if ((ret = fs_err(sdev, ret))) {
239 			ERROR("Operation rte_eth_dev_set_link_up failed for sub_device %d"
240 			      " with error %d", i, ret);
241 			fs_unlock(dev, 0);
242 			return ret;
243 		}
244 	}
245 	fs_unlock(dev, 0);
246 	return 0;
247 }
248 
249 static int
250 fs_dev_set_link_down(struct rte_eth_dev *dev)
251 {
252 	struct sub_device *sdev;
253 	uint8_t i;
254 	int ret;
255 
256 	ret = fs_lock(dev, 0);
257 	if (ret != 0)
258 		return ret;
259 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
260 		DEBUG("Calling rte_eth_dev_set_link_down on sub_device %d", i);
261 		ret = rte_eth_dev_set_link_down(PORT_ID(sdev));
262 		if ((ret = fs_err(sdev, ret))) {
263 			ERROR("Operation rte_eth_dev_set_link_down failed for sub_device %d"
264 			      " with error %d", i, ret);
265 			fs_unlock(dev, 0);
266 			return ret;
267 		}
268 	}
269 	fs_unlock(dev, 0);
270 	return 0;
271 }
272 
273 static int
274 fs_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
275 {
276 	struct sub_device *sdev;
277 	uint8_t i;
278 	int ret;
279 	int err = 0;
280 	bool failure = true;
281 
282 	ret = fs_lock(dev, 0);
283 	if (ret != 0)
284 		return ret;
285 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
286 		uint16_t port_id = ETH(sdev)->data->port_id;
287 
288 		ret = rte_eth_dev_rx_queue_stop(port_id, rx_queue_id);
289 		ret = fs_err(sdev, ret);
290 		if (ret) {
291 			ERROR("Rx queue stop failed for subdevice %d", i);
292 			err = ret;
293 		} else {
294 			failure = false;
295 		}
296 	}
297 	dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
298 	fs_unlock(dev, 0);
299 	/* Return 0 in case of at least one successful queue stop */
300 	return (failure) ? err : 0;
301 }
302 
303 static int
304 fs_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
305 {
306 	struct sub_device *sdev;
307 	uint8_t i;
308 	int ret;
309 
310 	ret = fs_lock(dev, 0);
311 	if (ret != 0)
312 		return ret;
313 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
314 		uint16_t port_id = ETH(sdev)->data->port_id;
315 
316 		ret = rte_eth_dev_rx_queue_start(port_id, rx_queue_id);
317 		ret = fs_err(sdev, ret);
318 		if (ret) {
319 			ERROR("Rx queue start failed for subdevice %d", i);
320 			fs_rx_queue_stop(dev, rx_queue_id);
321 			fs_unlock(dev, 0);
322 			return ret;
323 		}
324 	}
325 	dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
326 	fs_unlock(dev, 0);
327 	return 0;
328 }
329 
330 static int
331 fs_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
332 {
333 	struct sub_device *sdev;
334 	uint8_t i;
335 	int ret;
336 	int err = 0;
337 	bool failure = true;
338 
339 	ret = fs_lock(dev, 0);
340 	if (ret != 0)
341 		return ret;
342 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
343 		uint16_t port_id = ETH(sdev)->data->port_id;
344 
345 		ret = rte_eth_dev_tx_queue_stop(port_id, tx_queue_id);
346 		ret = fs_err(sdev, ret);
347 		if (ret) {
348 			ERROR("Tx queue stop failed for subdevice %d", i);
349 			err = ret;
350 		} else {
351 			failure = false;
352 		}
353 	}
354 	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
355 	fs_unlock(dev, 0);
356 	/* Return 0 in case of at least one successful queue stop */
357 	return (failure) ? err : 0;
358 }
359 
360 static int
361 fs_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
362 {
363 	struct sub_device *sdev;
364 	uint8_t i;
365 	int ret;
366 
367 	ret = fs_lock(dev, 0);
368 	if (ret != 0)
369 		return ret;
370 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
371 		uint16_t port_id = ETH(sdev)->data->port_id;
372 
373 		ret = rte_eth_dev_tx_queue_start(port_id, tx_queue_id);
374 		ret = fs_err(sdev, ret);
375 		if (ret) {
376 			ERROR("Tx queue start failed for subdevice %d", i);
377 			fs_tx_queue_stop(dev, tx_queue_id);
378 			fs_unlock(dev, 0);
379 			return ret;
380 		}
381 	}
382 	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
383 	fs_unlock(dev, 0);
384 	return 0;
385 }
386 
387 static void
388 fs_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
389 {
390 	struct sub_device *sdev;
391 	uint8_t i;
392 	struct rxq *rxq = dev->data->rx_queues[qid];
393 
394 	if (rxq == NULL)
395 		return;
396 	if (fs_lock(dev, 0) != 0)
397 		return;
398 	if (rxq->event_fd >= 0)
399 		close(rxq->event_fd);
400 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
401 		if (ETH(sdev)->data->rx_queues != NULL &&
402 		    ETH(sdev)->data->rx_queues[rxq->qid] != NULL)
403 			SUBOPS(sdev, rx_queue_release)(ETH(sdev), rxq->qid);
404 	}
405 	dev->data->rx_queues[rxq->qid] = NULL;
406 	rte_free(rxq);
407 	fs_unlock(dev, 0);
408 }
409 
410 static int
411 fs_rx_queue_setup(struct rte_eth_dev *dev,
412 		uint16_t rx_queue_id,
413 		uint16_t nb_rx_desc,
414 		unsigned int socket_id,
415 		const struct rte_eth_rxconf *rx_conf,
416 		struct rte_mempool *mb_pool)
417 {
418 	struct sub_device *sdev;
419 	struct rxq *rxq;
420 	uint8_t i;
421 	int ret;
422 
423 	ret = fs_lock(dev, 0);
424 	if (ret != 0)
425 		return ret;
426 	if (rx_conf->rx_deferred_start) {
427 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
428 			if (SUBOPS(sdev, rx_queue_start) == NULL) {
429 				ERROR("Rx queue deferred start is not "
430 					"supported for subdevice %d", i);
431 				fs_unlock(dev, 0);
432 				return -EINVAL;
433 			}
434 		}
435 	}
436 	rxq = dev->data->rx_queues[rx_queue_id];
437 	if (rxq != NULL) {
438 		fs_rx_queue_release(dev, rx_queue_id);
439 		dev->data->rx_queues[rx_queue_id] = NULL;
440 	}
441 	rxq = rte_zmalloc(NULL,
442 			  sizeof(*rxq) +
443 			  sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
444 			  RTE_CACHE_LINE_SIZE);
445 	if (rxq == NULL) {
446 		fs_unlock(dev, 0);
447 		return -ENOMEM;
448 	}
449 	FOREACH_SUBDEV(sdev, i, dev)
450 		rte_atomic64_init(&rxq->refcnt[i]);
451 	rxq->qid = rx_queue_id;
452 	rxq->socket_id = socket_id;
453 	rxq->info.mp = mb_pool;
454 	rxq->info.conf = *rx_conf;
455 	rxq->info.nb_desc = nb_rx_desc;
456 	rxq->priv = PRIV(dev);
457 	rxq->sdev = PRIV(dev)->subs;
458 #ifdef RTE_EXEC_ENV_LINUX
459 	rxq->event_fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
460 	if (rxq->event_fd < 0) {
461 		ERROR("Failed to create an eventfd: %s", strerror(errno));
462 		fs_unlock(dev, 0);
463 		return -errno;
464 	}
465 #else
466 	rxq->event_fd = -1;
467 #endif
468 	dev->data->rx_queues[rx_queue_id] = rxq;
469 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
470 		ret = rte_eth_rx_queue_setup(PORT_ID(sdev),
471 				rx_queue_id,
472 				nb_rx_desc, socket_id,
473 				rx_conf, mb_pool);
474 		if ((ret = fs_err(sdev, ret))) {
475 			ERROR("RX queue setup failed for sub_device %d", i);
476 			goto free_rxq;
477 		}
478 	}
479 	fs_unlock(dev, 0);
480 	return 0;
481 free_rxq:
482 	fs_rx_queue_release(dev, rx_queue_id);
483 	fs_unlock(dev, 0);
484 	return ret;
485 }
486 
487 static int
488 fs_rx_intr_enable(struct rte_eth_dev *dev, uint16_t idx)
489 {
490 	struct rxq *rxq;
491 	struct sub_device *sdev;
492 	uint8_t i;
493 	int ret;
494 	int rc = 0;
495 
496 	ret = fs_lock(dev, 0);
497 	if (ret != 0)
498 		return ret;
499 	if (idx >= dev->data->nb_rx_queues) {
500 		rc = -EINVAL;
501 		goto unlock;
502 	}
503 	rxq = dev->data->rx_queues[idx];
504 	if (rxq == NULL || rxq->event_fd <= 0) {
505 		rc = -EINVAL;
506 		goto unlock;
507 	}
508 	/* Fail if proxy service is nor running. */
509 	if (PRIV(dev)->rxp.sstate != SS_RUNNING) {
510 		ERROR("failsafe interrupt services are not running");
511 		rc = -EAGAIN;
512 		goto unlock;
513 	}
514 	rxq->enable_events = 1;
515 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
516 		ret = rte_eth_dev_rx_intr_enable(PORT_ID(sdev), idx);
517 		ret = fs_err(sdev, ret);
518 		if (ret)
519 			rc = ret;
520 	}
521 unlock:
522 	fs_unlock(dev, 0);
523 	if (rc)
524 		rte_errno = -rc;
525 	return rc;
526 }
527 
528 static int
529 fs_rx_intr_disable(struct rte_eth_dev *dev, uint16_t idx)
530 {
531 	struct rxq *rxq;
532 	struct sub_device *sdev;
533 	uint64_t u64;
534 	uint8_t i;
535 	int rc = 0;
536 	int ret;
537 
538 	ret = fs_lock(dev, 0);
539 	if (ret != 0)
540 		return ret;
541 	if (idx >= dev->data->nb_rx_queues) {
542 		rc = -EINVAL;
543 		goto unlock;
544 	}
545 	rxq = dev->data->rx_queues[idx];
546 	if (rxq == NULL || rxq->event_fd <= 0) {
547 		rc = -EINVAL;
548 		goto unlock;
549 	}
550 	rxq->enable_events = 0;
551 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
552 		ret = rte_eth_dev_rx_intr_disable(PORT_ID(sdev), idx);
553 		ret = fs_err(sdev, ret);
554 		if (ret)
555 			rc = ret;
556 	}
557 	/* Clear pending events */
558 	while (read(rxq->event_fd, &u64, sizeof(uint64_t)) >  0)
559 		;
560 unlock:
561 	fs_unlock(dev, 0);
562 	if (rc)
563 		rte_errno = -rc;
564 	return rc;
565 }
566 
567 static void
568 fs_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
569 {
570 	struct sub_device *sdev;
571 	uint8_t i;
572 	struct txq *txq = dev->data->tx_queues[qid];
573 
574 	if (txq == NULL)
575 		return;
576 	if (fs_lock(dev, 0) != 0)
577 		return;
578 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
579 		if (ETH(sdev)->data->tx_queues != NULL &&
580 		    ETH(sdev)->data->tx_queues[txq->qid] != NULL)
581 			SUBOPS(sdev, tx_queue_release)(ETH(sdev), txq->qid);
582 	}
583 	dev->data->tx_queues[txq->qid] = NULL;
584 	rte_free(txq);
585 	fs_unlock(dev, 0);
586 }
587 
588 static int
589 fs_tx_queue_setup(struct rte_eth_dev *dev,
590 		uint16_t tx_queue_id,
591 		uint16_t nb_tx_desc,
592 		unsigned int socket_id,
593 		const struct rte_eth_txconf *tx_conf)
594 {
595 	struct sub_device *sdev;
596 	struct txq *txq;
597 	uint8_t i;
598 	int ret;
599 
600 	ret = fs_lock(dev, 0);
601 	if (ret != 0)
602 		return ret;
603 	if (tx_conf->tx_deferred_start) {
604 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
605 			if (SUBOPS(sdev, tx_queue_start) == NULL) {
606 				ERROR("Tx queue deferred start is not "
607 					"supported for subdevice %d", i);
608 				fs_unlock(dev, 0);
609 				return -EINVAL;
610 			}
611 		}
612 	}
613 	txq = dev->data->tx_queues[tx_queue_id];
614 	if (txq != NULL) {
615 		fs_tx_queue_release(dev, tx_queue_id);
616 		dev->data->tx_queues[tx_queue_id] = NULL;
617 	}
618 	txq = rte_zmalloc("ethdev TX queue",
619 			  sizeof(*txq) +
620 			  sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
621 			  RTE_CACHE_LINE_SIZE);
622 	if (txq == NULL) {
623 		fs_unlock(dev, 0);
624 		return -ENOMEM;
625 	}
626 	FOREACH_SUBDEV(sdev, i, dev)
627 		rte_atomic64_init(&txq->refcnt[i]);
628 	txq->qid = tx_queue_id;
629 	txq->socket_id = socket_id;
630 	txq->info.conf = *tx_conf;
631 	txq->info.nb_desc = nb_tx_desc;
632 	txq->priv = PRIV(dev);
633 	dev->data->tx_queues[tx_queue_id] = txq;
634 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
635 		ret = rte_eth_tx_queue_setup(PORT_ID(sdev),
636 				tx_queue_id,
637 				nb_tx_desc, socket_id,
638 				tx_conf);
639 		if ((ret = fs_err(sdev, ret))) {
640 			ERROR("TX queue setup failed for sub_device %d", i);
641 			goto free_txq;
642 		}
643 	}
644 	fs_unlock(dev, 0);
645 	return 0;
646 free_txq:
647 	fs_tx_queue_release(dev, tx_queue_id);
648 	fs_unlock(dev, 0);
649 	return ret;
650 }
651 
652 static void
653 fs_dev_free_queues(struct rte_eth_dev *dev)
654 {
655 	uint16_t i;
656 
657 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
658 		fs_rx_queue_release(dev, i);
659 		dev->data->rx_queues[i] = NULL;
660 	}
661 	dev->data->nb_rx_queues = 0;
662 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
663 		fs_tx_queue_release(dev, i);
664 		dev->data->tx_queues[i] = NULL;
665 	}
666 	dev->data->nb_tx_queues = 0;
667 }
668 
669 int
670 failsafe_eth_dev_close(struct rte_eth_dev *dev)
671 {
672 	struct sub_device *sdev;
673 	uint8_t i;
674 	int err, ret = 0;
675 
676 	ret = fs_lock(dev, 0);
677 	if (ret != 0)
678 		return ret;
679 	failsafe_hotplug_alarm_cancel(dev);
680 	if (PRIV(dev)->state == DEV_STARTED) {
681 		ret = dev->dev_ops->dev_stop(dev);
682 		if (ret != 0) {
683 			fs_unlock(dev, 0);
684 			return ret;
685 		}
686 	}
687 	PRIV(dev)->state = DEV_ACTIVE - 1;
688 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
689 		DEBUG("Closing sub_device %d", i);
690 		failsafe_eth_dev_unregister_callbacks(sdev);
691 		err = rte_eth_dev_close(PORT_ID(sdev));
692 		if (err) {
693 			ret = ret ? ret : err;
694 			ERROR("Error while closing sub-device %u",
695 					PORT_ID(sdev));
696 		}
697 		sdev->state = DEV_ACTIVE - 1;
698 	}
699 	rte_eth_dev_callback_unregister(RTE_ETH_ALL, RTE_ETH_EVENT_NEW,
700 					failsafe_eth_new_event_callback, dev);
701 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
702 		fs_unlock(dev, 0);
703 		return ret;
704 	}
705 	fs_dev_free_queues(dev);
706 	err = failsafe_eal_uninit(dev);
707 	if (err) {
708 		ret = ret ? ret : err;
709 		ERROR("Error while uninitializing sub-EAL");
710 	}
711 	failsafe_args_free(dev);
712 	rte_free(PRIV(dev)->subs);
713 	rte_free(PRIV(dev)->mcast_addrs);
714 	/* mac_addrs must not be freed alone because part of dev_private */
715 	dev->data->mac_addrs = NULL;
716 	fs_unlock(dev, 0);
717 	err = pthread_mutex_destroy(&PRIV(dev)->hotplug_mutex);
718 	if (err) {
719 		ret = ret ? ret : err;
720 		ERROR("Error while destroying hotplug mutex");
721 	}
722 	return ret;
723 }
724 
725 static int
726 fs_promiscuous_enable(struct rte_eth_dev *dev)
727 {
728 	struct sub_device *sdev;
729 	uint8_t i;
730 	int ret = 0;
731 
732 	ret = fs_lock(dev, 0);
733 	if (ret != 0)
734 		return ret;
735 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
736 		ret = rte_eth_promiscuous_enable(PORT_ID(sdev));
737 		ret = fs_err(sdev, ret);
738 		if (ret != 0) {
739 			ERROR("Promiscuous mode enable failed for subdevice %d",
740 				PORT_ID(sdev));
741 			break;
742 		}
743 	}
744 	if (ret != 0) {
745 		/* Rollback in the case of failure */
746 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
747 			ret = rte_eth_promiscuous_disable(PORT_ID(sdev));
748 			ret = fs_err(sdev, ret);
749 			if (ret != 0)
750 				ERROR("Promiscuous mode disable during rollback failed for subdevice %d",
751 					PORT_ID(sdev));
752 		}
753 	}
754 	fs_unlock(dev, 0);
755 
756 	return ret;
757 }
758 
759 static int
760 fs_promiscuous_disable(struct rte_eth_dev *dev)
761 {
762 	struct sub_device *sdev;
763 	uint8_t i;
764 	int ret = 0;
765 
766 	ret = fs_lock(dev, 0);
767 	if (ret != 0)
768 		return ret;
769 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
770 		ret = rte_eth_promiscuous_disable(PORT_ID(sdev));
771 		ret = fs_err(sdev, ret);
772 		if (ret != 0) {
773 			ERROR("Promiscuous mode disable failed for subdevice %d",
774 				PORT_ID(sdev));
775 			break;
776 		}
777 	}
778 	if (ret != 0) {
779 		/* Rollback in the case of failure */
780 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
781 			ret = rte_eth_promiscuous_enable(PORT_ID(sdev));
782 			ret = fs_err(sdev, ret);
783 			if (ret != 0)
784 				ERROR("Promiscuous mode enable during rollback failed for subdevice %d",
785 					PORT_ID(sdev));
786 		}
787 	}
788 	fs_unlock(dev, 0);
789 
790 	return ret;
791 }
792 
793 static int
794 fs_allmulticast_enable(struct rte_eth_dev *dev)
795 {
796 	struct sub_device *sdev;
797 	uint8_t i;
798 	int ret = 0;
799 
800 	ret = fs_lock(dev, 0);
801 	if (ret != 0)
802 		return ret;
803 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
804 		ret = rte_eth_allmulticast_enable(PORT_ID(sdev));
805 		ret = fs_err(sdev, ret);
806 		if (ret != 0) {
807 			ERROR("All-multicast mode enable 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_disable(PORT_ID(sdev));
816 			ret = fs_err(sdev, ret);
817 			if (ret != 0)
818 				ERROR("All-multicast mode disable 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_allmulticast_disable(struct rte_eth_dev *dev)
829 {
830 	struct sub_device *sdev;
831 	uint8_t i;
832 	int ret = 0;
833 
834 	ret = fs_lock(dev, 0);
835 	if (ret != 0)
836 		return ret;
837 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
838 		ret = rte_eth_allmulticast_disable(PORT_ID(sdev));
839 		ret = fs_err(sdev, ret);
840 		if (ret != 0) {
841 			ERROR("All-multicast mode disable failed for subdevice %d",
842 				PORT_ID(sdev));
843 			break;
844 		}
845 	}
846 	if (ret != 0) {
847 		/* Rollback in the case of failure */
848 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
849 			ret = rte_eth_allmulticast_enable(PORT_ID(sdev));
850 			ret = fs_err(sdev, ret);
851 			if (ret != 0)
852 				ERROR("All-multicast mode enable during rollback failed for subdevice %d",
853 					PORT_ID(sdev));
854 		}
855 	}
856 	fs_unlock(dev, 0);
857 
858 	return ret;
859 }
860 
861 static int
862 fs_link_update(struct rte_eth_dev *dev,
863 		int wait_to_complete)
864 {
865 	struct sub_device *sdev;
866 	uint8_t i;
867 	int ret;
868 
869 	ret = fs_lock(dev, 0);
870 	if (ret != 0)
871 		return ret;
872 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
873 		DEBUG("Calling link_update on sub_device %d", i);
874 		ret = (SUBOPS(sdev, link_update))(ETH(sdev), wait_to_complete);
875 		if (ret && ret != -1 && sdev->remove == 0 &&
876 		    rte_eth_dev_is_removed(PORT_ID(sdev)) == 0) {
877 			ERROR("Link update failed for sub_device %d with error %d",
878 			      i, ret);
879 			fs_unlock(dev, 0);
880 			return ret;
881 		}
882 	}
883 	if (TX_SUBDEV(dev)) {
884 		struct rte_eth_link *l1;
885 		struct rte_eth_link *l2;
886 
887 		l1 = &dev->data->dev_link;
888 		l2 = &ETH(TX_SUBDEV(dev))->data->dev_link;
889 		if (memcmp(l1, l2, sizeof(*l1))) {
890 			*l1 = *l2;
891 			fs_unlock(dev, 0);
892 			return 0;
893 		}
894 	}
895 	fs_unlock(dev, 0);
896 	return -1;
897 }
898 
899 static int
900 fs_stats_get(struct rte_eth_dev *dev,
901 	     struct rte_eth_stats *stats)
902 {
903 	struct rte_eth_stats backup;
904 	struct sub_device *sdev;
905 	uint8_t i;
906 	int ret;
907 
908 	ret = fs_lock(dev, 0);
909 	if (ret != 0)
910 		return ret;
911 	rte_memcpy(stats, &PRIV(dev)->stats_accumulator, sizeof(*stats));
912 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
913 		struct rte_eth_stats *snapshot = &sdev->stats_snapshot.stats;
914 		uint64_t *timestamp = &sdev->stats_snapshot.timestamp;
915 
916 		rte_memcpy(&backup, snapshot, sizeof(backup));
917 		ret = rte_eth_stats_get(PORT_ID(sdev), snapshot);
918 		if (ret) {
919 			if (!fs_err(sdev, ret)) {
920 				rte_memcpy(snapshot, &backup, sizeof(backup));
921 				goto inc;
922 			}
923 			ERROR("Operation rte_eth_stats_get failed for sub_device %d with error %d",
924 				  i, ret);
925 			*timestamp = 0;
926 			fs_unlock(dev, 0);
927 			return ret;
928 		}
929 		*timestamp = rte_rdtsc();
930 inc:
931 		failsafe_stats_increment(stats, snapshot);
932 	}
933 	fs_unlock(dev, 0);
934 	return 0;
935 }
936 
937 static int
938 fs_stats_reset(struct rte_eth_dev *dev)
939 {
940 	struct sub_device *sdev;
941 	uint8_t i;
942 	int ret;
943 
944 	ret = fs_lock(dev, 0);
945 	if (ret != 0)
946 		return ret;
947 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
948 		ret = rte_eth_stats_reset(PORT_ID(sdev));
949 		if (ret) {
950 			if (!fs_err(sdev, ret))
951 				continue;
952 
953 			ERROR("Operation rte_eth_stats_reset failed for sub_device %d with error %d",
954 			      i, ret);
955 			fs_unlock(dev, 0);
956 			return ret;
957 		}
958 		memset(&sdev->stats_snapshot, 0, sizeof(struct rte_eth_stats));
959 	}
960 	memset(&PRIV(dev)->stats_accumulator, 0, sizeof(struct rte_eth_stats));
961 	fs_unlock(dev, 0);
962 
963 	return 0;
964 }
965 
966 static int
967 __fs_xstats_count(struct rte_eth_dev *dev)
968 {
969 	struct sub_device *sdev;
970 	int count = 0;
971 	uint8_t i;
972 	int ret;
973 
974 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
975 		ret = rte_eth_xstats_get_names(PORT_ID(sdev), NULL, 0);
976 		if (ret < 0)
977 			return ret;
978 		count += ret;
979 	}
980 
981 	return count;
982 }
983 
984 static int
985 __fs_xstats_get_names(struct rte_eth_dev *dev,
986 		    struct rte_eth_xstat_name *xstats_names,
987 		    unsigned int limit)
988 {
989 	struct sub_device *sdev;
990 	unsigned int count = 0;
991 	uint8_t i;
992 
993 	/* Caller only cares about count */
994 	if (!xstats_names)
995 		return  __fs_xstats_count(dev);
996 
997 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
998 		struct rte_eth_xstat_name *sub_names = xstats_names + count;
999 		int j, r;
1000 
1001 		if (count >= limit)
1002 			break;
1003 
1004 		r = rte_eth_xstats_get_names(PORT_ID(sdev),
1005 					     sub_names, limit - count);
1006 		if (r < 0)
1007 			return r;
1008 
1009 		/* add subN_ prefix to names */
1010 		for (j = 0; j < r; j++) {
1011 			char *xname = sub_names[j].name;
1012 			char tmp[RTE_ETH_XSTATS_NAME_SIZE];
1013 
1014 			if ((xname[0] == 't' || xname[0] == 'r') &&
1015 			    xname[1] == 'x' && xname[2] == '_')
1016 				snprintf(tmp, sizeof(tmp), "%.3ssub%u_%s",
1017 					 xname, i, xname + 3);
1018 			else
1019 				snprintf(tmp, sizeof(tmp), "sub%u_%s",
1020 					 i, xname);
1021 
1022 			strlcpy(xname, tmp, RTE_ETH_XSTATS_NAME_SIZE);
1023 		}
1024 		count += r;
1025 	}
1026 	return count;
1027 }
1028 
1029 static int
1030 fs_xstats_get_names(struct rte_eth_dev *dev,
1031 		    struct rte_eth_xstat_name *xstats_names,
1032 		    unsigned int limit)
1033 {
1034 	int ret;
1035 
1036 	ret = fs_lock(dev, 0);
1037 	if (ret != 0)
1038 		return ret;
1039 	ret = __fs_xstats_get_names(dev, xstats_names, limit);
1040 	fs_unlock(dev, 0);
1041 	return ret;
1042 }
1043 
1044 static int
1045 __fs_xstats_get(struct rte_eth_dev *dev,
1046 	      struct rte_eth_xstat *xstats,
1047 	      unsigned int n)
1048 {
1049 	unsigned int count = 0;
1050 	struct sub_device *sdev;
1051 	uint8_t i;
1052 	int j, ret;
1053 
1054 	ret = __fs_xstats_count(dev);
1055 	/*
1056 	 * if error
1057 	 * or caller did not give enough space
1058 	 * or just querying
1059 	 */
1060 	if (ret < 0 || ret > (int)n || xstats == NULL)
1061 		return ret;
1062 
1063 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1064 		ret = rte_eth_xstats_get(PORT_ID(sdev), xstats, n);
1065 		if (ret < 0)
1066 			return ret;
1067 
1068 		if (ret > (int)n)
1069 			return n + count;
1070 
1071 		/* add offset to id's from sub-device */
1072 		for (j = 0; j < ret; j++)
1073 			xstats[j].id += count;
1074 
1075 		xstats += ret;
1076 		n -= ret;
1077 		count += ret;
1078 	}
1079 
1080 	return count;
1081 }
1082 
1083 static int
1084 fs_xstats_get(struct rte_eth_dev *dev,
1085 	      struct rte_eth_xstat *xstats,
1086 	      unsigned int n)
1087 {
1088 	int ret;
1089 
1090 	ret = fs_lock(dev, 0);
1091 	if (ret != 0)
1092 		return ret;
1093 	ret = __fs_xstats_get(dev, xstats, n);
1094 	fs_unlock(dev, 0);
1095 
1096 	return ret;
1097 }
1098 
1099 
1100 static int
1101 fs_xstats_reset(struct rte_eth_dev *dev)
1102 {
1103 	struct sub_device *sdev;
1104 	uint8_t i;
1105 	int r;
1106 
1107 	r = fs_lock(dev, 0);
1108 	if (r != 0)
1109 		return r;
1110 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1111 		r = rte_eth_xstats_reset(PORT_ID(sdev));
1112 		if (r < 0)
1113 			break;
1114 	}
1115 	fs_unlock(dev, 0);
1116 
1117 	return r;
1118 }
1119 
1120 static void
1121 fs_dev_merge_desc_lim(struct rte_eth_desc_lim *to,
1122 		      const struct rte_eth_desc_lim *from)
1123 {
1124 	to->nb_max = RTE_MIN(to->nb_max, from->nb_max);
1125 	to->nb_min = RTE_MAX(to->nb_min, from->nb_min);
1126 	to->nb_align = RTE_MAX(to->nb_align, from->nb_align);
1127 
1128 	to->nb_seg_max = RTE_MIN(to->nb_seg_max, from->nb_seg_max);
1129 	to->nb_mtu_seg_max = RTE_MIN(to->nb_mtu_seg_max, from->nb_mtu_seg_max);
1130 }
1131 
1132 /*
1133  * Merge the information from sub-devices.
1134  *
1135  * The reported values must be the common subset of all sub devices
1136  */
1137 static void
1138 fs_dev_merge_info(struct rte_eth_dev_info *info,
1139 		  const struct rte_eth_dev_info *sinfo)
1140 {
1141 	info->min_mtu = RTE_MAX(info->min_mtu, sinfo->min_mtu);
1142 	info->max_mtu = RTE_MIN(info->max_mtu, sinfo->max_mtu);
1143 	info->max_rx_pktlen = RTE_MIN(info->max_rx_pktlen, sinfo->max_rx_pktlen);
1144 	info->max_rx_queues = RTE_MIN(info->max_rx_queues, sinfo->max_rx_queues);
1145 	info->max_tx_queues = RTE_MIN(info->max_tx_queues, sinfo->max_tx_queues);
1146 	info->max_mac_addrs = RTE_MIN(info->max_mac_addrs, sinfo->max_mac_addrs);
1147 	info->max_hash_mac_addrs = RTE_MIN(info->max_hash_mac_addrs,
1148 					sinfo->max_hash_mac_addrs);
1149 	info->max_vmdq_pools = RTE_MIN(info->max_vmdq_pools, sinfo->max_vmdq_pools);
1150 	info->max_vfs = RTE_MIN(info->max_vfs, sinfo->max_vfs);
1151 
1152 	fs_dev_merge_desc_lim(&info->rx_desc_lim, &sinfo->rx_desc_lim);
1153 	fs_dev_merge_desc_lim(&info->tx_desc_lim, &sinfo->tx_desc_lim);
1154 
1155 	info->rx_offload_capa &= sinfo->rx_offload_capa;
1156 	info->tx_offload_capa &= sinfo->tx_offload_capa;
1157 	info->rx_queue_offload_capa &= sinfo->rx_queue_offload_capa;
1158 	info->tx_queue_offload_capa &= sinfo->tx_queue_offload_capa;
1159 	info->flow_type_rss_offloads &= sinfo->flow_type_rss_offloads;
1160 
1161 	/*
1162 	 * RETA size is a GCD of RETA sizes indicated by sub-devices.
1163 	 * Each of these sizes is a power of 2, so use the lower one.
1164 	 */
1165 	info->reta_size = RTE_MIN(info->reta_size, sinfo->reta_size);
1166 
1167 	info->hash_key_size = RTE_MIN(info->hash_key_size,
1168 				      sinfo->hash_key_size);
1169 }
1170 
1171 /**
1172  * Fail-safe dev_infos_get rules:
1173  *
1174  * No sub_device:
1175  *   Numerables:
1176  *      Use the maximum possible values for any field, so as not
1177  *      to impede any further configuration effort.
1178  *   Capabilities:
1179  *      Limits capabilities to those that are understood by the
1180  *      fail-safe PMD. This understanding stems from the fail-safe
1181  *      being capable of verifying that the related capability is
1182  *      expressed within the device configuration (struct rte_eth_conf).
1183  *
1184  * At least one probed sub_device:
1185  *   Numerables:
1186  *      Uses values from the active probed sub_device
1187  *      The rationale here is that if any sub_device is less capable
1188  *      (for example concerning the number of queues) than the active
1189  *      sub_device, then its subsequent configuration will fail.
1190  *      It is impossible to foresee this failure when the failing sub_device
1191  *      is supposed to be plugged-in later on, so the configuration process
1192  *      is the single point of failure and error reporting.
1193  *   Capabilities:
1194  *      Uses a logical AND of RX capabilities among
1195  *      all sub_devices and the default capabilities.
1196  *      Uses a logical AND of TX capabilities among
1197  *      the active probed sub_device and the default capabilities.
1198  *      Uses a logical AND of device capabilities among
1199  *      all sub_devices and the default capabilities.
1200  *
1201  */
1202 static int
1203 fs_dev_infos_get(struct rte_eth_dev *dev,
1204 		  struct rte_eth_dev_info *infos)
1205 {
1206 	struct sub_device *sdev;
1207 	uint8_t i;
1208 	int ret;
1209 
1210 	/* Use maximum upper bounds by default */
1211 	infos->min_mtu = RTE_ETHER_MIN_MTU;
1212 	infos->max_mtu = UINT16_MAX;
1213 	infos->max_rx_pktlen = UINT32_MAX;
1214 	infos->max_rx_queues = RTE_MAX_QUEUES_PER_PORT;
1215 	infos->max_tx_queues = RTE_MAX_QUEUES_PER_PORT;
1216 	infos->max_mac_addrs = FAILSAFE_MAX_ETHADDR;
1217 	infos->max_hash_mac_addrs = UINT32_MAX;
1218 	infos->max_vfs = UINT16_MAX;
1219 	infos->max_vmdq_pools = UINT16_MAX;
1220 	infos->reta_size = UINT16_MAX;
1221 	infos->hash_key_size = UINT8_MAX;
1222 
1223 	/*
1224 	 * Set of capabilities that can be verified upon
1225 	 * configuring a sub-device.
1226 	 */
1227 	infos->rx_offload_capa =
1228 		RTE_ETH_RX_OFFLOAD_VLAN_STRIP |
1229 		RTE_ETH_RX_OFFLOAD_IPV4_CKSUM |
1230 		RTE_ETH_RX_OFFLOAD_UDP_CKSUM |
1231 		RTE_ETH_RX_OFFLOAD_TCP_CKSUM |
1232 		RTE_ETH_RX_OFFLOAD_TCP_LRO |
1233 		RTE_ETH_RX_OFFLOAD_QINQ_STRIP |
1234 		RTE_ETH_RX_OFFLOAD_OUTER_IPV4_CKSUM |
1235 		RTE_ETH_RX_OFFLOAD_MACSEC_STRIP |
1236 		RTE_ETH_RX_OFFLOAD_VLAN_FILTER |
1237 		RTE_ETH_RX_OFFLOAD_VLAN_EXTEND |
1238 		RTE_ETH_RX_OFFLOAD_SCATTER |
1239 		RTE_ETH_RX_OFFLOAD_TIMESTAMP |
1240 		RTE_ETH_RX_OFFLOAD_SECURITY |
1241 		RTE_ETH_RX_OFFLOAD_RSS_HASH;
1242 
1243 	infos->rx_queue_offload_capa =
1244 		RTE_ETH_RX_OFFLOAD_VLAN_STRIP |
1245 		RTE_ETH_RX_OFFLOAD_IPV4_CKSUM |
1246 		RTE_ETH_RX_OFFLOAD_UDP_CKSUM |
1247 		RTE_ETH_RX_OFFLOAD_TCP_CKSUM |
1248 		RTE_ETH_RX_OFFLOAD_TCP_LRO |
1249 		RTE_ETH_RX_OFFLOAD_QINQ_STRIP |
1250 		RTE_ETH_RX_OFFLOAD_OUTER_IPV4_CKSUM |
1251 		RTE_ETH_RX_OFFLOAD_MACSEC_STRIP |
1252 		RTE_ETH_RX_OFFLOAD_VLAN_FILTER |
1253 		RTE_ETH_RX_OFFLOAD_VLAN_EXTEND |
1254 		RTE_ETH_RX_OFFLOAD_SCATTER |
1255 		RTE_ETH_RX_OFFLOAD_TIMESTAMP |
1256 		RTE_ETH_RX_OFFLOAD_SECURITY |
1257 		RTE_ETH_RX_OFFLOAD_RSS_HASH;
1258 
1259 	infos->tx_offload_capa =
1260 		RTE_ETH_TX_OFFLOAD_MULTI_SEGS |
1261 		RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE |
1262 		RTE_ETH_TX_OFFLOAD_IPV4_CKSUM |
1263 		RTE_ETH_TX_OFFLOAD_UDP_CKSUM |
1264 		RTE_ETH_TX_OFFLOAD_TCP_CKSUM |
1265 		RTE_ETH_TX_OFFLOAD_TCP_TSO;
1266 
1267 	infos->flow_type_rss_offloads =
1268 		RTE_ETH_RSS_IP |
1269 		RTE_ETH_RSS_UDP |
1270 		RTE_ETH_RSS_TCP;
1271 	infos->dev_capa =
1272 		RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
1273 		RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
1274 	infos->dev_capa &= ~RTE_ETH_DEV_CAPA_FLOW_RULE_KEEP;
1275 
1276 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
1277 		struct rte_eth_dev_info sub_info;
1278 
1279 		ret = rte_eth_dev_info_get(PORT_ID(sdev), &sub_info);
1280 		ret = fs_err(sdev, ret);
1281 		if (ret != 0)
1282 			return ret;
1283 
1284 		fs_dev_merge_info(infos, &sub_info);
1285 	}
1286 
1287 	return 0;
1288 }
1289 
1290 static const uint32_t *
1291 fs_dev_supported_ptypes_get(struct rte_eth_dev *dev, size_t *no_of_elements)
1292 {
1293 	struct sub_device *sdev;
1294 	struct rte_eth_dev *edev;
1295 	const uint32_t *ret;
1296 
1297 	if (fs_lock(dev, 0) != 0)
1298 		return NULL;
1299 	sdev = TX_SUBDEV(dev);
1300 	if (sdev == NULL) {
1301 		ret = NULL;
1302 		goto unlock;
1303 	}
1304 	edev = ETH(sdev);
1305 	/* ENOTSUP: counts as no supported ptypes */
1306 	if (SUBOPS(sdev, dev_supported_ptypes_get) == NULL) {
1307 		ret = NULL;
1308 		goto unlock;
1309 	}
1310 	/*
1311 	 * The API does not permit to do a clean AND of all ptypes,
1312 	 * It is also incomplete by design and we do not really care
1313 	 * to have a best possible value in this context.
1314 	 * We just return the ptypes of the device of highest
1315 	 * priority, usually the PREFERRED device.
1316 	 */
1317 	ret = SUBOPS(sdev, dev_supported_ptypes_get)(edev, no_of_elements);
1318 unlock:
1319 	fs_unlock(dev, 0);
1320 	return ret;
1321 }
1322 
1323 static int
1324 fs_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1325 {
1326 	struct sub_device *sdev;
1327 	uint8_t i;
1328 	int ret;
1329 
1330 	ret = fs_lock(dev, 0);
1331 	if (ret != 0)
1332 		return ret;
1333 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1334 		DEBUG("Calling rte_eth_dev_set_mtu on sub_device %d", i);
1335 		ret = rte_eth_dev_set_mtu(PORT_ID(sdev), mtu);
1336 		if ((ret = fs_err(sdev, ret))) {
1337 			ERROR("Operation rte_eth_dev_set_mtu failed for sub_device %d with error %d",
1338 			      i, ret);
1339 			fs_unlock(dev, 0);
1340 			return ret;
1341 		}
1342 	}
1343 	fs_unlock(dev, 0);
1344 	return 0;
1345 }
1346 
1347 static int
1348 fs_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1349 {
1350 	struct sub_device *sdev;
1351 	uint8_t i;
1352 	int ret;
1353 
1354 	ret = fs_lock(dev, 0);
1355 	if (ret != 0)
1356 		return ret;
1357 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1358 		DEBUG("Calling rte_eth_dev_vlan_filter on sub_device %d", i);
1359 		ret = rte_eth_dev_vlan_filter(PORT_ID(sdev), vlan_id, on);
1360 		if ((ret = fs_err(sdev, ret))) {
1361 			ERROR("Operation rte_eth_dev_vlan_filter failed for sub_device %d"
1362 			      " with error %d", i, ret);
1363 			fs_unlock(dev, 0);
1364 			return ret;
1365 		}
1366 	}
1367 	fs_unlock(dev, 0);
1368 	return 0;
1369 }
1370 
1371 static int
1372 fs_flow_ctrl_get(struct rte_eth_dev *dev,
1373 		struct rte_eth_fc_conf *fc_conf)
1374 {
1375 	struct sub_device *sdev;
1376 	int ret;
1377 
1378 	ret = fs_lock(dev, 0);
1379 	if (ret != 0)
1380 		return ret;
1381 	sdev = TX_SUBDEV(dev);
1382 	if (sdev == NULL) {
1383 		ret = 0;
1384 		goto unlock;
1385 	}
1386 	if (SUBOPS(sdev, flow_ctrl_get) == NULL) {
1387 		ret = -ENOTSUP;
1388 		goto unlock;
1389 	}
1390 	ret = SUBOPS(sdev, flow_ctrl_get)(ETH(sdev), fc_conf);
1391 unlock:
1392 	fs_unlock(dev, 0);
1393 	return ret;
1394 }
1395 
1396 static int
1397 fs_flow_ctrl_set(struct rte_eth_dev *dev,
1398 		struct rte_eth_fc_conf *fc_conf)
1399 {
1400 	struct sub_device *sdev;
1401 	uint8_t i;
1402 	int ret;
1403 
1404 	ret = fs_lock(dev, 0);
1405 	if (ret != 0)
1406 		return ret;
1407 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1408 		DEBUG("Calling rte_eth_dev_flow_ctrl_set on sub_device %d", i);
1409 		ret = rte_eth_dev_flow_ctrl_set(PORT_ID(sdev), fc_conf);
1410 		if ((ret = fs_err(sdev, ret))) {
1411 			ERROR("Operation rte_eth_dev_flow_ctrl_set failed for sub_device %d"
1412 			      " with error %d", i, ret);
1413 			fs_unlock(dev, 0);
1414 			return ret;
1415 		}
1416 	}
1417 	fs_unlock(dev, 0);
1418 	return 0;
1419 }
1420 
1421 static void
1422 fs_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1423 {
1424 	struct sub_device *sdev;
1425 	uint8_t i;
1426 
1427 	if (fs_lock(dev, 0) != 0)
1428 		return;
1429 	/* No check: already done within the rte_eth_dev_mac_addr_remove
1430 	 * call for the fail-safe device.
1431 	 */
1432 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
1433 		rte_eth_dev_mac_addr_remove(PORT_ID(sdev),
1434 				&dev->data->mac_addrs[index]);
1435 	PRIV(dev)->mac_addr_pool[index] = 0;
1436 	fs_unlock(dev, 0);
1437 }
1438 
1439 static int
1440 fs_mac_addr_add(struct rte_eth_dev *dev,
1441 		struct rte_ether_addr *mac_addr,
1442 		uint32_t index,
1443 		uint32_t vmdq)
1444 {
1445 	struct sub_device *sdev;
1446 	int ret;
1447 	uint8_t i;
1448 
1449 	RTE_ASSERT(index < FAILSAFE_MAX_ETHADDR);
1450 	ret = fs_lock(dev, 0);
1451 	if (ret != 0)
1452 		return ret;
1453 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1454 		ret = rte_eth_dev_mac_addr_add(PORT_ID(sdev), mac_addr, vmdq);
1455 		if ((ret = fs_err(sdev, ret))) {
1456 			ERROR("Operation rte_eth_dev_mac_addr_add failed for sub_device %"
1457 			      PRIu8 " with error %d", i, ret);
1458 			fs_unlock(dev, 0);
1459 			return ret;
1460 		}
1461 	}
1462 	if (index >= PRIV(dev)->nb_mac_addr) {
1463 		DEBUG("Growing mac_addrs array");
1464 		PRIV(dev)->nb_mac_addr = index;
1465 	}
1466 	PRIV(dev)->mac_addr_pool[index] = vmdq;
1467 	fs_unlock(dev, 0);
1468 	return 0;
1469 }
1470 
1471 static int
1472 fs_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1473 {
1474 	struct sub_device *sdev;
1475 	uint8_t i;
1476 	int ret;
1477 
1478 	ret = fs_lock(dev, 0);
1479 	if (ret != 0)
1480 		return ret;
1481 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1482 		ret = rte_eth_dev_default_mac_addr_set(PORT_ID(sdev), mac_addr);
1483 		ret = fs_err(sdev, ret);
1484 		if (ret) {
1485 			ERROR("Operation rte_eth_dev_mac_addr_set failed for sub_device %d with error %d",
1486 				i, ret);
1487 			fs_unlock(dev, 0);
1488 			return ret;
1489 		}
1490 	}
1491 	fs_unlock(dev, 0);
1492 
1493 	return 0;
1494 }
1495 
1496 static int
1497 fs_set_mc_addr_list(struct rte_eth_dev *dev,
1498 		    struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1499 {
1500 	struct sub_device *sdev;
1501 	uint8_t i;
1502 	int ret;
1503 	void *mcast_addrs;
1504 
1505 	ret = fs_lock(dev, 0);
1506 	if (ret != 0)
1507 		return ret;
1508 
1509 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1510 		ret = rte_eth_dev_set_mc_addr_list(PORT_ID(sdev),
1511 						   mc_addr_set, nb_mc_addr);
1512 		if (ret != 0) {
1513 			ERROR("Operation rte_eth_dev_set_mc_addr_list failed for sub_device %d with error %d",
1514 			      i, ret);
1515 			goto rollback;
1516 		}
1517 	}
1518 
1519 	mcast_addrs = rte_realloc(PRIV(dev)->mcast_addrs,
1520 		nb_mc_addr * sizeof(PRIV(dev)->mcast_addrs[0]), 0);
1521 	if (mcast_addrs == NULL && nb_mc_addr > 0) {
1522 		ret = -ENOMEM;
1523 		goto rollback;
1524 	}
1525 	rte_memcpy(mcast_addrs, mc_addr_set,
1526 		   nb_mc_addr * sizeof(PRIV(dev)->mcast_addrs[0]));
1527 	PRIV(dev)->nb_mcast_addr = nb_mc_addr;
1528 	PRIV(dev)->mcast_addrs = mcast_addrs;
1529 
1530 	fs_unlock(dev, 0);
1531 	return 0;
1532 
1533 rollback:
1534 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1535 		int rc = rte_eth_dev_set_mc_addr_list(PORT_ID(sdev),
1536 			PRIV(dev)->mcast_addrs,	PRIV(dev)->nb_mcast_addr);
1537 		if (rc != 0) {
1538 			ERROR("Multicast MAC address list rollback for sub_device %d failed with error %d",
1539 			      i, rc);
1540 		}
1541 	}
1542 
1543 	fs_unlock(dev, 0);
1544 	return ret;
1545 }
1546 
1547 static int
1548 fs_rss_hash_update(struct rte_eth_dev *dev,
1549 			struct rte_eth_rss_conf *rss_conf)
1550 {
1551 	struct sub_device *sdev;
1552 	uint8_t i;
1553 	int ret;
1554 
1555 	ret = fs_lock(dev, 0);
1556 	if (ret != 0)
1557 		return ret;
1558 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1559 		ret = rte_eth_dev_rss_hash_update(PORT_ID(sdev), rss_conf);
1560 		ret = fs_err(sdev, ret);
1561 		if (ret) {
1562 			ERROR("Operation rte_eth_dev_rss_hash_update"
1563 				" failed for sub_device %d with error %d",
1564 				i, ret);
1565 			fs_unlock(dev, 0);
1566 			return ret;
1567 		}
1568 	}
1569 	fs_unlock(dev, 0);
1570 
1571 	return 0;
1572 }
1573 
1574 static int
1575 fs_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
1576 		const struct rte_flow_ops **ops)
1577 {
1578 	*ops = &fs_flow_ops;
1579 	return 0;
1580 }
1581 
1582 const struct eth_dev_ops failsafe_ops = {
1583 	.dev_configure = fs_dev_configure,
1584 	.dev_start = fs_dev_start,
1585 	.dev_stop = fs_dev_stop,
1586 	.dev_set_link_down = fs_dev_set_link_down,
1587 	.dev_set_link_up = fs_dev_set_link_up,
1588 	.dev_close = failsafe_eth_dev_close,
1589 	.promiscuous_enable = fs_promiscuous_enable,
1590 	.promiscuous_disable = fs_promiscuous_disable,
1591 	.allmulticast_enable = fs_allmulticast_enable,
1592 	.allmulticast_disable = fs_allmulticast_disable,
1593 	.link_update = fs_link_update,
1594 	.stats_get = fs_stats_get,
1595 	.stats_reset = fs_stats_reset,
1596 	.xstats_get = fs_xstats_get,
1597 	.xstats_get_names = fs_xstats_get_names,
1598 	.xstats_reset = fs_xstats_reset,
1599 	.dev_infos_get = fs_dev_infos_get,
1600 	.dev_supported_ptypes_get = fs_dev_supported_ptypes_get,
1601 	.mtu_set = fs_mtu_set,
1602 	.vlan_filter_set = fs_vlan_filter_set,
1603 	.rx_queue_start = fs_rx_queue_start,
1604 	.rx_queue_stop = fs_rx_queue_stop,
1605 	.tx_queue_start = fs_tx_queue_start,
1606 	.tx_queue_stop = fs_tx_queue_stop,
1607 	.rx_queue_setup = fs_rx_queue_setup,
1608 	.tx_queue_setup = fs_tx_queue_setup,
1609 	.rx_queue_release = fs_rx_queue_release,
1610 	.tx_queue_release = fs_tx_queue_release,
1611 	.rx_queue_intr_enable = fs_rx_intr_enable,
1612 	.rx_queue_intr_disable = fs_rx_intr_disable,
1613 	.flow_ctrl_get = fs_flow_ctrl_get,
1614 	.flow_ctrl_set = fs_flow_ctrl_set,
1615 	.mac_addr_remove = fs_mac_addr_remove,
1616 	.mac_addr_add = fs_mac_addr_add,
1617 	.mac_addr_set = fs_mac_addr_set,
1618 	.set_mc_addr_list = fs_set_mc_addr_list,
1619 	.rss_hash_update = fs_rss_hash_update,
1620 	.flow_ops_get = fs_flow_ops_get,
1621 };
1622