xref: /dpdk/drivers/net/failsafe/failsafe_ops.c (revision 42c3576d44dd48602aba0820b98c07e2c4278c0e)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright 2017 6WIND S.A.
5  *   Copyright 2017 Mellanox.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of 6WIND S.A. nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <stdbool.h>
35 #include <stdint.h>
36 
37 #include <rte_debug.h>
38 #include <rte_atomic.h>
39 #include <rte_ethdev_driver.h>
40 #include <rte_malloc.h>
41 #include <rte_flow.h>
42 #include <rte_cycles.h>
43 
44 #include "failsafe_private.h"
45 
46 static struct rte_eth_dev_info default_infos = {
47 	/* Max possible number of elements */
48 	.max_rx_pktlen = UINT32_MAX,
49 	.max_rx_queues = RTE_MAX_QUEUES_PER_PORT,
50 	.max_tx_queues = RTE_MAX_QUEUES_PER_PORT,
51 	.max_mac_addrs = FAILSAFE_MAX_ETHADDR,
52 	.max_hash_mac_addrs = UINT32_MAX,
53 	.max_vfs = UINT16_MAX,
54 	.max_vmdq_pools = UINT16_MAX,
55 	.rx_desc_lim = {
56 		.nb_max = UINT16_MAX,
57 		.nb_min = 0,
58 		.nb_align = 1,
59 		.nb_seg_max = UINT16_MAX,
60 		.nb_mtu_seg_max = UINT16_MAX,
61 	},
62 	.tx_desc_lim = {
63 		.nb_max = UINT16_MAX,
64 		.nb_min = 0,
65 		.nb_align = 1,
66 		.nb_seg_max = UINT16_MAX,
67 		.nb_mtu_seg_max = UINT16_MAX,
68 	},
69 	/*
70 	 * Set of capabilities that can be verified upon
71 	 * configuring a sub-device.
72 	 */
73 	.rx_offload_capa =
74 		DEV_RX_OFFLOAD_VLAN_STRIP |
75 		DEV_RX_OFFLOAD_IPV4_CKSUM |
76 		DEV_RX_OFFLOAD_UDP_CKSUM |
77 		DEV_RX_OFFLOAD_TCP_CKSUM |
78 		DEV_RX_OFFLOAD_TCP_LRO |
79 		DEV_RX_OFFLOAD_QINQ_STRIP |
80 		DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM |
81 		DEV_RX_OFFLOAD_MACSEC_STRIP |
82 		DEV_RX_OFFLOAD_HEADER_SPLIT |
83 		DEV_RX_OFFLOAD_VLAN_FILTER |
84 		DEV_RX_OFFLOAD_VLAN_EXTEND |
85 		DEV_RX_OFFLOAD_JUMBO_FRAME |
86 		DEV_RX_OFFLOAD_CRC_STRIP |
87 		DEV_RX_OFFLOAD_SCATTER |
88 		DEV_RX_OFFLOAD_TIMESTAMP |
89 		DEV_RX_OFFLOAD_SECURITY,
90 	.rx_queue_offload_capa =
91 		DEV_RX_OFFLOAD_VLAN_STRIP |
92 		DEV_RX_OFFLOAD_IPV4_CKSUM |
93 		DEV_RX_OFFLOAD_UDP_CKSUM |
94 		DEV_RX_OFFLOAD_TCP_CKSUM |
95 		DEV_RX_OFFLOAD_TCP_LRO |
96 		DEV_RX_OFFLOAD_QINQ_STRIP |
97 		DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM |
98 		DEV_RX_OFFLOAD_MACSEC_STRIP |
99 		DEV_RX_OFFLOAD_HEADER_SPLIT |
100 		DEV_RX_OFFLOAD_VLAN_FILTER |
101 		DEV_RX_OFFLOAD_VLAN_EXTEND |
102 		DEV_RX_OFFLOAD_JUMBO_FRAME |
103 		DEV_RX_OFFLOAD_CRC_STRIP |
104 		DEV_RX_OFFLOAD_SCATTER |
105 		DEV_RX_OFFLOAD_TIMESTAMP |
106 		DEV_RX_OFFLOAD_SECURITY,
107 	.tx_offload_capa = 0x0,
108 	.flow_type_rss_offloads = 0x0,
109 };
110 
111 static int
112 fs_dev_configure(struct rte_eth_dev *dev)
113 {
114 	struct sub_device *sdev;
115 	uint64_t supp_tx_offloads;
116 	uint64_t tx_offloads;
117 	uint8_t i;
118 	int ret;
119 
120 	supp_tx_offloads = PRIV(dev)->infos.tx_offload_capa;
121 	tx_offloads = dev->data->dev_conf.txmode.offloads;
122 	if ((tx_offloads & supp_tx_offloads) != tx_offloads) {
123 		rte_errno = ENOTSUP;
124 		ERROR("Some Tx offloads are not supported, "
125 		      "requested 0x%" PRIx64 " supported 0x%" PRIx64,
126 		      tx_offloads, supp_tx_offloads);
127 		return -rte_errno;
128 	}
129 	FOREACH_SUBDEV(sdev, i, dev) {
130 		int rmv_interrupt = 0;
131 		int lsc_interrupt = 0;
132 		int lsc_enabled;
133 
134 		if (sdev->state != DEV_PROBED)
135 			continue;
136 
137 		rmv_interrupt = ETH(sdev)->data->dev_flags &
138 				RTE_ETH_DEV_INTR_RMV;
139 		if (rmv_interrupt) {
140 			DEBUG("Enabling RMV interrupts for sub_device %d", i);
141 			dev->data->dev_conf.intr_conf.rmv = 1;
142 		} else {
143 			DEBUG("sub_device %d does not support RMV event", i);
144 		}
145 		lsc_enabled = dev->data->dev_conf.intr_conf.lsc;
146 		lsc_interrupt = lsc_enabled &&
147 				(ETH(sdev)->data->dev_flags &
148 				 RTE_ETH_DEV_INTR_LSC);
149 		if (lsc_interrupt) {
150 			DEBUG("Enabling LSC interrupts for sub_device %d", i);
151 			dev->data->dev_conf.intr_conf.lsc = 1;
152 		} else if (lsc_enabled && !lsc_interrupt) {
153 			DEBUG("Disabling LSC interrupts for sub_device %d", i);
154 			dev->data->dev_conf.intr_conf.lsc = 0;
155 		}
156 		DEBUG("Configuring sub-device %d", i);
157 		sdev->remove = 0;
158 		ret = rte_eth_dev_configure(PORT_ID(sdev),
159 					dev->data->nb_rx_queues,
160 					dev->data->nb_tx_queues,
161 					&dev->data->dev_conf);
162 		if (ret) {
163 			if (!fs_err(sdev, ret))
164 				continue;
165 			ERROR("Could not configure sub_device %d", i);
166 			return ret;
167 		}
168 		if (rmv_interrupt) {
169 			ret = rte_eth_dev_callback_register(PORT_ID(sdev),
170 					RTE_ETH_EVENT_INTR_RMV,
171 					failsafe_eth_rmv_event_callback,
172 					sdev);
173 			if (ret)
174 				WARN("Failed to register RMV callback for sub_device %d",
175 				     SUB_ID(sdev));
176 		}
177 		dev->data->dev_conf.intr_conf.rmv = 0;
178 		if (lsc_interrupt) {
179 			ret = rte_eth_dev_callback_register(PORT_ID(sdev),
180 						RTE_ETH_EVENT_INTR_LSC,
181 						failsafe_eth_lsc_event_callback,
182 						dev);
183 			if (ret)
184 				WARN("Failed to register LSC callback for sub_device %d",
185 				     SUB_ID(sdev));
186 		}
187 		dev->data->dev_conf.intr_conf.lsc = lsc_enabled;
188 		sdev->state = DEV_ACTIVE;
189 	}
190 	if (PRIV(dev)->state < DEV_ACTIVE)
191 		PRIV(dev)->state = DEV_ACTIVE;
192 	return 0;
193 }
194 
195 static int
196 fs_dev_start(struct rte_eth_dev *dev)
197 {
198 	struct sub_device *sdev;
199 	uint8_t i;
200 	int ret;
201 
202 	FOREACH_SUBDEV(sdev, i, dev) {
203 		if (sdev->state != DEV_ACTIVE)
204 			continue;
205 		DEBUG("Starting sub_device %d", i);
206 		ret = rte_eth_dev_start(PORT_ID(sdev));
207 		if (ret) {
208 			if (!fs_err(sdev, ret))
209 				continue;
210 			return ret;
211 		}
212 		sdev->state = DEV_STARTED;
213 	}
214 	if (PRIV(dev)->state < DEV_STARTED)
215 		PRIV(dev)->state = DEV_STARTED;
216 	fs_switch_dev(dev, NULL);
217 	return 0;
218 }
219 
220 static void
221 fs_dev_stop(struct rte_eth_dev *dev)
222 {
223 	struct sub_device *sdev;
224 	uint8_t i;
225 
226 	PRIV(dev)->state = DEV_STARTED - 1;
227 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_STARTED) {
228 		rte_eth_dev_stop(PORT_ID(sdev));
229 		sdev->state = DEV_STARTED - 1;
230 	}
231 }
232 
233 static int
234 fs_dev_set_link_up(struct rte_eth_dev *dev)
235 {
236 	struct sub_device *sdev;
237 	uint8_t i;
238 	int ret;
239 
240 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
241 		DEBUG("Calling rte_eth_dev_set_link_up on sub_device %d", i);
242 		ret = rte_eth_dev_set_link_up(PORT_ID(sdev));
243 		if ((ret = fs_err(sdev, ret))) {
244 			ERROR("Operation rte_eth_dev_set_link_up failed for sub_device %d"
245 			      " with error %d", i, ret);
246 			return ret;
247 		}
248 	}
249 	return 0;
250 }
251 
252 static int
253 fs_dev_set_link_down(struct rte_eth_dev *dev)
254 {
255 	struct sub_device *sdev;
256 	uint8_t i;
257 	int ret;
258 
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 			return ret;
266 		}
267 	}
268 	return 0;
269 }
270 
271 static void fs_dev_free_queues(struct rte_eth_dev *dev);
272 static void
273 fs_dev_close(struct rte_eth_dev *dev)
274 {
275 	struct sub_device *sdev;
276 	uint8_t i;
277 
278 	failsafe_hotplug_alarm_cancel(dev);
279 	if (PRIV(dev)->state == DEV_STARTED)
280 		dev->dev_ops->dev_stop(dev);
281 	PRIV(dev)->state = DEV_ACTIVE - 1;
282 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
283 		DEBUG("Closing sub_device %d", i);
284 		rte_eth_dev_close(PORT_ID(sdev));
285 		sdev->state = DEV_ACTIVE - 1;
286 	}
287 	fs_dev_free_queues(dev);
288 }
289 
290 static bool
291 fs_rxq_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
292 {
293 	uint64_t port_offloads;
294 	uint64_t queue_supp_offloads;
295 	uint64_t port_supp_offloads;
296 
297 	port_offloads = dev->data->dev_conf.rxmode.offloads;
298 	queue_supp_offloads = PRIV(dev)->infos.rx_queue_offload_capa;
299 	port_supp_offloads = PRIV(dev)->infos.rx_offload_capa;
300 	if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
301 	     offloads)
302 		return false;
303 	/* Verify we have no conflict with port offloads */
304 	if ((port_offloads ^ offloads) & port_supp_offloads)
305 		return false;
306 	return true;
307 }
308 
309 static void
310 fs_rx_queue_release(void *queue)
311 {
312 	struct rte_eth_dev *dev;
313 	struct sub_device *sdev;
314 	uint8_t i;
315 	struct rxq *rxq;
316 
317 	if (queue == NULL)
318 		return;
319 	rxq = queue;
320 	dev = rxq->priv->dev;
321 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
322 		SUBOPS(sdev, rx_queue_release)
323 			(ETH(sdev)->data->rx_queues[rxq->qid]);
324 	dev->data->rx_queues[rxq->qid] = NULL;
325 	rte_free(rxq);
326 }
327 
328 static int
329 fs_rx_queue_setup(struct rte_eth_dev *dev,
330 		uint16_t rx_queue_id,
331 		uint16_t nb_rx_desc,
332 		unsigned int socket_id,
333 		const struct rte_eth_rxconf *rx_conf,
334 		struct rte_mempool *mb_pool)
335 {
336 	struct sub_device *sdev;
337 	struct rxq *rxq;
338 	uint8_t i;
339 	int ret;
340 
341 	rxq = dev->data->rx_queues[rx_queue_id];
342 	if (rxq != NULL) {
343 		fs_rx_queue_release(rxq);
344 		dev->data->rx_queues[rx_queue_id] = NULL;
345 	}
346 	/* Verify application offloads are valid for our port and queue. */
347 	if (fs_rxq_offloads_valid(dev, rx_conf->offloads) == false) {
348 		rte_errno = ENOTSUP;
349 		ERROR("Rx queue offloads 0x%" PRIx64
350 		      " don't match port offloads 0x%" PRIx64
351 		      " or supported offloads 0x%" PRIx64,
352 		      rx_conf->offloads,
353 		      dev->data->dev_conf.rxmode.offloads,
354 		      PRIV(dev)->infos.rx_offload_capa |
355 		      PRIV(dev)->infos.rx_queue_offload_capa);
356 		return -rte_errno;
357 	}
358 	rxq = rte_zmalloc(NULL,
359 			  sizeof(*rxq) +
360 			  sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
361 			  RTE_CACHE_LINE_SIZE);
362 	if (rxq == NULL)
363 		return -ENOMEM;
364 	FOREACH_SUBDEV(sdev, i, dev)
365 		rte_atomic64_init(&rxq->refcnt[i]);
366 	rxq->qid = rx_queue_id;
367 	rxq->socket_id = socket_id;
368 	rxq->info.mp = mb_pool;
369 	rxq->info.conf = *rx_conf;
370 	rxq->info.nb_desc = nb_rx_desc;
371 	rxq->priv = PRIV(dev);
372 	rxq->sdev = PRIV(dev)->subs;
373 	dev->data->rx_queues[rx_queue_id] = rxq;
374 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
375 		ret = rte_eth_rx_queue_setup(PORT_ID(sdev),
376 				rx_queue_id,
377 				nb_rx_desc, socket_id,
378 				rx_conf, mb_pool);
379 		if ((ret = fs_err(sdev, ret))) {
380 			ERROR("RX queue setup failed for sub_device %d", i);
381 			goto free_rxq;
382 		}
383 	}
384 	return 0;
385 free_rxq:
386 	fs_rx_queue_release(rxq);
387 	return ret;
388 }
389 
390 static bool
391 fs_txq_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
392 {
393 	uint64_t port_offloads;
394 	uint64_t queue_supp_offloads;
395 	uint64_t port_supp_offloads;
396 
397 	port_offloads = dev->data->dev_conf.txmode.offloads;
398 	queue_supp_offloads = PRIV(dev)->infos.tx_queue_offload_capa;
399 	port_supp_offloads = PRIV(dev)->infos.tx_offload_capa;
400 	if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
401 	     offloads)
402 		return false;
403 	/* Verify we have no conflict with port offloads */
404 	if ((port_offloads ^ offloads) & port_supp_offloads)
405 		return false;
406 	return true;
407 }
408 
409 static void
410 fs_tx_queue_release(void *queue)
411 {
412 	struct rte_eth_dev *dev;
413 	struct sub_device *sdev;
414 	uint8_t i;
415 	struct txq *txq;
416 
417 	if (queue == NULL)
418 		return;
419 	txq = queue;
420 	dev = txq->priv->dev;
421 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
422 		SUBOPS(sdev, tx_queue_release)
423 			(ETH(sdev)->data->tx_queues[txq->qid]);
424 	dev->data->tx_queues[txq->qid] = NULL;
425 	rte_free(txq);
426 }
427 
428 static int
429 fs_tx_queue_setup(struct rte_eth_dev *dev,
430 		uint16_t tx_queue_id,
431 		uint16_t nb_tx_desc,
432 		unsigned int socket_id,
433 		const struct rte_eth_txconf *tx_conf)
434 {
435 	struct sub_device *sdev;
436 	struct txq *txq;
437 	uint8_t i;
438 	int ret;
439 
440 	txq = dev->data->tx_queues[tx_queue_id];
441 	if (txq != NULL) {
442 		fs_tx_queue_release(txq);
443 		dev->data->tx_queues[tx_queue_id] = NULL;
444 	}
445 	/*
446 	 * Don't verify queue offloads for applications which
447 	 * use the old API.
448 	 */
449 	if (tx_conf != NULL &&
450 	    (tx_conf->txq_flags & ETH_TXQ_FLAGS_IGNORE) &&
451 	    fs_txq_offloads_valid(dev, tx_conf->offloads) == false) {
452 		rte_errno = ENOTSUP;
453 		ERROR("Tx queue offloads 0x%" PRIx64
454 		      " don't match port offloads 0x%" PRIx64
455 		      " or supported offloads 0x%" PRIx64,
456 		      tx_conf->offloads,
457 		      dev->data->dev_conf.txmode.offloads,
458 		      PRIV(dev)->infos.tx_offload_capa |
459 		      PRIV(dev)->infos.tx_queue_offload_capa);
460 		return -rte_errno;
461 	}
462 	txq = rte_zmalloc("ethdev TX queue",
463 			  sizeof(*txq) +
464 			  sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
465 			  RTE_CACHE_LINE_SIZE);
466 	if (txq == NULL)
467 		return -ENOMEM;
468 	FOREACH_SUBDEV(sdev, i, dev)
469 		rte_atomic64_init(&txq->refcnt[i]);
470 	txq->qid = tx_queue_id;
471 	txq->socket_id = socket_id;
472 	txq->info.conf = *tx_conf;
473 	txq->info.nb_desc = nb_tx_desc;
474 	txq->priv = PRIV(dev);
475 	dev->data->tx_queues[tx_queue_id] = txq;
476 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
477 		ret = rte_eth_tx_queue_setup(PORT_ID(sdev),
478 				tx_queue_id,
479 				nb_tx_desc, socket_id,
480 				tx_conf);
481 		if ((ret = fs_err(sdev, ret))) {
482 			ERROR("TX queue setup failed for sub_device %d", i);
483 			goto free_txq;
484 		}
485 	}
486 	return 0;
487 free_txq:
488 	fs_tx_queue_release(txq);
489 	return ret;
490 }
491 
492 static void
493 fs_dev_free_queues(struct rte_eth_dev *dev)
494 {
495 	uint16_t i;
496 
497 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
498 		fs_rx_queue_release(dev->data->rx_queues[i]);
499 		dev->data->rx_queues[i] = NULL;
500 	}
501 	dev->data->nb_rx_queues = 0;
502 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
503 		fs_tx_queue_release(dev->data->tx_queues[i]);
504 		dev->data->tx_queues[i] = NULL;
505 	}
506 	dev->data->nb_tx_queues = 0;
507 }
508 
509 static void
510 fs_promiscuous_enable(struct rte_eth_dev *dev)
511 {
512 	struct sub_device *sdev;
513 	uint8_t i;
514 
515 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
516 		rte_eth_promiscuous_enable(PORT_ID(sdev));
517 }
518 
519 static void
520 fs_promiscuous_disable(struct rte_eth_dev *dev)
521 {
522 	struct sub_device *sdev;
523 	uint8_t i;
524 
525 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
526 		rte_eth_promiscuous_disable(PORT_ID(sdev));
527 }
528 
529 static void
530 fs_allmulticast_enable(struct rte_eth_dev *dev)
531 {
532 	struct sub_device *sdev;
533 	uint8_t i;
534 
535 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
536 		rte_eth_allmulticast_enable(PORT_ID(sdev));
537 }
538 
539 static void
540 fs_allmulticast_disable(struct rte_eth_dev *dev)
541 {
542 	struct sub_device *sdev;
543 	uint8_t i;
544 
545 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
546 		rte_eth_allmulticast_disable(PORT_ID(sdev));
547 }
548 
549 static int
550 fs_link_update(struct rte_eth_dev *dev,
551 		int wait_to_complete)
552 {
553 	struct sub_device *sdev;
554 	uint8_t i;
555 	int ret;
556 
557 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
558 		DEBUG("Calling link_update on sub_device %d", i);
559 		ret = (SUBOPS(sdev, link_update))(ETH(sdev), wait_to_complete);
560 		if (ret && ret != -1 && sdev->remove == 0 &&
561 		    rte_eth_dev_is_removed(PORT_ID(sdev)) == 0) {
562 			ERROR("Link update failed for sub_device %d with error %d",
563 			      i, ret);
564 			return ret;
565 		}
566 	}
567 	if (TX_SUBDEV(dev)) {
568 		struct rte_eth_link *l1;
569 		struct rte_eth_link *l2;
570 
571 		l1 = &dev->data->dev_link;
572 		l2 = &ETH(TX_SUBDEV(dev))->data->dev_link;
573 		if (memcmp(l1, l2, sizeof(*l1))) {
574 			*l1 = *l2;
575 			return 0;
576 		}
577 	}
578 	return -1;
579 }
580 
581 static int
582 fs_stats_get(struct rte_eth_dev *dev,
583 	     struct rte_eth_stats *stats)
584 {
585 	struct rte_eth_stats backup;
586 	struct sub_device *sdev;
587 	uint8_t i;
588 	int ret;
589 
590 	rte_memcpy(stats, &PRIV(dev)->stats_accumulator, sizeof(*stats));
591 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
592 		struct rte_eth_stats *snapshot = &sdev->stats_snapshot.stats;
593 		uint64_t *timestamp = &sdev->stats_snapshot.timestamp;
594 
595 		rte_memcpy(&backup, snapshot, sizeof(backup));
596 		ret = rte_eth_stats_get(PORT_ID(sdev), snapshot);
597 		if (ret) {
598 			if (!fs_err(sdev, ret)) {
599 				rte_memcpy(snapshot, &backup, sizeof(backup));
600 				goto inc;
601 			}
602 			ERROR("Operation rte_eth_stats_get failed for sub_device %d with error %d",
603 				  i, ret);
604 			*timestamp = 0;
605 			return ret;
606 		}
607 		*timestamp = rte_rdtsc();
608 inc:
609 		failsafe_stats_increment(stats, snapshot);
610 	}
611 	return 0;
612 }
613 
614 static void
615 fs_stats_reset(struct rte_eth_dev *dev)
616 {
617 	struct sub_device *sdev;
618 	uint8_t i;
619 
620 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
621 		rte_eth_stats_reset(PORT_ID(sdev));
622 		memset(&sdev->stats_snapshot, 0, sizeof(struct rte_eth_stats));
623 	}
624 	memset(&PRIV(dev)->stats_accumulator, 0, sizeof(struct rte_eth_stats));
625 }
626 
627 /**
628  * Fail-safe dev_infos_get rules:
629  *
630  * No sub_device:
631  *   Numerables:
632  *      Use the maximum possible values for any field, so as not
633  *      to impede any further configuration effort.
634  *   Capabilities:
635  *      Limits capabilities to those that are understood by the
636  *      fail-safe PMD. This understanding stems from the fail-safe
637  *      being capable of verifying that the related capability is
638  *      expressed within the device configuration (struct rte_eth_conf).
639  *
640  * At least one probed sub_device:
641  *   Numerables:
642  *      Uses values from the active probed sub_device
643  *      The rationale here is that if any sub_device is less capable
644  *      (for example concerning the number of queues) than the active
645  *      sub_device, then its subsequent configuration will fail.
646  *      It is impossible to foresee this failure when the failing sub_device
647  *      is supposed to be plugged-in later on, so the configuration process
648  *      is the single point of failure and error reporting.
649  *   Capabilities:
650  *      Uses a logical AND of RX capabilities among
651  *      all sub_devices and the default capabilities.
652  *      Uses a logical AND of TX capabilities among
653  *      the active probed sub_device and the default capabilities.
654  *
655  */
656 static void
657 fs_dev_infos_get(struct rte_eth_dev *dev,
658 		  struct rte_eth_dev_info *infos)
659 {
660 	struct sub_device *sdev;
661 	uint8_t i;
662 
663 	sdev = TX_SUBDEV(dev);
664 	if (sdev == NULL) {
665 		DEBUG("No probed device, using default infos");
666 		rte_memcpy(&PRIV(dev)->infos, &default_infos,
667 			   sizeof(default_infos));
668 	} else {
669 		uint64_t rx_offload_capa;
670 		uint64_t rxq_offload_capa;
671 
672 		rx_offload_capa = default_infos.rx_offload_capa;
673 		rxq_offload_capa = default_infos.rx_queue_offload_capa;
674 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
675 			rte_eth_dev_info_get(PORT_ID(sdev),
676 					&PRIV(dev)->infos);
677 			rx_offload_capa &= PRIV(dev)->infos.rx_offload_capa;
678 			rxq_offload_capa &=
679 					PRIV(dev)->infos.rx_queue_offload_capa;
680 		}
681 		sdev = TX_SUBDEV(dev);
682 		rte_eth_dev_info_get(PORT_ID(sdev), &PRIV(dev)->infos);
683 		PRIV(dev)->infos.rx_offload_capa = rx_offload_capa;
684 		PRIV(dev)->infos.rx_queue_offload_capa = rxq_offload_capa;
685 		PRIV(dev)->infos.tx_offload_capa &=
686 					default_infos.tx_offload_capa;
687 		PRIV(dev)->infos.tx_queue_offload_capa &=
688 					default_infos.tx_queue_offload_capa;
689 		PRIV(dev)->infos.flow_type_rss_offloads &=
690 					default_infos.flow_type_rss_offloads;
691 	}
692 	rte_memcpy(infos, &PRIV(dev)->infos, sizeof(*infos));
693 }
694 
695 static const uint32_t *
696 fs_dev_supported_ptypes_get(struct rte_eth_dev *dev)
697 {
698 	struct sub_device *sdev;
699 	struct rte_eth_dev *edev;
700 
701 	sdev = TX_SUBDEV(dev);
702 	if (sdev == NULL)
703 		return NULL;
704 	edev = ETH(sdev);
705 	/* ENOTSUP: counts as no supported ptypes */
706 	if (SUBOPS(sdev, dev_supported_ptypes_get) == NULL)
707 		return NULL;
708 	/*
709 	 * The API does not permit to do a clean AND of all ptypes,
710 	 * It is also incomplete by design and we do not really care
711 	 * to have a best possible value in this context.
712 	 * We just return the ptypes of the device of highest
713 	 * priority, usually the PREFERRED device.
714 	 */
715 	return SUBOPS(sdev, dev_supported_ptypes_get)(edev);
716 }
717 
718 static int
719 fs_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
720 {
721 	struct sub_device *sdev;
722 	uint8_t i;
723 	int ret;
724 
725 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
726 		DEBUG("Calling rte_eth_dev_set_mtu on sub_device %d", i);
727 		ret = rte_eth_dev_set_mtu(PORT_ID(sdev), mtu);
728 		if ((ret = fs_err(sdev, ret))) {
729 			ERROR("Operation rte_eth_dev_set_mtu failed for sub_device %d with error %d",
730 			      i, ret);
731 			return ret;
732 		}
733 	}
734 	return 0;
735 }
736 
737 static int
738 fs_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
739 {
740 	struct sub_device *sdev;
741 	uint8_t i;
742 	int ret;
743 
744 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
745 		DEBUG("Calling rte_eth_dev_vlan_filter on sub_device %d", i);
746 		ret = rte_eth_dev_vlan_filter(PORT_ID(sdev), vlan_id, on);
747 		if ((ret = fs_err(sdev, ret))) {
748 			ERROR("Operation rte_eth_dev_vlan_filter failed for sub_device %d"
749 			      " with error %d", i, ret);
750 			return ret;
751 		}
752 	}
753 	return 0;
754 }
755 
756 static int
757 fs_flow_ctrl_get(struct rte_eth_dev *dev,
758 		struct rte_eth_fc_conf *fc_conf)
759 {
760 	struct sub_device *sdev;
761 
762 	sdev = TX_SUBDEV(dev);
763 	if (sdev == NULL)
764 		return 0;
765 	if (SUBOPS(sdev, flow_ctrl_get) == NULL)
766 		return -ENOTSUP;
767 	return SUBOPS(sdev, flow_ctrl_get)(ETH(sdev), fc_conf);
768 }
769 
770 static int
771 fs_flow_ctrl_set(struct rte_eth_dev *dev,
772 		struct rte_eth_fc_conf *fc_conf)
773 {
774 	struct sub_device *sdev;
775 	uint8_t i;
776 	int ret;
777 
778 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
779 		DEBUG("Calling rte_eth_dev_flow_ctrl_set on sub_device %d", i);
780 		ret = rte_eth_dev_flow_ctrl_set(PORT_ID(sdev), fc_conf);
781 		if ((ret = fs_err(sdev, ret))) {
782 			ERROR("Operation rte_eth_dev_flow_ctrl_set failed for sub_device %d"
783 			      " with error %d", i, ret);
784 			return ret;
785 		}
786 	}
787 	return 0;
788 }
789 
790 static void
791 fs_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
792 {
793 	struct sub_device *sdev;
794 	uint8_t i;
795 
796 	/* No check: already done within the rte_eth_dev_mac_addr_remove
797 	 * call for the fail-safe device.
798 	 */
799 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
800 		rte_eth_dev_mac_addr_remove(PORT_ID(sdev),
801 				&dev->data->mac_addrs[index]);
802 	PRIV(dev)->mac_addr_pool[index] = 0;
803 }
804 
805 static int
806 fs_mac_addr_add(struct rte_eth_dev *dev,
807 		struct ether_addr *mac_addr,
808 		uint32_t index,
809 		uint32_t vmdq)
810 {
811 	struct sub_device *sdev;
812 	int ret;
813 	uint8_t i;
814 
815 	RTE_ASSERT(index < FAILSAFE_MAX_ETHADDR);
816 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
817 		ret = rte_eth_dev_mac_addr_add(PORT_ID(sdev), mac_addr, vmdq);
818 		if ((ret = fs_err(sdev, ret))) {
819 			ERROR("Operation rte_eth_dev_mac_addr_add failed for sub_device %"
820 			      PRIu8 " with error %d", i, ret);
821 			return ret;
822 		}
823 	}
824 	if (index >= PRIV(dev)->nb_mac_addr) {
825 		DEBUG("Growing mac_addrs array");
826 		PRIV(dev)->nb_mac_addr = index;
827 	}
828 	PRIV(dev)->mac_addr_pool[index] = vmdq;
829 	return 0;
830 }
831 
832 static void
833 fs_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
834 {
835 	struct sub_device *sdev;
836 	uint8_t i;
837 
838 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
839 		rte_eth_dev_default_mac_addr_set(PORT_ID(sdev), mac_addr);
840 }
841 
842 static int
843 fs_filter_ctrl(struct rte_eth_dev *dev,
844 		enum rte_filter_type type,
845 		enum rte_filter_op op,
846 		void *arg)
847 {
848 	struct sub_device *sdev;
849 	uint8_t i;
850 	int ret;
851 
852 	if (type == RTE_ETH_FILTER_GENERIC &&
853 	    op == RTE_ETH_FILTER_GET) {
854 		*(const void **)arg = &fs_flow_ops;
855 		return 0;
856 	}
857 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
858 		DEBUG("Calling rte_eth_dev_filter_ctrl on sub_device %d", i);
859 		ret = rte_eth_dev_filter_ctrl(PORT_ID(sdev), type, op, arg);
860 		if ((ret = fs_err(sdev, ret))) {
861 			ERROR("Operation rte_eth_dev_filter_ctrl failed for sub_device %d"
862 			      " with error %d", i, ret);
863 			return ret;
864 		}
865 	}
866 	return 0;
867 }
868 
869 const struct eth_dev_ops failsafe_ops = {
870 	.dev_configure = fs_dev_configure,
871 	.dev_start = fs_dev_start,
872 	.dev_stop = fs_dev_stop,
873 	.dev_set_link_down = fs_dev_set_link_down,
874 	.dev_set_link_up = fs_dev_set_link_up,
875 	.dev_close = fs_dev_close,
876 	.promiscuous_enable = fs_promiscuous_enable,
877 	.promiscuous_disable = fs_promiscuous_disable,
878 	.allmulticast_enable = fs_allmulticast_enable,
879 	.allmulticast_disable = fs_allmulticast_disable,
880 	.link_update = fs_link_update,
881 	.stats_get = fs_stats_get,
882 	.stats_reset = fs_stats_reset,
883 	.dev_infos_get = fs_dev_infos_get,
884 	.dev_supported_ptypes_get = fs_dev_supported_ptypes_get,
885 	.mtu_set = fs_mtu_set,
886 	.vlan_filter_set = fs_vlan_filter_set,
887 	.rx_queue_setup = fs_rx_queue_setup,
888 	.tx_queue_setup = fs_tx_queue_setup,
889 	.rx_queue_release = fs_rx_queue_release,
890 	.tx_queue_release = fs_tx_queue_release,
891 	.flow_ctrl_get = fs_flow_ctrl_get,
892 	.flow_ctrl_set = fs_flow_ctrl_set,
893 	.mac_addr_remove = fs_mac_addr_remove,
894 	.mac_addr_add = fs_mac_addr_add,
895 	.mac_addr_set = fs_mac_addr_set,
896 	.filter_ctrl = fs_filter_ctrl,
897 };
898