xref: /dpdk/drivers/net/failsafe/failsafe_ether.c (revision 7bd6f76ee678ec6aa81cb53562f852a43e842718)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2017 6WIND S.A.
3  * Copyright 2017 Mellanox Technologies, Ltd
4  */
5 
6 #include <unistd.h>
7 
8 #include <rte_flow.h>
9 #include <rte_flow_driver.h>
10 #include <rte_cycles.h>
11 
12 #include "failsafe_private.h"
13 
14 /** Print a message out of a flow error. */
15 static int
16 fs_flow_complain(struct rte_flow_error *error)
17 {
18 	static const char *const errstrlist[] = {
19 		[RTE_FLOW_ERROR_TYPE_NONE] = "no error",
20 		[RTE_FLOW_ERROR_TYPE_UNSPECIFIED] = "cause unspecified",
21 		[RTE_FLOW_ERROR_TYPE_HANDLE] = "flow rule (handle)",
22 		[RTE_FLOW_ERROR_TYPE_ATTR_GROUP] = "group field",
23 		[RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY] = "priority field",
24 		[RTE_FLOW_ERROR_TYPE_ATTR_INGRESS] = "ingress field",
25 		[RTE_FLOW_ERROR_TYPE_ATTR_EGRESS] = "egress field",
26 		[RTE_FLOW_ERROR_TYPE_ATTR] = "attributes structure",
27 		[RTE_FLOW_ERROR_TYPE_ITEM_NUM] = "pattern length",
28 		[RTE_FLOW_ERROR_TYPE_ITEM] = "specific pattern item",
29 		[RTE_FLOW_ERROR_TYPE_ACTION_NUM] = "number of actions",
30 		[RTE_FLOW_ERROR_TYPE_ACTION] = "specific action",
31 	};
32 	const char *errstr;
33 	char buf[32];
34 	int err = rte_errno;
35 
36 	if ((unsigned int)error->type >= RTE_DIM(errstrlist) ||
37 			!errstrlist[error->type])
38 		errstr = "unknown type";
39 	else
40 		errstr = errstrlist[error->type];
41 	ERROR("Caught error type %d (%s): %s%s\n",
42 		error->type, errstr,
43 		error->cause ? (snprintf(buf, sizeof(buf), "cause: %p, ",
44 				error->cause), buf) : "",
45 		error->message ? error->message : "(no stated reason)");
46 	return -err;
47 }
48 
49 static int
50 eth_dev_flow_isolate_set(struct rte_eth_dev *dev,
51 			 struct sub_device *sdev)
52 {
53 	struct rte_flow_error ferror;
54 	int ret;
55 
56 	if (!PRIV(dev)->flow_isolated) {
57 		DEBUG("Flow isolation already disabled");
58 	} else {
59 		DEBUG("Enabling flow isolation");
60 		ret = rte_flow_isolate(PORT_ID(sdev),
61 				       PRIV(dev)->flow_isolated,
62 				       &ferror);
63 		if (ret) {
64 			fs_flow_complain(&ferror);
65 			return ret;
66 		}
67 	}
68 	return 0;
69 }
70 
71 static int
72 fs_eth_dev_conf_apply(struct rte_eth_dev *dev,
73 		struct sub_device *sdev)
74 {
75 	struct rte_eth_dev *edev;
76 	struct rte_vlan_filter_conf *vfc1;
77 	struct rte_vlan_filter_conf *vfc2;
78 	struct rte_flow *flow;
79 	struct rte_flow_error ferror;
80 	uint32_t i;
81 	int ret;
82 
83 	edev = ETH(sdev);
84 	/* RX queue setup */
85 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
86 		struct rxq *rxq;
87 
88 		rxq = dev->data->rx_queues[i];
89 		ret = rte_eth_rx_queue_setup(PORT_ID(sdev), i,
90 				rxq->info.nb_desc, rxq->socket_id,
91 				&rxq->info.conf, rxq->info.mp);
92 		if (ret) {
93 			ERROR("rx_queue_setup failed");
94 			return ret;
95 		}
96 	}
97 	/* TX queue setup */
98 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
99 		struct txq *txq;
100 
101 		txq = dev->data->tx_queues[i];
102 		ret = rte_eth_tx_queue_setup(PORT_ID(sdev), i,
103 				txq->info.nb_desc, txq->socket_id,
104 				&txq->info.conf);
105 		if (ret) {
106 			ERROR("tx_queue_setup failed");
107 			return ret;
108 		}
109 	}
110 	/* dev_link.link_status */
111 	if (dev->data->dev_link.link_status !=
112 	    edev->data->dev_link.link_status) {
113 		DEBUG("Configuring link_status");
114 		if (dev->data->dev_link.link_status)
115 			ret = rte_eth_dev_set_link_up(PORT_ID(sdev));
116 		else
117 			ret = rte_eth_dev_set_link_down(PORT_ID(sdev));
118 		if (ret) {
119 			ERROR("Failed to apply link_status");
120 			return ret;
121 		}
122 	} else {
123 		DEBUG("link_status already set");
124 	}
125 	/* promiscuous */
126 	if (dev->data->promiscuous != edev->data->promiscuous) {
127 		DEBUG("Configuring promiscuous");
128 		if (dev->data->promiscuous)
129 			rte_eth_promiscuous_enable(PORT_ID(sdev));
130 		else
131 			rte_eth_promiscuous_disable(PORT_ID(sdev));
132 	} else {
133 		DEBUG("promiscuous already set");
134 	}
135 	/* all_multicast */
136 	if (dev->data->all_multicast != edev->data->all_multicast) {
137 		DEBUG("Configuring all_multicast");
138 		if (dev->data->all_multicast)
139 			rte_eth_allmulticast_enable(PORT_ID(sdev));
140 		else
141 			rte_eth_allmulticast_disable(PORT_ID(sdev));
142 	} else {
143 		DEBUG("all_multicast already set");
144 	}
145 	/* MTU */
146 	if (dev->data->mtu != edev->data->mtu) {
147 		DEBUG("Configuring MTU");
148 		ret = rte_eth_dev_set_mtu(PORT_ID(sdev), dev->data->mtu);
149 		if (ret) {
150 			ERROR("Failed to apply MTU");
151 			return ret;
152 		}
153 	} else {
154 		DEBUG("MTU already set");
155 	}
156 	/* default MAC */
157 	DEBUG("Configuring default MAC address");
158 	ret = rte_eth_dev_default_mac_addr_set(PORT_ID(sdev),
159 			&dev->data->mac_addrs[0]);
160 	if (ret) {
161 		ERROR("Setting default MAC address failed");
162 		return ret;
163 	}
164 	/* additional MAC */
165 	if (PRIV(dev)->nb_mac_addr > 1)
166 		DEBUG("Configure additional MAC address%s",
167 			(PRIV(dev)->nb_mac_addr > 2 ? "es" : ""));
168 	for (i = 1; i < PRIV(dev)->nb_mac_addr; i++) {
169 		struct ether_addr *ea;
170 
171 		ea = &dev->data->mac_addrs[i];
172 		ret = rte_eth_dev_mac_addr_add(PORT_ID(sdev), ea,
173 				PRIV(dev)->mac_addr_pool[i]);
174 		if (ret) {
175 			char ea_fmt[ETHER_ADDR_FMT_SIZE];
176 
177 			ether_format_addr(ea_fmt, ETHER_ADDR_FMT_SIZE, ea);
178 			ERROR("Adding MAC address %s failed", ea_fmt);
179 			return ret;
180 		}
181 	}
182 	/*
183 	 * Propagate multicast MAC addresses to sub-devices,
184 	 * if non zero number of addresses is set.
185 	 * The condition is required to avoid breakage of failsafe
186 	 * for sub-devices which do not support the operation
187 	 * if the feature is really not used.
188 	 */
189 	if (PRIV(dev)->nb_mcast_addr > 0) {
190 		DEBUG("Configuring multicast MAC addresses");
191 		ret = rte_eth_dev_set_mc_addr_list(PORT_ID(sdev),
192 						   PRIV(dev)->mcast_addrs,
193 						   PRIV(dev)->nb_mcast_addr);
194 		if (ret) {
195 			ERROR("Failed to apply multicast MAC addresses");
196 			return ret;
197 		}
198 	}
199 	/* VLAN filter */
200 	vfc1 = &dev->data->vlan_filter_conf;
201 	vfc2 = &edev->data->vlan_filter_conf;
202 	if (memcmp(vfc1, vfc2, sizeof(struct rte_vlan_filter_conf))) {
203 		uint64_t vbit;
204 		uint64_t ids;
205 		size_t i;
206 		uint16_t vlan_id;
207 
208 		DEBUG("Configuring VLAN filter");
209 		for (i = 0; i < RTE_DIM(vfc1->ids); i++) {
210 			if (vfc1->ids[i] == 0)
211 				continue;
212 			ids = vfc1->ids[i];
213 			while (ids) {
214 				vlan_id = 64 * i;
215 				/* count trailing zeroes */
216 				vbit = ~ids & (ids - 1);
217 				/* clear least significant bit set */
218 				ids ^= (ids ^ (ids - 1)) ^ vbit;
219 				for (; vbit; vlan_id++)
220 					vbit >>= 1;
221 				ret = rte_eth_dev_vlan_filter(
222 					PORT_ID(sdev), vlan_id, 1);
223 				if (ret) {
224 					ERROR("Failed to apply VLAN filter %hu",
225 						vlan_id);
226 					return ret;
227 				}
228 			}
229 		}
230 	} else {
231 		DEBUG("VLAN filter already set");
232 	}
233 	/* rte_flow */
234 	if (TAILQ_EMPTY(&PRIV(dev)->flow_list)) {
235 		DEBUG("rte_flow already set");
236 	} else {
237 		DEBUG("Resetting rte_flow configuration");
238 		ret = rte_flow_flush(PORT_ID(sdev), &ferror);
239 		if (ret) {
240 			fs_flow_complain(&ferror);
241 			return ret;
242 		}
243 		i = 0;
244 		rte_errno = 0;
245 		DEBUG("Configuring rte_flow");
246 		TAILQ_FOREACH(flow, &PRIV(dev)->flow_list, next) {
247 			DEBUG("Creating flow #%" PRIu32, i++);
248 			flow->flows[SUB_ID(sdev)] =
249 				rte_flow_create(PORT_ID(sdev),
250 						&flow->fd->attr,
251 						flow->fd->items,
252 						flow->fd->actions,
253 						&ferror);
254 			ret = rte_errno;
255 			if (ret)
256 				break;
257 		}
258 		if (ret) {
259 			fs_flow_complain(&ferror);
260 			return ret;
261 		}
262 	}
263 	return 0;
264 }
265 
266 static void
267 fs_dev_remove(struct sub_device *sdev)
268 {
269 	int ret;
270 
271 	if (sdev == NULL)
272 		return;
273 	switch (sdev->state) {
274 	case DEV_STARTED:
275 		failsafe_rx_intr_uninstall_subdevice(sdev);
276 		rte_eth_dev_stop(PORT_ID(sdev));
277 		sdev->state = DEV_ACTIVE;
278 		/* fallthrough */
279 	case DEV_ACTIVE:
280 		failsafe_eth_dev_unregister_callbacks(sdev);
281 		rte_eth_dev_close(PORT_ID(sdev));
282 		sdev->state = DEV_PROBED;
283 		/* fallthrough */
284 	case DEV_PROBED:
285 		ret = rte_eal_hotplug_remove(sdev->bus->name,
286 					     sdev->dev->name);
287 		if (ret) {
288 			ERROR("Bus detach failed for sub_device %u",
289 			      SUB_ID(sdev));
290 		} else {
291 			rte_eth_dev_release_port(ETH(sdev));
292 		}
293 		sdev->state = DEV_PARSED;
294 		/* fallthrough */
295 	case DEV_PARSED:
296 	case DEV_UNDEFINED:
297 		sdev->state = DEV_UNDEFINED;
298 		/* the end */
299 		break;
300 	}
301 	sdev->remove = 0;
302 	failsafe_hotplug_alarm_install(sdev->fs_dev);
303 }
304 
305 static void
306 fs_dev_stats_save(struct sub_device *sdev)
307 {
308 	struct rte_eth_stats stats;
309 	int err;
310 
311 	/* Attempt to read current stats. */
312 	err = rte_eth_stats_get(PORT_ID(sdev), &stats);
313 	if (err) {
314 		uint64_t timestamp = sdev->stats_snapshot.timestamp;
315 
316 		WARN("Could not access latest statistics from sub-device %d.\n",
317 			 SUB_ID(sdev));
318 		if (timestamp != 0)
319 			WARN("Using latest snapshot taken before %"PRIu64" seconds.\n",
320 				 (rte_rdtsc() - timestamp) / rte_get_tsc_hz());
321 	}
322 	failsafe_stats_increment(&PRIV(sdev->fs_dev)->stats_accumulator,
323 			err ? &sdev->stats_snapshot.stats : &stats);
324 	memset(&sdev->stats_snapshot, 0, sizeof(sdev->stats_snapshot));
325 }
326 
327 static inline int
328 fs_rxtx_clean(struct sub_device *sdev)
329 {
330 	uint16_t i;
331 
332 	for (i = 0; i < ETH(sdev)->data->nb_rx_queues; i++)
333 		if (FS_ATOMIC_RX(sdev, i))
334 			return 0;
335 	for (i = 0; i < ETH(sdev)->data->nb_tx_queues; i++)
336 		if (FS_ATOMIC_TX(sdev, i))
337 			return 0;
338 	return 1;
339 }
340 
341 void
342 failsafe_eth_dev_unregister_callbacks(struct sub_device *sdev)
343 {
344 	int ret;
345 
346 	if (sdev == NULL)
347 		return;
348 	if (sdev->rmv_callback) {
349 		ret = rte_eth_dev_callback_unregister(PORT_ID(sdev),
350 						RTE_ETH_EVENT_INTR_RMV,
351 						failsafe_eth_rmv_event_callback,
352 						sdev);
353 		if (ret)
354 			WARN("Failed to unregister RMV callback for sub_device"
355 			     " %d", SUB_ID(sdev));
356 		sdev->rmv_callback = 0;
357 	}
358 	if (sdev->lsc_callback) {
359 		ret = rte_eth_dev_callback_unregister(PORT_ID(sdev),
360 						RTE_ETH_EVENT_INTR_LSC,
361 						failsafe_eth_lsc_event_callback,
362 						sdev);
363 		if (ret)
364 			WARN("Failed to unregister LSC callback for sub_device"
365 			     " %d", SUB_ID(sdev));
366 		sdev->lsc_callback = 0;
367 	}
368 }
369 
370 void
371 failsafe_dev_remove(struct rte_eth_dev *dev)
372 {
373 	struct sub_device *sdev;
374 	uint8_t i;
375 
376 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
377 		if (sdev->remove && fs_rxtx_clean(sdev)) {
378 			if (fs_lock(dev, 1) != 0)
379 				return;
380 			fs_dev_stats_save(sdev);
381 			fs_dev_remove(sdev);
382 			fs_unlock(dev, 1);
383 		}
384 }
385 
386 static int
387 failsafe_eth_dev_rx_queues_sync(struct rte_eth_dev *dev)
388 {
389 	struct rxq *rxq;
390 	int ret;
391 	uint16_t i;
392 
393 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
394 		rxq = dev->data->rx_queues[i];
395 
396 		if (rxq->info.conf.rx_deferred_start &&
397 		    dev->data->rx_queue_state[i] ==
398 						RTE_ETH_QUEUE_STATE_STARTED) {
399 			/*
400 			 * The subdevice Rx queue does not launch on device
401 			 * start if deferred start flag is set. It needs to be
402 			 * started manually in case an appropriate failsafe Rx
403 			 * queue has been started earlier.
404 			 */
405 			ret = dev->dev_ops->rx_queue_start(dev, i);
406 			if (ret) {
407 				ERROR("Could not synchronize Rx queue %d", i);
408 				return ret;
409 			}
410 		} else if (dev->data->rx_queue_state[i] ==
411 						RTE_ETH_QUEUE_STATE_STOPPED) {
412 			/*
413 			 * The subdevice Rx queue needs to be stopped manually
414 			 * in case an appropriate failsafe Rx queue has been
415 			 * stopped earlier.
416 			 */
417 			ret = dev->dev_ops->rx_queue_stop(dev, i);
418 			if (ret) {
419 				ERROR("Could not synchronize Rx queue %d", i);
420 				return ret;
421 			}
422 		}
423 	}
424 	return 0;
425 }
426 
427 static int
428 failsafe_eth_dev_tx_queues_sync(struct rte_eth_dev *dev)
429 {
430 	struct txq *txq;
431 	int ret;
432 	uint16_t i;
433 
434 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
435 		txq = dev->data->tx_queues[i];
436 
437 		if (txq->info.conf.tx_deferred_start &&
438 		    dev->data->tx_queue_state[i] ==
439 						RTE_ETH_QUEUE_STATE_STARTED) {
440 			/*
441 			 * The subdevice Tx queue does not launch on device
442 			 * start if deferred start flag is set. It needs to be
443 			 * started manually in case an appropriate failsafe Tx
444 			 * queue has been started earlier.
445 			 */
446 			ret = dev->dev_ops->tx_queue_start(dev, i);
447 			if (ret) {
448 				ERROR("Could not synchronize Tx queue %d", i);
449 				return ret;
450 			}
451 		} else if (dev->data->tx_queue_state[i] ==
452 						RTE_ETH_QUEUE_STATE_STOPPED) {
453 			/*
454 			 * The subdevice Tx queue needs to be stopped manually
455 			 * in case an appropriate failsafe Tx queue has been
456 			 * stopped earlier.
457 			 */
458 			ret = dev->dev_ops->tx_queue_stop(dev, i);
459 			if (ret) {
460 				ERROR("Could not synchronize Tx queue %d", i);
461 				return ret;
462 			}
463 		}
464 	}
465 	return 0;
466 }
467 
468 int
469 failsafe_eth_dev_state_sync(struct rte_eth_dev *dev)
470 {
471 	struct sub_device *sdev;
472 	uint32_t inactive;
473 	int ret;
474 	uint8_t i;
475 
476 	if (PRIV(dev)->state < DEV_PARSED)
477 		return 0;
478 
479 	ret = failsafe_args_parse_subs(dev);
480 	if (ret)
481 		goto err_remove;
482 
483 	if (PRIV(dev)->state < DEV_PROBED)
484 		return 0;
485 	ret = failsafe_eal_init(dev);
486 	if (ret)
487 		goto err_remove;
488 	if (PRIV(dev)->state < DEV_ACTIVE)
489 		return 0;
490 	inactive = 0;
491 	FOREACH_SUBDEV(sdev, i, dev) {
492 		if (sdev->state == DEV_PROBED) {
493 			inactive |= UINT32_C(1) << i;
494 			ret = eth_dev_flow_isolate_set(dev, sdev);
495 			if (ret) {
496 				ERROR("Could not apply configuration to sub_device %d",
497 				      i);
498 				goto err_remove;
499 			}
500 		}
501 	}
502 	ret = dev->dev_ops->dev_configure(dev);
503 	if (ret)
504 		goto err_remove;
505 	FOREACH_SUBDEV(sdev, i, dev) {
506 		if (inactive & (UINT32_C(1) << i)) {
507 			ret = fs_eth_dev_conf_apply(dev, sdev);
508 			if (ret) {
509 				ERROR("Could not apply configuration to sub_device %d",
510 				      i);
511 				goto err_remove;
512 			}
513 		}
514 	}
515 	/*
516 	 * If new devices have been configured, check if
517 	 * the link state has changed.
518 	 */
519 	if (inactive)
520 		dev->dev_ops->link_update(dev, 1);
521 	if (PRIV(dev)->state < DEV_STARTED)
522 		return 0;
523 	ret = dev->dev_ops->dev_start(dev);
524 	if (ret)
525 		goto err_remove;
526 	ret = failsafe_eth_dev_rx_queues_sync(dev);
527 	if (ret)
528 		goto err_remove;
529 	ret = failsafe_eth_dev_tx_queues_sync(dev);
530 	if (ret)
531 		goto err_remove;
532 	return 0;
533 err_remove:
534 	FOREACH_SUBDEV(sdev, i, dev)
535 		if (sdev->state != PRIV(dev)->state)
536 			sdev->remove = 1;
537 	return ret;
538 }
539 
540 void
541 failsafe_stats_increment(struct rte_eth_stats *to, struct rte_eth_stats *from)
542 {
543 	uint32_t i;
544 
545 	RTE_ASSERT(to != NULL && from != NULL);
546 	to->ipackets += from->ipackets;
547 	to->opackets += from->opackets;
548 	to->ibytes += from->ibytes;
549 	to->obytes += from->obytes;
550 	to->imissed += from->imissed;
551 	to->ierrors += from->ierrors;
552 	to->oerrors += from->oerrors;
553 	to->rx_nombuf += from->rx_nombuf;
554 	for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS; i++) {
555 		to->q_ipackets[i] += from->q_ipackets[i];
556 		to->q_opackets[i] += from->q_opackets[i];
557 		to->q_ibytes[i] += from->q_ibytes[i];
558 		to->q_obytes[i] += from->q_obytes[i];
559 		to->q_errors[i] += from->q_errors[i];
560 	}
561 }
562 
563 int
564 failsafe_eth_rmv_event_callback(uint16_t port_id __rte_unused,
565 				enum rte_eth_event_type event __rte_unused,
566 				void *cb_arg, void *out __rte_unused)
567 {
568 	struct sub_device *sdev = cb_arg;
569 
570 	fs_lock(sdev->fs_dev, 0);
571 	/* Switch as soon as possible tx_dev. */
572 	fs_switch_dev(sdev->fs_dev, sdev);
573 	/* Use safe bursts in any case. */
574 	set_burst_fn(sdev->fs_dev, 1);
575 	/*
576 	 * Async removal, the sub-PMD will try to unregister
577 	 * the callback at the source of the current thread context.
578 	 */
579 	sdev->remove = 1;
580 	fs_unlock(sdev->fs_dev, 0);
581 	return 0;
582 }
583 
584 int
585 failsafe_eth_lsc_event_callback(uint16_t port_id __rte_unused,
586 				enum rte_eth_event_type event __rte_unused,
587 				void *cb_arg, void *out __rte_unused)
588 {
589 	struct rte_eth_dev *dev = cb_arg;
590 	int ret;
591 
592 	ret = dev->dev_ops->link_update(dev, 0);
593 	/* We must pass on the LSC event */
594 	if (ret)
595 		return _rte_eth_dev_callback_process(dev,
596 						     RTE_ETH_EVENT_INTR_LSC,
597 						     NULL);
598 	else
599 		return 0;
600 }
601 
602 /* Take sub-device ownership before it becomes exposed to the application. */
603 int
604 failsafe_eth_new_event_callback(uint16_t port_id,
605 				enum rte_eth_event_type event __rte_unused,
606 				void *cb_arg, void *out __rte_unused)
607 {
608 	struct rte_eth_dev *fs_dev = cb_arg;
609 	struct sub_device *sdev;
610 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
611 	uint8_t i;
612 
613 	FOREACH_SUBDEV_STATE(sdev, i, fs_dev, DEV_PARSED) {
614 		if (sdev->state >= DEV_PROBED)
615 			continue;
616 		if (strcmp(sdev->devargs.name, dev->device->name) != 0)
617 			continue;
618 		rte_eth_dev_owner_set(port_id, &PRIV(fs_dev)->my_owner);
619 		/* The actual owner will be checked after the port probing. */
620 		break;
621 	}
622 	return 0;
623 }
624