xref: /dpdk/drivers/net/failsafe/failsafe_ether.c (revision 117eaa70584b73eebf6f648cf3ee6f2ab03264a0)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2017 6WIND S.A.
3  * Copyright 2017 Mellanox.
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 	/* VLAN filter */
183 	vfc1 = &dev->data->vlan_filter_conf;
184 	vfc2 = &edev->data->vlan_filter_conf;
185 	if (memcmp(vfc1, vfc2, sizeof(struct rte_vlan_filter_conf))) {
186 		uint64_t vbit;
187 		uint64_t ids;
188 		size_t i;
189 		uint16_t vlan_id;
190 
191 		DEBUG("Configuring VLAN filter");
192 		for (i = 0; i < RTE_DIM(vfc1->ids); i++) {
193 			if (vfc1->ids[i] == 0)
194 				continue;
195 			ids = vfc1->ids[i];
196 			while (ids) {
197 				vlan_id = 64 * i;
198 				/* count trailing zeroes */
199 				vbit = ~ids & (ids - 1);
200 				/* clear least significant bit set */
201 				ids ^= (ids ^ (ids - 1)) ^ vbit;
202 				for (; vbit; vlan_id++)
203 					vbit >>= 1;
204 				ret = rte_eth_dev_vlan_filter(
205 					PORT_ID(sdev), vlan_id, 1);
206 				if (ret) {
207 					ERROR("Failed to apply VLAN filter %hu",
208 						vlan_id);
209 					return ret;
210 				}
211 			}
212 		}
213 	} else {
214 		DEBUG("VLAN filter already set");
215 	}
216 	/* rte_flow */
217 	if (TAILQ_EMPTY(&PRIV(dev)->flow_list)) {
218 		DEBUG("rte_flow already set");
219 	} else {
220 		DEBUG("Resetting rte_flow configuration");
221 		ret = rte_flow_flush(PORT_ID(sdev), &ferror);
222 		if (ret) {
223 			fs_flow_complain(&ferror);
224 			return ret;
225 		}
226 		i = 0;
227 		rte_errno = 0;
228 		DEBUG("Configuring rte_flow");
229 		TAILQ_FOREACH(flow, &PRIV(dev)->flow_list, next) {
230 			DEBUG("Creating flow #%" PRIu32, i++);
231 			flow->flows[SUB_ID(sdev)] =
232 				rte_flow_create(PORT_ID(sdev),
233 						&flow->fd->attr,
234 						flow->fd->items,
235 						flow->fd->actions,
236 						&ferror);
237 			ret = rte_errno;
238 			if (ret)
239 				break;
240 		}
241 		if (ret) {
242 			fs_flow_complain(&ferror);
243 			return ret;
244 		}
245 	}
246 	return 0;
247 }
248 
249 static void
250 fs_dev_remove(struct sub_device *sdev)
251 {
252 	int ret;
253 
254 	if (sdev == NULL)
255 		return;
256 	switch (sdev->state) {
257 	case DEV_STARTED:
258 		failsafe_rx_intr_uninstall_subdevice(sdev);
259 		rte_eth_dev_stop(PORT_ID(sdev));
260 		sdev->state = DEV_ACTIVE;
261 		/* fallthrough */
262 	case DEV_ACTIVE:
263 		rte_eth_dev_close(PORT_ID(sdev));
264 		sdev->state = DEV_PROBED;
265 		/* fallthrough */
266 	case DEV_PROBED:
267 		ret = rte_eal_hotplug_remove(sdev->bus->name,
268 					     sdev->dev->name);
269 		if (ret) {
270 			ERROR("Bus detach failed for sub_device %u",
271 			      SUB_ID(sdev));
272 		} else {
273 			rte_eth_dev_release_port(ETH(sdev));
274 		}
275 		sdev->state = DEV_PARSED;
276 		/* fallthrough */
277 	case DEV_PARSED:
278 	case DEV_UNDEFINED:
279 		sdev->state = DEV_UNDEFINED;
280 		/* the end */
281 		break;
282 	}
283 	failsafe_hotplug_alarm_install(sdev->fs_dev);
284 }
285 
286 static void
287 fs_dev_stats_save(struct sub_device *sdev)
288 {
289 	struct rte_eth_stats stats;
290 	int err;
291 
292 	/* Attempt to read current stats. */
293 	err = rte_eth_stats_get(PORT_ID(sdev), &stats);
294 	if (err) {
295 		uint64_t timestamp = sdev->stats_snapshot.timestamp;
296 
297 		WARN("Could not access latest statistics from sub-device %d.\n",
298 			 SUB_ID(sdev));
299 		if (timestamp != 0)
300 			WARN("Using latest snapshot taken before %"PRIu64" seconds.\n",
301 				 (rte_rdtsc() - timestamp) / rte_get_tsc_hz());
302 	}
303 	failsafe_stats_increment(&PRIV(sdev->fs_dev)->stats_accumulator,
304 			err ? &sdev->stats_snapshot.stats : &stats);
305 	memset(&sdev->stats_snapshot, 0, sizeof(sdev->stats_snapshot));
306 }
307 
308 static inline int
309 fs_rxtx_clean(struct sub_device *sdev)
310 {
311 	uint16_t i;
312 
313 	for (i = 0; i < ETH(sdev)->data->nb_rx_queues; i++)
314 		if (FS_ATOMIC_RX(sdev, i))
315 			return 0;
316 	for (i = 0; i < ETH(sdev)->data->nb_tx_queues; i++)
317 		if (FS_ATOMIC_TX(sdev, i))
318 			return 0;
319 	return 1;
320 }
321 
322 void
323 failsafe_dev_remove(struct rte_eth_dev *dev)
324 {
325 	struct sub_device *sdev;
326 	uint8_t i;
327 
328 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
329 		if (sdev->remove && fs_rxtx_clean(sdev)) {
330 			fs_dev_stats_save(sdev);
331 			fs_dev_remove(sdev);
332 		}
333 }
334 
335 int
336 failsafe_eth_dev_state_sync(struct rte_eth_dev *dev)
337 {
338 	struct sub_device *sdev;
339 	uint32_t inactive;
340 	int ret;
341 	uint8_t i;
342 
343 	if (PRIV(dev)->state < DEV_PARSED)
344 		return 0;
345 
346 	ret = failsafe_args_parse_subs(dev);
347 	if (ret)
348 		goto err_remove;
349 
350 	if (PRIV(dev)->state < DEV_PROBED)
351 		return 0;
352 	ret = failsafe_eal_init(dev);
353 	if (ret)
354 		goto err_remove;
355 	if (PRIV(dev)->state < DEV_ACTIVE)
356 		return 0;
357 	inactive = 0;
358 	FOREACH_SUBDEV(sdev, i, dev) {
359 		if (sdev->state == DEV_PROBED) {
360 			inactive |= UINT32_C(1) << i;
361 			ret = eth_dev_flow_isolate_set(dev, sdev);
362 			if (ret) {
363 				ERROR("Could not apply configuration to sub_device %d",
364 				      i);
365 				goto err_remove;
366 			}
367 		}
368 	}
369 	ret = dev->dev_ops->dev_configure(dev);
370 	if (ret)
371 		goto err_remove;
372 	FOREACH_SUBDEV(sdev, i, dev) {
373 		if (inactive & (UINT32_C(1) << i)) {
374 			ret = fs_eth_dev_conf_apply(dev, sdev);
375 			if (ret) {
376 				ERROR("Could not apply configuration to sub_device %d",
377 				      i);
378 				goto err_remove;
379 			}
380 		}
381 	}
382 	/*
383 	 * If new devices have been configured, check if
384 	 * the link state has changed.
385 	 */
386 	if (inactive)
387 		dev->dev_ops->link_update(dev, 1);
388 	if (PRIV(dev)->state < DEV_STARTED)
389 		return 0;
390 	ret = dev->dev_ops->dev_start(dev);
391 	if (ret)
392 		goto err_remove;
393 	return 0;
394 err_remove:
395 	FOREACH_SUBDEV(sdev, i, dev)
396 		if (sdev->state != PRIV(dev)->state)
397 			sdev->remove = 1;
398 	return ret;
399 }
400 
401 void
402 failsafe_stats_increment(struct rte_eth_stats *to, struct rte_eth_stats *from)
403 {
404 	uint32_t i;
405 
406 	RTE_ASSERT(to != NULL && from != NULL);
407 	to->ipackets += from->ipackets;
408 	to->opackets += from->opackets;
409 	to->ibytes += from->ibytes;
410 	to->obytes += from->obytes;
411 	to->imissed += from->imissed;
412 	to->ierrors += from->ierrors;
413 	to->oerrors += from->oerrors;
414 	to->rx_nombuf += from->rx_nombuf;
415 	for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS; i++) {
416 		to->q_ipackets[i] += from->q_ipackets[i];
417 		to->q_opackets[i] += from->q_opackets[i];
418 		to->q_ibytes[i] += from->q_ibytes[i];
419 		to->q_obytes[i] += from->q_obytes[i];
420 		to->q_errors[i] += from->q_errors[i];
421 	}
422 }
423 
424 int
425 failsafe_eth_rmv_event_callback(uint16_t port_id __rte_unused,
426 				enum rte_eth_event_type event __rte_unused,
427 				void *cb_arg, void *out __rte_unused)
428 {
429 	struct sub_device *sdev = cb_arg;
430 
431 	/* Switch as soon as possible tx_dev. */
432 	fs_switch_dev(sdev->fs_dev, sdev);
433 	/* Use safe bursts in any case. */
434 	set_burst_fn(sdev->fs_dev, 1);
435 	/*
436 	 * Async removal, the sub-PMD will try to unregister
437 	 * the callback at the source of the current thread context.
438 	 */
439 	sdev->remove = 1;
440 	return 0;
441 }
442 
443 int
444 failsafe_eth_lsc_event_callback(uint16_t port_id __rte_unused,
445 				enum rte_eth_event_type event __rte_unused,
446 				void *cb_arg, void *out __rte_unused)
447 {
448 	struct rte_eth_dev *dev = cb_arg;
449 	int ret;
450 
451 	ret = dev->dev_ops->link_update(dev, 0);
452 	/* We must pass on the LSC event */
453 	if (ret)
454 		return _rte_eth_dev_callback_process(dev,
455 						     RTE_ETH_EVENT_INTR_LSC,
456 						     NULL);
457 	else
458 		return 0;
459 }
460