1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2017 6WIND S.A.
3 * Copyright 2017 Mellanox Technologies, Ltd
9 #include <rte_flow_driver.h>
10 #include <rte_cycles.h>
12 #include "failsafe_private.h"
14 /** Print a message out of a flow error. */
16 fs_flow_complain(struct rte_flow_error *error)
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",
36 if ((unsigned int)error->type >= RTE_DIM(errstrlist) ||
37 !errstrlist[error->type])
38 errstr = "unknown type";
40 errstr = errstrlist[error->type];
41 ERROR("Caught error type %d (%s): %s%s\n",
43 error->cause ? (snprintf(buf, sizeof(buf), "cause: %p, ",
44 error->cause), buf) : "",
45 error->message ? error->message : "(no stated reason)");
50 eth_dev_flow_isolate_set(struct rte_eth_dev *dev,
51 struct sub_device *sdev)
53 struct rte_flow_error ferror;
56 if (!PRIV(dev)->flow_isolated) {
57 DEBUG("Flow isolation already disabled");
59 DEBUG("Enabling flow isolation");
60 ret = rte_flow_isolate(PORT_ID(sdev),
61 PRIV(dev)->flow_isolated,
64 fs_flow_complain(&ferror);
72 fs_eth_dev_conf_apply(struct rte_eth_dev *dev,
73 struct sub_device *sdev)
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;
85 for (i = 0; i < dev->data->nb_rx_queues; i++) {
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);
93 ERROR("rx_queue_setup failed");
98 for (i = 0; i < dev->data->nb_tx_queues; i++) {
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,
106 ERROR("tx_queue_setup failed");
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));
117 ret = rte_eth_dev_set_link_down(PORT_ID(sdev));
119 ERROR("Failed to apply link_status");
123 DEBUG("link_status already set");
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));
131 rte_eth_promiscuous_disable(PORT_ID(sdev));
133 DEBUG("promiscuous already set");
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));
141 rte_eth_allmulticast_disable(PORT_ID(sdev));
143 DEBUG("all_multicast already set");
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);
150 ERROR("Failed to apply MTU");
154 DEBUG("MTU already set");
157 DEBUG("Configuring default MAC address");
158 ret = rte_eth_dev_default_mac_addr_set(PORT_ID(sdev),
159 &dev->data->mac_addrs[0]);
161 ERROR("Setting default MAC address failed");
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;
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]);
175 char ea_fmt[ETHER_ADDR_FMT_SIZE];
177 ether_format_addr(ea_fmt, ETHER_ADDR_FMT_SIZE, ea);
178 ERROR("Adding MAC address %s failed", ea_fmt);
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.
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);
195 ERROR("Failed to apply multicast MAC addresses");
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))) {
208 DEBUG("Configuring VLAN filter");
209 for (i = 0; i < RTE_DIM(vfc1->ids); i++) {
210 if (vfc1->ids[i] == 0)
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++)
221 ret = rte_eth_dev_vlan_filter(
222 PORT_ID(sdev), vlan_id, 1);
224 ERROR("Failed to apply VLAN filter %hu",
231 DEBUG("VLAN filter already set");
234 if (TAILQ_EMPTY(&PRIV(dev)->flow_list)) {
235 DEBUG("rte_flow already set");
237 DEBUG("Resetting rte_flow configuration");
238 ret = rte_flow_flush(PORT_ID(sdev), &ferror);
240 fs_flow_complain(&ferror);
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),
259 fs_flow_complain(&ferror);
267 fs_dev_remove(struct sub_device *sdev)
273 switch (sdev->state) {
275 failsafe_rx_intr_uninstall_subdevice(sdev);
276 rte_eth_dev_stop(PORT_ID(sdev));
277 sdev->state = DEV_ACTIVE;
280 failsafe_eth_dev_unregister_callbacks(sdev);
281 rte_eth_dev_close(PORT_ID(sdev));
282 sdev->state = DEV_PROBED;
285 ret = rte_dev_remove(sdev->dev);
287 ERROR("Bus detach failed for sub_device %u",
290 rte_eth_dev_release_port(ETH(sdev));
292 sdev->state = DEV_PARSED;
296 sdev->state = DEV_UNDEFINED;
301 failsafe_hotplug_alarm_install(sdev->fs_dev);
305 fs_dev_stats_save(struct sub_device *sdev)
307 struct rte_eth_stats stats;
310 /* Attempt to read current stats. */
311 err = rte_eth_stats_get(PORT_ID(sdev), &stats);
313 uint64_t timestamp = sdev->stats_snapshot.timestamp;
315 WARN("Could not access latest statistics from sub-device %d.\n",
318 WARN("Using latest snapshot taken before %"PRIu64" seconds.\n",
319 (rte_rdtsc() - timestamp) / rte_get_tsc_hz());
321 failsafe_stats_increment(&PRIV(sdev->fs_dev)->stats_accumulator,
322 err ? &sdev->stats_snapshot.stats : &stats);
323 memset(&sdev->stats_snapshot, 0, sizeof(sdev->stats_snapshot));
327 fs_rxtx_clean(struct sub_device *sdev)
331 for (i = 0; i < ETH(sdev)->data->nb_rx_queues; i++)
332 if (FS_ATOMIC_RX(sdev, i))
334 for (i = 0; i < ETH(sdev)->data->nb_tx_queues; i++)
335 if (FS_ATOMIC_TX(sdev, i))
341 failsafe_eth_dev_unregister_callbacks(struct sub_device *sdev)
347 if (sdev->rmv_callback) {
348 ret = rte_eth_dev_callback_unregister(PORT_ID(sdev),
349 RTE_ETH_EVENT_INTR_RMV,
350 failsafe_eth_rmv_event_callback,
353 WARN("Failed to unregister RMV callback for sub_device"
354 " %d", SUB_ID(sdev));
355 sdev->rmv_callback = 0;
357 if (sdev->lsc_callback) {
358 ret = rte_eth_dev_callback_unregister(PORT_ID(sdev),
359 RTE_ETH_EVENT_INTR_LSC,
360 failsafe_eth_lsc_event_callback,
363 WARN("Failed to unregister LSC callback for sub_device"
364 " %d", SUB_ID(sdev));
365 sdev->lsc_callback = 0;
370 failsafe_dev_remove(struct rte_eth_dev *dev)
372 struct sub_device *sdev;
375 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
376 if (sdev->remove && fs_rxtx_clean(sdev)) {
377 if (fs_lock(dev, 1) != 0)
379 fs_dev_stats_save(sdev);
386 failsafe_eth_dev_rx_queues_sync(struct rte_eth_dev *dev)
392 for (i = 0; i < dev->data->nb_rx_queues; i++) {
393 rxq = dev->data->rx_queues[i];
395 if (rxq->info.conf.rx_deferred_start &&
396 dev->data->rx_queue_state[i] ==
397 RTE_ETH_QUEUE_STATE_STARTED) {
399 * The subdevice Rx queue does not launch on device
400 * start if deferred start flag is set. It needs to be
401 * started manually in case an appropriate failsafe Rx
402 * queue has been started earlier.
404 ret = dev->dev_ops->rx_queue_start(dev, i);
406 ERROR("Could not synchronize Rx queue %d", i);
409 } else if (dev->data->rx_queue_state[i] ==
410 RTE_ETH_QUEUE_STATE_STOPPED) {
412 * The subdevice Rx queue needs to be stopped manually
413 * in case an appropriate failsafe Rx queue has been
416 ret = dev->dev_ops->rx_queue_stop(dev, i);
418 ERROR("Could not synchronize Rx queue %d", i);
427 failsafe_eth_dev_tx_queues_sync(struct rte_eth_dev *dev)
433 for (i = 0; i < dev->data->nb_tx_queues; i++) {
434 txq = dev->data->tx_queues[i];
436 if (txq->info.conf.tx_deferred_start &&
437 dev->data->tx_queue_state[i] ==
438 RTE_ETH_QUEUE_STATE_STARTED) {
440 * The subdevice Tx queue does not launch on device
441 * start if deferred start flag is set. It needs to be
442 * started manually in case an appropriate failsafe Tx
443 * queue has been started earlier.
445 ret = dev->dev_ops->tx_queue_start(dev, i);
447 ERROR("Could not synchronize Tx queue %d", i);
450 } else if (dev->data->tx_queue_state[i] ==
451 RTE_ETH_QUEUE_STATE_STOPPED) {
453 * The subdevice Tx queue needs to be stopped manually
454 * in case an appropriate failsafe Tx queue has been
457 ret = dev->dev_ops->tx_queue_stop(dev, i);
459 ERROR("Could not synchronize Tx queue %d", i);
468 failsafe_eth_dev_state_sync(struct rte_eth_dev *dev)
470 struct sub_device *sdev;
475 if (PRIV(dev)->state < DEV_PARSED)
478 ret = failsafe_args_parse_subs(dev);
482 if (PRIV(dev)->state < DEV_PROBED)
484 ret = failsafe_eal_init(dev);
487 if (PRIV(dev)->state < DEV_ACTIVE)
490 FOREACH_SUBDEV(sdev, i, dev) {
491 if (sdev->state == DEV_PROBED) {
492 inactive |= UINT32_C(1) << i;
493 ret = eth_dev_flow_isolate_set(dev, sdev);
495 ERROR("Could not apply configuration to sub_device %d",
501 ret = dev->dev_ops->dev_configure(dev);
504 FOREACH_SUBDEV(sdev, i, dev) {
505 if (inactive & (UINT32_C(1) << i)) {
506 ret = fs_eth_dev_conf_apply(dev, sdev);
508 ERROR("Could not apply configuration to sub_device %d",
515 * If new devices have been configured, check if
516 * the link state has changed.
519 dev->dev_ops->link_update(dev, 1);
520 if (PRIV(dev)->state < DEV_STARTED)
522 ret = dev->dev_ops->dev_start(dev);
525 ret = failsafe_eth_dev_rx_queues_sync(dev);
528 ret = failsafe_eth_dev_tx_queues_sync(dev);
533 FOREACH_SUBDEV(sdev, i, dev)
534 if (sdev->state != PRIV(dev)->state)
540 failsafe_stats_increment(struct rte_eth_stats *to, struct rte_eth_stats *from)
544 RTE_ASSERT(to != NULL && from != NULL);
545 to->ipackets += from->ipackets;
546 to->opackets += from->opackets;
547 to->ibytes += from->ibytes;
548 to->obytes += from->obytes;
549 to->imissed += from->imissed;
550 to->ierrors += from->ierrors;
551 to->oerrors += from->oerrors;
552 to->rx_nombuf += from->rx_nombuf;
553 for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS; i++) {
554 to->q_ipackets[i] += from->q_ipackets[i];
555 to->q_opackets[i] += from->q_opackets[i];
556 to->q_ibytes[i] += from->q_ibytes[i];
557 to->q_obytes[i] += from->q_obytes[i];
558 to->q_errors[i] += from->q_errors[i];
563 failsafe_eth_rmv_event_callback(uint16_t port_id __rte_unused,
564 enum rte_eth_event_type event __rte_unused,
565 void *cb_arg, void *out __rte_unused)
567 struct sub_device *sdev = cb_arg;
569 fs_lock(sdev->fs_dev, 0);
570 /* Switch as soon as possible tx_dev. */
571 fs_switch_dev(sdev->fs_dev, sdev);
572 /* Use safe bursts in any case. */
573 failsafe_set_burst_fn(sdev->fs_dev, 1);
575 * Async removal, the sub-PMD will try to unregister
576 * the callback at the source of the current thread context.
579 fs_unlock(sdev->fs_dev, 0);
584 failsafe_eth_lsc_event_callback(uint16_t port_id __rte_unused,
585 enum rte_eth_event_type event __rte_unused,
586 void *cb_arg, void *out __rte_unused)
588 struct rte_eth_dev *dev = cb_arg;
591 ret = dev->dev_ops->link_update(dev, 0);
592 /* We must pass on the LSC event */
594 return _rte_eth_dev_callback_process(dev,
595 RTE_ETH_EVENT_INTR_LSC,
601 /* Take sub-device ownership before it becomes exposed to the application. */
603 failsafe_eth_new_event_callback(uint16_t port_id,
604 enum rte_eth_event_type event __rte_unused,
605 void *cb_arg, void *out __rte_unused)
607 struct rte_eth_dev *fs_dev = cb_arg;
608 struct sub_device *sdev;
609 struct rte_eth_dev *dev = &rte_eth_devices[port_id];
612 FOREACH_SUBDEV_STATE(sdev, i, fs_dev, DEV_PARSED) {
613 if (sdev->state >= DEV_PROBED)
615 if (strcmp(sdev->devargs.name, dev->device->name) != 0)
617 rte_eth_dev_owner_set(port_id, &PRIV(fs_dev)->my_owner);
618 /* The actual owner will be checked after the port probing. */