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 ret = rte_eth_promiscuous_enable(PORT_ID(sdev));
131 ret = rte_eth_promiscuous_disable(PORT_ID(sdev));
133 ERROR("Failed to apply promiscuous mode");
137 DEBUG("promiscuous already set");
140 if (dev->data->all_multicast != edev->data->all_multicast) {
141 DEBUG("Configuring all_multicast");
142 if (dev->data->all_multicast)
143 ret = rte_eth_allmulticast_enable(PORT_ID(sdev));
145 ret = rte_eth_allmulticast_disable(PORT_ID(sdev));
147 ERROR("Failed to apply allmulticast mode");
151 DEBUG("all_multicast already set");
154 if (dev->data->mtu != edev->data->mtu) {
155 DEBUG("Configuring MTU");
156 ret = rte_eth_dev_set_mtu(PORT_ID(sdev), dev->data->mtu);
158 ERROR("Failed to apply MTU");
162 DEBUG("MTU already set");
165 DEBUG("Configuring default MAC address");
166 ret = rte_eth_dev_default_mac_addr_set(PORT_ID(sdev),
167 &dev->data->mac_addrs[0]);
169 ERROR("Setting default MAC address failed");
173 if (PRIV(dev)->nb_mac_addr > 1)
174 DEBUG("Configure additional MAC address%s",
175 (PRIV(dev)->nb_mac_addr > 2 ? "es" : ""));
176 for (i = 1; i < PRIV(dev)->nb_mac_addr; i++) {
177 struct rte_ether_addr *ea;
179 ea = &dev->data->mac_addrs[i];
180 ret = rte_eth_dev_mac_addr_add(PORT_ID(sdev), ea,
181 PRIV(dev)->mac_addr_pool[i]);
183 char ea_fmt[RTE_ETHER_ADDR_FMT_SIZE];
185 rte_ether_format_addr(ea_fmt,
186 RTE_ETHER_ADDR_FMT_SIZE, ea);
187 ERROR("Adding MAC address %s failed", ea_fmt);
192 * Propagate multicast MAC addresses to sub-devices,
193 * if non zero number of addresses is set.
194 * The condition is required to avoid breakage of failsafe
195 * for sub-devices which do not support the operation
196 * if the feature is really not used.
198 if (PRIV(dev)->nb_mcast_addr > 0) {
199 DEBUG("Configuring multicast MAC addresses");
200 ret = rte_eth_dev_set_mc_addr_list(PORT_ID(sdev),
201 PRIV(dev)->mcast_addrs,
202 PRIV(dev)->nb_mcast_addr);
204 ERROR("Failed to apply multicast MAC addresses");
209 vfc1 = &dev->data->vlan_filter_conf;
210 vfc2 = &edev->data->vlan_filter_conf;
211 if (memcmp(vfc1, vfc2, sizeof(struct rte_vlan_filter_conf))) {
217 DEBUG("Configuring VLAN filter");
218 for (i = 0; i < RTE_DIM(vfc1->ids); i++) {
219 if (vfc1->ids[i] == 0)
224 /* count trailing zeroes */
225 vbit = ~ids & (ids - 1);
226 /* clear least significant bit set */
227 ids ^= (ids ^ (ids - 1)) ^ vbit;
228 for (; vbit; vlan_id++)
230 ret = rte_eth_dev_vlan_filter(
231 PORT_ID(sdev), vlan_id, 1);
233 ERROR("Failed to apply VLAN filter %hu",
240 DEBUG("VLAN filter already set");
243 if (TAILQ_EMPTY(&PRIV(dev)->flow_list)) {
244 DEBUG("rte_flow already set");
246 DEBUG("Resetting rte_flow configuration");
247 ret = rte_flow_flush(PORT_ID(sdev), &ferror);
249 fs_flow_complain(&ferror);
254 DEBUG("Configuring rte_flow");
255 TAILQ_FOREACH(flow, &PRIV(dev)->flow_list, next) {
256 DEBUG("Creating flow #%" PRIu32, i++);
257 flow->flows[SUB_ID(sdev)] =
258 rte_flow_create(PORT_ID(sdev),
268 fs_flow_complain(&ferror);
276 fs_dev_remove(struct sub_device *sdev)
282 switch (sdev->state) {
284 failsafe_rx_intr_uninstall_subdevice(sdev);
285 rte_eth_dev_stop(PORT_ID(sdev));
286 sdev->state = DEV_ACTIVE;
289 failsafe_eth_dev_unregister_callbacks(sdev);
290 ret = rte_eth_dev_close(PORT_ID(sdev));
292 ERROR("Port close failed for sub-device %u",
295 sdev->state = DEV_PROBED;
298 ret = rte_dev_remove(sdev->dev);
300 ERROR("Bus detach failed for sub_device %u",
303 rte_eth_dev_release_port(ETH(sdev));
305 sdev->state = DEV_PARSED;
309 sdev->state = DEV_UNDEFINED;
310 sdev->sdev_port_id = RTE_MAX_ETHPORTS;
315 failsafe_hotplug_alarm_install(fs_dev(sdev));
319 fs_dev_stats_save(struct sub_device *sdev)
321 struct rte_eth_stats stats;
324 /* Attempt to read current stats. */
325 err = rte_eth_stats_get(PORT_ID(sdev), &stats);
327 uint64_t timestamp = sdev->stats_snapshot.timestamp;
329 WARN("Could not access latest statistics from sub-device %d.",
332 WARN("Using latest snapshot taken before %"PRIu64" seconds.",
333 (rte_rdtsc() - timestamp) / rte_get_tsc_hz());
335 failsafe_stats_increment
336 (&PRIV(fs_dev(sdev))->stats_accumulator,
337 err ? &sdev->stats_snapshot.stats : &stats);
338 memset(&sdev->stats_snapshot, 0, sizeof(sdev->stats_snapshot));
342 fs_rxtx_clean(struct sub_device *sdev)
346 for (i = 0; i < ETH(sdev)->data->nb_rx_queues; i++)
347 if (FS_ATOMIC_RX(sdev, i))
349 for (i = 0; i < ETH(sdev)->data->nb_tx_queues; i++)
350 if (FS_ATOMIC_TX(sdev, i))
356 failsafe_eth_dev_unregister_callbacks(struct sub_device *sdev)
362 if (sdev->rmv_callback) {
363 ret = rte_eth_dev_callback_unregister(PORT_ID(sdev),
364 RTE_ETH_EVENT_INTR_RMV,
365 failsafe_eth_rmv_event_callback,
368 WARN("Failed to unregister RMV callback for sub_device"
369 " %d", SUB_ID(sdev));
370 sdev->rmv_callback = 0;
372 if (sdev->lsc_callback) {
373 ret = rte_eth_dev_callback_unregister(PORT_ID(sdev),
374 RTE_ETH_EVENT_INTR_LSC,
375 failsafe_eth_lsc_event_callback,
378 WARN("Failed to unregister LSC callback for sub_device"
379 " %d", SUB_ID(sdev));
380 sdev->lsc_callback = 0;
385 failsafe_dev_remove(struct rte_eth_dev *dev)
387 struct sub_device *sdev;
390 FOREACH_SUBDEV(sdev, i, dev) {
394 /* Active devices must have finished their burst and
395 * their stats must be saved.
397 if (sdev->state >= DEV_ACTIVE &&
398 fs_rxtx_clean(sdev) == 0)
400 if (fs_lock(dev, 1) != 0)
402 if (sdev->state >= DEV_ACTIVE)
403 fs_dev_stats_save(sdev);
410 failsafe_eth_dev_rx_queues_sync(struct rte_eth_dev *dev)
416 for (i = 0; i < dev->data->nb_rx_queues; i++) {
417 rxq = dev->data->rx_queues[i];
419 if (rxq->info.conf.rx_deferred_start &&
420 dev->data->rx_queue_state[i] ==
421 RTE_ETH_QUEUE_STATE_STARTED) {
423 * The subdevice Rx queue does not launch on device
424 * start if deferred start flag is set. It needs to be
425 * started manually in case an appropriate failsafe Rx
426 * queue has been started earlier.
428 ret = dev->dev_ops->rx_queue_start(dev, i);
430 ERROR("Could not synchronize Rx queue %d", i);
433 } else if (dev->data->rx_queue_state[i] ==
434 RTE_ETH_QUEUE_STATE_STOPPED) {
436 * The subdevice Rx queue needs to be stopped manually
437 * in case an appropriate failsafe Rx queue has been
440 ret = dev->dev_ops->rx_queue_stop(dev, i);
442 ERROR("Could not synchronize Rx queue %d", i);
451 failsafe_eth_dev_tx_queues_sync(struct rte_eth_dev *dev)
457 for (i = 0; i < dev->data->nb_tx_queues; i++) {
458 txq = dev->data->tx_queues[i];
460 if (txq->info.conf.tx_deferred_start &&
461 dev->data->tx_queue_state[i] ==
462 RTE_ETH_QUEUE_STATE_STARTED) {
464 * The subdevice Tx queue does not launch on device
465 * start if deferred start flag is set. It needs to be
466 * started manually in case an appropriate failsafe Tx
467 * queue has been started earlier.
469 ret = dev->dev_ops->tx_queue_start(dev, i);
471 ERROR("Could not synchronize Tx queue %d", i);
474 } else if (dev->data->tx_queue_state[i] ==
475 RTE_ETH_QUEUE_STATE_STOPPED) {
477 * The subdevice Tx queue needs to be stopped manually
478 * in case an appropriate failsafe Tx queue has been
481 ret = dev->dev_ops->tx_queue_stop(dev, i);
483 ERROR("Could not synchronize Tx queue %d", i);
492 failsafe_eth_dev_state_sync(struct rte_eth_dev *dev)
494 struct sub_device *sdev;
499 if (PRIV(dev)->state < DEV_PARSED)
502 ret = failsafe_args_parse_subs(dev);
506 if (PRIV(dev)->state < DEV_PROBED)
508 ret = failsafe_eal_init(dev);
511 if (PRIV(dev)->state < DEV_ACTIVE)
514 FOREACH_SUBDEV(sdev, i, dev) {
515 if (sdev->state == DEV_PROBED) {
516 inactive |= UINT32_C(1) << i;
517 ret = eth_dev_flow_isolate_set(dev, sdev);
519 ERROR("Could not apply configuration to sub_device %d",
525 ret = dev->dev_ops->dev_configure(dev);
528 FOREACH_SUBDEV(sdev, i, dev) {
529 if (inactive & (UINT32_C(1) << i)) {
530 ret = fs_eth_dev_conf_apply(dev, sdev);
532 ERROR("Could not apply configuration to sub_device %d",
539 * If new devices have been configured, check if
540 * the link state has changed.
543 dev->dev_ops->link_update(dev, 1);
544 if (PRIV(dev)->state < DEV_STARTED)
546 ret = dev->dev_ops->dev_start(dev);
549 ret = failsafe_eth_dev_rx_queues_sync(dev);
552 ret = failsafe_eth_dev_tx_queues_sync(dev);
557 FOREACH_SUBDEV(sdev, i, dev)
558 if (sdev->state != PRIV(dev)->state)
564 failsafe_stats_increment(struct rte_eth_stats *to, struct rte_eth_stats *from)
568 RTE_ASSERT(to != NULL && from != NULL);
569 to->ipackets += from->ipackets;
570 to->opackets += from->opackets;
571 to->ibytes += from->ibytes;
572 to->obytes += from->obytes;
573 to->imissed += from->imissed;
574 to->ierrors += from->ierrors;
575 to->oerrors += from->oerrors;
576 to->rx_nombuf += from->rx_nombuf;
577 for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS; i++) {
578 to->q_ipackets[i] += from->q_ipackets[i];
579 to->q_opackets[i] += from->q_opackets[i];
580 to->q_ibytes[i] += from->q_ibytes[i];
581 to->q_obytes[i] += from->q_obytes[i];
582 to->q_errors[i] += from->q_errors[i];
587 failsafe_eth_rmv_event_callback(uint16_t port_id __rte_unused,
588 enum rte_eth_event_type event __rte_unused,
589 void *cb_arg, void *out __rte_unused)
591 struct sub_device *sdev = cb_arg;
593 fs_lock(fs_dev(sdev), 0);
594 /* Switch as soon as possible tx_dev. */
595 fs_switch_dev(fs_dev(sdev), sdev);
596 /* Use safe bursts in any case. */
597 failsafe_set_burst_fn(fs_dev(sdev), 1);
599 * Async removal, the sub-PMD will try to unregister
600 * the callback at the source of the current thread context.
603 fs_unlock(fs_dev(sdev), 0);
608 failsafe_eth_lsc_event_callback(uint16_t port_id __rte_unused,
609 enum rte_eth_event_type event __rte_unused,
610 void *cb_arg, void *out __rte_unused)
612 struct rte_eth_dev *dev = cb_arg;
615 ret = dev->dev_ops->link_update(dev, 0);
616 /* We must pass on the LSC event */
618 return rte_eth_dev_callback_process(dev,
619 RTE_ETH_EVENT_INTR_LSC,
625 /* Take sub-device ownership before it becomes exposed to the application. */
627 failsafe_eth_new_event_callback(uint16_t port_id,
628 enum rte_eth_event_type event __rte_unused,
629 void *cb_arg, void *out __rte_unused)
631 struct rte_eth_dev *fs_dev = cb_arg;
632 struct sub_device *sdev;
633 struct rte_eth_dev *dev = &rte_eth_devices[port_id];
636 FOREACH_SUBDEV_STATE(sdev, i, fs_dev, DEV_PARSED) {
637 if (sdev->state >= DEV_PROBED)
639 if (dev->device == NULL) {
640 WARN("Trying to probe malformed device %s.\n",
644 if (strcmp(sdev->devargs.name, dev->device->name) != 0)
646 rte_eth_dev_owner_set(port_id, &PRIV(fs_dev)->my_owner);
647 /* The actual owner will be checked after the port probing. */