1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2017 6WIND S.A.
3 * Copyright 2017 Mellanox Technologies, Ltd
10 #include <rte_debug.h>
11 #include <rte_atomic.h>
12 #include <ethdev_driver.h>
13 #include <rte_malloc.h>
15 #include <rte_cycles.h>
16 #include <rte_ethdev.h>
17 #include <rte_string_fns.h>
19 #include "failsafe_private.h"
22 fs_dev_configure(struct rte_eth_dev *dev)
24 struct sub_device *sdev;
29 FOREACH_SUBDEV(sdev, i, dev) {
30 int rmv_interrupt = 0;
31 int lsc_interrupt = 0;
34 if (sdev->state != DEV_PROBED &&
35 !(PRIV(dev)->alarm_lock == 0 && sdev->state == DEV_ACTIVE))
38 rmv_interrupt = ETH(sdev)->data->dev_flags &
41 DEBUG("Enabling RMV interrupts for sub_device %d", i);
42 dev->data->dev_conf.intr_conf.rmv = 1;
44 DEBUG("sub_device %d does not support RMV event", i);
46 lsc_enabled = dev->data->dev_conf.intr_conf.lsc;
47 lsc_interrupt = lsc_enabled &&
48 (ETH(sdev)->data->dev_flags &
49 RTE_ETH_DEV_INTR_LSC);
51 DEBUG("Enabling LSC interrupts for sub_device %d", i);
52 dev->data->dev_conf.intr_conf.lsc = 1;
53 } else if (lsc_enabled && !lsc_interrupt) {
54 DEBUG("Disabling LSC interrupts for sub_device %d", i);
55 dev->data->dev_conf.intr_conf.lsc = 0;
57 DEBUG("Configuring sub-device %d", i);
58 ret = rte_eth_dev_configure(PORT_ID(sdev),
59 dev->data->nb_rx_queues,
60 dev->data->nb_tx_queues,
61 &dev->data->dev_conf);
63 if (!fs_err(sdev, ret))
65 ERROR("Could not configure sub_device %d", i);
69 if (rmv_interrupt && sdev->rmv_callback == 0) {
70 ret = rte_eth_dev_callback_register(PORT_ID(sdev),
71 RTE_ETH_EVENT_INTR_RMV,
72 failsafe_eth_rmv_event_callback,
75 WARN("Failed to register RMV callback for sub_device %d",
78 sdev->rmv_callback = 1;
80 dev->data->dev_conf.intr_conf.rmv = 0;
81 if (lsc_interrupt && sdev->lsc_callback == 0) {
82 ret = rte_eth_dev_callback_register(PORT_ID(sdev),
83 RTE_ETH_EVENT_INTR_LSC,
84 failsafe_eth_lsc_event_callback,
87 WARN("Failed to register LSC callback for sub_device %d",
90 sdev->lsc_callback = 1;
92 dev->data->dev_conf.intr_conf.lsc = lsc_enabled;
93 sdev->state = DEV_ACTIVE;
95 if (PRIV(dev)->state < DEV_ACTIVE)
96 PRIV(dev)->state = DEV_ACTIVE;
102 fs_set_queues_state_start(struct rte_eth_dev *dev)
108 for (i = 0; i < dev->data->nb_rx_queues; i++) {
109 rxq = dev->data->rx_queues[i];
110 if (rxq != NULL && !rxq->info.conf.rx_deferred_start)
111 dev->data->rx_queue_state[i] =
112 RTE_ETH_QUEUE_STATE_STARTED;
114 for (i = 0; i < dev->data->nb_tx_queues; i++) {
115 txq = dev->data->tx_queues[i];
116 if (txq != NULL && !txq->info.conf.tx_deferred_start)
117 dev->data->tx_queue_state[i] =
118 RTE_ETH_QUEUE_STATE_STARTED;
123 fs_dev_start(struct rte_eth_dev *dev)
125 struct sub_device *sdev;
130 ret = failsafe_rx_intr_install(dev);
135 FOREACH_SUBDEV(sdev, i, dev) {
136 if (sdev->state != DEV_ACTIVE)
138 DEBUG("Starting sub_device %d", i);
139 ret = rte_eth_dev_start(PORT_ID(sdev));
141 if (!fs_err(sdev, ret))
146 ret = failsafe_rx_intr_install_subdevice(sdev);
148 if (!fs_err(sdev, ret))
150 if (fs_err(sdev, rte_eth_dev_stop(PORT_ID(sdev))) < 0)
151 ERROR("Failed to stop sub-device %u",
156 sdev->state = DEV_STARTED;
158 if (PRIV(dev)->state < DEV_STARTED) {
159 PRIV(dev)->state = DEV_STARTED;
160 fs_set_queues_state_start(dev);
162 fs_switch_dev(dev, NULL);
168 fs_set_queues_state_stop(struct rte_eth_dev *dev)
172 for (i = 0; i < dev->data->nb_rx_queues; i++)
173 if (dev->data->rx_queues[i] != NULL)
174 dev->data->rx_queue_state[i] =
175 RTE_ETH_QUEUE_STATE_STOPPED;
176 for (i = 0; i < dev->data->nb_tx_queues; i++)
177 if (dev->data->tx_queues[i] != NULL)
178 dev->data->tx_queue_state[i] =
179 RTE_ETH_QUEUE_STATE_STOPPED;
183 fs_dev_stop(struct rte_eth_dev *dev)
185 struct sub_device *sdev;
190 PRIV(dev)->state = DEV_STARTED - 1;
191 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_STARTED) {
192 ret = rte_eth_dev_stop(PORT_ID(sdev));
193 if (fs_err(sdev, ret) < 0) {
194 ERROR("Failed to stop device %u",
196 PRIV(dev)->state = DEV_STARTED + 1;
200 failsafe_rx_intr_uninstall_subdevice(sdev);
201 sdev->state = DEV_STARTED - 1;
203 failsafe_rx_intr_uninstall(dev);
204 fs_set_queues_state_stop(dev);
211 fs_dev_set_link_up(struct rte_eth_dev *dev)
213 struct sub_device *sdev;
218 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
219 DEBUG("Calling rte_eth_dev_set_link_up on sub_device %d", i);
220 ret = rte_eth_dev_set_link_up(PORT_ID(sdev));
221 if ((ret = fs_err(sdev, ret))) {
222 ERROR("Operation rte_eth_dev_set_link_up failed for sub_device %d"
223 " with error %d", i, ret);
233 fs_dev_set_link_down(struct rte_eth_dev *dev)
235 struct sub_device *sdev;
240 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
241 DEBUG("Calling rte_eth_dev_set_link_down on sub_device %d", i);
242 ret = rte_eth_dev_set_link_down(PORT_ID(sdev));
243 if ((ret = fs_err(sdev, ret))) {
244 ERROR("Operation rte_eth_dev_set_link_down failed for sub_device %d"
245 " with error %d", i, ret);
255 fs_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
257 struct sub_device *sdev;
264 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
265 uint16_t port_id = ETH(sdev)->data->port_id;
267 ret = rte_eth_dev_rx_queue_stop(port_id, rx_queue_id);
268 ret = fs_err(sdev, ret);
270 ERROR("Rx queue stop failed for subdevice %d", i);
276 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
278 /* Return 0 in case of at least one successful queue stop */
279 return (failure) ? err : 0;
283 fs_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
285 struct sub_device *sdev;
290 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
291 uint16_t port_id = ETH(sdev)->data->port_id;
293 ret = rte_eth_dev_rx_queue_start(port_id, rx_queue_id);
294 ret = fs_err(sdev, ret);
296 ERROR("Rx queue start failed for subdevice %d", i);
297 fs_rx_queue_stop(dev, rx_queue_id);
302 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
308 fs_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
310 struct sub_device *sdev;
317 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
318 uint16_t port_id = ETH(sdev)->data->port_id;
320 ret = rte_eth_dev_tx_queue_stop(port_id, tx_queue_id);
321 ret = fs_err(sdev, ret);
323 ERROR("Tx queue stop failed for subdevice %d", i);
329 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
331 /* Return 0 in case of at least one successful queue stop */
332 return (failure) ? err : 0;
336 fs_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
338 struct sub_device *sdev;
343 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
344 uint16_t port_id = ETH(sdev)->data->port_id;
346 ret = rte_eth_dev_tx_queue_start(port_id, tx_queue_id);
347 ret = fs_err(sdev, ret);
349 ERROR("Tx queue start failed for subdevice %d", i);
350 fs_tx_queue_stop(dev, tx_queue_id);
355 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
361 fs_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
363 struct sub_device *sdev;
365 struct rxq *rxq = dev->data->rx_queues[qid];
370 if (rxq->event_fd >= 0)
371 close(rxq->event_fd);
372 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
373 if (ETH(sdev)->data->rx_queues != NULL &&
374 ETH(sdev)->data->rx_queues[rxq->qid] != NULL)
375 SUBOPS(sdev, rx_queue_release)(ETH(sdev), rxq->qid);
377 dev->data->rx_queues[rxq->qid] = NULL;
383 fs_rx_queue_setup(struct rte_eth_dev *dev,
384 uint16_t rx_queue_id,
386 unsigned int socket_id,
387 const struct rte_eth_rxconf *rx_conf,
388 struct rte_mempool *mb_pool)
391 * FIXME: Add a proper interface in rte_eal_interrupts for
392 * allocating eventfd as an interrupt vector.
393 * For the time being, fake as if we are using MSIX interrupts,
394 * this will cause rte_intr_efd_enable to allocate an eventfd for us.
396 struct rte_intr_handle intr_handle = {
397 .type = RTE_INTR_HANDLE_VFIO_MSIX,
400 struct sub_device *sdev;
406 if (rx_conf->rx_deferred_start) {
407 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
408 if (SUBOPS(sdev, rx_queue_start) == NULL) {
409 ERROR("Rx queue deferred start is not "
410 "supported for subdevice %d", i);
416 rxq = dev->data->rx_queues[rx_queue_id];
418 fs_rx_queue_release(dev, rx_queue_id);
419 dev->data->rx_queues[rx_queue_id] = NULL;
421 rxq = rte_zmalloc(NULL,
423 sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
424 RTE_CACHE_LINE_SIZE);
429 FOREACH_SUBDEV(sdev, i, dev)
430 rte_atomic64_init(&rxq->refcnt[i]);
431 rxq->qid = rx_queue_id;
432 rxq->socket_id = socket_id;
433 rxq->info.mp = mb_pool;
434 rxq->info.conf = *rx_conf;
435 rxq->info.nb_desc = nb_rx_desc;
436 rxq->priv = PRIV(dev);
437 rxq->sdev = PRIV(dev)->subs;
438 ret = rte_intr_efd_enable(&intr_handle, 1);
443 rxq->event_fd = intr_handle.efds[0];
444 dev->data->rx_queues[rx_queue_id] = rxq;
445 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
446 ret = rte_eth_rx_queue_setup(PORT_ID(sdev),
448 nb_rx_desc, socket_id,
450 if ((ret = fs_err(sdev, ret))) {
451 ERROR("RX queue setup failed for sub_device %d", i);
458 fs_rx_queue_release(dev, rx_queue_id);
464 fs_rx_intr_enable(struct rte_eth_dev *dev, uint16_t idx)
467 struct sub_device *sdev;
473 if (idx >= dev->data->nb_rx_queues) {
477 rxq = dev->data->rx_queues[idx];
478 if (rxq == NULL || rxq->event_fd <= 0) {
482 /* Fail if proxy service is nor running. */
483 if (PRIV(dev)->rxp.sstate != SS_RUNNING) {
484 ERROR("failsafe interrupt services are not running");
488 rxq->enable_events = 1;
489 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
490 ret = rte_eth_dev_rx_intr_enable(PORT_ID(sdev), idx);
491 ret = fs_err(sdev, ret);
503 fs_rx_intr_disable(struct rte_eth_dev *dev, uint16_t idx)
506 struct sub_device *sdev;
513 if (idx >= dev->data->nb_rx_queues) {
517 rxq = dev->data->rx_queues[idx];
518 if (rxq == NULL || rxq->event_fd <= 0) {
522 rxq->enable_events = 0;
523 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
524 ret = rte_eth_dev_rx_intr_disable(PORT_ID(sdev), idx);
525 ret = fs_err(sdev, ret);
529 /* Clear pending events */
530 while (read(rxq->event_fd, &u64, sizeof(uint64_t)) > 0)
540 fs_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
542 struct sub_device *sdev;
544 struct txq *txq = dev->data->tx_queues[qid];
549 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
550 if (ETH(sdev)->data->tx_queues != NULL &&
551 ETH(sdev)->data->tx_queues[txq->qid] != NULL)
552 SUBOPS(sdev, tx_queue_release)(ETH(sdev), txq->qid);
554 dev->data->tx_queues[txq->qid] = NULL;
560 fs_tx_queue_setup(struct rte_eth_dev *dev,
561 uint16_t tx_queue_id,
563 unsigned int socket_id,
564 const struct rte_eth_txconf *tx_conf)
566 struct sub_device *sdev;
572 if (tx_conf->tx_deferred_start) {
573 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
574 if (SUBOPS(sdev, tx_queue_start) == NULL) {
575 ERROR("Tx queue deferred start is not "
576 "supported for subdevice %d", i);
582 txq = dev->data->tx_queues[tx_queue_id];
584 fs_tx_queue_release(dev, tx_queue_id);
585 dev->data->tx_queues[tx_queue_id] = NULL;
587 txq = rte_zmalloc("ethdev TX queue",
589 sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
590 RTE_CACHE_LINE_SIZE);
595 FOREACH_SUBDEV(sdev, i, dev)
596 rte_atomic64_init(&txq->refcnt[i]);
597 txq->qid = tx_queue_id;
598 txq->socket_id = socket_id;
599 txq->info.conf = *tx_conf;
600 txq->info.nb_desc = nb_tx_desc;
601 txq->priv = PRIV(dev);
602 dev->data->tx_queues[tx_queue_id] = txq;
603 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
604 ret = rte_eth_tx_queue_setup(PORT_ID(sdev),
606 nb_tx_desc, socket_id,
608 if ((ret = fs_err(sdev, ret))) {
609 ERROR("TX queue setup failed for sub_device %d", i);
616 fs_tx_queue_release(dev, tx_queue_id);
622 fs_dev_free_queues(struct rte_eth_dev *dev)
626 for (i = 0; i < dev->data->nb_rx_queues; i++) {
627 fs_rx_queue_release(dev, i);
628 dev->data->rx_queues[i] = NULL;
630 dev->data->nb_rx_queues = 0;
631 for (i = 0; i < dev->data->nb_tx_queues; i++) {
632 fs_tx_queue_release(dev, i);
633 dev->data->tx_queues[i] = NULL;
635 dev->data->nb_tx_queues = 0;
639 failsafe_eth_dev_close(struct rte_eth_dev *dev)
641 struct sub_device *sdev;
646 failsafe_hotplug_alarm_cancel(dev);
647 if (PRIV(dev)->state == DEV_STARTED) {
648 ret = dev->dev_ops->dev_stop(dev);
654 PRIV(dev)->state = DEV_ACTIVE - 1;
655 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
656 DEBUG("Closing sub_device %d", i);
657 failsafe_eth_dev_unregister_callbacks(sdev);
658 err = rte_eth_dev_close(PORT_ID(sdev));
660 ret = ret ? ret : err;
661 ERROR("Error while closing sub-device %u",
664 sdev->state = DEV_ACTIVE - 1;
666 rte_eth_dev_callback_unregister(RTE_ETH_ALL, RTE_ETH_EVENT_NEW,
667 failsafe_eth_new_event_callback, dev);
668 if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
672 fs_dev_free_queues(dev);
673 err = failsafe_eal_uninit(dev);
675 ret = ret ? ret : err;
676 ERROR("Error while uninitializing sub-EAL");
678 failsafe_args_free(dev);
679 rte_free(PRIV(dev)->subs);
680 rte_free(PRIV(dev)->mcast_addrs);
681 /* mac_addrs must not be freed alone because part of dev_private */
682 dev->data->mac_addrs = NULL;
684 err = pthread_mutex_destroy(&PRIV(dev)->hotplug_mutex);
686 ret = ret ? ret : err;
687 ERROR("Error while destroying hotplug mutex");
693 fs_promiscuous_enable(struct rte_eth_dev *dev)
695 struct sub_device *sdev;
700 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
701 ret = rte_eth_promiscuous_enable(PORT_ID(sdev));
702 ret = fs_err(sdev, ret);
704 ERROR("Promiscuous mode enable failed for subdevice %d",
710 /* Rollback in the case of failure */
711 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
712 ret = rte_eth_promiscuous_disable(PORT_ID(sdev));
713 ret = fs_err(sdev, ret);
715 ERROR("Promiscuous mode disable during rollback failed for subdevice %d",
725 fs_promiscuous_disable(struct rte_eth_dev *dev)
727 struct sub_device *sdev;
732 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
733 ret = rte_eth_promiscuous_disable(PORT_ID(sdev));
734 ret = fs_err(sdev, ret);
736 ERROR("Promiscuous mode disable failed for subdevice %d",
742 /* Rollback in the case of failure */
743 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
744 ret = rte_eth_promiscuous_enable(PORT_ID(sdev));
745 ret = fs_err(sdev, ret);
747 ERROR("Promiscuous mode enable during rollback failed for subdevice %d",
757 fs_allmulticast_enable(struct rte_eth_dev *dev)
759 struct sub_device *sdev;
764 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
765 ret = rte_eth_allmulticast_enable(PORT_ID(sdev));
766 ret = fs_err(sdev, ret);
768 ERROR("All-multicast mode enable failed for subdevice %d",
774 /* Rollback in the case of failure */
775 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
776 ret = rte_eth_allmulticast_disable(PORT_ID(sdev));
777 ret = fs_err(sdev, ret);
779 ERROR("All-multicast mode disable during rollback failed for subdevice %d",
789 fs_allmulticast_disable(struct rte_eth_dev *dev)
791 struct sub_device *sdev;
796 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
797 ret = rte_eth_allmulticast_disable(PORT_ID(sdev));
798 ret = fs_err(sdev, ret);
800 ERROR("All-multicast mode disable failed for subdevice %d",
806 /* Rollback in the case of failure */
807 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
808 ret = rte_eth_allmulticast_enable(PORT_ID(sdev));
809 ret = fs_err(sdev, ret);
811 ERROR("All-multicast mode enable during rollback failed for subdevice %d",
821 fs_link_update(struct rte_eth_dev *dev,
822 int wait_to_complete)
824 struct sub_device *sdev;
829 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
830 DEBUG("Calling link_update on sub_device %d", i);
831 ret = (SUBOPS(sdev, link_update))(ETH(sdev), wait_to_complete);
832 if (ret && ret != -1 && sdev->remove == 0 &&
833 rte_eth_dev_is_removed(PORT_ID(sdev)) == 0) {
834 ERROR("Link update failed for sub_device %d with error %d",
840 if (TX_SUBDEV(dev)) {
841 struct rte_eth_link *l1;
842 struct rte_eth_link *l2;
844 l1 = &dev->data->dev_link;
845 l2 = Ð(TX_SUBDEV(dev))->data->dev_link;
846 if (memcmp(l1, l2, sizeof(*l1))) {
857 fs_stats_get(struct rte_eth_dev *dev,
858 struct rte_eth_stats *stats)
860 struct rte_eth_stats backup;
861 struct sub_device *sdev;
866 rte_memcpy(stats, &PRIV(dev)->stats_accumulator, sizeof(*stats));
867 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
868 struct rte_eth_stats *snapshot = &sdev->stats_snapshot.stats;
869 uint64_t *timestamp = &sdev->stats_snapshot.timestamp;
871 rte_memcpy(&backup, snapshot, sizeof(backup));
872 ret = rte_eth_stats_get(PORT_ID(sdev), snapshot);
874 if (!fs_err(sdev, ret)) {
875 rte_memcpy(snapshot, &backup, sizeof(backup));
878 ERROR("Operation rte_eth_stats_get failed for sub_device %d with error %d",
884 *timestamp = rte_rdtsc();
886 failsafe_stats_increment(stats, snapshot);
893 fs_stats_reset(struct rte_eth_dev *dev)
895 struct sub_device *sdev;
900 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
901 ret = rte_eth_stats_reset(PORT_ID(sdev));
903 if (!fs_err(sdev, ret))
906 ERROR("Operation rte_eth_stats_reset failed for sub_device %d with error %d",
911 memset(&sdev->stats_snapshot, 0, sizeof(struct rte_eth_stats));
913 memset(&PRIV(dev)->stats_accumulator, 0, sizeof(struct rte_eth_stats));
920 __fs_xstats_count(struct rte_eth_dev *dev)
922 struct sub_device *sdev;
927 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
928 ret = rte_eth_xstats_get_names(PORT_ID(sdev), NULL, 0);
938 __fs_xstats_get_names(struct rte_eth_dev *dev,
939 struct rte_eth_xstat_name *xstats_names,
942 struct sub_device *sdev;
943 unsigned int count = 0;
946 /* Caller only cares about count */
948 return __fs_xstats_count(dev);
950 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
951 struct rte_eth_xstat_name *sub_names = xstats_names + count;
957 r = rte_eth_xstats_get_names(PORT_ID(sdev),
958 sub_names, limit - count);
962 /* add subN_ prefix to names */
963 for (j = 0; j < r; j++) {
964 char *xname = sub_names[j].name;
965 char tmp[RTE_ETH_XSTATS_NAME_SIZE];
967 if ((xname[0] == 't' || xname[0] == 'r') &&
968 xname[1] == 'x' && xname[2] == '_')
969 snprintf(tmp, sizeof(tmp), "%.3ssub%u_%s",
970 xname, i, xname + 3);
972 snprintf(tmp, sizeof(tmp), "sub%u_%s",
975 strlcpy(xname, tmp, RTE_ETH_XSTATS_NAME_SIZE);
983 fs_xstats_get_names(struct rte_eth_dev *dev,
984 struct rte_eth_xstat_name *xstats_names,
990 ret = __fs_xstats_get_names(dev, xstats_names, limit);
996 __fs_xstats_get(struct rte_eth_dev *dev,
997 struct rte_eth_xstat *xstats,
1000 unsigned int count = 0;
1001 struct sub_device *sdev;
1005 ret = __fs_xstats_count(dev);
1008 * or caller did not give enough space
1011 if (ret < 0 || ret > (int)n || xstats == NULL)
1014 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1015 ret = rte_eth_xstats_get(PORT_ID(sdev), xstats, n);
1022 /* add offset to id's from sub-device */
1023 for (j = 0; j < ret; j++)
1024 xstats[j].id += count;
1035 fs_xstats_get(struct rte_eth_dev *dev,
1036 struct rte_eth_xstat *xstats,
1042 ret = __fs_xstats_get(dev, xstats, n);
1050 fs_xstats_reset(struct rte_eth_dev *dev)
1052 struct sub_device *sdev;
1057 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1058 r = rte_eth_xstats_reset(PORT_ID(sdev));
1068 fs_dev_merge_desc_lim(struct rte_eth_desc_lim *to,
1069 const struct rte_eth_desc_lim *from)
1071 to->nb_max = RTE_MIN(to->nb_max, from->nb_max);
1072 to->nb_min = RTE_MAX(to->nb_min, from->nb_min);
1073 to->nb_align = RTE_MAX(to->nb_align, from->nb_align);
1075 to->nb_seg_max = RTE_MIN(to->nb_seg_max, from->nb_seg_max);
1076 to->nb_mtu_seg_max = RTE_MIN(to->nb_mtu_seg_max, from->nb_mtu_seg_max);
1080 * Merge the information from sub-devices.
1082 * The reported values must be the common subset of all sub devices
1085 fs_dev_merge_info(struct rte_eth_dev_info *info,
1086 const struct rte_eth_dev_info *sinfo)
1088 info->min_mtu = RTE_MAX(info->min_mtu, sinfo->min_mtu);
1089 info->max_mtu = RTE_MIN(info->max_mtu, sinfo->max_mtu);
1090 info->max_rx_pktlen = RTE_MIN(info->max_rx_pktlen, sinfo->max_rx_pktlen);
1091 info->max_rx_queues = RTE_MIN(info->max_rx_queues, sinfo->max_rx_queues);
1092 info->max_tx_queues = RTE_MIN(info->max_tx_queues, sinfo->max_tx_queues);
1093 info->max_mac_addrs = RTE_MIN(info->max_mac_addrs, sinfo->max_mac_addrs);
1094 info->max_hash_mac_addrs = RTE_MIN(info->max_hash_mac_addrs,
1095 sinfo->max_hash_mac_addrs);
1096 info->max_vmdq_pools = RTE_MIN(info->max_vmdq_pools, sinfo->max_vmdq_pools);
1097 info->max_vfs = RTE_MIN(info->max_vfs, sinfo->max_vfs);
1099 fs_dev_merge_desc_lim(&info->rx_desc_lim, &sinfo->rx_desc_lim);
1100 fs_dev_merge_desc_lim(&info->tx_desc_lim, &sinfo->tx_desc_lim);
1102 info->rx_offload_capa &= sinfo->rx_offload_capa;
1103 info->tx_offload_capa &= sinfo->tx_offload_capa;
1104 info->rx_queue_offload_capa &= sinfo->rx_queue_offload_capa;
1105 info->tx_queue_offload_capa &= sinfo->tx_queue_offload_capa;
1106 info->flow_type_rss_offloads &= sinfo->flow_type_rss_offloads;
1109 * RETA size is a GCD of RETA sizes indicated by sub-devices.
1110 * Each of these sizes is a power of 2, so use the lower one.
1112 info->reta_size = RTE_MIN(info->reta_size, sinfo->reta_size);
1114 info->hash_key_size = RTE_MIN(info->hash_key_size,
1115 sinfo->hash_key_size);
1119 * Fail-safe dev_infos_get rules:
1123 * Use the maximum possible values for any field, so as not
1124 * to impede any further configuration effort.
1126 * Limits capabilities to those that are understood by the
1127 * fail-safe PMD. This understanding stems from the fail-safe
1128 * being capable of verifying that the related capability is
1129 * expressed within the device configuration (struct rte_eth_conf).
1131 * At least one probed sub_device:
1133 * Uses values from the active probed sub_device
1134 * The rationale here is that if any sub_device is less capable
1135 * (for example concerning the number of queues) than the active
1136 * sub_device, then its subsequent configuration will fail.
1137 * It is impossible to foresee this failure when the failing sub_device
1138 * is supposed to be plugged-in later on, so the configuration process
1139 * is the single point of failure and error reporting.
1141 * Uses a logical AND of RX capabilities among
1142 * all sub_devices and the default capabilities.
1143 * Uses a logical AND of TX capabilities among
1144 * the active probed sub_device and the default capabilities.
1145 * Uses a logical AND of device capabilities among
1146 * all sub_devices and the default capabilities.
1150 fs_dev_infos_get(struct rte_eth_dev *dev,
1151 struct rte_eth_dev_info *infos)
1153 struct sub_device *sdev;
1157 /* Use maximum upper bounds by default */
1158 infos->min_mtu = RTE_ETHER_MIN_MTU;
1159 infos->max_mtu = UINT16_MAX;
1160 infos->max_rx_pktlen = UINT32_MAX;
1161 infos->max_rx_queues = RTE_MAX_QUEUES_PER_PORT;
1162 infos->max_tx_queues = RTE_MAX_QUEUES_PER_PORT;
1163 infos->max_mac_addrs = FAILSAFE_MAX_ETHADDR;
1164 infos->max_hash_mac_addrs = UINT32_MAX;
1165 infos->max_vfs = UINT16_MAX;
1166 infos->max_vmdq_pools = UINT16_MAX;
1167 infos->reta_size = UINT16_MAX;
1168 infos->hash_key_size = UINT8_MAX;
1171 * Set of capabilities that can be verified upon
1172 * configuring a sub-device.
1174 infos->rx_offload_capa =
1175 RTE_ETH_RX_OFFLOAD_VLAN_STRIP |
1176 RTE_ETH_RX_OFFLOAD_IPV4_CKSUM |
1177 RTE_ETH_RX_OFFLOAD_UDP_CKSUM |
1178 RTE_ETH_RX_OFFLOAD_TCP_CKSUM |
1179 RTE_ETH_RX_OFFLOAD_TCP_LRO |
1180 RTE_ETH_RX_OFFLOAD_QINQ_STRIP |
1181 RTE_ETH_RX_OFFLOAD_OUTER_IPV4_CKSUM |
1182 RTE_ETH_RX_OFFLOAD_MACSEC_STRIP |
1183 RTE_ETH_RX_OFFLOAD_HEADER_SPLIT |
1184 RTE_ETH_RX_OFFLOAD_VLAN_FILTER |
1185 RTE_ETH_RX_OFFLOAD_VLAN_EXTEND |
1186 RTE_ETH_RX_OFFLOAD_SCATTER |
1187 RTE_ETH_RX_OFFLOAD_TIMESTAMP |
1188 RTE_ETH_RX_OFFLOAD_SECURITY |
1189 RTE_ETH_RX_OFFLOAD_RSS_HASH;
1191 infos->rx_queue_offload_capa =
1192 RTE_ETH_RX_OFFLOAD_VLAN_STRIP |
1193 RTE_ETH_RX_OFFLOAD_IPV4_CKSUM |
1194 RTE_ETH_RX_OFFLOAD_UDP_CKSUM |
1195 RTE_ETH_RX_OFFLOAD_TCP_CKSUM |
1196 RTE_ETH_RX_OFFLOAD_TCP_LRO |
1197 RTE_ETH_RX_OFFLOAD_QINQ_STRIP |
1198 RTE_ETH_RX_OFFLOAD_OUTER_IPV4_CKSUM |
1199 RTE_ETH_RX_OFFLOAD_MACSEC_STRIP |
1200 RTE_ETH_RX_OFFLOAD_HEADER_SPLIT |
1201 RTE_ETH_RX_OFFLOAD_VLAN_FILTER |
1202 RTE_ETH_RX_OFFLOAD_VLAN_EXTEND |
1203 RTE_ETH_RX_OFFLOAD_SCATTER |
1204 RTE_ETH_RX_OFFLOAD_TIMESTAMP |
1205 RTE_ETH_RX_OFFLOAD_SECURITY |
1206 RTE_ETH_RX_OFFLOAD_RSS_HASH;
1208 infos->tx_offload_capa =
1209 RTE_ETH_TX_OFFLOAD_MULTI_SEGS |
1210 RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE |
1211 RTE_ETH_TX_OFFLOAD_IPV4_CKSUM |
1212 RTE_ETH_TX_OFFLOAD_UDP_CKSUM |
1213 RTE_ETH_TX_OFFLOAD_TCP_CKSUM |
1214 RTE_ETH_TX_OFFLOAD_TCP_TSO;
1216 infos->flow_type_rss_offloads =
1221 RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
1222 RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
1224 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
1225 struct rte_eth_dev_info sub_info;
1227 ret = rte_eth_dev_info_get(PORT_ID(sdev), &sub_info);
1228 ret = fs_err(sdev, ret);
1232 fs_dev_merge_info(infos, &sub_info);
1238 static const uint32_t *
1239 fs_dev_supported_ptypes_get(struct rte_eth_dev *dev)
1241 struct sub_device *sdev;
1242 struct rte_eth_dev *edev;
1243 const uint32_t *ret;
1246 sdev = TX_SUBDEV(dev);
1252 /* ENOTSUP: counts as no supported ptypes */
1253 if (SUBOPS(sdev, dev_supported_ptypes_get) == NULL) {
1258 * The API does not permit to do a clean AND of all ptypes,
1259 * It is also incomplete by design and we do not really care
1260 * to have a best possible value in this context.
1261 * We just return the ptypes of the device of highest
1262 * priority, usually the PREFERRED device.
1264 ret = SUBOPS(sdev, dev_supported_ptypes_get)(edev);
1271 fs_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1273 struct sub_device *sdev;
1278 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1279 DEBUG("Calling rte_eth_dev_set_mtu on sub_device %d", i);
1280 ret = rte_eth_dev_set_mtu(PORT_ID(sdev), mtu);
1281 if ((ret = fs_err(sdev, ret))) {
1282 ERROR("Operation rte_eth_dev_set_mtu failed for sub_device %d with error %d",
1293 fs_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1295 struct sub_device *sdev;
1300 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1301 DEBUG("Calling rte_eth_dev_vlan_filter on sub_device %d", i);
1302 ret = rte_eth_dev_vlan_filter(PORT_ID(sdev), vlan_id, on);
1303 if ((ret = fs_err(sdev, ret))) {
1304 ERROR("Operation rte_eth_dev_vlan_filter failed for sub_device %d"
1305 " with error %d", i, ret);
1315 fs_flow_ctrl_get(struct rte_eth_dev *dev,
1316 struct rte_eth_fc_conf *fc_conf)
1318 struct sub_device *sdev;
1322 sdev = TX_SUBDEV(dev);
1327 if (SUBOPS(sdev, flow_ctrl_get) == NULL) {
1331 ret = SUBOPS(sdev, flow_ctrl_get)(ETH(sdev), fc_conf);
1338 fs_flow_ctrl_set(struct rte_eth_dev *dev,
1339 struct rte_eth_fc_conf *fc_conf)
1341 struct sub_device *sdev;
1346 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1347 DEBUG("Calling rte_eth_dev_flow_ctrl_set on sub_device %d", i);
1348 ret = rte_eth_dev_flow_ctrl_set(PORT_ID(sdev), fc_conf);
1349 if ((ret = fs_err(sdev, ret))) {
1350 ERROR("Operation rte_eth_dev_flow_ctrl_set failed for sub_device %d"
1351 " with error %d", i, ret);
1361 fs_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1363 struct sub_device *sdev;
1367 /* No check: already done within the rte_eth_dev_mac_addr_remove
1368 * call for the fail-safe device.
1370 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
1371 rte_eth_dev_mac_addr_remove(PORT_ID(sdev),
1372 &dev->data->mac_addrs[index]);
1373 PRIV(dev)->mac_addr_pool[index] = 0;
1378 fs_mac_addr_add(struct rte_eth_dev *dev,
1379 struct rte_ether_addr *mac_addr,
1383 struct sub_device *sdev;
1387 RTE_ASSERT(index < FAILSAFE_MAX_ETHADDR);
1389 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1390 ret = rte_eth_dev_mac_addr_add(PORT_ID(sdev), mac_addr, vmdq);
1391 if ((ret = fs_err(sdev, ret))) {
1392 ERROR("Operation rte_eth_dev_mac_addr_add failed for sub_device %"
1393 PRIu8 " with error %d", i, ret);
1398 if (index >= PRIV(dev)->nb_mac_addr) {
1399 DEBUG("Growing mac_addrs array");
1400 PRIV(dev)->nb_mac_addr = index;
1402 PRIV(dev)->mac_addr_pool[index] = vmdq;
1408 fs_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1410 struct sub_device *sdev;
1415 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1416 ret = rte_eth_dev_default_mac_addr_set(PORT_ID(sdev), mac_addr);
1417 ret = fs_err(sdev, ret);
1419 ERROR("Operation rte_eth_dev_mac_addr_set failed for sub_device %d with error %d",
1431 fs_set_mc_addr_list(struct rte_eth_dev *dev,
1432 struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1434 struct sub_device *sdev;
1441 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1442 ret = rte_eth_dev_set_mc_addr_list(PORT_ID(sdev),
1443 mc_addr_set, nb_mc_addr);
1445 ERROR("Operation rte_eth_dev_set_mc_addr_list failed for sub_device %d with error %d",
1451 mcast_addrs = rte_realloc(PRIV(dev)->mcast_addrs,
1452 nb_mc_addr * sizeof(PRIV(dev)->mcast_addrs[0]), 0);
1453 if (mcast_addrs == NULL && nb_mc_addr > 0) {
1457 rte_memcpy(mcast_addrs, mc_addr_set,
1458 nb_mc_addr * sizeof(PRIV(dev)->mcast_addrs[0]));
1459 PRIV(dev)->nb_mcast_addr = nb_mc_addr;
1460 PRIV(dev)->mcast_addrs = mcast_addrs;
1466 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1467 int rc = rte_eth_dev_set_mc_addr_list(PORT_ID(sdev),
1468 PRIV(dev)->mcast_addrs, PRIV(dev)->nb_mcast_addr);
1470 ERROR("Multicast MAC address list rollback for sub_device %d failed with error %d",
1480 fs_rss_hash_update(struct rte_eth_dev *dev,
1481 struct rte_eth_rss_conf *rss_conf)
1483 struct sub_device *sdev;
1488 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1489 ret = rte_eth_dev_rss_hash_update(PORT_ID(sdev), rss_conf);
1490 ret = fs_err(sdev, ret);
1492 ERROR("Operation rte_eth_dev_rss_hash_update"
1493 " failed for sub_device %d with error %d",
1505 fs_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
1506 const struct rte_flow_ops **ops)
1508 *ops = &fs_flow_ops;
1512 const struct eth_dev_ops failsafe_ops = {
1513 .dev_configure = fs_dev_configure,
1514 .dev_start = fs_dev_start,
1515 .dev_stop = fs_dev_stop,
1516 .dev_set_link_down = fs_dev_set_link_down,
1517 .dev_set_link_up = fs_dev_set_link_up,
1518 .dev_close = failsafe_eth_dev_close,
1519 .promiscuous_enable = fs_promiscuous_enable,
1520 .promiscuous_disable = fs_promiscuous_disable,
1521 .allmulticast_enable = fs_allmulticast_enable,
1522 .allmulticast_disable = fs_allmulticast_disable,
1523 .link_update = fs_link_update,
1524 .stats_get = fs_stats_get,
1525 .stats_reset = fs_stats_reset,
1526 .xstats_get = fs_xstats_get,
1527 .xstats_get_names = fs_xstats_get_names,
1528 .xstats_reset = fs_xstats_reset,
1529 .dev_infos_get = fs_dev_infos_get,
1530 .dev_supported_ptypes_get = fs_dev_supported_ptypes_get,
1531 .mtu_set = fs_mtu_set,
1532 .vlan_filter_set = fs_vlan_filter_set,
1533 .rx_queue_start = fs_rx_queue_start,
1534 .rx_queue_stop = fs_rx_queue_stop,
1535 .tx_queue_start = fs_tx_queue_start,
1536 .tx_queue_stop = fs_tx_queue_stop,
1537 .rx_queue_setup = fs_rx_queue_setup,
1538 .tx_queue_setup = fs_tx_queue_setup,
1539 .rx_queue_release = fs_rx_queue_release,
1540 .tx_queue_release = fs_tx_queue_release,
1541 .rx_queue_intr_enable = fs_rx_intr_enable,
1542 .rx_queue_intr_disable = fs_rx_intr_disable,
1543 .flow_ctrl_get = fs_flow_ctrl_get,
1544 .flow_ctrl_set = fs_flow_ctrl_set,
1545 .mac_addr_remove = fs_mac_addr_remove,
1546 .mac_addr_add = fs_mac_addr_add,
1547 .mac_addr_set = fs_mac_addr_set,
1548 .set_mc_addr_list = fs_set_mc_addr_list,
1549 .rss_hash_update = fs_rss_hash_update,
1550 .flow_ops_get = fs_flow_ops_get,