1 /* SPDX-License-Identifier: BSD-3-Clause
3 * Copyright (c) 2016-2018 Solarflare Communications Inc.
6 * This software was jointly developed between OKTET Labs (under contract
7 * for Solarflare) and Solarflare Communications, Inc.
11 #include <rte_ethdev_driver.h>
12 #include <rte_ethdev_pci.h>
14 #include <rte_bus_pci.h>
15 #include <rte_errno.h>
16 #include <rte_string_fns.h>
21 #include "sfc_debug.h"
23 #include "sfc_kvargs.h"
29 #include "sfc_dp_rx.h"
31 uint32_t sfc_logtype_driver;
33 static struct sfc_dp_list sfc_dp_head =
34 TAILQ_HEAD_INITIALIZER(sfc_dp_head);
37 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
39 struct sfc_adapter *sa = dev->data->dev_private;
40 efx_nic_fw_info_t enfi;
45 * Return value of the callback is likely supposed to be
46 * equal to or greater than 0, nevertheless, if an error
47 * occurs, it will be desirable to pass it to the caller
49 if ((fw_version == NULL) || (fw_size == 0))
52 rc = efx_nic_get_fw_version(sa->nic, &enfi);
56 ret = snprintf(fw_version, fw_size,
57 "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
58 enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
59 enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
63 if (enfi.enfi_dpcpu_fw_ids_valid) {
64 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
67 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
68 fw_size - dpcpu_fw_ids_offset,
69 " rx%" PRIx16 " tx%" PRIx16,
70 enfi.enfi_rx_dpcpu_fw_id,
71 enfi.enfi_tx_dpcpu_fw_id);
78 if (fw_size < (size_t)(++ret))
85 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
87 struct sfc_adapter *sa = dev->data->dev_private;
88 struct sfc_rss *rss = &sa->rss;
89 uint64_t txq_offloads_def = 0;
91 sfc_log_init(sa, "entry");
93 dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
95 /* Autonegotiation may be disabled */
96 dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
97 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
98 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
99 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
100 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
101 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_25000FDX)
102 dev_info->speed_capa |= ETH_LINK_SPEED_25G;
103 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
104 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
105 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_50000FDX)
106 dev_info->speed_capa |= ETH_LINK_SPEED_50G;
107 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_100000FDX)
108 dev_info->speed_capa |= ETH_LINK_SPEED_100G;
110 dev_info->max_rx_queues = sa->rxq_max;
111 dev_info->max_tx_queues = sa->txq_max;
113 /* By default packets are dropped if no descriptors are available */
114 dev_info->default_rxconf.rx_drop_en = 1;
116 dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
119 * rx_offload_capa includes both device and queue offloads since
120 * the latter may be requested on a per device basis which makes
121 * sense when some offloads are needed to be set on all queues.
123 dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
124 dev_info->rx_queue_offload_capa;
126 dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
129 * tx_offload_capa includes both device and queue offloads since
130 * the latter may be requested on a per device basis which makes
131 * sense when some offloads are needed to be set on all queues.
133 dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
134 dev_info->tx_queue_offload_capa;
136 if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
137 txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
139 dev_info->default_txconf.offloads |= txq_offloads_def;
141 if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
145 for (i = 0; i < rss->hf_map_nb_entries; ++i)
146 rte_hf |= rss->hf_map[i].rte;
148 dev_info->reta_size = EFX_RSS_TBL_SIZE;
149 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
150 dev_info->flow_type_rss_offloads = rte_hf;
153 /* Initialize to hardware limits */
154 dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
155 dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
156 /* The RXQ hardware requires that the descriptor count is a power
157 * of 2, but rx_desc_lim cannot properly describe that constraint.
159 dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
161 /* Initialize to hardware limits */
162 dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
163 dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
165 * The TXQ hardware requires that the descriptor count is a power
166 * of 2, but tx_desc_lim cannot properly describe that constraint
168 dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
170 if (sa->dp_rx->get_dev_info != NULL)
171 sa->dp_rx->get_dev_info(dev_info);
172 if (sa->dp_tx->get_dev_info != NULL)
173 sa->dp_tx->get_dev_info(dev_info);
176 static const uint32_t *
177 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
179 struct sfc_adapter *sa = dev->data->dev_private;
180 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
181 uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
183 return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
187 sfc_dev_configure(struct rte_eth_dev *dev)
189 struct rte_eth_dev_data *dev_data = dev->data;
190 struct sfc_adapter *sa = dev_data->dev_private;
193 sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
194 dev_data->nb_rx_queues, dev_data->nb_tx_queues);
196 sfc_adapter_lock(sa);
198 case SFC_ADAPTER_CONFIGURED:
200 case SFC_ADAPTER_INITIALIZED:
201 rc = sfc_configure(sa);
204 sfc_err(sa, "unexpected adapter state %u to configure",
209 sfc_adapter_unlock(sa);
211 sfc_log_init(sa, "done %d", rc);
217 sfc_dev_start(struct rte_eth_dev *dev)
219 struct sfc_adapter *sa = dev->data->dev_private;
222 sfc_log_init(sa, "entry");
224 sfc_adapter_lock(sa);
226 sfc_adapter_unlock(sa);
228 sfc_log_init(sa, "done %d", rc);
234 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
236 struct sfc_adapter *sa = dev->data->dev_private;
237 struct rte_eth_link current_link;
240 sfc_log_init(sa, "entry");
242 if (sa->state != SFC_ADAPTER_STARTED) {
243 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, ¤t_link);
244 } else if (wait_to_complete) {
245 efx_link_mode_t link_mode;
247 if (efx_port_poll(sa->nic, &link_mode) != 0)
248 link_mode = EFX_LINK_UNKNOWN;
249 sfc_port_link_mode_to_info(link_mode, ¤t_link);
252 sfc_ev_mgmt_qpoll(sa);
253 rte_eth_linkstatus_get(dev, ¤t_link);
256 ret = rte_eth_linkstatus_set(dev, ¤t_link);
258 sfc_notice(sa, "Link status is %s",
259 current_link.link_status ? "UP" : "DOWN");
265 sfc_dev_stop(struct rte_eth_dev *dev)
267 struct sfc_adapter *sa = dev->data->dev_private;
269 sfc_log_init(sa, "entry");
271 sfc_adapter_lock(sa);
273 sfc_adapter_unlock(sa);
275 sfc_log_init(sa, "done");
279 sfc_dev_set_link_up(struct rte_eth_dev *dev)
281 struct sfc_adapter *sa = dev->data->dev_private;
284 sfc_log_init(sa, "entry");
286 sfc_adapter_lock(sa);
288 sfc_adapter_unlock(sa);
295 sfc_dev_set_link_down(struct rte_eth_dev *dev)
297 struct sfc_adapter *sa = dev->data->dev_private;
299 sfc_log_init(sa, "entry");
301 sfc_adapter_lock(sa);
303 sfc_adapter_unlock(sa);
309 sfc_dev_close(struct rte_eth_dev *dev)
311 struct sfc_adapter *sa = dev->data->dev_private;
313 sfc_log_init(sa, "entry");
315 sfc_adapter_lock(sa);
317 case SFC_ADAPTER_STARTED:
319 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
321 case SFC_ADAPTER_CONFIGURED:
323 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
325 case SFC_ADAPTER_INITIALIZED:
328 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
331 sfc_adapter_unlock(sa);
333 sfc_log_init(sa, "done");
337 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
340 struct sfc_port *port;
342 struct sfc_adapter *sa = dev->data->dev_private;
343 boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
344 const char *desc = (allmulti) ? "all-multi" : "promiscuous";
346 sfc_adapter_lock(sa);
349 toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
351 if (*toggle != enabled) {
354 if (port->isolated) {
355 sfc_warn(sa, "isolated mode is active on the port");
356 sfc_warn(sa, "the change is to be applied on the next "
357 "start provided that isolated mode is "
358 "disabled prior the next start");
359 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
360 (sfc_set_rx_mode(sa) != 0)) {
361 *toggle = !(enabled);
362 sfc_warn(sa, "Failed to %s %s mode",
363 ((enabled) ? "enable" : "disable"), desc);
367 sfc_adapter_unlock(sa);
371 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
373 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
377 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
379 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
383 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
385 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
389 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
391 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
395 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
396 uint16_t nb_rx_desc, unsigned int socket_id,
397 const struct rte_eth_rxconf *rx_conf,
398 struct rte_mempool *mb_pool)
400 struct sfc_adapter *sa = dev->data->dev_private;
403 sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
404 rx_queue_id, nb_rx_desc, socket_id);
406 sfc_adapter_lock(sa);
408 rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
413 dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
415 sfc_adapter_unlock(sa);
420 sfc_adapter_unlock(sa);
426 sfc_rx_queue_release(void *queue)
428 struct sfc_dp_rxq *dp_rxq = queue;
430 struct sfc_adapter *sa;
431 unsigned int sw_index;
436 rxq = sfc_rxq_by_dp_rxq(dp_rxq);
438 sfc_adapter_lock(sa);
440 sw_index = sfc_rxq_sw_index(rxq);
442 sfc_log_init(sa, "RxQ=%u", sw_index);
444 sa->eth_dev->data->rx_queues[sw_index] = NULL;
446 sfc_rx_qfini(sa, sw_index);
448 sfc_adapter_unlock(sa);
452 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
453 uint16_t nb_tx_desc, unsigned int socket_id,
454 const struct rte_eth_txconf *tx_conf)
456 struct sfc_adapter *sa = dev->data->dev_private;
459 sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
460 tx_queue_id, nb_tx_desc, socket_id);
462 sfc_adapter_lock(sa);
464 rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
468 dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
470 sfc_adapter_unlock(sa);
474 sfc_adapter_unlock(sa);
480 sfc_tx_queue_release(void *queue)
482 struct sfc_dp_txq *dp_txq = queue;
484 unsigned int sw_index;
485 struct sfc_adapter *sa;
490 txq = sfc_txq_by_dp_txq(dp_txq);
491 sw_index = sfc_txq_sw_index(txq);
493 SFC_ASSERT(txq->evq != NULL);
496 sfc_log_init(sa, "TxQ = %u", sw_index);
498 sfc_adapter_lock(sa);
500 SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
501 sa->eth_dev->data->tx_queues[sw_index] = NULL;
503 sfc_tx_qfini(sa, sw_index);
505 sfc_adapter_unlock(sa);
509 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
511 struct sfc_adapter *sa = dev->data->dev_private;
512 struct sfc_port *port = &sa->port;
516 rte_spinlock_lock(&port->mac_stats_lock);
518 ret = sfc_port_update_mac_stats(sa);
522 mac_stats = port->mac_stats_buf;
524 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
525 EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
527 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
528 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
529 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
531 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
532 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
533 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
535 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
536 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
537 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
539 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
540 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
541 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
542 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
543 stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
544 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
546 stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
547 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
548 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
549 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
551 * Take into account stats which are whenever supported
552 * on EF10. If some stat is not supported by current
553 * firmware variant or HW revision, it is guaranteed
554 * to be zero in mac_stats.
557 mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
558 mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
559 mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
560 mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
561 mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
562 mac_stats[EFX_MAC_PM_TRUNC_QBB] +
563 mac_stats[EFX_MAC_PM_DISCARD_QBB] +
564 mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
565 mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
566 mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
568 mac_stats[EFX_MAC_RX_FCS_ERRORS] +
569 mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
570 mac_stats[EFX_MAC_RX_JABBER_PKTS];
571 /* no oerrors counters supported on EF10 */
575 rte_spinlock_unlock(&port->mac_stats_lock);
576 SFC_ASSERT(ret >= 0);
581 sfc_stats_reset(struct rte_eth_dev *dev)
583 struct sfc_adapter *sa = dev->data->dev_private;
584 struct sfc_port *port = &sa->port;
587 if (sa->state != SFC_ADAPTER_STARTED) {
589 * The operation cannot be done if port is not started; it
590 * will be scheduled to be done during the next port start
592 port->mac_stats_reset_pending = B_TRUE;
596 rc = sfc_port_reset_mac_stats(sa);
598 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
602 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
603 unsigned int xstats_count)
605 struct sfc_adapter *sa = dev->data->dev_private;
606 struct sfc_port *port = &sa->port;
612 rte_spinlock_lock(&port->mac_stats_lock);
614 rc = sfc_port_update_mac_stats(sa);
621 mac_stats = port->mac_stats_buf;
623 for (i = 0; i < EFX_MAC_NSTATS; ++i) {
624 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
625 if (xstats != NULL && nstats < (int)xstats_count) {
626 xstats[nstats].id = nstats;
627 xstats[nstats].value = mac_stats[i];
634 rte_spinlock_unlock(&port->mac_stats_lock);
640 sfc_xstats_get_names(struct rte_eth_dev *dev,
641 struct rte_eth_xstat_name *xstats_names,
642 unsigned int xstats_count)
644 struct sfc_adapter *sa = dev->data->dev_private;
645 struct sfc_port *port = &sa->port;
647 unsigned int nstats = 0;
649 for (i = 0; i < EFX_MAC_NSTATS; ++i) {
650 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
651 if (xstats_names != NULL && nstats < xstats_count)
652 strlcpy(xstats_names[nstats].name,
653 efx_mac_stat_name(sa->nic, i),
654 sizeof(xstats_names[0].name));
663 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
664 uint64_t *values, unsigned int n)
666 struct sfc_adapter *sa = dev->data->dev_private;
667 struct sfc_port *port = &sa->port;
669 unsigned int nb_supported = 0;
670 unsigned int nb_written = 0;
675 if (unlikely(values == NULL) ||
676 unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
677 return port->mac_stats_nb_supported;
679 rte_spinlock_lock(&port->mac_stats_lock);
681 rc = sfc_port_update_mac_stats(sa);
688 mac_stats = port->mac_stats_buf;
690 for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
691 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
694 if ((ids == NULL) || (ids[nb_written] == nb_supported))
695 values[nb_written++] = mac_stats[i];
703 rte_spinlock_unlock(&port->mac_stats_lock);
709 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
710 struct rte_eth_xstat_name *xstats_names,
711 const uint64_t *ids, unsigned int size)
713 struct sfc_adapter *sa = dev->data->dev_private;
714 struct sfc_port *port = &sa->port;
715 unsigned int nb_supported = 0;
716 unsigned int nb_written = 0;
719 if (unlikely(xstats_names == NULL) ||
720 unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
721 return port->mac_stats_nb_supported;
723 for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
724 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
727 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
728 char *name = xstats_names[nb_written++].name;
730 strlcpy(name, efx_mac_stat_name(sa->nic, i),
731 sizeof(xstats_names[0].name));
741 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
743 struct sfc_adapter *sa = dev->data->dev_private;
744 unsigned int wanted_fc, link_fc;
746 memset(fc_conf, 0, sizeof(*fc_conf));
748 sfc_adapter_lock(sa);
750 if (sa->state == SFC_ADAPTER_STARTED)
751 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
753 link_fc = sa->port.flow_ctrl;
757 fc_conf->mode = RTE_FC_NONE;
759 case EFX_FCNTL_RESPOND:
760 fc_conf->mode = RTE_FC_RX_PAUSE;
762 case EFX_FCNTL_GENERATE:
763 fc_conf->mode = RTE_FC_TX_PAUSE;
765 case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
766 fc_conf->mode = RTE_FC_FULL;
769 sfc_err(sa, "%s: unexpected flow control value %#x",
773 fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
775 sfc_adapter_unlock(sa);
781 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
783 struct sfc_adapter *sa = dev->data->dev_private;
784 struct sfc_port *port = &sa->port;
788 if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
789 fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
790 fc_conf->mac_ctrl_frame_fwd != 0) {
791 sfc_err(sa, "unsupported flow control settings specified");
796 switch (fc_conf->mode) {
800 case RTE_FC_RX_PAUSE:
801 fcntl = EFX_FCNTL_RESPOND;
803 case RTE_FC_TX_PAUSE:
804 fcntl = EFX_FCNTL_GENERATE;
807 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
814 sfc_adapter_lock(sa);
816 if (sa->state == SFC_ADAPTER_STARTED) {
817 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
819 goto fail_mac_fcntl_set;
822 port->flow_ctrl = fcntl;
823 port->flow_ctrl_autoneg = fc_conf->autoneg;
825 sfc_adapter_unlock(sa);
830 sfc_adapter_unlock(sa);
837 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
839 struct sfc_adapter *sa = dev->data->dev_private;
840 size_t pdu = EFX_MAC_PDU(mtu);
844 sfc_log_init(sa, "mtu=%u", mtu);
847 if (pdu < EFX_MAC_PDU_MIN) {
848 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
849 (unsigned int)mtu, (unsigned int)pdu,
853 if (pdu > EFX_MAC_PDU_MAX) {
854 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
855 (unsigned int)mtu, (unsigned int)pdu,
860 sfc_adapter_lock(sa);
862 if (pdu != sa->port.pdu) {
863 if (sa->state == SFC_ADAPTER_STARTED) {
866 old_pdu = sa->port.pdu;
877 * The driver does not use it, but other PMDs update jumbo_frame
878 * flag and max_rx_pkt_len when MTU is set.
880 if (mtu > ETHER_MAX_LEN) {
881 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
883 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
884 rxmode->jumbo_frame = 1;
887 dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
889 sfc_adapter_unlock(sa);
891 sfc_log_init(sa, "done");
895 sa->port.pdu = old_pdu;
896 if (sfc_start(sa) != 0)
897 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
898 "PDU max size - port is stopped",
899 (unsigned int)pdu, (unsigned int)old_pdu);
900 sfc_adapter_unlock(sa);
903 sfc_log_init(sa, "failed %d", rc);
908 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
910 struct sfc_adapter *sa = dev->data->dev_private;
911 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
912 struct sfc_port *port = &sa->port;
913 struct ether_addr *old_addr = &dev->data->mac_addrs[0];
916 sfc_adapter_lock(sa);
919 * Copy the address to the device private data so that
920 * it could be recalled in the case of adapter restart.
922 ether_addr_copy(mac_addr, &port->default_mac_addr);
925 * Neither of the two following checks can return
926 * an error. The new MAC address is preserved in
927 * the device private data and can be activated
928 * on the next port start if the user prevents
929 * isolated mode from being enabled.
931 if (port->isolated) {
932 sfc_warn(sa, "isolated mode is active on the port");
933 sfc_warn(sa, "will not set MAC address");
937 if (sa->state != SFC_ADAPTER_STARTED) {
938 sfc_notice(sa, "the port is not started");
939 sfc_notice(sa, "the new MAC address will be set on port start");
944 if (encp->enc_allow_set_mac_with_installed_filters) {
945 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
947 sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
952 * Changing the MAC address by means of MCDI request
953 * has no effect on received traffic, therefore
954 * we also need to update unicast filters
956 rc = sfc_set_rx_mode(sa);
958 sfc_err(sa, "cannot set filter (rc = %u)", rc);
959 /* Rollback the old address */
960 (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
961 (void)sfc_set_rx_mode(sa);
964 sfc_warn(sa, "cannot set MAC address with filters installed");
965 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
966 sfc_warn(sa, "(some traffic may be dropped)");
969 * Since setting MAC address with filters installed is not
970 * allowed on the adapter, the new MAC address will be set
971 * by means of adapter restart. sfc_start() shall retrieve
972 * the new address from the device private data and set it.
977 sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
982 ether_addr_copy(old_addr, &port->default_mac_addr);
984 sfc_adapter_unlock(sa);
992 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
995 struct sfc_adapter *sa = dev->data->dev_private;
996 struct sfc_port *port = &sa->port;
997 uint8_t *mc_addrs = port->mcast_addrs;
1001 if (port->isolated) {
1002 sfc_err(sa, "isolated mode is active on the port");
1003 sfc_err(sa, "will not set multicast address list");
1007 if (mc_addrs == NULL)
1010 if (nb_mc_addr > port->max_mcast_addrs) {
1011 sfc_err(sa, "too many multicast addresses: %u > %u",
1012 nb_mc_addr, port->max_mcast_addrs);
1016 for (i = 0; i < nb_mc_addr; ++i) {
1017 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1019 mc_addrs += EFX_MAC_ADDR_LEN;
1022 port->nb_mcast_addrs = nb_mc_addr;
1024 if (sa->state != SFC_ADAPTER_STARTED)
1027 rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1028 port->nb_mcast_addrs);
1030 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1037 * The function is used by the secondary process as well. It must not
1038 * use any process-local pointers from the adapter data.
1041 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1042 struct rte_eth_rxq_info *qinfo)
1044 struct sfc_adapter *sa = dev->data->dev_private;
1045 struct sfc_rxq_info *rxq_info;
1046 struct sfc_rxq *rxq;
1048 sfc_adapter_lock(sa);
1050 SFC_ASSERT(rx_queue_id < sa->rxq_count);
1052 rxq_info = &sa->rxq_info[rx_queue_id];
1053 rxq = rxq_info->rxq;
1054 SFC_ASSERT(rxq != NULL);
1056 qinfo->mp = rxq->refill_mb_pool;
1057 qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1058 qinfo->conf.rx_drop_en = 1;
1059 qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1060 qinfo->conf.offloads = DEV_RX_OFFLOAD_IPV4_CKSUM |
1061 DEV_RX_OFFLOAD_UDP_CKSUM |
1062 DEV_RX_OFFLOAD_TCP_CKSUM;
1063 if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1064 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1065 qinfo->scattered_rx = 1;
1067 qinfo->nb_desc = rxq_info->entries;
1069 sfc_adapter_unlock(sa);
1073 * The function is used by the secondary process as well. It must not
1074 * use any process-local pointers from the adapter data.
1077 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1078 struct rte_eth_txq_info *qinfo)
1080 struct sfc_adapter *sa = dev->data->dev_private;
1081 struct sfc_txq_info *txq_info;
1083 sfc_adapter_lock(sa);
1085 SFC_ASSERT(tx_queue_id < sa->txq_count);
1087 txq_info = &sa->txq_info[tx_queue_id];
1088 SFC_ASSERT(txq_info->txq != NULL);
1090 memset(qinfo, 0, sizeof(*qinfo));
1092 qinfo->conf.txq_flags = txq_info->txq->flags;
1093 qinfo->conf.offloads = txq_info->txq->offloads;
1094 qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1095 qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1096 qinfo->nb_desc = txq_info->entries;
1098 sfc_adapter_unlock(sa);
1102 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1104 struct sfc_adapter *sa = dev->data->dev_private;
1106 sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1108 return sfc_rx_qdesc_npending(sa, rx_queue_id);
1112 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1114 struct sfc_dp_rxq *dp_rxq = queue;
1116 return sfc_rx_qdesc_done(dp_rxq, offset);
1120 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1122 struct sfc_dp_rxq *dp_rxq = queue;
1123 struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1125 return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1129 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1131 struct sfc_dp_txq *dp_txq = queue;
1132 struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1134 return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1138 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1140 struct sfc_adapter *sa = dev->data->dev_private;
1143 sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1145 sfc_adapter_lock(sa);
1148 if (sa->state != SFC_ADAPTER_STARTED)
1149 goto fail_not_started;
1151 rc = sfc_rx_qstart(sa, rx_queue_id);
1153 goto fail_rx_qstart;
1155 sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1157 sfc_adapter_unlock(sa);
1163 sfc_adapter_unlock(sa);
1169 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1171 struct sfc_adapter *sa = dev->data->dev_private;
1173 sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1175 sfc_adapter_lock(sa);
1176 sfc_rx_qstop(sa, rx_queue_id);
1178 sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1180 sfc_adapter_unlock(sa);
1186 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1188 struct sfc_adapter *sa = dev->data->dev_private;
1191 sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1193 sfc_adapter_lock(sa);
1196 if (sa->state != SFC_ADAPTER_STARTED)
1197 goto fail_not_started;
1199 rc = sfc_tx_qstart(sa, tx_queue_id);
1201 goto fail_tx_qstart;
1203 sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1205 sfc_adapter_unlock(sa);
1211 sfc_adapter_unlock(sa);
1217 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1219 struct sfc_adapter *sa = dev->data->dev_private;
1221 sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1223 sfc_adapter_lock(sa);
1225 sfc_tx_qstop(sa, tx_queue_id);
1227 sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1229 sfc_adapter_unlock(sa);
1233 static efx_tunnel_protocol_t
1234 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1237 case RTE_TUNNEL_TYPE_VXLAN:
1238 return EFX_TUNNEL_PROTOCOL_VXLAN;
1239 case RTE_TUNNEL_TYPE_GENEVE:
1240 return EFX_TUNNEL_PROTOCOL_GENEVE;
1242 return EFX_TUNNEL_NPROTOS;
1246 enum sfc_udp_tunnel_op_e {
1247 SFC_UDP_TUNNEL_ADD_PORT,
1248 SFC_UDP_TUNNEL_DEL_PORT,
1252 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1253 struct rte_eth_udp_tunnel *tunnel_udp,
1254 enum sfc_udp_tunnel_op_e op)
1256 struct sfc_adapter *sa = dev->data->dev_private;
1257 efx_tunnel_protocol_t tunnel_proto;
1260 sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1261 (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1262 (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1263 tunnel_udp->udp_port, tunnel_udp->prot_type);
1266 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1267 if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1269 goto fail_bad_proto;
1272 sfc_adapter_lock(sa);
1275 case SFC_UDP_TUNNEL_ADD_PORT:
1276 rc = efx_tunnel_config_udp_add(sa->nic,
1277 tunnel_udp->udp_port,
1280 case SFC_UDP_TUNNEL_DEL_PORT:
1281 rc = efx_tunnel_config_udp_remove(sa->nic,
1282 tunnel_udp->udp_port,
1293 if (sa->state == SFC_ADAPTER_STARTED) {
1294 rc = efx_tunnel_reconfigure(sa->nic);
1297 * Configuration is accepted by FW and MC reboot
1298 * is initiated to apply the changes. MC reboot
1299 * will be handled in a usual way (MC reboot
1300 * event on management event queue and adapter
1304 } else if (rc != 0) {
1305 goto fail_reconfigure;
1309 sfc_adapter_unlock(sa);
1313 /* Remove/restore entry since the change makes the trouble */
1315 case SFC_UDP_TUNNEL_ADD_PORT:
1316 (void)efx_tunnel_config_udp_remove(sa->nic,
1317 tunnel_udp->udp_port,
1320 case SFC_UDP_TUNNEL_DEL_PORT:
1321 (void)efx_tunnel_config_udp_add(sa->nic,
1322 tunnel_udp->udp_port,
1329 sfc_adapter_unlock(sa);
1337 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1338 struct rte_eth_udp_tunnel *tunnel_udp)
1340 return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1344 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1345 struct rte_eth_udp_tunnel *tunnel_udp)
1347 return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1351 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1352 struct rte_eth_rss_conf *rss_conf)
1354 struct sfc_adapter *sa = dev->data->dev_private;
1355 struct sfc_rss *rss = &sa->rss;
1356 struct sfc_port *port = &sa->port;
1358 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1361 if (rss->channels == 0)
1364 sfc_adapter_lock(sa);
1367 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1368 * hence, conversion is done here to derive a correct set of ETH_RSS
1369 * flags which corresponds to the active EFX configuration stored
1370 * locally in 'sfc_adapter' and kept up-to-date
1372 rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(sa, rss->hash_types);
1373 rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1374 if (rss_conf->rss_key != NULL)
1375 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1377 sfc_adapter_unlock(sa);
1383 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1384 struct rte_eth_rss_conf *rss_conf)
1386 struct sfc_adapter *sa = dev->data->dev_private;
1387 struct sfc_rss *rss = &sa->rss;
1388 struct sfc_port *port = &sa->port;
1389 unsigned int efx_hash_types;
1395 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1396 sfc_err(sa, "RSS is not available");
1400 if (rss->channels == 0) {
1401 sfc_err(sa, "RSS is not configured");
1405 if ((rss_conf->rss_key != NULL) &&
1406 (rss_conf->rss_key_len != sizeof(rss->key))) {
1407 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1412 sfc_adapter_lock(sa);
1414 rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1416 goto fail_rx_hf_rte_to_efx;
1418 rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1419 rss->hash_alg, efx_hash_types, B_TRUE);
1421 goto fail_scale_mode_set;
1423 if (rss_conf->rss_key != NULL) {
1424 if (sa->state == SFC_ADAPTER_STARTED) {
1425 rc = efx_rx_scale_key_set(sa->nic,
1426 EFX_RSS_CONTEXT_DEFAULT,
1430 goto fail_scale_key_set;
1433 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1436 rss->hash_types = efx_hash_types;
1438 sfc_adapter_unlock(sa);
1443 if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1444 EFX_RX_HASHALG_TOEPLITZ,
1445 rss->hash_types, B_TRUE) != 0)
1446 sfc_err(sa, "failed to restore RSS mode");
1448 fail_scale_mode_set:
1449 fail_rx_hf_rte_to_efx:
1450 sfc_adapter_unlock(sa);
1455 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1456 struct rte_eth_rss_reta_entry64 *reta_conf,
1459 struct sfc_adapter *sa = dev->data->dev_private;
1460 struct sfc_rss *rss = &sa->rss;
1461 struct sfc_port *port = &sa->port;
1464 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1467 if (rss->channels == 0)
1470 if (reta_size != EFX_RSS_TBL_SIZE)
1473 sfc_adapter_lock(sa);
1475 for (entry = 0; entry < reta_size; entry++) {
1476 int grp = entry / RTE_RETA_GROUP_SIZE;
1477 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1479 if ((reta_conf[grp].mask >> grp_idx) & 1)
1480 reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1483 sfc_adapter_unlock(sa);
1489 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1490 struct rte_eth_rss_reta_entry64 *reta_conf,
1493 struct sfc_adapter *sa = dev->data->dev_private;
1494 struct sfc_rss *rss = &sa->rss;
1495 struct sfc_port *port = &sa->port;
1496 unsigned int *rss_tbl_new;
1504 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1505 sfc_err(sa, "RSS is not available");
1509 if (rss->channels == 0) {
1510 sfc_err(sa, "RSS is not configured");
1514 if (reta_size != EFX_RSS_TBL_SIZE) {
1515 sfc_err(sa, "RETA size is wrong (should be %u)",
1520 rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1521 if (rss_tbl_new == NULL)
1524 sfc_adapter_lock(sa);
1526 rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1528 for (entry = 0; entry < reta_size; entry++) {
1529 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1530 struct rte_eth_rss_reta_entry64 *grp;
1532 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1534 if (grp->mask & (1ull << grp_idx)) {
1535 if (grp->reta[grp_idx] >= rss->channels) {
1537 goto bad_reta_entry;
1539 rss_tbl_new[entry] = grp->reta[grp_idx];
1543 if (sa->state == SFC_ADAPTER_STARTED) {
1544 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1545 rss_tbl_new, EFX_RSS_TBL_SIZE);
1547 goto fail_scale_tbl_set;
1550 rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1554 sfc_adapter_unlock(sa);
1556 rte_free(rss_tbl_new);
1558 SFC_ASSERT(rc >= 0);
1563 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1564 enum rte_filter_op filter_op,
1567 struct sfc_adapter *sa = dev->data->dev_private;
1570 sfc_log_init(sa, "entry");
1572 switch (filter_type) {
1573 case RTE_ETH_FILTER_NONE:
1574 sfc_err(sa, "Global filters configuration not supported");
1576 case RTE_ETH_FILTER_MACVLAN:
1577 sfc_err(sa, "MACVLAN filters not supported");
1579 case RTE_ETH_FILTER_ETHERTYPE:
1580 sfc_err(sa, "EtherType filters not supported");
1582 case RTE_ETH_FILTER_FLEXIBLE:
1583 sfc_err(sa, "Flexible filters not supported");
1585 case RTE_ETH_FILTER_SYN:
1586 sfc_err(sa, "SYN filters not supported");
1588 case RTE_ETH_FILTER_NTUPLE:
1589 sfc_err(sa, "NTUPLE filters not supported");
1591 case RTE_ETH_FILTER_TUNNEL:
1592 sfc_err(sa, "Tunnel filters not supported");
1594 case RTE_ETH_FILTER_FDIR:
1595 sfc_err(sa, "Flow Director filters not supported");
1597 case RTE_ETH_FILTER_HASH:
1598 sfc_err(sa, "Hash filters not supported");
1600 case RTE_ETH_FILTER_GENERIC:
1601 if (filter_op != RTE_ETH_FILTER_GET) {
1604 *(const void **)arg = &sfc_flow_ops;
1609 sfc_err(sa, "Unknown filter type %u", filter_type);
1613 sfc_log_init(sa, "exit: %d", -rc);
1614 SFC_ASSERT(rc >= 0);
1619 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1621 struct sfc_adapter *sa = dev->data->dev_private;
1624 * If Rx datapath does not provide callback to check mempool,
1625 * all pools are supported.
1627 if (sa->dp_rx->pool_ops_supported == NULL)
1630 return sa->dp_rx->pool_ops_supported(pool);
1633 static const struct eth_dev_ops sfc_eth_dev_ops = {
1634 .dev_configure = sfc_dev_configure,
1635 .dev_start = sfc_dev_start,
1636 .dev_stop = sfc_dev_stop,
1637 .dev_set_link_up = sfc_dev_set_link_up,
1638 .dev_set_link_down = sfc_dev_set_link_down,
1639 .dev_close = sfc_dev_close,
1640 .promiscuous_enable = sfc_dev_promisc_enable,
1641 .promiscuous_disable = sfc_dev_promisc_disable,
1642 .allmulticast_enable = sfc_dev_allmulti_enable,
1643 .allmulticast_disable = sfc_dev_allmulti_disable,
1644 .link_update = sfc_dev_link_update,
1645 .stats_get = sfc_stats_get,
1646 .stats_reset = sfc_stats_reset,
1647 .xstats_get = sfc_xstats_get,
1648 .xstats_reset = sfc_stats_reset,
1649 .xstats_get_names = sfc_xstats_get_names,
1650 .dev_infos_get = sfc_dev_infos_get,
1651 .dev_supported_ptypes_get = sfc_dev_supported_ptypes_get,
1652 .mtu_set = sfc_dev_set_mtu,
1653 .rx_queue_start = sfc_rx_queue_start,
1654 .rx_queue_stop = sfc_rx_queue_stop,
1655 .tx_queue_start = sfc_tx_queue_start,
1656 .tx_queue_stop = sfc_tx_queue_stop,
1657 .rx_queue_setup = sfc_rx_queue_setup,
1658 .rx_queue_release = sfc_rx_queue_release,
1659 .rx_queue_count = sfc_rx_queue_count,
1660 .rx_descriptor_done = sfc_rx_descriptor_done,
1661 .rx_descriptor_status = sfc_rx_descriptor_status,
1662 .tx_descriptor_status = sfc_tx_descriptor_status,
1663 .tx_queue_setup = sfc_tx_queue_setup,
1664 .tx_queue_release = sfc_tx_queue_release,
1665 .flow_ctrl_get = sfc_flow_ctrl_get,
1666 .flow_ctrl_set = sfc_flow_ctrl_set,
1667 .mac_addr_set = sfc_mac_addr_set,
1668 .udp_tunnel_port_add = sfc_dev_udp_tunnel_port_add,
1669 .udp_tunnel_port_del = sfc_dev_udp_tunnel_port_del,
1670 .reta_update = sfc_dev_rss_reta_update,
1671 .reta_query = sfc_dev_rss_reta_query,
1672 .rss_hash_update = sfc_dev_rss_hash_update,
1673 .rss_hash_conf_get = sfc_dev_rss_hash_conf_get,
1674 .filter_ctrl = sfc_dev_filter_ctrl,
1675 .set_mc_addr_list = sfc_set_mc_addr_list,
1676 .rxq_info_get = sfc_rx_queue_info_get,
1677 .txq_info_get = sfc_tx_queue_info_get,
1678 .fw_version_get = sfc_fw_version_get,
1679 .xstats_get_by_id = sfc_xstats_get_by_id,
1680 .xstats_get_names_by_id = sfc_xstats_get_names_by_id,
1681 .pool_ops_supported = sfc_pool_ops_supported,
1685 * Duplicate a string in potentially shared memory required for
1686 * multi-process support.
1688 * strdup() allocates from process-local heap/memory.
1691 sfc_strdup(const char *str)
1699 size = strlen(str) + 1;
1700 copy = rte_malloc(__func__, size, 0);
1702 rte_memcpy(copy, str, size);
1708 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1710 struct sfc_adapter *sa = dev->data->dev_private;
1711 const efx_nic_cfg_t *encp;
1712 unsigned int avail_caps = 0;
1713 const char *rx_name = NULL;
1714 const char *tx_name = NULL;
1717 switch (sa->family) {
1718 case EFX_FAMILY_HUNTINGTON:
1719 case EFX_FAMILY_MEDFORD:
1720 case EFX_FAMILY_MEDFORD2:
1721 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1727 encp = efx_nic_cfg_get(sa->nic);
1728 if (encp->enc_rx_es_super_buffer_supported)
1729 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1731 rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1732 sfc_kvarg_string_handler, &rx_name);
1734 goto fail_kvarg_rx_datapath;
1736 if (rx_name != NULL) {
1737 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1738 if (sa->dp_rx == NULL) {
1739 sfc_err(sa, "Rx datapath %s not found", rx_name);
1743 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1745 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1748 goto fail_dp_rx_caps;
1751 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1752 if (sa->dp_rx == NULL) {
1753 sfc_err(sa, "Rx datapath by caps %#x not found",
1760 sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1761 if (sa->dp_rx_name == NULL) {
1763 goto fail_dp_rx_name;
1766 sfc_notice(sa, "use %s Rx datapath", sa->dp_rx_name);
1768 dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1770 rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1771 sfc_kvarg_string_handler, &tx_name);
1773 goto fail_kvarg_tx_datapath;
1775 if (tx_name != NULL) {
1776 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1777 if (sa->dp_tx == NULL) {
1778 sfc_err(sa, "Tx datapath %s not found", tx_name);
1782 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1784 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1787 goto fail_dp_tx_caps;
1790 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1791 if (sa->dp_tx == NULL) {
1792 sfc_err(sa, "Tx datapath by caps %#x not found",
1799 sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1800 if (sa->dp_tx_name == NULL) {
1802 goto fail_dp_tx_name;
1805 sfc_notice(sa, "use %s Tx datapath", sa->dp_tx_name);
1807 dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1809 dev->dev_ops = &sfc_eth_dev_ops;
1818 fail_kvarg_tx_datapath:
1819 rte_free(sa->dp_rx_name);
1820 sa->dp_rx_name = NULL;
1827 fail_kvarg_rx_datapath:
1832 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1834 struct sfc_adapter *sa = dev->data->dev_private;
1836 dev->dev_ops = NULL;
1837 dev->rx_pkt_burst = NULL;
1838 dev->tx_pkt_burst = NULL;
1840 rte_free(sa->dp_tx_name);
1841 sa->dp_tx_name = NULL;
1844 rte_free(sa->dp_rx_name);
1845 sa->dp_rx_name = NULL;
1849 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1850 .rxq_info_get = sfc_rx_queue_info_get,
1851 .txq_info_get = sfc_tx_queue_info_get,
1855 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1858 * Device private data has really many process-local pointers.
1859 * Below code should be extremely careful to use data located
1860 * in shared memory only.
1862 struct sfc_adapter *sa = dev->data->dev_private;
1863 const struct sfc_dp_rx *dp_rx;
1864 const struct sfc_dp_tx *dp_tx;
1867 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1868 if (dp_rx == NULL) {
1869 sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1873 if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1874 sfc_err(sa, "%s Rx datapath does not support multi-process",
1877 goto fail_dp_rx_multi_process;
1880 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1881 if (dp_tx == NULL) {
1882 sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1886 if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1887 sfc_err(sa, "%s Tx datapath does not support multi-process",
1890 goto fail_dp_tx_multi_process;
1893 dev->rx_pkt_burst = dp_rx->pkt_burst;
1894 dev->tx_pkt_burst = dp_tx->pkt_burst;
1895 dev->dev_ops = &sfc_eth_dev_secondary_ops;
1899 fail_dp_tx_multi_process:
1901 fail_dp_rx_multi_process:
1907 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1909 dev->dev_ops = NULL;
1910 dev->tx_pkt_burst = NULL;
1911 dev->rx_pkt_burst = NULL;
1915 sfc_register_dp(void)
1918 if (TAILQ_EMPTY(&sfc_dp_head)) {
1919 /* Prefer EF10 datapath */
1920 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
1921 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1922 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1924 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1925 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1926 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1931 sfc_eth_dev_init(struct rte_eth_dev *dev)
1933 struct sfc_adapter *sa = dev->data->dev_private;
1934 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1936 const efx_nic_cfg_t *encp;
1937 const struct ether_addr *from;
1941 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1942 return -sfc_eth_dev_secondary_set_ops(dev);
1944 /* Required for logging */
1945 sa->pci_addr = pci_dev->addr;
1946 sa->port_id = dev->data->port_id;
1950 /* Copy PCI device info to the dev->data */
1951 rte_eth_copy_pci_info(dev, pci_dev);
1953 sa->logtype_main = sfc_register_logtype(sa, SFC_LOGTYPE_MAIN_STR,
1956 rc = sfc_kvargs_parse(sa);
1958 goto fail_kvargs_parse;
1960 sfc_log_init(sa, "entry");
1962 dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1963 if (dev->data->mac_addrs == NULL) {
1965 goto fail_mac_addrs;
1968 sfc_adapter_lock_init(sa);
1969 sfc_adapter_lock(sa);
1971 sfc_log_init(sa, "probing");
1976 sfc_log_init(sa, "set device ops");
1977 rc = sfc_eth_dev_set_ops(dev);
1981 sfc_log_init(sa, "attaching");
1982 rc = sfc_attach(sa);
1986 encp = efx_nic_cfg_get(sa->nic);
1989 * The arguments are really reverse order in comparison to
1990 * Linux kernel. Copy from NIC config to Ethernet device data.
1992 from = (const struct ether_addr *)(encp->enc_mac_addr);
1993 ether_addr_copy(from, &dev->data->mac_addrs[0]);
1995 sfc_adapter_unlock(sa);
1997 sfc_log_init(sa, "done");
2001 sfc_eth_dev_clear_ops(dev);
2007 sfc_adapter_unlock(sa);
2008 sfc_adapter_lock_fini(sa);
2009 rte_free(dev->data->mac_addrs);
2010 dev->data->mac_addrs = NULL;
2013 sfc_kvargs_cleanup(sa);
2016 sfc_log_init(sa, "failed %d", rc);
2022 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2024 struct sfc_adapter *sa;
2026 if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2027 sfc_eth_dev_secondary_clear_ops(dev);
2031 sa = dev->data->dev_private;
2032 sfc_log_init(sa, "entry");
2034 sfc_adapter_lock(sa);
2036 sfc_eth_dev_clear_ops(dev);
2041 rte_free(dev->data->mac_addrs);
2042 dev->data->mac_addrs = NULL;
2044 sfc_kvargs_cleanup(sa);
2046 sfc_adapter_unlock(sa);
2047 sfc_adapter_lock_fini(sa);
2049 sfc_log_init(sa, "done");
2051 /* Required for logging, so cleanup last */
2056 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2057 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2058 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2059 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2060 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2061 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2062 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2063 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2064 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2065 { .vendor_id = 0 /* sentinel */ }
2068 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2069 struct rte_pci_device *pci_dev)
2071 return rte_eth_dev_pci_generic_probe(pci_dev,
2072 sizeof(struct sfc_adapter), sfc_eth_dev_init);
2075 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2077 return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2080 static struct rte_pci_driver sfc_efx_pmd = {
2081 .id_table = pci_id_sfc_efx_map,
2083 RTE_PCI_DRV_INTR_LSC |
2084 RTE_PCI_DRV_NEED_MAPPING,
2085 .probe = sfc_eth_dev_pci_probe,
2086 .remove = sfc_eth_dev_pci_remove,
2089 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2090 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2091 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2092 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2093 SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2094 SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2095 SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2096 SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2097 SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2098 SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2100 RTE_INIT(sfc_driver_register_logtype);
2102 sfc_driver_register_logtype(void)
2106 ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2108 sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;