1 /* SPDX-License-Identifier: BSD-3-Clause
3 * Copyright(c) 2019-2021 Xilinx, Inc.
4 * Copyright(c) 2016-2019 Solarflare Communications Inc.
6 * This software was jointly developed between OKTET Labs (under contract
7 * for Solarflare) and Solarflare Communications, Inc.
11 #include <ethdev_driver.h>
12 #include <ethdev_pci.h>
14 #include <rte_bus_pci.h>
15 #include <rte_errno.h>
16 #include <rte_string_fns.h>
17 #include <rte_ether.h>
22 #include "sfc_debug.h"
24 #include "sfc_kvargs.h"
29 #include "sfc_flow_tunnel.h"
31 #include "sfc_dp_rx.h"
33 #include "sfc_sw_stats.h"
34 #include "sfc_switch.h"
35 #include "sfc_nic_dma.h"
37 #define SFC_XSTAT_ID_INVALID_VAL UINT64_MAX
38 #define SFC_XSTAT_ID_INVALID_NAME '\0'
40 uint32_t sfc_logtype_driver;
42 static struct sfc_dp_list sfc_dp_head =
43 TAILQ_HEAD_INITIALIZER(sfc_dp_head);
46 static void sfc_eth_dev_clear_ops(struct rte_eth_dev *dev);
50 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
52 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
53 efx_nic_fw_info_t enfi;
57 rc = efx_nic_get_fw_version(sa->nic, &enfi);
61 ret = snprintf(fw_version, fw_size,
62 "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
63 enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
64 enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
68 if (enfi.enfi_dpcpu_fw_ids_valid) {
69 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
72 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
73 fw_size - dpcpu_fw_ids_offset,
74 " rx%" PRIx16 " tx%" PRIx16,
75 enfi.enfi_rx_dpcpu_fw_id,
76 enfi.enfi_tx_dpcpu_fw_id);
83 if (fw_size < (size_t)(++ret))
90 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
92 const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
93 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
94 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
95 struct sfc_rss *rss = &sas->rss;
96 struct sfc_mae *mae = &sa->mae;
97 uint64_t txq_offloads_def = 0;
99 sfc_log_init(sa, "entry");
101 dev_info->min_mtu = RTE_ETHER_MIN_MTU;
102 dev_info->max_mtu = EFX_MAC_SDU_MAX;
104 dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
106 dev_info->max_vfs = sa->sriov.num_vfs;
108 /* Autonegotiation may be disabled */
109 dev_info->speed_capa = RTE_ETH_LINK_SPEED_FIXED;
110 if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_1000FDX))
111 dev_info->speed_capa |= RTE_ETH_LINK_SPEED_1G;
112 if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_10000FDX))
113 dev_info->speed_capa |= RTE_ETH_LINK_SPEED_10G;
114 if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_25000FDX))
115 dev_info->speed_capa |= RTE_ETH_LINK_SPEED_25G;
116 if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_40000FDX))
117 dev_info->speed_capa |= RTE_ETH_LINK_SPEED_40G;
118 if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_50000FDX))
119 dev_info->speed_capa |= RTE_ETH_LINK_SPEED_50G;
120 if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_100000FDX))
121 dev_info->speed_capa |= RTE_ETH_LINK_SPEED_100G;
123 dev_info->max_rx_queues = sa->rxq_max;
124 dev_info->max_tx_queues = sa->txq_max;
126 /* By default packets are dropped if no descriptors are available */
127 dev_info->default_rxconf.rx_drop_en = 1;
129 dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
132 * rx_offload_capa includes both device and queue offloads since
133 * the latter may be requested on a per device basis which makes
134 * sense when some offloads are needed to be set on all queues.
136 dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
137 dev_info->rx_queue_offload_capa;
139 dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
142 * tx_offload_capa includes both device and queue offloads since
143 * the latter may be requested on a per device basis which makes
144 * sense when some offloads are needed to be set on all queues.
146 dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
147 dev_info->tx_queue_offload_capa;
149 if (dev_info->tx_offload_capa & RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE)
150 txq_offloads_def |= RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE;
152 dev_info->default_txconf.offloads |= txq_offloads_def;
154 if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
158 for (i = 0; i < rss->hf_map_nb_entries; ++i)
159 rte_hf |= rss->hf_map[i].rte;
161 dev_info->reta_size = EFX_RSS_TBL_SIZE;
162 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
163 dev_info->flow_type_rss_offloads = rte_hf;
166 /* Initialize to hardware limits */
167 dev_info->rx_desc_lim.nb_max = sa->rxq_max_entries;
168 dev_info->rx_desc_lim.nb_min = sa->rxq_min_entries;
169 /* The RXQ hardware requires that the descriptor count is a power
170 * of 2, but rx_desc_lim cannot properly describe that constraint.
172 dev_info->rx_desc_lim.nb_align = sa->rxq_min_entries;
174 /* Initialize to hardware limits */
175 dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
176 dev_info->tx_desc_lim.nb_min = sa->txq_min_entries;
178 * The TXQ hardware requires that the descriptor count is a power
179 * of 2, but tx_desc_lim cannot properly describe that constraint
181 dev_info->tx_desc_lim.nb_align = sa->txq_min_entries;
183 if (sap->dp_rx->get_dev_info != NULL)
184 sap->dp_rx->get_dev_info(dev_info);
185 if (sap->dp_tx->get_dev_info != NULL)
186 sap->dp_tx->get_dev_info(dev_info);
188 dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
189 RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
190 dev_info->dev_capa &= ~RTE_ETH_DEV_CAPA_FLOW_RULE_KEEP;
192 if (mae->status == SFC_MAE_STATUS_SUPPORTED ||
193 mae->status == SFC_MAE_STATUS_ADMIN) {
194 dev_info->switch_info.name = dev->device->driver->name;
195 dev_info->switch_info.domain_id = mae->switch_domain_id;
196 dev_info->switch_info.port_id = mae->switch_port_id;
202 static const uint32_t *
203 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
205 const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
207 return sap->dp_rx->supported_ptypes_get(sap->shared->tunnel_encaps);
211 sfc_dev_configure(struct rte_eth_dev *dev)
213 struct rte_eth_dev_data *dev_data = dev->data;
214 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
217 sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
218 dev_data->nb_rx_queues, dev_data->nb_tx_queues);
220 sfc_adapter_lock(sa);
222 case SFC_ETHDEV_CONFIGURED:
224 case SFC_ETHDEV_INITIALIZED:
225 rc = sfc_configure(sa);
228 sfc_err(sa, "unexpected adapter state %u to configure",
233 sfc_adapter_unlock(sa);
235 sfc_log_init(sa, "done %d", rc);
241 sfc_dev_start(struct rte_eth_dev *dev)
243 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
246 sfc_log_init(sa, "entry");
248 sfc_adapter_lock(sa);
250 sfc_adapter_unlock(sa);
252 sfc_log_init(sa, "done %d", rc);
258 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
260 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
261 struct rte_eth_link current_link;
264 sfc_log_init(sa, "entry");
266 if (sa->state != SFC_ETHDEV_STARTED) {
267 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, ¤t_link);
268 } else if (wait_to_complete) {
269 efx_link_mode_t link_mode;
271 if (efx_port_poll(sa->nic, &link_mode) != 0)
272 link_mode = EFX_LINK_UNKNOWN;
273 sfc_port_link_mode_to_info(link_mode, ¤t_link);
276 sfc_ev_mgmt_qpoll(sa);
277 rte_eth_linkstatus_get(dev, ¤t_link);
280 ret = rte_eth_linkstatus_set(dev, ¤t_link);
282 sfc_notice(sa, "Link status is %s",
283 current_link.link_status ? "UP" : "DOWN");
289 sfc_dev_stop(struct rte_eth_dev *dev)
291 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
293 sfc_log_init(sa, "entry");
295 sfc_adapter_lock(sa);
297 sfc_adapter_unlock(sa);
299 sfc_log_init(sa, "done");
305 sfc_dev_set_link_up(struct rte_eth_dev *dev)
307 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
310 sfc_log_init(sa, "entry");
312 sfc_adapter_lock(sa);
314 sfc_adapter_unlock(sa);
321 sfc_dev_set_link_down(struct rte_eth_dev *dev)
323 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
325 sfc_log_init(sa, "entry");
327 sfc_adapter_lock(sa);
329 sfc_adapter_unlock(sa);
335 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
337 free(dev->process_private);
338 rte_eth_dev_release_port(dev);
342 sfc_dev_close(struct rte_eth_dev *dev)
344 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
346 sfc_log_init(sa, "entry");
348 if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
349 sfc_eth_dev_secondary_clear_ops(dev);
355 sfc_adapter_lock(sa);
357 case SFC_ETHDEV_STARTED:
359 SFC_ASSERT(sa->state == SFC_ETHDEV_CONFIGURED);
361 case SFC_ETHDEV_CONFIGURED:
363 SFC_ASSERT(sa->state == SFC_ETHDEV_INITIALIZED);
365 case SFC_ETHDEV_INITIALIZED:
368 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
373 * Cleanup all resources.
374 * Rollback primary process sfc_eth_dev_init() below.
377 sfc_eth_dev_clear_ops(dev);
379 sfc_nic_dma_detach(sa);
383 sfc_kvargs_cleanup(sa);
385 sfc_adapter_unlock(sa);
386 sfc_adapter_lock_fini(sa);
388 sfc_log_init(sa, "done");
390 /* Required for logging, so cleanup last */
399 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
402 struct sfc_port *port;
404 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
405 boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
406 const char *desc = (allmulti) ? "all-multi" : "promiscuous";
409 sfc_adapter_lock(sa);
412 toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
414 if (*toggle != enabled) {
417 if (sfc_sa2shared(sa)->isolated) {
418 sfc_warn(sa, "isolated mode is active on the port");
419 sfc_warn(sa, "the change is to be applied on the next "
420 "start provided that isolated mode is "
421 "disabled prior the next start");
422 } else if ((sa->state == SFC_ETHDEV_STARTED) &&
423 ((rc = sfc_set_rx_mode(sa)) != 0)) {
424 *toggle = !(enabled);
425 sfc_warn(sa, "Failed to %s %s mode, rc = %d",
426 ((enabled) ? "enable" : "disable"), desc, rc);
429 * For promiscuous and all-multicast filters a
430 * permission failure should be reported as an
431 * unsupported filter.
438 sfc_adapter_unlock(sa);
443 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
445 int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
452 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
454 int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
461 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
463 int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
470 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
472 int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
479 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid,
480 uint16_t nb_rx_desc, unsigned int socket_id,
481 const struct rte_eth_rxconf *rx_conf,
482 struct rte_mempool *mb_pool)
484 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
485 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
486 sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
487 struct sfc_rxq_info *rxq_info;
488 sfc_sw_index_t sw_index;
491 sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
492 ethdev_qid, nb_rx_desc, socket_id);
494 sfc_adapter_lock(sa);
496 sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
497 rc = sfc_rx_qinit(sa, sw_index, nb_rx_desc, socket_id,
502 rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
503 dev->data->rx_queues[ethdev_qid] = rxq_info->dp;
505 sfc_adapter_unlock(sa);
510 sfc_adapter_unlock(sa);
516 sfc_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
518 struct sfc_dp_rxq *dp_rxq = dev->data->rx_queues[qid];
520 struct sfc_adapter *sa;
521 sfc_sw_index_t sw_index;
526 rxq = sfc_rxq_by_dp_rxq(dp_rxq);
528 sfc_adapter_lock(sa);
530 sw_index = dp_rxq->dpq.queue_id;
532 sfc_log_init(sa, "RxQ=%u", sw_index);
534 sfc_rx_qfini(sa, sw_index);
536 sfc_adapter_unlock(sa);
540 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid,
541 uint16_t nb_tx_desc, unsigned int socket_id,
542 const struct rte_eth_txconf *tx_conf)
544 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
545 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
546 struct sfc_txq_info *txq_info;
547 sfc_sw_index_t sw_index;
550 sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
551 ethdev_qid, nb_tx_desc, socket_id);
553 sfc_adapter_lock(sa);
555 sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
556 rc = sfc_tx_qinit(sa, sw_index, nb_tx_desc, socket_id, tx_conf);
560 txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
561 dev->data->tx_queues[ethdev_qid] = txq_info->dp;
563 sfc_adapter_unlock(sa);
567 sfc_adapter_unlock(sa);
573 sfc_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
575 struct sfc_dp_txq *dp_txq = dev->data->tx_queues[qid];
577 sfc_sw_index_t sw_index;
578 struct sfc_adapter *sa;
583 txq = sfc_txq_by_dp_txq(dp_txq);
584 sw_index = dp_txq->dpq.queue_id;
586 SFC_ASSERT(txq->evq != NULL);
589 sfc_log_init(sa, "TxQ = %u", sw_index);
591 sfc_adapter_lock(sa);
593 sfc_tx_qfini(sa, sw_index);
595 sfc_adapter_unlock(sa);
599 sfc_stats_get_dp_rx(struct sfc_adapter *sa, uint64_t *pkts, uint64_t *bytes)
601 struct sfc_adapter_shared *sas = sfc_sa2shared(sa);
602 uint64_t pkts_sum = 0;
603 uint64_t bytes_sum = 0;
606 for (i = 0; i < sas->ethdev_rxq_count; ++i) {
607 struct sfc_rxq_info *rxq_info;
609 rxq_info = sfc_rxq_info_by_ethdev_qid(sas, i);
610 if (rxq_info->state & SFC_RXQ_INITIALIZED) {
611 union sfc_pkts_bytes qstats;
613 sfc_pkts_bytes_get(&rxq_info->dp->dpq.stats, &qstats);
614 pkts_sum += qstats.pkts -
615 sa->sw_stats.reset_rx_pkts[i];
616 bytes_sum += qstats.bytes -
617 sa->sw_stats.reset_rx_bytes[i];
626 sfc_stats_get_dp_tx(struct sfc_adapter *sa, uint64_t *pkts, uint64_t *bytes)
628 struct sfc_adapter_shared *sas = sfc_sa2shared(sa);
629 uint64_t pkts_sum = 0;
630 uint64_t bytes_sum = 0;
633 for (i = 0; i < sas->ethdev_txq_count; ++i) {
634 struct sfc_txq_info *txq_info;
636 txq_info = sfc_txq_info_by_ethdev_qid(sas, i);
637 if (txq_info->state & SFC_TXQ_INITIALIZED) {
638 union sfc_pkts_bytes qstats;
640 sfc_pkts_bytes_get(&txq_info->dp->dpq.stats, &qstats);
641 pkts_sum += qstats.pkts -
642 sa->sw_stats.reset_tx_pkts[i];
643 bytes_sum += qstats.bytes -
644 sa->sw_stats.reset_tx_bytes[i];
653 * Some statistics are computed as A - B where A and B each increase
654 * monotonically with some hardware counter(s) and the counters are read
657 * If packet X is counted in A, but not counted in B yet, computed value is
660 * If packet X is not counted in A at the moment of reading the counter,
661 * but counted in B at the moment of reading the counter, computed value
664 * However, counter which grows backward is worse evil than slightly wrong
665 * value. So, let's try to guarantee that it never happens except may be
666 * the case when the MAC stats are zeroed as a result of a NIC reset.
669 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
671 if ((int64_t)(newval - *stat) > 0 || newval == 0)
676 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
678 const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
679 bool have_dp_rx_stats = sap->dp_rx->features & SFC_DP_RX_FEAT_STATS;
680 bool have_dp_tx_stats = sap->dp_tx->features & SFC_DP_TX_FEAT_STATS;
681 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
682 struct sfc_port *port = &sa->port;
686 sfc_adapter_lock(sa);
688 if (have_dp_rx_stats)
689 sfc_stats_get_dp_rx(sa, &stats->ipackets, &stats->ibytes);
690 if (have_dp_tx_stats)
691 sfc_stats_get_dp_tx(sa, &stats->opackets, &stats->obytes);
693 ret = sfc_port_update_mac_stats(sa, B_FALSE);
697 mac_stats = port->mac_stats_buf;
699 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
700 EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
701 if (!have_dp_rx_stats) {
703 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
704 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
705 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
707 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
708 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
709 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
711 /* CRC is included in these stats, but shouldn't be */
712 stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
714 if (!have_dp_tx_stats) {
716 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
717 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
718 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
720 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
721 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
722 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
724 /* CRC is included in these stats, but shouldn't be */
725 stats->obytes -= stats->opackets * RTE_ETHER_CRC_LEN;
727 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
728 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
730 if (!have_dp_tx_stats) {
731 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
732 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS] -
733 mac_stats[EFX_MAC_TX_PKTS] * RTE_ETHER_CRC_LEN;
737 * Take into account stats which are whenever supported
738 * on EF10. If some stat is not supported by current
739 * firmware variant or HW revision, it is guaranteed
740 * to be zero in mac_stats.
743 mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
744 mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
745 mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
746 mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
747 mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
748 mac_stats[EFX_MAC_PM_TRUNC_QBB] +
749 mac_stats[EFX_MAC_PM_DISCARD_QBB] +
750 mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
751 mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
752 mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
754 mac_stats[EFX_MAC_RX_FCS_ERRORS] +
755 mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
756 mac_stats[EFX_MAC_RX_JABBER_PKTS];
757 /* no oerrors counters supported on EF10 */
759 if (!have_dp_rx_stats) {
760 /* Exclude missed, errors and pauses from Rx packets */
761 sfc_update_diff_stat(&port->ipackets,
762 mac_stats[EFX_MAC_RX_PKTS] -
763 mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
764 stats->imissed - stats->ierrors);
765 stats->ipackets = port->ipackets;
766 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS] -
767 mac_stats[EFX_MAC_RX_PKTS] * RTE_ETHER_CRC_LEN;
772 sfc_adapter_unlock(sa);
773 SFC_ASSERT(ret >= 0);
778 sfc_stats_reset(struct rte_eth_dev *dev)
780 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
781 struct sfc_port *port = &sa->port;
784 sfc_adapter_lock(sa);
786 if (sa->state != SFC_ETHDEV_STARTED) {
788 * The operation cannot be done if port is not started; it
789 * will be scheduled to be done during the next port start
791 port->mac_stats_reset_pending = B_TRUE;
792 sfc_adapter_unlock(sa);
796 rc = sfc_port_reset_mac_stats(sa);
798 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
800 sfc_sw_xstats_reset(sa);
802 sfc_adapter_unlock(sa);
809 sfc_xstats_get_nb_supported(struct sfc_adapter *sa)
811 struct sfc_port *port = &sa->port;
812 unsigned int nb_supported;
814 sfc_adapter_lock(sa);
815 nb_supported = port->mac_stats_nb_supported +
816 sfc_sw_xstats_get_nb_supported(sa);
817 sfc_adapter_unlock(sa);
823 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
824 unsigned int xstats_count)
826 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
827 unsigned int nb_written = 0;
828 unsigned int nb_supported = 0;
831 if (unlikely(xstats == NULL))
832 return sfc_xstats_get_nb_supported(sa);
834 rc = sfc_port_get_mac_stats(sa, xstats, xstats_count, &nb_written);
839 sfc_sw_xstats_get_vals(sa, xstats, xstats_count, &nb_written,
846 sfc_xstats_get_names(struct rte_eth_dev *dev,
847 struct rte_eth_xstat_name *xstats_names,
848 unsigned int xstats_count)
850 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
851 struct sfc_port *port = &sa->port;
853 unsigned int nstats = 0;
854 unsigned int nb_written = 0;
857 if (unlikely(xstats_names == NULL))
858 return sfc_xstats_get_nb_supported(sa);
860 for (i = 0; i < EFX_MAC_NSTATS; ++i) {
861 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
862 if (nstats < xstats_count) {
863 strlcpy(xstats_names[nstats].name,
864 efx_mac_stat_name(sa->nic, i),
865 sizeof(xstats_names[0].name));
872 ret = sfc_sw_xstats_get_names(sa, xstats_names, xstats_count,
873 &nb_written, &nstats);
883 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
884 uint64_t *values, unsigned int n)
886 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
887 struct sfc_port *port = &sa->port;
888 unsigned int nb_supported;
892 if (unlikely(ids == NULL || values == NULL))
896 * Values array could be filled in nonsequential order. Fill values with
897 * constant indicating invalid ID first.
899 for (i = 0; i < n; i++)
900 values[i] = SFC_XSTAT_ID_INVALID_VAL;
902 rc = sfc_port_get_mac_stats_by_id(sa, ids, values, n);
906 nb_supported = port->mac_stats_nb_supported;
907 sfc_sw_xstats_get_vals_by_id(sa, ids, values, n, &nb_supported);
909 /* Return number of written stats before invalid ID is encountered. */
910 for (i = 0; i < n; i++) {
911 if (values[i] == SFC_XSTAT_ID_INVALID_VAL)
919 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
921 struct rte_eth_xstat_name *xstats_names,
924 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
925 struct sfc_port *port = &sa->port;
926 unsigned int nb_supported;
930 if (unlikely(xstats_names == NULL && ids != NULL) ||
931 unlikely(xstats_names != NULL && ids == NULL))
934 if (unlikely(xstats_names == NULL && ids == NULL))
935 return sfc_xstats_get_nb_supported(sa);
938 * Names array could be filled in nonsequential order. Fill names with
939 * string indicating invalid ID first.
941 for (i = 0; i < size; i++)
942 xstats_names[i].name[0] = SFC_XSTAT_ID_INVALID_NAME;
944 sfc_adapter_lock(sa);
946 SFC_ASSERT(port->mac_stats_nb_supported <=
947 RTE_DIM(port->mac_stats_by_id));
949 for (i = 0; i < size; i++) {
950 if (ids[i] < port->mac_stats_nb_supported) {
951 strlcpy(xstats_names[i].name,
952 efx_mac_stat_name(sa->nic,
953 port->mac_stats_by_id[ids[i]]),
954 sizeof(xstats_names[0].name));
958 nb_supported = port->mac_stats_nb_supported;
960 sfc_adapter_unlock(sa);
962 ret = sfc_sw_xstats_get_names_by_id(sa, ids, xstats_names, size,
969 /* Return number of written names before invalid ID is encountered. */
970 for (i = 0; i < size; i++) {
971 if (xstats_names[i].name[0] == SFC_XSTAT_ID_INVALID_NAME)
979 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
981 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
982 unsigned int wanted_fc, link_fc;
984 memset(fc_conf, 0, sizeof(*fc_conf));
986 sfc_adapter_lock(sa);
988 if (sa->state == SFC_ETHDEV_STARTED)
989 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
991 link_fc = sa->port.flow_ctrl;
995 fc_conf->mode = RTE_ETH_FC_NONE;
997 case EFX_FCNTL_RESPOND:
998 fc_conf->mode = RTE_ETH_FC_RX_PAUSE;
1000 case EFX_FCNTL_GENERATE:
1001 fc_conf->mode = RTE_ETH_FC_TX_PAUSE;
1003 case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
1004 fc_conf->mode = RTE_ETH_FC_FULL;
1007 sfc_err(sa, "%s: unexpected flow control value %#x",
1011 fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
1013 sfc_adapter_unlock(sa);
1019 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1021 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1022 struct sfc_port *port = &sa->port;
1026 if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
1027 fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
1028 fc_conf->mac_ctrl_frame_fwd != 0) {
1029 sfc_err(sa, "unsupported flow control settings specified");
1034 switch (fc_conf->mode) {
1035 case RTE_ETH_FC_NONE:
1038 case RTE_ETH_FC_RX_PAUSE:
1039 fcntl = EFX_FCNTL_RESPOND;
1041 case RTE_ETH_FC_TX_PAUSE:
1042 fcntl = EFX_FCNTL_GENERATE;
1044 case RTE_ETH_FC_FULL:
1045 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
1052 sfc_adapter_lock(sa);
1054 if (sa->state == SFC_ETHDEV_STARTED) {
1055 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
1057 goto fail_mac_fcntl_set;
1060 port->flow_ctrl = fcntl;
1061 port->flow_ctrl_autoneg = fc_conf->autoneg;
1063 sfc_adapter_unlock(sa);
1068 sfc_adapter_unlock(sa);
1075 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
1077 struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
1078 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1079 boolean_t scatter_enabled;
1083 for (i = 0; i < sas->rxq_count; i++) {
1084 if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
1087 scatter_enabled = (sas->rxq_info[i].type_flags &
1088 EFX_RXQ_FLAG_SCATTER);
1090 if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
1091 encp->enc_rx_prefix_size,
1093 encp->enc_rx_scatter_max, &error)) {
1094 sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
1104 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
1106 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1107 size_t pdu = EFX_MAC_PDU(mtu);
1111 sfc_log_init(sa, "mtu=%u", mtu);
1114 if (pdu < EFX_MAC_PDU_MIN) {
1115 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
1116 (unsigned int)mtu, (unsigned int)pdu,
1120 if (pdu > EFX_MAC_PDU_MAX) {
1121 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
1122 (unsigned int)mtu, (unsigned int)pdu,
1123 (unsigned int)EFX_MAC_PDU_MAX);
1127 sfc_adapter_lock(sa);
1129 rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
1131 goto fail_check_scatter;
1133 if (pdu != sa->port.pdu) {
1134 if (sa->state == SFC_ETHDEV_STARTED) {
1137 old_pdu = sa->port.pdu;
1147 sfc_adapter_unlock(sa);
1149 sfc_log_init(sa, "done");
1153 sa->port.pdu = old_pdu;
1154 if (sfc_start(sa) != 0)
1155 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1156 "PDU max size - port is stopped",
1157 (unsigned int)pdu, (unsigned int)old_pdu);
1160 sfc_adapter_unlock(sa);
1163 sfc_log_init(sa, "failed %d", rc);
1168 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1170 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1171 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1172 struct sfc_port *port = &sa->port;
1173 struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1176 sfc_adapter_lock(sa);
1178 if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr))
1182 * Copy the address to the device private data so that
1183 * it could be recalled in the case of adapter restart.
1185 rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1188 * Neither of the two following checks can return
1189 * an error. The new MAC address is preserved in
1190 * the device private data and can be activated
1191 * on the next port start if the user prevents
1192 * isolated mode from being enabled.
1194 if (sfc_sa2shared(sa)->isolated) {
1195 sfc_warn(sa, "isolated mode is active on the port");
1196 sfc_warn(sa, "will not set MAC address");
1200 if (sa->state != SFC_ETHDEV_STARTED) {
1201 sfc_notice(sa, "the port is not started");
1202 sfc_notice(sa, "the new MAC address will be set on port start");
1207 if (encp->enc_allow_set_mac_with_installed_filters) {
1208 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1210 sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1215 * Changing the MAC address by means of MCDI request
1216 * has no effect on received traffic, therefore
1217 * we also need to update unicast filters
1219 rc = sfc_set_rx_mode_unchecked(sa);
1221 sfc_err(sa, "cannot set filter (rc = %u)", rc);
1222 /* Rollback the old address */
1223 (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1224 (void)sfc_set_rx_mode_unchecked(sa);
1227 sfc_warn(sa, "cannot set MAC address with filters installed");
1228 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1229 sfc_warn(sa, "(some traffic may be dropped)");
1232 * Since setting MAC address with filters installed is not
1233 * allowed on the adapter, the new MAC address will be set
1234 * by means of adapter restart. sfc_start() shall retrieve
1235 * the new address from the device private data and set it.
1240 sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1245 rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1247 sfc_adapter_unlock(sa);
1249 SFC_ASSERT(rc >= 0);
1255 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1256 struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1258 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1259 struct sfc_port *port = &sa->port;
1260 uint8_t *mc_addrs = port->mcast_addrs;
1264 if (sfc_sa2shared(sa)->isolated) {
1265 sfc_err(sa, "isolated mode is active on the port");
1266 sfc_err(sa, "will not set multicast address list");
1270 if (mc_addrs == NULL)
1273 if (nb_mc_addr > port->max_mcast_addrs) {
1274 sfc_err(sa, "too many multicast addresses: %u > %u",
1275 nb_mc_addr, port->max_mcast_addrs);
1279 for (i = 0; i < nb_mc_addr; ++i) {
1280 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1282 mc_addrs += EFX_MAC_ADDR_LEN;
1285 port->nb_mcast_addrs = nb_mc_addr;
1287 if (sa->state != SFC_ETHDEV_STARTED)
1290 rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1291 port->nb_mcast_addrs);
1293 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1295 SFC_ASSERT(rc >= 0);
1300 * The function is used by the secondary process as well. It must not
1301 * use any process-local pointers from the adapter data.
1304 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1305 struct rte_eth_rxq_info *qinfo)
1307 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1308 sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1309 struct sfc_rxq_info *rxq_info;
1311 rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1313 qinfo->mp = rxq_info->refill_mb_pool;
1314 qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1315 qinfo->conf.rx_drop_en = 1;
1316 qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1317 qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1318 if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1319 qinfo->conf.offloads |= RTE_ETH_RX_OFFLOAD_SCATTER;
1320 qinfo->scattered_rx = 1;
1322 qinfo->nb_desc = rxq_info->entries;
1326 * The function is used by the secondary process as well. It must not
1327 * use any process-local pointers from the adapter data.
1330 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1331 struct rte_eth_txq_info *qinfo)
1333 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1334 struct sfc_txq_info *txq_info;
1336 SFC_ASSERT(ethdev_qid < sas->ethdev_txq_count);
1338 txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1340 memset(qinfo, 0, sizeof(*qinfo));
1342 qinfo->conf.offloads = txq_info->offloads;
1343 qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1344 qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1345 qinfo->nb_desc = txq_info->entries;
1349 * The function is used by the secondary process as well. It must not
1350 * use any process-local pointers from the adapter data.
1353 sfc_rx_queue_count(void *rx_queue)
1355 struct sfc_dp_rxq *dp_rxq = rx_queue;
1356 const struct sfc_dp_rx *dp_rx;
1357 struct sfc_rxq_info *rxq_info;
1359 dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1360 rxq_info = sfc_rxq_info_by_dp_rxq(dp_rxq);
1362 if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1365 return dp_rx->qdesc_npending(dp_rxq);
1369 * The function is used by the secondary process as well. It must not
1370 * use any process-local pointers from the adapter data.
1373 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1375 struct sfc_dp_rxq *dp_rxq = queue;
1376 const struct sfc_dp_rx *dp_rx;
1378 dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1380 return dp_rx->qdesc_status(dp_rxq, offset);
1384 * The function is used by the secondary process as well. It must not
1385 * use any process-local pointers from the adapter data.
1388 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1390 struct sfc_dp_txq *dp_txq = queue;
1391 const struct sfc_dp_tx *dp_tx;
1393 dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1395 return dp_tx->qdesc_status(dp_txq, offset);
1399 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1401 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1402 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1403 sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1404 struct sfc_rxq_info *rxq_info;
1405 sfc_sw_index_t sw_index;
1408 sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1410 sfc_adapter_lock(sa);
1413 if (sa->state != SFC_ETHDEV_STARTED)
1414 goto fail_not_started;
1416 rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1417 if (rxq_info->state != SFC_RXQ_INITIALIZED)
1418 goto fail_not_setup;
1420 sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1421 rc = sfc_rx_qstart(sa, sw_index);
1423 goto fail_rx_qstart;
1425 rxq_info->deferred_started = B_TRUE;
1427 sfc_adapter_unlock(sa);
1434 sfc_adapter_unlock(sa);
1440 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1442 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1443 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1444 sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1445 struct sfc_rxq_info *rxq_info;
1446 sfc_sw_index_t sw_index;
1448 sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1450 sfc_adapter_lock(sa);
1452 sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1453 sfc_rx_qstop(sa, sw_index);
1455 rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1456 rxq_info->deferred_started = B_FALSE;
1458 sfc_adapter_unlock(sa);
1464 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1466 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1467 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1468 struct sfc_txq_info *txq_info;
1469 sfc_sw_index_t sw_index;
1472 sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1474 sfc_adapter_lock(sa);
1477 if (sa->state != SFC_ETHDEV_STARTED)
1478 goto fail_not_started;
1480 txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1481 if (txq_info->state != SFC_TXQ_INITIALIZED)
1482 goto fail_not_setup;
1484 sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1485 rc = sfc_tx_qstart(sa, sw_index);
1487 goto fail_tx_qstart;
1489 txq_info->deferred_started = B_TRUE;
1491 sfc_adapter_unlock(sa);
1498 sfc_adapter_unlock(sa);
1504 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1506 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1507 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1508 struct sfc_txq_info *txq_info;
1509 sfc_sw_index_t sw_index;
1511 sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1513 sfc_adapter_lock(sa);
1515 sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1516 sfc_tx_qstop(sa, sw_index);
1518 txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1519 txq_info->deferred_started = B_FALSE;
1521 sfc_adapter_unlock(sa);
1525 static efx_tunnel_protocol_t
1526 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1529 case RTE_ETH_TUNNEL_TYPE_VXLAN:
1530 return EFX_TUNNEL_PROTOCOL_VXLAN;
1531 case RTE_ETH_TUNNEL_TYPE_GENEVE:
1532 return EFX_TUNNEL_PROTOCOL_GENEVE;
1534 return EFX_TUNNEL_NPROTOS;
1538 enum sfc_udp_tunnel_op_e {
1539 SFC_UDP_TUNNEL_ADD_PORT,
1540 SFC_UDP_TUNNEL_DEL_PORT,
1544 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1545 struct rte_eth_udp_tunnel *tunnel_udp,
1546 enum sfc_udp_tunnel_op_e op)
1548 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1549 efx_tunnel_protocol_t tunnel_proto;
1552 sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1553 (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1554 (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1555 tunnel_udp->udp_port, tunnel_udp->prot_type);
1558 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1559 if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1561 goto fail_bad_proto;
1564 sfc_adapter_lock(sa);
1567 case SFC_UDP_TUNNEL_ADD_PORT:
1568 rc = efx_tunnel_config_udp_add(sa->nic,
1569 tunnel_udp->udp_port,
1572 case SFC_UDP_TUNNEL_DEL_PORT:
1573 rc = efx_tunnel_config_udp_remove(sa->nic,
1574 tunnel_udp->udp_port,
1585 if (sa->state == SFC_ETHDEV_STARTED) {
1586 rc = efx_tunnel_reconfigure(sa->nic);
1589 * Configuration is accepted by FW and MC reboot
1590 * is initiated to apply the changes. MC reboot
1591 * will be handled in a usual way (MC reboot
1592 * event on management event queue and adapter
1596 } else if (rc != 0) {
1597 goto fail_reconfigure;
1601 sfc_adapter_unlock(sa);
1605 /* Remove/restore entry since the change makes the trouble */
1607 case SFC_UDP_TUNNEL_ADD_PORT:
1608 (void)efx_tunnel_config_udp_remove(sa->nic,
1609 tunnel_udp->udp_port,
1612 case SFC_UDP_TUNNEL_DEL_PORT:
1613 (void)efx_tunnel_config_udp_add(sa->nic,
1614 tunnel_udp->udp_port,
1621 sfc_adapter_unlock(sa);
1629 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1630 struct rte_eth_udp_tunnel *tunnel_udp)
1632 return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1636 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1637 struct rte_eth_udp_tunnel *tunnel_udp)
1639 return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1643 * The function is used by the secondary process as well. It must not
1644 * use any process-local pointers from the adapter data.
1647 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1648 struct rte_eth_rss_conf *rss_conf)
1650 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1651 struct sfc_rss *rss = &sas->rss;
1653 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1657 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1658 * hence, conversion is done here to derive a correct set of RTE_ETH_RSS
1659 * flags which corresponds to the active EFX configuration stored
1660 * locally in 'sfc_adapter' and kept up-to-date
1662 rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1663 rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1664 if (rss_conf->rss_key != NULL)
1665 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1671 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1672 struct rte_eth_rss_conf *rss_conf)
1674 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1675 struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1676 unsigned int efx_hash_types;
1677 uint32_t contexts[] = {EFX_RSS_CONTEXT_DEFAULT, rss->dummy_rss_context};
1678 unsigned int n_contexts;
1679 unsigned int mode_i = 0;
1680 unsigned int key_i = 0;
1684 n_contexts = rss->dummy_rss_context == EFX_RSS_CONTEXT_DEFAULT ? 1 : 2;
1686 if (sfc_sa2shared(sa)->isolated)
1689 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1690 sfc_err(sa, "RSS is not available");
1694 if (rss->channels == 0) {
1695 sfc_err(sa, "RSS is not configured");
1699 if ((rss_conf->rss_key != NULL) &&
1700 (rss_conf->rss_key_len != sizeof(rss->key))) {
1701 sfc_err(sa, "RSS key size is wrong (should be %zu)",
1706 sfc_adapter_lock(sa);
1708 rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1710 goto fail_rx_hf_rte_to_efx;
1712 for (mode_i = 0; mode_i < n_contexts; mode_i++) {
1713 rc = efx_rx_scale_mode_set(sa->nic, contexts[mode_i],
1714 rss->hash_alg, efx_hash_types,
1717 goto fail_scale_mode_set;
1720 if (rss_conf->rss_key != NULL) {
1721 if (sa->state == SFC_ETHDEV_STARTED) {
1722 for (key_i = 0; key_i < n_contexts; key_i++) {
1723 rc = efx_rx_scale_key_set(sa->nic,
1728 goto fail_scale_key_set;
1732 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1735 rss->hash_types = efx_hash_types;
1737 sfc_adapter_unlock(sa);
1742 for (i = 0; i < key_i; i++) {
1743 if (efx_rx_scale_key_set(sa->nic, contexts[i], rss->key,
1744 sizeof(rss->key)) != 0)
1745 sfc_err(sa, "failed to restore RSS key");
1748 fail_scale_mode_set:
1749 for (i = 0; i < mode_i; i++) {
1750 if (efx_rx_scale_mode_set(sa->nic, contexts[i],
1751 EFX_RX_HASHALG_TOEPLITZ,
1752 rss->hash_types, B_TRUE) != 0)
1753 sfc_err(sa, "failed to restore RSS mode");
1756 fail_rx_hf_rte_to_efx:
1757 sfc_adapter_unlock(sa);
1762 * The function is used by the secondary process as well. It must not
1763 * use any process-local pointers from the adapter data.
1766 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1767 struct rte_eth_rss_reta_entry64 *reta_conf,
1770 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1771 struct sfc_rss *rss = &sas->rss;
1774 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1777 if (rss->channels == 0)
1780 if (reta_size != EFX_RSS_TBL_SIZE)
1783 for (entry = 0; entry < reta_size; entry++) {
1784 int grp = entry / RTE_ETH_RETA_GROUP_SIZE;
1785 int grp_idx = entry % RTE_ETH_RETA_GROUP_SIZE;
1787 if ((reta_conf[grp].mask >> grp_idx) & 1)
1788 reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1795 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1796 struct rte_eth_rss_reta_entry64 *reta_conf,
1799 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1800 struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1801 unsigned int *rss_tbl_new;
1806 if (sfc_sa2shared(sa)->isolated)
1809 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1810 sfc_err(sa, "RSS is not available");
1814 if (rss->channels == 0) {
1815 sfc_err(sa, "RSS is not configured");
1819 if (reta_size != EFX_RSS_TBL_SIZE) {
1820 sfc_err(sa, "RETA size is wrong (should be %u)",
1825 rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1826 if (rss_tbl_new == NULL)
1829 sfc_adapter_lock(sa);
1831 rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1833 for (entry = 0; entry < reta_size; entry++) {
1834 int grp_idx = entry % RTE_ETH_RETA_GROUP_SIZE;
1835 struct rte_eth_rss_reta_entry64 *grp;
1837 grp = &reta_conf[entry / RTE_ETH_RETA_GROUP_SIZE];
1839 if (grp->mask & (1ull << grp_idx)) {
1840 if (grp->reta[grp_idx] >= rss->channels) {
1842 goto bad_reta_entry;
1844 rss_tbl_new[entry] = grp->reta[grp_idx];
1848 if (sa->state == SFC_ETHDEV_STARTED) {
1849 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1850 rss_tbl_new, EFX_RSS_TBL_SIZE);
1852 goto fail_scale_tbl_set;
1855 rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1859 sfc_adapter_unlock(sa);
1861 rte_free(rss_tbl_new);
1863 SFC_ASSERT(rc >= 0);
1868 sfc_dev_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
1869 const struct rte_flow_ops **ops)
1871 *ops = &sfc_flow_ops;
1876 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1878 const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1881 * If Rx datapath does not provide callback to check mempool,
1882 * all pools are supported.
1884 if (sap->dp_rx->pool_ops_supported == NULL)
1887 return sap->dp_rx->pool_ops_supported(pool);
1891 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1893 const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1894 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1895 sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1896 struct sfc_rxq_info *rxq_info;
1898 rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1900 return sap->dp_rx->intr_enable(rxq_info->dp);
1904 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1906 const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1907 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1908 sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1909 struct sfc_rxq_info *rxq_info;
1911 rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1913 return sap->dp_rx->intr_disable(rxq_info->dp);
1916 struct sfc_mport_journal_ctx {
1917 struct sfc_adapter *sa;
1918 uint16_t switch_domain_id;
1919 uint32_t mcdi_handle;
1920 bool controllers_assigned;
1921 efx_pcie_interface_t *controllers;
1922 size_t nb_controllers;
1926 sfc_journal_ctx_add_controller(struct sfc_mport_journal_ctx *ctx,
1927 efx_pcie_interface_t intf)
1929 efx_pcie_interface_t *new_controllers;
1933 if (ctx->controllers == NULL) {
1934 ctx->controllers = rte_malloc("sfc_controller_mapping",
1935 sizeof(ctx->controllers[0]), 0);
1936 if (ctx->controllers == NULL)
1939 ctx->controllers[0] = intf;
1940 ctx->nb_controllers = 1;
1945 for (i = 0; i < ctx->nb_controllers; i++) {
1946 if (ctx->controllers[i] == intf)
1948 if (ctx->controllers[i] > intf)
1953 ctx->nb_controllers += 1;
1954 new_size = ctx->nb_controllers * sizeof(ctx->controllers[0]);
1956 new_controllers = rte_realloc(ctx->controllers, new_size, 0);
1957 if (new_controllers == NULL) {
1958 rte_free(ctx->controllers);
1961 ctx->controllers = new_controllers;
1963 for (i = target + 1; i < ctx->nb_controllers; i++)
1964 ctx->controllers[i] = ctx->controllers[i - 1];
1966 ctx->controllers[target] = intf;
1972 sfc_process_mport_journal_entry(struct sfc_mport_journal_ctx *ctx,
1973 efx_mport_desc_t *mport)
1975 struct sfc_mae_switch_port_request req;
1976 efx_mport_sel_t entity_selector;
1977 efx_mport_sel_t ethdev_mport;
1978 uint16_t switch_port_id;
1983 "processing mport id %u (controller %u pf %u vf %u)",
1984 mport->emd_id.id, mport->emd_vnic.ev_intf,
1985 mport->emd_vnic.ev_pf, mport->emd_vnic.ev_vf);
1986 efx_mae_mport_invalid(ðdev_mport);
1988 if (!ctx->controllers_assigned) {
1989 rc = sfc_journal_ctx_add_controller(ctx,
1990 mport->emd_vnic.ev_intf);
1995 /* Build Mport selector */
1996 efx_rc = efx_mae_mport_by_pcie_mh_function(mport->emd_vnic.ev_intf,
1997 mport->emd_vnic.ev_pf,
1998 mport->emd_vnic.ev_vf,
2001 sfc_err(ctx->sa, "failed to build entity mport selector for c%upf%uvf%u",
2002 mport->emd_vnic.ev_intf,
2003 mport->emd_vnic.ev_pf,
2004 mport->emd_vnic.ev_vf);
2008 rc = sfc_mae_switch_port_id_by_entity(ctx->switch_domain_id,
2010 SFC_MAE_SWITCH_PORT_REPRESENTOR,
2014 /* Already registered */
2018 * No representor has been created for this entity.
2019 * Create a dummy switch registry entry with an invalid ethdev
2020 * mport selector. When a corresponding representor is created,
2021 * this entry will be updated.
2023 req.type = SFC_MAE_SWITCH_PORT_REPRESENTOR;
2024 req.entity_mportp = &entity_selector;
2025 req.ethdev_mportp = ðdev_mport;
2026 req.ethdev_port_id = RTE_MAX_ETHPORTS;
2027 req.port_data.repr.intf = mport->emd_vnic.ev_intf;
2028 req.port_data.repr.pf = mport->emd_vnic.ev_pf;
2029 req.port_data.repr.vf = mport->emd_vnic.ev_vf;
2031 rc = sfc_mae_assign_switch_port(ctx->switch_domain_id,
2032 &req, &switch_port_id);
2035 "failed to assign MAE switch port for c%upf%uvf%u: %s",
2036 mport->emd_vnic.ev_intf,
2037 mport->emd_vnic.ev_pf,
2038 mport->emd_vnic.ev_vf,
2044 sfc_err(ctx->sa, "failed to find MAE switch port for c%upf%uvf%u: %s",
2045 mport->emd_vnic.ev_intf,
2046 mport->emd_vnic.ev_pf,
2047 mport->emd_vnic.ev_vf,
2056 sfc_process_mport_journal_cb(void *data, efx_mport_desc_t *mport,
2059 struct sfc_mport_journal_ctx *ctx = data;
2061 if (ctx == NULL || ctx->sa == NULL) {
2062 sfc_err(ctx->sa, "received NULL context or SFC adapter");
2066 if (mport_len != sizeof(*mport)) {
2067 sfc_err(ctx->sa, "actual and expected mport buffer sizes differ");
2071 SFC_ASSERT(sfc_adapter_is_locked(ctx->sa));
2074 * If a zombie flag is set, it means the mport has been marked for
2075 * deletion and cannot be used for any new operations. The mport will
2076 * be destroyed completely once all references to it are released.
2078 if (mport->emd_zombie) {
2079 sfc_dbg(ctx->sa, "mport is a zombie, skipping");
2082 if (mport->emd_type != EFX_MPORT_TYPE_VNIC) {
2083 sfc_dbg(ctx->sa, "mport is not a VNIC, skipping");
2086 if (mport->emd_vnic.ev_client_type != EFX_MPORT_VNIC_CLIENT_FUNCTION) {
2087 sfc_dbg(ctx->sa, "mport is not a function, skipping");
2090 if (mport->emd_vnic.ev_handle == ctx->mcdi_handle) {
2091 sfc_dbg(ctx->sa, "mport is this driver instance, skipping");
2095 return sfc_process_mport_journal_entry(ctx, mport);
2099 sfc_process_mport_journal(struct sfc_adapter *sa)
2101 struct sfc_mport_journal_ctx ctx;
2102 const efx_pcie_interface_t *controllers;
2103 size_t nb_controllers;
2107 memset(&ctx, 0, sizeof(ctx));
2109 ctx.switch_domain_id = sa->mae.switch_domain_id;
2111 efx_rc = efx_mcdi_get_own_client_handle(sa->nic, &ctx.mcdi_handle);
2113 sfc_err(sa, "failed to get own MCDI handle");
2114 SFC_ASSERT(efx_rc > 0);
2118 rc = sfc_mae_switch_domain_controllers(ctx.switch_domain_id,
2119 &controllers, &nb_controllers);
2121 sfc_err(sa, "failed to get controller mapping");
2125 ctx.controllers_assigned = controllers != NULL;
2126 ctx.controllers = NULL;
2127 ctx.nb_controllers = 0;
2129 efx_rc = efx_mae_read_mport_journal(sa->nic,
2130 sfc_process_mport_journal_cb, &ctx);
2132 sfc_err(sa, "failed to process MAE mport journal");
2133 SFC_ASSERT(efx_rc > 0);
2137 if (controllers == NULL) {
2138 rc = sfc_mae_switch_domain_map_controllers(ctx.switch_domain_id,
2140 ctx.nb_controllers);
2149 sfc_count_representors_cb(enum sfc_mae_switch_port_type type,
2150 const efx_mport_sel_t *ethdev_mportp __rte_unused,
2151 uint16_t ethdev_port_id __rte_unused,
2152 const efx_mport_sel_t *entity_mportp __rte_unused,
2153 uint16_t switch_port_id __rte_unused,
2154 union sfc_mae_switch_port_data *port_datap
2158 int *counter = user_datap;
2160 SFC_ASSERT(counter != NULL);
2162 if (type == SFC_MAE_SWITCH_PORT_REPRESENTOR)
2166 struct sfc_get_representors_ctx {
2167 struct rte_eth_representor_info *info;
2168 struct sfc_adapter *sa;
2169 uint16_t switch_domain_id;
2170 const efx_pcie_interface_t *controllers;
2171 size_t nb_controllers;
2175 sfc_get_representors_cb(enum sfc_mae_switch_port_type type,
2176 const efx_mport_sel_t *ethdev_mportp __rte_unused,
2177 uint16_t ethdev_port_id __rte_unused,
2178 const efx_mport_sel_t *entity_mportp __rte_unused,
2179 uint16_t switch_port_id,
2180 union sfc_mae_switch_port_data *port_datap,
2183 struct sfc_get_representors_ctx *ctx = user_datap;
2184 struct rte_eth_representor_range *range;
2188 SFC_ASSERT(ctx != NULL);
2189 SFC_ASSERT(ctx->info != NULL);
2190 SFC_ASSERT(ctx->sa != NULL);
2192 if (type != SFC_MAE_SWITCH_PORT_REPRESENTOR) {
2193 sfc_dbg(ctx->sa, "not a representor, skipping");
2196 if (ctx->info->nb_ranges >= ctx->info->nb_ranges_alloc) {
2197 sfc_dbg(ctx->sa, "info structure is full already");
2201 range = &ctx->info->ranges[ctx->info->nb_ranges];
2202 rc = sfc_mae_switch_controller_from_mapping(ctx->controllers,
2203 ctx->nb_controllers,
2204 port_datap->repr.intf,
2205 &range->controller);
2207 sfc_err(ctx->sa, "invalid representor controller: %d",
2208 port_datap->repr.intf);
2209 range->controller = -1;
2211 range->pf = port_datap->repr.pf;
2212 range->id_base = switch_port_id;
2213 range->id_end = switch_port_id;
2215 if (port_datap->repr.vf != EFX_PCI_VF_INVALID) {
2216 range->type = RTE_ETH_REPRESENTOR_VF;
2217 range->vf = port_datap->repr.vf;
2218 ret = snprintf(range->name, RTE_DEV_NAME_MAX_LEN,
2219 "c%dpf%dvf%d", range->controller, range->pf,
2222 range->type = RTE_ETH_REPRESENTOR_PF;
2223 ret = snprintf(range->name, RTE_DEV_NAME_MAX_LEN,
2224 "c%dpf%d", range->controller, range->pf);
2226 if (ret >= RTE_DEV_NAME_MAX_LEN) {
2227 sfc_err(ctx->sa, "representor name has been truncated: %s",
2231 ctx->info->nb_ranges++;
2235 sfc_representor_info_get(struct rte_eth_dev *dev,
2236 struct rte_eth_representor_info *info)
2238 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2239 struct sfc_get_representors_ctx get_repr_ctx;
2240 const efx_nic_cfg_t *nic_cfg;
2241 uint16_t switch_domain_id;
2246 sfc_adapter_lock(sa);
2248 if (sa->mae.status != SFC_MAE_STATUS_ADMIN) {
2249 sfc_adapter_unlock(sa);
2253 rc = sfc_process_mport_journal(sa);
2255 sfc_adapter_unlock(sa);
2260 switch_domain_id = sa->mae.switch_domain_id;
2263 rc = sfc_mae_switch_ports_iterate(switch_domain_id,
2264 sfc_count_representors_cb,
2267 sfc_adapter_unlock(sa);
2273 sfc_adapter_unlock(sa);
2277 rc = sfc_mae_switch_domain_controllers(switch_domain_id,
2278 &get_repr_ctx.controllers,
2279 &get_repr_ctx.nb_controllers);
2281 sfc_adapter_unlock(sa);
2286 nic_cfg = efx_nic_cfg_get(sa->nic);
2288 rc = sfc_mae_switch_domain_get_controller(switch_domain_id,
2292 sfc_err(sa, "invalid controller: %d", nic_cfg->enc_intf);
2296 info->controller = controller;
2297 info->pf = nic_cfg->enc_pf;
2299 get_repr_ctx.info = info;
2300 get_repr_ctx.sa = sa;
2301 get_repr_ctx.switch_domain_id = switch_domain_id;
2302 rc = sfc_mae_switch_ports_iterate(switch_domain_id,
2303 sfc_get_representors_cb,
2306 sfc_adapter_unlock(sa);
2311 sfc_adapter_unlock(sa);
2316 sfc_rx_metadata_negotiate(struct rte_eth_dev *dev, uint64_t *features)
2318 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2319 uint64_t supported = 0;
2321 sfc_adapter_lock(sa);
2323 if ((sa->priv.dp_rx->features & SFC_DP_RX_FEAT_FLOW_FLAG) != 0)
2324 supported |= RTE_ETH_RX_METADATA_USER_FLAG;
2326 if ((sa->priv.dp_rx->features & SFC_DP_RX_FEAT_FLOW_MARK) != 0)
2327 supported |= RTE_ETH_RX_METADATA_USER_MARK;
2329 if (sfc_flow_tunnel_is_supported(sa))
2330 supported |= RTE_ETH_RX_METADATA_TUNNEL_ID;
2332 sa->negotiated_rx_metadata = supported & *features;
2333 *features = sa->negotiated_rx_metadata;
2335 sfc_adapter_unlock(sa);
2340 static const struct eth_dev_ops sfc_eth_dev_ops = {
2341 .dev_configure = sfc_dev_configure,
2342 .dev_start = sfc_dev_start,
2343 .dev_stop = sfc_dev_stop,
2344 .dev_set_link_up = sfc_dev_set_link_up,
2345 .dev_set_link_down = sfc_dev_set_link_down,
2346 .dev_close = sfc_dev_close,
2347 .promiscuous_enable = sfc_dev_promisc_enable,
2348 .promiscuous_disable = sfc_dev_promisc_disable,
2349 .allmulticast_enable = sfc_dev_allmulti_enable,
2350 .allmulticast_disable = sfc_dev_allmulti_disable,
2351 .link_update = sfc_dev_link_update,
2352 .stats_get = sfc_stats_get,
2353 .stats_reset = sfc_stats_reset,
2354 .xstats_get = sfc_xstats_get,
2355 .xstats_reset = sfc_stats_reset,
2356 .xstats_get_names = sfc_xstats_get_names,
2357 .dev_infos_get = sfc_dev_infos_get,
2358 .dev_supported_ptypes_get = sfc_dev_supported_ptypes_get,
2359 .mtu_set = sfc_dev_set_mtu,
2360 .rx_queue_start = sfc_rx_queue_start,
2361 .rx_queue_stop = sfc_rx_queue_stop,
2362 .tx_queue_start = sfc_tx_queue_start,
2363 .tx_queue_stop = sfc_tx_queue_stop,
2364 .rx_queue_setup = sfc_rx_queue_setup,
2365 .rx_queue_release = sfc_rx_queue_release,
2366 .rx_queue_intr_enable = sfc_rx_queue_intr_enable,
2367 .rx_queue_intr_disable = sfc_rx_queue_intr_disable,
2368 .tx_queue_setup = sfc_tx_queue_setup,
2369 .tx_queue_release = sfc_tx_queue_release,
2370 .flow_ctrl_get = sfc_flow_ctrl_get,
2371 .flow_ctrl_set = sfc_flow_ctrl_set,
2372 .mac_addr_set = sfc_mac_addr_set,
2373 .udp_tunnel_port_add = sfc_dev_udp_tunnel_port_add,
2374 .udp_tunnel_port_del = sfc_dev_udp_tunnel_port_del,
2375 .reta_update = sfc_dev_rss_reta_update,
2376 .reta_query = sfc_dev_rss_reta_query,
2377 .rss_hash_update = sfc_dev_rss_hash_update,
2378 .rss_hash_conf_get = sfc_dev_rss_hash_conf_get,
2379 .flow_ops_get = sfc_dev_flow_ops_get,
2380 .set_mc_addr_list = sfc_set_mc_addr_list,
2381 .rxq_info_get = sfc_rx_queue_info_get,
2382 .txq_info_get = sfc_tx_queue_info_get,
2383 .fw_version_get = sfc_fw_version_get,
2384 .xstats_get_by_id = sfc_xstats_get_by_id,
2385 .xstats_get_names_by_id = sfc_xstats_get_names_by_id,
2386 .pool_ops_supported = sfc_pool_ops_supported,
2387 .representor_info_get = sfc_representor_info_get,
2388 .rx_metadata_negotiate = sfc_rx_metadata_negotiate,
2391 struct sfc_ethdev_init_data {
2392 uint16_t nb_representors;
2396 * Duplicate a string in potentially shared memory required for
2397 * multi-process support.
2399 * strdup() allocates from process-local heap/memory.
2402 sfc_strdup(const char *str)
2410 size = strlen(str) + 1;
2411 copy = rte_malloc(__func__, size, 0);
2413 rte_memcpy(copy, str, size);
2419 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
2421 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2422 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2423 const struct sfc_dp_rx *dp_rx;
2424 const struct sfc_dp_tx *dp_tx;
2425 const efx_nic_cfg_t *encp;
2426 unsigned int avail_caps = 0;
2427 const char *rx_name = NULL;
2428 const char *tx_name = NULL;
2431 switch (sa->family) {
2432 case EFX_FAMILY_HUNTINGTON:
2433 case EFX_FAMILY_MEDFORD:
2434 case EFX_FAMILY_MEDFORD2:
2435 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
2436 avail_caps |= SFC_DP_HW_FW_CAP_RX_EFX;
2437 avail_caps |= SFC_DP_HW_FW_CAP_TX_EFX;
2439 case EFX_FAMILY_RIVERHEAD:
2440 avail_caps |= SFC_DP_HW_FW_CAP_EF100;
2446 encp = efx_nic_cfg_get(sa->nic);
2447 if (encp->enc_rx_es_super_buffer_supported)
2448 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
2450 rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
2451 sfc_kvarg_string_handler, &rx_name);
2453 goto fail_kvarg_rx_datapath;
2455 if (rx_name != NULL) {
2456 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
2457 if (dp_rx == NULL) {
2458 sfc_err(sa, "Rx datapath %s not found", rx_name);
2462 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
2464 "Insufficient Hw/FW capabilities to use Rx datapath %s",
2467 goto fail_dp_rx_caps;
2470 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
2471 if (dp_rx == NULL) {
2472 sfc_err(sa, "Rx datapath by caps %#x not found",
2479 sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
2480 if (sas->dp_rx_name == NULL) {
2482 goto fail_dp_rx_name;
2485 if (strcmp(dp_rx->dp.name, SFC_KVARG_DATAPATH_EF10_ESSB) == 0) {
2486 /* FLAG and MARK are always available from Rx prefix. */
2487 sa->negotiated_rx_metadata |= RTE_ETH_RX_METADATA_USER_FLAG;
2488 sa->negotiated_rx_metadata |= RTE_ETH_RX_METADATA_USER_MARK;
2491 sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
2493 rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
2494 sfc_kvarg_string_handler, &tx_name);
2496 goto fail_kvarg_tx_datapath;
2498 if (tx_name != NULL) {
2499 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
2500 if (dp_tx == NULL) {
2501 sfc_err(sa, "Tx datapath %s not found", tx_name);
2505 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
2507 "Insufficient Hw/FW capabilities to use Tx datapath %s",
2510 goto fail_dp_tx_caps;
2513 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
2514 if (dp_tx == NULL) {
2515 sfc_err(sa, "Tx datapath by caps %#x not found",
2522 sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
2523 if (sas->dp_tx_name == NULL) {
2525 goto fail_dp_tx_name;
2528 sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
2530 sa->priv.dp_rx = dp_rx;
2531 sa->priv.dp_tx = dp_tx;
2533 dev->rx_pkt_burst = dp_rx->pkt_burst;
2534 dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2535 dev->tx_pkt_burst = dp_tx->pkt_burst;
2537 dev->rx_queue_count = sfc_rx_queue_count;
2538 dev->rx_descriptor_status = sfc_rx_descriptor_status;
2539 dev->tx_descriptor_status = sfc_tx_descriptor_status;
2540 dev->dev_ops = &sfc_eth_dev_ops;
2547 fail_kvarg_tx_datapath:
2548 rte_free(sas->dp_rx_name);
2549 sas->dp_rx_name = NULL;
2554 fail_kvarg_rx_datapath:
2559 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
2561 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2562 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2564 dev->dev_ops = NULL;
2565 dev->tx_pkt_prepare = NULL;
2566 dev->rx_pkt_burst = NULL;
2567 dev->tx_pkt_burst = NULL;
2569 rte_free(sas->dp_tx_name);
2570 sas->dp_tx_name = NULL;
2571 sa->priv.dp_tx = NULL;
2573 rte_free(sas->dp_rx_name);
2574 sas->dp_rx_name = NULL;
2575 sa->priv.dp_rx = NULL;
2578 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2579 .dev_supported_ptypes_get = sfc_dev_supported_ptypes_get,
2580 .reta_query = sfc_dev_rss_reta_query,
2581 .rss_hash_conf_get = sfc_dev_rss_hash_conf_get,
2582 .rxq_info_get = sfc_rx_queue_info_get,
2583 .txq_info_get = sfc_tx_queue_info_get,
2587 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2589 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2590 struct sfc_adapter_priv *sap;
2591 const struct sfc_dp_rx *dp_rx;
2592 const struct sfc_dp_tx *dp_tx;
2596 * Allocate process private data from heap, since it should not
2597 * be located in shared memory allocated using rte_malloc() API.
2599 sap = calloc(1, sizeof(*sap));
2602 goto fail_alloc_priv;
2605 sap->logtype_main = logtype_main;
2607 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2608 if (dp_rx == NULL) {
2609 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2610 "cannot find %s Rx datapath", sas->dp_rx_name);
2614 if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2615 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2616 "%s Rx datapath does not support multi-process",
2619 goto fail_dp_rx_multi_process;
2622 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2623 if (dp_tx == NULL) {
2624 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2625 "cannot find %s Tx datapath", sas->dp_tx_name);
2629 if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2630 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2631 "%s Tx datapath does not support multi-process",
2634 goto fail_dp_tx_multi_process;
2640 dev->process_private = sap;
2641 dev->rx_pkt_burst = dp_rx->pkt_burst;
2642 dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2643 dev->tx_pkt_burst = dp_tx->pkt_burst;
2644 dev->rx_queue_count = sfc_rx_queue_count;
2645 dev->rx_descriptor_status = sfc_rx_descriptor_status;
2646 dev->tx_descriptor_status = sfc_tx_descriptor_status;
2647 dev->dev_ops = &sfc_eth_dev_secondary_ops;
2651 fail_dp_tx_multi_process:
2653 fail_dp_rx_multi_process:
2662 sfc_register_dp(void)
2665 if (TAILQ_EMPTY(&sfc_dp_head)) {
2666 /* Prefer EF10 datapath */
2667 sfc_dp_register(&sfc_dp_head, &sfc_ef100_rx.dp);
2668 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2669 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2670 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2672 sfc_dp_register(&sfc_dp_head, &sfc_ef100_tx.dp);
2673 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2674 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2675 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2680 sfc_parse_switch_mode(struct sfc_adapter *sa, bool has_representors)
2682 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
2683 const char *switch_mode = NULL;
2686 sfc_log_init(sa, "entry");
2688 rc = sfc_kvargs_process(sa, SFC_KVARG_SWITCH_MODE,
2689 sfc_kvarg_string_handler, &switch_mode);
2693 if (switch_mode == NULL) {
2694 sa->switchdev = encp->enc_mae_admin &&
2695 (!encp->enc_datapath_cap_evb ||
2697 } else if (strcasecmp(switch_mode, SFC_KVARG_SWITCH_MODE_LEGACY) == 0) {
2698 sa->switchdev = false;
2699 } else if (strcasecmp(switch_mode,
2700 SFC_KVARG_SWITCH_MODE_SWITCHDEV) == 0) {
2701 sa->switchdev = true;
2703 sfc_err(sa, "invalid switch mode device argument '%s'",
2709 sfc_log_init(sa, "done");
2715 sfc_log_init(sa, "failed: %s", rte_strerror(rc));
2721 sfc_eth_dev_init(struct rte_eth_dev *dev, void *init_params)
2723 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2724 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2725 struct sfc_ethdev_init_data *init_data = init_params;
2726 uint32_t logtype_main;
2727 struct sfc_adapter *sa;
2729 const efx_nic_cfg_t *encp;
2730 const struct rte_ether_addr *from;
2733 if (sfc_efx_dev_class_get(pci_dev->device.devargs) !=
2734 SFC_EFX_DEV_CLASS_NET) {
2735 SFC_GENERIC_LOG(DEBUG,
2736 "Incompatible device class: skip probing, should be probed by other sfc driver.");
2740 rc = sfc_dp_mport_register();
2746 logtype_main = sfc_register_logtype(&pci_dev->addr,
2747 SFC_LOGTYPE_MAIN_STR,
2750 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2751 return -sfc_eth_dev_secondary_init(dev, logtype_main);
2753 /* Required for logging */
2754 ret = snprintf(sas->log_prefix, sizeof(sas->log_prefix),
2755 "PMD: sfc_efx " PCI_PRI_FMT " #%" PRIu16 ": ",
2756 pci_dev->addr.domain, pci_dev->addr.bus,
2757 pci_dev->addr.devid, pci_dev->addr.function,
2758 dev->data->port_id);
2759 if (ret < 0 || ret >= (int)sizeof(sas->log_prefix)) {
2760 SFC_GENERIC_LOG(ERR,
2761 "reserved log prefix is too short for " PCI_PRI_FMT,
2762 pci_dev->addr.domain, pci_dev->addr.bus,
2763 pci_dev->addr.devid, pci_dev->addr.function);
2766 sas->pci_addr = pci_dev->addr;
2767 sas->port_id = dev->data->port_id;
2770 * Allocate process private data from heap, since it should not
2771 * be located in shared memory allocated using rte_malloc() API.
2773 sa = calloc(1, sizeof(*sa));
2779 dev->process_private = sa;
2781 /* Required for logging */
2782 sa->priv.shared = sas;
2783 sa->priv.logtype_main = logtype_main;
2787 /* Copy PCI device info to the dev->data */
2788 rte_eth_copy_pci_info(dev, pci_dev);
2789 dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE;
2791 rc = sfc_kvargs_parse(sa);
2793 goto fail_kvargs_parse;
2795 sfc_log_init(sa, "entry");
2797 dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2798 if (dev->data->mac_addrs == NULL) {
2800 goto fail_mac_addrs;
2803 sfc_adapter_lock_init(sa);
2804 sfc_adapter_lock(sa);
2806 sfc_log_init(sa, "probing");
2812 * Selecting a default switch mode requires the NIC to be probed and
2813 * to have its capabilities filled in.
2815 rc = sfc_parse_switch_mode(sa, init_data->nb_representors > 0);
2817 goto fail_switch_mode;
2819 sfc_log_init(sa, "set device ops");
2820 rc = sfc_eth_dev_set_ops(dev);
2824 sfc_log_init(sa, "attaching");
2825 rc = sfc_attach(sa);
2829 if (sa->switchdev && sa->mae.status != SFC_MAE_STATUS_ADMIN) {
2831 "failed to enable switchdev mode without admin MAE privilege");
2833 goto fail_switchdev_no_mae;
2836 encp = efx_nic_cfg_get(sa->nic);
2839 * The arguments are really reverse order in comparison to
2840 * Linux kernel. Copy from NIC config to Ethernet device data.
2842 from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2843 rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2846 * Setup the NIC DMA mapping handler. All internal mempools
2847 * MUST be created on attach before this point, and the
2848 * adapter MUST NOT create mempools with the adapter lock
2849 * held after this point.
2851 rc = sfc_nic_dma_attach(sa);
2853 goto fail_nic_dma_attach;
2855 sfc_adapter_unlock(sa);
2857 sfc_log_init(sa, "done");
2860 fail_nic_dma_attach:
2861 fail_switchdev_no_mae:
2865 sfc_eth_dev_clear_ops(dev);
2872 sfc_adapter_unlock(sa);
2873 sfc_adapter_lock_fini(sa);
2874 rte_free(dev->data->mac_addrs);
2875 dev->data->mac_addrs = NULL;
2878 sfc_kvargs_cleanup(sa);
2881 sfc_log_init(sa, "failed %d", rc);
2882 dev->process_private = NULL;
2891 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2898 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2899 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2900 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2901 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2902 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2903 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2904 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2905 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2906 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2907 { RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD) },
2908 { RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD_VF) },
2909 { .vendor_id = 0 /* sentinel */ }
2913 sfc_parse_rte_devargs(const char *args, struct rte_eth_devargs *devargs)
2915 struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
2919 rc = rte_eth_devargs_parse(args, ð_da);
2921 SFC_GENERIC_LOG(ERR,
2922 "Failed to parse generic devargs '%s'",
2934 sfc_eth_dev_find_or_create(struct rte_pci_device *pci_dev,
2935 struct sfc_ethdev_init_data *init_data,
2936 struct rte_eth_dev **devp,
2939 struct rte_eth_dev *dev;
2940 bool created = false;
2943 dev = rte_eth_dev_allocated(pci_dev->device.name);
2945 rc = rte_eth_dev_create(&pci_dev->device, pci_dev->device.name,
2946 sizeof(struct sfc_adapter_shared),
2947 eth_dev_pci_specific_init, pci_dev,
2948 sfc_eth_dev_init, init_data);
2950 SFC_GENERIC_LOG(ERR, "Failed to create sfc ethdev '%s'",
2951 pci_dev->device.name);
2957 dev = rte_eth_dev_allocated(pci_dev->device.name);
2959 SFC_GENERIC_LOG(ERR,
2960 "Failed to find allocated sfc ethdev '%s'",
2961 pci_dev->device.name);
2967 *dev_created = created;
2973 sfc_eth_dev_create_repr(struct sfc_adapter *sa,
2974 efx_pcie_interface_t controller,
2977 enum rte_eth_representor_type type)
2979 struct sfc_repr_entity_info entity;
2980 efx_mport_sel_t mport_sel;
2984 case RTE_ETH_REPRESENTOR_NONE:
2986 case RTE_ETH_REPRESENTOR_VF:
2987 case RTE_ETH_REPRESENTOR_PF:
2989 case RTE_ETH_REPRESENTOR_SF:
2990 sfc_err(sa, "SF representors are not supported");
2993 sfc_err(sa, "unknown representor type: %d", type);
2997 rc = efx_mae_mport_by_pcie_mh_function(controller,
3003 "failed to get m-port selector for controller %u port %u repr_port %u: %s",
3004 controller, port, repr_port, rte_strerror(-rc));
3008 memset(&entity, 0, sizeof(entity));
3010 entity.intf = controller;
3012 entity.vf = repr_port;
3014 rc = sfc_repr_create(sa->eth_dev, &entity, sa->mae.switch_domain_id,
3018 "failed to create representor for controller %u port %u repr_port %u: %s",
3019 controller, port, repr_port, rte_strerror(-rc));
3027 sfc_eth_dev_create_repr_port(struct sfc_adapter *sa,
3028 const struct rte_eth_devargs *eth_da,
3029 efx_pcie_interface_t controller,
3032 int first_error = 0;
3036 if (eth_da->type == RTE_ETH_REPRESENTOR_PF) {
3037 return sfc_eth_dev_create_repr(sa, controller, port,
3042 for (i = 0; i < eth_da->nb_representor_ports; i++) {
3043 rc = sfc_eth_dev_create_repr(sa, controller, port,
3044 eth_da->representor_ports[i],
3046 if (rc != 0 && first_error == 0)
3054 sfc_eth_dev_create_repr_controller(struct sfc_adapter *sa,
3055 const struct rte_eth_devargs *eth_da,
3056 efx_pcie_interface_t controller)
3058 const efx_nic_cfg_t *encp;
3059 int first_error = 0;
3060 uint16_t default_port;
3064 if (eth_da->nb_ports == 0) {
3065 encp = efx_nic_cfg_get(sa->nic);
3066 default_port = encp->enc_intf == controller ? encp->enc_pf : 0;
3067 return sfc_eth_dev_create_repr_port(sa, eth_da, controller,
3071 for (i = 0; i < eth_da->nb_ports; i++) {
3072 rc = sfc_eth_dev_create_repr_port(sa, eth_da, controller,
3074 if (rc != 0 && first_error == 0)
3082 sfc_eth_dev_create_representors(struct rte_eth_dev *dev,
3083 const struct rte_eth_devargs *eth_da)
3085 efx_pcie_interface_t intf;
3086 const efx_nic_cfg_t *encp;
3087 struct sfc_adapter *sa;
3088 uint16_t switch_domain_id;
3092 sa = sfc_adapter_by_eth_dev(dev);
3093 switch_domain_id = sa->mae.switch_domain_id;
3095 switch (eth_da->type) {
3096 case RTE_ETH_REPRESENTOR_NONE:
3098 case RTE_ETH_REPRESENTOR_PF:
3099 case RTE_ETH_REPRESENTOR_VF:
3101 case RTE_ETH_REPRESENTOR_SF:
3102 sfc_err(sa, "SF representors are not supported");
3105 sfc_err(sa, "unknown representor type: %d",
3110 if (!sa->switchdev) {
3111 sfc_err(sa, "cannot create representors in non-switchdev mode");
3115 if (!sfc_repr_available(sfc_sa2shared(sa))) {
3116 sfc_err(sa, "cannot create representors: unsupported");
3122 * This is needed to construct the DPDK controller -> EFX interface
3125 sfc_adapter_lock(sa);
3126 rc = sfc_process_mport_journal(sa);
3127 sfc_adapter_unlock(sa);
3133 if (eth_da->nb_mh_controllers > 0) {
3134 for (i = 0; i < eth_da->nb_mh_controllers; i++) {
3135 rc = sfc_mae_switch_domain_get_intf(switch_domain_id,
3136 eth_da->mh_controllers[i],
3139 sfc_err(sa, "failed to get representor");
3142 sfc_eth_dev_create_repr_controller(sa, eth_da, intf);
3145 encp = efx_nic_cfg_get(sa->nic);
3146 sfc_eth_dev_create_repr_controller(sa, eth_da, encp->enc_intf);
3152 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
3153 struct rte_pci_device *pci_dev)
3155 struct sfc_ethdev_init_data init_data;
3156 struct rte_eth_devargs eth_da;
3157 struct rte_eth_dev *dev;
3161 if (pci_dev->device.devargs != NULL) {
3162 rc = sfc_parse_rte_devargs(pci_dev->device.devargs->args,
3167 memset(ð_da, 0, sizeof(eth_da));
3170 /* If no VF representors specified, check for PF ones */
3171 if (eth_da.nb_representor_ports > 0)
3172 init_data.nb_representors = eth_da.nb_representor_ports;
3174 init_data.nb_representors = eth_da.nb_ports;
3176 if (init_data.nb_representors > 0 &&
3177 rte_eal_process_type() != RTE_PROC_PRIMARY) {
3178 SFC_GENERIC_LOG(ERR,
3179 "Create representors from secondary process not supported, dev '%s'",
3180 pci_dev->device.name);
3185 * Driver supports RTE_PCI_DRV_PROBE_AGAIN. Hence create device only
3186 * if it does not already exist. Re-probing an existing device is
3187 * expected to allow additional representors to be configured.
3189 rc = sfc_eth_dev_find_or_create(pci_dev, &init_data, &dev,
3194 rc = sfc_eth_dev_create_representors(dev, ð_da);
3197 (void)rte_eth_dev_destroy(dev, sfc_eth_dev_uninit);
3205 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
3207 return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
3210 static struct rte_pci_driver sfc_efx_pmd = {
3211 .id_table = pci_id_sfc_efx_map,
3213 RTE_PCI_DRV_INTR_LSC |
3214 RTE_PCI_DRV_NEED_MAPPING |
3215 RTE_PCI_DRV_PROBE_AGAIN,
3216 .probe = sfc_eth_dev_pci_probe,
3217 .remove = sfc_eth_dev_pci_remove,
3220 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
3221 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
3222 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
3223 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
3224 SFC_KVARG_SWITCH_MODE "=" SFC_KVARG_VALUES_SWITCH_MODE " "
3225 SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
3226 SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
3227 SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
3228 SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
3229 SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
3230 SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
3232 RTE_INIT(sfc_driver_register_logtype)
3236 ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
3238 sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;