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.h>
12 #include <rte_ethdev_pci.h>
14 #include <rte_bus_pci.h>
15 #include <rte_errno.h>
20 #include "sfc_debug.h"
22 #include "sfc_kvargs.h"
28 #include "sfc_dp_rx.h"
30 static struct sfc_dp_list sfc_dp_head =
31 TAILQ_HEAD_INITIALIZER(sfc_dp_head);
34 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
36 struct sfc_adapter *sa = dev->data->dev_private;
37 efx_nic_fw_info_t enfi;
42 * Return value of the callback is likely supposed to be
43 * equal to or greater than 0, nevertheless, if an error
44 * occurs, it will be desirable to pass it to the caller
46 if ((fw_version == NULL) || (fw_size == 0))
49 rc = efx_nic_get_fw_version(sa->nic, &enfi);
53 ret = snprintf(fw_version, fw_size,
54 "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
55 enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
56 enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
60 if (enfi.enfi_dpcpu_fw_ids_valid) {
61 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
64 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
65 fw_size - dpcpu_fw_ids_offset,
66 " rx%" PRIx16 " tx%" PRIx16,
67 enfi.enfi_rx_dpcpu_fw_id,
68 enfi.enfi_tx_dpcpu_fw_id);
75 if (fw_size < (size_t)(++ret))
82 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
84 struct sfc_adapter *sa = dev->data->dev_private;
85 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
87 sfc_log_init(sa, "entry");
89 dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
90 dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
92 /* Autonegotiation may be disabled */
93 dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
94 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
95 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
96 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
97 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
98 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
99 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
101 dev_info->max_rx_queues = sa->rxq_max;
102 dev_info->max_tx_queues = sa->txq_max;
104 /* By default packets are dropped if no descriptors are available */
105 dev_info->default_rxconf.rx_drop_en = 1;
107 dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
110 * rx_offload_capa includes both device and queue offloads since
111 * the latter may be requested on a per device basis which makes
112 * sense when some offloads are needed to be set on all queues.
114 dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
115 dev_info->rx_queue_offload_capa;
117 dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa);
119 dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
120 if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
121 !encp->enc_hw_tx_insert_vlan_enabled)
122 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
124 if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
125 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
127 if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_POOL)
128 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTMEMP;
130 if (~sa->dp_tx->features & SFC_DP_TX_FEAT_REFCNT)
131 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOREFCOUNT;
133 #if EFSYS_OPT_RX_SCALE
134 if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
135 dev_info->reta_size = EFX_RSS_TBL_SIZE;
136 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
137 dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
141 /* Initialize to hardware limits */
142 dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
143 dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
144 /* The RXQ hardware requires that the descriptor count is a power
145 * of 2, but rx_desc_lim cannot properly describe that constraint.
147 dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
149 /* Initialize to hardware limits */
150 dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
151 dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
153 * The TXQ hardware requires that the descriptor count is a power
154 * of 2, but tx_desc_lim cannot properly describe that constraint
156 dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
158 if (sa->dp_rx->get_dev_info != NULL)
159 sa->dp_rx->get_dev_info(dev_info);
160 if (sa->dp_tx->get_dev_info != NULL)
161 sa->dp_tx->get_dev_info(dev_info);
164 static const uint32_t *
165 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
167 struct sfc_adapter *sa = dev->data->dev_private;
168 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
169 uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
171 return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
175 sfc_dev_configure(struct rte_eth_dev *dev)
177 struct rte_eth_dev_data *dev_data = dev->data;
178 struct sfc_adapter *sa = dev_data->dev_private;
181 sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
182 dev_data->nb_rx_queues, dev_data->nb_tx_queues);
184 sfc_adapter_lock(sa);
186 case SFC_ADAPTER_CONFIGURED:
188 case SFC_ADAPTER_INITIALIZED:
189 rc = sfc_configure(sa);
192 sfc_err(sa, "unexpected adapter state %u to configure",
197 sfc_adapter_unlock(sa);
199 sfc_log_init(sa, "done %d", rc);
205 sfc_dev_start(struct rte_eth_dev *dev)
207 struct sfc_adapter *sa = dev->data->dev_private;
210 sfc_log_init(sa, "entry");
212 sfc_adapter_lock(sa);
214 sfc_adapter_unlock(sa);
216 sfc_log_init(sa, "done %d", rc);
222 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
224 struct sfc_adapter *sa = dev->data->dev_private;
225 struct rte_eth_link *dev_link = &dev->data->dev_link;
226 struct rte_eth_link old_link;
227 struct rte_eth_link current_link;
229 sfc_log_init(sa, "entry");
232 EFX_STATIC_ASSERT(sizeof(*dev_link) == sizeof(rte_atomic64_t));
233 *(int64_t *)&old_link = rte_atomic64_read((rte_atomic64_t *)dev_link);
235 if (sa->state != SFC_ADAPTER_STARTED) {
236 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, ¤t_link);
237 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
238 *(uint64_t *)&old_link,
239 *(uint64_t *)¤t_link))
241 } else if (wait_to_complete) {
242 efx_link_mode_t link_mode;
244 if (efx_port_poll(sa->nic, &link_mode) != 0)
245 link_mode = EFX_LINK_UNKNOWN;
246 sfc_port_link_mode_to_info(link_mode, ¤t_link);
248 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
249 *(uint64_t *)&old_link,
250 *(uint64_t *)¤t_link))
253 sfc_ev_mgmt_qpoll(sa);
254 *(int64_t *)¤t_link =
255 rte_atomic64_read((rte_atomic64_t *)dev_link);
258 if (old_link.link_status != current_link.link_status)
259 sfc_info(sa, "Link status is %s",
260 current_link.link_status ? "UP" : "DOWN");
262 return old_link.link_status == current_link.link_status ? 0 : -1;
266 sfc_dev_stop(struct rte_eth_dev *dev)
268 struct sfc_adapter *sa = dev->data->dev_private;
270 sfc_log_init(sa, "entry");
272 sfc_adapter_lock(sa);
274 sfc_adapter_unlock(sa);
276 sfc_log_init(sa, "done");
280 sfc_dev_set_link_up(struct rte_eth_dev *dev)
282 struct sfc_adapter *sa = dev->data->dev_private;
285 sfc_log_init(sa, "entry");
287 sfc_adapter_lock(sa);
289 sfc_adapter_unlock(sa);
296 sfc_dev_set_link_down(struct rte_eth_dev *dev)
298 struct sfc_adapter *sa = dev->data->dev_private;
300 sfc_log_init(sa, "entry");
302 sfc_adapter_lock(sa);
304 sfc_adapter_unlock(sa);
310 sfc_dev_close(struct rte_eth_dev *dev)
312 struct sfc_adapter *sa = dev->data->dev_private;
314 sfc_log_init(sa, "entry");
316 sfc_adapter_lock(sa);
318 case SFC_ADAPTER_STARTED:
320 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
322 case SFC_ADAPTER_CONFIGURED:
324 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
326 case SFC_ADAPTER_INITIALIZED:
329 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
332 sfc_adapter_unlock(sa);
334 sfc_log_init(sa, "done");
338 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
341 struct sfc_port *port;
343 struct sfc_adapter *sa = dev->data->dev_private;
344 boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
345 const char *desc = (allmulti) ? "all-multi" : "promiscuous";
347 sfc_adapter_lock(sa);
350 toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
352 if (*toggle != enabled) {
355 if (port->isolated) {
356 sfc_warn(sa, "isolated mode is active on the port");
357 sfc_warn(sa, "the change is to be applied on the next "
358 "start provided that isolated mode is "
359 "disabled prior the next start");
360 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
361 (sfc_set_rx_mode(sa) != 0)) {
362 *toggle = !(enabled);
363 sfc_warn(sa, "Failed to %s %s mode",
364 ((enabled) ? "enable" : "disable"), desc);
368 sfc_adapter_unlock(sa);
372 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
374 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
378 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
380 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
384 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
386 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
390 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
392 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
396 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
397 uint16_t nb_rx_desc, unsigned int socket_id,
398 const struct rte_eth_rxconf *rx_conf,
399 struct rte_mempool *mb_pool)
401 struct sfc_adapter *sa = dev->data->dev_private;
404 sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
405 rx_queue_id, nb_rx_desc, socket_id);
407 sfc_adapter_lock(sa);
409 rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
414 dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
416 sfc_adapter_unlock(sa);
421 sfc_adapter_unlock(sa);
427 sfc_rx_queue_release(void *queue)
429 struct sfc_dp_rxq *dp_rxq = queue;
431 struct sfc_adapter *sa;
432 unsigned int sw_index;
437 rxq = sfc_rxq_by_dp_rxq(dp_rxq);
439 sfc_adapter_lock(sa);
441 sw_index = sfc_rxq_sw_index(rxq);
443 sfc_log_init(sa, "RxQ=%u", sw_index);
445 sa->eth_dev->data->rx_queues[sw_index] = NULL;
447 sfc_rx_qfini(sa, sw_index);
449 sfc_adapter_unlock(sa);
453 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
454 uint16_t nb_tx_desc, unsigned int socket_id,
455 const struct rte_eth_txconf *tx_conf)
457 struct sfc_adapter *sa = dev->data->dev_private;
460 sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
461 tx_queue_id, nb_tx_desc, socket_id);
463 sfc_adapter_lock(sa);
465 rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
469 dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
471 sfc_adapter_unlock(sa);
475 sfc_adapter_unlock(sa);
481 sfc_tx_queue_release(void *queue)
483 struct sfc_dp_txq *dp_txq = queue;
485 unsigned int sw_index;
486 struct sfc_adapter *sa;
491 txq = sfc_txq_by_dp_txq(dp_txq);
492 sw_index = sfc_txq_sw_index(txq);
494 SFC_ASSERT(txq->evq != NULL);
497 sfc_log_init(sa, "TxQ = %u", sw_index);
499 sfc_adapter_lock(sa);
501 SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
502 sa->eth_dev->data->tx_queues[sw_index] = NULL;
504 sfc_tx_qfini(sa, sw_index);
506 sfc_adapter_unlock(sa);
510 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
512 struct sfc_adapter *sa = dev->data->dev_private;
513 struct sfc_port *port = &sa->port;
517 rte_spinlock_lock(&port->mac_stats_lock);
519 ret = sfc_port_update_mac_stats(sa);
523 mac_stats = port->mac_stats_buf;
525 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
526 EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
528 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
529 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
530 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
532 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
533 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
534 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
536 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
537 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
538 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
540 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
541 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
542 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
543 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
544 stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
545 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
547 stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
548 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
549 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
550 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
552 * Take into account stats which are whenever supported
553 * on EF10. If some stat is not supported by current
554 * firmware variant or HW revision, it is guaranteed
555 * to be zero in mac_stats.
558 mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
559 mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
560 mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
561 mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
562 mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
563 mac_stats[EFX_MAC_PM_TRUNC_QBB] +
564 mac_stats[EFX_MAC_PM_DISCARD_QBB] +
565 mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
566 mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
567 mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
569 mac_stats[EFX_MAC_RX_FCS_ERRORS] +
570 mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
571 mac_stats[EFX_MAC_RX_JABBER_PKTS];
572 /* no oerrors counters supported on EF10 */
576 rte_spinlock_unlock(&port->mac_stats_lock);
577 SFC_ASSERT(ret >= 0);
582 sfc_stats_reset(struct rte_eth_dev *dev)
584 struct sfc_adapter *sa = dev->data->dev_private;
585 struct sfc_port *port = &sa->port;
588 if (sa->state != SFC_ADAPTER_STARTED) {
590 * The operation cannot be done if port is not started; it
591 * will be scheduled to be done during the next port start
593 port->mac_stats_reset_pending = B_TRUE;
597 rc = sfc_port_reset_mac_stats(sa);
599 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
603 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
604 unsigned int xstats_count)
606 struct sfc_adapter *sa = dev->data->dev_private;
607 struct sfc_port *port = &sa->port;
613 rte_spinlock_lock(&port->mac_stats_lock);
615 rc = sfc_port_update_mac_stats(sa);
622 mac_stats = port->mac_stats_buf;
624 for (i = 0; i < EFX_MAC_NSTATS; ++i) {
625 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
626 if (xstats != NULL && nstats < (int)xstats_count) {
627 xstats[nstats].id = nstats;
628 xstats[nstats].value = mac_stats[i];
635 rte_spinlock_unlock(&port->mac_stats_lock);
641 sfc_xstats_get_names(struct rte_eth_dev *dev,
642 struct rte_eth_xstat_name *xstats_names,
643 unsigned int xstats_count)
645 struct sfc_adapter *sa = dev->data->dev_private;
646 struct sfc_port *port = &sa->port;
648 unsigned int nstats = 0;
650 for (i = 0; i < EFX_MAC_NSTATS; ++i) {
651 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
652 if (xstats_names != NULL && nstats < xstats_count)
653 strncpy(xstats_names[nstats].name,
654 efx_mac_stat_name(sa->nic, i),
655 sizeof(xstats_names[0].name));
664 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
665 uint64_t *values, unsigned int n)
667 struct sfc_adapter *sa = dev->data->dev_private;
668 struct sfc_port *port = &sa->port;
670 unsigned int nb_supported = 0;
671 unsigned int nb_written = 0;
676 if (unlikely(values == NULL) ||
677 unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
678 return port->mac_stats_nb_supported;
680 rte_spinlock_lock(&port->mac_stats_lock);
682 rc = sfc_port_update_mac_stats(sa);
689 mac_stats = port->mac_stats_buf;
691 for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
692 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
695 if ((ids == NULL) || (ids[nb_written] == nb_supported))
696 values[nb_written++] = mac_stats[i];
704 rte_spinlock_unlock(&port->mac_stats_lock);
710 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
711 struct rte_eth_xstat_name *xstats_names,
712 const uint64_t *ids, unsigned int size)
714 struct sfc_adapter *sa = dev->data->dev_private;
715 struct sfc_port *port = &sa->port;
716 unsigned int nb_supported = 0;
717 unsigned int nb_written = 0;
720 if (unlikely(xstats_names == NULL) ||
721 unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
722 return port->mac_stats_nb_supported;
724 for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
725 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
728 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
729 char *name = xstats_names[nb_written++].name;
731 strncpy(name, efx_mac_stat_name(sa->nic, i),
732 sizeof(xstats_names[0].name));
733 name[sizeof(xstats_names[0].name) - 1] = '\0';
743 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
745 struct sfc_adapter *sa = dev->data->dev_private;
746 unsigned int wanted_fc, link_fc;
748 memset(fc_conf, 0, sizeof(*fc_conf));
750 sfc_adapter_lock(sa);
752 if (sa->state == SFC_ADAPTER_STARTED)
753 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
755 link_fc = sa->port.flow_ctrl;
759 fc_conf->mode = RTE_FC_NONE;
761 case EFX_FCNTL_RESPOND:
762 fc_conf->mode = RTE_FC_RX_PAUSE;
764 case EFX_FCNTL_GENERATE:
765 fc_conf->mode = RTE_FC_TX_PAUSE;
767 case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
768 fc_conf->mode = RTE_FC_FULL;
771 sfc_err(sa, "%s: unexpected flow control value %#x",
775 fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
777 sfc_adapter_unlock(sa);
783 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
785 struct sfc_adapter *sa = dev->data->dev_private;
786 struct sfc_port *port = &sa->port;
790 if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
791 fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
792 fc_conf->mac_ctrl_frame_fwd != 0) {
793 sfc_err(sa, "unsupported flow control settings specified");
798 switch (fc_conf->mode) {
802 case RTE_FC_RX_PAUSE:
803 fcntl = EFX_FCNTL_RESPOND;
805 case RTE_FC_TX_PAUSE:
806 fcntl = EFX_FCNTL_GENERATE;
809 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
816 sfc_adapter_lock(sa);
818 if (sa->state == SFC_ADAPTER_STARTED) {
819 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
821 goto fail_mac_fcntl_set;
824 port->flow_ctrl = fcntl;
825 port->flow_ctrl_autoneg = fc_conf->autoneg;
827 sfc_adapter_unlock(sa);
832 sfc_adapter_unlock(sa);
839 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
841 struct sfc_adapter *sa = dev->data->dev_private;
842 size_t pdu = EFX_MAC_PDU(mtu);
846 sfc_log_init(sa, "mtu=%u", mtu);
849 if (pdu < EFX_MAC_PDU_MIN) {
850 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
851 (unsigned int)mtu, (unsigned int)pdu,
855 if (pdu > EFX_MAC_PDU_MAX) {
856 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
857 (unsigned int)mtu, (unsigned int)pdu,
862 sfc_adapter_lock(sa);
864 if (pdu != sa->port.pdu) {
865 if (sa->state == SFC_ADAPTER_STARTED) {
868 old_pdu = sa->port.pdu;
879 * The driver does not use it, but other PMDs update jumbo_frame
880 * flag and max_rx_pkt_len when MTU is set.
882 if (mtu > ETHER_MAX_LEN) {
883 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
885 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
886 rxmode->jumbo_frame = 1;
889 dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
891 sfc_adapter_unlock(sa);
893 sfc_log_init(sa, "done");
897 sa->port.pdu = old_pdu;
898 if (sfc_start(sa) != 0)
899 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
900 "PDU max size - port is stopped",
901 (unsigned int)pdu, (unsigned int)old_pdu);
902 sfc_adapter_unlock(sa);
905 sfc_log_init(sa, "failed %d", rc);
910 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
912 struct sfc_adapter *sa = dev->data->dev_private;
913 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
914 struct sfc_port *port = &sa->port;
917 sfc_adapter_lock(sa);
920 * Copy the address to the device private data so that
921 * it could be recalled in the case of adapter restart.
923 ether_addr_copy(mac_addr, &port->default_mac_addr);
925 if (port->isolated) {
926 sfc_err(sa, "isolated mode is active on the port");
927 sfc_err(sa, "will not set MAC address");
931 if (sa->state != SFC_ADAPTER_STARTED) {
932 sfc_info(sa, "the port is not started");
933 sfc_info(sa, "the new MAC address will be set on port start");
938 if (encp->enc_allow_set_mac_with_installed_filters) {
939 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
941 sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
946 * Changing the MAC address by means of MCDI request
947 * has no effect on received traffic, therefore
948 * we also need to update unicast filters
950 rc = sfc_set_rx_mode(sa);
952 sfc_err(sa, "cannot set filter (rc = %u)", rc);
954 sfc_warn(sa, "cannot set MAC address with filters installed");
955 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
956 sfc_warn(sa, "(some traffic may be dropped)");
959 * Since setting MAC address with filters installed is not
960 * allowed on the adapter, the new MAC address will be set
961 * by means of adapter restart. sfc_start() shall retrieve
962 * the new address from the device private data and set it.
967 sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
972 * In the case of failure sa->port->default_mac_addr does not
973 * need rollback since no error code is returned, and the upper
974 * API will anyway update the external MAC address storage.
975 * To be consistent with that new value it is better to keep
976 * the device private value the same.
978 sfc_adapter_unlock(sa);
983 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
986 struct sfc_adapter *sa = dev->data->dev_private;
987 struct sfc_port *port = &sa->port;
988 uint8_t *mc_addrs = port->mcast_addrs;
992 if (port->isolated) {
993 sfc_err(sa, "isolated mode is active on the port");
994 sfc_err(sa, "will not set multicast address list");
998 if (mc_addrs == NULL)
1001 if (nb_mc_addr > port->max_mcast_addrs) {
1002 sfc_err(sa, "too many multicast addresses: %u > %u",
1003 nb_mc_addr, port->max_mcast_addrs);
1007 for (i = 0; i < nb_mc_addr; ++i) {
1008 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1010 mc_addrs += EFX_MAC_ADDR_LEN;
1013 port->nb_mcast_addrs = nb_mc_addr;
1015 if (sa->state != SFC_ADAPTER_STARTED)
1018 rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1019 port->nb_mcast_addrs);
1021 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1028 * The function is used by the secondary process as well. It must not
1029 * use any process-local pointers from the adapter data.
1032 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1033 struct rte_eth_rxq_info *qinfo)
1035 struct sfc_adapter *sa = dev->data->dev_private;
1036 struct sfc_rxq_info *rxq_info;
1037 struct sfc_rxq *rxq;
1039 sfc_adapter_lock(sa);
1041 SFC_ASSERT(rx_queue_id < sa->rxq_count);
1043 rxq_info = &sa->rxq_info[rx_queue_id];
1044 rxq = rxq_info->rxq;
1045 SFC_ASSERT(rxq != NULL);
1047 qinfo->mp = rxq->refill_mb_pool;
1048 qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1049 qinfo->conf.rx_drop_en = 1;
1050 qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1051 qinfo->conf.offloads = DEV_RX_OFFLOAD_IPV4_CKSUM |
1052 DEV_RX_OFFLOAD_UDP_CKSUM |
1053 DEV_RX_OFFLOAD_TCP_CKSUM;
1054 if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1055 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1056 qinfo->scattered_rx = 1;
1058 qinfo->nb_desc = rxq_info->entries;
1060 sfc_adapter_unlock(sa);
1064 * The function is used by the secondary process as well. It must not
1065 * use any process-local pointers from the adapter data.
1068 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1069 struct rte_eth_txq_info *qinfo)
1071 struct sfc_adapter *sa = dev->data->dev_private;
1072 struct sfc_txq_info *txq_info;
1074 sfc_adapter_lock(sa);
1076 SFC_ASSERT(tx_queue_id < sa->txq_count);
1078 txq_info = &sa->txq_info[tx_queue_id];
1079 SFC_ASSERT(txq_info->txq != NULL);
1081 memset(qinfo, 0, sizeof(*qinfo));
1083 qinfo->conf.txq_flags = txq_info->txq->flags;
1084 qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1085 qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1086 qinfo->nb_desc = txq_info->entries;
1088 sfc_adapter_unlock(sa);
1092 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1094 struct sfc_adapter *sa = dev->data->dev_private;
1096 sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1098 return sfc_rx_qdesc_npending(sa, rx_queue_id);
1102 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1104 struct sfc_dp_rxq *dp_rxq = queue;
1106 return sfc_rx_qdesc_done(dp_rxq, offset);
1110 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1112 struct sfc_dp_rxq *dp_rxq = queue;
1113 struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1115 return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1119 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1121 struct sfc_dp_txq *dp_txq = queue;
1122 struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1124 return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1128 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1130 struct sfc_adapter *sa = dev->data->dev_private;
1133 sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1135 sfc_adapter_lock(sa);
1138 if (sa->state != SFC_ADAPTER_STARTED)
1139 goto fail_not_started;
1141 rc = sfc_rx_qstart(sa, rx_queue_id);
1143 goto fail_rx_qstart;
1145 sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1147 sfc_adapter_unlock(sa);
1153 sfc_adapter_unlock(sa);
1159 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1161 struct sfc_adapter *sa = dev->data->dev_private;
1163 sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1165 sfc_adapter_lock(sa);
1166 sfc_rx_qstop(sa, rx_queue_id);
1168 sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1170 sfc_adapter_unlock(sa);
1176 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1178 struct sfc_adapter *sa = dev->data->dev_private;
1181 sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1183 sfc_adapter_lock(sa);
1186 if (sa->state != SFC_ADAPTER_STARTED)
1187 goto fail_not_started;
1189 rc = sfc_tx_qstart(sa, tx_queue_id);
1191 goto fail_tx_qstart;
1193 sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1195 sfc_adapter_unlock(sa);
1201 sfc_adapter_unlock(sa);
1207 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1209 struct sfc_adapter *sa = dev->data->dev_private;
1211 sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1213 sfc_adapter_lock(sa);
1215 sfc_tx_qstop(sa, tx_queue_id);
1217 sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1219 sfc_adapter_unlock(sa);
1223 static efx_tunnel_protocol_t
1224 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1227 case RTE_TUNNEL_TYPE_VXLAN:
1228 return EFX_TUNNEL_PROTOCOL_VXLAN;
1229 case RTE_TUNNEL_TYPE_GENEVE:
1230 return EFX_TUNNEL_PROTOCOL_GENEVE;
1232 return EFX_TUNNEL_NPROTOS;
1236 enum sfc_udp_tunnel_op_e {
1237 SFC_UDP_TUNNEL_ADD_PORT,
1238 SFC_UDP_TUNNEL_DEL_PORT,
1242 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1243 struct rte_eth_udp_tunnel *tunnel_udp,
1244 enum sfc_udp_tunnel_op_e op)
1246 struct sfc_adapter *sa = dev->data->dev_private;
1247 efx_tunnel_protocol_t tunnel_proto;
1250 sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1251 (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1252 (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1253 tunnel_udp->udp_port, tunnel_udp->prot_type);
1256 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1257 if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1259 goto fail_bad_proto;
1262 sfc_adapter_lock(sa);
1265 case SFC_UDP_TUNNEL_ADD_PORT:
1266 rc = efx_tunnel_config_udp_add(sa->nic,
1267 tunnel_udp->udp_port,
1270 case SFC_UDP_TUNNEL_DEL_PORT:
1271 rc = efx_tunnel_config_udp_remove(sa->nic,
1272 tunnel_udp->udp_port,
1283 if (sa->state == SFC_ADAPTER_STARTED) {
1284 rc = efx_tunnel_reconfigure(sa->nic);
1287 * Configuration is accepted by FW and MC reboot
1288 * is initiated to apply the changes. MC reboot
1289 * will be handled in a usual way (MC reboot
1290 * event on management event queue and adapter
1294 } else if (rc != 0) {
1295 goto fail_reconfigure;
1299 sfc_adapter_unlock(sa);
1303 /* Remove/restore entry since the change makes the trouble */
1305 case SFC_UDP_TUNNEL_ADD_PORT:
1306 (void)efx_tunnel_config_udp_remove(sa->nic,
1307 tunnel_udp->udp_port,
1310 case SFC_UDP_TUNNEL_DEL_PORT:
1311 (void)efx_tunnel_config_udp_add(sa->nic,
1312 tunnel_udp->udp_port,
1319 sfc_adapter_unlock(sa);
1327 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1328 struct rte_eth_udp_tunnel *tunnel_udp)
1330 return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1334 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1335 struct rte_eth_udp_tunnel *tunnel_udp)
1337 return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1340 #if EFSYS_OPT_RX_SCALE
1342 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1343 struct rte_eth_rss_conf *rss_conf)
1345 struct sfc_adapter *sa = dev->data->dev_private;
1346 struct sfc_port *port = &sa->port;
1348 if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1351 if (sa->rss_channels == 0)
1354 sfc_adapter_lock(sa);
1357 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1358 * hence, conversion is done here to derive a correct set of ETH_RSS
1359 * flags which corresponds to the active EFX configuration stored
1360 * locally in 'sfc_adapter' and kept up-to-date
1362 rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1363 rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1364 if (rss_conf->rss_key != NULL)
1365 rte_memcpy(rss_conf->rss_key, sa->rss_key, EFX_RSS_KEY_SIZE);
1367 sfc_adapter_unlock(sa);
1373 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1374 struct rte_eth_rss_conf *rss_conf)
1376 struct sfc_adapter *sa = dev->data->dev_private;
1377 struct sfc_port *port = &sa->port;
1378 unsigned int efx_hash_types;
1384 if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1385 sfc_err(sa, "RSS is not available");
1389 if (sa->rss_channels == 0) {
1390 sfc_err(sa, "RSS is not configured");
1394 if ((rss_conf->rss_key != NULL) &&
1395 (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1396 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1397 sizeof(sa->rss_key));
1401 if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1402 sfc_err(sa, "unsupported hash functions requested");
1406 sfc_adapter_lock(sa);
1408 efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1410 rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1411 EFX_RX_HASHALG_TOEPLITZ,
1412 efx_hash_types, B_TRUE);
1414 goto fail_scale_mode_set;
1416 if (rss_conf->rss_key != NULL) {
1417 if (sa->state == SFC_ADAPTER_STARTED) {
1418 rc = efx_rx_scale_key_set(sa->nic,
1419 EFX_RSS_CONTEXT_DEFAULT,
1421 sizeof(sa->rss_key));
1423 goto fail_scale_key_set;
1426 rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1429 sa->rss_hash_types = efx_hash_types;
1431 sfc_adapter_unlock(sa);
1436 if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1437 EFX_RX_HASHALG_TOEPLITZ,
1438 sa->rss_hash_types, B_TRUE) != 0)
1439 sfc_err(sa, "failed to restore RSS mode");
1441 fail_scale_mode_set:
1442 sfc_adapter_unlock(sa);
1447 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1448 struct rte_eth_rss_reta_entry64 *reta_conf,
1451 struct sfc_adapter *sa = dev->data->dev_private;
1452 struct sfc_port *port = &sa->port;
1455 if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1458 if (sa->rss_channels == 0)
1461 if (reta_size != EFX_RSS_TBL_SIZE)
1464 sfc_adapter_lock(sa);
1466 for (entry = 0; entry < reta_size; entry++) {
1467 int grp = entry / RTE_RETA_GROUP_SIZE;
1468 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1470 if ((reta_conf[grp].mask >> grp_idx) & 1)
1471 reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1474 sfc_adapter_unlock(sa);
1480 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1481 struct rte_eth_rss_reta_entry64 *reta_conf,
1484 struct sfc_adapter *sa = dev->data->dev_private;
1485 struct sfc_port *port = &sa->port;
1486 unsigned int *rss_tbl_new;
1494 if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1495 sfc_err(sa, "RSS is not available");
1499 if (sa->rss_channels == 0) {
1500 sfc_err(sa, "RSS is not configured");
1504 if (reta_size != EFX_RSS_TBL_SIZE) {
1505 sfc_err(sa, "RETA size is wrong (should be %u)",
1510 rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1511 if (rss_tbl_new == NULL)
1514 sfc_adapter_lock(sa);
1516 rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1518 for (entry = 0; entry < reta_size; entry++) {
1519 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1520 struct rte_eth_rss_reta_entry64 *grp;
1522 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1524 if (grp->mask & (1ull << grp_idx)) {
1525 if (grp->reta[grp_idx] >= sa->rss_channels) {
1527 goto bad_reta_entry;
1529 rss_tbl_new[entry] = grp->reta[grp_idx];
1533 if (sa->state == SFC_ADAPTER_STARTED) {
1534 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1535 rss_tbl_new, EFX_RSS_TBL_SIZE);
1537 goto fail_scale_tbl_set;
1540 rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1544 sfc_adapter_unlock(sa);
1546 rte_free(rss_tbl_new);
1548 SFC_ASSERT(rc >= 0);
1554 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1555 enum rte_filter_op filter_op,
1558 struct sfc_adapter *sa = dev->data->dev_private;
1561 sfc_log_init(sa, "entry");
1563 switch (filter_type) {
1564 case RTE_ETH_FILTER_NONE:
1565 sfc_err(sa, "Global filters configuration not supported");
1567 case RTE_ETH_FILTER_MACVLAN:
1568 sfc_err(sa, "MACVLAN filters not supported");
1570 case RTE_ETH_FILTER_ETHERTYPE:
1571 sfc_err(sa, "EtherType filters not supported");
1573 case RTE_ETH_FILTER_FLEXIBLE:
1574 sfc_err(sa, "Flexible filters not supported");
1576 case RTE_ETH_FILTER_SYN:
1577 sfc_err(sa, "SYN filters not supported");
1579 case RTE_ETH_FILTER_NTUPLE:
1580 sfc_err(sa, "NTUPLE filters not supported");
1582 case RTE_ETH_FILTER_TUNNEL:
1583 sfc_err(sa, "Tunnel filters not supported");
1585 case RTE_ETH_FILTER_FDIR:
1586 sfc_err(sa, "Flow Director filters not supported");
1588 case RTE_ETH_FILTER_HASH:
1589 sfc_err(sa, "Hash filters not supported");
1591 case RTE_ETH_FILTER_GENERIC:
1592 if (filter_op != RTE_ETH_FILTER_GET) {
1595 *(const void **)arg = &sfc_flow_ops;
1600 sfc_err(sa, "Unknown filter type %u", filter_type);
1604 sfc_log_init(sa, "exit: %d", -rc);
1605 SFC_ASSERT(rc >= 0);
1609 static const struct eth_dev_ops sfc_eth_dev_ops = {
1610 .dev_configure = sfc_dev_configure,
1611 .dev_start = sfc_dev_start,
1612 .dev_stop = sfc_dev_stop,
1613 .dev_set_link_up = sfc_dev_set_link_up,
1614 .dev_set_link_down = sfc_dev_set_link_down,
1615 .dev_close = sfc_dev_close,
1616 .promiscuous_enable = sfc_dev_promisc_enable,
1617 .promiscuous_disable = sfc_dev_promisc_disable,
1618 .allmulticast_enable = sfc_dev_allmulti_enable,
1619 .allmulticast_disable = sfc_dev_allmulti_disable,
1620 .link_update = sfc_dev_link_update,
1621 .stats_get = sfc_stats_get,
1622 .stats_reset = sfc_stats_reset,
1623 .xstats_get = sfc_xstats_get,
1624 .xstats_reset = sfc_stats_reset,
1625 .xstats_get_names = sfc_xstats_get_names,
1626 .dev_infos_get = sfc_dev_infos_get,
1627 .dev_supported_ptypes_get = sfc_dev_supported_ptypes_get,
1628 .mtu_set = sfc_dev_set_mtu,
1629 .rx_queue_start = sfc_rx_queue_start,
1630 .rx_queue_stop = sfc_rx_queue_stop,
1631 .tx_queue_start = sfc_tx_queue_start,
1632 .tx_queue_stop = sfc_tx_queue_stop,
1633 .rx_queue_setup = sfc_rx_queue_setup,
1634 .rx_queue_release = sfc_rx_queue_release,
1635 .rx_queue_count = sfc_rx_queue_count,
1636 .rx_descriptor_done = sfc_rx_descriptor_done,
1637 .rx_descriptor_status = sfc_rx_descriptor_status,
1638 .tx_descriptor_status = sfc_tx_descriptor_status,
1639 .tx_queue_setup = sfc_tx_queue_setup,
1640 .tx_queue_release = sfc_tx_queue_release,
1641 .flow_ctrl_get = sfc_flow_ctrl_get,
1642 .flow_ctrl_set = sfc_flow_ctrl_set,
1643 .mac_addr_set = sfc_mac_addr_set,
1644 .udp_tunnel_port_add = sfc_dev_udp_tunnel_port_add,
1645 .udp_tunnel_port_del = sfc_dev_udp_tunnel_port_del,
1646 #if EFSYS_OPT_RX_SCALE
1647 .reta_update = sfc_dev_rss_reta_update,
1648 .reta_query = sfc_dev_rss_reta_query,
1649 .rss_hash_update = sfc_dev_rss_hash_update,
1650 .rss_hash_conf_get = sfc_dev_rss_hash_conf_get,
1652 .filter_ctrl = sfc_dev_filter_ctrl,
1653 .set_mc_addr_list = sfc_set_mc_addr_list,
1654 .rxq_info_get = sfc_rx_queue_info_get,
1655 .txq_info_get = sfc_tx_queue_info_get,
1656 .fw_version_get = sfc_fw_version_get,
1657 .xstats_get_by_id = sfc_xstats_get_by_id,
1658 .xstats_get_names_by_id = sfc_xstats_get_names_by_id,
1662 * Duplicate a string in potentially shared memory required for
1663 * multi-process support.
1665 * strdup() allocates from process-local heap/memory.
1668 sfc_strdup(const char *str)
1676 size = strlen(str) + 1;
1677 copy = rte_malloc(__func__, size, 0);
1679 rte_memcpy(copy, str, size);
1685 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1687 struct sfc_adapter *sa = dev->data->dev_private;
1688 unsigned int avail_caps = 0;
1689 const char *rx_name = NULL;
1690 const char *tx_name = NULL;
1693 switch (sa->family) {
1694 case EFX_FAMILY_HUNTINGTON:
1695 case EFX_FAMILY_MEDFORD:
1696 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1702 rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1703 sfc_kvarg_string_handler, &rx_name);
1705 goto fail_kvarg_rx_datapath;
1707 if (rx_name != NULL) {
1708 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1709 if (sa->dp_rx == NULL) {
1710 sfc_err(sa, "Rx datapath %s not found", rx_name);
1714 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1716 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1719 goto fail_dp_rx_caps;
1722 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1723 if (sa->dp_rx == NULL) {
1724 sfc_err(sa, "Rx datapath by caps %#x not found",
1731 sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1732 if (sa->dp_rx_name == NULL) {
1734 goto fail_dp_rx_name;
1737 sfc_info(sa, "use %s Rx datapath", sa->dp_rx_name);
1739 dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1741 rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1742 sfc_kvarg_string_handler, &tx_name);
1744 goto fail_kvarg_tx_datapath;
1746 if (tx_name != NULL) {
1747 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1748 if (sa->dp_tx == NULL) {
1749 sfc_err(sa, "Tx datapath %s not found", tx_name);
1753 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1755 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1758 goto fail_dp_tx_caps;
1761 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1762 if (sa->dp_tx == NULL) {
1763 sfc_err(sa, "Tx datapath by caps %#x not found",
1770 sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1771 if (sa->dp_tx_name == NULL) {
1773 goto fail_dp_tx_name;
1776 sfc_info(sa, "use %s Tx datapath", sa->dp_tx_name);
1778 dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1780 dev->dev_ops = &sfc_eth_dev_ops;
1789 fail_kvarg_tx_datapath:
1790 rte_free(sa->dp_rx_name);
1791 sa->dp_rx_name = NULL;
1798 fail_kvarg_rx_datapath:
1803 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1805 struct sfc_adapter *sa = dev->data->dev_private;
1807 dev->dev_ops = NULL;
1808 dev->rx_pkt_burst = NULL;
1809 dev->tx_pkt_burst = NULL;
1811 rte_free(sa->dp_tx_name);
1812 sa->dp_tx_name = NULL;
1815 rte_free(sa->dp_rx_name);
1816 sa->dp_rx_name = NULL;
1820 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1821 .rxq_info_get = sfc_rx_queue_info_get,
1822 .txq_info_get = sfc_tx_queue_info_get,
1826 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1829 * Device private data has really many process-local pointers.
1830 * Below code should be extremely careful to use data located
1831 * in shared memory only.
1833 struct sfc_adapter *sa = dev->data->dev_private;
1834 const struct sfc_dp_rx *dp_rx;
1835 const struct sfc_dp_tx *dp_tx;
1838 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1839 if (dp_rx == NULL) {
1840 sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1844 if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1845 sfc_err(sa, "%s Rx datapath does not support multi-process",
1848 goto fail_dp_rx_multi_process;
1851 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1852 if (dp_tx == NULL) {
1853 sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1857 if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1858 sfc_err(sa, "%s Tx datapath does not support multi-process",
1861 goto fail_dp_tx_multi_process;
1864 dev->rx_pkt_burst = dp_rx->pkt_burst;
1865 dev->tx_pkt_burst = dp_tx->pkt_burst;
1866 dev->dev_ops = &sfc_eth_dev_secondary_ops;
1870 fail_dp_tx_multi_process:
1872 fail_dp_rx_multi_process:
1878 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1880 dev->dev_ops = NULL;
1881 dev->tx_pkt_burst = NULL;
1882 dev->rx_pkt_burst = NULL;
1886 sfc_register_dp(void)
1889 if (TAILQ_EMPTY(&sfc_dp_head)) {
1890 /* Prefer EF10 datapath */
1891 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1892 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1894 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1895 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1896 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1901 sfc_eth_dev_init(struct rte_eth_dev *dev)
1903 struct sfc_adapter *sa = dev->data->dev_private;
1904 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1906 const efx_nic_cfg_t *encp;
1907 const struct ether_addr *from;
1911 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1912 return -sfc_eth_dev_secondary_set_ops(dev);
1914 /* Required for logging */
1915 sa->pci_addr = pci_dev->addr;
1916 sa->port_id = dev->data->port_id;
1920 /* Copy PCI device info to the dev->data */
1921 rte_eth_copy_pci_info(dev, pci_dev);
1923 rc = sfc_kvargs_parse(sa);
1925 goto fail_kvargs_parse;
1927 rc = sfc_kvargs_process(sa, SFC_KVARG_DEBUG_INIT,
1928 sfc_kvarg_bool_handler, &sa->debug_init);
1930 goto fail_kvarg_debug_init;
1932 sfc_log_init(sa, "entry");
1934 dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1935 if (dev->data->mac_addrs == NULL) {
1937 goto fail_mac_addrs;
1940 sfc_adapter_lock_init(sa);
1941 sfc_adapter_lock(sa);
1943 sfc_log_init(sa, "probing");
1948 sfc_log_init(sa, "set device ops");
1949 rc = sfc_eth_dev_set_ops(dev);
1953 sfc_log_init(sa, "attaching");
1954 rc = sfc_attach(sa);
1958 encp = efx_nic_cfg_get(sa->nic);
1961 * The arguments are really reverse order in comparison to
1962 * Linux kernel. Copy from NIC config to Ethernet device data.
1964 from = (const struct ether_addr *)(encp->enc_mac_addr);
1965 ether_addr_copy(from, &dev->data->mac_addrs[0]);
1967 sfc_adapter_unlock(sa);
1969 sfc_log_init(sa, "done");
1973 sfc_eth_dev_clear_ops(dev);
1979 sfc_adapter_unlock(sa);
1980 sfc_adapter_lock_fini(sa);
1981 rte_free(dev->data->mac_addrs);
1982 dev->data->mac_addrs = NULL;
1985 fail_kvarg_debug_init:
1986 sfc_kvargs_cleanup(sa);
1989 sfc_log_init(sa, "failed %d", rc);
1995 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
1997 struct sfc_adapter *sa;
1999 if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2000 sfc_eth_dev_secondary_clear_ops(dev);
2004 sa = dev->data->dev_private;
2005 sfc_log_init(sa, "entry");
2007 sfc_adapter_lock(sa);
2009 sfc_eth_dev_clear_ops(dev);
2014 rte_free(dev->data->mac_addrs);
2015 dev->data->mac_addrs = NULL;
2017 sfc_kvargs_cleanup(sa);
2019 sfc_adapter_unlock(sa);
2020 sfc_adapter_lock_fini(sa);
2022 sfc_log_init(sa, "done");
2024 /* Required for logging, so cleanup last */
2029 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2030 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2031 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2032 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2033 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2034 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2035 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2036 { .vendor_id = 0 /* sentinel */ }
2039 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2040 struct rte_pci_device *pci_dev)
2042 return rte_eth_dev_pci_generic_probe(pci_dev,
2043 sizeof(struct sfc_adapter), sfc_eth_dev_init);
2046 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2048 return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2051 static struct rte_pci_driver sfc_efx_pmd = {
2052 .id_table = pci_id_sfc_efx_map,
2054 RTE_PCI_DRV_INTR_LSC |
2055 RTE_PCI_DRV_NEED_MAPPING,
2056 .probe = sfc_eth_dev_pci_probe,
2057 .remove = sfc_eth_dev_pci_remove,
2060 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2061 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2062 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2063 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2064 SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2065 SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2066 SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2067 SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long> "
2068 SFC_KVARG_MCDI_LOGGING "=" SFC_KVARG_VALUES_BOOL " "
2069 SFC_KVARG_DEBUG_INIT "=" SFC_KVARG_VALUES_BOOL);