1 /* SPDX-License-Identifier: BSD-3-Clause
3 * Copyright (c) 2016-2018 Solarflare Communications Inc.
6 * This software was jointly developed between OKTET Labs (under contract
7 * for Solarflare) and Solarflare Communications, Inc.
11 #include <rte_ethdev_driver.h>
12 #include <rte_ethdev_pci.h>
14 #include <rte_bus_pci.h>
15 #include <rte_errno.h>
16 #include <rte_string_fns.h>
17 #include <rte_ether.h>
22 #include "sfc_debug.h"
24 #include "sfc_kvargs.h"
30 #include "sfc_dp_rx.h"
32 uint32_t sfc_logtype_driver;
34 static struct sfc_dp_list sfc_dp_head =
35 TAILQ_HEAD_INITIALIZER(sfc_dp_head);
38 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
40 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
41 efx_nic_fw_info_t enfi;
46 * Return value of the callback is likely supposed to be
47 * equal to or greater than 0, nevertheless, if an error
48 * occurs, it will be desirable to pass it to the caller
50 if ((fw_version == NULL) || (fw_size == 0))
53 rc = efx_nic_get_fw_version(sa->nic, &enfi);
57 ret = snprintf(fw_version, fw_size,
58 "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
59 enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
60 enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
64 if (enfi.enfi_dpcpu_fw_ids_valid) {
65 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
68 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
69 fw_size - dpcpu_fw_ids_offset,
70 " rx%" PRIx16 " tx%" PRIx16,
71 enfi.enfi_rx_dpcpu_fw_id,
72 enfi.enfi_tx_dpcpu_fw_id);
79 if (fw_size < (size_t)(++ret))
86 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
88 const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
89 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
90 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
91 struct sfc_rss *rss = &sas->rss;
92 uint64_t txq_offloads_def = 0;
94 sfc_log_init(sa, "entry");
96 dev_info->min_mtu = RTE_ETHER_MIN_MTU;
97 dev_info->max_mtu = EFX_MAC_SDU_MAX;
99 dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
101 /* Autonegotiation may be disabled */
102 dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
103 if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_1000FDX))
104 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
105 if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_10000FDX))
106 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
107 if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_25000FDX))
108 dev_info->speed_capa |= ETH_LINK_SPEED_25G;
109 if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_40000FDX))
110 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
111 if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_50000FDX))
112 dev_info->speed_capa |= ETH_LINK_SPEED_50G;
113 if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_100000FDX))
114 dev_info->speed_capa |= ETH_LINK_SPEED_100G;
116 dev_info->max_rx_queues = sa->rxq_max;
117 dev_info->max_tx_queues = sa->txq_max;
119 /* By default packets are dropped if no descriptors are available */
120 dev_info->default_rxconf.rx_drop_en = 1;
122 dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
125 * rx_offload_capa includes both device and queue offloads since
126 * the latter may be requested on a per device basis which makes
127 * sense when some offloads are needed to be set on all queues.
129 dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
130 dev_info->rx_queue_offload_capa;
132 dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
135 * tx_offload_capa includes both device and queue offloads since
136 * the latter may be requested on a per device basis which makes
137 * sense when some offloads are needed to be set on all queues.
139 dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
140 dev_info->tx_queue_offload_capa;
142 if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
143 txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
145 dev_info->default_txconf.offloads |= txq_offloads_def;
147 if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
151 for (i = 0; i < rss->hf_map_nb_entries; ++i)
152 rte_hf |= rss->hf_map[i].rte;
154 dev_info->reta_size = EFX_RSS_TBL_SIZE;
155 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
156 dev_info->flow_type_rss_offloads = rte_hf;
159 /* Initialize to hardware limits */
160 dev_info->rx_desc_lim.nb_max = sa->rxq_max_entries;
161 dev_info->rx_desc_lim.nb_min = sa->rxq_min_entries;
162 /* The RXQ hardware requires that the descriptor count is a power
163 * of 2, but rx_desc_lim cannot properly describe that constraint.
165 dev_info->rx_desc_lim.nb_align = sa->rxq_min_entries;
167 /* Initialize to hardware limits */
168 dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
169 dev_info->tx_desc_lim.nb_min = sa->txq_min_entries;
171 * The TXQ hardware requires that the descriptor count is a power
172 * of 2, but tx_desc_lim cannot properly describe that constraint
174 dev_info->tx_desc_lim.nb_align = sa->txq_min_entries;
176 if (sap->dp_rx->get_dev_info != NULL)
177 sap->dp_rx->get_dev_info(dev_info);
178 if (sap->dp_tx->get_dev_info != NULL)
179 sap->dp_tx->get_dev_info(dev_info);
181 dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
182 RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
185 static const uint32_t *
186 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
188 const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
190 return sap->dp_rx->supported_ptypes_get(sap->shared->tunnel_encaps);
194 sfc_dev_configure(struct rte_eth_dev *dev)
196 struct rte_eth_dev_data *dev_data = dev->data;
197 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
200 sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
201 dev_data->nb_rx_queues, dev_data->nb_tx_queues);
203 sfc_adapter_lock(sa);
205 case SFC_ADAPTER_CONFIGURED:
207 case SFC_ADAPTER_INITIALIZED:
208 rc = sfc_configure(sa);
211 sfc_err(sa, "unexpected adapter state %u to configure",
216 sfc_adapter_unlock(sa);
218 sfc_log_init(sa, "done %d", rc);
224 sfc_dev_start(struct rte_eth_dev *dev)
226 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
229 sfc_log_init(sa, "entry");
231 sfc_adapter_lock(sa);
233 sfc_adapter_unlock(sa);
235 sfc_log_init(sa, "done %d", rc);
241 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
243 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
244 struct rte_eth_link current_link;
247 sfc_log_init(sa, "entry");
249 if (sa->state != SFC_ADAPTER_STARTED) {
250 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, ¤t_link);
251 } else if (wait_to_complete) {
252 efx_link_mode_t link_mode;
254 if (efx_port_poll(sa->nic, &link_mode) != 0)
255 link_mode = EFX_LINK_UNKNOWN;
256 sfc_port_link_mode_to_info(link_mode, ¤t_link);
259 sfc_ev_mgmt_qpoll(sa);
260 rte_eth_linkstatus_get(dev, ¤t_link);
263 ret = rte_eth_linkstatus_set(dev, ¤t_link);
265 sfc_notice(sa, "Link status is %s",
266 current_link.link_status ? "UP" : "DOWN");
272 sfc_dev_stop(struct rte_eth_dev *dev)
274 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
276 sfc_log_init(sa, "entry");
278 sfc_adapter_lock(sa);
280 sfc_adapter_unlock(sa);
282 sfc_log_init(sa, "done");
286 sfc_dev_set_link_up(struct rte_eth_dev *dev)
288 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
291 sfc_log_init(sa, "entry");
293 sfc_adapter_lock(sa);
295 sfc_adapter_unlock(sa);
302 sfc_dev_set_link_down(struct rte_eth_dev *dev)
304 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
306 sfc_log_init(sa, "entry");
308 sfc_adapter_lock(sa);
310 sfc_adapter_unlock(sa);
316 sfc_dev_close(struct rte_eth_dev *dev)
318 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
320 sfc_log_init(sa, "entry");
322 sfc_adapter_lock(sa);
324 case SFC_ADAPTER_STARTED:
326 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
328 case SFC_ADAPTER_CONFIGURED:
330 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
332 case SFC_ADAPTER_INITIALIZED:
335 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
338 sfc_adapter_unlock(sa);
340 sfc_log_init(sa, "done");
344 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
347 struct sfc_port *port;
349 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
350 boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
351 const char *desc = (allmulti) ? "all-multi" : "promiscuous";
353 sfc_adapter_lock(sa);
356 toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
358 if (*toggle != enabled) {
361 if (sfc_sa2shared(sa)->isolated) {
362 sfc_warn(sa, "isolated mode is active on the port");
363 sfc_warn(sa, "the change is to be applied on the next "
364 "start provided that isolated mode is "
365 "disabled prior the next start");
366 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
367 (sfc_set_rx_mode(sa) != 0)) {
368 *toggle = !(enabled);
369 sfc_warn(sa, "Failed to %s %s mode",
370 ((enabled) ? "enable" : "disable"), desc);
374 sfc_adapter_unlock(sa);
378 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
380 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
384 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
386 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
390 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
392 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
396 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
398 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
402 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
403 uint16_t nb_rx_desc, unsigned int socket_id,
404 const struct rte_eth_rxconf *rx_conf,
405 struct rte_mempool *mb_pool)
407 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
408 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
411 sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
412 rx_queue_id, nb_rx_desc, socket_id);
414 sfc_adapter_lock(sa);
416 rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
421 dev->data->rx_queues[rx_queue_id] = sas->rxq_info[rx_queue_id].dp;
423 sfc_adapter_unlock(sa);
428 sfc_adapter_unlock(sa);
434 sfc_rx_queue_release(void *queue)
436 struct sfc_dp_rxq *dp_rxq = queue;
438 struct sfc_adapter *sa;
439 unsigned int sw_index;
444 rxq = sfc_rxq_by_dp_rxq(dp_rxq);
446 sfc_adapter_lock(sa);
448 sw_index = dp_rxq->dpq.queue_id;
450 sfc_log_init(sa, "RxQ=%u", sw_index);
452 sfc_rx_qfini(sa, sw_index);
454 sfc_adapter_unlock(sa);
458 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
459 uint16_t nb_tx_desc, unsigned int socket_id,
460 const struct rte_eth_txconf *tx_conf)
462 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
463 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
466 sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
467 tx_queue_id, nb_tx_desc, socket_id);
469 sfc_adapter_lock(sa);
471 rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
475 dev->data->tx_queues[tx_queue_id] = sas->txq_info[tx_queue_id].dp;
477 sfc_adapter_unlock(sa);
481 sfc_adapter_unlock(sa);
487 sfc_tx_queue_release(void *queue)
489 struct sfc_dp_txq *dp_txq = queue;
491 unsigned int sw_index;
492 struct sfc_adapter *sa;
497 txq = sfc_txq_by_dp_txq(dp_txq);
498 sw_index = dp_txq->dpq.queue_id;
500 SFC_ASSERT(txq->evq != NULL);
503 sfc_log_init(sa, "TxQ = %u", sw_index);
505 sfc_adapter_lock(sa);
507 sfc_tx_qfini(sa, sw_index);
509 sfc_adapter_unlock(sa);
513 * Some statistics are computed as A - B where A and B each increase
514 * monotonically with some hardware counter(s) and the counters are read
517 * If packet X is counted in A, but not counted in B yet, computed value is
520 * If packet X is not counted in A at the moment of reading the counter,
521 * but counted in B at the moment of reading the counter, computed value
524 * However, counter which grows backward is worse evil than slightly wrong
525 * value. So, let's try to guarantee that it never happens except may be
526 * the case when the MAC stats are zeroed as a result of a NIC reset.
529 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
531 if ((int64_t)(newval - *stat) > 0 || newval == 0)
536 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
538 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
539 struct sfc_port *port = &sa->port;
543 rte_spinlock_lock(&port->mac_stats_lock);
545 ret = sfc_port_update_mac_stats(sa);
549 mac_stats = port->mac_stats_buf;
551 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
552 EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
554 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
555 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
556 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
558 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
559 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
560 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
562 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
563 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
564 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
566 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
567 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
568 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
569 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
570 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
572 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
573 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
574 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
576 * Take into account stats which are whenever supported
577 * on EF10. If some stat is not supported by current
578 * firmware variant or HW revision, it is guaranteed
579 * to be zero in mac_stats.
582 mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
583 mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
584 mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
585 mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
586 mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
587 mac_stats[EFX_MAC_PM_TRUNC_QBB] +
588 mac_stats[EFX_MAC_PM_DISCARD_QBB] +
589 mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
590 mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
591 mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
593 mac_stats[EFX_MAC_RX_FCS_ERRORS] +
594 mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
595 mac_stats[EFX_MAC_RX_JABBER_PKTS];
596 /* no oerrors counters supported on EF10 */
598 /* Exclude missed, errors and pauses from Rx packets */
599 sfc_update_diff_stat(&port->ipackets,
600 mac_stats[EFX_MAC_RX_PKTS] -
601 mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
602 stats->imissed - stats->ierrors);
603 stats->ipackets = port->ipackets;
607 rte_spinlock_unlock(&port->mac_stats_lock);
608 SFC_ASSERT(ret >= 0);
613 sfc_stats_reset(struct rte_eth_dev *dev)
615 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
616 struct sfc_port *port = &sa->port;
619 if (sa->state != SFC_ADAPTER_STARTED) {
621 * The operation cannot be done if port is not started; it
622 * will be scheduled to be done during the next port start
624 port->mac_stats_reset_pending = B_TRUE;
628 rc = sfc_port_reset_mac_stats(sa);
630 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
634 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
635 unsigned int xstats_count)
637 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
638 struct sfc_port *port = &sa->port;
644 rte_spinlock_lock(&port->mac_stats_lock);
646 rc = sfc_port_update_mac_stats(sa);
653 mac_stats = port->mac_stats_buf;
655 for (i = 0; i < EFX_MAC_NSTATS; ++i) {
656 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
657 if (xstats != NULL && nstats < (int)xstats_count) {
658 xstats[nstats].id = nstats;
659 xstats[nstats].value = mac_stats[i];
666 rte_spinlock_unlock(&port->mac_stats_lock);
672 sfc_xstats_get_names(struct rte_eth_dev *dev,
673 struct rte_eth_xstat_name *xstats_names,
674 unsigned int xstats_count)
676 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
677 struct sfc_port *port = &sa->port;
679 unsigned int nstats = 0;
681 for (i = 0; i < EFX_MAC_NSTATS; ++i) {
682 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
683 if (xstats_names != NULL && nstats < xstats_count)
684 strlcpy(xstats_names[nstats].name,
685 efx_mac_stat_name(sa->nic, i),
686 sizeof(xstats_names[0].name));
695 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
696 uint64_t *values, unsigned int n)
698 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
699 struct sfc_port *port = &sa->port;
701 unsigned int nb_supported = 0;
702 unsigned int nb_written = 0;
707 if (unlikely(values == NULL) ||
708 unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
709 return port->mac_stats_nb_supported;
711 rte_spinlock_lock(&port->mac_stats_lock);
713 rc = sfc_port_update_mac_stats(sa);
720 mac_stats = port->mac_stats_buf;
722 for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
723 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
726 if ((ids == NULL) || (ids[nb_written] == nb_supported))
727 values[nb_written++] = mac_stats[i];
735 rte_spinlock_unlock(&port->mac_stats_lock);
741 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
742 struct rte_eth_xstat_name *xstats_names,
743 const uint64_t *ids, unsigned int size)
745 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
746 struct sfc_port *port = &sa->port;
747 unsigned int nb_supported = 0;
748 unsigned int nb_written = 0;
751 if (unlikely(xstats_names == NULL) ||
752 unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
753 return port->mac_stats_nb_supported;
755 for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
756 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
759 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
760 char *name = xstats_names[nb_written++].name;
762 strlcpy(name, efx_mac_stat_name(sa->nic, i),
763 sizeof(xstats_names[0].name));
773 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
775 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
776 unsigned int wanted_fc, link_fc;
778 memset(fc_conf, 0, sizeof(*fc_conf));
780 sfc_adapter_lock(sa);
782 if (sa->state == SFC_ADAPTER_STARTED)
783 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
785 link_fc = sa->port.flow_ctrl;
789 fc_conf->mode = RTE_FC_NONE;
791 case EFX_FCNTL_RESPOND:
792 fc_conf->mode = RTE_FC_RX_PAUSE;
794 case EFX_FCNTL_GENERATE:
795 fc_conf->mode = RTE_FC_TX_PAUSE;
797 case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
798 fc_conf->mode = RTE_FC_FULL;
801 sfc_err(sa, "%s: unexpected flow control value %#x",
805 fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
807 sfc_adapter_unlock(sa);
813 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
815 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
816 struct sfc_port *port = &sa->port;
820 if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
821 fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
822 fc_conf->mac_ctrl_frame_fwd != 0) {
823 sfc_err(sa, "unsupported flow control settings specified");
828 switch (fc_conf->mode) {
832 case RTE_FC_RX_PAUSE:
833 fcntl = EFX_FCNTL_RESPOND;
835 case RTE_FC_TX_PAUSE:
836 fcntl = EFX_FCNTL_GENERATE;
839 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
846 sfc_adapter_lock(sa);
848 if (sa->state == SFC_ADAPTER_STARTED) {
849 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
851 goto fail_mac_fcntl_set;
854 port->flow_ctrl = fcntl;
855 port->flow_ctrl_autoneg = fc_conf->autoneg;
857 sfc_adapter_unlock(sa);
862 sfc_adapter_unlock(sa);
869 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
871 struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
872 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
873 boolean_t scatter_enabled;
877 for (i = 0; i < sas->rxq_count; i++) {
878 if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
881 scatter_enabled = (sas->rxq_info[i].type_flags &
882 EFX_RXQ_FLAG_SCATTER);
884 if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
885 encp->enc_rx_prefix_size,
886 scatter_enabled, &error)) {
887 sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
897 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
899 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
900 size_t pdu = EFX_MAC_PDU(mtu);
904 sfc_log_init(sa, "mtu=%u", mtu);
907 if (pdu < EFX_MAC_PDU_MIN) {
908 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
909 (unsigned int)mtu, (unsigned int)pdu,
913 if (pdu > EFX_MAC_PDU_MAX) {
914 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
915 (unsigned int)mtu, (unsigned int)pdu,
920 sfc_adapter_lock(sa);
922 rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
924 goto fail_check_scatter;
926 if (pdu != sa->port.pdu) {
927 if (sa->state == SFC_ADAPTER_STARTED) {
930 old_pdu = sa->port.pdu;
941 * The driver does not use it, but other PMDs update jumbo frame
942 * flag and max_rx_pkt_len when MTU is set.
944 if (mtu > RTE_ETHER_MAX_LEN) {
945 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
946 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
949 dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
951 sfc_adapter_unlock(sa);
953 sfc_log_init(sa, "done");
957 sa->port.pdu = old_pdu;
958 if (sfc_start(sa) != 0)
959 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
960 "PDU max size - port is stopped",
961 (unsigned int)pdu, (unsigned int)old_pdu);
964 sfc_adapter_unlock(sa);
967 sfc_log_init(sa, "failed %d", rc);
972 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
974 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
975 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
976 struct sfc_port *port = &sa->port;
977 struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
980 sfc_adapter_lock(sa);
983 * Copy the address to the device private data so that
984 * it could be recalled in the case of adapter restart.
986 rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
989 * Neither of the two following checks can return
990 * an error. The new MAC address is preserved in
991 * the device private data and can be activated
992 * on the next port start if the user prevents
993 * isolated mode from being enabled.
995 if (sfc_sa2shared(sa)->isolated) {
996 sfc_warn(sa, "isolated mode is active on the port");
997 sfc_warn(sa, "will not set MAC address");
1001 if (sa->state != SFC_ADAPTER_STARTED) {
1002 sfc_notice(sa, "the port is not started");
1003 sfc_notice(sa, "the new MAC address will be set on port start");
1008 if (encp->enc_allow_set_mac_with_installed_filters) {
1009 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1011 sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1016 * Changing the MAC address by means of MCDI request
1017 * has no effect on received traffic, therefore
1018 * we also need to update unicast filters
1020 rc = sfc_set_rx_mode(sa);
1022 sfc_err(sa, "cannot set filter (rc = %u)", rc);
1023 /* Rollback the old address */
1024 (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1025 (void)sfc_set_rx_mode(sa);
1028 sfc_warn(sa, "cannot set MAC address with filters installed");
1029 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1030 sfc_warn(sa, "(some traffic may be dropped)");
1033 * Since setting MAC address with filters installed is not
1034 * allowed on the adapter, the new MAC address will be set
1035 * by means of adapter restart. sfc_start() shall retrieve
1036 * the new address from the device private data and set it.
1041 sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1046 rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1048 sfc_adapter_unlock(sa);
1050 SFC_ASSERT(rc >= 0);
1056 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1057 struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1059 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1060 struct sfc_port *port = &sa->port;
1061 uint8_t *mc_addrs = port->mcast_addrs;
1065 if (sfc_sa2shared(sa)->isolated) {
1066 sfc_err(sa, "isolated mode is active on the port");
1067 sfc_err(sa, "will not set multicast address list");
1071 if (mc_addrs == NULL)
1074 if (nb_mc_addr > port->max_mcast_addrs) {
1075 sfc_err(sa, "too many multicast addresses: %u > %u",
1076 nb_mc_addr, port->max_mcast_addrs);
1080 for (i = 0; i < nb_mc_addr; ++i) {
1081 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1083 mc_addrs += EFX_MAC_ADDR_LEN;
1086 port->nb_mcast_addrs = nb_mc_addr;
1088 if (sa->state != SFC_ADAPTER_STARTED)
1091 rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1092 port->nb_mcast_addrs);
1094 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1096 SFC_ASSERT(rc >= 0);
1101 * The function is used by the secondary process as well. It must not
1102 * use any process-local pointers from the adapter data.
1105 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1106 struct rte_eth_rxq_info *qinfo)
1108 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1109 struct sfc_rxq_info *rxq_info;
1111 SFC_ASSERT(rx_queue_id < sas->rxq_count);
1113 rxq_info = &sas->rxq_info[rx_queue_id];
1115 qinfo->mp = rxq_info->refill_mb_pool;
1116 qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1117 qinfo->conf.rx_drop_en = 1;
1118 qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1119 qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1120 if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1121 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1122 qinfo->scattered_rx = 1;
1124 qinfo->nb_desc = rxq_info->entries;
1128 * The function is used by the secondary process as well. It must not
1129 * use any process-local pointers from the adapter data.
1132 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1133 struct rte_eth_txq_info *qinfo)
1135 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1136 struct sfc_txq_info *txq_info;
1138 SFC_ASSERT(tx_queue_id < sas->txq_count);
1140 txq_info = &sas->txq_info[tx_queue_id];
1142 memset(qinfo, 0, sizeof(*qinfo));
1144 qinfo->conf.offloads = txq_info->offloads;
1145 qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1146 qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1147 qinfo->nb_desc = txq_info->entries;
1151 * The function is used by the secondary process as well. It must not
1152 * use any process-local pointers from the adapter data.
1155 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1157 const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1158 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1159 struct sfc_rxq_info *rxq_info;
1161 SFC_ASSERT(rx_queue_id < sas->rxq_count);
1162 rxq_info = &sas->rxq_info[rx_queue_id];
1164 if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1167 return sap->dp_rx->qdesc_npending(rxq_info->dp);
1171 * The function is used by the secondary process as well. It must not
1172 * use any process-local pointers from the adapter data.
1175 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1177 struct sfc_dp_rxq *dp_rxq = queue;
1178 const struct sfc_dp_rx *dp_rx;
1180 dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1182 return offset < dp_rx->qdesc_npending(dp_rxq);
1186 * The function is used by the secondary process as well. It must not
1187 * use any process-local pointers from the adapter data.
1190 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1192 struct sfc_dp_rxq *dp_rxq = queue;
1193 const struct sfc_dp_rx *dp_rx;
1195 dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1197 return dp_rx->qdesc_status(dp_rxq, offset);
1201 * The function is used by the secondary process as well. It must not
1202 * use any process-local pointers from the adapter data.
1205 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1207 struct sfc_dp_txq *dp_txq = queue;
1208 const struct sfc_dp_tx *dp_tx;
1210 dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1212 return dp_tx->qdesc_status(dp_txq, offset);
1216 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1218 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1219 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1222 sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1224 sfc_adapter_lock(sa);
1227 if (sa->state != SFC_ADAPTER_STARTED)
1228 goto fail_not_started;
1230 if (sas->rxq_info[rx_queue_id].state != SFC_RXQ_INITIALIZED)
1231 goto fail_not_setup;
1233 rc = sfc_rx_qstart(sa, rx_queue_id);
1235 goto fail_rx_qstart;
1237 sas->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1239 sfc_adapter_unlock(sa);
1246 sfc_adapter_unlock(sa);
1252 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1254 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1255 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1257 sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1259 sfc_adapter_lock(sa);
1260 sfc_rx_qstop(sa, rx_queue_id);
1262 sas->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1264 sfc_adapter_unlock(sa);
1270 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1272 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1273 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1276 sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1278 sfc_adapter_lock(sa);
1281 if (sa->state != SFC_ADAPTER_STARTED)
1282 goto fail_not_started;
1284 if (sas->txq_info[tx_queue_id].state != SFC_TXQ_INITIALIZED)
1285 goto fail_not_setup;
1287 rc = sfc_tx_qstart(sa, tx_queue_id);
1289 goto fail_tx_qstart;
1291 sas->txq_info[tx_queue_id].deferred_started = B_TRUE;
1293 sfc_adapter_unlock(sa);
1300 sfc_adapter_unlock(sa);
1306 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1308 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1309 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1311 sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1313 sfc_adapter_lock(sa);
1315 sfc_tx_qstop(sa, tx_queue_id);
1317 sas->txq_info[tx_queue_id].deferred_started = B_FALSE;
1319 sfc_adapter_unlock(sa);
1323 static efx_tunnel_protocol_t
1324 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1327 case RTE_TUNNEL_TYPE_VXLAN:
1328 return EFX_TUNNEL_PROTOCOL_VXLAN;
1329 case RTE_TUNNEL_TYPE_GENEVE:
1330 return EFX_TUNNEL_PROTOCOL_GENEVE;
1332 return EFX_TUNNEL_NPROTOS;
1336 enum sfc_udp_tunnel_op_e {
1337 SFC_UDP_TUNNEL_ADD_PORT,
1338 SFC_UDP_TUNNEL_DEL_PORT,
1342 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1343 struct rte_eth_udp_tunnel *tunnel_udp,
1344 enum sfc_udp_tunnel_op_e op)
1346 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1347 efx_tunnel_protocol_t tunnel_proto;
1350 sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1351 (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1352 (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1353 tunnel_udp->udp_port, tunnel_udp->prot_type);
1356 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1357 if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1359 goto fail_bad_proto;
1362 sfc_adapter_lock(sa);
1365 case SFC_UDP_TUNNEL_ADD_PORT:
1366 rc = efx_tunnel_config_udp_add(sa->nic,
1367 tunnel_udp->udp_port,
1370 case SFC_UDP_TUNNEL_DEL_PORT:
1371 rc = efx_tunnel_config_udp_remove(sa->nic,
1372 tunnel_udp->udp_port,
1383 if (sa->state == SFC_ADAPTER_STARTED) {
1384 rc = efx_tunnel_reconfigure(sa->nic);
1387 * Configuration is accepted by FW and MC reboot
1388 * is initiated to apply the changes. MC reboot
1389 * will be handled in a usual way (MC reboot
1390 * event on management event queue and adapter
1394 } else if (rc != 0) {
1395 goto fail_reconfigure;
1399 sfc_adapter_unlock(sa);
1403 /* Remove/restore entry since the change makes the trouble */
1405 case SFC_UDP_TUNNEL_ADD_PORT:
1406 (void)efx_tunnel_config_udp_remove(sa->nic,
1407 tunnel_udp->udp_port,
1410 case SFC_UDP_TUNNEL_DEL_PORT:
1411 (void)efx_tunnel_config_udp_add(sa->nic,
1412 tunnel_udp->udp_port,
1419 sfc_adapter_unlock(sa);
1427 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1428 struct rte_eth_udp_tunnel *tunnel_udp)
1430 return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1434 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1435 struct rte_eth_udp_tunnel *tunnel_udp)
1437 return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1441 * The function is used by the secondary process as well. It must not
1442 * use any process-local pointers from the adapter data.
1445 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1446 struct rte_eth_rss_conf *rss_conf)
1448 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1449 struct sfc_rss *rss = &sas->rss;
1451 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1455 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1456 * hence, conversion is done here to derive a correct set of ETH_RSS
1457 * flags which corresponds to the active EFX configuration stored
1458 * locally in 'sfc_adapter' and kept up-to-date
1460 rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1461 rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1462 if (rss_conf->rss_key != NULL)
1463 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1469 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1470 struct rte_eth_rss_conf *rss_conf)
1472 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1473 struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1474 unsigned int efx_hash_types;
1477 if (sfc_sa2shared(sa)->isolated)
1480 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1481 sfc_err(sa, "RSS is not available");
1485 if (rss->channels == 0) {
1486 sfc_err(sa, "RSS is not configured");
1490 if ((rss_conf->rss_key != NULL) &&
1491 (rss_conf->rss_key_len != sizeof(rss->key))) {
1492 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1497 sfc_adapter_lock(sa);
1499 rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1501 goto fail_rx_hf_rte_to_efx;
1503 rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1504 rss->hash_alg, efx_hash_types, B_TRUE);
1506 goto fail_scale_mode_set;
1508 if (rss_conf->rss_key != NULL) {
1509 if (sa->state == SFC_ADAPTER_STARTED) {
1510 rc = efx_rx_scale_key_set(sa->nic,
1511 EFX_RSS_CONTEXT_DEFAULT,
1515 goto fail_scale_key_set;
1518 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1521 rss->hash_types = efx_hash_types;
1523 sfc_adapter_unlock(sa);
1528 if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1529 EFX_RX_HASHALG_TOEPLITZ,
1530 rss->hash_types, B_TRUE) != 0)
1531 sfc_err(sa, "failed to restore RSS mode");
1533 fail_scale_mode_set:
1534 fail_rx_hf_rte_to_efx:
1535 sfc_adapter_unlock(sa);
1540 * The function is used by the secondary process as well. It must not
1541 * use any process-local pointers from the adapter data.
1544 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1545 struct rte_eth_rss_reta_entry64 *reta_conf,
1548 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1549 struct sfc_rss *rss = &sas->rss;
1552 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1555 if (rss->channels == 0)
1558 if (reta_size != EFX_RSS_TBL_SIZE)
1561 for (entry = 0; entry < reta_size; entry++) {
1562 int grp = entry / RTE_RETA_GROUP_SIZE;
1563 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1565 if ((reta_conf[grp].mask >> grp_idx) & 1)
1566 reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1573 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1574 struct rte_eth_rss_reta_entry64 *reta_conf,
1577 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1578 struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1579 unsigned int *rss_tbl_new;
1584 if (sfc_sa2shared(sa)->isolated)
1587 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1588 sfc_err(sa, "RSS is not available");
1592 if (rss->channels == 0) {
1593 sfc_err(sa, "RSS is not configured");
1597 if (reta_size != EFX_RSS_TBL_SIZE) {
1598 sfc_err(sa, "RETA size is wrong (should be %u)",
1603 rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1604 if (rss_tbl_new == NULL)
1607 sfc_adapter_lock(sa);
1609 rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1611 for (entry = 0; entry < reta_size; entry++) {
1612 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1613 struct rte_eth_rss_reta_entry64 *grp;
1615 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1617 if (grp->mask & (1ull << grp_idx)) {
1618 if (grp->reta[grp_idx] >= rss->channels) {
1620 goto bad_reta_entry;
1622 rss_tbl_new[entry] = grp->reta[grp_idx];
1626 if (sa->state == SFC_ADAPTER_STARTED) {
1627 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1628 rss_tbl_new, EFX_RSS_TBL_SIZE);
1630 goto fail_scale_tbl_set;
1633 rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1637 sfc_adapter_unlock(sa);
1639 rte_free(rss_tbl_new);
1641 SFC_ASSERT(rc >= 0);
1646 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1647 enum rte_filter_op filter_op,
1650 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1653 sfc_log_init(sa, "entry");
1655 switch (filter_type) {
1656 case RTE_ETH_FILTER_NONE:
1657 sfc_err(sa, "Global filters configuration not supported");
1659 case RTE_ETH_FILTER_MACVLAN:
1660 sfc_err(sa, "MACVLAN filters not supported");
1662 case RTE_ETH_FILTER_ETHERTYPE:
1663 sfc_err(sa, "EtherType filters not supported");
1665 case RTE_ETH_FILTER_FLEXIBLE:
1666 sfc_err(sa, "Flexible filters not supported");
1668 case RTE_ETH_FILTER_SYN:
1669 sfc_err(sa, "SYN filters not supported");
1671 case RTE_ETH_FILTER_NTUPLE:
1672 sfc_err(sa, "NTUPLE filters not supported");
1674 case RTE_ETH_FILTER_TUNNEL:
1675 sfc_err(sa, "Tunnel filters not supported");
1677 case RTE_ETH_FILTER_FDIR:
1678 sfc_err(sa, "Flow Director filters not supported");
1680 case RTE_ETH_FILTER_HASH:
1681 sfc_err(sa, "Hash filters not supported");
1683 case RTE_ETH_FILTER_GENERIC:
1684 if (filter_op != RTE_ETH_FILTER_GET) {
1687 *(const void **)arg = &sfc_flow_ops;
1692 sfc_err(sa, "Unknown filter type %u", filter_type);
1696 sfc_log_init(sa, "exit: %d", -rc);
1697 SFC_ASSERT(rc >= 0);
1702 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1704 const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1707 * If Rx datapath does not provide callback to check mempool,
1708 * all pools are supported.
1710 if (sap->dp_rx->pool_ops_supported == NULL)
1713 return sap->dp_rx->pool_ops_supported(pool);
1716 static const struct eth_dev_ops sfc_eth_dev_ops = {
1717 .dev_configure = sfc_dev_configure,
1718 .dev_start = sfc_dev_start,
1719 .dev_stop = sfc_dev_stop,
1720 .dev_set_link_up = sfc_dev_set_link_up,
1721 .dev_set_link_down = sfc_dev_set_link_down,
1722 .dev_close = sfc_dev_close,
1723 .promiscuous_enable = sfc_dev_promisc_enable,
1724 .promiscuous_disable = sfc_dev_promisc_disable,
1725 .allmulticast_enable = sfc_dev_allmulti_enable,
1726 .allmulticast_disable = sfc_dev_allmulti_disable,
1727 .link_update = sfc_dev_link_update,
1728 .stats_get = sfc_stats_get,
1729 .stats_reset = sfc_stats_reset,
1730 .xstats_get = sfc_xstats_get,
1731 .xstats_reset = sfc_stats_reset,
1732 .xstats_get_names = sfc_xstats_get_names,
1733 .dev_infos_get = sfc_dev_infos_get,
1734 .dev_supported_ptypes_get = sfc_dev_supported_ptypes_get,
1735 .mtu_set = sfc_dev_set_mtu,
1736 .rx_queue_start = sfc_rx_queue_start,
1737 .rx_queue_stop = sfc_rx_queue_stop,
1738 .tx_queue_start = sfc_tx_queue_start,
1739 .tx_queue_stop = sfc_tx_queue_stop,
1740 .rx_queue_setup = sfc_rx_queue_setup,
1741 .rx_queue_release = sfc_rx_queue_release,
1742 .rx_queue_count = sfc_rx_queue_count,
1743 .rx_descriptor_done = sfc_rx_descriptor_done,
1744 .rx_descriptor_status = sfc_rx_descriptor_status,
1745 .tx_descriptor_status = sfc_tx_descriptor_status,
1746 .tx_queue_setup = sfc_tx_queue_setup,
1747 .tx_queue_release = sfc_tx_queue_release,
1748 .flow_ctrl_get = sfc_flow_ctrl_get,
1749 .flow_ctrl_set = sfc_flow_ctrl_set,
1750 .mac_addr_set = sfc_mac_addr_set,
1751 .udp_tunnel_port_add = sfc_dev_udp_tunnel_port_add,
1752 .udp_tunnel_port_del = sfc_dev_udp_tunnel_port_del,
1753 .reta_update = sfc_dev_rss_reta_update,
1754 .reta_query = sfc_dev_rss_reta_query,
1755 .rss_hash_update = sfc_dev_rss_hash_update,
1756 .rss_hash_conf_get = sfc_dev_rss_hash_conf_get,
1757 .filter_ctrl = sfc_dev_filter_ctrl,
1758 .set_mc_addr_list = sfc_set_mc_addr_list,
1759 .rxq_info_get = sfc_rx_queue_info_get,
1760 .txq_info_get = sfc_tx_queue_info_get,
1761 .fw_version_get = sfc_fw_version_get,
1762 .xstats_get_by_id = sfc_xstats_get_by_id,
1763 .xstats_get_names_by_id = sfc_xstats_get_names_by_id,
1764 .pool_ops_supported = sfc_pool_ops_supported,
1768 * Duplicate a string in potentially shared memory required for
1769 * multi-process support.
1771 * strdup() allocates from process-local heap/memory.
1774 sfc_strdup(const char *str)
1782 size = strlen(str) + 1;
1783 copy = rte_malloc(__func__, size, 0);
1785 rte_memcpy(copy, str, size);
1791 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1793 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1794 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1795 const struct sfc_dp_rx *dp_rx;
1796 const struct sfc_dp_tx *dp_tx;
1797 const efx_nic_cfg_t *encp;
1798 unsigned int avail_caps = 0;
1799 const char *rx_name = NULL;
1800 const char *tx_name = NULL;
1803 switch (sa->family) {
1804 case EFX_FAMILY_HUNTINGTON:
1805 case EFX_FAMILY_MEDFORD:
1806 case EFX_FAMILY_MEDFORD2:
1807 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1813 encp = efx_nic_cfg_get(sa->nic);
1814 if (encp->enc_rx_es_super_buffer_supported)
1815 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1817 rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1818 sfc_kvarg_string_handler, &rx_name);
1820 goto fail_kvarg_rx_datapath;
1822 if (rx_name != NULL) {
1823 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1824 if (dp_rx == NULL) {
1825 sfc_err(sa, "Rx datapath %s not found", rx_name);
1829 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1831 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1834 goto fail_dp_rx_caps;
1837 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1838 if (dp_rx == NULL) {
1839 sfc_err(sa, "Rx datapath by caps %#x not found",
1846 sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1847 if (sas->dp_rx_name == NULL) {
1849 goto fail_dp_rx_name;
1852 sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
1854 rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1855 sfc_kvarg_string_handler, &tx_name);
1857 goto fail_kvarg_tx_datapath;
1859 if (tx_name != NULL) {
1860 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1861 if (dp_tx == NULL) {
1862 sfc_err(sa, "Tx datapath %s not found", tx_name);
1866 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
1868 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1871 goto fail_dp_tx_caps;
1874 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1875 if (dp_tx == NULL) {
1876 sfc_err(sa, "Tx datapath by caps %#x not found",
1883 sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
1884 if (sas->dp_tx_name == NULL) {
1886 goto fail_dp_tx_name;
1889 sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
1891 sa->priv.dp_rx = dp_rx;
1892 sa->priv.dp_tx = dp_tx;
1894 dev->rx_pkt_burst = dp_rx->pkt_burst;
1895 dev->tx_pkt_prepare = dp_tx->pkt_prepare;
1896 dev->tx_pkt_burst = dp_tx->pkt_burst;
1898 dev->dev_ops = &sfc_eth_dev_ops;
1905 fail_kvarg_tx_datapath:
1906 rte_free(sas->dp_rx_name);
1907 sas->dp_rx_name = NULL;
1912 fail_kvarg_rx_datapath:
1917 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1919 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1920 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1922 dev->dev_ops = NULL;
1923 dev->tx_pkt_prepare = NULL;
1924 dev->rx_pkt_burst = NULL;
1925 dev->tx_pkt_burst = NULL;
1927 rte_free(sas->dp_tx_name);
1928 sas->dp_tx_name = NULL;
1929 sa->priv.dp_tx = NULL;
1931 rte_free(sas->dp_rx_name);
1932 sas->dp_rx_name = NULL;
1933 sa->priv.dp_rx = NULL;
1936 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1937 .dev_supported_ptypes_get = sfc_dev_supported_ptypes_get,
1938 .rx_queue_count = sfc_rx_queue_count,
1939 .rx_descriptor_done = sfc_rx_descriptor_done,
1940 .rx_descriptor_status = sfc_rx_descriptor_status,
1941 .tx_descriptor_status = sfc_tx_descriptor_status,
1942 .reta_query = sfc_dev_rss_reta_query,
1943 .rss_hash_conf_get = sfc_dev_rss_hash_conf_get,
1944 .rxq_info_get = sfc_rx_queue_info_get,
1945 .txq_info_get = sfc_tx_queue_info_get,
1949 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
1951 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1952 struct sfc_adapter_priv *sap;
1953 const struct sfc_dp_rx *dp_rx;
1954 const struct sfc_dp_tx *dp_tx;
1958 * Allocate process private data from heap, since it should not
1959 * be located in shared memory allocated using rte_malloc() API.
1961 sap = calloc(1, sizeof(*sap));
1964 goto fail_alloc_priv;
1967 sap->logtype_main = logtype_main;
1969 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
1970 if (dp_rx == NULL) {
1971 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1972 "cannot find %s Rx datapath", sas->dp_rx_name);
1976 if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1977 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1978 "%s Rx datapath does not support multi-process",
1981 goto fail_dp_rx_multi_process;
1984 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
1985 if (dp_tx == NULL) {
1986 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1987 "cannot find %s Tx datapath", sas->dp_tx_name);
1991 if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1992 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1993 "%s Tx datapath does not support multi-process",
1996 goto fail_dp_tx_multi_process;
2002 dev->process_private = sap;
2003 dev->rx_pkt_burst = dp_rx->pkt_burst;
2004 dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2005 dev->tx_pkt_burst = dp_tx->pkt_burst;
2006 dev->dev_ops = &sfc_eth_dev_secondary_ops;
2010 fail_dp_tx_multi_process:
2012 fail_dp_rx_multi_process:
2021 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
2023 free(dev->process_private);
2024 dev->process_private = NULL;
2025 dev->dev_ops = NULL;
2026 dev->tx_pkt_prepare = NULL;
2027 dev->tx_pkt_burst = NULL;
2028 dev->rx_pkt_burst = NULL;
2032 sfc_register_dp(void)
2035 if (TAILQ_EMPTY(&sfc_dp_head)) {
2036 /* Prefer EF10 datapath */
2037 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2038 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2039 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2041 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2042 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2043 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2048 sfc_eth_dev_init(struct rte_eth_dev *dev)
2050 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2051 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2052 uint32_t logtype_main;
2053 struct sfc_adapter *sa;
2055 const efx_nic_cfg_t *encp;
2056 const struct rte_ether_addr *from;
2060 logtype_main = sfc_register_logtype(&pci_dev->addr,
2061 SFC_LOGTYPE_MAIN_STR,
2064 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2065 return -sfc_eth_dev_secondary_init(dev, logtype_main);
2067 /* Required for logging */
2068 sas->pci_addr = pci_dev->addr;
2069 sas->port_id = dev->data->port_id;
2072 * Allocate process private data from heap, since it should not
2073 * be located in shared memory allocated using rte_malloc() API.
2075 sa = calloc(1, sizeof(*sa));
2081 dev->process_private = sa;
2083 /* Required for logging */
2084 sa->priv.shared = sas;
2085 sa->priv.logtype_main = logtype_main;
2089 /* Copy PCI device info to the dev->data */
2090 rte_eth_copy_pci_info(dev, pci_dev);
2092 rc = sfc_kvargs_parse(sa);
2094 goto fail_kvargs_parse;
2096 sfc_log_init(sa, "entry");
2098 dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2099 if (dev->data->mac_addrs == NULL) {
2101 goto fail_mac_addrs;
2104 sfc_adapter_lock_init(sa);
2105 sfc_adapter_lock(sa);
2107 sfc_log_init(sa, "probing");
2112 sfc_log_init(sa, "set device ops");
2113 rc = sfc_eth_dev_set_ops(dev);
2117 sfc_log_init(sa, "attaching");
2118 rc = sfc_attach(sa);
2122 encp = efx_nic_cfg_get(sa->nic);
2125 * The arguments are really reverse order in comparison to
2126 * Linux kernel. Copy from NIC config to Ethernet device data.
2128 from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2129 rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2131 sfc_adapter_unlock(sa);
2133 sfc_log_init(sa, "done");
2137 sfc_eth_dev_clear_ops(dev);
2143 sfc_adapter_unlock(sa);
2144 sfc_adapter_lock_fini(sa);
2145 rte_free(dev->data->mac_addrs);
2146 dev->data->mac_addrs = NULL;
2149 sfc_kvargs_cleanup(sa);
2152 sfc_log_init(sa, "failed %d", rc);
2153 dev->process_private = NULL;
2162 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2164 struct sfc_adapter *sa;
2166 if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2167 sfc_eth_dev_secondary_clear_ops(dev);
2171 sa = sfc_adapter_by_eth_dev(dev);
2172 sfc_log_init(sa, "entry");
2174 sfc_adapter_lock(sa);
2176 sfc_eth_dev_clear_ops(dev);
2181 sfc_kvargs_cleanup(sa);
2183 sfc_adapter_unlock(sa);
2184 sfc_adapter_lock_fini(sa);
2186 sfc_log_init(sa, "done");
2188 /* Required for logging, so cleanup last */
2191 dev->process_private = NULL;
2197 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2198 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2199 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2200 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2201 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2202 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2203 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2204 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2205 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2206 { .vendor_id = 0 /* sentinel */ }
2209 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2210 struct rte_pci_device *pci_dev)
2212 return rte_eth_dev_pci_generic_probe(pci_dev,
2213 sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
2216 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2218 return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2221 static struct rte_pci_driver sfc_efx_pmd = {
2222 .id_table = pci_id_sfc_efx_map,
2224 RTE_PCI_DRV_INTR_LSC |
2225 RTE_PCI_DRV_NEED_MAPPING,
2226 .probe = sfc_eth_dev_pci_probe,
2227 .remove = sfc_eth_dev_pci_remove,
2230 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2231 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2232 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2233 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2234 SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2235 SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2236 SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2237 SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2238 SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2239 SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2241 RTE_INIT(sfc_driver_register_logtype)
2245 ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2247 sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;