1 /* SPDX-License-Identifier: BSD-3-Clause
3 * Copyright (c) 2016-2018 Solarflare Communications Inc.
6 * This software was jointly developed between OKTET Labs (under contract
7 * for Solarflare) and Solarflare Communications, Inc.
11 #include <rte_ethdev_driver.h>
12 #include <rte_ethdev_pci.h>
14 #include <rte_bus_pci.h>
15 #include <rte_errno.h>
16 #include <rte_string_fns.h>
21 #include "sfc_debug.h"
23 #include "sfc_kvargs.h"
29 #include "sfc_dp_rx.h"
31 uint32_t sfc_logtype_driver;
33 static struct sfc_dp_list sfc_dp_head =
34 TAILQ_HEAD_INITIALIZER(sfc_dp_head);
37 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
39 struct sfc_adapter *sa = dev->data->dev_private;
40 efx_nic_fw_info_t enfi;
45 * Return value of the callback is likely supposed to be
46 * equal to or greater than 0, nevertheless, if an error
47 * occurs, it will be desirable to pass it to the caller
49 if ((fw_version == NULL) || (fw_size == 0))
52 rc = efx_nic_get_fw_version(sa->nic, &enfi);
56 ret = snprintf(fw_version, fw_size,
57 "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
58 enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
59 enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
63 if (enfi.enfi_dpcpu_fw_ids_valid) {
64 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
67 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
68 fw_size - dpcpu_fw_ids_offset,
69 " rx%" PRIx16 " tx%" PRIx16,
70 enfi.enfi_rx_dpcpu_fw_id,
71 enfi.enfi_tx_dpcpu_fw_id);
78 if (fw_size < (size_t)(++ret))
85 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
87 struct sfc_adapter *sa = dev->data->dev_private;
88 struct sfc_rss *rss = &sa->rss;
89 uint64_t txq_offloads_def = 0;
91 sfc_log_init(sa, "entry");
93 dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
95 /* Autonegotiation may be disabled */
96 dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
97 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
98 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
99 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
100 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
101 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_25000FDX)
102 dev_info->speed_capa |= ETH_LINK_SPEED_25G;
103 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
104 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
105 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_50000FDX)
106 dev_info->speed_capa |= ETH_LINK_SPEED_50G;
107 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_100000FDX)
108 dev_info->speed_capa |= ETH_LINK_SPEED_100G;
110 dev_info->max_rx_queues = sa->rxq_max;
111 dev_info->max_tx_queues = sa->txq_max;
113 /* By default packets are dropped if no descriptors are available */
114 dev_info->default_rxconf.rx_drop_en = 1;
116 dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
119 * rx_offload_capa includes both device and queue offloads since
120 * the latter may be requested on a per device basis which makes
121 * sense when some offloads are needed to be set on all queues.
123 dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
124 dev_info->rx_queue_offload_capa;
126 dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
129 * tx_offload_capa includes both device and queue offloads since
130 * the latter may be requested on a per device basis which makes
131 * sense when some offloads are needed to be set on all queues.
133 dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
134 dev_info->tx_queue_offload_capa;
136 if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
137 txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
139 dev_info->default_txconf.offloads |= txq_offloads_def;
141 if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
145 for (i = 0; i < rss->hf_map_nb_entries; ++i)
146 rte_hf |= rss->hf_map[i].rte;
148 dev_info->reta_size = EFX_RSS_TBL_SIZE;
149 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
150 dev_info->flow_type_rss_offloads = rte_hf;
153 /* Initialize to hardware limits */
154 dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
155 dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
156 /* The RXQ hardware requires that the descriptor count is a power
157 * of 2, but rx_desc_lim cannot properly describe that constraint.
159 dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
161 /* Initialize to hardware limits */
162 dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
163 dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
165 * The TXQ hardware requires that the descriptor count is a power
166 * of 2, but tx_desc_lim cannot properly describe that constraint
168 dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
170 if (sa->dp_rx->get_dev_info != NULL)
171 sa->dp_rx->get_dev_info(dev_info);
172 if (sa->dp_tx->get_dev_info != NULL)
173 sa->dp_tx->get_dev_info(dev_info);
175 dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
176 RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
179 static const uint32_t *
180 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
182 struct sfc_adapter *sa = dev->data->dev_private;
183 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
184 uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
186 return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
190 sfc_dev_configure(struct rte_eth_dev *dev)
192 struct rte_eth_dev_data *dev_data = dev->data;
193 struct sfc_adapter *sa = dev_data->dev_private;
196 sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
197 dev_data->nb_rx_queues, dev_data->nb_tx_queues);
199 sfc_adapter_lock(sa);
201 case SFC_ADAPTER_CONFIGURED:
203 case SFC_ADAPTER_INITIALIZED:
204 rc = sfc_configure(sa);
207 sfc_err(sa, "unexpected adapter state %u to configure",
212 sfc_adapter_unlock(sa);
214 sfc_log_init(sa, "done %d", rc);
220 sfc_dev_start(struct rte_eth_dev *dev)
222 struct sfc_adapter *sa = dev->data->dev_private;
225 sfc_log_init(sa, "entry");
227 sfc_adapter_lock(sa);
229 sfc_adapter_unlock(sa);
231 sfc_log_init(sa, "done %d", rc);
237 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
239 struct sfc_adapter *sa = dev->data->dev_private;
240 struct rte_eth_link current_link;
243 sfc_log_init(sa, "entry");
245 if (sa->state != SFC_ADAPTER_STARTED) {
246 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, ¤t_link);
247 } else if (wait_to_complete) {
248 efx_link_mode_t link_mode;
250 if (efx_port_poll(sa->nic, &link_mode) != 0)
251 link_mode = EFX_LINK_UNKNOWN;
252 sfc_port_link_mode_to_info(link_mode, ¤t_link);
255 sfc_ev_mgmt_qpoll(sa);
256 rte_eth_linkstatus_get(dev, ¤t_link);
259 ret = rte_eth_linkstatus_set(dev, ¤t_link);
261 sfc_notice(sa, "Link status is %s",
262 current_link.link_status ? "UP" : "DOWN");
268 sfc_dev_stop(struct rte_eth_dev *dev)
270 struct sfc_adapter *sa = dev->data->dev_private;
272 sfc_log_init(sa, "entry");
274 sfc_adapter_lock(sa);
276 sfc_adapter_unlock(sa);
278 sfc_log_init(sa, "done");
282 sfc_dev_set_link_up(struct rte_eth_dev *dev)
284 struct sfc_adapter *sa = dev->data->dev_private;
287 sfc_log_init(sa, "entry");
289 sfc_adapter_lock(sa);
291 sfc_adapter_unlock(sa);
298 sfc_dev_set_link_down(struct rte_eth_dev *dev)
300 struct sfc_adapter *sa = dev->data->dev_private;
302 sfc_log_init(sa, "entry");
304 sfc_adapter_lock(sa);
306 sfc_adapter_unlock(sa);
312 sfc_dev_close(struct rte_eth_dev *dev)
314 struct sfc_adapter *sa = dev->data->dev_private;
316 sfc_log_init(sa, "entry");
318 sfc_adapter_lock(sa);
320 case SFC_ADAPTER_STARTED:
322 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
324 case SFC_ADAPTER_CONFIGURED:
326 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
328 case SFC_ADAPTER_INITIALIZED:
331 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
334 sfc_adapter_unlock(sa);
336 sfc_log_init(sa, "done");
340 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
343 struct sfc_port *port;
345 struct sfc_adapter *sa = dev->data->dev_private;
346 boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
347 const char *desc = (allmulti) ? "all-multi" : "promiscuous";
349 sfc_adapter_lock(sa);
352 toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
354 if (*toggle != enabled) {
357 if (port->isolated) {
358 sfc_warn(sa, "isolated mode is active on the port");
359 sfc_warn(sa, "the change is to be applied on the next "
360 "start provided that isolated mode is "
361 "disabled prior the next start");
362 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
363 (sfc_set_rx_mode(sa) != 0)) {
364 *toggle = !(enabled);
365 sfc_warn(sa, "Failed to %s %s mode",
366 ((enabled) ? "enable" : "disable"), desc);
370 sfc_adapter_unlock(sa);
374 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
376 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
380 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
382 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
386 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
388 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
392 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
394 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
398 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
399 uint16_t nb_rx_desc, unsigned int socket_id,
400 const struct rte_eth_rxconf *rx_conf,
401 struct rte_mempool *mb_pool)
403 struct sfc_adapter *sa = dev->data->dev_private;
406 sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
407 rx_queue_id, nb_rx_desc, socket_id);
409 sfc_adapter_lock(sa);
411 rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
416 dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
418 sfc_adapter_unlock(sa);
423 sfc_adapter_unlock(sa);
429 sfc_rx_queue_release(void *queue)
431 struct sfc_dp_rxq *dp_rxq = queue;
433 struct sfc_adapter *sa;
434 unsigned int sw_index;
439 rxq = sfc_rxq_by_dp_rxq(dp_rxq);
441 sfc_adapter_lock(sa);
443 sw_index = sfc_rxq_sw_index(rxq);
445 sfc_log_init(sa, "RxQ=%u", sw_index);
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_tx_qfini(sa, sw_index);
503 sfc_adapter_unlock(sa);
507 * Some statistics are computed as A - B where A and B each increase
508 * monotonically with some hardware counter(s) and the counters are read
511 * If packet X is counted in A, but not counted in B yet, computed value is
514 * If packet X is not counted in A at the moment of reading the counter,
515 * but counted in B at the moment of reading the counter, computed value
518 * However, counter which grows backward is worse evil than slightly wrong
519 * value. So, let's try to guarantee that it never happens except may be
520 * the case when the MAC stats are zeroed as a result of a NIC reset.
523 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
525 if ((int64_t)(newval - *stat) > 0 || newval == 0)
530 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
532 struct sfc_adapter *sa = dev->data->dev_private;
533 struct sfc_port *port = &sa->port;
537 rte_spinlock_lock(&port->mac_stats_lock);
539 ret = sfc_port_update_mac_stats(sa);
543 mac_stats = port->mac_stats_buf;
545 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
546 EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
548 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
549 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
550 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
552 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
553 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
554 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
556 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
557 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
558 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
560 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
561 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
562 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
563 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
564 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
566 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
567 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
568 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
570 * Take into account stats which are whenever supported
571 * on EF10. If some stat is not supported by current
572 * firmware variant or HW revision, it is guaranteed
573 * to be zero in mac_stats.
576 mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
577 mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
578 mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
579 mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
580 mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
581 mac_stats[EFX_MAC_PM_TRUNC_QBB] +
582 mac_stats[EFX_MAC_PM_DISCARD_QBB] +
583 mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
584 mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
585 mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
587 mac_stats[EFX_MAC_RX_FCS_ERRORS] +
588 mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
589 mac_stats[EFX_MAC_RX_JABBER_PKTS];
590 /* no oerrors counters supported on EF10 */
592 /* Exclude missed, errors and pauses from Rx packets */
593 sfc_update_diff_stat(&port->ipackets,
594 mac_stats[EFX_MAC_RX_PKTS] -
595 mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
596 stats->imissed - stats->ierrors);
597 stats->ipackets = port->ipackets;
601 rte_spinlock_unlock(&port->mac_stats_lock);
602 SFC_ASSERT(ret >= 0);
607 sfc_stats_reset(struct rte_eth_dev *dev)
609 struct sfc_adapter *sa = dev->data->dev_private;
610 struct sfc_port *port = &sa->port;
613 if (sa->state != SFC_ADAPTER_STARTED) {
615 * The operation cannot be done if port is not started; it
616 * will be scheduled to be done during the next port start
618 port->mac_stats_reset_pending = B_TRUE;
622 rc = sfc_port_reset_mac_stats(sa);
624 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
628 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
629 unsigned int xstats_count)
631 struct sfc_adapter *sa = dev->data->dev_private;
632 struct sfc_port *port = &sa->port;
638 rte_spinlock_lock(&port->mac_stats_lock);
640 rc = sfc_port_update_mac_stats(sa);
647 mac_stats = port->mac_stats_buf;
649 for (i = 0; i < EFX_MAC_NSTATS; ++i) {
650 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
651 if (xstats != NULL && nstats < (int)xstats_count) {
652 xstats[nstats].id = nstats;
653 xstats[nstats].value = mac_stats[i];
660 rte_spinlock_unlock(&port->mac_stats_lock);
666 sfc_xstats_get_names(struct rte_eth_dev *dev,
667 struct rte_eth_xstat_name *xstats_names,
668 unsigned int xstats_count)
670 struct sfc_adapter *sa = dev->data->dev_private;
671 struct sfc_port *port = &sa->port;
673 unsigned int nstats = 0;
675 for (i = 0; i < EFX_MAC_NSTATS; ++i) {
676 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
677 if (xstats_names != NULL && nstats < xstats_count)
678 strlcpy(xstats_names[nstats].name,
679 efx_mac_stat_name(sa->nic, i),
680 sizeof(xstats_names[0].name));
689 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
690 uint64_t *values, unsigned int n)
692 struct sfc_adapter *sa = dev->data->dev_private;
693 struct sfc_port *port = &sa->port;
695 unsigned int nb_supported = 0;
696 unsigned int nb_written = 0;
701 if (unlikely(values == NULL) ||
702 unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
703 return port->mac_stats_nb_supported;
705 rte_spinlock_lock(&port->mac_stats_lock);
707 rc = sfc_port_update_mac_stats(sa);
714 mac_stats = port->mac_stats_buf;
716 for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
717 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
720 if ((ids == NULL) || (ids[nb_written] == nb_supported))
721 values[nb_written++] = mac_stats[i];
729 rte_spinlock_unlock(&port->mac_stats_lock);
735 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
736 struct rte_eth_xstat_name *xstats_names,
737 const uint64_t *ids, unsigned int size)
739 struct sfc_adapter *sa = dev->data->dev_private;
740 struct sfc_port *port = &sa->port;
741 unsigned int nb_supported = 0;
742 unsigned int nb_written = 0;
745 if (unlikely(xstats_names == NULL) ||
746 unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
747 return port->mac_stats_nb_supported;
749 for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
750 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
753 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
754 char *name = xstats_names[nb_written++].name;
756 strlcpy(name, efx_mac_stat_name(sa->nic, i),
757 sizeof(xstats_names[0].name));
767 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
769 struct sfc_adapter *sa = dev->data->dev_private;
770 unsigned int wanted_fc, link_fc;
772 memset(fc_conf, 0, sizeof(*fc_conf));
774 sfc_adapter_lock(sa);
776 if (sa->state == SFC_ADAPTER_STARTED)
777 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
779 link_fc = sa->port.flow_ctrl;
783 fc_conf->mode = RTE_FC_NONE;
785 case EFX_FCNTL_RESPOND:
786 fc_conf->mode = RTE_FC_RX_PAUSE;
788 case EFX_FCNTL_GENERATE:
789 fc_conf->mode = RTE_FC_TX_PAUSE;
791 case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
792 fc_conf->mode = RTE_FC_FULL;
795 sfc_err(sa, "%s: unexpected flow control value %#x",
799 fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
801 sfc_adapter_unlock(sa);
807 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
809 struct sfc_adapter *sa = dev->data->dev_private;
810 struct sfc_port *port = &sa->port;
814 if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
815 fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
816 fc_conf->mac_ctrl_frame_fwd != 0) {
817 sfc_err(sa, "unsupported flow control settings specified");
822 switch (fc_conf->mode) {
826 case RTE_FC_RX_PAUSE:
827 fcntl = EFX_FCNTL_RESPOND;
829 case RTE_FC_TX_PAUSE:
830 fcntl = EFX_FCNTL_GENERATE;
833 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
840 sfc_adapter_lock(sa);
842 if (sa->state == SFC_ADAPTER_STARTED) {
843 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
845 goto fail_mac_fcntl_set;
848 port->flow_ctrl = fcntl;
849 port->flow_ctrl_autoneg = fc_conf->autoneg;
851 sfc_adapter_unlock(sa);
856 sfc_adapter_unlock(sa);
863 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
865 struct sfc_adapter *sa = dev->data->dev_private;
866 size_t pdu = EFX_MAC_PDU(mtu);
870 sfc_log_init(sa, "mtu=%u", mtu);
873 if (pdu < EFX_MAC_PDU_MIN) {
874 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
875 (unsigned int)mtu, (unsigned int)pdu,
879 if (pdu > EFX_MAC_PDU_MAX) {
880 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
881 (unsigned int)mtu, (unsigned int)pdu,
886 sfc_adapter_lock(sa);
888 if (pdu != sa->port.pdu) {
889 if (sa->state == SFC_ADAPTER_STARTED) {
892 old_pdu = sa->port.pdu;
903 * The driver does not use it, but other PMDs update jumbo frame
904 * flag and max_rx_pkt_len when MTU is set.
906 if (mtu > ETHER_MAX_LEN) {
907 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
908 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
911 dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
913 sfc_adapter_unlock(sa);
915 sfc_log_init(sa, "done");
919 sa->port.pdu = old_pdu;
920 if (sfc_start(sa) != 0)
921 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
922 "PDU max size - port is stopped",
923 (unsigned int)pdu, (unsigned int)old_pdu);
924 sfc_adapter_unlock(sa);
927 sfc_log_init(sa, "failed %d", rc);
932 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
934 struct sfc_adapter *sa = dev->data->dev_private;
935 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
936 struct sfc_port *port = &sa->port;
937 struct ether_addr *old_addr = &dev->data->mac_addrs[0];
940 sfc_adapter_lock(sa);
943 * Copy the address to the device private data so that
944 * it could be recalled in the case of adapter restart.
946 ether_addr_copy(mac_addr, &port->default_mac_addr);
949 * Neither of the two following checks can return
950 * an error. The new MAC address is preserved in
951 * the device private data and can be activated
952 * on the next port start if the user prevents
953 * isolated mode from being enabled.
955 if (port->isolated) {
956 sfc_warn(sa, "isolated mode is active on the port");
957 sfc_warn(sa, "will not set MAC address");
961 if (sa->state != SFC_ADAPTER_STARTED) {
962 sfc_notice(sa, "the port is not started");
963 sfc_notice(sa, "the new MAC address will be set on port start");
968 if (encp->enc_allow_set_mac_with_installed_filters) {
969 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
971 sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
976 * Changing the MAC address by means of MCDI request
977 * has no effect on received traffic, therefore
978 * we also need to update unicast filters
980 rc = sfc_set_rx_mode(sa);
982 sfc_err(sa, "cannot set filter (rc = %u)", rc);
983 /* Rollback the old address */
984 (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
985 (void)sfc_set_rx_mode(sa);
988 sfc_warn(sa, "cannot set MAC address with filters installed");
989 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
990 sfc_warn(sa, "(some traffic may be dropped)");
993 * Since setting MAC address with filters installed is not
994 * allowed on the adapter, the new MAC address will be set
995 * by means of adapter restart. sfc_start() shall retrieve
996 * the new address from the device private data and set it.
1001 sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1006 ether_addr_copy(old_addr, &port->default_mac_addr);
1008 sfc_adapter_unlock(sa);
1010 SFC_ASSERT(rc >= 0);
1016 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
1017 uint32_t nb_mc_addr)
1019 struct sfc_adapter *sa = dev->data->dev_private;
1020 struct sfc_port *port = &sa->port;
1021 uint8_t *mc_addrs = port->mcast_addrs;
1025 if (port->isolated) {
1026 sfc_err(sa, "isolated mode is active on the port");
1027 sfc_err(sa, "will not set multicast address list");
1031 if (mc_addrs == NULL)
1034 if (nb_mc_addr > port->max_mcast_addrs) {
1035 sfc_err(sa, "too many multicast addresses: %u > %u",
1036 nb_mc_addr, port->max_mcast_addrs);
1040 for (i = 0; i < nb_mc_addr; ++i) {
1041 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1043 mc_addrs += EFX_MAC_ADDR_LEN;
1046 port->nb_mcast_addrs = nb_mc_addr;
1048 if (sa->state != SFC_ADAPTER_STARTED)
1051 rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1052 port->nb_mcast_addrs);
1054 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1056 SFC_ASSERT(rc >= 0);
1061 * The function is used by the secondary process as well. It must not
1062 * use any process-local pointers from the adapter data.
1065 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1066 struct rte_eth_rxq_info *qinfo)
1068 struct sfc_adapter *sa = dev->data->dev_private;
1069 struct sfc_rxq_info *rxq_info;
1071 sfc_adapter_lock(sa);
1073 SFC_ASSERT(rx_queue_id < sa->rxq_count);
1075 rxq_info = &sa->rxq_info[rx_queue_id];
1077 qinfo->mp = rxq_info->refill_mb_pool;
1078 qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1079 qinfo->conf.rx_drop_en = 1;
1080 qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1081 qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1082 if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1083 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1084 qinfo->scattered_rx = 1;
1086 qinfo->nb_desc = rxq_info->entries;
1088 sfc_adapter_unlock(sa);
1092 * The function is used by the secondary process as well. It must not
1093 * use any process-local pointers from the adapter data.
1096 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1097 struct rte_eth_txq_info *qinfo)
1099 struct sfc_adapter *sa = dev->data->dev_private;
1100 struct sfc_txq_info *txq_info;
1102 sfc_adapter_lock(sa);
1104 SFC_ASSERT(tx_queue_id < sa->txq_count);
1106 txq_info = &sa->txq_info[tx_queue_id];
1108 memset(qinfo, 0, sizeof(*qinfo));
1110 qinfo->conf.offloads = txq_info->offloads;
1111 qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1112 qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1113 qinfo->nb_desc = txq_info->entries;
1115 sfc_adapter_unlock(sa);
1119 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1121 struct sfc_adapter *sa = dev->data->dev_private;
1123 return sfc_rx_qdesc_npending(sa, rx_queue_id);
1127 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1129 struct sfc_dp_rxq *dp_rxq = queue;
1131 return sfc_rx_qdesc_done(dp_rxq, offset);
1135 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1137 struct sfc_dp_rxq *dp_rxq = queue;
1138 struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1140 return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1144 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1146 struct sfc_dp_txq *dp_txq = queue;
1147 struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1149 return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1153 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1155 struct sfc_adapter *sa = dev->data->dev_private;
1158 sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1160 sfc_adapter_lock(sa);
1163 if (sa->state != SFC_ADAPTER_STARTED)
1164 goto fail_not_started;
1166 if (sa->rxq_info[rx_queue_id].rxq == NULL)
1167 goto fail_not_setup;
1169 rc = sfc_rx_qstart(sa, rx_queue_id);
1171 goto fail_rx_qstart;
1173 sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1175 sfc_adapter_unlock(sa);
1182 sfc_adapter_unlock(sa);
1188 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1190 struct sfc_adapter *sa = dev->data->dev_private;
1192 sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1194 sfc_adapter_lock(sa);
1195 sfc_rx_qstop(sa, rx_queue_id);
1197 sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1199 sfc_adapter_unlock(sa);
1205 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1207 struct sfc_adapter *sa = dev->data->dev_private;
1210 sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1212 sfc_adapter_lock(sa);
1215 if (sa->state != SFC_ADAPTER_STARTED)
1216 goto fail_not_started;
1218 if (sa->txq_info[tx_queue_id].txq == NULL)
1219 goto fail_not_setup;
1221 rc = sfc_tx_qstart(sa, tx_queue_id);
1223 goto fail_tx_qstart;
1225 sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1227 sfc_adapter_unlock(sa);
1234 sfc_adapter_unlock(sa);
1240 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1242 struct sfc_adapter *sa = dev->data->dev_private;
1244 sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1246 sfc_adapter_lock(sa);
1248 sfc_tx_qstop(sa, tx_queue_id);
1250 sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1252 sfc_adapter_unlock(sa);
1256 static efx_tunnel_protocol_t
1257 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1260 case RTE_TUNNEL_TYPE_VXLAN:
1261 return EFX_TUNNEL_PROTOCOL_VXLAN;
1262 case RTE_TUNNEL_TYPE_GENEVE:
1263 return EFX_TUNNEL_PROTOCOL_GENEVE;
1265 return EFX_TUNNEL_NPROTOS;
1269 enum sfc_udp_tunnel_op_e {
1270 SFC_UDP_TUNNEL_ADD_PORT,
1271 SFC_UDP_TUNNEL_DEL_PORT,
1275 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1276 struct rte_eth_udp_tunnel *tunnel_udp,
1277 enum sfc_udp_tunnel_op_e op)
1279 struct sfc_adapter *sa = dev->data->dev_private;
1280 efx_tunnel_protocol_t tunnel_proto;
1283 sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1284 (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1285 (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1286 tunnel_udp->udp_port, tunnel_udp->prot_type);
1289 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1290 if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1292 goto fail_bad_proto;
1295 sfc_adapter_lock(sa);
1298 case SFC_UDP_TUNNEL_ADD_PORT:
1299 rc = efx_tunnel_config_udp_add(sa->nic,
1300 tunnel_udp->udp_port,
1303 case SFC_UDP_TUNNEL_DEL_PORT:
1304 rc = efx_tunnel_config_udp_remove(sa->nic,
1305 tunnel_udp->udp_port,
1316 if (sa->state == SFC_ADAPTER_STARTED) {
1317 rc = efx_tunnel_reconfigure(sa->nic);
1320 * Configuration is accepted by FW and MC reboot
1321 * is initiated to apply the changes. MC reboot
1322 * will be handled in a usual way (MC reboot
1323 * event on management event queue and adapter
1327 } else if (rc != 0) {
1328 goto fail_reconfigure;
1332 sfc_adapter_unlock(sa);
1336 /* Remove/restore entry since the change makes the trouble */
1338 case SFC_UDP_TUNNEL_ADD_PORT:
1339 (void)efx_tunnel_config_udp_remove(sa->nic,
1340 tunnel_udp->udp_port,
1343 case SFC_UDP_TUNNEL_DEL_PORT:
1344 (void)efx_tunnel_config_udp_add(sa->nic,
1345 tunnel_udp->udp_port,
1352 sfc_adapter_unlock(sa);
1360 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1361 struct rte_eth_udp_tunnel *tunnel_udp)
1363 return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1367 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1368 struct rte_eth_udp_tunnel *tunnel_udp)
1370 return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1374 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1375 struct rte_eth_rss_conf *rss_conf)
1377 struct sfc_adapter *sa = dev->data->dev_private;
1378 struct sfc_rss *rss = &sa->rss;
1380 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1383 sfc_adapter_lock(sa);
1386 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1387 * hence, conversion is done here to derive a correct set of ETH_RSS
1388 * flags which corresponds to the active EFX configuration stored
1389 * locally in 'sfc_adapter' and kept up-to-date
1391 rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(sa, rss->hash_types);
1392 rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1393 if (rss_conf->rss_key != NULL)
1394 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1396 sfc_adapter_unlock(sa);
1402 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1403 struct rte_eth_rss_conf *rss_conf)
1405 struct sfc_adapter *sa = dev->data->dev_private;
1406 struct sfc_rss *rss = &sa->rss;
1407 struct sfc_port *port = &sa->port;
1408 unsigned int efx_hash_types;
1414 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1415 sfc_err(sa, "RSS is not available");
1419 if (rss->channels == 0) {
1420 sfc_err(sa, "RSS is not configured");
1424 if ((rss_conf->rss_key != NULL) &&
1425 (rss_conf->rss_key_len != sizeof(rss->key))) {
1426 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1431 sfc_adapter_lock(sa);
1433 rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1435 goto fail_rx_hf_rte_to_efx;
1437 rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1438 rss->hash_alg, efx_hash_types, B_TRUE);
1440 goto fail_scale_mode_set;
1442 if (rss_conf->rss_key != NULL) {
1443 if (sa->state == SFC_ADAPTER_STARTED) {
1444 rc = efx_rx_scale_key_set(sa->nic,
1445 EFX_RSS_CONTEXT_DEFAULT,
1449 goto fail_scale_key_set;
1452 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1455 rss->hash_types = efx_hash_types;
1457 sfc_adapter_unlock(sa);
1462 if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1463 EFX_RX_HASHALG_TOEPLITZ,
1464 rss->hash_types, B_TRUE) != 0)
1465 sfc_err(sa, "failed to restore RSS mode");
1467 fail_scale_mode_set:
1468 fail_rx_hf_rte_to_efx:
1469 sfc_adapter_unlock(sa);
1474 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1475 struct rte_eth_rss_reta_entry64 *reta_conf,
1478 struct sfc_adapter *sa = dev->data->dev_private;
1479 struct sfc_rss *rss = &sa->rss;
1480 struct sfc_port *port = &sa->port;
1483 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1486 if (rss->channels == 0)
1489 if (reta_size != EFX_RSS_TBL_SIZE)
1492 sfc_adapter_lock(sa);
1494 for (entry = 0; entry < reta_size; entry++) {
1495 int grp = entry / RTE_RETA_GROUP_SIZE;
1496 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1498 if ((reta_conf[grp].mask >> grp_idx) & 1)
1499 reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1502 sfc_adapter_unlock(sa);
1508 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1509 struct rte_eth_rss_reta_entry64 *reta_conf,
1512 struct sfc_adapter *sa = dev->data->dev_private;
1513 struct sfc_rss *rss = &sa->rss;
1514 struct sfc_port *port = &sa->port;
1515 unsigned int *rss_tbl_new;
1523 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1524 sfc_err(sa, "RSS is not available");
1528 if (rss->channels == 0) {
1529 sfc_err(sa, "RSS is not configured");
1533 if (reta_size != EFX_RSS_TBL_SIZE) {
1534 sfc_err(sa, "RETA size is wrong (should be %u)",
1539 rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1540 if (rss_tbl_new == NULL)
1543 sfc_adapter_lock(sa);
1545 rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1547 for (entry = 0; entry < reta_size; entry++) {
1548 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1549 struct rte_eth_rss_reta_entry64 *grp;
1551 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1553 if (grp->mask & (1ull << grp_idx)) {
1554 if (grp->reta[grp_idx] >= rss->channels) {
1556 goto bad_reta_entry;
1558 rss_tbl_new[entry] = grp->reta[grp_idx];
1562 if (sa->state == SFC_ADAPTER_STARTED) {
1563 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1564 rss_tbl_new, EFX_RSS_TBL_SIZE);
1566 goto fail_scale_tbl_set;
1569 rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1573 sfc_adapter_unlock(sa);
1575 rte_free(rss_tbl_new);
1577 SFC_ASSERT(rc >= 0);
1582 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1583 enum rte_filter_op filter_op,
1586 struct sfc_adapter *sa = dev->data->dev_private;
1589 sfc_log_init(sa, "entry");
1591 switch (filter_type) {
1592 case RTE_ETH_FILTER_NONE:
1593 sfc_err(sa, "Global filters configuration not supported");
1595 case RTE_ETH_FILTER_MACVLAN:
1596 sfc_err(sa, "MACVLAN filters not supported");
1598 case RTE_ETH_FILTER_ETHERTYPE:
1599 sfc_err(sa, "EtherType filters not supported");
1601 case RTE_ETH_FILTER_FLEXIBLE:
1602 sfc_err(sa, "Flexible filters not supported");
1604 case RTE_ETH_FILTER_SYN:
1605 sfc_err(sa, "SYN filters not supported");
1607 case RTE_ETH_FILTER_NTUPLE:
1608 sfc_err(sa, "NTUPLE filters not supported");
1610 case RTE_ETH_FILTER_TUNNEL:
1611 sfc_err(sa, "Tunnel filters not supported");
1613 case RTE_ETH_FILTER_FDIR:
1614 sfc_err(sa, "Flow Director filters not supported");
1616 case RTE_ETH_FILTER_HASH:
1617 sfc_err(sa, "Hash filters not supported");
1619 case RTE_ETH_FILTER_GENERIC:
1620 if (filter_op != RTE_ETH_FILTER_GET) {
1623 *(const void **)arg = &sfc_flow_ops;
1628 sfc_err(sa, "Unknown filter type %u", filter_type);
1632 sfc_log_init(sa, "exit: %d", -rc);
1633 SFC_ASSERT(rc >= 0);
1638 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1640 struct sfc_adapter *sa = dev->data->dev_private;
1643 * If Rx datapath does not provide callback to check mempool,
1644 * all pools are supported.
1646 if (sa->dp_rx->pool_ops_supported == NULL)
1649 return sa->dp_rx->pool_ops_supported(pool);
1652 static const struct eth_dev_ops sfc_eth_dev_ops = {
1653 .dev_configure = sfc_dev_configure,
1654 .dev_start = sfc_dev_start,
1655 .dev_stop = sfc_dev_stop,
1656 .dev_set_link_up = sfc_dev_set_link_up,
1657 .dev_set_link_down = sfc_dev_set_link_down,
1658 .dev_close = sfc_dev_close,
1659 .promiscuous_enable = sfc_dev_promisc_enable,
1660 .promiscuous_disable = sfc_dev_promisc_disable,
1661 .allmulticast_enable = sfc_dev_allmulti_enable,
1662 .allmulticast_disable = sfc_dev_allmulti_disable,
1663 .link_update = sfc_dev_link_update,
1664 .stats_get = sfc_stats_get,
1665 .stats_reset = sfc_stats_reset,
1666 .xstats_get = sfc_xstats_get,
1667 .xstats_reset = sfc_stats_reset,
1668 .xstats_get_names = sfc_xstats_get_names,
1669 .dev_infos_get = sfc_dev_infos_get,
1670 .dev_supported_ptypes_get = sfc_dev_supported_ptypes_get,
1671 .mtu_set = sfc_dev_set_mtu,
1672 .rx_queue_start = sfc_rx_queue_start,
1673 .rx_queue_stop = sfc_rx_queue_stop,
1674 .tx_queue_start = sfc_tx_queue_start,
1675 .tx_queue_stop = sfc_tx_queue_stop,
1676 .rx_queue_setup = sfc_rx_queue_setup,
1677 .rx_queue_release = sfc_rx_queue_release,
1678 .rx_queue_count = sfc_rx_queue_count,
1679 .rx_descriptor_done = sfc_rx_descriptor_done,
1680 .rx_descriptor_status = sfc_rx_descriptor_status,
1681 .tx_descriptor_status = sfc_tx_descriptor_status,
1682 .tx_queue_setup = sfc_tx_queue_setup,
1683 .tx_queue_release = sfc_tx_queue_release,
1684 .flow_ctrl_get = sfc_flow_ctrl_get,
1685 .flow_ctrl_set = sfc_flow_ctrl_set,
1686 .mac_addr_set = sfc_mac_addr_set,
1687 .udp_tunnel_port_add = sfc_dev_udp_tunnel_port_add,
1688 .udp_tunnel_port_del = sfc_dev_udp_tunnel_port_del,
1689 .reta_update = sfc_dev_rss_reta_update,
1690 .reta_query = sfc_dev_rss_reta_query,
1691 .rss_hash_update = sfc_dev_rss_hash_update,
1692 .rss_hash_conf_get = sfc_dev_rss_hash_conf_get,
1693 .filter_ctrl = sfc_dev_filter_ctrl,
1694 .set_mc_addr_list = sfc_set_mc_addr_list,
1695 .rxq_info_get = sfc_rx_queue_info_get,
1696 .txq_info_get = sfc_tx_queue_info_get,
1697 .fw_version_get = sfc_fw_version_get,
1698 .xstats_get_by_id = sfc_xstats_get_by_id,
1699 .xstats_get_names_by_id = sfc_xstats_get_names_by_id,
1700 .pool_ops_supported = sfc_pool_ops_supported,
1704 * Duplicate a string in potentially shared memory required for
1705 * multi-process support.
1707 * strdup() allocates from process-local heap/memory.
1710 sfc_strdup(const char *str)
1718 size = strlen(str) + 1;
1719 copy = rte_malloc(__func__, size, 0);
1721 rte_memcpy(copy, str, size);
1727 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1729 struct sfc_adapter *sa = dev->data->dev_private;
1730 const efx_nic_cfg_t *encp;
1731 unsigned int avail_caps = 0;
1732 const char *rx_name = NULL;
1733 const char *tx_name = NULL;
1736 switch (sa->family) {
1737 case EFX_FAMILY_HUNTINGTON:
1738 case EFX_FAMILY_MEDFORD:
1739 case EFX_FAMILY_MEDFORD2:
1740 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1746 encp = efx_nic_cfg_get(sa->nic);
1747 if (encp->enc_rx_es_super_buffer_supported)
1748 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1750 rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1751 sfc_kvarg_string_handler, &rx_name);
1753 goto fail_kvarg_rx_datapath;
1755 if (rx_name != NULL) {
1756 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1757 if (sa->dp_rx == NULL) {
1758 sfc_err(sa, "Rx datapath %s not found", rx_name);
1762 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1764 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1767 goto fail_dp_rx_caps;
1770 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1771 if (sa->dp_rx == NULL) {
1772 sfc_err(sa, "Rx datapath by caps %#x not found",
1779 sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1780 if (sa->dp_rx_name == NULL) {
1782 goto fail_dp_rx_name;
1785 sfc_notice(sa, "use %s Rx datapath", sa->dp_rx_name);
1787 dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1789 rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1790 sfc_kvarg_string_handler, &tx_name);
1792 goto fail_kvarg_tx_datapath;
1794 if (tx_name != NULL) {
1795 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1796 if (sa->dp_tx == NULL) {
1797 sfc_err(sa, "Tx datapath %s not found", tx_name);
1801 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1803 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1806 goto fail_dp_tx_caps;
1809 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1810 if (sa->dp_tx == NULL) {
1811 sfc_err(sa, "Tx datapath by caps %#x not found",
1818 sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1819 if (sa->dp_tx_name == NULL) {
1821 goto fail_dp_tx_name;
1824 sfc_notice(sa, "use %s Tx datapath", sa->dp_tx_name);
1826 dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1828 dev->dev_ops = &sfc_eth_dev_ops;
1837 fail_kvarg_tx_datapath:
1838 rte_free(sa->dp_rx_name);
1839 sa->dp_rx_name = NULL;
1846 fail_kvarg_rx_datapath:
1851 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1853 struct sfc_adapter *sa = dev->data->dev_private;
1855 dev->dev_ops = NULL;
1856 dev->rx_pkt_burst = NULL;
1857 dev->tx_pkt_burst = NULL;
1859 rte_free(sa->dp_tx_name);
1860 sa->dp_tx_name = NULL;
1863 rte_free(sa->dp_rx_name);
1864 sa->dp_rx_name = NULL;
1868 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1869 .rxq_info_get = sfc_rx_queue_info_get,
1870 .txq_info_get = sfc_tx_queue_info_get,
1874 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev, uint32_t logtype_main)
1877 * Device private data has really many process-local pointers.
1878 * Below code should be extremely careful to use data located
1879 * in shared memory only.
1881 struct sfc_adapter *sa = dev->data->dev_private;
1882 const struct sfc_dp_rx *dp_rx;
1883 const struct sfc_dp_tx *dp_tx;
1886 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1887 if (dp_rx == NULL) {
1888 SFC_LOG(sa, RTE_LOG_ERR, logtype_main,
1889 "cannot find %s Rx datapath", sa->dp_rx_name);
1893 if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1894 SFC_LOG(sa, RTE_LOG_ERR, logtype_main,
1895 "%s Rx datapath does not support multi-process",
1898 goto fail_dp_rx_multi_process;
1901 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1902 if (dp_tx == NULL) {
1903 SFC_LOG(sa, RTE_LOG_ERR, logtype_main,
1904 "cannot find %s Tx datapath", sa->dp_tx_name);
1908 if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1909 SFC_LOG(sa, RTE_LOG_ERR, logtype_main,
1910 "%s Tx datapath does not support multi-process",
1913 goto fail_dp_tx_multi_process;
1916 dev->rx_pkt_burst = dp_rx->pkt_burst;
1917 dev->tx_pkt_burst = dp_tx->pkt_burst;
1918 dev->dev_ops = &sfc_eth_dev_secondary_ops;
1922 fail_dp_tx_multi_process:
1924 fail_dp_rx_multi_process:
1930 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1932 dev->dev_ops = NULL;
1933 dev->tx_pkt_burst = NULL;
1934 dev->rx_pkt_burst = NULL;
1938 sfc_register_dp(void)
1941 if (TAILQ_EMPTY(&sfc_dp_head)) {
1942 /* Prefer EF10 datapath */
1943 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
1944 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1945 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1947 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1948 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1949 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1954 sfc_eth_dev_init(struct rte_eth_dev *dev)
1956 struct sfc_adapter *sa = dev->data->dev_private;
1957 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1958 uint32_t logtype_main;
1960 const efx_nic_cfg_t *encp;
1961 const struct ether_addr *from;
1965 logtype_main = sfc_register_logtype(&pci_dev->addr,
1966 SFC_LOGTYPE_MAIN_STR,
1969 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1970 return -sfc_eth_dev_secondary_set_ops(dev, logtype_main);
1972 /* Required for logging */
1973 sa->pci_addr = pci_dev->addr;
1974 sa->port_id = dev->data->port_id;
1975 sa->logtype_main = logtype_main;
1979 /* Copy PCI device info to the dev->data */
1980 rte_eth_copy_pci_info(dev, pci_dev);
1982 rc = sfc_kvargs_parse(sa);
1984 goto fail_kvargs_parse;
1986 sfc_log_init(sa, "entry");
1988 dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1989 if (dev->data->mac_addrs == NULL) {
1991 goto fail_mac_addrs;
1994 sfc_adapter_lock_init(sa);
1995 sfc_adapter_lock(sa);
1997 sfc_log_init(sa, "probing");
2002 sfc_log_init(sa, "set device ops");
2003 rc = sfc_eth_dev_set_ops(dev);
2007 sfc_log_init(sa, "attaching");
2008 rc = sfc_attach(sa);
2012 encp = efx_nic_cfg_get(sa->nic);
2015 * The arguments are really reverse order in comparison to
2016 * Linux kernel. Copy from NIC config to Ethernet device data.
2018 from = (const struct ether_addr *)(encp->enc_mac_addr);
2019 ether_addr_copy(from, &dev->data->mac_addrs[0]);
2021 sfc_adapter_unlock(sa);
2023 sfc_log_init(sa, "done");
2027 sfc_eth_dev_clear_ops(dev);
2033 sfc_adapter_unlock(sa);
2034 sfc_adapter_lock_fini(sa);
2035 rte_free(dev->data->mac_addrs);
2036 dev->data->mac_addrs = NULL;
2039 sfc_kvargs_cleanup(sa);
2042 sfc_log_init(sa, "failed %d", rc);
2048 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2050 struct sfc_adapter *sa;
2052 if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2053 sfc_eth_dev_secondary_clear_ops(dev);
2057 sa = dev->data->dev_private;
2058 sfc_log_init(sa, "entry");
2060 sfc_adapter_lock(sa);
2062 sfc_eth_dev_clear_ops(dev);
2067 sfc_kvargs_cleanup(sa);
2069 sfc_adapter_unlock(sa);
2070 sfc_adapter_lock_fini(sa);
2072 sfc_log_init(sa, "done");
2074 /* Required for logging, so cleanup last */
2079 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2080 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2081 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2082 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2083 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2084 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2085 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2086 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2087 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2088 { .vendor_id = 0 /* sentinel */ }
2091 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2092 struct rte_pci_device *pci_dev)
2094 return rte_eth_dev_pci_generic_probe(pci_dev,
2095 sizeof(struct sfc_adapter), sfc_eth_dev_init);
2098 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2100 return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2103 static struct rte_pci_driver sfc_efx_pmd = {
2104 .id_table = pci_id_sfc_efx_map,
2106 RTE_PCI_DRV_INTR_LSC |
2107 RTE_PCI_DRV_NEED_MAPPING,
2108 .probe = sfc_eth_dev_pci_probe,
2109 .remove = sfc_eth_dev_pci_remove,
2112 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2113 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2114 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2115 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2116 SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2117 SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2118 SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2119 SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2120 SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2121 SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2123 RTE_INIT(sfc_driver_register_logtype)
2127 ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2129 sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;