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 const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
88 struct sfc_adapter *sa = dev->data->dev_private;
89 struct sfc_rss *rss = &sa->rss;
90 uint64_t txq_offloads_def = 0;
92 sfc_log_init(sa, "entry");
94 dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
96 /* Autonegotiation may be disabled */
97 dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
98 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
99 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
100 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
101 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
102 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_25000FDX)
103 dev_info->speed_capa |= ETH_LINK_SPEED_25G;
104 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
105 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
106 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_50000FDX)
107 dev_info->speed_capa |= ETH_LINK_SPEED_50G;
108 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_100000FDX)
109 dev_info->speed_capa |= ETH_LINK_SPEED_100G;
111 dev_info->max_rx_queues = sa->rxq_max;
112 dev_info->max_tx_queues = sa->txq_max;
114 /* By default packets are dropped if no descriptors are available */
115 dev_info->default_rxconf.rx_drop_en = 1;
117 dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
120 * rx_offload_capa includes both device and queue offloads since
121 * the latter may be requested on a per device basis which makes
122 * sense when some offloads are needed to be set on all queues.
124 dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
125 dev_info->rx_queue_offload_capa;
127 dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
130 * tx_offload_capa includes both device and queue offloads since
131 * the latter may be requested on a per device basis which makes
132 * sense when some offloads are needed to be set on all queues.
134 dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
135 dev_info->tx_queue_offload_capa;
137 if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
138 txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
140 dev_info->default_txconf.offloads |= txq_offloads_def;
142 if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
146 for (i = 0; i < rss->hf_map_nb_entries; ++i)
147 rte_hf |= rss->hf_map[i].rte;
149 dev_info->reta_size = EFX_RSS_TBL_SIZE;
150 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
151 dev_info->flow_type_rss_offloads = rte_hf;
154 /* Initialize to hardware limits */
155 dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
156 dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
157 /* The RXQ hardware requires that the descriptor count is a power
158 * of 2, but rx_desc_lim cannot properly describe that constraint.
160 dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
162 /* Initialize to hardware limits */
163 dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
164 dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
166 * The TXQ hardware requires that the descriptor count is a power
167 * of 2, but tx_desc_lim cannot properly describe that constraint
169 dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
171 if (sap->dp_rx->get_dev_info != NULL)
172 sap->dp_rx->get_dev_info(dev_info);
173 if (sap->dp_tx->get_dev_info != NULL)
174 sap->dp_tx->get_dev_info(dev_info);
176 dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
177 RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
180 static const uint32_t *
181 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
183 const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
184 struct sfc_adapter *sa = dev->data->dev_private;
185 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
186 uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
188 return sap->dp_rx->supported_ptypes_get(tunnel_encaps);
192 sfc_dev_configure(struct rte_eth_dev *dev)
194 struct rte_eth_dev_data *dev_data = dev->data;
195 struct sfc_adapter *sa = dev_data->dev_private;
198 sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
199 dev_data->nb_rx_queues, dev_data->nb_tx_queues);
201 sfc_adapter_lock(sa);
203 case SFC_ADAPTER_CONFIGURED:
205 case SFC_ADAPTER_INITIALIZED:
206 rc = sfc_configure(sa);
209 sfc_err(sa, "unexpected adapter state %u to configure",
214 sfc_adapter_unlock(sa);
216 sfc_log_init(sa, "done %d", rc);
222 sfc_dev_start(struct rte_eth_dev *dev)
224 struct sfc_adapter *sa = dev->data->dev_private;
227 sfc_log_init(sa, "entry");
229 sfc_adapter_lock(sa);
231 sfc_adapter_unlock(sa);
233 sfc_log_init(sa, "done %d", rc);
239 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
241 struct sfc_adapter *sa = dev->data->dev_private;
242 struct rte_eth_link current_link;
245 sfc_log_init(sa, "entry");
247 if (sa->state != SFC_ADAPTER_STARTED) {
248 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, ¤t_link);
249 } else if (wait_to_complete) {
250 efx_link_mode_t link_mode;
252 if (efx_port_poll(sa->nic, &link_mode) != 0)
253 link_mode = EFX_LINK_UNKNOWN;
254 sfc_port_link_mode_to_info(link_mode, ¤t_link);
257 sfc_ev_mgmt_qpoll(sa);
258 rte_eth_linkstatus_get(dev, ¤t_link);
261 ret = rte_eth_linkstatus_set(dev, ¤t_link);
263 sfc_notice(sa, "Link status is %s",
264 current_link.link_status ? "UP" : "DOWN");
270 sfc_dev_stop(struct rte_eth_dev *dev)
272 struct sfc_adapter *sa = dev->data->dev_private;
274 sfc_log_init(sa, "entry");
276 sfc_adapter_lock(sa);
278 sfc_adapter_unlock(sa);
280 sfc_log_init(sa, "done");
284 sfc_dev_set_link_up(struct rte_eth_dev *dev)
286 struct sfc_adapter *sa = dev->data->dev_private;
289 sfc_log_init(sa, "entry");
291 sfc_adapter_lock(sa);
293 sfc_adapter_unlock(sa);
300 sfc_dev_set_link_down(struct rte_eth_dev *dev)
302 struct sfc_adapter *sa = dev->data->dev_private;
304 sfc_log_init(sa, "entry");
306 sfc_adapter_lock(sa);
308 sfc_adapter_unlock(sa);
314 sfc_dev_close(struct rte_eth_dev *dev)
316 struct sfc_adapter *sa = dev->data->dev_private;
318 sfc_log_init(sa, "entry");
320 sfc_adapter_lock(sa);
322 case SFC_ADAPTER_STARTED:
324 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
326 case SFC_ADAPTER_CONFIGURED:
328 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
330 case SFC_ADAPTER_INITIALIZED:
333 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
336 sfc_adapter_unlock(sa);
338 sfc_log_init(sa, "done");
342 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
345 struct sfc_port *port;
347 struct sfc_adapter *sa = dev->data->dev_private;
348 boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
349 const char *desc = (allmulti) ? "all-multi" : "promiscuous";
351 sfc_adapter_lock(sa);
354 toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
356 if (*toggle != enabled) {
359 if (port->isolated) {
360 sfc_warn(sa, "isolated mode is active on the port");
361 sfc_warn(sa, "the change is to be applied on the next "
362 "start provided that isolated mode is "
363 "disabled prior the next start");
364 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
365 (sfc_set_rx_mode(sa) != 0)) {
366 *toggle = !(enabled);
367 sfc_warn(sa, "Failed to %s %s mode",
368 ((enabled) ? "enable" : "disable"), desc);
372 sfc_adapter_unlock(sa);
376 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
378 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
382 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
384 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
388 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
390 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
394 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
396 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
400 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
401 uint16_t nb_rx_desc, unsigned int socket_id,
402 const struct rte_eth_rxconf *rx_conf,
403 struct rte_mempool *mb_pool)
405 struct sfc_adapter *sa = dev->data->dev_private;
408 sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
409 rx_queue_id, nb_rx_desc, socket_id);
411 sfc_adapter_lock(sa);
413 rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
418 dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].dp;
420 sfc_adapter_unlock(sa);
425 sfc_adapter_unlock(sa);
431 sfc_rx_queue_release(void *queue)
433 struct sfc_dp_rxq *dp_rxq = queue;
435 struct sfc_adapter *sa;
436 unsigned int sw_index;
441 rxq = sfc_rxq_by_dp_rxq(dp_rxq);
443 sfc_adapter_lock(sa);
445 sw_index = sfc_rxq_sw_index(rxq);
447 sfc_log_init(sa, "RxQ=%u", sw_index);
449 sfc_rx_qfini(sa, sw_index);
451 sfc_adapter_unlock(sa);
455 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
456 uint16_t nb_tx_desc, unsigned int socket_id,
457 const struct rte_eth_txconf *tx_conf)
459 struct sfc_adapter *sa = dev->data->dev_private;
462 sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
463 tx_queue_id, nb_tx_desc, socket_id);
465 sfc_adapter_lock(sa);
467 rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
471 dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
473 sfc_adapter_unlock(sa);
477 sfc_adapter_unlock(sa);
483 sfc_tx_queue_release(void *queue)
485 struct sfc_dp_txq *dp_txq = queue;
487 unsigned int sw_index;
488 struct sfc_adapter *sa;
493 txq = sfc_txq_by_dp_txq(dp_txq);
494 sw_index = sfc_txq_sw_index(txq);
496 SFC_ASSERT(txq->evq != NULL);
499 sfc_log_init(sa, "TxQ = %u", sw_index);
501 sfc_adapter_lock(sa);
503 sfc_tx_qfini(sa, sw_index);
505 sfc_adapter_unlock(sa);
509 * Some statistics are computed as A - B where A and B each increase
510 * monotonically with some hardware counter(s) and the counters are read
513 * If packet X is counted in A, but not counted in B yet, computed value is
516 * If packet X is not counted in A at the moment of reading the counter,
517 * but counted in B at the moment of reading the counter, computed value
520 * However, counter which grows backward is worse evil than slightly wrong
521 * value. So, let's try to guarantee that it never happens except may be
522 * the case when the MAC stats are zeroed as a result of a NIC reset.
525 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
527 if ((int64_t)(newval - *stat) > 0 || newval == 0)
532 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
534 struct sfc_adapter *sa = dev->data->dev_private;
535 struct sfc_port *port = &sa->port;
539 rte_spinlock_lock(&port->mac_stats_lock);
541 ret = sfc_port_update_mac_stats(sa);
545 mac_stats = port->mac_stats_buf;
547 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
548 EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
550 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
551 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
552 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
554 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
555 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
556 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
558 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
559 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
560 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
562 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
563 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
564 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
565 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
566 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
568 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
569 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
570 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
572 * Take into account stats which are whenever supported
573 * on EF10. If some stat is not supported by current
574 * firmware variant or HW revision, it is guaranteed
575 * to be zero in mac_stats.
578 mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
579 mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
580 mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
581 mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
582 mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
583 mac_stats[EFX_MAC_PM_TRUNC_QBB] +
584 mac_stats[EFX_MAC_PM_DISCARD_QBB] +
585 mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
586 mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
587 mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
589 mac_stats[EFX_MAC_RX_FCS_ERRORS] +
590 mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
591 mac_stats[EFX_MAC_RX_JABBER_PKTS];
592 /* no oerrors counters supported on EF10 */
594 /* Exclude missed, errors and pauses from Rx packets */
595 sfc_update_diff_stat(&port->ipackets,
596 mac_stats[EFX_MAC_RX_PKTS] -
597 mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
598 stats->imissed - stats->ierrors);
599 stats->ipackets = port->ipackets;
603 rte_spinlock_unlock(&port->mac_stats_lock);
604 SFC_ASSERT(ret >= 0);
609 sfc_stats_reset(struct rte_eth_dev *dev)
611 struct sfc_adapter *sa = dev->data->dev_private;
612 struct sfc_port *port = &sa->port;
615 if (sa->state != SFC_ADAPTER_STARTED) {
617 * The operation cannot be done if port is not started; it
618 * will be scheduled to be done during the next port start
620 port->mac_stats_reset_pending = B_TRUE;
624 rc = sfc_port_reset_mac_stats(sa);
626 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
630 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
631 unsigned int xstats_count)
633 struct sfc_adapter *sa = dev->data->dev_private;
634 struct sfc_port *port = &sa->port;
640 rte_spinlock_lock(&port->mac_stats_lock);
642 rc = sfc_port_update_mac_stats(sa);
649 mac_stats = port->mac_stats_buf;
651 for (i = 0; i < EFX_MAC_NSTATS; ++i) {
652 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
653 if (xstats != NULL && nstats < (int)xstats_count) {
654 xstats[nstats].id = nstats;
655 xstats[nstats].value = mac_stats[i];
662 rte_spinlock_unlock(&port->mac_stats_lock);
668 sfc_xstats_get_names(struct rte_eth_dev *dev,
669 struct rte_eth_xstat_name *xstats_names,
670 unsigned int xstats_count)
672 struct sfc_adapter *sa = dev->data->dev_private;
673 struct sfc_port *port = &sa->port;
675 unsigned int nstats = 0;
677 for (i = 0; i < EFX_MAC_NSTATS; ++i) {
678 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
679 if (xstats_names != NULL && nstats < xstats_count)
680 strlcpy(xstats_names[nstats].name,
681 efx_mac_stat_name(sa->nic, i),
682 sizeof(xstats_names[0].name));
691 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
692 uint64_t *values, unsigned int n)
694 struct sfc_adapter *sa = dev->data->dev_private;
695 struct sfc_port *port = &sa->port;
697 unsigned int nb_supported = 0;
698 unsigned int nb_written = 0;
703 if (unlikely(values == NULL) ||
704 unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
705 return port->mac_stats_nb_supported;
707 rte_spinlock_lock(&port->mac_stats_lock);
709 rc = sfc_port_update_mac_stats(sa);
716 mac_stats = port->mac_stats_buf;
718 for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
719 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
722 if ((ids == NULL) || (ids[nb_written] == nb_supported))
723 values[nb_written++] = mac_stats[i];
731 rte_spinlock_unlock(&port->mac_stats_lock);
737 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
738 struct rte_eth_xstat_name *xstats_names,
739 const uint64_t *ids, unsigned int size)
741 struct sfc_adapter *sa = dev->data->dev_private;
742 struct sfc_port *port = &sa->port;
743 unsigned int nb_supported = 0;
744 unsigned int nb_written = 0;
747 if (unlikely(xstats_names == NULL) ||
748 unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
749 return port->mac_stats_nb_supported;
751 for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
752 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
755 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
756 char *name = xstats_names[nb_written++].name;
758 strlcpy(name, efx_mac_stat_name(sa->nic, i),
759 sizeof(xstats_names[0].name));
769 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
771 struct sfc_adapter *sa = dev->data->dev_private;
772 unsigned int wanted_fc, link_fc;
774 memset(fc_conf, 0, sizeof(*fc_conf));
776 sfc_adapter_lock(sa);
778 if (sa->state == SFC_ADAPTER_STARTED)
779 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
781 link_fc = sa->port.flow_ctrl;
785 fc_conf->mode = RTE_FC_NONE;
787 case EFX_FCNTL_RESPOND:
788 fc_conf->mode = RTE_FC_RX_PAUSE;
790 case EFX_FCNTL_GENERATE:
791 fc_conf->mode = RTE_FC_TX_PAUSE;
793 case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
794 fc_conf->mode = RTE_FC_FULL;
797 sfc_err(sa, "%s: unexpected flow control value %#x",
801 fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
803 sfc_adapter_unlock(sa);
809 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
811 struct sfc_adapter *sa = dev->data->dev_private;
812 struct sfc_port *port = &sa->port;
816 if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
817 fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
818 fc_conf->mac_ctrl_frame_fwd != 0) {
819 sfc_err(sa, "unsupported flow control settings specified");
824 switch (fc_conf->mode) {
828 case RTE_FC_RX_PAUSE:
829 fcntl = EFX_FCNTL_RESPOND;
831 case RTE_FC_TX_PAUSE:
832 fcntl = EFX_FCNTL_GENERATE;
835 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
842 sfc_adapter_lock(sa);
844 if (sa->state == SFC_ADAPTER_STARTED) {
845 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
847 goto fail_mac_fcntl_set;
850 port->flow_ctrl = fcntl;
851 port->flow_ctrl_autoneg = fc_conf->autoneg;
853 sfc_adapter_unlock(sa);
858 sfc_adapter_unlock(sa);
865 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
867 struct sfc_adapter *sa = dev->data->dev_private;
868 size_t pdu = EFX_MAC_PDU(mtu);
872 sfc_log_init(sa, "mtu=%u", mtu);
875 if (pdu < EFX_MAC_PDU_MIN) {
876 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
877 (unsigned int)mtu, (unsigned int)pdu,
881 if (pdu > EFX_MAC_PDU_MAX) {
882 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
883 (unsigned int)mtu, (unsigned int)pdu,
888 sfc_adapter_lock(sa);
890 if (pdu != sa->port.pdu) {
891 if (sa->state == SFC_ADAPTER_STARTED) {
894 old_pdu = sa->port.pdu;
905 * The driver does not use it, but other PMDs update jumbo frame
906 * flag and max_rx_pkt_len when MTU is set.
908 if (mtu > ETHER_MAX_LEN) {
909 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
910 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
913 dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
915 sfc_adapter_unlock(sa);
917 sfc_log_init(sa, "done");
921 sa->port.pdu = old_pdu;
922 if (sfc_start(sa) != 0)
923 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
924 "PDU max size - port is stopped",
925 (unsigned int)pdu, (unsigned int)old_pdu);
926 sfc_adapter_unlock(sa);
929 sfc_log_init(sa, "failed %d", rc);
934 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
936 struct sfc_adapter *sa = dev->data->dev_private;
937 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
938 struct sfc_port *port = &sa->port;
939 struct ether_addr *old_addr = &dev->data->mac_addrs[0];
942 sfc_adapter_lock(sa);
945 * Copy the address to the device private data so that
946 * it could be recalled in the case of adapter restart.
948 ether_addr_copy(mac_addr, &port->default_mac_addr);
951 * Neither of the two following checks can return
952 * an error. The new MAC address is preserved in
953 * the device private data and can be activated
954 * on the next port start if the user prevents
955 * isolated mode from being enabled.
957 if (port->isolated) {
958 sfc_warn(sa, "isolated mode is active on the port");
959 sfc_warn(sa, "will not set MAC address");
963 if (sa->state != SFC_ADAPTER_STARTED) {
964 sfc_notice(sa, "the port is not started");
965 sfc_notice(sa, "the new MAC address will be set on port start");
970 if (encp->enc_allow_set_mac_with_installed_filters) {
971 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
973 sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
978 * Changing the MAC address by means of MCDI request
979 * has no effect on received traffic, therefore
980 * we also need to update unicast filters
982 rc = sfc_set_rx_mode(sa);
984 sfc_err(sa, "cannot set filter (rc = %u)", rc);
985 /* Rollback the old address */
986 (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
987 (void)sfc_set_rx_mode(sa);
990 sfc_warn(sa, "cannot set MAC address with filters installed");
991 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
992 sfc_warn(sa, "(some traffic may be dropped)");
995 * Since setting MAC address with filters installed is not
996 * allowed on the adapter, the new MAC address will be set
997 * by means of adapter restart. sfc_start() shall retrieve
998 * the new address from the device private data and set it.
1003 sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1008 ether_addr_copy(old_addr, &port->default_mac_addr);
1010 sfc_adapter_unlock(sa);
1012 SFC_ASSERT(rc >= 0);
1018 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
1019 uint32_t nb_mc_addr)
1021 struct sfc_adapter *sa = dev->data->dev_private;
1022 struct sfc_port *port = &sa->port;
1023 uint8_t *mc_addrs = port->mcast_addrs;
1027 if (port->isolated) {
1028 sfc_err(sa, "isolated mode is active on the port");
1029 sfc_err(sa, "will not set multicast address list");
1033 if (mc_addrs == NULL)
1036 if (nb_mc_addr > port->max_mcast_addrs) {
1037 sfc_err(sa, "too many multicast addresses: %u > %u",
1038 nb_mc_addr, port->max_mcast_addrs);
1042 for (i = 0; i < nb_mc_addr; ++i) {
1043 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1045 mc_addrs += EFX_MAC_ADDR_LEN;
1048 port->nb_mcast_addrs = nb_mc_addr;
1050 if (sa->state != SFC_ADAPTER_STARTED)
1053 rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1054 port->nb_mcast_addrs);
1056 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1058 SFC_ASSERT(rc >= 0);
1063 * The function is used by the secondary process as well. It must not
1064 * use any process-local pointers from the adapter data.
1067 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1068 struct rte_eth_rxq_info *qinfo)
1070 struct sfc_adapter *sa = dev->data->dev_private;
1071 struct sfc_rxq_info *rxq_info;
1073 sfc_adapter_lock(sa);
1075 SFC_ASSERT(rx_queue_id < sa->rxq_count);
1077 rxq_info = &sa->rxq_info[rx_queue_id];
1079 qinfo->mp = rxq_info->refill_mb_pool;
1080 qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1081 qinfo->conf.rx_drop_en = 1;
1082 qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1083 qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1084 if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1085 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1086 qinfo->scattered_rx = 1;
1088 qinfo->nb_desc = rxq_info->entries;
1090 sfc_adapter_unlock(sa);
1094 * The function is used by the secondary process as well. It must not
1095 * use any process-local pointers from the adapter data.
1098 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1099 struct rte_eth_txq_info *qinfo)
1101 struct sfc_adapter *sa = dev->data->dev_private;
1102 struct sfc_txq_info *txq_info;
1104 sfc_adapter_lock(sa);
1106 SFC_ASSERT(tx_queue_id < sa->txq_count);
1108 txq_info = &sa->txq_info[tx_queue_id];
1110 memset(qinfo, 0, sizeof(*qinfo));
1112 qinfo->conf.offloads = txq_info->offloads;
1113 qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1114 qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1115 qinfo->nb_desc = txq_info->entries;
1117 sfc_adapter_unlock(sa);
1121 * The function is used by the secondary process as well. It must not
1122 * use any process-local pointers from the adapter data.
1125 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1127 const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1128 struct sfc_adapter *sa = dev->data->dev_private;
1129 struct sfc_rxq_info *rxq_info;
1131 SFC_ASSERT(rx_queue_id < sa->rxq_count);
1132 rxq_info = &sa->rxq_info[rx_queue_id];
1134 if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1137 return sap->dp_rx->qdesc_npending(rxq_info->dp);
1141 * The function is used by the secondary process as well. It must not
1142 * use any process-local pointers from the adapter data.
1145 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1147 struct sfc_dp_rxq *dp_rxq = queue;
1148 const struct sfc_dp_rx *dp_rx;
1150 dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1152 return offset < dp_rx->qdesc_npending(dp_rxq);
1156 * The function is used by the secondary process as well. It must not
1157 * use any process-local pointers from the adapter data.
1160 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1162 struct sfc_dp_rxq *dp_rxq = queue;
1163 const struct sfc_dp_rx *dp_rx;
1165 dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1167 return dp_rx->qdesc_status(dp_rxq, offset);
1171 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1173 struct sfc_dp_txq *dp_txq = queue;
1174 struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1176 return txq->evq->sa->priv.dp_tx->qdesc_status(dp_txq, offset);
1180 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1182 struct sfc_adapter *sa = dev->data->dev_private;
1185 sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1187 sfc_adapter_lock(sa);
1190 if (sa->state != SFC_ADAPTER_STARTED)
1191 goto fail_not_started;
1193 if (sa->rxq_info[rx_queue_id].rxq == NULL)
1194 goto fail_not_setup;
1196 rc = sfc_rx_qstart(sa, rx_queue_id);
1198 goto fail_rx_qstart;
1200 sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1202 sfc_adapter_unlock(sa);
1209 sfc_adapter_unlock(sa);
1215 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1217 struct sfc_adapter *sa = dev->data->dev_private;
1219 sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1221 sfc_adapter_lock(sa);
1222 sfc_rx_qstop(sa, rx_queue_id);
1224 sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1226 sfc_adapter_unlock(sa);
1232 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1234 struct sfc_adapter *sa = dev->data->dev_private;
1237 sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1239 sfc_adapter_lock(sa);
1242 if (sa->state != SFC_ADAPTER_STARTED)
1243 goto fail_not_started;
1245 if (sa->txq_info[tx_queue_id].txq == NULL)
1246 goto fail_not_setup;
1248 rc = sfc_tx_qstart(sa, tx_queue_id);
1250 goto fail_tx_qstart;
1252 sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1254 sfc_adapter_unlock(sa);
1261 sfc_adapter_unlock(sa);
1267 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1269 struct sfc_adapter *sa = dev->data->dev_private;
1271 sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1273 sfc_adapter_lock(sa);
1275 sfc_tx_qstop(sa, tx_queue_id);
1277 sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1279 sfc_adapter_unlock(sa);
1283 static efx_tunnel_protocol_t
1284 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1287 case RTE_TUNNEL_TYPE_VXLAN:
1288 return EFX_TUNNEL_PROTOCOL_VXLAN;
1289 case RTE_TUNNEL_TYPE_GENEVE:
1290 return EFX_TUNNEL_PROTOCOL_GENEVE;
1292 return EFX_TUNNEL_NPROTOS;
1296 enum sfc_udp_tunnel_op_e {
1297 SFC_UDP_TUNNEL_ADD_PORT,
1298 SFC_UDP_TUNNEL_DEL_PORT,
1302 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1303 struct rte_eth_udp_tunnel *tunnel_udp,
1304 enum sfc_udp_tunnel_op_e op)
1306 struct sfc_adapter *sa = dev->data->dev_private;
1307 efx_tunnel_protocol_t tunnel_proto;
1310 sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1311 (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1312 (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1313 tunnel_udp->udp_port, tunnel_udp->prot_type);
1316 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1317 if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1319 goto fail_bad_proto;
1322 sfc_adapter_lock(sa);
1325 case SFC_UDP_TUNNEL_ADD_PORT:
1326 rc = efx_tunnel_config_udp_add(sa->nic,
1327 tunnel_udp->udp_port,
1330 case SFC_UDP_TUNNEL_DEL_PORT:
1331 rc = efx_tunnel_config_udp_remove(sa->nic,
1332 tunnel_udp->udp_port,
1343 if (sa->state == SFC_ADAPTER_STARTED) {
1344 rc = efx_tunnel_reconfigure(sa->nic);
1347 * Configuration is accepted by FW and MC reboot
1348 * is initiated to apply the changes. MC reboot
1349 * will be handled in a usual way (MC reboot
1350 * event on management event queue and adapter
1354 } else if (rc != 0) {
1355 goto fail_reconfigure;
1359 sfc_adapter_unlock(sa);
1363 /* Remove/restore entry since the change makes the trouble */
1365 case SFC_UDP_TUNNEL_ADD_PORT:
1366 (void)efx_tunnel_config_udp_remove(sa->nic,
1367 tunnel_udp->udp_port,
1370 case SFC_UDP_TUNNEL_DEL_PORT:
1371 (void)efx_tunnel_config_udp_add(sa->nic,
1372 tunnel_udp->udp_port,
1379 sfc_adapter_unlock(sa);
1387 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1388 struct rte_eth_udp_tunnel *tunnel_udp)
1390 return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1394 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1395 struct rte_eth_udp_tunnel *tunnel_udp)
1397 return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1401 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1402 struct rte_eth_rss_conf *rss_conf)
1404 struct sfc_adapter *sa = dev->data->dev_private;
1405 struct sfc_rss *rss = &sa->rss;
1407 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1410 sfc_adapter_lock(sa);
1413 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1414 * hence, conversion is done here to derive a correct set of ETH_RSS
1415 * flags which corresponds to the active EFX configuration stored
1416 * locally in 'sfc_adapter' and kept up-to-date
1418 rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(sa, rss->hash_types);
1419 rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1420 if (rss_conf->rss_key != NULL)
1421 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1423 sfc_adapter_unlock(sa);
1429 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1430 struct rte_eth_rss_conf *rss_conf)
1432 struct sfc_adapter *sa = dev->data->dev_private;
1433 struct sfc_rss *rss = &sa->rss;
1434 struct sfc_port *port = &sa->port;
1435 unsigned int efx_hash_types;
1441 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1442 sfc_err(sa, "RSS is not available");
1446 if (rss->channels == 0) {
1447 sfc_err(sa, "RSS is not configured");
1451 if ((rss_conf->rss_key != NULL) &&
1452 (rss_conf->rss_key_len != sizeof(rss->key))) {
1453 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1458 sfc_adapter_lock(sa);
1460 rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1462 goto fail_rx_hf_rte_to_efx;
1464 rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1465 rss->hash_alg, efx_hash_types, B_TRUE);
1467 goto fail_scale_mode_set;
1469 if (rss_conf->rss_key != NULL) {
1470 if (sa->state == SFC_ADAPTER_STARTED) {
1471 rc = efx_rx_scale_key_set(sa->nic,
1472 EFX_RSS_CONTEXT_DEFAULT,
1476 goto fail_scale_key_set;
1479 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1482 rss->hash_types = efx_hash_types;
1484 sfc_adapter_unlock(sa);
1489 if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1490 EFX_RX_HASHALG_TOEPLITZ,
1491 rss->hash_types, B_TRUE) != 0)
1492 sfc_err(sa, "failed to restore RSS mode");
1494 fail_scale_mode_set:
1495 fail_rx_hf_rte_to_efx:
1496 sfc_adapter_unlock(sa);
1501 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1502 struct rte_eth_rss_reta_entry64 *reta_conf,
1505 struct sfc_adapter *sa = dev->data->dev_private;
1506 struct sfc_rss *rss = &sa->rss;
1507 struct sfc_port *port = &sa->port;
1510 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1513 if (rss->channels == 0)
1516 if (reta_size != EFX_RSS_TBL_SIZE)
1519 sfc_adapter_lock(sa);
1521 for (entry = 0; entry < reta_size; entry++) {
1522 int grp = entry / RTE_RETA_GROUP_SIZE;
1523 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1525 if ((reta_conf[grp].mask >> grp_idx) & 1)
1526 reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1529 sfc_adapter_unlock(sa);
1535 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1536 struct rte_eth_rss_reta_entry64 *reta_conf,
1539 struct sfc_adapter *sa = dev->data->dev_private;
1540 struct sfc_rss *rss = &sa->rss;
1541 struct sfc_port *port = &sa->port;
1542 unsigned int *rss_tbl_new;
1550 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1551 sfc_err(sa, "RSS is not available");
1555 if (rss->channels == 0) {
1556 sfc_err(sa, "RSS is not configured");
1560 if (reta_size != EFX_RSS_TBL_SIZE) {
1561 sfc_err(sa, "RETA size is wrong (should be %u)",
1566 rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1567 if (rss_tbl_new == NULL)
1570 sfc_adapter_lock(sa);
1572 rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1574 for (entry = 0; entry < reta_size; entry++) {
1575 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1576 struct rte_eth_rss_reta_entry64 *grp;
1578 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1580 if (grp->mask & (1ull << grp_idx)) {
1581 if (grp->reta[grp_idx] >= rss->channels) {
1583 goto bad_reta_entry;
1585 rss_tbl_new[entry] = grp->reta[grp_idx];
1589 if (sa->state == SFC_ADAPTER_STARTED) {
1590 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1591 rss_tbl_new, EFX_RSS_TBL_SIZE);
1593 goto fail_scale_tbl_set;
1596 rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1600 sfc_adapter_unlock(sa);
1602 rte_free(rss_tbl_new);
1604 SFC_ASSERT(rc >= 0);
1609 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1610 enum rte_filter_op filter_op,
1613 struct sfc_adapter *sa = dev->data->dev_private;
1616 sfc_log_init(sa, "entry");
1618 switch (filter_type) {
1619 case RTE_ETH_FILTER_NONE:
1620 sfc_err(sa, "Global filters configuration not supported");
1622 case RTE_ETH_FILTER_MACVLAN:
1623 sfc_err(sa, "MACVLAN filters not supported");
1625 case RTE_ETH_FILTER_ETHERTYPE:
1626 sfc_err(sa, "EtherType filters not supported");
1628 case RTE_ETH_FILTER_FLEXIBLE:
1629 sfc_err(sa, "Flexible filters not supported");
1631 case RTE_ETH_FILTER_SYN:
1632 sfc_err(sa, "SYN filters not supported");
1634 case RTE_ETH_FILTER_NTUPLE:
1635 sfc_err(sa, "NTUPLE filters not supported");
1637 case RTE_ETH_FILTER_TUNNEL:
1638 sfc_err(sa, "Tunnel filters not supported");
1640 case RTE_ETH_FILTER_FDIR:
1641 sfc_err(sa, "Flow Director filters not supported");
1643 case RTE_ETH_FILTER_HASH:
1644 sfc_err(sa, "Hash filters not supported");
1646 case RTE_ETH_FILTER_GENERIC:
1647 if (filter_op != RTE_ETH_FILTER_GET) {
1650 *(const void **)arg = &sfc_flow_ops;
1655 sfc_err(sa, "Unknown filter type %u", filter_type);
1659 sfc_log_init(sa, "exit: %d", -rc);
1660 SFC_ASSERT(rc >= 0);
1665 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1667 const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1670 * If Rx datapath does not provide callback to check mempool,
1671 * all pools are supported.
1673 if (sap->dp_rx->pool_ops_supported == NULL)
1676 return sap->dp_rx->pool_ops_supported(pool);
1679 static const struct eth_dev_ops sfc_eth_dev_ops = {
1680 .dev_configure = sfc_dev_configure,
1681 .dev_start = sfc_dev_start,
1682 .dev_stop = sfc_dev_stop,
1683 .dev_set_link_up = sfc_dev_set_link_up,
1684 .dev_set_link_down = sfc_dev_set_link_down,
1685 .dev_close = sfc_dev_close,
1686 .promiscuous_enable = sfc_dev_promisc_enable,
1687 .promiscuous_disable = sfc_dev_promisc_disable,
1688 .allmulticast_enable = sfc_dev_allmulti_enable,
1689 .allmulticast_disable = sfc_dev_allmulti_disable,
1690 .link_update = sfc_dev_link_update,
1691 .stats_get = sfc_stats_get,
1692 .stats_reset = sfc_stats_reset,
1693 .xstats_get = sfc_xstats_get,
1694 .xstats_reset = sfc_stats_reset,
1695 .xstats_get_names = sfc_xstats_get_names,
1696 .dev_infos_get = sfc_dev_infos_get,
1697 .dev_supported_ptypes_get = sfc_dev_supported_ptypes_get,
1698 .mtu_set = sfc_dev_set_mtu,
1699 .rx_queue_start = sfc_rx_queue_start,
1700 .rx_queue_stop = sfc_rx_queue_stop,
1701 .tx_queue_start = sfc_tx_queue_start,
1702 .tx_queue_stop = sfc_tx_queue_stop,
1703 .rx_queue_setup = sfc_rx_queue_setup,
1704 .rx_queue_release = sfc_rx_queue_release,
1705 .rx_queue_count = sfc_rx_queue_count,
1706 .rx_descriptor_done = sfc_rx_descriptor_done,
1707 .rx_descriptor_status = sfc_rx_descriptor_status,
1708 .tx_descriptor_status = sfc_tx_descriptor_status,
1709 .tx_queue_setup = sfc_tx_queue_setup,
1710 .tx_queue_release = sfc_tx_queue_release,
1711 .flow_ctrl_get = sfc_flow_ctrl_get,
1712 .flow_ctrl_set = sfc_flow_ctrl_set,
1713 .mac_addr_set = sfc_mac_addr_set,
1714 .udp_tunnel_port_add = sfc_dev_udp_tunnel_port_add,
1715 .udp_tunnel_port_del = sfc_dev_udp_tunnel_port_del,
1716 .reta_update = sfc_dev_rss_reta_update,
1717 .reta_query = sfc_dev_rss_reta_query,
1718 .rss_hash_update = sfc_dev_rss_hash_update,
1719 .rss_hash_conf_get = sfc_dev_rss_hash_conf_get,
1720 .filter_ctrl = sfc_dev_filter_ctrl,
1721 .set_mc_addr_list = sfc_set_mc_addr_list,
1722 .rxq_info_get = sfc_rx_queue_info_get,
1723 .txq_info_get = sfc_tx_queue_info_get,
1724 .fw_version_get = sfc_fw_version_get,
1725 .xstats_get_by_id = sfc_xstats_get_by_id,
1726 .xstats_get_names_by_id = sfc_xstats_get_names_by_id,
1727 .pool_ops_supported = sfc_pool_ops_supported,
1731 * Duplicate a string in potentially shared memory required for
1732 * multi-process support.
1734 * strdup() allocates from process-local heap/memory.
1737 sfc_strdup(const char *str)
1745 size = strlen(str) + 1;
1746 copy = rte_malloc(__func__, size, 0);
1748 rte_memcpy(copy, str, size);
1754 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1756 struct sfc_adapter *sa = dev->data->dev_private;
1757 const struct sfc_dp_rx *dp_rx;
1758 const struct sfc_dp_tx *dp_tx;
1759 const efx_nic_cfg_t *encp;
1760 unsigned int avail_caps = 0;
1761 const char *rx_name = NULL;
1762 const char *tx_name = NULL;
1765 switch (sa->family) {
1766 case EFX_FAMILY_HUNTINGTON:
1767 case EFX_FAMILY_MEDFORD:
1768 case EFX_FAMILY_MEDFORD2:
1769 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1775 encp = efx_nic_cfg_get(sa->nic);
1776 if (encp->enc_rx_es_super_buffer_supported)
1777 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1779 rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1780 sfc_kvarg_string_handler, &rx_name);
1782 goto fail_kvarg_rx_datapath;
1784 if (rx_name != NULL) {
1785 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1786 if (dp_rx == NULL) {
1787 sfc_err(sa, "Rx datapath %s not found", rx_name);
1791 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1793 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1796 goto fail_dp_rx_caps;
1799 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1800 if (dp_rx == NULL) {
1801 sfc_err(sa, "Rx datapath by caps %#x not found",
1808 sa->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1809 if (sa->dp_rx_name == NULL) {
1811 goto fail_dp_rx_name;
1814 sfc_notice(sa, "use %s Rx datapath", sa->dp_rx_name);
1816 rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1817 sfc_kvarg_string_handler, &tx_name);
1819 goto fail_kvarg_tx_datapath;
1821 if (tx_name != NULL) {
1822 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1823 if (dp_tx == NULL) {
1824 sfc_err(sa, "Tx datapath %s not found", tx_name);
1828 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
1830 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1833 goto fail_dp_tx_caps;
1836 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1837 if (dp_tx == NULL) {
1838 sfc_err(sa, "Tx datapath by caps %#x not found",
1845 sa->dp_tx_name = sfc_strdup(dp_tx->dp.name);
1846 if (sa->dp_tx_name == NULL) {
1848 goto fail_dp_tx_name;
1851 sfc_notice(sa, "use %s Tx datapath", sa->dp_tx_name);
1853 sa->priv.dp_rx = dp_rx;
1854 sa->priv.dp_tx = dp_tx;
1856 dev->rx_pkt_burst = dp_rx->pkt_burst;
1857 dev->tx_pkt_burst = dp_tx->pkt_burst;
1859 dev->dev_ops = &sfc_eth_dev_ops;
1866 fail_kvarg_tx_datapath:
1867 rte_free(sa->dp_rx_name);
1868 sa->dp_rx_name = NULL;
1873 fail_kvarg_rx_datapath:
1878 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1880 struct sfc_adapter *sa = dev->data->dev_private;
1882 dev->dev_ops = NULL;
1883 dev->rx_pkt_burst = NULL;
1884 dev->tx_pkt_burst = NULL;
1886 rte_free(sa->dp_tx_name);
1887 sa->dp_tx_name = NULL;
1888 sa->priv.dp_tx = NULL;
1890 rte_free(sa->dp_rx_name);
1891 sa->dp_rx_name = NULL;
1892 sa->priv.dp_rx = NULL;
1895 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1896 .rx_queue_count = sfc_rx_queue_count,
1897 .rx_descriptor_done = sfc_rx_descriptor_done,
1898 .rx_descriptor_status = sfc_rx_descriptor_status,
1899 .rxq_info_get = sfc_rx_queue_info_get,
1900 .txq_info_get = sfc_tx_queue_info_get,
1904 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
1907 * Device private data has really many process-local pointers.
1908 * Below code should be extremely careful to use data located
1909 * in shared memory only.
1911 struct sfc_adapter *sa = dev->data->dev_private;
1912 struct sfc_adapter_priv *sap;
1913 const struct sfc_dp_rx *dp_rx;
1914 const struct sfc_dp_tx *dp_tx;
1918 * Allocate process private data from heap, since it should not
1919 * be located in shared memory allocated using rte_malloc() API.
1921 sap = calloc(1, sizeof(*sap));
1924 goto fail_alloc_priv;
1927 sap->logtype_main = logtype_main;
1929 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1930 if (dp_rx == NULL) {
1931 SFC_LOG(sa, RTE_LOG_ERR, logtype_main,
1932 "cannot find %s Rx datapath", sa->dp_rx_name);
1936 if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1937 SFC_LOG(sa, RTE_LOG_ERR, logtype_main,
1938 "%s Rx datapath does not support multi-process",
1941 goto fail_dp_rx_multi_process;
1944 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1945 if (dp_tx == NULL) {
1946 SFC_LOG(sa, RTE_LOG_ERR, logtype_main,
1947 "cannot find %s Tx datapath", sa->dp_tx_name);
1951 if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1952 SFC_LOG(sa, RTE_LOG_ERR, logtype_main,
1953 "%s Tx datapath does not support multi-process",
1956 goto fail_dp_tx_multi_process;
1962 dev->process_private = sap;
1963 dev->rx_pkt_burst = dp_rx->pkt_burst;
1964 dev->tx_pkt_burst = dp_tx->pkt_burst;
1965 dev->dev_ops = &sfc_eth_dev_secondary_ops;
1969 fail_dp_tx_multi_process:
1971 fail_dp_rx_multi_process:
1980 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1982 free(dev->process_private);
1983 dev->process_private = NULL;
1984 dev->dev_ops = NULL;
1985 dev->tx_pkt_burst = NULL;
1986 dev->rx_pkt_burst = NULL;
1990 sfc_register_dp(void)
1993 if (TAILQ_EMPTY(&sfc_dp_head)) {
1994 /* Prefer EF10 datapath */
1995 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
1996 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1997 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1999 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2000 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2001 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2006 sfc_eth_dev_init(struct rte_eth_dev *dev)
2008 struct sfc_adapter *sa = dev->data->dev_private;
2009 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2010 uint32_t logtype_main;
2012 const efx_nic_cfg_t *encp;
2013 const struct ether_addr *from;
2017 logtype_main = sfc_register_logtype(&pci_dev->addr,
2018 SFC_LOGTYPE_MAIN_STR,
2021 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2022 return -sfc_eth_dev_secondary_init(dev, logtype_main);
2025 * sfc_adapter is a mixture of shared and process private data.
2026 * During transition period use it in both kinds. When the
2027 * driver becomes ready to separate it, sfc_adapter will become
2028 * primary process private only.
2030 dev->process_private = sa;
2032 /* Required for logging */
2033 sa->pci_addr = pci_dev->addr;
2034 sa->port_id = dev->data->port_id;
2035 sa->priv.logtype_main = logtype_main;
2039 /* Copy PCI device info to the dev->data */
2040 rte_eth_copy_pci_info(dev, pci_dev);
2042 rc = sfc_kvargs_parse(sa);
2044 goto fail_kvargs_parse;
2046 sfc_log_init(sa, "entry");
2048 dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
2049 if (dev->data->mac_addrs == NULL) {
2051 goto fail_mac_addrs;
2054 sfc_adapter_lock_init(sa);
2055 sfc_adapter_lock(sa);
2057 sfc_log_init(sa, "probing");
2062 sfc_log_init(sa, "set device ops");
2063 rc = sfc_eth_dev_set_ops(dev);
2067 sfc_log_init(sa, "attaching");
2068 rc = sfc_attach(sa);
2072 encp = efx_nic_cfg_get(sa->nic);
2075 * The arguments are really reverse order in comparison to
2076 * Linux kernel. Copy from NIC config to Ethernet device data.
2078 from = (const struct ether_addr *)(encp->enc_mac_addr);
2079 ether_addr_copy(from, &dev->data->mac_addrs[0]);
2081 sfc_adapter_unlock(sa);
2083 sfc_log_init(sa, "done");
2087 sfc_eth_dev_clear_ops(dev);
2093 sfc_adapter_unlock(sa);
2094 sfc_adapter_lock_fini(sa);
2095 rte_free(dev->data->mac_addrs);
2096 dev->data->mac_addrs = NULL;
2099 sfc_kvargs_cleanup(sa);
2102 sfc_log_init(sa, "failed %d", rc);
2103 dev->process_private = NULL;
2109 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2111 struct sfc_adapter *sa;
2113 if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2114 sfc_eth_dev_secondary_clear_ops(dev);
2118 sa = dev->data->dev_private;
2119 sfc_log_init(sa, "entry");
2121 sfc_adapter_lock(sa);
2123 sfc_eth_dev_clear_ops(dev);
2128 sfc_kvargs_cleanup(sa);
2130 sfc_adapter_unlock(sa);
2131 sfc_adapter_lock_fini(sa);
2133 sfc_log_init(sa, "done");
2135 /* Required for logging, so cleanup last */
2140 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2141 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2142 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2143 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2144 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2145 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2146 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2147 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2148 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2149 { .vendor_id = 0 /* sentinel */ }
2152 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2153 struct rte_pci_device *pci_dev)
2155 return rte_eth_dev_pci_generic_probe(pci_dev,
2156 sizeof(struct sfc_adapter), sfc_eth_dev_init);
2159 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2161 return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2164 static struct rte_pci_driver sfc_efx_pmd = {
2165 .id_table = pci_id_sfc_efx_map,
2167 RTE_PCI_DRV_INTR_LSC |
2168 RTE_PCI_DRV_NEED_MAPPING,
2169 .probe = sfc_eth_dev_pci_probe,
2170 .remove = sfc_eth_dev_pci_remove,
2173 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2174 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2175 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2176 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2177 SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2178 SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2179 SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2180 SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2181 SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2182 SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2184 RTE_INIT(sfc_driver_register_logtype)
2188 ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2190 sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;