4 * Copyright (c) 2016-2017 Solarflare Communications Inc.
7 * This software was jointly developed between OKTET Labs (under contract
8 * for Solarflare) and Solarflare Communications, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions are met:
13 * 1. Redistributions of source code must retain the above copyright notice,
14 * this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright notice,
16 * this list of conditions and the following disclaimer in the documentation
17 * and/or other materials provided with the distribution.
19 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
20 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
21 * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
23 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
24 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
25 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
26 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
27 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
28 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
29 * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33 #include <rte_ethdev.h>
34 #include <rte_ethdev_pci.h>
36 #include <rte_bus_pci.h>
37 #include <rte_errno.h>
42 #include "sfc_debug.h"
44 #include "sfc_kvargs.h"
50 #include "sfc_dp_rx.h"
52 static struct sfc_dp_list sfc_dp_head =
53 TAILQ_HEAD_INITIALIZER(sfc_dp_head);
56 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
58 struct sfc_adapter *sa = dev->data->dev_private;
59 efx_nic_fw_info_t enfi;
64 * Return value of the callback is likely supposed to be
65 * equal to or greater than 0, nevertheless, if an error
66 * occurs, it will be desirable to pass it to the caller
68 if ((fw_version == NULL) || (fw_size == 0))
71 rc = efx_nic_get_fw_version(sa->nic, &enfi);
75 ret = snprintf(fw_version, fw_size,
76 "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
77 enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
78 enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
82 if (enfi.enfi_dpcpu_fw_ids_valid) {
83 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
86 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
87 fw_size - dpcpu_fw_ids_offset,
88 " rx%" PRIx16 " tx%" PRIx16,
89 enfi.enfi_rx_dpcpu_fw_id,
90 enfi.enfi_tx_dpcpu_fw_id);
97 if (fw_size < (size_t)(++ret))
104 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
106 struct sfc_adapter *sa = dev->data->dev_private;
107 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
109 sfc_log_init(sa, "entry");
111 dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
112 dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
114 /* Autonegotiation may be disabled */
115 dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
116 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
117 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
118 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
119 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
120 if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
121 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
123 dev_info->max_rx_queues = sa->rxq_max;
124 dev_info->max_tx_queues = sa->txq_max;
126 /* By default packets are dropped if no descriptors are available */
127 dev_info->default_rxconf.rx_drop_en = 1;
129 dev_info->rx_offload_capa =
130 DEV_RX_OFFLOAD_IPV4_CKSUM |
131 DEV_RX_OFFLOAD_UDP_CKSUM |
132 DEV_RX_OFFLOAD_TCP_CKSUM;
134 if ((encp->enc_tunnel_encapsulations_supported != 0) &&
135 (sa->dp_rx->features & SFC_DP_RX_FEAT_TUNNELS))
136 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM;
138 dev_info->tx_offload_capa =
139 DEV_TX_OFFLOAD_IPV4_CKSUM |
140 DEV_TX_OFFLOAD_UDP_CKSUM |
141 DEV_TX_OFFLOAD_TCP_CKSUM;
143 dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
144 if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
145 !encp->enc_hw_tx_insert_vlan_enabled)
146 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
148 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_VLAN_INSERT;
150 if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
151 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
153 if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_POOL)
154 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTMEMP;
156 if (~sa->dp_tx->features & SFC_DP_TX_FEAT_REFCNT)
157 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOREFCOUNT;
159 #if EFSYS_OPT_RX_SCALE
160 if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
161 dev_info->reta_size = EFX_RSS_TBL_SIZE;
162 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
163 dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
168 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
170 dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
171 dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
172 /* The RXQ hardware requires that the descriptor count is a power
173 * of 2, but rx_desc_lim cannot properly describe that constraint.
175 dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
177 dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
178 dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
180 * The TXQ hardware requires that the descriptor count is a power
181 * of 2, but tx_desc_lim cannot properly describe that constraint
183 dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
186 static const uint32_t *
187 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
189 struct sfc_adapter *sa = dev->data->dev_private;
190 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
191 uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
193 return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
197 sfc_dev_configure(struct rte_eth_dev *dev)
199 struct rte_eth_dev_data *dev_data = dev->data;
200 struct sfc_adapter *sa = dev_data->dev_private;
203 sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
204 dev_data->nb_rx_queues, dev_data->nb_tx_queues);
206 sfc_adapter_lock(sa);
208 case SFC_ADAPTER_CONFIGURED:
210 case SFC_ADAPTER_INITIALIZED:
211 rc = sfc_configure(sa);
214 sfc_err(sa, "unexpected adapter state %u to configure",
219 sfc_adapter_unlock(sa);
221 sfc_log_init(sa, "done %d", rc);
227 sfc_dev_start(struct rte_eth_dev *dev)
229 struct sfc_adapter *sa = dev->data->dev_private;
232 sfc_log_init(sa, "entry");
234 sfc_adapter_lock(sa);
236 sfc_adapter_unlock(sa);
238 sfc_log_init(sa, "done %d", rc);
244 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
246 struct sfc_adapter *sa = dev->data->dev_private;
247 struct rte_eth_link *dev_link = &dev->data->dev_link;
248 struct rte_eth_link old_link;
249 struct rte_eth_link current_link;
251 sfc_log_init(sa, "entry");
254 EFX_STATIC_ASSERT(sizeof(*dev_link) == sizeof(rte_atomic64_t));
255 *(int64_t *)&old_link = rte_atomic64_read((rte_atomic64_t *)dev_link);
257 if (sa->state != SFC_ADAPTER_STARTED) {
258 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, ¤t_link);
259 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
260 *(uint64_t *)&old_link,
261 *(uint64_t *)¤t_link))
263 } else if (wait_to_complete) {
264 efx_link_mode_t link_mode;
266 if (efx_port_poll(sa->nic, &link_mode) != 0)
267 link_mode = EFX_LINK_UNKNOWN;
268 sfc_port_link_mode_to_info(link_mode, ¤t_link);
270 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
271 *(uint64_t *)&old_link,
272 *(uint64_t *)¤t_link))
275 sfc_ev_mgmt_qpoll(sa);
276 *(int64_t *)¤t_link =
277 rte_atomic64_read((rte_atomic64_t *)dev_link);
280 if (old_link.link_status != current_link.link_status)
281 sfc_info(sa, "Link status is %s",
282 current_link.link_status ? "UP" : "DOWN");
284 return old_link.link_status == current_link.link_status ? 0 : -1;
288 sfc_dev_stop(struct rte_eth_dev *dev)
290 struct sfc_adapter *sa = dev->data->dev_private;
292 sfc_log_init(sa, "entry");
294 sfc_adapter_lock(sa);
296 sfc_adapter_unlock(sa);
298 sfc_log_init(sa, "done");
302 sfc_dev_set_link_up(struct rte_eth_dev *dev)
304 struct sfc_adapter *sa = dev->data->dev_private;
307 sfc_log_init(sa, "entry");
309 sfc_adapter_lock(sa);
311 sfc_adapter_unlock(sa);
318 sfc_dev_set_link_down(struct rte_eth_dev *dev)
320 struct sfc_adapter *sa = dev->data->dev_private;
322 sfc_log_init(sa, "entry");
324 sfc_adapter_lock(sa);
326 sfc_adapter_unlock(sa);
332 sfc_dev_close(struct rte_eth_dev *dev)
334 struct sfc_adapter *sa = dev->data->dev_private;
336 sfc_log_init(sa, "entry");
338 sfc_adapter_lock(sa);
340 case SFC_ADAPTER_STARTED:
342 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
344 case SFC_ADAPTER_CONFIGURED:
346 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
348 case SFC_ADAPTER_INITIALIZED:
351 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
354 sfc_adapter_unlock(sa);
356 sfc_log_init(sa, "done");
360 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
363 struct sfc_port *port;
365 struct sfc_adapter *sa = dev->data->dev_private;
366 boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
367 const char *desc = (allmulti) ? "all-multi" : "promiscuous";
369 sfc_adapter_lock(sa);
372 toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
374 if (*toggle != enabled) {
377 if (port->isolated) {
378 sfc_warn(sa, "isolated mode is active on the port");
379 sfc_warn(sa, "the change is to be applied on the next "
380 "start provided that isolated mode is "
381 "disabled prior the next start");
382 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
383 (sfc_set_rx_mode(sa) != 0)) {
384 *toggle = !(enabled);
385 sfc_warn(sa, "Failed to %s %s mode",
386 ((enabled) ? "enable" : "disable"), desc);
390 sfc_adapter_unlock(sa);
394 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
396 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
400 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
402 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
406 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
408 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
412 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
414 sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
418 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
419 uint16_t nb_rx_desc, unsigned int socket_id,
420 const struct rte_eth_rxconf *rx_conf,
421 struct rte_mempool *mb_pool)
423 struct sfc_adapter *sa = dev->data->dev_private;
426 sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
427 rx_queue_id, nb_rx_desc, socket_id);
429 sfc_adapter_lock(sa);
431 rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
436 dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
438 sfc_adapter_unlock(sa);
443 sfc_adapter_unlock(sa);
449 sfc_rx_queue_release(void *queue)
451 struct sfc_dp_rxq *dp_rxq = queue;
453 struct sfc_adapter *sa;
454 unsigned int sw_index;
459 rxq = sfc_rxq_by_dp_rxq(dp_rxq);
461 sfc_adapter_lock(sa);
463 sw_index = sfc_rxq_sw_index(rxq);
465 sfc_log_init(sa, "RxQ=%u", sw_index);
467 sa->eth_dev->data->rx_queues[sw_index] = NULL;
469 sfc_rx_qfini(sa, sw_index);
471 sfc_adapter_unlock(sa);
475 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
476 uint16_t nb_tx_desc, unsigned int socket_id,
477 const struct rte_eth_txconf *tx_conf)
479 struct sfc_adapter *sa = dev->data->dev_private;
482 sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
483 tx_queue_id, nb_tx_desc, socket_id);
485 sfc_adapter_lock(sa);
487 rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
491 dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
493 sfc_adapter_unlock(sa);
497 sfc_adapter_unlock(sa);
503 sfc_tx_queue_release(void *queue)
505 struct sfc_dp_txq *dp_txq = queue;
507 unsigned int sw_index;
508 struct sfc_adapter *sa;
513 txq = sfc_txq_by_dp_txq(dp_txq);
514 sw_index = sfc_txq_sw_index(txq);
516 SFC_ASSERT(txq->evq != NULL);
519 sfc_log_init(sa, "TxQ = %u", sw_index);
521 sfc_adapter_lock(sa);
523 SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
524 sa->eth_dev->data->tx_queues[sw_index] = NULL;
526 sfc_tx_qfini(sa, sw_index);
528 sfc_adapter_unlock(sa);
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_OVERFLOW];
566 stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
567 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
569 stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
570 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
571 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
572 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
574 * Take into account stats which are whenever supported
575 * on EF10. If some stat is not supported by current
576 * firmware variant or HW revision, it is guaranteed
577 * to be zero in mac_stats.
580 mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
581 mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
582 mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
583 mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
584 mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
585 mac_stats[EFX_MAC_PM_TRUNC_QBB] +
586 mac_stats[EFX_MAC_PM_DISCARD_QBB] +
587 mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
588 mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
589 mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
591 mac_stats[EFX_MAC_RX_FCS_ERRORS] +
592 mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
593 mac_stats[EFX_MAC_RX_JABBER_PKTS];
594 /* no oerrors counters supported on EF10 */
598 rte_spinlock_unlock(&port->mac_stats_lock);
599 SFC_ASSERT(ret >= 0);
604 sfc_stats_reset(struct rte_eth_dev *dev)
606 struct sfc_adapter *sa = dev->data->dev_private;
607 struct sfc_port *port = &sa->port;
610 if (sa->state != SFC_ADAPTER_STARTED) {
612 * The operation cannot be done if port is not started; it
613 * will be scheduled to be done during the next port start
615 port->mac_stats_reset_pending = B_TRUE;
619 rc = sfc_port_reset_mac_stats(sa);
621 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
625 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
626 unsigned int xstats_count)
628 struct sfc_adapter *sa = dev->data->dev_private;
629 struct sfc_port *port = &sa->port;
635 rte_spinlock_lock(&port->mac_stats_lock);
637 rc = sfc_port_update_mac_stats(sa);
644 mac_stats = port->mac_stats_buf;
646 for (i = 0; i < EFX_MAC_NSTATS; ++i) {
647 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
648 if (xstats != NULL && nstats < (int)xstats_count) {
649 xstats[nstats].id = nstats;
650 xstats[nstats].value = mac_stats[i];
657 rte_spinlock_unlock(&port->mac_stats_lock);
663 sfc_xstats_get_names(struct rte_eth_dev *dev,
664 struct rte_eth_xstat_name *xstats_names,
665 unsigned int xstats_count)
667 struct sfc_adapter *sa = dev->data->dev_private;
668 struct sfc_port *port = &sa->port;
670 unsigned int nstats = 0;
672 for (i = 0; i < EFX_MAC_NSTATS; ++i) {
673 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
674 if (xstats_names != NULL && nstats < xstats_count)
675 strncpy(xstats_names[nstats].name,
676 efx_mac_stat_name(sa->nic, i),
677 sizeof(xstats_names[0].name));
686 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
687 uint64_t *values, unsigned int n)
689 struct sfc_adapter *sa = dev->data->dev_private;
690 struct sfc_port *port = &sa->port;
692 unsigned int nb_supported = 0;
693 unsigned int nb_written = 0;
698 if (unlikely(values == NULL) ||
699 unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
700 return port->mac_stats_nb_supported;
702 rte_spinlock_lock(&port->mac_stats_lock);
704 rc = sfc_port_update_mac_stats(sa);
711 mac_stats = port->mac_stats_buf;
713 for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
714 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
717 if ((ids == NULL) || (ids[nb_written] == nb_supported))
718 values[nb_written++] = mac_stats[i];
726 rte_spinlock_unlock(&port->mac_stats_lock);
732 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
733 struct rte_eth_xstat_name *xstats_names,
734 const uint64_t *ids, unsigned int size)
736 struct sfc_adapter *sa = dev->data->dev_private;
737 struct sfc_port *port = &sa->port;
738 unsigned int nb_supported = 0;
739 unsigned int nb_written = 0;
742 if (unlikely(xstats_names == NULL) ||
743 unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
744 return port->mac_stats_nb_supported;
746 for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
747 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
750 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
751 char *name = xstats_names[nb_written++].name;
753 strncpy(name, efx_mac_stat_name(sa->nic, i),
754 sizeof(xstats_names[0].name));
755 name[sizeof(xstats_names[0].name) - 1] = '\0';
765 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
767 struct sfc_adapter *sa = dev->data->dev_private;
768 unsigned int wanted_fc, link_fc;
770 memset(fc_conf, 0, sizeof(*fc_conf));
772 sfc_adapter_lock(sa);
774 if (sa->state == SFC_ADAPTER_STARTED)
775 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
777 link_fc = sa->port.flow_ctrl;
781 fc_conf->mode = RTE_FC_NONE;
783 case EFX_FCNTL_RESPOND:
784 fc_conf->mode = RTE_FC_RX_PAUSE;
786 case EFX_FCNTL_GENERATE:
787 fc_conf->mode = RTE_FC_TX_PAUSE;
789 case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
790 fc_conf->mode = RTE_FC_FULL;
793 sfc_err(sa, "%s: unexpected flow control value %#x",
797 fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
799 sfc_adapter_unlock(sa);
805 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
807 struct sfc_adapter *sa = dev->data->dev_private;
808 struct sfc_port *port = &sa->port;
812 if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
813 fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
814 fc_conf->mac_ctrl_frame_fwd != 0) {
815 sfc_err(sa, "unsupported flow control settings specified");
820 switch (fc_conf->mode) {
824 case RTE_FC_RX_PAUSE:
825 fcntl = EFX_FCNTL_RESPOND;
827 case RTE_FC_TX_PAUSE:
828 fcntl = EFX_FCNTL_GENERATE;
831 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
838 sfc_adapter_lock(sa);
840 if (sa->state == SFC_ADAPTER_STARTED) {
841 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
843 goto fail_mac_fcntl_set;
846 port->flow_ctrl = fcntl;
847 port->flow_ctrl_autoneg = fc_conf->autoneg;
849 sfc_adapter_unlock(sa);
854 sfc_adapter_unlock(sa);
861 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
863 struct sfc_adapter *sa = dev->data->dev_private;
864 size_t pdu = EFX_MAC_PDU(mtu);
868 sfc_log_init(sa, "mtu=%u", mtu);
871 if (pdu < EFX_MAC_PDU_MIN) {
872 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
873 (unsigned int)mtu, (unsigned int)pdu,
877 if (pdu > EFX_MAC_PDU_MAX) {
878 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
879 (unsigned int)mtu, (unsigned int)pdu,
884 sfc_adapter_lock(sa);
886 if (pdu != sa->port.pdu) {
887 if (sa->state == SFC_ADAPTER_STARTED) {
890 old_pdu = sa->port.pdu;
901 * The driver does not use it, but other PMDs update jumbo_frame
902 * flag and max_rx_pkt_len when MTU is set.
904 dev->data->dev_conf.rxmode.jumbo_frame = (mtu > ETHER_MAX_LEN);
905 dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
907 sfc_adapter_unlock(sa);
909 sfc_log_init(sa, "done");
913 sa->port.pdu = old_pdu;
914 if (sfc_start(sa) != 0)
915 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
916 "PDU max size - port is stopped",
917 (unsigned int)pdu, (unsigned int)old_pdu);
918 sfc_adapter_unlock(sa);
921 sfc_log_init(sa, "failed %d", rc);
926 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
928 struct sfc_adapter *sa = dev->data->dev_private;
929 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
930 struct sfc_port *port = &sa->port;
933 sfc_adapter_lock(sa);
936 * Copy the address to the device private data so that
937 * it could be recalled in the case of adapter restart.
939 ether_addr_copy(mac_addr, &port->default_mac_addr);
941 if (port->isolated) {
942 sfc_err(sa, "isolated mode is active on the port");
943 sfc_err(sa, "will not set MAC address");
947 if (sa->state != SFC_ADAPTER_STARTED) {
948 sfc_info(sa, "the port is not started");
949 sfc_info(sa, "the new MAC address will be set on port start");
954 if (encp->enc_allow_set_mac_with_installed_filters) {
955 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
957 sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
962 * Changing the MAC address by means of MCDI request
963 * has no effect on received traffic, therefore
964 * we also need to update unicast filters
966 rc = sfc_set_rx_mode(sa);
968 sfc_err(sa, "cannot set filter (rc = %u)", rc);
970 sfc_warn(sa, "cannot set MAC address with filters installed");
971 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
972 sfc_warn(sa, "(some traffic may be dropped)");
975 * Since setting MAC address with filters installed is not
976 * allowed on the adapter, the new MAC address will be set
977 * by means of adapter restart. sfc_start() shall retrieve
978 * the new address from the device private data and set it.
983 sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
988 * In the case of failure sa->port->default_mac_addr does not
989 * need rollback since no error code is returned, and the upper
990 * API will anyway update the external MAC address storage.
991 * To be consistent with that new value it is better to keep
992 * the device private value the same.
994 sfc_adapter_unlock(sa);
999 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
1000 uint32_t nb_mc_addr)
1002 struct sfc_adapter *sa = dev->data->dev_private;
1003 struct sfc_port *port = &sa->port;
1004 uint8_t *mc_addrs = port->mcast_addrs;
1008 if (port->isolated) {
1009 sfc_err(sa, "isolated mode is active on the port");
1010 sfc_err(sa, "will not set multicast address list");
1014 if (mc_addrs == NULL)
1017 if (nb_mc_addr > port->max_mcast_addrs) {
1018 sfc_err(sa, "too many multicast addresses: %u > %u",
1019 nb_mc_addr, port->max_mcast_addrs);
1023 for (i = 0; i < nb_mc_addr; ++i) {
1024 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1026 mc_addrs += EFX_MAC_ADDR_LEN;
1029 port->nb_mcast_addrs = nb_mc_addr;
1031 if (sa->state != SFC_ADAPTER_STARTED)
1034 rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1035 port->nb_mcast_addrs);
1037 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1044 * The function is used by the secondary process as well. It must not
1045 * use any process-local pointers from the adapter data.
1048 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1049 struct rte_eth_rxq_info *qinfo)
1051 struct sfc_adapter *sa = dev->data->dev_private;
1052 struct sfc_rxq_info *rxq_info;
1053 struct sfc_rxq *rxq;
1055 sfc_adapter_lock(sa);
1057 SFC_ASSERT(rx_queue_id < sa->rxq_count);
1059 rxq_info = &sa->rxq_info[rx_queue_id];
1060 rxq = rxq_info->rxq;
1061 SFC_ASSERT(rxq != NULL);
1063 qinfo->mp = rxq->refill_mb_pool;
1064 qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1065 qinfo->conf.rx_drop_en = 1;
1066 qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1067 qinfo->scattered_rx =
1068 ((rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) != 0);
1069 qinfo->nb_desc = rxq_info->entries;
1071 sfc_adapter_unlock(sa);
1075 * The function is used by the secondary process as well. It must not
1076 * use any process-local pointers from the adapter data.
1079 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1080 struct rte_eth_txq_info *qinfo)
1082 struct sfc_adapter *sa = dev->data->dev_private;
1083 struct sfc_txq_info *txq_info;
1085 sfc_adapter_lock(sa);
1087 SFC_ASSERT(tx_queue_id < sa->txq_count);
1089 txq_info = &sa->txq_info[tx_queue_id];
1090 SFC_ASSERT(txq_info->txq != NULL);
1092 memset(qinfo, 0, sizeof(*qinfo));
1094 qinfo->conf.txq_flags = txq_info->txq->flags;
1095 qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1096 qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1097 qinfo->nb_desc = txq_info->entries;
1099 sfc_adapter_unlock(sa);
1103 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1105 struct sfc_adapter *sa = dev->data->dev_private;
1107 sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1109 return sfc_rx_qdesc_npending(sa, rx_queue_id);
1113 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1115 struct sfc_dp_rxq *dp_rxq = queue;
1117 return sfc_rx_qdesc_done(dp_rxq, offset);
1121 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1123 struct sfc_dp_rxq *dp_rxq = queue;
1124 struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1126 return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1130 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1132 struct sfc_dp_txq *dp_txq = queue;
1133 struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1135 return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1139 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1141 struct sfc_adapter *sa = dev->data->dev_private;
1144 sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1146 sfc_adapter_lock(sa);
1149 if (sa->state != SFC_ADAPTER_STARTED)
1150 goto fail_not_started;
1152 rc = sfc_rx_qstart(sa, rx_queue_id);
1154 goto fail_rx_qstart;
1156 sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1158 sfc_adapter_unlock(sa);
1164 sfc_adapter_unlock(sa);
1170 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1172 struct sfc_adapter *sa = dev->data->dev_private;
1174 sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1176 sfc_adapter_lock(sa);
1177 sfc_rx_qstop(sa, rx_queue_id);
1179 sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1181 sfc_adapter_unlock(sa);
1187 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1189 struct sfc_adapter *sa = dev->data->dev_private;
1192 sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1194 sfc_adapter_lock(sa);
1197 if (sa->state != SFC_ADAPTER_STARTED)
1198 goto fail_not_started;
1200 rc = sfc_tx_qstart(sa, tx_queue_id);
1202 goto fail_tx_qstart;
1204 sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1206 sfc_adapter_unlock(sa);
1212 sfc_adapter_unlock(sa);
1218 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1220 struct sfc_adapter *sa = dev->data->dev_private;
1222 sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1224 sfc_adapter_lock(sa);
1226 sfc_tx_qstop(sa, tx_queue_id);
1228 sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1230 sfc_adapter_unlock(sa);
1234 static efx_tunnel_protocol_t
1235 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1238 case RTE_TUNNEL_TYPE_VXLAN:
1239 return EFX_TUNNEL_PROTOCOL_VXLAN;
1240 case RTE_TUNNEL_TYPE_GENEVE:
1241 return EFX_TUNNEL_PROTOCOL_GENEVE;
1243 return EFX_TUNNEL_NPROTOS;
1247 enum sfc_udp_tunnel_op_e {
1248 SFC_UDP_TUNNEL_ADD_PORT,
1249 SFC_UDP_TUNNEL_DEL_PORT,
1253 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1254 struct rte_eth_udp_tunnel *tunnel_udp,
1255 enum sfc_udp_tunnel_op_e op)
1257 struct sfc_adapter *sa = dev->data->dev_private;
1258 efx_tunnel_protocol_t tunnel_proto;
1261 sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1262 (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1263 (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1264 tunnel_udp->udp_port, tunnel_udp->prot_type);
1267 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1268 if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1270 goto fail_bad_proto;
1273 sfc_adapter_lock(sa);
1276 case SFC_UDP_TUNNEL_ADD_PORT:
1277 rc = efx_tunnel_config_udp_add(sa->nic,
1278 tunnel_udp->udp_port,
1281 case SFC_UDP_TUNNEL_DEL_PORT:
1282 rc = efx_tunnel_config_udp_remove(sa->nic,
1283 tunnel_udp->udp_port,
1294 if (sa->state == SFC_ADAPTER_STARTED) {
1295 rc = efx_tunnel_reconfigure(sa->nic);
1298 * Configuration is accepted by FW and MC reboot
1299 * is initiated to apply the changes. MC reboot
1300 * will be handled in a usual way (MC reboot
1301 * event on management event queue and adapter
1305 } else if (rc != 0) {
1306 goto fail_reconfigure;
1310 sfc_adapter_unlock(sa);
1314 /* Remove/restore entry since the change makes the trouble */
1316 case SFC_UDP_TUNNEL_ADD_PORT:
1317 (void)efx_tunnel_config_udp_remove(sa->nic,
1318 tunnel_udp->udp_port,
1321 case SFC_UDP_TUNNEL_DEL_PORT:
1322 (void)efx_tunnel_config_udp_add(sa->nic,
1323 tunnel_udp->udp_port,
1330 sfc_adapter_unlock(sa);
1338 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1339 struct rte_eth_udp_tunnel *tunnel_udp)
1341 return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1345 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1346 struct rte_eth_udp_tunnel *tunnel_udp)
1348 return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1351 #if EFSYS_OPT_RX_SCALE
1353 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1354 struct rte_eth_rss_conf *rss_conf)
1356 struct sfc_adapter *sa = dev->data->dev_private;
1357 struct sfc_port *port = &sa->port;
1359 if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1362 if (sa->rss_channels == 0)
1365 sfc_adapter_lock(sa);
1368 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1369 * hence, conversion is done here to derive a correct set of ETH_RSS
1370 * flags which corresponds to the active EFX configuration stored
1371 * locally in 'sfc_adapter' and kept up-to-date
1373 rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1374 rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1375 if (rss_conf->rss_key != NULL)
1376 rte_memcpy(rss_conf->rss_key, sa->rss_key, EFX_RSS_KEY_SIZE);
1378 sfc_adapter_unlock(sa);
1384 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1385 struct rte_eth_rss_conf *rss_conf)
1387 struct sfc_adapter *sa = dev->data->dev_private;
1388 struct sfc_port *port = &sa->port;
1389 unsigned int efx_hash_types;
1395 if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1396 sfc_err(sa, "RSS is not available");
1400 if (sa->rss_channels == 0) {
1401 sfc_err(sa, "RSS is not configured");
1405 if ((rss_conf->rss_key != NULL) &&
1406 (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1407 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1408 sizeof(sa->rss_key));
1412 if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1413 sfc_err(sa, "unsupported hash functions requested");
1417 sfc_adapter_lock(sa);
1419 efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1421 rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1422 EFX_RX_HASHALG_TOEPLITZ,
1423 efx_hash_types, B_TRUE);
1425 goto fail_scale_mode_set;
1427 if (rss_conf->rss_key != NULL) {
1428 if (sa->state == SFC_ADAPTER_STARTED) {
1429 rc = efx_rx_scale_key_set(sa->nic,
1430 EFX_RSS_CONTEXT_DEFAULT,
1432 sizeof(sa->rss_key));
1434 goto fail_scale_key_set;
1437 rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1440 sa->rss_hash_types = efx_hash_types;
1442 sfc_adapter_unlock(sa);
1447 if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1448 EFX_RX_HASHALG_TOEPLITZ,
1449 sa->rss_hash_types, B_TRUE) != 0)
1450 sfc_err(sa, "failed to restore RSS mode");
1452 fail_scale_mode_set:
1453 sfc_adapter_unlock(sa);
1458 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1459 struct rte_eth_rss_reta_entry64 *reta_conf,
1462 struct sfc_adapter *sa = dev->data->dev_private;
1463 struct sfc_port *port = &sa->port;
1466 if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1469 if (sa->rss_channels == 0)
1472 if (reta_size != EFX_RSS_TBL_SIZE)
1475 sfc_adapter_lock(sa);
1477 for (entry = 0; entry < reta_size; entry++) {
1478 int grp = entry / RTE_RETA_GROUP_SIZE;
1479 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1481 if ((reta_conf[grp].mask >> grp_idx) & 1)
1482 reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1485 sfc_adapter_unlock(sa);
1491 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1492 struct rte_eth_rss_reta_entry64 *reta_conf,
1495 struct sfc_adapter *sa = dev->data->dev_private;
1496 struct sfc_port *port = &sa->port;
1497 unsigned int *rss_tbl_new;
1505 if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1506 sfc_err(sa, "RSS is not available");
1510 if (sa->rss_channels == 0) {
1511 sfc_err(sa, "RSS is not configured");
1515 if (reta_size != EFX_RSS_TBL_SIZE) {
1516 sfc_err(sa, "RETA size is wrong (should be %u)",
1521 rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1522 if (rss_tbl_new == NULL)
1525 sfc_adapter_lock(sa);
1527 rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1529 for (entry = 0; entry < reta_size; entry++) {
1530 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1531 struct rte_eth_rss_reta_entry64 *grp;
1533 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1535 if (grp->mask & (1ull << grp_idx)) {
1536 if (grp->reta[grp_idx] >= sa->rss_channels) {
1538 goto bad_reta_entry;
1540 rss_tbl_new[entry] = grp->reta[grp_idx];
1544 if (sa->state == SFC_ADAPTER_STARTED) {
1545 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1546 rss_tbl_new, EFX_RSS_TBL_SIZE);
1548 goto fail_scale_tbl_set;
1551 rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1555 sfc_adapter_unlock(sa);
1557 rte_free(rss_tbl_new);
1559 SFC_ASSERT(rc >= 0);
1565 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1566 enum rte_filter_op filter_op,
1569 struct sfc_adapter *sa = dev->data->dev_private;
1572 sfc_log_init(sa, "entry");
1574 switch (filter_type) {
1575 case RTE_ETH_FILTER_NONE:
1576 sfc_err(sa, "Global filters configuration not supported");
1578 case RTE_ETH_FILTER_MACVLAN:
1579 sfc_err(sa, "MACVLAN filters not supported");
1581 case RTE_ETH_FILTER_ETHERTYPE:
1582 sfc_err(sa, "EtherType filters not supported");
1584 case RTE_ETH_FILTER_FLEXIBLE:
1585 sfc_err(sa, "Flexible filters not supported");
1587 case RTE_ETH_FILTER_SYN:
1588 sfc_err(sa, "SYN filters not supported");
1590 case RTE_ETH_FILTER_NTUPLE:
1591 sfc_err(sa, "NTUPLE filters not supported");
1593 case RTE_ETH_FILTER_TUNNEL:
1594 sfc_err(sa, "Tunnel filters not supported");
1596 case RTE_ETH_FILTER_FDIR:
1597 sfc_err(sa, "Flow Director filters not supported");
1599 case RTE_ETH_FILTER_HASH:
1600 sfc_err(sa, "Hash filters not supported");
1602 case RTE_ETH_FILTER_GENERIC:
1603 if (filter_op != RTE_ETH_FILTER_GET) {
1606 *(const void **)arg = &sfc_flow_ops;
1611 sfc_err(sa, "Unknown filter type %u", filter_type);
1615 sfc_log_init(sa, "exit: %d", -rc);
1616 SFC_ASSERT(rc >= 0);
1620 static const struct eth_dev_ops sfc_eth_dev_ops = {
1621 .dev_configure = sfc_dev_configure,
1622 .dev_start = sfc_dev_start,
1623 .dev_stop = sfc_dev_stop,
1624 .dev_set_link_up = sfc_dev_set_link_up,
1625 .dev_set_link_down = sfc_dev_set_link_down,
1626 .dev_close = sfc_dev_close,
1627 .promiscuous_enable = sfc_dev_promisc_enable,
1628 .promiscuous_disable = sfc_dev_promisc_disable,
1629 .allmulticast_enable = sfc_dev_allmulti_enable,
1630 .allmulticast_disable = sfc_dev_allmulti_disable,
1631 .link_update = sfc_dev_link_update,
1632 .stats_get = sfc_stats_get,
1633 .stats_reset = sfc_stats_reset,
1634 .xstats_get = sfc_xstats_get,
1635 .xstats_reset = sfc_stats_reset,
1636 .xstats_get_names = sfc_xstats_get_names,
1637 .dev_infos_get = sfc_dev_infos_get,
1638 .dev_supported_ptypes_get = sfc_dev_supported_ptypes_get,
1639 .mtu_set = sfc_dev_set_mtu,
1640 .rx_queue_start = sfc_rx_queue_start,
1641 .rx_queue_stop = sfc_rx_queue_stop,
1642 .tx_queue_start = sfc_tx_queue_start,
1643 .tx_queue_stop = sfc_tx_queue_stop,
1644 .rx_queue_setup = sfc_rx_queue_setup,
1645 .rx_queue_release = sfc_rx_queue_release,
1646 .rx_queue_count = sfc_rx_queue_count,
1647 .rx_descriptor_done = sfc_rx_descriptor_done,
1648 .rx_descriptor_status = sfc_rx_descriptor_status,
1649 .tx_descriptor_status = sfc_tx_descriptor_status,
1650 .tx_queue_setup = sfc_tx_queue_setup,
1651 .tx_queue_release = sfc_tx_queue_release,
1652 .flow_ctrl_get = sfc_flow_ctrl_get,
1653 .flow_ctrl_set = sfc_flow_ctrl_set,
1654 .mac_addr_set = sfc_mac_addr_set,
1655 .udp_tunnel_port_add = sfc_dev_udp_tunnel_port_add,
1656 .udp_tunnel_port_del = sfc_dev_udp_tunnel_port_del,
1657 #if EFSYS_OPT_RX_SCALE
1658 .reta_update = sfc_dev_rss_reta_update,
1659 .reta_query = sfc_dev_rss_reta_query,
1660 .rss_hash_update = sfc_dev_rss_hash_update,
1661 .rss_hash_conf_get = sfc_dev_rss_hash_conf_get,
1663 .filter_ctrl = sfc_dev_filter_ctrl,
1664 .set_mc_addr_list = sfc_set_mc_addr_list,
1665 .rxq_info_get = sfc_rx_queue_info_get,
1666 .txq_info_get = sfc_tx_queue_info_get,
1667 .fw_version_get = sfc_fw_version_get,
1668 .xstats_get_by_id = sfc_xstats_get_by_id,
1669 .xstats_get_names_by_id = sfc_xstats_get_names_by_id,
1673 * Duplicate a string in potentially shared memory required for
1674 * multi-process support.
1676 * strdup() allocates from process-local heap/memory.
1679 sfc_strdup(const char *str)
1687 size = strlen(str) + 1;
1688 copy = rte_malloc(__func__, size, 0);
1690 rte_memcpy(copy, str, size);
1696 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1698 struct sfc_adapter *sa = dev->data->dev_private;
1699 unsigned int avail_caps = 0;
1700 const char *rx_name = NULL;
1701 const char *tx_name = NULL;
1704 switch (sa->family) {
1705 case EFX_FAMILY_HUNTINGTON:
1706 case EFX_FAMILY_MEDFORD:
1707 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1713 rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1714 sfc_kvarg_string_handler, &rx_name);
1716 goto fail_kvarg_rx_datapath;
1718 if (rx_name != NULL) {
1719 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1720 if (sa->dp_rx == NULL) {
1721 sfc_err(sa, "Rx datapath %s not found", rx_name);
1725 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1727 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1730 goto fail_dp_rx_caps;
1733 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1734 if (sa->dp_rx == NULL) {
1735 sfc_err(sa, "Rx datapath by caps %#x not found",
1742 sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1743 if (sa->dp_rx_name == NULL) {
1745 goto fail_dp_rx_name;
1748 sfc_info(sa, "use %s Rx datapath", sa->dp_rx_name);
1750 dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1752 rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1753 sfc_kvarg_string_handler, &tx_name);
1755 goto fail_kvarg_tx_datapath;
1757 if (tx_name != NULL) {
1758 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1759 if (sa->dp_tx == NULL) {
1760 sfc_err(sa, "Tx datapath %s not found", tx_name);
1764 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1766 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1769 goto fail_dp_tx_caps;
1772 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1773 if (sa->dp_tx == NULL) {
1774 sfc_err(sa, "Tx datapath by caps %#x not found",
1781 sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1782 if (sa->dp_tx_name == NULL) {
1784 goto fail_dp_tx_name;
1787 sfc_info(sa, "use %s Tx datapath", sa->dp_tx_name);
1789 dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1791 dev->dev_ops = &sfc_eth_dev_ops;
1800 fail_kvarg_tx_datapath:
1801 rte_free(sa->dp_rx_name);
1802 sa->dp_rx_name = NULL;
1809 fail_kvarg_rx_datapath:
1814 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1816 struct sfc_adapter *sa = dev->data->dev_private;
1818 dev->dev_ops = NULL;
1819 dev->rx_pkt_burst = NULL;
1820 dev->tx_pkt_burst = NULL;
1822 rte_free(sa->dp_tx_name);
1823 sa->dp_tx_name = NULL;
1826 rte_free(sa->dp_rx_name);
1827 sa->dp_rx_name = NULL;
1831 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1832 .rxq_info_get = sfc_rx_queue_info_get,
1833 .txq_info_get = sfc_tx_queue_info_get,
1837 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1840 * Device private data has really many process-local pointers.
1841 * Below code should be extremely careful to use data located
1842 * in shared memory only.
1844 struct sfc_adapter *sa = dev->data->dev_private;
1845 const struct sfc_dp_rx *dp_rx;
1846 const struct sfc_dp_tx *dp_tx;
1849 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1850 if (dp_rx == NULL) {
1851 sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1855 if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1856 sfc_err(sa, "%s Rx datapath does not support multi-process",
1859 goto fail_dp_rx_multi_process;
1862 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1863 if (dp_tx == NULL) {
1864 sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1868 if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1869 sfc_err(sa, "%s Tx datapath does not support multi-process",
1872 goto fail_dp_tx_multi_process;
1875 dev->rx_pkt_burst = dp_rx->pkt_burst;
1876 dev->tx_pkt_burst = dp_tx->pkt_burst;
1877 dev->dev_ops = &sfc_eth_dev_secondary_ops;
1881 fail_dp_tx_multi_process:
1883 fail_dp_rx_multi_process:
1889 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1891 dev->dev_ops = NULL;
1892 dev->tx_pkt_burst = NULL;
1893 dev->rx_pkt_burst = NULL;
1897 sfc_register_dp(void)
1900 if (TAILQ_EMPTY(&sfc_dp_head)) {
1901 /* Prefer EF10 datapath */
1902 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1903 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1905 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1906 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1907 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1912 sfc_eth_dev_init(struct rte_eth_dev *dev)
1914 struct sfc_adapter *sa = dev->data->dev_private;
1915 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1917 const efx_nic_cfg_t *encp;
1918 const struct ether_addr *from;
1922 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1923 return -sfc_eth_dev_secondary_set_ops(dev);
1925 /* Required for logging */
1926 sa->pci_addr = pci_dev->addr;
1927 sa->port_id = dev->data->port_id;
1931 /* Copy PCI device info to the dev->data */
1932 rte_eth_copy_pci_info(dev, pci_dev);
1934 rc = sfc_kvargs_parse(sa);
1936 goto fail_kvargs_parse;
1938 rc = sfc_kvargs_process(sa, SFC_KVARG_DEBUG_INIT,
1939 sfc_kvarg_bool_handler, &sa->debug_init);
1941 goto fail_kvarg_debug_init;
1943 sfc_log_init(sa, "entry");
1945 dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1946 if (dev->data->mac_addrs == NULL) {
1948 goto fail_mac_addrs;
1951 sfc_adapter_lock_init(sa);
1952 sfc_adapter_lock(sa);
1954 sfc_log_init(sa, "probing");
1959 sfc_log_init(sa, "set device ops");
1960 rc = sfc_eth_dev_set_ops(dev);
1964 sfc_log_init(sa, "attaching");
1965 rc = sfc_attach(sa);
1969 encp = efx_nic_cfg_get(sa->nic);
1972 * The arguments are really reverse order in comparison to
1973 * Linux kernel. Copy from NIC config to Ethernet device data.
1975 from = (const struct ether_addr *)(encp->enc_mac_addr);
1976 ether_addr_copy(from, &dev->data->mac_addrs[0]);
1978 sfc_adapter_unlock(sa);
1980 sfc_log_init(sa, "done");
1984 sfc_eth_dev_clear_ops(dev);
1990 sfc_adapter_unlock(sa);
1991 sfc_adapter_lock_fini(sa);
1992 rte_free(dev->data->mac_addrs);
1993 dev->data->mac_addrs = NULL;
1996 fail_kvarg_debug_init:
1997 sfc_kvargs_cleanup(sa);
2000 sfc_log_init(sa, "failed %d", rc);
2006 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2008 struct sfc_adapter *sa;
2010 if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2011 sfc_eth_dev_secondary_clear_ops(dev);
2015 sa = dev->data->dev_private;
2016 sfc_log_init(sa, "entry");
2018 sfc_adapter_lock(sa);
2020 sfc_eth_dev_clear_ops(dev);
2025 rte_free(dev->data->mac_addrs);
2026 dev->data->mac_addrs = NULL;
2028 sfc_kvargs_cleanup(sa);
2030 sfc_adapter_unlock(sa);
2031 sfc_adapter_lock_fini(sa);
2033 sfc_log_init(sa, "done");
2035 /* Required for logging, so cleanup last */
2040 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2041 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2042 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2043 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2044 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2045 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2046 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2047 { .vendor_id = 0 /* sentinel */ }
2050 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2051 struct rte_pci_device *pci_dev)
2053 return rte_eth_dev_pci_generic_probe(pci_dev,
2054 sizeof(struct sfc_adapter), sfc_eth_dev_init);
2057 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2059 return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2062 static struct rte_pci_driver sfc_efx_pmd = {
2063 .id_table = pci_id_sfc_efx_map,
2065 RTE_PCI_DRV_INTR_LSC |
2066 RTE_PCI_DRV_NEED_MAPPING,
2067 .probe = sfc_eth_dev_pci_probe,
2068 .remove = sfc_eth_dev_pci_remove,
2071 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2072 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2073 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2074 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2075 SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2076 SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2077 SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2078 SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long> "
2079 SFC_KVARG_MCDI_LOGGING "=" SFC_KVARG_VALUES_BOOL " "
2080 SFC_KVARG_DEBUG_INIT "=" SFC_KVARG_VALUES_BOOL);