xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision 5dba3b9c4c131b88a78bcecfef39db23ebc47873)
1 /*-
2  *   BSD LICENSE
3  *
4  * Copyright (c) 2016-2017 Solarflare Communications Inc.
5  * All rights reserved.
6  *
7  * This software was jointly developed between OKTET Labs (under contract
8  * for Solarflare) and Solarflare Communications, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions are met:
12  *
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.
18  *
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.
30  */
31 
32 #include <rte_dev.h>
33 #include <rte_ethdev.h>
34 #include <rte_ethdev_pci.h>
35 #include <rte_pci.h>
36 #include <rte_bus_pci.h>
37 #include <rte_errno.h>
38 
39 #include "efx.h"
40 
41 #include "sfc.h"
42 #include "sfc_debug.h"
43 #include "sfc_log.h"
44 #include "sfc_kvargs.h"
45 #include "sfc_ev.h"
46 #include "sfc_rx.h"
47 #include "sfc_tx.h"
48 #include "sfc_flow.h"
49 #include "sfc_dp.h"
50 #include "sfc_dp_rx.h"
51 
52 static struct sfc_dp_list sfc_dp_head =
53 	TAILQ_HEAD_INITIALIZER(sfc_dp_head);
54 
55 static int
56 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
57 {
58 	struct sfc_adapter *sa = dev->data->dev_private;
59 	efx_nic_fw_info_t enfi;
60 	int ret;
61 	int rc;
62 
63 	/*
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
67 	 */
68 	if ((fw_version == NULL) || (fw_size == 0))
69 		return -EINVAL;
70 
71 	rc = efx_nic_get_fw_version(sa->nic, &enfi);
72 	if (rc != 0)
73 		return -rc;
74 
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]);
79 	if (ret < 0)
80 		return ret;
81 
82 	if (enfi.enfi_dpcpu_fw_ids_valid) {
83 		size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
84 		int ret_extra;
85 
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);
91 		if (ret_extra < 0)
92 			return ret_extra;
93 
94 		ret += ret_extra;
95 	}
96 
97 	if (fw_size < (size_t)(++ret))
98 		return ret;
99 	else
100 		return 0;
101 }
102 
103 static void
104 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
105 {
106 	struct sfc_adapter *sa = dev->data->dev_private;
107 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
108 
109 	sfc_log_init(sa, "entry");
110 
111 	dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
112 	dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
113 
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;
122 
123 	dev_info->max_rx_queues = sa->rxq_max;
124 	dev_info->max_tx_queues = sa->txq_max;
125 
126 	/* By default packets are dropped if no descriptors are available */
127 	dev_info->default_rxconf.rx_drop_en = 1;
128 
129 	dev_info->rx_offload_capa =
130 		DEV_RX_OFFLOAD_IPV4_CKSUM |
131 		DEV_RX_OFFLOAD_UDP_CKSUM |
132 		DEV_RX_OFFLOAD_TCP_CKSUM;
133 
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;
137 
138 	dev_info->tx_offload_capa =
139 		DEV_TX_OFFLOAD_IPV4_CKSUM |
140 		DEV_TX_OFFLOAD_UDP_CKSUM |
141 		DEV_TX_OFFLOAD_TCP_CKSUM;
142 
143 	if (encp->enc_tunnel_encapsulations_supported != 0)
144 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM;
145 
146 	dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
147 	if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
148 	    !encp->enc_hw_tx_insert_vlan_enabled)
149 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
150 	else
151 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_VLAN_INSERT;
152 
153 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
154 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
155 
156 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_POOL)
157 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTMEMP;
158 
159 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_REFCNT)
160 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOREFCOUNT;
161 
162 #if EFSYS_OPT_RX_SCALE
163 	if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
164 		dev_info->reta_size = EFX_RSS_TBL_SIZE;
165 		dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
166 		dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
167 	}
168 #endif
169 
170 	if (sa->tso)
171 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
172 
173 	dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
174 	dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
175 	/* The RXQ hardware requires that the descriptor count is a power
176 	 * of 2, but rx_desc_lim cannot properly describe that constraint.
177 	 */
178 	dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
179 
180 	dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
181 	dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
182 	/*
183 	 * The TXQ hardware requires that the descriptor count is a power
184 	 * of 2, but tx_desc_lim cannot properly describe that constraint
185 	 */
186 	dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
187 }
188 
189 static const uint32_t *
190 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
191 {
192 	struct sfc_adapter *sa = dev->data->dev_private;
193 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
194 	uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
195 
196 	return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
197 }
198 
199 static int
200 sfc_dev_configure(struct rte_eth_dev *dev)
201 {
202 	struct rte_eth_dev_data *dev_data = dev->data;
203 	struct sfc_adapter *sa = dev_data->dev_private;
204 	int rc;
205 
206 	sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
207 		     dev_data->nb_rx_queues, dev_data->nb_tx_queues);
208 
209 	sfc_adapter_lock(sa);
210 	switch (sa->state) {
211 	case SFC_ADAPTER_CONFIGURED:
212 		/* FALLTHROUGH */
213 	case SFC_ADAPTER_INITIALIZED:
214 		rc = sfc_configure(sa);
215 		break;
216 	default:
217 		sfc_err(sa, "unexpected adapter state %u to configure",
218 			sa->state);
219 		rc = EINVAL;
220 		break;
221 	}
222 	sfc_adapter_unlock(sa);
223 
224 	sfc_log_init(sa, "done %d", rc);
225 	SFC_ASSERT(rc >= 0);
226 	return -rc;
227 }
228 
229 static int
230 sfc_dev_start(struct rte_eth_dev *dev)
231 {
232 	struct sfc_adapter *sa = dev->data->dev_private;
233 	int rc;
234 
235 	sfc_log_init(sa, "entry");
236 
237 	sfc_adapter_lock(sa);
238 	rc = sfc_start(sa);
239 	sfc_adapter_unlock(sa);
240 
241 	sfc_log_init(sa, "done %d", rc);
242 	SFC_ASSERT(rc >= 0);
243 	return -rc;
244 }
245 
246 static int
247 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
248 {
249 	struct sfc_adapter *sa = dev->data->dev_private;
250 	struct rte_eth_link *dev_link = &dev->data->dev_link;
251 	struct rte_eth_link old_link;
252 	struct rte_eth_link current_link;
253 
254 	sfc_log_init(sa, "entry");
255 
256 retry:
257 	EFX_STATIC_ASSERT(sizeof(*dev_link) == sizeof(rte_atomic64_t));
258 	*(int64_t *)&old_link = rte_atomic64_read((rte_atomic64_t *)dev_link);
259 
260 	if (sa->state != SFC_ADAPTER_STARTED) {
261 		sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
262 		if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
263 					 *(uint64_t *)&old_link,
264 					 *(uint64_t *)&current_link))
265 			goto retry;
266 	} else if (wait_to_complete) {
267 		efx_link_mode_t link_mode;
268 
269 		if (efx_port_poll(sa->nic, &link_mode) != 0)
270 			link_mode = EFX_LINK_UNKNOWN;
271 		sfc_port_link_mode_to_info(link_mode, &current_link);
272 
273 		if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
274 					 *(uint64_t *)&old_link,
275 					 *(uint64_t *)&current_link))
276 			goto retry;
277 	} else {
278 		sfc_ev_mgmt_qpoll(sa);
279 		*(int64_t *)&current_link =
280 			rte_atomic64_read((rte_atomic64_t *)dev_link);
281 	}
282 
283 	if (old_link.link_status != current_link.link_status)
284 		sfc_info(sa, "Link status is %s",
285 			 current_link.link_status ? "UP" : "DOWN");
286 
287 	return old_link.link_status == current_link.link_status ? 0 : -1;
288 }
289 
290 static void
291 sfc_dev_stop(struct rte_eth_dev *dev)
292 {
293 	struct sfc_adapter *sa = dev->data->dev_private;
294 
295 	sfc_log_init(sa, "entry");
296 
297 	sfc_adapter_lock(sa);
298 	sfc_stop(sa);
299 	sfc_adapter_unlock(sa);
300 
301 	sfc_log_init(sa, "done");
302 }
303 
304 static int
305 sfc_dev_set_link_up(struct rte_eth_dev *dev)
306 {
307 	struct sfc_adapter *sa = dev->data->dev_private;
308 	int rc;
309 
310 	sfc_log_init(sa, "entry");
311 
312 	sfc_adapter_lock(sa);
313 	rc = sfc_start(sa);
314 	sfc_adapter_unlock(sa);
315 
316 	SFC_ASSERT(rc >= 0);
317 	return -rc;
318 }
319 
320 static int
321 sfc_dev_set_link_down(struct rte_eth_dev *dev)
322 {
323 	struct sfc_adapter *sa = dev->data->dev_private;
324 
325 	sfc_log_init(sa, "entry");
326 
327 	sfc_adapter_lock(sa);
328 	sfc_stop(sa);
329 	sfc_adapter_unlock(sa);
330 
331 	return 0;
332 }
333 
334 static void
335 sfc_dev_close(struct rte_eth_dev *dev)
336 {
337 	struct sfc_adapter *sa = dev->data->dev_private;
338 
339 	sfc_log_init(sa, "entry");
340 
341 	sfc_adapter_lock(sa);
342 	switch (sa->state) {
343 	case SFC_ADAPTER_STARTED:
344 		sfc_stop(sa);
345 		SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
346 		/* FALLTHROUGH */
347 	case SFC_ADAPTER_CONFIGURED:
348 		sfc_close(sa);
349 		SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
350 		/* FALLTHROUGH */
351 	case SFC_ADAPTER_INITIALIZED:
352 		break;
353 	default:
354 		sfc_err(sa, "unexpected adapter state %u on close", sa->state);
355 		break;
356 	}
357 	sfc_adapter_unlock(sa);
358 
359 	sfc_log_init(sa, "done");
360 }
361 
362 static void
363 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
364 		   boolean_t enabled)
365 {
366 	struct sfc_port *port;
367 	boolean_t *toggle;
368 	struct sfc_adapter *sa = dev->data->dev_private;
369 	boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
370 	const char *desc = (allmulti) ? "all-multi" : "promiscuous";
371 
372 	sfc_adapter_lock(sa);
373 
374 	port = &sa->port;
375 	toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
376 
377 	if (*toggle != enabled) {
378 		*toggle = enabled;
379 
380 		if (port->isolated) {
381 			sfc_warn(sa, "isolated mode is active on the port");
382 			sfc_warn(sa, "the change is to be applied on the next "
383 				     "start provided that isolated mode is "
384 				     "disabled prior the next start");
385 		} else if ((sa->state == SFC_ADAPTER_STARTED) &&
386 			   (sfc_set_rx_mode(sa) != 0)) {
387 			*toggle = !(enabled);
388 			sfc_warn(sa, "Failed to %s %s mode",
389 				 ((enabled) ? "enable" : "disable"), desc);
390 		}
391 	}
392 
393 	sfc_adapter_unlock(sa);
394 }
395 
396 static void
397 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
398 {
399 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
400 }
401 
402 static void
403 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
404 {
405 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
406 }
407 
408 static void
409 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
410 {
411 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
412 }
413 
414 static void
415 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
416 {
417 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
418 }
419 
420 static int
421 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
422 		   uint16_t nb_rx_desc, unsigned int socket_id,
423 		   const struct rte_eth_rxconf *rx_conf,
424 		   struct rte_mempool *mb_pool)
425 {
426 	struct sfc_adapter *sa = dev->data->dev_private;
427 	int rc;
428 
429 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
430 		     rx_queue_id, nb_rx_desc, socket_id);
431 
432 	sfc_adapter_lock(sa);
433 
434 	rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
435 			  rx_conf, mb_pool);
436 	if (rc != 0)
437 		goto fail_rx_qinit;
438 
439 	dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
440 
441 	sfc_adapter_unlock(sa);
442 
443 	return 0;
444 
445 fail_rx_qinit:
446 	sfc_adapter_unlock(sa);
447 	SFC_ASSERT(rc > 0);
448 	return -rc;
449 }
450 
451 static void
452 sfc_rx_queue_release(void *queue)
453 {
454 	struct sfc_dp_rxq *dp_rxq = queue;
455 	struct sfc_rxq *rxq;
456 	struct sfc_adapter *sa;
457 	unsigned int sw_index;
458 
459 	if (dp_rxq == NULL)
460 		return;
461 
462 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
463 	sa = rxq->evq->sa;
464 	sfc_adapter_lock(sa);
465 
466 	sw_index = sfc_rxq_sw_index(rxq);
467 
468 	sfc_log_init(sa, "RxQ=%u", sw_index);
469 
470 	sa->eth_dev->data->rx_queues[sw_index] = NULL;
471 
472 	sfc_rx_qfini(sa, sw_index);
473 
474 	sfc_adapter_unlock(sa);
475 }
476 
477 static int
478 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
479 		   uint16_t nb_tx_desc, unsigned int socket_id,
480 		   const struct rte_eth_txconf *tx_conf)
481 {
482 	struct sfc_adapter *sa = dev->data->dev_private;
483 	int rc;
484 
485 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
486 		     tx_queue_id, nb_tx_desc, socket_id);
487 
488 	sfc_adapter_lock(sa);
489 
490 	rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
491 	if (rc != 0)
492 		goto fail_tx_qinit;
493 
494 	dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
495 
496 	sfc_adapter_unlock(sa);
497 	return 0;
498 
499 fail_tx_qinit:
500 	sfc_adapter_unlock(sa);
501 	SFC_ASSERT(rc > 0);
502 	return -rc;
503 }
504 
505 static void
506 sfc_tx_queue_release(void *queue)
507 {
508 	struct sfc_dp_txq *dp_txq = queue;
509 	struct sfc_txq *txq;
510 	unsigned int sw_index;
511 	struct sfc_adapter *sa;
512 
513 	if (dp_txq == NULL)
514 		return;
515 
516 	txq = sfc_txq_by_dp_txq(dp_txq);
517 	sw_index = sfc_txq_sw_index(txq);
518 
519 	SFC_ASSERT(txq->evq != NULL);
520 	sa = txq->evq->sa;
521 
522 	sfc_log_init(sa, "TxQ = %u", sw_index);
523 
524 	sfc_adapter_lock(sa);
525 
526 	SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
527 	sa->eth_dev->data->tx_queues[sw_index] = NULL;
528 
529 	sfc_tx_qfini(sa, sw_index);
530 
531 	sfc_adapter_unlock(sa);
532 }
533 
534 static int
535 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
536 {
537 	struct sfc_adapter *sa = dev->data->dev_private;
538 	struct sfc_port *port = &sa->port;
539 	uint64_t *mac_stats;
540 	int ret;
541 
542 	rte_spinlock_lock(&port->mac_stats_lock);
543 
544 	ret = sfc_port_update_mac_stats(sa);
545 	if (ret != 0)
546 		goto unlock;
547 
548 	mac_stats = port->mac_stats_buf;
549 
550 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
551 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
552 		stats->ipackets =
553 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
554 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
555 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
556 		stats->opackets =
557 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
558 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
559 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
560 		stats->ibytes =
561 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
562 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
563 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
564 		stats->obytes =
565 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
566 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
567 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
568 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
569 		stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
570 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
571 	} else {
572 		stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
573 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
574 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
575 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
576 		/*
577 		 * Take into account stats which are whenever supported
578 		 * on EF10. If some stat is not supported by current
579 		 * firmware variant or HW revision, it is guaranteed
580 		 * to be zero in mac_stats.
581 		 */
582 		stats->imissed =
583 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
584 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
585 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
586 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
587 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
588 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
589 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
590 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
591 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
592 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
593 		stats->ierrors =
594 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
595 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
596 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
597 		/* no oerrors counters supported on EF10 */
598 	}
599 
600 unlock:
601 	rte_spinlock_unlock(&port->mac_stats_lock);
602 	SFC_ASSERT(ret >= 0);
603 	return -ret;
604 }
605 
606 static void
607 sfc_stats_reset(struct rte_eth_dev *dev)
608 {
609 	struct sfc_adapter *sa = dev->data->dev_private;
610 	struct sfc_port *port = &sa->port;
611 	int rc;
612 
613 	if (sa->state != SFC_ADAPTER_STARTED) {
614 		/*
615 		 * The operation cannot be done if port is not started; it
616 		 * will be scheduled to be done during the next port start
617 		 */
618 		port->mac_stats_reset_pending = B_TRUE;
619 		return;
620 	}
621 
622 	rc = sfc_port_reset_mac_stats(sa);
623 	if (rc != 0)
624 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
625 }
626 
627 static int
628 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
629 	       unsigned int xstats_count)
630 {
631 	struct sfc_adapter *sa = dev->data->dev_private;
632 	struct sfc_port *port = &sa->port;
633 	uint64_t *mac_stats;
634 	int rc;
635 	unsigned int i;
636 	int nstats = 0;
637 
638 	rte_spinlock_lock(&port->mac_stats_lock);
639 
640 	rc = sfc_port_update_mac_stats(sa);
641 	if (rc != 0) {
642 		SFC_ASSERT(rc > 0);
643 		nstats = -rc;
644 		goto unlock;
645 	}
646 
647 	mac_stats = port->mac_stats_buf;
648 
649 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
650 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
651 			if (xstats != NULL && nstats < (int)xstats_count) {
652 				xstats[nstats].id = nstats;
653 				xstats[nstats].value = mac_stats[i];
654 			}
655 			nstats++;
656 		}
657 	}
658 
659 unlock:
660 	rte_spinlock_unlock(&port->mac_stats_lock);
661 
662 	return nstats;
663 }
664 
665 static int
666 sfc_xstats_get_names(struct rte_eth_dev *dev,
667 		     struct rte_eth_xstat_name *xstats_names,
668 		     unsigned int xstats_count)
669 {
670 	struct sfc_adapter *sa = dev->data->dev_private;
671 	struct sfc_port *port = &sa->port;
672 	unsigned int i;
673 	unsigned int nstats = 0;
674 
675 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
676 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
677 			if (xstats_names != NULL && nstats < xstats_count)
678 				strncpy(xstats_names[nstats].name,
679 					efx_mac_stat_name(sa->nic, i),
680 					sizeof(xstats_names[0].name));
681 			nstats++;
682 		}
683 	}
684 
685 	return nstats;
686 }
687 
688 static int
689 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
690 		     uint64_t *values, unsigned int n)
691 {
692 	struct sfc_adapter *sa = dev->data->dev_private;
693 	struct sfc_port *port = &sa->port;
694 	uint64_t *mac_stats;
695 	unsigned int nb_supported = 0;
696 	unsigned int nb_written = 0;
697 	unsigned int i;
698 	int ret;
699 	int rc;
700 
701 	if (unlikely(values == NULL) ||
702 	    unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
703 		return port->mac_stats_nb_supported;
704 
705 	rte_spinlock_lock(&port->mac_stats_lock);
706 
707 	rc = sfc_port_update_mac_stats(sa);
708 	if (rc != 0) {
709 		SFC_ASSERT(rc > 0);
710 		ret = -rc;
711 		goto unlock;
712 	}
713 
714 	mac_stats = port->mac_stats_buf;
715 
716 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
717 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
718 			continue;
719 
720 		if ((ids == NULL) || (ids[nb_written] == nb_supported))
721 			values[nb_written++] = mac_stats[i];
722 
723 		++nb_supported;
724 	}
725 
726 	ret = nb_written;
727 
728 unlock:
729 	rte_spinlock_unlock(&port->mac_stats_lock);
730 
731 	return ret;
732 }
733 
734 static int
735 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
736 			   struct rte_eth_xstat_name *xstats_names,
737 			   const uint64_t *ids, unsigned int size)
738 {
739 	struct sfc_adapter *sa = dev->data->dev_private;
740 	struct sfc_port *port = &sa->port;
741 	unsigned int nb_supported = 0;
742 	unsigned int nb_written = 0;
743 	unsigned int i;
744 
745 	if (unlikely(xstats_names == NULL) ||
746 	    unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
747 		return port->mac_stats_nb_supported;
748 
749 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
750 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
751 			continue;
752 
753 		if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
754 			char *name = xstats_names[nb_written++].name;
755 
756 			strncpy(name, efx_mac_stat_name(sa->nic, i),
757 				sizeof(xstats_names[0].name));
758 			name[sizeof(xstats_names[0].name) - 1] = '\0';
759 		}
760 
761 		++nb_supported;
762 	}
763 
764 	return nb_written;
765 }
766 
767 static int
768 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
769 {
770 	struct sfc_adapter *sa = dev->data->dev_private;
771 	unsigned int wanted_fc, link_fc;
772 
773 	memset(fc_conf, 0, sizeof(*fc_conf));
774 
775 	sfc_adapter_lock(sa);
776 
777 	if (sa->state == SFC_ADAPTER_STARTED)
778 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
779 	else
780 		link_fc = sa->port.flow_ctrl;
781 
782 	switch (link_fc) {
783 	case 0:
784 		fc_conf->mode = RTE_FC_NONE;
785 		break;
786 	case EFX_FCNTL_RESPOND:
787 		fc_conf->mode = RTE_FC_RX_PAUSE;
788 		break;
789 	case EFX_FCNTL_GENERATE:
790 		fc_conf->mode = RTE_FC_TX_PAUSE;
791 		break;
792 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
793 		fc_conf->mode = RTE_FC_FULL;
794 		break;
795 	default:
796 		sfc_err(sa, "%s: unexpected flow control value %#x",
797 			__func__, link_fc);
798 	}
799 
800 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
801 
802 	sfc_adapter_unlock(sa);
803 
804 	return 0;
805 }
806 
807 static int
808 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
809 {
810 	struct sfc_adapter *sa = dev->data->dev_private;
811 	struct sfc_port *port = &sa->port;
812 	unsigned int fcntl;
813 	int rc;
814 
815 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
816 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
817 	    fc_conf->mac_ctrl_frame_fwd != 0) {
818 		sfc_err(sa, "unsupported flow control settings specified");
819 		rc = EINVAL;
820 		goto fail_inval;
821 	}
822 
823 	switch (fc_conf->mode) {
824 	case RTE_FC_NONE:
825 		fcntl = 0;
826 		break;
827 	case RTE_FC_RX_PAUSE:
828 		fcntl = EFX_FCNTL_RESPOND;
829 		break;
830 	case RTE_FC_TX_PAUSE:
831 		fcntl = EFX_FCNTL_GENERATE;
832 		break;
833 	case RTE_FC_FULL:
834 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
835 		break;
836 	default:
837 		rc = EINVAL;
838 		goto fail_inval;
839 	}
840 
841 	sfc_adapter_lock(sa);
842 
843 	if (sa->state == SFC_ADAPTER_STARTED) {
844 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
845 		if (rc != 0)
846 			goto fail_mac_fcntl_set;
847 	}
848 
849 	port->flow_ctrl = fcntl;
850 	port->flow_ctrl_autoneg = fc_conf->autoneg;
851 
852 	sfc_adapter_unlock(sa);
853 
854 	return 0;
855 
856 fail_mac_fcntl_set:
857 	sfc_adapter_unlock(sa);
858 fail_inval:
859 	SFC_ASSERT(rc > 0);
860 	return -rc;
861 }
862 
863 static int
864 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
865 {
866 	struct sfc_adapter *sa = dev->data->dev_private;
867 	size_t pdu = EFX_MAC_PDU(mtu);
868 	size_t old_pdu;
869 	int rc;
870 
871 	sfc_log_init(sa, "mtu=%u", mtu);
872 
873 	rc = EINVAL;
874 	if (pdu < EFX_MAC_PDU_MIN) {
875 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
876 			(unsigned int)mtu, (unsigned int)pdu,
877 			EFX_MAC_PDU_MIN);
878 		goto fail_inval;
879 	}
880 	if (pdu > EFX_MAC_PDU_MAX) {
881 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
882 			(unsigned int)mtu, (unsigned int)pdu,
883 			EFX_MAC_PDU_MAX);
884 		goto fail_inval;
885 	}
886 
887 	sfc_adapter_lock(sa);
888 
889 	if (pdu != sa->port.pdu) {
890 		if (sa->state == SFC_ADAPTER_STARTED) {
891 			sfc_stop(sa);
892 
893 			old_pdu = sa->port.pdu;
894 			sa->port.pdu = pdu;
895 			rc = sfc_start(sa);
896 			if (rc != 0)
897 				goto fail_start;
898 		} else {
899 			sa->port.pdu = pdu;
900 		}
901 	}
902 
903 	/*
904 	 * The driver does not use it, but other PMDs update jumbo_frame
905 	 * flag and max_rx_pkt_len when MTU is set.
906 	 */
907 	dev->data->dev_conf.rxmode.jumbo_frame = (mtu > ETHER_MAX_LEN);
908 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
909 
910 	sfc_adapter_unlock(sa);
911 
912 	sfc_log_init(sa, "done");
913 	return 0;
914 
915 fail_start:
916 	sa->port.pdu = old_pdu;
917 	if (sfc_start(sa) != 0)
918 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
919 			"PDU max size - port is stopped",
920 			(unsigned int)pdu, (unsigned int)old_pdu);
921 	sfc_adapter_unlock(sa);
922 
923 fail_inval:
924 	sfc_log_init(sa, "failed %d", rc);
925 	SFC_ASSERT(rc > 0);
926 	return -rc;
927 }
928 static void
929 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
930 {
931 	struct sfc_adapter *sa = dev->data->dev_private;
932 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
933 	struct sfc_port *port = &sa->port;
934 	int rc;
935 
936 	sfc_adapter_lock(sa);
937 
938 	/*
939 	 * Copy the address to the device private data so that
940 	 * it could be recalled in the case of adapter restart.
941 	 */
942 	ether_addr_copy(mac_addr, &port->default_mac_addr);
943 
944 	if (port->isolated) {
945 		sfc_err(sa, "isolated mode is active on the port");
946 		sfc_err(sa, "will not set MAC address");
947 		goto unlock;
948 	}
949 
950 	if (sa->state != SFC_ADAPTER_STARTED) {
951 		sfc_info(sa, "the port is not started");
952 		sfc_info(sa, "the new MAC address will be set on port start");
953 
954 		goto unlock;
955 	}
956 
957 	if (encp->enc_allow_set_mac_with_installed_filters) {
958 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
959 		if (rc != 0) {
960 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
961 			goto unlock;
962 		}
963 
964 		/*
965 		 * Changing the MAC address by means of MCDI request
966 		 * has no effect on received traffic, therefore
967 		 * we also need to update unicast filters
968 		 */
969 		rc = sfc_set_rx_mode(sa);
970 		if (rc != 0)
971 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
972 	} else {
973 		sfc_warn(sa, "cannot set MAC address with filters installed");
974 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
975 		sfc_warn(sa, "(some traffic may be dropped)");
976 
977 		/*
978 		 * Since setting MAC address with filters installed is not
979 		 * allowed on the adapter, the new MAC address will be set
980 		 * by means of adapter restart. sfc_start() shall retrieve
981 		 * the new address from the device private data and set it.
982 		 */
983 		sfc_stop(sa);
984 		rc = sfc_start(sa);
985 		if (rc != 0)
986 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
987 	}
988 
989 unlock:
990 	/*
991 	 * In the case of failure sa->port->default_mac_addr does not
992 	 * need rollback since no error code is returned, and the upper
993 	 * API will anyway update the external MAC address storage.
994 	 * To be consistent with that new value it is better to keep
995 	 * the device private value the same.
996 	 */
997 	sfc_adapter_unlock(sa);
998 }
999 
1000 
1001 static int
1002 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
1003 		     uint32_t nb_mc_addr)
1004 {
1005 	struct sfc_adapter *sa = dev->data->dev_private;
1006 	struct sfc_port *port = &sa->port;
1007 	uint8_t *mc_addrs = port->mcast_addrs;
1008 	int rc;
1009 	unsigned int i;
1010 
1011 	if (port->isolated) {
1012 		sfc_err(sa, "isolated mode is active on the port");
1013 		sfc_err(sa, "will not set multicast address list");
1014 		return -ENOTSUP;
1015 	}
1016 
1017 	if (mc_addrs == NULL)
1018 		return -ENOBUFS;
1019 
1020 	if (nb_mc_addr > port->max_mcast_addrs) {
1021 		sfc_err(sa, "too many multicast addresses: %u > %u",
1022 			 nb_mc_addr, port->max_mcast_addrs);
1023 		return -EINVAL;
1024 	}
1025 
1026 	for (i = 0; i < nb_mc_addr; ++i) {
1027 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1028 				 EFX_MAC_ADDR_LEN);
1029 		mc_addrs += EFX_MAC_ADDR_LEN;
1030 	}
1031 
1032 	port->nb_mcast_addrs = nb_mc_addr;
1033 
1034 	if (sa->state != SFC_ADAPTER_STARTED)
1035 		return 0;
1036 
1037 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1038 					port->nb_mcast_addrs);
1039 	if (rc != 0)
1040 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1041 
1042 	SFC_ASSERT(rc > 0);
1043 	return -rc;
1044 }
1045 
1046 /*
1047  * The function is used by the secondary process as well. It must not
1048  * use any process-local pointers from the adapter data.
1049  */
1050 static void
1051 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1052 		      struct rte_eth_rxq_info *qinfo)
1053 {
1054 	struct sfc_adapter *sa = dev->data->dev_private;
1055 	struct sfc_rxq_info *rxq_info;
1056 	struct sfc_rxq *rxq;
1057 
1058 	sfc_adapter_lock(sa);
1059 
1060 	SFC_ASSERT(rx_queue_id < sa->rxq_count);
1061 
1062 	rxq_info = &sa->rxq_info[rx_queue_id];
1063 	rxq = rxq_info->rxq;
1064 	SFC_ASSERT(rxq != NULL);
1065 
1066 	qinfo->mp = rxq->refill_mb_pool;
1067 	qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1068 	qinfo->conf.rx_drop_en = 1;
1069 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1070 	qinfo->scattered_rx =
1071 		((rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) != 0);
1072 	qinfo->nb_desc = rxq_info->entries;
1073 
1074 	sfc_adapter_unlock(sa);
1075 }
1076 
1077 /*
1078  * The function is used by the secondary process as well. It must not
1079  * use any process-local pointers from the adapter data.
1080  */
1081 static void
1082 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1083 		      struct rte_eth_txq_info *qinfo)
1084 {
1085 	struct sfc_adapter *sa = dev->data->dev_private;
1086 	struct sfc_txq_info *txq_info;
1087 
1088 	sfc_adapter_lock(sa);
1089 
1090 	SFC_ASSERT(tx_queue_id < sa->txq_count);
1091 
1092 	txq_info = &sa->txq_info[tx_queue_id];
1093 	SFC_ASSERT(txq_info->txq != NULL);
1094 
1095 	memset(qinfo, 0, sizeof(*qinfo));
1096 
1097 	qinfo->conf.txq_flags = txq_info->txq->flags;
1098 	qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1099 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1100 	qinfo->nb_desc = txq_info->entries;
1101 
1102 	sfc_adapter_unlock(sa);
1103 }
1104 
1105 static uint32_t
1106 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1107 {
1108 	struct sfc_adapter *sa = dev->data->dev_private;
1109 
1110 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1111 
1112 	return sfc_rx_qdesc_npending(sa, rx_queue_id);
1113 }
1114 
1115 static int
1116 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1117 {
1118 	struct sfc_dp_rxq *dp_rxq = queue;
1119 
1120 	return sfc_rx_qdesc_done(dp_rxq, offset);
1121 }
1122 
1123 static int
1124 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1125 {
1126 	struct sfc_dp_rxq *dp_rxq = queue;
1127 	struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1128 
1129 	return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1130 }
1131 
1132 static int
1133 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1134 {
1135 	struct sfc_dp_txq *dp_txq = queue;
1136 	struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1137 
1138 	return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1139 }
1140 
1141 static int
1142 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1143 {
1144 	struct sfc_adapter *sa = dev->data->dev_private;
1145 	int rc;
1146 
1147 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1148 
1149 	sfc_adapter_lock(sa);
1150 
1151 	rc = EINVAL;
1152 	if (sa->state != SFC_ADAPTER_STARTED)
1153 		goto fail_not_started;
1154 
1155 	rc = sfc_rx_qstart(sa, rx_queue_id);
1156 	if (rc != 0)
1157 		goto fail_rx_qstart;
1158 
1159 	sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1160 
1161 	sfc_adapter_unlock(sa);
1162 
1163 	return 0;
1164 
1165 fail_rx_qstart:
1166 fail_not_started:
1167 	sfc_adapter_unlock(sa);
1168 	SFC_ASSERT(rc > 0);
1169 	return -rc;
1170 }
1171 
1172 static int
1173 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1174 {
1175 	struct sfc_adapter *sa = dev->data->dev_private;
1176 
1177 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1178 
1179 	sfc_adapter_lock(sa);
1180 	sfc_rx_qstop(sa, rx_queue_id);
1181 
1182 	sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1183 
1184 	sfc_adapter_unlock(sa);
1185 
1186 	return 0;
1187 }
1188 
1189 static int
1190 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1191 {
1192 	struct sfc_adapter *sa = dev->data->dev_private;
1193 	int rc;
1194 
1195 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1196 
1197 	sfc_adapter_lock(sa);
1198 
1199 	rc = EINVAL;
1200 	if (sa->state != SFC_ADAPTER_STARTED)
1201 		goto fail_not_started;
1202 
1203 	rc = sfc_tx_qstart(sa, tx_queue_id);
1204 	if (rc != 0)
1205 		goto fail_tx_qstart;
1206 
1207 	sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1208 
1209 	sfc_adapter_unlock(sa);
1210 	return 0;
1211 
1212 fail_tx_qstart:
1213 
1214 fail_not_started:
1215 	sfc_adapter_unlock(sa);
1216 	SFC_ASSERT(rc > 0);
1217 	return -rc;
1218 }
1219 
1220 static int
1221 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1222 {
1223 	struct sfc_adapter *sa = dev->data->dev_private;
1224 
1225 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1226 
1227 	sfc_adapter_lock(sa);
1228 
1229 	sfc_tx_qstop(sa, tx_queue_id);
1230 
1231 	sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1232 
1233 	sfc_adapter_unlock(sa);
1234 	return 0;
1235 }
1236 
1237 static efx_tunnel_protocol_t
1238 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1239 {
1240 	switch (rte_type) {
1241 	case RTE_TUNNEL_TYPE_VXLAN:
1242 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1243 	case RTE_TUNNEL_TYPE_GENEVE:
1244 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1245 	default:
1246 		return EFX_TUNNEL_NPROTOS;
1247 	}
1248 }
1249 
1250 enum sfc_udp_tunnel_op_e {
1251 	SFC_UDP_TUNNEL_ADD_PORT,
1252 	SFC_UDP_TUNNEL_DEL_PORT,
1253 };
1254 
1255 static int
1256 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1257 		      struct rte_eth_udp_tunnel *tunnel_udp,
1258 		      enum sfc_udp_tunnel_op_e op)
1259 {
1260 	struct sfc_adapter *sa = dev->data->dev_private;
1261 	efx_tunnel_protocol_t tunnel_proto;
1262 	int rc;
1263 
1264 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1265 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1266 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1267 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1268 
1269 	tunnel_proto =
1270 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1271 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1272 		rc = ENOTSUP;
1273 		goto fail_bad_proto;
1274 	}
1275 
1276 	sfc_adapter_lock(sa);
1277 
1278 	switch (op) {
1279 	case SFC_UDP_TUNNEL_ADD_PORT:
1280 		rc = efx_tunnel_config_udp_add(sa->nic,
1281 					       tunnel_udp->udp_port,
1282 					       tunnel_proto);
1283 		break;
1284 	case SFC_UDP_TUNNEL_DEL_PORT:
1285 		rc = efx_tunnel_config_udp_remove(sa->nic,
1286 						  tunnel_udp->udp_port,
1287 						  tunnel_proto);
1288 		break;
1289 	default:
1290 		rc = EINVAL;
1291 		goto fail_bad_op;
1292 	}
1293 
1294 	if (rc != 0)
1295 		goto fail_op;
1296 
1297 	if (sa->state == SFC_ADAPTER_STARTED) {
1298 		rc = efx_tunnel_reconfigure(sa->nic);
1299 		if (rc == EAGAIN) {
1300 			/*
1301 			 * Configuration is accepted by FW and MC reboot
1302 			 * is initiated to apply the changes. MC reboot
1303 			 * will be handled in a usual way (MC reboot
1304 			 * event on management event queue and adapter
1305 			 * restart).
1306 			 */
1307 			rc = 0;
1308 		} else if (rc != 0) {
1309 			goto fail_reconfigure;
1310 		}
1311 	}
1312 
1313 	sfc_adapter_unlock(sa);
1314 	return 0;
1315 
1316 fail_reconfigure:
1317 	/* Remove/restore entry since the change makes the trouble */
1318 	switch (op) {
1319 	case SFC_UDP_TUNNEL_ADD_PORT:
1320 		(void)efx_tunnel_config_udp_remove(sa->nic,
1321 						   tunnel_udp->udp_port,
1322 						   tunnel_proto);
1323 		break;
1324 	case SFC_UDP_TUNNEL_DEL_PORT:
1325 		(void)efx_tunnel_config_udp_add(sa->nic,
1326 						tunnel_udp->udp_port,
1327 						tunnel_proto);
1328 		break;
1329 	}
1330 
1331 fail_op:
1332 fail_bad_op:
1333 	sfc_adapter_unlock(sa);
1334 
1335 fail_bad_proto:
1336 	SFC_ASSERT(rc > 0);
1337 	return -rc;
1338 }
1339 
1340 static int
1341 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1342 			    struct rte_eth_udp_tunnel *tunnel_udp)
1343 {
1344 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1345 }
1346 
1347 static int
1348 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1349 			    struct rte_eth_udp_tunnel *tunnel_udp)
1350 {
1351 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1352 }
1353 
1354 #if EFSYS_OPT_RX_SCALE
1355 static int
1356 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1357 			  struct rte_eth_rss_conf *rss_conf)
1358 {
1359 	struct sfc_adapter *sa = dev->data->dev_private;
1360 	struct sfc_port *port = &sa->port;
1361 
1362 	if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1363 		return -ENOTSUP;
1364 
1365 	if (sa->rss_channels == 0)
1366 		return -EINVAL;
1367 
1368 	sfc_adapter_lock(sa);
1369 
1370 	/*
1371 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1372 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1373 	 * flags which corresponds to the active EFX configuration stored
1374 	 * locally in 'sfc_adapter' and kept up-to-date
1375 	 */
1376 	rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1377 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1378 	if (rss_conf->rss_key != NULL)
1379 		rte_memcpy(rss_conf->rss_key, sa->rss_key, EFX_RSS_KEY_SIZE);
1380 
1381 	sfc_adapter_unlock(sa);
1382 
1383 	return 0;
1384 }
1385 
1386 static int
1387 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1388 			struct rte_eth_rss_conf *rss_conf)
1389 {
1390 	struct sfc_adapter *sa = dev->data->dev_private;
1391 	struct sfc_port *port = &sa->port;
1392 	unsigned int efx_hash_types;
1393 	int rc = 0;
1394 
1395 	if (port->isolated)
1396 		return -ENOTSUP;
1397 
1398 	if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1399 		sfc_err(sa, "RSS is not available");
1400 		return -ENOTSUP;
1401 	}
1402 
1403 	if (sa->rss_channels == 0) {
1404 		sfc_err(sa, "RSS is not configured");
1405 		return -EINVAL;
1406 	}
1407 
1408 	if ((rss_conf->rss_key != NULL) &&
1409 	    (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1410 		sfc_err(sa, "RSS key size is wrong (should be %lu)",
1411 			sizeof(sa->rss_key));
1412 		return -EINVAL;
1413 	}
1414 
1415 	if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1416 		sfc_err(sa, "unsupported hash functions requested");
1417 		return -EINVAL;
1418 	}
1419 
1420 	sfc_adapter_lock(sa);
1421 
1422 	efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1423 
1424 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1425 				   EFX_RX_HASHALG_TOEPLITZ,
1426 				   efx_hash_types, B_TRUE);
1427 	if (rc != 0)
1428 		goto fail_scale_mode_set;
1429 
1430 	if (rss_conf->rss_key != NULL) {
1431 		if (sa->state == SFC_ADAPTER_STARTED) {
1432 			rc = efx_rx_scale_key_set(sa->nic,
1433 						  EFX_RSS_CONTEXT_DEFAULT,
1434 						  rss_conf->rss_key,
1435 						  sizeof(sa->rss_key));
1436 			if (rc != 0)
1437 				goto fail_scale_key_set;
1438 		}
1439 
1440 		rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1441 	}
1442 
1443 	sa->rss_hash_types = efx_hash_types;
1444 
1445 	sfc_adapter_unlock(sa);
1446 
1447 	return 0;
1448 
1449 fail_scale_key_set:
1450 	if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1451 				  EFX_RX_HASHALG_TOEPLITZ,
1452 				  sa->rss_hash_types, B_TRUE) != 0)
1453 		sfc_err(sa, "failed to restore RSS mode");
1454 
1455 fail_scale_mode_set:
1456 	sfc_adapter_unlock(sa);
1457 	return -rc;
1458 }
1459 
1460 static int
1461 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1462 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1463 		       uint16_t reta_size)
1464 {
1465 	struct sfc_adapter *sa = dev->data->dev_private;
1466 	struct sfc_port *port = &sa->port;
1467 	int entry;
1468 
1469 	if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1470 		return -ENOTSUP;
1471 
1472 	if (sa->rss_channels == 0)
1473 		return -EINVAL;
1474 
1475 	if (reta_size != EFX_RSS_TBL_SIZE)
1476 		return -EINVAL;
1477 
1478 	sfc_adapter_lock(sa);
1479 
1480 	for (entry = 0; entry < reta_size; entry++) {
1481 		int grp = entry / RTE_RETA_GROUP_SIZE;
1482 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1483 
1484 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1485 			reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1486 	}
1487 
1488 	sfc_adapter_unlock(sa);
1489 
1490 	return 0;
1491 }
1492 
1493 static int
1494 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1495 			struct rte_eth_rss_reta_entry64 *reta_conf,
1496 			uint16_t reta_size)
1497 {
1498 	struct sfc_adapter *sa = dev->data->dev_private;
1499 	struct sfc_port *port = &sa->port;
1500 	unsigned int *rss_tbl_new;
1501 	uint16_t entry;
1502 	int rc = 0;
1503 
1504 
1505 	if (port->isolated)
1506 		return -ENOTSUP;
1507 
1508 	if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1509 		sfc_err(sa, "RSS is not available");
1510 		return -ENOTSUP;
1511 	}
1512 
1513 	if (sa->rss_channels == 0) {
1514 		sfc_err(sa, "RSS is not configured");
1515 		return -EINVAL;
1516 	}
1517 
1518 	if (reta_size != EFX_RSS_TBL_SIZE) {
1519 		sfc_err(sa, "RETA size is wrong (should be %u)",
1520 			EFX_RSS_TBL_SIZE);
1521 		return -EINVAL;
1522 	}
1523 
1524 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1525 	if (rss_tbl_new == NULL)
1526 		return -ENOMEM;
1527 
1528 	sfc_adapter_lock(sa);
1529 
1530 	rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1531 
1532 	for (entry = 0; entry < reta_size; entry++) {
1533 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1534 		struct rte_eth_rss_reta_entry64 *grp;
1535 
1536 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1537 
1538 		if (grp->mask & (1ull << grp_idx)) {
1539 			if (grp->reta[grp_idx] >= sa->rss_channels) {
1540 				rc = EINVAL;
1541 				goto bad_reta_entry;
1542 			}
1543 			rss_tbl_new[entry] = grp->reta[grp_idx];
1544 		}
1545 	}
1546 
1547 	if (sa->state == SFC_ADAPTER_STARTED) {
1548 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1549 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1550 		if (rc != 0)
1551 			goto fail_scale_tbl_set;
1552 	}
1553 
1554 	rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1555 
1556 fail_scale_tbl_set:
1557 bad_reta_entry:
1558 	sfc_adapter_unlock(sa);
1559 
1560 	rte_free(rss_tbl_new);
1561 
1562 	SFC_ASSERT(rc >= 0);
1563 	return -rc;
1564 }
1565 #endif
1566 
1567 static int
1568 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1569 		    enum rte_filter_op filter_op,
1570 		    void *arg)
1571 {
1572 	struct sfc_adapter *sa = dev->data->dev_private;
1573 	int rc = ENOTSUP;
1574 
1575 	sfc_log_init(sa, "entry");
1576 
1577 	switch (filter_type) {
1578 	case RTE_ETH_FILTER_NONE:
1579 		sfc_err(sa, "Global filters configuration not supported");
1580 		break;
1581 	case RTE_ETH_FILTER_MACVLAN:
1582 		sfc_err(sa, "MACVLAN filters not supported");
1583 		break;
1584 	case RTE_ETH_FILTER_ETHERTYPE:
1585 		sfc_err(sa, "EtherType filters not supported");
1586 		break;
1587 	case RTE_ETH_FILTER_FLEXIBLE:
1588 		sfc_err(sa, "Flexible filters not supported");
1589 		break;
1590 	case RTE_ETH_FILTER_SYN:
1591 		sfc_err(sa, "SYN filters not supported");
1592 		break;
1593 	case RTE_ETH_FILTER_NTUPLE:
1594 		sfc_err(sa, "NTUPLE filters not supported");
1595 		break;
1596 	case RTE_ETH_FILTER_TUNNEL:
1597 		sfc_err(sa, "Tunnel filters not supported");
1598 		break;
1599 	case RTE_ETH_FILTER_FDIR:
1600 		sfc_err(sa, "Flow Director filters not supported");
1601 		break;
1602 	case RTE_ETH_FILTER_HASH:
1603 		sfc_err(sa, "Hash filters not supported");
1604 		break;
1605 	case RTE_ETH_FILTER_GENERIC:
1606 		if (filter_op != RTE_ETH_FILTER_GET) {
1607 			rc = EINVAL;
1608 		} else {
1609 			*(const void **)arg = &sfc_flow_ops;
1610 			rc = 0;
1611 		}
1612 		break;
1613 	default:
1614 		sfc_err(sa, "Unknown filter type %u", filter_type);
1615 		break;
1616 	}
1617 
1618 	sfc_log_init(sa, "exit: %d", -rc);
1619 	SFC_ASSERT(rc >= 0);
1620 	return -rc;
1621 }
1622 
1623 static const struct eth_dev_ops sfc_eth_dev_ops = {
1624 	.dev_configure			= sfc_dev_configure,
1625 	.dev_start			= sfc_dev_start,
1626 	.dev_stop			= sfc_dev_stop,
1627 	.dev_set_link_up		= sfc_dev_set_link_up,
1628 	.dev_set_link_down		= sfc_dev_set_link_down,
1629 	.dev_close			= sfc_dev_close,
1630 	.promiscuous_enable		= sfc_dev_promisc_enable,
1631 	.promiscuous_disable		= sfc_dev_promisc_disable,
1632 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1633 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1634 	.link_update			= sfc_dev_link_update,
1635 	.stats_get			= sfc_stats_get,
1636 	.stats_reset			= sfc_stats_reset,
1637 	.xstats_get			= sfc_xstats_get,
1638 	.xstats_reset			= sfc_stats_reset,
1639 	.xstats_get_names		= sfc_xstats_get_names,
1640 	.dev_infos_get			= sfc_dev_infos_get,
1641 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1642 	.mtu_set			= sfc_dev_set_mtu,
1643 	.rx_queue_start			= sfc_rx_queue_start,
1644 	.rx_queue_stop			= sfc_rx_queue_stop,
1645 	.tx_queue_start			= sfc_tx_queue_start,
1646 	.tx_queue_stop			= sfc_tx_queue_stop,
1647 	.rx_queue_setup			= sfc_rx_queue_setup,
1648 	.rx_queue_release		= sfc_rx_queue_release,
1649 	.rx_queue_count			= sfc_rx_queue_count,
1650 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1651 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1652 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1653 	.tx_queue_setup			= sfc_tx_queue_setup,
1654 	.tx_queue_release		= sfc_tx_queue_release,
1655 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1656 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1657 	.mac_addr_set			= sfc_mac_addr_set,
1658 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1659 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
1660 #if EFSYS_OPT_RX_SCALE
1661 	.reta_update			= sfc_dev_rss_reta_update,
1662 	.reta_query			= sfc_dev_rss_reta_query,
1663 	.rss_hash_update		= sfc_dev_rss_hash_update,
1664 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1665 #endif
1666 	.filter_ctrl			= sfc_dev_filter_ctrl,
1667 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1668 	.rxq_info_get			= sfc_rx_queue_info_get,
1669 	.txq_info_get			= sfc_tx_queue_info_get,
1670 	.fw_version_get			= sfc_fw_version_get,
1671 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1672 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1673 };
1674 
1675 /**
1676  * Duplicate a string in potentially shared memory required for
1677  * multi-process support.
1678  *
1679  * strdup() allocates from process-local heap/memory.
1680  */
1681 static char *
1682 sfc_strdup(const char *str)
1683 {
1684 	size_t size;
1685 	char *copy;
1686 
1687 	if (str == NULL)
1688 		return NULL;
1689 
1690 	size = strlen(str) + 1;
1691 	copy = rte_malloc(__func__, size, 0);
1692 	if (copy != NULL)
1693 		rte_memcpy(copy, str, size);
1694 
1695 	return copy;
1696 }
1697 
1698 static int
1699 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1700 {
1701 	struct sfc_adapter *sa = dev->data->dev_private;
1702 	unsigned int avail_caps = 0;
1703 	const char *rx_name = NULL;
1704 	const char *tx_name = NULL;
1705 	int rc;
1706 
1707 	switch (sa->family) {
1708 	case EFX_FAMILY_HUNTINGTON:
1709 	case EFX_FAMILY_MEDFORD:
1710 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1711 		break;
1712 	default:
1713 		break;
1714 	}
1715 
1716 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1717 				sfc_kvarg_string_handler, &rx_name);
1718 	if (rc != 0)
1719 		goto fail_kvarg_rx_datapath;
1720 
1721 	if (rx_name != NULL) {
1722 		sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1723 		if (sa->dp_rx == NULL) {
1724 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1725 			rc = ENOENT;
1726 			goto fail_dp_rx;
1727 		}
1728 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1729 			sfc_err(sa,
1730 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1731 				rx_name);
1732 			rc = EINVAL;
1733 			goto fail_dp_rx_caps;
1734 		}
1735 	} else {
1736 		sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1737 		if (sa->dp_rx == NULL) {
1738 			sfc_err(sa, "Rx datapath by caps %#x not found",
1739 				avail_caps);
1740 			rc = ENOENT;
1741 			goto fail_dp_rx;
1742 		}
1743 	}
1744 
1745 	sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1746 	if (sa->dp_rx_name == NULL) {
1747 		rc = ENOMEM;
1748 		goto fail_dp_rx_name;
1749 	}
1750 
1751 	sfc_info(sa, "use %s Rx datapath", sa->dp_rx_name);
1752 
1753 	dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1754 
1755 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1756 				sfc_kvarg_string_handler, &tx_name);
1757 	if (rc != 0)
1758 		goto fail_kvarg_tx_datapath;
1759 
1760 	if (tx_name != NULL) {
1761 		sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1762 		if (sa->dp_tx == NULL) {
1763 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1764 			rc = ENOENT;
1765 			goto fail_dp_tx;
1766 		}
1767 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1768 			sfc_err(sa,
1769 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1770 				tx_name);
1771 			rc = EINVAL;
1772 			goto fail_dp_tx_caps;
1773 		}
1774 	} else {
1775 		sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1776 		if (sa->dp_tx == NULL) {
1777 			sfc_err(sa, "Tx datapath by caps %#x not found",
1778 				avail_caps);
1779 			rc = ENOENT;
1780 			goto fail_dp_tx;
1781 		}
1782 	}
1783 
1784 	sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1785 	if (sa->dp_tx_name == NULL) {
1786 		rc = ENOMEM;
1787 		goto fail_dp_tx_name;
1788 	}
1789 
1790 	sfc_info(sa, "use %s Tx datapath", sa->dp_tx_name);
1791 
1792 	dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1793 
1794 	dev->dev_ops = &sfc_eth_dev_ops;
1795 
1796 	return 0;
1797 
1798 fail_dp_tx_name:
1799 fail_dp_tx_caps:
1800 	sa->dp_tx = NULL;
1801 
1802 fail_dp_tx:
1803 fail_kvarg_tx_datapath:
1804 	rte_free(sa->dp_rx_name);
1805 	sa->dp_rx_name = NULL;
1806 
1807 fail_dp_rx_name:
1808 fail_dp_rx_caps:
1809 	sa->dp_rx = NULL;
1810 
1811 fail_dp_rx:
1812 fail_kvarg_rx_datapath:
1813 	return rc;
1814 }
1815 
1816 static void
1817 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1818 {
1819 	struct sfc_adapter *sa = dev->data->dev_private;
1820 
1821 	dev->dev_ops = NULL;
1822 	dev->rx_pkt_burst = NULL;
1823 	dev->tx_pkt_burst = NULL;
1824 
1825 	rte_free(sa->dp_tx_name);
1826 	sa->dp_tx_name = NULL;
1827 	sa->dp_tx = NULL;
1828 
1829 	rte_free(sa->dp_rx_name);
1830 	sa->dp_rx_name = NULL;
1831 	sa->dp_rx = NULL;
1832 }
1833 
1834 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1835 	.rxq_info_get			= sfc_rx_queue_info_get,
1836 	.txq_info_get			= sfc_tx_queue_info_get,
1837 };
1838 
1839 static int
1840 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1841 {
1842 	/*
1843 	 * Device private data has really many process-local pointers.
1844 	 * Below code should be extremely careful to use data located
1845 	 * in shared memory only.
1846 	 */
1847 	struct sfc_adapter *sa = dev->data->dev_private;
1848 	const struct sfc_dp_rx *dp_rx;
1849 	const struct sfc_dp_tx *dp_tx;
1850 	int rc;
1851 
1852 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1853 	if (dp_rx == NULL) {
1854 		sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1855 		rc = ENOENT;
1856 		goto fail_dp_rx;
1857 	}
1858 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1859 		sfc_err(sa, "%s Rx datapath does not support multi-process",
1860 			sa->dp_tx_name);
1861 		rc = EINVAL;
1862 		goto fail_dp_rx_multi_process;
1863 	}
1864 
1865 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1866 	if (dp_tx == NULL) {
1867 		sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1868 		rc = ENOENT;
1869 		goto fail_dp_tx;
1870 	}
1871 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1872 		sfc_err(sa, "%s Tx datapath does not support multi-process",
1873 			sa->dp_tx_name);
1874 		rc = EINVAL;
1875 		goto fail_dp_tx_multi_process;
1876 	}
1877 
1878 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1879 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1880 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
1881 
1882 	return 0;
1883 
1884 fail_dp_tx_multi_process:
1885 fail_dp_tx:
1886 fail_dp_rx_multi_process:
1887 fail_dp_rx:
1888 	return rc;
1889 }
1890 
1891 static void
1892 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1893 {
1894 	dev->dev_ops = NULL;
1895 	dev->tx_pkt_burst = NULL;
1896 	dev->rx_pkt_burst = NULL;
1897 }
1898 
1899 static void
1900 sfc_register_dp(void)
1901 {
1902 	/* Register once */
1903 	if (TAILQ_EMPTY(&sfc_dp_head)) {
1904 		/* Prefer EF10 datapath */
1905 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1906 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1907 
1908 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1909 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1910 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1911 	}
1912 }
1913 
1914 static int
1915 sfc_eth_dev_init(struct rte_eth_dev *dev)
1916 {
1917 	struct sfc_adapter *sa = dev->data->dev_private;
1918 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1919 	int rc;
1920 	const efx_nic_cfg_t *encp;
1921 	const struct ether_addr *from;
1922 
1923 	sfc_register_dp();
1924 
1925 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1926 		return -sfc_eth_dev_secondary_set_ops(dev);
1927 
1928 	/* Required for logging */
1929 	sa->pci_addr = pci_dev->addr;
1930 	sa->port_id = dev->data->port_id;
1931 
1932 	sa->eth_dev = dev;
1933 
1934 	/* Copy PCI device info to the dev->data */
1935 	rte_eth_copy_pci_info(dev, pci_dev);
1936 
1937 	rc = sfc_kvargs_parse(sa);
1938 	if (rc != 0)
1939 		goto fail_kvargs_parse;
1940 
1941 	rc = sfc_kvargs_process(sa, SFC_KVARG_DEBUG_INIT,
1942 				sfc_kvarg_bool_handler, &sa->debug_init);
1943 	if (rc != 0)
1944 		goto fail_kvarg_debug_init;
1945 
1946 	sfc_log_init(sa, "entry");
1947 
1948 	dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1949 	if (dev->data->mac_addrs == NULL) {
1950 		rc = ENOMEM;
1951 		goto fail_mac_addrs;
1952 	}
1953 
1954 	sfc_adapter_lock_init(sa);
1955 	sfc_adapter_lock(sa);
1956 
1957 	sfc_log_init(sa, "probing");
1958 	rc = sfc_probe(sa);
1959 	if (rc != 0)
1960 		goto fail_probe;
1961 
1962 	sfc_log_init(sa, "set device ops");
1963 	rc = sfc_eth_dev_set_ops(dev);
1964 	if (rc != 0)
1965 		goto fail_set_ops;
1966 
1967 	sfc_log_init(sa, "attaching");
1968 	rc = sfc_attach(sa);
1969 	if (rc != 0)
1970 		goto fail_attach;
1971 
1972 	encp = efx_nic_cfg_get(sa->nic);
1973 
1974 	/*
1975 	 * The arguments are really reverse order in comparison to
1976 	 * Linux kernel. Copy from NIC config to Ethernet device data.
1977 	 */
1978 	from = (const struct ether_addr *)(encp->enc_mac_addr);
1979 	ether_addr_copy(from, &dev->data->mac_addrs[0]);
1980 
1981 	sfc_adapter_unlock(sa);
1982 
1983 	sfc_log_init(sa, "done");
1984 	return 0;
1985 
1986 fail_attach:
1987 	sfc_eth_dev_clear_ops(dev);
1988 
1989 fail_set_ops:
1990 	sfc_unprobe(sa);
1991 
1992 fail_probe:
1993 	sfc_adapter_unlock(sa);
1994 	sfc_adapter_lock_fini(sa);
1995 	rte_free(dev->data->mac_addrs);
1996 	dev->data->mac_addrs = NULL;
1997 
1998 fail_mac_addrs:
1999 fail_kvarg_debug_init:
2000 	sfc_kvargs_cleanup(sa);
2001 
2002 fail_kvargs_parse:
2003 	sfc_log_init(sa, "failed %d", rc);
2004 	SFC_ASSERT(rc > 0);
2005 	return -rc;
2006 }
2007 
2008 static int
2009 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2010 {
2011 	struct sfc_adapter *sa;
2012 
2013 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2014 		sfc_eth_dev_secondary_clear_ops(dev);
2015 		return 0;
2016 	}
2017 
2018 	sa = dev->data->dev_private;
2019 	sfc_log_init(sa, "entry");
2020 
2021 	sfc_adapter_lock(sa);
2022 
2023 	sfc_eth_dev_clear_ops(dev);
2024 
2025 	sfc_detach(sa);
2026 	sfc_unprobe(sa);
2027 
2028 	rte_free(dev->data->mac_addrs);
2029 	dev->data->mac_addrs = NULL;
2030 
2031 	sfc_kvargs_cleanup(sa);
2032 
2033 	sfc_adapter_unlock(sa);
2034 	sfc_adapter_lock_fini(sa);
2035 
2036 	sfc_log_init(sa, "done");
2037 
2038 	/* Required for logging, so cleanup last */
2039 	sa->eth_dev = NULL;
2040 	return 0;
2041 }
2042 
2043 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2044 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2045 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2046 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2047 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2048 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2049 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2050 	{ .vendor_id = 0 /* sentinel */ }
2051 };
2052 
2053 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2054 	struct rte_pci_device *pci_dev)
2055 {
2056 	return rte_eth_dev_pci_generic_probe(pci_dev,
2057 		sizeof(struct sfc_adapter), sfc_eth_dev_init);
2058 }
2059 
2060 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2061 {
2062 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2063 }
2064 
2065 static struct rte_pci_driver sfc_efx_pmd = {
2066 	.id_table = pci_id_sfc_efx_map,
2067 	.drv_flags =
2068 		RTE_PCI_DRV_INTR_LSC |
2069 		RTE_PCI_DRV_NEED_MAPPING,
2070 	.probe = sfc_eth_dev_pci_probe,
2071 	.remove = sfc_eth_dev_pci_remove,
2072 };
2073 
2074 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2075 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2076 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2077 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2078 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2079 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2080 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2081 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long> "
2082 	SFC_KVARG_MCDI_LOGGING "=" SFC_KVARG_VALUES_BOOL " "
2083 	SFC_KVARG_DEBUG_INIT "=" SFC_KVARG_VALUES_BOOL);
2084