xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision c7dadc9fb5b45e809adb2e8ab6c5d9888a3a41bc)
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 	/* Initialize to hardware limits */
174 	dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
175 	dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
176 	/* The RXQ hardware requires that the descriptor count is a power
177 	 * of 2, but rx_desc_lim cannot properly describe that constraint.
178 	 */
179 	dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
180 
181 	/* Initialize to hardware limits */
182 	dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
183 	dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
184 	/*
185 	 * The TXQ hardware requires that the descriptor count is a power
186 	 * of 2, but tx_desc_lim cannot properly describe that constraint
187 	 */
188 	dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
189 
190 	if (sa->dp_rx->get_dev_info != NULL)
191 		sa->dp_rx->get_dev_info(dev_info);
192 	if (sa->dp_tx->get_dev_info != NULL)
193 		sa->dp_tx->get_dev_info(dev_info);
194 }
195 
196 static const uint32_t *
197 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
198 {
199 	struct sfc_adapter *sa = dev->data->dev_private;
200 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
201 	uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
202 
203 	return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
204 }
205 
206 static int
207 sfc_dev_configure(struct rte_eth_dev *dev)
208 {
209 	struct rte_eth_dev_data *dev_data = dev->data;
210 	struct sfc_adapter *sa = dev_data->dev_private;
211 	int rc;
212 
213 	sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
214 		     dev_data->nb_rx_queues, dev_data->nb_tx_queues);
215 
216 	sfc_adapter_lock(sa);
217 	switch (sa->state) {
218 	case SFC_ADAPTER_CONFIGURED:
219 		/* FALLTHROUGH */
220 	case SFC_ADAPTER_INITIALIZED:
221 		rc = sfc_configure(sa);
222 		break;
223 	default:
224 		sfc_err(sa, "unexpected adapter state %u to configure",
225 			sa->state);
226 		rc = EINVAL;
227 		break;
228 	}
229 	sfc_adapter_unlock(sa);
230 
231 	sfc_log_init(sa, "done %d", rc);
232 	SFC_ASSERT(rc >= 0);
233 	return -rc;
234 }
235 
236 static int
237 sfc_dev_start(struct rte_eth_dev *dev)
238 {
239 	struct sfc_adapter *sa = dev->data->dev_private;
240 	int rc;
241 
242 	sfc_log_init(sa, "entry");
243 
244 	sfc_adapter_lock(sa);
245 	rc = sfc_start(sa);
246 	sfc_adapter_unlock(sa);
247 
248 	sfc_log_init(sa, "done %d", rc);
249 	SFC_ASSERT(rc >= 0);
250 	return -rc;
251 }
252 
253 static int
254 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
255 {
256 	struct sfc_adapter *sa = dev->data->dev_private;
257 	struct rte_eth_link *dev_link = &dev->data->dev_link;
258 	struct rte_eth_link old_link;
259 	struct rte_eth_link current_link;
260 
261 	sfc_log_init(sa, "entry");
262 
263 retry:
264 	EFX_STATIC_ASSERT(sizeof(*dev_link) == sizeof(rte_atomic64_t));
265 	*(int64_t *)&old_link = rte_atomic64_read((rte_atomic64_t *)dev_link);
266 
267 	if (sa->state != SFC_ADAPTER_STARTED) {
268 		sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
269 		if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
270 					 *(uint64_t *)&old_link,
271 					 *(uint64_t *)&current_link))
272 			goto retry;
273 	} else if (wait_to_complete) {
274 		efx_link_mode_t link_mode;
275 
276 		if (efx_port_poll(sa->nic, &link_mode) != 0)
277 			link_mode = EFX_LINK_UNKNOWN;
278 		sfc_port_link_mode_to_info(link_mode, &current_link);
279 
280 		if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
281 					 *(uint64_t *)&old_link,
282 					 *(uint64_t *)&current_link))
283 			goto retry;
284 	} else {
285 		sfc_ev_mgmt_qpoll(sa);
286 		*(int64_t *)&current_link =
287 			rte_atomic64_read((rte_atomic64_t *)dev_link);
288 	}
289 
290 	if (old_link.link_status != current_link.link_status)
291 		sfc_info(sa, "Link status is %s",
292 			 current_link.link_status ? "UP" : "DOWN");
293 
294 	return old_link.link_status == current_link.link_status ? 0 : -1;
295 }
296 
297 static void
298 sfc_dev_stop(struct rte_eth_dev *dev)
299 {
300 	struct sfc_adapter *sa = dev->data->dev_private;
301 
302 	sfc_log_init(sa, "entry");
303 
304 	sfc_adapter_lock(sa);
305 	sfc_stop(sa);
306 	sfc_adapter_unlock(sa);
307 
308 	sfc_log_init(sa, "done");
309 }
310 
311 static int
312 sfc_dev_set_link_up(struct rte_eth_dev *dev)
313 {
314 	struct sfc_adapter *sa = dev->data->dev_private;
315 	int rc;
316 
317 	sfc_log_init(sa, "entry");
318 
319 	sfc_adapter_lock(sa);
320 	rc = sfc_start(sa);
321 	sfc_adapter_unlock(sa);
322 
323 	SFC_ASSERT(rc >= 0);
324 	return -rc;
325 }
326 
327 static int
328 sfc_dev_set_link_down(struct rte_eth_dev *dev)
329 {
330 	struct sfc_adapter *sa = dev->data->dev_private;
331 
332 	sfc_log_init(sa, "entry");
333 
334 	sfc_adapter_lock(sa);
335 	sfc_stop(sa);
336 	sfc_adapter_unlock(sa);
337 
338 	return 0;
339 }
340 
341 static void
342 sfc_dev_close(struct rte_eth_dev *dev)
343 {
344 	struct sfc_adapter *sa = dev->data->dev_private;
345 
346 	sfc_log_init(sa, "entry");
347 
348 	sfc_adapter_lock(sa);
349 	switch (sa->state) {
350 	case SFC_ADAPTER_STARTED:
351 		sfc_stop(sa);
352 		SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
353 		/* FALLTHROUGH */
354 	case SFC_ADAPTER_CONFIGURED:
355 		sfc_close(sa);
356 		SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
357 		/* FALLTHROUGH */
358 	case SFC_ADAPTER_INITIALIZED:
359 		break;
360 	default:
361 		sfc_err(sa, "unexpected adapter state %u on close", sa->state);
362 		break;
363 	}
364 	sfc_adapter_unlock(sa);
365 
366 	sfc_log_init(sa, "done");
367 }
368 
369 static void
370 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
371 		   boolean_t enabled)
372 {
373 	struct sfc_port *port;
374 	boolean_t *toggle;
375 	struct sfc_adapter *sa = dev->data->dev_private;
376 	boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
377 	const char *desc = (allmulti) ? "all-multi" : "promiscuous";
378 
379 	sfc_adapter_lock(sa);
380 
381 	port = &sa->port;
382 	toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
383 
384 	if (*toggle != enabled) {
385 		*toggle = enabled;
386 
387 		if (port->isolated) {
388 			sfc_warn(sa, "isolated mode is active on the port");
389 			sfc_warn(sa, "the change is to be applied on the next "
390 				     "start provided that isolated mode is "
391 				     "disabled prior the next start");
392 		} else if ((sa->state == SFC_ADAPTER_STARTED) &&
393 			   (sfc_set_rx_mode(sa) != 0)) {
394 			*toggle = !(enabled);
395 			sfc_warn(sa, "Failed to %s %s mode",
396 				 ((enabled) ? "enable" : "disable"), desc);
397 		}
398 	}
399 
400 	sfc_adapter_unlock(sa);
401 }
402 
403 static void
404 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
405 {
406 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
407 }
408 
409 static void
410 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
411 {
412 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
413 }
414 
415 static void
416 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
417 {
418 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
419 }
420 
421 static void
422 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
423 {
424 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
425 }
426 
427 static int
428 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
429 		   uint16_t nb_rx_desc, unsigned int socket_id,
430 		   const struct rte_eth_rxconf *rx_conf,
431 		   struct rte_mempool *mb_pool)
432 {
433 	struct sfc_adapter *sa = dev->data->dev_private;
434 	int rc;
435 
436 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
437 		     rx_queue_id, nb_rx_desc, socket_id);
438 
439 	sfc_adapter_lock(sa);
440 
441 	rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
442 			  rx_conf, mb_pool);
443 	if (rc != 0)
444 		goto fail_rx_qinit;
445 
446 	dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
447 
448 	sfc_adapter_unlock(sa);
449 
450 	return 0;
451 
452 fail_rx_qinit:
453 	sfc_adapter_unlock(sa);
454 	SFC_ASSERT(rc > 0);
455 	return -rc;
456 }
457 
458 static void
459 sfc_rx_queue_release(void *queue)
460 {
461 	struct sfc_dp_rxq *dp_rxq = queue;
462 	struct sfc_rxq *rxq;
463 	struct sfc_adapter *sa;
464 	unsigned int sw_index;
465 
466 	if (dp_rxq == NULL)
467 		return;
468 
469 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
470 	sa = rxq->evq->sa;
471 	sfc_adapter_lock(sa);
472 
473 	sw_index = sfc_rxq_sw_index(rxq);
474 
475 	sfc_log_init(sa, "RxQ=%u", sw_index);
476 
477 	sa->eth_dev->data->rx_queues[sw_index] = NULL;
478 
479 	sfc_rx_qfini(sa, sw_index);
480 
481 	sfc_adapter_unlock(sa);
482 }
483 
484 static int
485 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
486 		   uint16_t nb_tx_desc, unsigned int socket_id,
487 		   const struct rte_eth_txconf *tx_conf)
488 {
489 	struct sfc_adapter *sa = dev->data->dev_private;
490 	int rc;
491 
492 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
493 		     tx_queue_id, nb_tx_desc, socket_id);
494 
495 	sfc_adapter_lock(sa);
496 
497 	rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
498 	if (rc != 0)
499 		goto fail_tx_qinit;
500 
501 	dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
502 
503 	sfc_adapter_unlock(sa);
504 	return 0;
505 
506 fail_tx_qinit:
507 	sfc_adapter_unlock(sa);
508 	SFC_ASSERT(rc > 0);
509 	return -rc;
510 }
511 
512 static void
513 sfc_tx_queue_release(void *queue)
514 {
515 	struct sfc_dp_txq *dp_txq = queue;
516 	struct sfc_txq *txq;
517 	unsigned int sw_index;
518 	struct sfc_adapter *sa;
519 
520 	if (dp_txq == NULL)
521 		return;
522 
523 	txq = sfc_txq_by_dp_txq(dp_txq);
524 	sw_index = sfc_txq_sw_index(txq);
525 
526 	SFC_ASSERT(txq->evq != NULL);
527 	sa = txq->evq->sa;
528 
529 	sfc_log_init(sa, "TxQ = %u", sw_index);
530 
531 	sfc_adapter_lock(sa);
532 
533 	SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
534 	sa->eth_dev->data->tx_queues[sw_index] = NULL;
535 
536 	sfc_tx_qfini(sa, sw_index);
537 
538 	sfc_adapter_unlock(sa);
539 }
540 
541 static int
542 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
543 {
544 	struct sfc_adapter *sa = dev->data->dev_private;
545 	struct sfc_port *port = &sa->port;
546 	uint64_t *mac_stats;
547 	int ret;
548 
549 	rte_spinlock_lock(&port->mac_stats_lock);
550 
551 	ret = sfc_port_update_mac_stats(sa);
552 	if (ret != 0)
553 		goto unlock;
554 
555 	mac_stats = port->mac_stats_buf;
556 
557 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
558 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
559 		stats->ipackets =
560 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
561 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
562 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
563 		stats->opackets =
564 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
565 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
566 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
567 		stats->ibytes =
568 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
569 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
570 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
571 		stats->obytes =
572 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
573 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
574 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
575 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
576 		stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
577 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
578 	} else {
579 		stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
580 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
581 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
582 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
583 		/*
584 		 * Take into account stats which are whenever supported
585 		 * on EF10. If some stat is not supported by current
586 		 * firmware variant or HW revision, it is guaranteed
587 		 * to be zero in mac_stats.
588 		 */
589 		stats->imissed =
590 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
591 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
592 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
593 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
594 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
595 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
596 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
597 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
598 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
599 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
600 		stats->ierrors =
601 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
602 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
603 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
604 		/* no oerrors counters supported on EF10 */
605 	}
606 
607 unlock:
608 	rte_spinlock_unlock(&port->mac_stats_lock);
609 	SFC_ASSERT(ret >= 0);
610 	return -ret;
611 }
612 
613 static void
614 sfc_stats_reset(struct rte_eth_dev *dev)
615 {
616 	struct sfc_adapter *sa = dev->data->dev_private;
617 	struct sfc_port *port = &sa->port;
618 	int rc;
619 
620 	if (sa->state != SFC_ADAPTER_STARTED) {
621 		/*
622 		 * The operation cannot be done if port is not started; it
623 		 * will be scheduled to be done during the next port start
624 		 */
625 		port->mac_stats_reset_pending = B_TRUE;
626 		return;
627 	}
628 
629 	rc = sfc_port_reset_mac_stats(sa);
630 	if (rc != 0)
631 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
632 }
633 
634 static int
635 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
636 	       unsigned int xstats_count)
637 {
638 	struct sfc_adapter *sa = dev->data->dev_private;
639 	struct sfc_port *port = &sa->port;
640 	uint64_t *mac_stats;
641 	int rc;
642 	unsigned int i;
643 	int nstats = 0;
644 
645 	rte_spinlock_lock(&port->mac_stats_lock);
646 
647 	rc = sfc_port_update_mac_stats(sa);
648 	if (rc != 0) {
649 		SFC_ASSERT(rc > 0);
650 		nstats = -rc;
651 		goto unlock;
652 	}
653 
654 	mac_stats = port->mac_stats_buf;
655 
656 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
657 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
658 			if (xstats != NULL && nstats < (int)xstats_count) {
659 				xstats[nstats].id = nstats;
660 				xstats[nstats].value = mac_stats[i];
661 			}
662 			nstats++;
663 		}
664 	}
665 
666 unlock:
667 	rte_spinlock_unlock(&port->mac_stats_lock);
668 
669 	return nstats;
670 }
671 
672 static int
673 sfc_xstats_get_names(struct rte_eth_dev *dev,
674 		     struct rte_eth_xstat_name *xstats_names,
675 		     unsigned int xstats_count)
676 {
677 	struct sfc_adapter *sa = dev->data->dev_private;
678 	struct sfc_port *port = &sa->port;
679 	unsigned int i;
680 	unsigned int nstats = 0;
681 
682 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
683 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
684 			if (xstats_names != NULL && nstats < xstats_count)
685 				strncpy(xstats_names[nstats].name,
686 					efx_mac_stat_name(sa->nic, i),
687 					sizeof(xstats_names[0].name));
688 			nstats++;
689 		}
690 	}
691 
692 	return nstats;
693 }
694 
695 static int
696 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
697 		     uint64_t *values, unsigned int n)
698 {
699 	struct sfc_adapter *sa = dev->data->dev_private;
700 	struct sfc_port *port = &sa->port;
701 	uint64_t *mac_stats;
702 	unsigned int nb_supported = 0;
703 	unsigned int nb_written = 0;
704 	unsigned int i;
705 	int ret;
706 	int rc;
707 
708 	if (unlikely(values == NULL) ||
709 	    unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
710 		return port->mac_stats_nb_supported;
711 
712 	rte_spinlock_lock(&port->mac_stats_lock);
713 
714 	rc = sfc_port_update_mac_stats(sa);
715 	if (rc != 0) {
716 		SFC_ASSERT(rc > 0);
717 		ret = -rc;
718 		goto unlock;
719 	}
720 
721 	mac_stats = port->mac_stats_buf;
722 
723 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
724 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
725 			continue;
726 
727 		if ((ids == NULL) || (ids[nb_written] == nb_supported))
728 			values[nb_written++] = mac_stats[i];
729 
730 		++nb_supported;
731 	}
732 
733 	ret = nb_written;
734 
735 unlock:
736 	rte_spinlock_unlock(&port->mac_stats_lock);
737 
738 	return ret;
739 }
740 
741 static int
742 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
743 			   struct rte_eth_xstat_name *xstats_names,
744 			   const uint64_t *ids, unsigned int size)
745 {
746 	struct sfc_adapter *sa = dev->data->dev_private;
747 	struct sfc_port *port = &sa->port;
748 	unsigned int nb_supported = 0;
749 	unsigned int nb_written = 0;
750 	unsigned int i;
751 
752 	if (unlikely(xstats_names == NULL) ||
753 	    unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
754 		return port->mac_stats_nb_supported;
755 
756 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
757 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
758 			continue;
759 
760 		if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
761 			char *name = xstats_names[nb_written++].name;
762 
763 			strncpy(name, efx_mac_stat_name(sa->nic, i),
764 				sizeof(xstats_names[0].name));
765 			name[sizeof(xstats_names[0].name) - 1] = '\0';
766 		}
767 
768 		++nb_supported;
769 	}
770 
771 	return nb_written;
772 }
773 
774 static int
775 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
776 {
777 	struct sfc_adapter *sa = dev->data->dev_private;
778 	unsigned int wanted_fc, link_fc;
779 
780 	memset(fc_conf, 0, sizeof(*fc_conf));
781 
782 	sfc_adapter_lock(sa);
783 
784 	if (sa->state == SFC_ADAPTER_STARTED)
785 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
786 	else
787 		link_fc = sa->port.flow_ctrl;
788 
789 	switch (link_fc) {
790 	case 0:
791 		fc_conf->mode = RTE_FC_NONE;
792 		break;
793 	case EFX_FCNTL_RESPOND:
794 		fc_conf->mode = RTE_FC_RX_PAUSE;
795 		break;
796 	case EFX_FCNTL_GENERATE:
797 		fc_conf->mode = RTE_FC_TX_PAUSE;
798 		break;
799 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
800 		fc_conf->mode = RTE_FC_FULL;
801 		break;
802 	default:
803 		sfc_err(sa, "%s: unexpected flow control value %#x",
804 			__func__, link_fc);
805 	}
806 
807 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
808 
809 	sfc_adapter_unlock(sa);
810 
811 	return 0;
812 }
813 
814 static int
815 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
816 {
817 	struct sfc_adapter *sa = dev->data->dev_private;
818 	struct sfc_port *port = &sa->port;
819 	unsigned int fcntl;
820 	int rc;
821 
822 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
823 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
824 	    fc_conf->mac_ctrl_frame_fwd != 0) {
825 		sfc_err(sa, "unsupported flow control settings specified");
826 		rc = EINVAL;
827 		goto fail_inval;
828 	}
829 
830 	switch (fc_conf->mode) {
831 	case RTE_FC_NONE:
832 		fcntl = 0;
833 		break;
834 	case RTE_FC_RX_PAUSE:
835 		fcntl = EFX_FCNTL_RESPOND;
836 		break;
837 	case RTE_FC_TX_PAUSE:
838 		fcntl = EFX_FCNTL_GENERATE;
839 		break;
840 	case RTE_FC_FULL:
841 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
842 		break;
843 	default:
844 		rc = EINVAL;
845 		goto fail_inval;
846 	}
847 
848 	sfc_adapter_lock(sa);
849 
850 	if (sa->state == SFC_ADAPTER_STARTED) {
851 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
852 		if (rc != 0)
853 			goto fail_mac_fcntl_set;
854 	}
855 
856 	port->flow_ctrl = fcntl;
857 	port->flow_ctrl_autoneg = fc_conf->autoneg;
858 
859 	sfc_adapter_unlock(sa);
860 
861 	return 0;
862 
863 fail_mac_fcntl_set:
864 	sfc_adapter_unlock(sa);
865 fail_inval:
866 	SFC_ASSERT(rc > 0);
867 	return -rc;
868 }
869 
870 static int
871 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
872 {
873 	struct sfc_adapter *sa = dev->data->dev_private;
874 	size_t pdu = EFX_MAC_PDU(mtu);
875 	size_t old_pdu;
876 	int rc;
877 
878 	sfc_log_init(sa, "mtu=%u", mtu);
879 
880 	rc = EINVAL;
881 	if (pdu < EFX_MAC_PDU_MIN) {
882 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
883 			(unsigned int)mtu, (unsigned int)pdu,
884 			EFX_MAC_PDU_MIN);
885 		goto fail_inval;
886 	}
887 	if (pdu > EFX_MAC_PDU_MAX) {
888 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
889 			(unsigned int)mtu, (unsigned int)pdu,
890 			EFX_MAC_PDU_MAX);
891 		goto fail_inval;
892 	}
893 
894 	sfc_adapter_lock(sa);
895 
896 	if (pdu != sa->port.pdu) {
897 		if (sa->state == SFC_ADAPTER_STARTED) {
898 			sfc_stop(sa);
899 
900 			old_pdu = sa->port.pdu;
901 			sa->port.pdu = pdu;
902 			rc = sfc_start(sa);
903 			if (rc != 0)
904 				goto fail_start;
905 		} else {
906 			sa->port.pdu = pdu;
907 		}
908 	}
909 
910 	/*
911 	 * The driver does not use it, but other PMDs update jumbo_frame
912 	 * flag and max_rx_pkt_len when MTU is set.
913 	 */
914 	dev->data->dev_conf.rxmode.jumbo_frame = (mtu > ETHER_MAX_LEN);
915 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
916 
917 	sfc_adapter_unlock(sa);
918 
919 	sfc_log_init(sa, "done");
920 	return 0;
921 
922 fail_start:
923 	sa->port.pdu = old_pdu;
924 	if (sfc_start(sa) != 0)
925 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
926 			"PDU max size - port is stopped",
927 			(unsigned int)pdu, (unsigned int)old_pdu);
928 	sfc_adapter_unlock(sa);
929 
930 fail_inval:
931 	sfc_log_init(sa, "failed %d", rc);
932 	SFC_ASSERT(rc > 0);
933 	return -rc;
934 }
935 static void
936 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
937 {
938 	struct sfc_adapter *sa = dev->data->dev_private;
939 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
940 	struct sfc_port *port = &sa->port;
941 	int rc;
942 
943 	sfc_adapter_lock(sa);
944 
945 	/*
946 	 * Copy the address to the device private data so that
947 	 * it could be recalled in the case of adapter restart.
948 	 */
949 	ether_addr_copy(mac_addr, &port->default_mac_addr);
950 
951 	if (port->isolated) {
952 		sfc_err(sa, "isolated mode is active on the port");
953 		sfc_err(sa, "will not set MAC address");
954 		goto unlock;
955 	}
956 
957 	if (sa->state != SFC_ADAPTER_STARTED) {
958 		sfc_info(sa, "the port is not started");
959 		sfc_info(sa, "the new MAC address will be set on port start");
960 
961 		goto unlock;
962 	}
963 
964 	if (encp->enc_allow_set_mac_with_installed_filters) {
965 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
966 		if (rc != 0) {
967 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
968 			goto unlock;
969 		}
970 
971 		/*
972 		 * Changing the MAC address by means of MCDI request
973 		 * has no effect on received traffic, therefore
974 		 * we also need to update unicast filters
975 		 */
976 		rc = sfc_set_rx_mode(sa);
977 		if (rc != 0)
978 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
979 	} else {
980 		sfc_warn(sa, "cannot set MAC address with filters installed");
981 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
982 		sfc_warn(sa, "(some traffic may be dropped)");
983 
984 		/*
985 		 * Since setting MAC address with filters installed is not
986 		 * allowed on the adapter, the new MAC address will be set
987 		 * by means of adapter restart. sfc_start() shall retrieve
988 		 * the new address from the device private data and set it.
989 		 */
990 		sfc_stop(sa);
991 		rc = sfc_start(sa);
992 		if (rc != 0)
993 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
994 	}
995 
996 unlock:
997 	/*
998 	 * In the case of failure sa->port->default_mac_addr does not
999 	 * need rollback since no error code is returned, and the upper
1000 	 * API will anyway update the external MAC address storage.
1001 	 * To be consistent with that new value it is better to keep
1002 	 * the device private value the same.
1003 	 */
1004 	sfc_adapter_unlock(sa);
1005 }
1006 
1007 
1008 static int
1009 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
1010 		     uint32_t nb_mc_addr)
1011 {
1012 	struct sfc_adapter *sa = dev->data->dev_private;
1013 	struct sfc_port *port = &sa->port;
1014 	uint8_t *mc_addrs = port->mcast_addrs;
1015 	int rc;
1016 	unsigned int i;
1017 
1018 	if (port->isolated) {
1019 		sfc_err(sa, "isolated mode is active on the port");
1020 		sfc_err(sa, "will not set multicast address list");
1021 		return -ENOTSUP;
1022 	}
1023 
1024 	if (mc_addrs == NULL)
1025 		return -ENOBUFS;
1026 
1027 	if (nb_mc_addr > port->max_mcast_addrs) {
1028 		sfc_err(sa, "too many multicast addresses: %u > %u",
1029 			 nb_mc_addr, port->max_mcast_addrs);
1030 		return -EINVAL;
1031 	}
1032 
1033 	for (i = 0; i < nb_mc_addr; ++i) {
1034 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1035 				 EFX_MAC_ADDR_LEN);
1036 		mc_addrs += EFX_MAC_ADDR_LEN;
1037 	}
1038 
1039 	port->nb_mcast_addrs = nb_mc_addr;
1040 
1041 	if (sa->state != SFC_ADAPTER_STARTED)
1042 		return 0;
1043 
1044 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1045 					port->nb_mcast_addrs);
1046 	if (rc != 0)
1047 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1048 
1049 	SFC_ASSERT(rc > 0);
1050 	return -rc;
1051 }
1052 
1053 /*
1054  * The function is used by the secondary process as well. It must not
1055  * use any process-local pointers from the adapter data.
1056  */
1057 static void
1058 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1059 		      struct rte_eth_rxq_info *qinfo)
1060 {
1061 	struct sfc_adapter *sa = dev->data->dev_private;
1062 	struct sfc_rxq_info *rxq_info;
1063 	struct sfc_rxq *rxq;
1064 
1065 	sfc_adapter_lock(sa);
1066 
1067 	SFC_ASSERT(rx_queue_id < sa->rxq_count);
1068 
1069 	rxq_info = &sa->rxq_info[rx_queue_id];
1070 	rxq = rxq_info->rxq;
1071 	SFC_ASSERT(rxq != NULL);
1072 
1073 	qinfo->mp = rxq->refill_mb_pool;
1074 	qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1075 	qinfo->conf.rx_drop_en = 1;
1076 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1077 	qinfo->scattered_rx =
1078 		((rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) != 0);
1079 	qinfo->nb_desc = rxq_info->entries;
1080 
1081 	sfc_adapter_unlock(sa);
1082 }
1083 
1084 /*
1085  * The function is used by the secondary process as well. It must not
1086  * use any process-local pointers from the adapter data.
1087  */
1088 static void
1089 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1090 		      struct rte_eth_txq_info *qinfo)
1091 {
1092 	struct sfc_adapter *sa = dev->data->dev_private;
1093 	struct sfc_txq_info *txq_info;
1094 
1095 	sfc_adapter_lock(sa);
1096 
1097 	SFC_ASSERT(tx_queue_id < sa->txq_count);
1098 
1099 	txq_info = &sa->txq_info[tx_queue_id];
1100 	SFC_ASSERT(txq_info->txq != NULL);
1101 
1102 	memset(qinfo, 0, sizeof(*qinfo));
1103 
1104 	qinfo->conf.txq_flags = txq_info->txq->flags;
1105 	qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1106 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1107 	qinfo->nb_desc = txq_info->entries;
1108 
1109 	sfc_adapter_unlock(sa);
1110 }
1111 
1112 static uint32_t
1113 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1114 {
1115 	struct sfc_adapter *sa = dev->data->dev_private;
1116 
1117 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1118 
1119 	return sfc_rx_qdesc_npending(sa, rx_queue_id);
1120 }
1121 
1122 static int
1123 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1124 {
1125 	struct sfc_dp_rxq *dp_rxq = queue;
1126 
1127 	return sfc_rx_qdesc_done(dp_rxq, offset);
1128 }
1129 
1130 static int
1131 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1132 {
1133 	struct sfc_dp_rxq *dp_rxq = queue;
1134 	struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1135 
1136 	return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1137 }
1138 
1139 static int
1140 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1141 {
1142 	struct sfc_dp_txq *dp_txq = queue;
1143 	struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1144 
1145 	return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1146 }
1147 
1148 static int
1149 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1150 {
1151 	struct sfc_adapter *sa = dev->data->dev_private;
1152 	int rc;
1153 
1154 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1155 
1156 	sfc_adapter_lock(sa);
1157 
1158 	rc = EINVAL;
1159 	if (sa->state != SFC_ADAPTER_STARTED)
1160 		goto fail_not_started;
1161 
1162 	rc = sfc_rx_qstart(sa, rx_queue_id);
1163 	if (rc != 0)
1164 		goto fail_rx_qstart;
1165 
1166 	sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1167 
1168 	sfc_adapter_unlock(sa);
1169 
1170 	return 0;
1171 
1172 fail_rx_qstart:
1173 fail_not_started:
1174 	sfc_adapter_unlock(sa);
1175 	SFC_ASSERT(rc > 0);
1176 	return -rc;
1177 }
1178 
1179 static int
1180 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1181 {
1182 	struct sfc_adapter *sa = dev->data->dev_private;
1183 
1184 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1185 
1186 	sfc_adapter_lock(sa);
1187 	sfc_rx_qstop(sa, rx_queue_id);
1188 
1189 	sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1190 
1191 	sfc_adapter_unlock(sa);
1192 
1193 	return 0;
1194 }
1195 
1196 static int
1197 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1198 {
1199 	struct sfc_adapter *sa = dev->data->dev_private;
1200 	int rc;
1201 
1202 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1203 
1204 	sfc_adapter_lock(sa);
1205 
1206 	rc = EINVAL;
1207 	if (sa->state != SFC_ADAPTER_STARTED)
1208 		goto fail_not_started;
1209 
1210 	rc = sfc_tx_qstart(sa, tx_queue_id);
1211 	if (rc != 0)
1212 		goto fail_tx_qstart;
1213 
1214 	sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1215 
1216 	sfc_adapter_unlock(sa);
1217 	return 0;
1218 
1219 fail_tx_qstart:
1220 
1221 fail_not_started:
1222 	sfc_adapter_unlock(sa);
1223 	SFC_ASSERT(rc > 0);
1224 	return -rc;
1225 }
1226 
1227 static int
1228 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1229 {
1230 	struct sfc_adapter *sa = dev->data->dev_private;
1231 
1232 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1233 
1234 	sfc_adapter_lock(sa);
1235 
1236 	sfc_tx_qstop(sa, tx_queue_id);
1237 
1238 	sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1239 
1240 	sfc_adapter_unlock(sa);
1241 	return 0;
1242 }
1243 
1244 static efx_tunnel_protocol_t
1245 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1246 {
1247 	switch (rte_type) {
1248 	case RTE_TUNNEL_TYPE_VXLAN:
1249 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1250 	case RTE_TUNNEL_TYPE_GENEVE:
1251 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1252 	default:
1253 		return EFX_TUNNEL_NPROTOS;
1254 	}
1255 }
1256 
1257 enum sfc_udp_tunnel_op_e {
1258 	SFC_UDP_TUNNEL_ADD_PORT,
1259 	SFC_UDP_TUNNEL_DEL_PORT,
1260 };
1261 
1262 static int
1263 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1264 		      struct rte_eth_udp_tunnel *tunnel_udp,
1265 		      enum sfc_udp_tunnel_op_e op)
1266 {
1267 	struct sfc_adapter *sa = dev->data->dev_private;
1268 	efx_tunnel_protocol_t tunnel_proto;
1269 	int rc;
1270 
1271 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1272 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1273 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1274 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1275 
1276 	tunnel_proto =
1277 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1278 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1279 		rc = ENOTSUP;
1280 		goto fail_bad_proto;
1281 	}
1282 
1283 	sfc_adapter_lock(sa);
1284 
1285 	switch (op) {
1286 	case SFC_UDP_TUNNEL_ADD_PORT:
1287 		rc = efx_tunnel_config_udp_add(sa->nic,
1288 					       tunnel_udp->udp_port,
1289 					       tunnel_proto);
1290 		break;
1291 	case SFC_UDP_TUNNEL_DEL_PORT:
1292 		rc = efx_tunnel_config_udp_remove(sa->nic,
1293 						  tunnel_udp->udp_port,
1294 						  tunnel_proto);
1295 		break;
1296 	default:
1297 		rc = EINVAL;
1298 		goto fail_bad_op;
1299 	}
1300 
1301 	if (rc != 0)
1302 		goto fail_op;
1303 
1304 	if (sa->state == SFC_ADAPTER_STARTED) {
1305 		rc = efx_tunnel_reconfigure(sa->nic);
1306 		if (rc == EAGAIN) {
1307 			/*
1308 			 * Configuration is accepted by FW and MC reboot
1309 			 * is initiated to apply the changes. MC reboot
1310 			 * will be handled in a usual way (MC reboot
1311 			 * event on management event queue and adapter
1312 			 * restart).
1313 			 */
1314 			rc = 0;
1315 		} else if (rc != 0) {
1316 			goto fail_reconfigure;
1317 		}
1318 	}
1319 
1320 	sfc_adapter_unlock(sa);
1321 	return 0;
1322 
1323 fail_reconfigure:
1324 	/* Remove/restore entry since the change makes the trouble */
1325 	switch (op) {
1326 	case SFC_UDP_TUNNEL_ADD_PORT:
1327 		(void)efx_tunnel_config_udp_remove(sa->nic,
1328 						   tunnel_udp->udp_port,
1329 						   tunnel_proto);
1330 		break;
1331 	case SFC_UDP_TUNNEL_DEL_PORT:
1332 		(void)efx_tunnel_config_udp_add(sa->nic,
1333 						tunnel_udp->udp_port,
1334 						tunnel_proto);
1335 		break;
1336 	}
1337 
1338 fail_op:
1339 fail_bad_op:
1340 	sfc_adapter_unlock(sa);
1341 
1342 fail_bad_proto:
1343 	SFC_ASSERT(rc > 0);
1344 	return -rc;
1345 }
1346 
1347 static int
1348 sfc_dev_udp_tunnel_port_add(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_ADD_PORT);
1352 }
1353 
1354 static int
1355 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1356 			    struct rte_eth_udp_tunnel *tunnel_udp)
1357 {
1358 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1359 }
1360 
1361 #if EFSYS_OPT_RX_SCALE
1362 static int
1363 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1364 			  struct rte_eth_rss_conf *rss_conf)
1365 {
1366 	struct sfc_adapter *sa = dev->data->dev_private;
1367 	struct sfc_port *port = &sa->port;
1368 
1369 	if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1370 		return -ENOTSUP;
1371 
1372 	if (sa->rss_channels == 0)
1373 		return -EINVAL;
1374 
1375 	sfc_adapter_lock(sa);
1376 
1377 	/*
1378 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1379 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1380 	 * flags which corresponds to the active EFX configuration stored
1381 	 * locally in 'sfc_adapter' and kept up-to-date
1382 	 */
1383 	rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1384 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1385 	if (rss_conf->rss_key != NULL)
1386 		rte_memcpy(rss_conf->rss_key, sa->rss_key, EFX_RSS_KEY_SIZE);
1387 
1388 	sfc_adapter_unlock(sa);
1389 
1390 	return 0;
1391 }
1392 
1393 static int
1394 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1395 			struct rte_eth_rss_conf *rss_conf)
1396 {
1397 	struct sfc_adapter *sa = dev->data->dev_private;
1398 	struct sfc_port *port = &sa->port;
1399 	unsigned int efx_hash_types;
1400 	int rc = 0;
1401 
1402 	if (port->isolated)
1403 		return -ENOTSUP;
1404 
1405 	if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1406 		sfc_err(sa, "RSS is not available");
1407 		return -ENOTSUP;
1408 	}
1409 
1410 	if (sa->rss_channels == 0) {
1411 		sfc_err(sa, "RSS is not configured");
1412 		return -EINVAL;
1413 	}
1414 
1415 	if ((rss_conf->rss_key != NULL) &&
1416 	    (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1417 		sfc_err(sa, "RSS key size is wrong (should be %lu)",
1418 			sizeof(sa->rss_key));
1419 		return -EINVAL;
1420 	}
1421 
1422 	if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1423 		sfc_err(sa, "unsupported hash functions requested");
1424 		return -EINVAL;
1425 	}
1426 
1427 	sfc_adapter_lock(sa);
1428 
1429 	efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1430 
1431 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1432 				   EFX_RX_HASHALG_TOEPLITZ,
1433 				   efx_hash_types, B_TRUE);
1434 	if (rc != 0)
1435 		goto fail_scale_mode_set;
1436 
1437 	if (rss_conf->rss_key != NULL) {
1438 		if (sa->state == SFC_ADAPTER_STARTED) {
1439 			rc = efx_rx_scale_key_set(sa->nic,
1440 						  EFX_RSS_CONTEXT_DEFAULT,
1441 						  rss_conf->rss_key,
1442 						  sizeof(sa->rss_key));
1443 			if (rc != 0)
1444 				goto fail_scale_key_set;
1445 		}
1446 
1447 		rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1448 	}
1449 
1450 	sa->rss_hash_types = efx_hash_types;
1451 
1452 	sfc_adapter_unlock(sa);
1453 
1454 	return 0;
1455 
1456 fail_scale_key_set:
1457 	if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1458 				  EFX_RX_HASHALG_TOEPLITZ,
1459 				  sa->rss_hash_types, B_TRUE) != 0)
1460 		sfc_err(sa, "failed to restore RSS mode");
1461 
1462 fail_scale_mode_set:
1463 	sfc_adapter_unlock(sa);
1464 	return -rc;
1465 }
1466 
1467 static int
1468 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1469 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1470 		       uint16_t reta_size)
1471 {
1472 	struct sfc_adapter *sa = dev->data->dev_private;
1473 	struct sfc_port *port = &sa->port;
1474 	int entry;
1475 
1476 	if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1477 		return -ENOTSUP;
1478 
1479 	if (sa->rss_channels == 0)
1480 		return -EINVAL;
1481 
1482 	if (reta_size != EFX_RSS_TBL_SIZE)
1483 		return -EINVAL;
1484 
1485 	sfc_adapter_lock(sa);
1486 
1487 	for (entry = 0; entry < reta_size; entry++) {
1488 		int grp = entry / RTE_RETA_GROUP_SIZE;
1489 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1490 
1491 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1492 			reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1493 	}
1494 
1495 	sfc_adapter_unlock(sa);
1496 
1497 	return 0;
1498 }
1499 
1500 static int
1501 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1502 			struct rte_eth_rss_reta_entry64 *reta_conf,
1503 			uint16_t reta_size)
1504 {
1505 	struct sfc_adapter *sa = dev->data->dev_private;
1506 	struct sfc_port *port = &sa->port;
1507 	unsigned int *rss_tbl_new;
1508 	uint16_t entry;
1509 	int rc = 0;
1510 
1511 
1512 	if (port->isolated)
1513 		return -ENOTSUP;
1514 
1515 	if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1516 		sfc_err(sa, "RSS is not available");
1517 		return -ENOTSUP;
1518 	}
1519 
1520 	if (sa->rss_channels == 0) {
1521 		sfc_err(sa, "RSS is not configured");
1522 		return -EINVAL;
1523 	}
1524 
1525 	if (reta_size != EFX_RSS_TBL_SIZE) {
1526 		sfc_err(sa, "RETA size is wrong (should be %u)",
1527 			EFX_RSS_TBL_SIZE);
1528 		return -EINVAL;
1529 	}
1530 
1531 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1532 	if (rss_tbl_new == NULL)
1533 		return -ENOMEM;
1534 
1535 	sfc_adapter_lock(sa);
1536 
1537 	rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1538 
1539 	for (entry = 0; entry < reta_size; entry++) {
1540 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1541 		struct rte_eth_rss_reta_entry64 *grp;
1542 
1543 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1544 
1545 		if (grp->mask & (1ull << grp_idx)) {
1546 			if (grp->reta[grp_idx] >= sa->rss_channels) {
1547 				rc = EINVAL;
1548 				goto bad_reta_entry;
1549 			}
1550 			rss_tbl_new[entry] = grp->reta[grp_idx];
1551 		}
1552 	}
1553 
1554 	if (sa->state == SFC_ADAPTER_STARTED) {
1555 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1556 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1557 		if (rc != 0)
1558 			goto fail_scale_tbl_set;
1559 	}
1560 
1561 	rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1562 
1563 fail_scale_tbl_set:
1564 bad_reta_entry:
1565 	sfc_adapter_unlock(sa);
1566 
1567 	rte_free(rss_tbl_new);
1568 
1569 	SFC_ASSERT(rc >= 0);
1570 	return -rc;
1571 }
1572 #endif
1573 
1574 static int
1575 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1576 		    enum rte_filter_op filter_op,
1577 		    void *arg)
1578 {
1579 	struct sfc_adapter *sa = dev->data->dev_private;
1580 	int rc = ENOTSUP;
1581 
1582 	sfc_log_init(sa, "entry");
1583 
1584 	switch (filter_type) {
1585 	case RTE_ETH_FILTER_NONE:
1586 		sfc_err(sa, "Global filters configuration not supported");
1587 		break;
1588 	case RTE_ETH_FILTER_MACVLAN:
1589 		sfc_err(sa, "MACVLAN filters not supported");
1590 		break;
1591 	case RTE_ETH_FILTER_ETHERTYPE:
1592 		sfc_err(sa, "EtherType filters not supported");
1593 		break;
1594 	case RTE_ETH_FILTER_FLEXIBLE:
1595 		sfc_err(sa, "Flexible filters not supported");
1596 		break;
1597 	case RTE_ETH_FILTER_SYN:
1598 		sfc_err(sa, "SYN filters not supported");
1599 		break;
1600 	case RTE_ETH_FILTER_NTUPLE:
1601 		sfc_err(sa, "NTUPLE filters not supported");
1602 		break;
1603 	case RTE_ETH_FILTER_TUNNEL:
1604 		sfc_err(sa, "Tunnel filters not supported");
1605 		break;
1606 	case RTE_ETH_FILTER_FDIR:
1607 		sfc_err(sa, "Flow Director filters not supported");
1608 		break;
1609 	case RTE_ETH_FILTER_HASH:
1610 		sfc_err(sa, "Hash filters not supported");
1611 		break;
1612 	case RTE_ETH_FILTER_GENERIC:
1613 		if (filter_op != RTE_ETH_FILTER_GET) {
1614 			rc = EINVAL;
1615 		} else {
1616 			*(const void **)arg = &sfc_flow_ops;
1617 			rc = 0;
1618 		}
1619 		break;
1620 	default:
1621 		sfc_err(sa, "Unknown filter type %u", filter_type);
1622 		break;
1623 	}
1624 
1625 	sfc_log_init(sa, "exit: %d", -rc);
1626 	SFC_ASSERT(rc >= 0);
1627 	return -rc;
1628 }
1629 
1630 static const struct eth_dev_ops sfc_eth_dev_ops = {
1631 	.dev_configure			= sfc_dev_configure,
1632 	.dev_start			= sfc_dev_start,
1633 	.dev_stop			= sfc_dev_stop,
1634 	.dev_set_link_up		= sfc_dev_set_link_up,
1635 	.dev_set_link_down		= sfc_dev_set_link_down,
1636 	.dev_close			= sfc_dev_close,
1637 	.promiscuous_enable		= sfc_dev_promisc_enable,
1638 	.promiscuous_disable		= sfc_dev_promisc_disable,
1639 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1640 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1641 	.link_update			= sfc_dev_link_update,
1642 	.stats_get			= sfc_stats_get,
1643 	.stats_reset			= sfc_stats_reset,
1644 	.xstats_get			= sfc_xstats_get,
1645 	.xstats_reset			= sfc_stats_reset,
1646 	.xstats_get_names		= sfc_xstats_get_names,
1647 	.dev_infos_get			= sfc_dev_infos_get,
1648 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1649 	.mtu_set			= sfc_dev_set_mtu,
1650 	.rx_queue_start			= sfc_rx_queue_start,
1651 	.rx_queue_stop			= sfc_rx_queue_stop,
1652 	.tx_queue_start			= sfc_tx_queue_start,
1653 	.tx_queue_stop			= sfc_tx_queue_stop,
1654 	.rx_queue_setup			= sfc_rx_queue_setup,
1655 	.rx_queue_release		= sfc_rx_queue_release,
1656 	.rx_queue_count			= sfc_rx_queue_count,
1657 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1658 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1659 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1660 	.tx_queue_setup			= sfc_tx_queue_setup,
1661 	.tx_queue_release		= sfc_tx_queue_release,
1662 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1663 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1664 	.mac_addr_set			= sfc_mac_addr_set,
1665 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1666 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
1667 #if EFSYS_OPT_RX_SCALE
1668 	.reta_update			= sfc_dev_rss_reta_update,
1669 	.reta_query			= sfc_dev_rss_reta_query,
1670 	.rss_hash_update		= sfc_dev_rss_hash_update,
1671 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1672 #endif
1673 	.filter_ctrl			= sfc_dev_filter_ctrl,
1674 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1675 	.rxq_info_get			= sfc_rx_queue_info_get,
1676 	.txq_info_get			= sfc_tx_queue_info_get,
1677 	.fw_version_get			= sfc_fw_version_get,
1678 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1679 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1680 };
1681 
1682 /**
1683  * Duplicate a string in potentially shared memory required for
1684  * multi-process support.
1685  *
1686  * strdup() allocates from process-local heap/memory.
1687  */
1688 static char *
1689 sfc_strdup(const char *str)
1690 {
1691 	size_t size;
1692 	char *copy;
1693 
1694 	if (str == NULL)
1695 		return NULL;
1696 
1697 	size = strlen(str) + 1;
1698 	copy = rte_malloc(__func__, size, 0);
1699 	if (copy != NULL)
1700 		rte_memcpy(copy, str, size);
1701 
1702 	return copy;
1703 }
1704 
1705 static int
1706 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1707 {
1708 	struct sfc_adapter *sa = dev->data->dev_private;
1709 	unsigned int avail_caps = 0;
1710 	const char *rx_name = NULL;
1711 	const char *tx_name = NULL;
1712 	int rc;
1713 
1714 	switch (sa->family) {
1715 	case EFX_FAMILY_HUNTINGTON:
1716 	case EFX_FAMILY_MEDFORD:
1717 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1718 		break;
1719 	default:
1720 		break;
1721 	}
1722 
1723 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1724 				sfc_kvarg_string_handler, &rx_name);
1725 	if (rc != 0)
1726 		goto fail_kvarg_rx_datapath;
1727 
1728 	if (rx_name != NULL) {
1729 		sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1730 		if (sa->dp_rx == NULL) {
1731 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1732 			rc = ENOENT;
1733 			goto fail_dp_rx;
1734 		}
1735 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1736 			sfc_err(sa,
1737 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1738 				rx_name);
1739 			rc = EINVAL;
1740 			goto fail_dp_rx_caps;
1741 		}
1742 	} else {
1743 		sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1744 		if (sa->dp_rx == NULL) {
1745 			sfc_err(sa, "Rx datapath by caps %#x not found",
1746 				avail_caps);
1747 			rc = ENOENT;
1748 			goto fail_dp_rx;
1749 		}
1750 	}
1751 
1752 	sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1753 	if (sa->dp_rx_name == NULL) {
1754 		rc = ENOMEM;
1755 		goto fail_dp_rx_name;
1756 	}
1757 
1758 	sfc_info(sa, "use %s Rx datapath", sa->dp_rx_name);
1759 
1760 	dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1761 
1762 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1763 				sfc_kvarg_string_handler, &tx_name);
1764 	if (rc != 0)
1765 		goto fail_kvarg_tx_datapath;
1766 
1767 	if (tx_name != NULL) {
1768 		sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1769 		if (sa->dp_tx == NULL) {
1770 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1771 			rc = ENOENT;
1772 			goto fail_dp_tx;
1773 		}
1774 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1775 			sfc_err(sa,
1776 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1777 				tx_name);
1778 			rc = EINVAL;
1779 			goto fail_dp_tx_caps;
1780 		}
1781 	} else {
1782 		sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1783 		if (sa->dp_tx == NULL) {
1784 			sfc_err(sa, "Tx datapath by caps %#x not found",
1785 				avail_caps);
1786 			rc = ENOENT;
1787 			goto fail_dp_tx;
1788 		}
1789 	}
1790 
1791 	sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1792 	if (sa->dp_tx_name == NULL) {
1793 		rc = ENOMEM;
1794 		goto fail_dp_tx_name;
1795 	}
1796 
1797 	sfc_info(sa, "use %s Tx datapath", sa->dp_tx_name);
1798 
1799 	dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1800 
1801 	dev->dev_ops = &sfc_eth_dev_ops;
1802 
1803 	return 0;
1804 
1805 fail_dp_tx_name:
1806 fail_dp_tx_caps:
1807 	sa->dp_tx = NULL;
1808 
1809 fail_dp_tx:
1810 fail_kvarg_tx_datapath:
1811 	rte_free(sa->dp_rx_name);
1812 	sa->dp_rx_name = NULL;
1813 
1814 fail_dp_rx_name:
1815 fail_dp_rx_caps:
1816 	sa->dp_rx = NULL;
1817 
1818 fail_dp_rx:
1819 fail_kvarg_rx_datapath:
1820 	return rc;
1821 }
1822 
1823 static void
1824 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1825 {
1826 	struct sfc_adapter *sa = dev->data->dev_private;
1827 
1828 	dev->dev_ops = NULL;
1829 	dev->rx_pkt_burst = NULL;
1830 	dev->tx_pkt_burst = NULL;
1831 
1832 	rte_free(sa->dp_tx_name);
1833 	sa->dp_tx_name = NULL;
1834 	sa->dp_tx = NULL;
1835 
1836 	rte_free(sa->dp_rx_name);
1837 	sa->dp_rx_name = NULL;
1838 	sa->dp_rx = NULL;
1839 }
1840 
1841 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1842 	.rxq_info_get			= sfc_rx_queue_info_get,
1843 	.txq_info_get			= sfc_tx_queue_info_get,
1844 };
1845 
1846 static int
1847 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1848 {
1849 	/*
1850 	 * Device private data has really many process-local pointers.
1851 	 * Below code should be extremely careful to use data located
1852 	 * in shared memory only.
1853 	 */
1854 	struct sfc_adapter *sa = dev->data->dev_private;
1855 	const struct sfc_dp_rx *dp_rx;
1856 	const struct sfc_dp_tx *dp_tx;
1857 	int rc;
1858 
1859 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1860 	if (dp_rx == NULL) {
1861 		sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1862 		rc = ENOENT;
1863 		goto fail_dp_rx;
1864 	}
1865 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1866 		sfc_err(sa, "%s Rx datapath does not support multi-process",
1867 			sa->dp_tx_name);
1868 		rc = EINVAL;
1869 		goto fail_dp_rx_multi_process;
1870 	}
1871 
1872 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1873 	if (dp_tx == NULL) {
1874 		sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1875 		rc = ENOENT;
1876 		goto fail_dp_tx;
1877 	}
1878 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1879 		sfc_err(sa, "%s Tx datapath does not support multi-process",
1880 			sa->dp_tx_name);
1881 		rc = EINVAL;
1882 		goto fail_dp_tx_multi_process;
1883 	}
1884 
1885 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1886 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1887 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
1888 
1889 	return 0;
1890 
1891 fail_dp_tx_multi_process:
1892 fail_dp_tx:
1893 fail_dp_rx_multi_process:
1894 fail_dp_rx:
1895 	return rc;
1896 }
1897 
1898 static void
1899 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1900 {
1901 	dev->dev_ops = NULL;
1902 	dev->tx_pkt_burst = NULL;
1903 	dev->rx_pkt_burst = NULL;
1904 }
1905 
1906 static void
1907 sfc_register_dp(void)
1908 {
1909 	/* Register once */
1910 	if (TAILQ_EMPTY(&sfc_dp_head)) {
1911 		/* Prefer EF10 datapath */
1912 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1913 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1914 
1915 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1916 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1917 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1918 	}
1919 }
1920 
1921 static int
1922 sfc_eth_dev_init(struct rte_eth_dev *dev)
1923 {
1924 	struct sfc_adapter *sa = dev->data->dev_private;
1925 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1926 	int rc;
1927 	const efx_nic_cfg_t *encp;
1928 	const struct ether_addr *from;
1929 
1930 	sfc_register_dp();
1931 
1932 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1933 		return -sfc_eth_dev_secondary_set_ops(dev);
1934 
1935 	/* Required for logging */
1936 	sa->pci_addr = pci_dev->addr;
1937 	sa->port_id = dev->data->port_id;
1938 
1939 	sa->eth_dev = dev;
1940 
1941 	/* Copy PCI device info to the dev->data */
1942 	rte_eth_copy_pci_info(dev, pci_dev);
1943 
1944 	rc = sfc_kvargs_parse(sa);
1945 	if (rc != 0)
1946 		goto fail_kvargs_parse;
1947 
1948 	rc = sfc_kvargs_process(sa, SFC_KVARG_DEBUG_INIT,
1949 				sfc_kvarg_bool_handler, &sa->debug_init);
1950 	if (rc != 0)
1951 		goto fail_kvarg_debug_init;
1952 
1953 	sfc_log_init(sa, "entry");
1954 
1955 	dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1956 	if (dev->data->mac_addrs == NULL) {
1957 		rc = ENOMEM;
1958 		goto fail_mac_addrs;
1959 	}
1960 
1961 	sfc_adapter_lock_init(sa);
1962 	sfc_adapter_lock(sa);
1963 
1964 	sfc_log_init(sa, "probing");
1965 	rc = sfc_probe(sa);
1966 	if (rc != 0)
1967 		goto fail_probe;
1968 
1969 	sfc_log_init(sa, "set device ops");
1970 	rc = sfc_eth_dev_set_ops(dev);
1971 	if (rc != 0)
1972 		goto fail_set_ops;
1973 
1974 	sfc_log_init(sa, "attaching");
1975 	rc = sfc_attach(sa);
1976 	if (rc != 0)
1977 		goto fail_attach;
1978 
1979 	encp = efx_nic_cfg_get(sa->nic);
1980 
1981 	/*
1982 	 * The arguments are really reverse order in comparison to
1983 	 * Linux kernel. Copy from NIC config to Ethernet device data.
1984 	 */
1985 	from = (const struct ether_addr *)(encp->enc_mac_addr);
1986 	ether_addr_copy(from, &dev->data->mac_addrs[0]);
1987 
1988 	sfc_adapter_unlock(sa);
1989 
1990 	sfc_log_init(sa, "done");
1991 	return 0;
1992 
1993 fail_attach:
1994 	sfc_eth_dev_clear_ops(dev);
1995 
1996 fail_set_ops:
1997 	sfc_unprobe(sa);
1998 
1999 fail_probe:
2000 	sfc_adapter_unlock(sa);
2001 	sfc_adapter_lock_fini(sa);
2002 	rte_free(dev->data->mac_addrs);
2003 	dev->data->mac_addrs = NULL;
2004 
2005 fail_mac_addrs:
2006 fail_kvarg_debug_init:
2007 	sfc_kvargs_cleanup(sa);
2008 
2009 fail_kvargs_parse:
2010 	sfc_log_init(sa, "failed %d", rc);
2011 	SFC_ASSERT(rc > 0);
2012 	return -rc;
2013 }
2014 
2015 static int
2016 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2017 {
2018 	struct sfc_adapter *sa;
2019 
2020 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2021 		sfc_eth_dev_secondary_clear_ops(dev);
2022 		return 0;
2023 	}
2024 
2025 	sa = dev->data->dev_private;
2026 	sfc_log_init(sa, "entry");
2027 
2028 	sfc_adapter_lock(sa);
2029 
2030 	sfc_eth_dev_clear_ops(dev);
2031 
2032 	sfc_detach(sa);
2033 	sfc_unprobe(sa);
2034 
2035 	rte_free(dev->data->mac_addrs);
2036 	dev->data->mac_addrs = NULL;
2037 
2038 	sfc_kvargs_cleanup(sa);
2039 
2040 	sfc_adapter_unlock(sa);
2041 	sfc_adapter_lock_fini(sa);
2042 
2043 	sfc_log_init(sa, "done");
2044 
2045 	/* Required for logging, so cleanup last */
2046 	sa->eth_dev = NULL;
2047 	return 0;
2048 }
2049 
2050 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2051 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2052 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2053 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2054 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2055 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2056 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2057 	{ .vendor_id = 0 /* sentinel */ }
2058 };
2059 
2060 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2061 	struct rte_pci_device *pci_dev)
2062 {
2063 	return rte_eth_dev_pci_generic_probe(pci_dev,
2064 		sizeof(struct sfc_adapter), sfc_eth_dev_init);
2065 }
2066 
2067 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2068 {
2069 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2070 }
2071 
2072 static struct rte_pci_driver sfc_efx_pmd = {
2073 	.id_table = pci_id_sfc_efx_map,
2074 	.drv_flags =
2075 		RTE_PCI_DRV_INTR_LSC |
2076 		RTE_PCI_DRV_NEED_MAPPING,
2077 	.probe = sfc_eth_dev_pci_probe,
2078 	.remove = sfc_eth_dev_pci_remove,
2079 };
2080 
2081 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2082 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2083 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2084 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2085 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2086 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2087 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2088 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long> "
2089 	SFC_KVARG_MCDI_LOGGING "=" SFC_KVARG_VALUES_BOOL " "
2090 	SFC_KVARG_DEBUG_INIT "=" SFC_KVARG_VALUES_BOOL);
2091