xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision 1311ec23f6bfd3a71a76cd3ceaffdeec2ed623c5)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  *
3  * Copyright(c) 2019-2021 Xilinx, Inc.
4  * Copyright(c) 2016-2019 Solarflare Communications Inc.
5  *
6  * This software was jointly developed between OKTET Labs (under contract
7  * for Solarflare) and Solarflare Communications, Inc.
8  */
9 
10 #include <dev_driver.h>
11 #include <ethdev_driver.h>
12 #include <ethdev_pci.h>
13 #include <rte_pci.h>
14 #include <bus_pci_driver.h>
15 #include <rte_errno.h>
16 #include <rte_string_fns.h>
17 #include <rte_ether.h>
18 
19 #include "efx.h"
20 
21 #include "sfc.h"
22 #include "sfc_debug.h"
23 #include "sfc_log.h"
24 #include "sfc_kvargs.h"
25 #include "sfc_ev.h"
26 #include "sfc_rx.h"
27 #include "sfc_tx.h"
28 #include "sfc_flow.h"
29 #include "sfc_flow_tunnel.h"
30 #include "sfc_dp.h"
31 #include "sfc_dp_rx.h"
32 #include "sfc_repr.h"
33 #include "sfc_sw_stats.h"
34 #include "sfc_switch.h"
35 #include "sfc_nic_dma.h"
36 
37 #define SFC_XSTAT_ID_INVALID_VAL  UINT64_MAX
38 #define SFC_XSTAT_ID_INVALID_NAME '\0'
39 
40 static struct sfc_dp_list sfc_dp_head =
41 	TAILQ_HEAD_INITIALIZER(sfc_dp_head);
42 
43 
44 static void sfc_eth_dev_clear_ops(struct rte_eth_dev *dev);
45 
46 
47 static int
48 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
49 {
50 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
51 	efx_nic_fw_info_t enfi;
52 	int ret;
53 	int rc;
54 
55 	rc = efx_nic_get_fw_version(sa->nic, &enfi);
56 	if (rc != 0)
57 		return -rc;
58 
59 	ret = snprintf(fw_version, fw_size,
60 		       "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
61 		       enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
62 		       enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
63 	if (ret < 0)
64 		return ret;
65 
66 	if (enfi.enfi_dpcpu_fw_ids_valid) {
67 		size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
68 		int ret_extra;
69 
70 		ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
71 				     fw_size - dpcpu_fw_ids_offset,
72 				     " rx%" PRIx16 " tx%" PRIx16,
73 				     enfi.enfi_rx_dpcpu_fw_id,
74 				     enfi.enfi_tx_dpcpu_fw_id);
75 		if (ret_extra < 0)
76 			return ret_extra;
77 
78 		ret += ret_extra;
79 	}
80 
81 	if (fw_size < (size_t)(++ret))
82 		return ret;
83 	else
84 		return 0;
85 }
86 
87 static int
88 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
89 {
90 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
91 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
92 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
93 	struct sfc_rss *rss = &sas->rss;
94 	struct sfc_mae *mae = &sa->mae;
95 
96 	sfc_log_init(sa, "entry");
97 
98 	dev_info->min_mtu = RTE_ETHER_MIN_MTU;
99 	dev_info->max_mtu = EFX_MAC_SDU_MAX;
100 
101 	dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
102 
103 	dev_info->max_vfs = sa->sriov.num_vfs;
104 
105 	/* Autonegotiation may be disabled */
106 	dev_info->speed_capa = RTE_ETH_LINK_SPEED_FIXED;
107 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_1000FDX))
108 		dev_info->speed_capa |= RTE_ETH_LINK_SPEED_1G;
109 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_10000FDX))
110 		dev_info->speed_capa |= RTE_ETH_LINK_SPEED_10G;
111 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_25000FDX))
112 		dev_info->speed_capa |= RTE_ETH_LINK_SPEED_25G;
113 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_40000FDX))
114 		dev_info->speed_capa |= RTE_ETH_LINK_SPEED_40G;
115 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_50000FDX))
116 		dev_info->speed_capa |= RTE_ETH_LINK_SPEED_50G;
117 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_100000FDX))
118 		dev_info->speed_capa |= RTE_ETH_LINK_SPEED_100G;
119 
120 	dev_info->max_rx_queues = sa->rxq_max;
121 	dev_info->max_tx_queues = sa->txq_max;
122 
123 	/* By default packets are dropped if no descriptors are available */
124 	dev_info->default_rxconf.rx_drop_en = 1;
125 
126 	dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
127 
128 	/*
129 	 * rx_offload_capa includes both device and queue offloads since
130 	 * the latter may be requested on a per device basis which makes
131 	 * sense when some offloads are needed to be set on all queues.
132 	 */
133 	dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
134 				    dev_info->rx_queue_offload_capa;
135 
136 	dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
137 
138 	/*
139 	 * tx_offload_capa includes both device and queue offloads since
140 	 * the latter may be requested on a per device basis which makes
141 	 * sense when some offloads are needed to be set on all queues.
142 	 */
143 	dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
144 				    dev_info->tx_queue_offload_capa;
145 
146 	if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
147 		uint64_t rte_hf = 0;
148 		unsigned int i;
149 
150 		for (i = 0; i < rss->hf_map_nb_entries; ++i)
151 			rte_hf |= rss->hf_map[i].rte;
152 
153 		dev_info->reta_size = EFX_RSS_TBL_SIZE;
154 		dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
155 		dev_info->flow_type_rss_offloads = rte_hf;
156 	}
157 
158 	/* Initialize to hardware limits */
159 	dev_info->rx_desc_lim.nb_max = sa->rxq_max_entries;
160 	dev_info->rx_desc_lim.nb_min = sa->rxq_min_entries;
161 	/* The RXQ hardware requires that the descriptor count is a power
162 	 * of 2, but rx_desc_lim cannot properly describe that constraint.
163 	 */
164 	dev_info->rx_desc_lim.nb_align = sa->rxq_min_entries;
165 
166 	/* Initialize to hardware limits */
167 	dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
168 	dev_info->tx_desc_lim.nb_min = sa->txq_min_entries;
169 	/*
170 	 * The TXQ hardware requires that the descriptor count is a power
171 	 * of 2, but tx_desc_lim cannot properly describe that constraint
172 	 */
173 	dev_info->tx_desc_lim.nb_align = sa->txq_min_entries;
174 
175 	if (sap->dp_rx->get_dev_info != NULL)
176 		sap->dp_rx->get_dev_info(dev_info);
177 	if (sap->dp_tx->get_dev_info != NULL)
178 		sap->dp_tx->get_dev_info(dev_info);
179 
180 	dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
181 			     RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
182 	dev_info->dev_capa &= ~RTE_ETH_DEV_CAPA_FLOW_RULE_KEEP;
183 
184 	if (mae->status == SFC_MAE_STATUS_SUPPORTED ||
185 	    mae->status == SFC_MAE_STATUS_ADMIN) {
186 		dev_info->switch_info.name = dev->device->driver->name;
187 		dev_info->switch_info.domain_id = mae->switch_domain_id;
188 		dev_info->switch_info.port_id = mae->switch_port_id;
189 	}
190 
191 	return 0;
192 }
193 
194 static const uint32_t *
195 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev, size_t *no_of_elements)
196 {
197 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
198 
199 	return sap->dp_rx->supported_ptypes_get(sap->shared->tunnel_encaps,
200 						no_of_elements);
201 }
202 
203 static int
204 sfc_dev_configure(struct rte_eth_dev *dev)
205 {
206 	struct rte_eth_dev_data *dev_data = dev->data;
207 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
208 	int rc;
209 
210 	sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
211 		     dev_data->nb_rx_queues, dev_data->nb_tx_queues);
212 
213 	sfc_adapter_lock(sa);
214 	switch (sa->state) {
215 	case SFC_ETHDEV_CONFIGURED:
216 		/* FALLTHROUGH */
217 	case SFC_ETHDEV_INITIALIZED:
218 		rc = sfc_configure(sa);
219 		break;
220 	default:
221 		sfc_err(sa, "unexpected adapter state %u to configure",
222 			sa->state);
223 		rc = EINVAL;
224 		break;
225 	}
226 	sfc_adapter_unlock(sa);
227 
228 	sfc_log_init(sa, "done %d", rc);
229 	SFC_ASSERT(rc >= 0);
230 	return -rc;
231 }
232 
233 static int
234 sfc_dev_start(struct rte_eth_dev *dev)
235 {
236 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
237 	int rc;
238 
239 	sfc_log_init(sa, "entry");
240 
241 	sfc_adapter_lock(sa);
242 	rc = sfc_start(sa);
243 	sfc_adapter_unlock(sa);
244 
245 	sfc_log_init(sa, "done %d", rc);
246 	SFC_ASSERT(rc >= 0);
247 	return -rc;
248 }
249 
250 static void
251 sfc_dev_get_rte_link(struct rte_eth_dev *dev, int wait_to_complete,
252 		     struct rte_eth_link *link)
253 {
254 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
255 
256 	SFC_ASSERT(link != NULL);
257 
258 	if (sa->state != SFC_ETHDEV_STARTED) {
259 		sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, link);
260 	} else if (wait_to_complete) {
261 		efx_link_mode_t link_mode;
262 
263 		if (efx_port_poll(sa->nic, &link_mode) != 0)
264 			link_mode = EFX_LINK_UNKNOWN;
265 		sfc_port_link_mode_to_info(link_mode, link);
266 	} else {
267 		sfc_ev_mgmt_qpoll(sa);
268 		rte_eth_linkstatus_get(dev, link);
269 	}
270 }
271 
272 static int
273 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
274 {
275 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
276 	struct rte_eth_link current_link;
277 	int ret;
278 
279 	sfc_log_init(sa, "entry");
280 
281 	sfc_dev_get_rte_link(dev, wait_to_complete, &current_link);
282 
283 	ret = rte_eth_linkstatus_set(dev, &current_link);
284 	if (ret == 0)
285 		sfc_notice(sa, "Link status is %s",
286 			   current_link.link_status ? "UP" : "DOWN");
287 
288 	return ret;
289 }
290 
291 static int
292 sfc_dev_stop(struct rte_eth_dev *dev)
293 {
294 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
295 
296 	sfc_log_init(sa, "entry");
297 
298 	sfc_adapter_lock(sa);
299 	sfc_stop(sa);
300 	sfc_adapter_unlock(sa);
301 
302 	sfc_log_init(sa, "done");
303 
304 	return 0;
305 }
306 
307 static int
308 sfc_dev_set_link_up(struct rte_eth_dev *dev)
309 {
310 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
311 	int rc;
312 
313 	sfc_log_init(sa, "entry");
314 
315 	sfc_adapter_lock(sa);
316 	rc = sfc_start(sa);
317 	sfc_adapter_unlock(sa);
318 
319 	SFC_ASSERT(rc >= 0);
320 	return -rc;
321 }
322 
323 static int
324 sfc_dev_set_link_down(struct rte_eth_dev *dev)
325 {
326 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
327 
328 	sfc_log_init(sa, "entry");
329 
330 	sfc_adapter_lock(sa);
331 	sfc_stop(sa);
332 	sfc_adapter_unlock(sa);
333 
334 	return 0;
335 }
336 
337 static void
338 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
339 {
340 	free(dev->process_private);
341 	rte_eth_dev_release_port(dev);
342 }
343 
344 static int
345 sfc_dev_close(struct rte_eth_dev *dev)
346 {
347 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
348 
349 	sfc_log_init(sa, "entry");
350 
351 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
352 		sfc_eth_dev_secondary_clear_ops(dev);
353 		return 0;
354 	}
355 
356 	sfc_pre_detach(sa);
357 
358 	sfc_adapter_lock(sa);
359 	switch (sa->state) {
360 	case SFC_ETHDEV_STARTED:
361 		sfc_stop(sa);
362 		SFC_ASSERT(sa->state == SFC_ETHDEV_CONFIGURED);
363 		/* FALLTHROUGH */
364 	case SFC_ETHDEV_CONFIGURED:
365 		sfc_close(sa);
366 		SFC_ASSERT(sa->state == SFC_ETHDEV_INITIALIZED);
367 		/* FALLTHROUGH */
368 	case SFC_ETHDEV_INITIALIZED:
369 		break;
370 	default:
371 		sfc_err(sa, "unexpected adapter state %u on close", sa->state);
372 		break;
373 	}
374 
375 	/*
376 	 * Cleanup all resources.
377 	 * Rollback primary process sfc_eth_dev_init() below.
378 	 */
379 
380 	sfc_eth_dev_clear_ops(dev);
381 
382 	sfc_nic_dma_detach(sa);
383 	sfc_detach(sa);
384 	sfc_unprobe(sa);
385 
386 	sfc_kvargs_cleanup(sa);
387 
388 	sfc_adapter_unlock(sa);
389 	sfc_adapter_lock_fini(sa);
390 
391 	sfc_log_init(sa, "done");
392 
393 	/* Required for logging, so cleanup last */
394 	sa->eth_dev = NULL;
395 
396 	free(sa);
397 
398 	return 0;
399 }
400 
401 static int
402 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
403 		   boolean_t enabled)
404 {
405 	struct sfc_port *port;
406 	boolean_t *toggle;
407 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
408 	boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
409 	const char *desc = (allmulti) ? "all-multi" : "promiscuous";
410 	int rc = 0;
411 
412 	sfc_adapter_lock(sa);
413 
414 	port = &sa->port;
415 	toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
416 
417 	if (*toggle != enabled) {
418 		*toggle = enabled;
419 
420 		if (sfc_sa2shared(sa)->isolated) {
421 			sfc_warn(sa, "isolated mode is active on the port");
422 			sfc_warn(sa, "the change is to be applied on the next "
423 				     "start provided that isolated mode is "
424 				     "disabled prior the next start");
425 		} else if ((sa->state == SFC_ETHDEV_STARTED) &&
426 			   ((rc = sfc_set_rx_mode(sa)) != 0)) {
427 			*toggle = !(enabled);
428 			sfc_warn(sa, "Failed to %s %s mode, rc = %d",
429 				 ((enabled) ? "enable" : "disable"), desc, rc);
430 
431 			/*
432 			 * For promiscuous and all-multicast filters a
433 			 * permission failure should be reported as an
434 			 * unsupported filter.
435 			 */
436 			if (rc == EPERM)
437 				rc = ENOTSUP;
438 		}
439 	}
440 
441 	sfc_adapter_unlock(sa);
442 	return rc;
443 }
444 
445 static int
446 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
447 {
448 	int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
449 
450 	SFC_ASSERT(rc >= 0);
451 	return -rc;
452 }
453 
454 static int
455 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
456 {
457 	int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
458 
459 	SFC_ASSERT(rc >= 0);
460 	return -rc;
461 }
462 
463 static int
464 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
465 {
466 	int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
467 
468 	SFC_ASSERT(rc >= 0);
469 	return -rc;
470 }
471 
472 static int
473 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
474 {
475 	int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
476 
477 	SFC_ASSERT(rc >= 0);
478 	return -rc;
479 }
480 
481 static int
482 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid,
483 		   uint16_t nb_rx_desc, unsigned int socket_id,
484 		   const struct rte_eth_rxconf *rx_conf,
485 		   struct rte_mempool *mb_pool)
486 {
487 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
488 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
489 	sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
490 	struct sfc_rxq_info *rxq_info;
491 	sfc_sw_index_t sw_index;
492 	int rc;
493 
494 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
495 		     ethdev_qid, nb_rx_desc, socket_id);
496 
497 	sfc_adapter_lock(sa);
498 
499 	sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
500 	rc = sfc_rx_qinit(sa, sw_index, nb_rx_desc, socket_id,
501 			  rx_conf, mb_pool);
502 	if (rc != 0)
503 		goto fail_rx_qinit;
504 
505 	rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
506 	dev->data->rx_queues[ethdev_qid] = rxq_info->dp;
507 
508 	sfc_adapter_unlock(sa);
509 
510 	return 0;
511 
512 fail_rx_qinit:
513 	sfc_adapter_unlock(sa);
514 	SFC_ASSERT(rc > 0);
515 	return -rc;
516 }
517 
518 static void
519 sfc_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
520 {
521 	struct sfc_dp_rxq *dp_rxq = dev->data->rx_queues[qid];
522 	struct sfc_rxq *rxq;
523 	struct sfc_adapter *sa;
524 	sfc_sw_index_t sw_index;
525 
526 	if (dp_rxq == NULL)
527 		return;
528 
529 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
530 	sa = rxq->evq->sa;
531 	sfc_adapter_lock(sa);
532 
533 	sw_index = dp_rxq->dpq.queue_id;
534 
535 	sfc_log_init(sa, "RxQ=%u", sw_index);
536 
537 	sfc_rx_qfini(sa, sw_index);
538 
539 	sfc_adapter_unlock(sa);
540 }
541 
542 static int
543 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid,
544 		   uint16_t nb_tx_desc, unsigned int socket_id,
545 		   const struct rte_eth_txconf *tx_conf)
546 {
547 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
548 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
549 	struct sfc_txq_info *txq_info;
550 	sfc_sw_index_t sw_index;
551 	int rc;
552 
553 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
554 		     ethdev_qid, nb_tx_desc, socket_id);
555 
556 	sfc_adapter_lock(sa);
557 
558 	sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
559 	rc = sfc_tx_qinit(sa, sw_index, nb_tx_desc, socket_id, tx_conf);
560 	if (rc != 0)
561 		goto fail_tx_qinit;
562 
563 	txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
564 	dev->data->tx_queues[ethdev_qid] = txq_info->dp;
565 
566 	sfc_adapter_unlock(sa);
567 	return 0;
568 
569 fail_tx_qinit:
570 	sfc_adapter_unlock(sa);
571 	SFC_ASSERT(rc > 0);
572 	return -rc;
573 }
574 
575 static void
576 sfc_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
577 {
578 	struct sfc_dp_txq *dp_txq = dev->data->tx_queues[qid];
579 	struct sfc_txq *txq;
580 	sfc_sw_index_t sw_index;
581 	struct sfc_adapter *sa;
582 
583 	if (dp_txq == NULL)
584 		return;
585 
586 	txq = sfc_txq_by_dp_txq(dp_txq);
587 	sw_index = dp_txq->dpq.queue_id;
588 
589 	SFC_ASSERT(txq->evq != NULL);
590 	sa = txq->evq->sa;
591 
592 	sfc_log_init(sa, "TxQ = %u", sw_index);
593 
594 	sfc_adapter_lock(sa);
595 
596 	sfc_tx_qfini(sa, sw_index);
597 
598 	sfc_adapter_unlock(sa);
599 }
600 
601 static void
602 sfc_stats_get_dp_rx(struct sfc_adapter *sa, uint64_t *pkts, uint64_t *bytes)
603 {
604 	struct sfc_adapter_shared *sas = sfc_sa2shared(sa);
605 	uint64_t pkts_sum = 0;
606 	uint64_t bytes_sum = 0;
607 	unsigned int i;
608 
609 	for (i = 0; i < sas->ethdev_rxq_count; ++i) {
610 		struct sfc_rxq_info *rxq_info;
611 
612 		rxq_info = sfc_rxq_info_by_ethdev_qid(sas, i);
613 		if (rxq_info->state & SFC_RXQ_INITIALIZED) {
614 			union sfc_pkts_bytes qstats;
615 
616 			sfc_pkts_bytes_get(&rxq_info->dp->dpq.stats, &qstats);
617 			pkts_sum += qstats.pkts -
618 					sa->sw_stats.reset_rx_pkts[i];
619 			bytes_sum += qstats.bytes -
620 					sa->sw_stats.reset_rx_bytes[i];
621 		}
622 	}
623 
624 	*pkts = pkts_sum;
625 	*bytes = bytes_sum;
626 }
627 
628 static void
629 sfc_stats_get_dp_tx(struct sfc_adapter *sa, uint64_t *pkts, uint64_t *bytes)
630 {
631 	struct sfc_adapter_shared *sas = sfc_sa2shared(sa);
632 	uint64_t pkts_sum = 0;
633 	uint64_t bytes_sum = 0;
634 	unsigned int i;
635 
636 	for (i = 0; i < sas->ethdev_txq_count; ++i) {
637 		struct sfc_txq_info *txq_info;
638 
639 		txq_info = sfc_txq_info_by_ethdev_qid(sas, i);
640 		if (txq_info->state & SFC_TXQ_INITIALIZED) {
641 			union sfc_pkts_bytes qstats;
642 
643 			sfc_pkts_bytes_get(&txq_info->dp->dpq.stats, &qstats);
644 			pkts_sum += qstats.pkts -
645 					sa->sw_stats.reset_tx_pkts[i];
646 			bytes_sum += qstats.bytes -
647 					sa->sw_stats.reset_tx_bytes[i];
648 		}
649 	}
650 
651 	*pkts = pkts_sum;
652 	*bytes = bytes_sum;
653 }
654 
655 /*
656  * Some statistics are computed as A - B where A and B each increase
657  * monotonically with some hardware counter(s) and the counters are read
658  * asynchronously.
659  *
660  * If packet X is counted in A, but not counted in B yet, computed value is
661  * greater than real.
662  *
663  * If packet X is not counted in A at the moment of reading the counter,
664  * but counted in B at the moment of reading the counter, computed value
665  * is less than real.
666  *
667  * However, counter which grows backward is worse evil than slightly wrong
668  * value. So, let's try to guarantee that it never happens except may be
669  * the case when the MAC stats are zeroed as a result of a NIC reset.
670  */
671 static void
672 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
673 {
674 	if ((int64_t)(newval - *stat) > 0 || newval == 0)
675 		*stat = newval;
676 }
677 
678 static int
679 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
680 {
681 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
682 	bool have_dp_rx_stats = sap->dp_rx->features & SFC_DP_RX_FEAT_STATS;
683 	bool have_dp_tx_stats = sap->dp_tx->features & SFC_DP_TX_FEAT_STATS;
684 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
685 	struct sfc_port *port = &sa->port;
686 	uint64_t *mac_stats;
687 	int ret;
688 
689 	sfc_adapter_lock(sa);
690 
691 	if (have_dp_rx_stats) {
692 		sfc_stats_get_dp_rx(sa, &stats->ipackets, &stats->ibytes);
693 		if (dev->data->dev_conf.rxmode.offloads &
694 		    RTE_ETH_RX_OFFLOAD_KEEP_CRC) {
695 			stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
696 		}
697 	}
698 	if (have_dp_tx_stats)
699 		sfc_stats_get_dp_tx(sa, &stats->opackets, &stats->obytes);
700 
701 	ret = sfc_port_update_mac_stats(sa, B_FALSE);
702 	if (ret != 0)
703 		goto unlock;
704 
705 	mac_stats = port->mac_stats_buf;
706 
707 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
708 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
709 		if (!have_dp_rx_stats) {
710 			stats->ipackets =
711 				mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
712 				mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
713 				mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
714 			stats->ibytes =
715 				mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
716 				mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
717 				mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
718 
719 			/* CRC is included in these stats, but shouldn't be */
720 			stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
721 		}
722 		if (!have_dp_tx_stats) {
723 			stats->opackets =
724 				mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
725 				mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
726 				mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
727 			stats->obytes =
728 				mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
729 				mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
730 				mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
731 
732 			/* CRC is included in these stats, but shouldn't be */
733 			stats->obytes -= stats->opackets * RTE_ETHER_CRC_LEN;
734 		}
735 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
736 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
737 	} else {
738 		if (!have_dp_tx_stats) {
739 			stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
740 			stats->obytes = mac_stats[EFX_MAC_TX_OCTETS] -
741 				mac_stats[EFX_MAC_TX_PKTS] * RTE_ETHER_CRC_LEN;
742 		}
743 
744 		/*
745 		 * Take into account stats which are whenever supported
746 		 * on EF10. If some stat is not supported by current
747 		 * firmware variant or HW revision, it is guaranteed
748 		 * to be zero in mac_stats.
749 		 */
750 		stats->imissed =
751 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
752 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
753 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
754 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
755 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
756 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
757 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
758 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
759 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
760 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
761 		stats->ierrors =
762 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
763 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
764 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
765 		/* no oerrors counters supported on EF10 */
766 
767 		if (!have_dp_rx_stats) {
768 			/* Exclude missed, errors and pauses from Rx packets */
769 			sfc_update_diff_stat(&port->ipackets,
770 				mac_stats[EFX_MAC_RX_PKTS] -
771 				mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
772 				stats->imissed - stats->ierrors);
773 			stats->ipackets = port->ipackets;
774 			stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS] -
775 				mac_stats[EFX_MAC_RX_PKTS] * RTE_ETHER_CRC_LEN;
776 		}
777 	}
778 
779 unlock:
780 	sfc_adapter_unlock(sa);
781 	SFC_ASSERT(ret >= 0);
782 	return -ret;
783 }
784 
785 static int
786 sfc_stats_reset(struct rte_eth_dev *dev)
787 {
788 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
789 	struct sfc_port *port = &sa->port;
790 	int rc;
791 
792 	sfc_adapter_lock(sa);
793 
794 	if (sa->state != SFC_ETHDEV_STARTED) {
795 		/*
796 		 * The operation cannot be done if port is not started; it
797 		 * will be scheduled to be done during the next port start
798 		 */
799 		port->mac_stats_reset_pending = B_TRUE;
800 		sfc_adapter_unlock(sa);
801 		return 0;
802 	}
803 
804 	rc = sfc_port_reset_mac_stats(sa);
805 	if (rc != 0)
806 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
807 
808 	sfc_sw_xstats_reset(sa);
809 
810 	sfc_adapter_unlock(sa);
811 
812 	SFC_ASSERT(rc >= 0);
813 	return -rc;
814 }
815 
816 static unsigned int
817 sfc_xstats_get_nb_supported(struct sfc_adapter *sa)
818 {
819 	struct sfc_port *port = &sa->port;
820 	unsigned int nb_supported;
821 
822 	sfc_adapter_lock(sa);
823 	nb_supported = port->mac_stats_nb_supported +
824 		       sfc_sw_xstats_get_nb_supported(sa);
825 	sfc_adapter_unlock(sa);
826 
827 	return nb_supported;
828 }
829 
830 static int
831 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
832 	       unsigned int xstats_count)
833 {
834 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
835 	unsigned int nb_written = 0;
836 	unsigned int nb_supported = 0;
837 	int rc;
838 
839 	if (unlikely(xstats == NULL))
840 		return sfc_xstats_get_nb_supported(sa);
841 
842 	rc = sfc_port_get_mac_stats(sa, xstats, xstats_count, &nb_written);
843 	if (rc < 0)
844 		return rc;
845 
846 	nb_supported = rc;
847 	sfc_sw_xstats_get_vals(sa, xstats, xstats_count, &nb_written,
848 			       &nb_supported);
849 
850 	return nb_supported;
851 }
852 
853 static int
854 sfc_xstats_get_names(struct rte_eth_dev *dev,
855 		     struct rte_eth_xstat_name *xstats_names,
856 		     unsigned int xstats_count)
857 {
858 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
859 	struct sfc_port *port = &sa->port;
860 	unsigned int i;
861 	unsigned int nstats = 0;
862 	unsigned int nb_written = 0;
863 	int ret;
864 
865 	if (unlikely(xstats_names == NULL))
866 		return sfc_xstats_get_nb_supported(sa);
867 
868 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
869 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
870 			if (nstats < xstats_count) {
871 				strlcpy(xstats_names[nstats].name,
872 					efx_mac_stat_name(sa->nic, i),
873 					sizeof(xstats_names[0].name));
874 				nb_written++;
875 			}
876 			nstats++;
877 		}
878 	}
879 
880 	ret = sfc_sw_xstats_get_names(sa, xstats_names, xstats_count,
881 				      &nb_written, &nstats);
882 	if (ret != 0) {
883 		SFC_ASSERT(ret < 0);
884 		return ret;
885 	}
886 
887 	return nstats;
888 }
889 
890 static int
891 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
892 		     uint64_t *values, unsigned int n)
893 {
894 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
895 	struct sfc_port *port = &sa->port;
896 	unsigned int nb_supported;
897 	unsigned int i;
898 	int rc;
899 
900 	if (unlikely(ids == NULL || values == NULL))
901 		return -EINVAL;
902 
903 	/*
904 	 * Values array could be filled in nonsequential order. Fill values with
905 	 * constant indicating invalid ID first.
906 	 */
907 	for (i = 0; i < n; i++)
908 		values[i] = SFC_XSTAT_ID_INVALID_VAL;
909 
910 	rc = sfc_port_get_mac_stats_by_id(sa, ids, values, n);
911 	if (rc != 0)
912 		return rc;
913 
914 	nb_supported = port->mac_stats_nb_supported;
915 	sfc_sw_xstats_get_vals_by_id(sa, ids, values, n, &nb_supported);
916 
917 	/* Return number of written stats before invalid ID is encountered. */
918 	for (i = 0; i < n; i++) {
919 		if (values[i] == SFC_XSTAT_ID_INVALID_VAL)
920 			return i;
921 	}
922 
923 	return n;
924 }
925 
926 static int
927 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
928 			   const uint64_t *ids,
929 			   struct rte_eth_xstat_name *xstats_names,
930 			   unsigned int size)
931 {
932 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
933 	struct sfc_port *port = &sa->port;
934 	unsigned int nb_supported;
935 	unsigned int i;
936 	int ret;
937 
938 	if (unlikely(xstats_names == NULL && ids != NULL) ||
939 	    unlikely(xstats_names != NULL && ids == NULL))
940 		return -EINVAL;
941 
942 	if (unlikely(xstats_names == NULL && ids == NULL))
943 		return sfc_xstats_get_nb_supported(sa);
944 
945 	/*
946 	 * Names array could be filled in nonsequential order. Fill names with
947 	 * string indicating invalid ID first.
948 	 */
949 	for (i = 0; i < size; i++)
950 		xstats_names[i].name[0] = SFC_XSTAT_ID_INVALID_NAME;
951 
952 	sfc_adapter_lock(sa);
953 
954 	SFC_ASSERT(port->mac_stats_nb_supported <=
955 		   RTE_DIM(port->mac_stats_by_id));
956 
957 	for (i = 0; i < size; i++) {
958 		if (ids[i] < port->mac_stats_nb_supported) {
959 			strlcpy(xstats_names[i].name,
960 				efx_mac_stat_name(sa->nic,
961 						 port->mac_stats_by_id[ids[i]]),
962 				sizeof(xstats_names[0].name));
963 		}
964 	}
965 
966 	nb_supported = port->mac_stats_nb_supported;
967 
968 	sfc_adapter_unlock(sa);
969 
970 	ret = sfc_sw_xstats_get_names_by_id(sa, ids, xstats_names, size,
971 					    &nb_supported);
972 	if (ret != 0) {
973 		SFC_ASSERT(ret < 0);
974 		return ret;
975 	}
976 
977 	/* Return number of written names before invalid ID is encountered. */
978 	for (i = 0; i < size; i++) {
979 		if (xstats_names[i].name[0] == SFC_XSTAT_ID_INVALID_NAME)
980 			return i;
981 	}
982 
983 	return size;
984 }
985 
986 static int
987 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
988 {
989 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
990 	unsigned int wanted_fc, link_fc;
991 
992 	memset(fc_conf, 0, sizeof(*fc_conf));
993 
994 	sfc_adapter_lock(sa);
995 
996 	if (sa->state == SFC_ETHDEV_STARTED)
997 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
998 	else
999 		link_fc = sa->port.flow_ctrl;
1000 
1001 	switch (link_fc) {
1002 	case 0:
1003 		fc_conf->mode = RTE_ETH_FC_NONE;
1004 		break;
1005 	case EFX_FCNTL_RESPOND:
1006 		fc_conf->mode = RTE_ETH_FC_RX_PAUSE;
1007 		break;
1008 	case EFX_FCNTL_GENERATE:
1009 		fc_conf->mode = RTE_ETH_FC_TX_PAUSE;
1010 		break;
1011 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
1012 		fc_conf->mode = RTE_ETH_FC_FULL;
1013 		break;
1014 	default:
1015 		sfc_err(sa, "%s: unexpected flow control value %#x",
1016 			__func__, link_fc);
1017 	}
1018 
1019 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
1020 
1021 	sfc_adapter_unlock(sa);
1022 
1023 	return 0;
1024 }
1025 
1026 static int
1027 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1028 {
1029 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1030 	struct sfc_port *port = &sa->port;
1031 	unsigned int fcntl;
1032 	int rc;
1033 
1034 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
1035 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
1036 	    fc_conf->mac_ctrl_frame_fwd != 0) {
1037 		sfc_err(sa, "unsupported flow control settings specified");
1038 		rc = EINVAL;
1039 		goto fail_inval;
1040 	}
1041 
1042 	switch (fc_conf->mode) {
1043 	case RTE_ETH_FC_NONE:
1044 		fcntl = 0;
1045 		break;
1046 	case RTE_ETH_FC_RX_PAUSE:
1047 		fcntl = EFX_FCNTL_RESPOND;
1048 		break;
1049 	case RTE_ETH_FC_TX_PAUSE:
1050 		fcntl = EFX_FCNTL_GENERATE;
1051 		break;
1052 	case RTE_ETH_FC_FULL:
1053 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
1054 		break;
1055 	default:
1056 		rc = EINVAL;
1057 		goto fail_inval;
1058 	}
1059 
1060 	sfc_adapter_lock(sa);
1061 
1062 	if (sa->state == SFC_ETHDEV_STARTED) {
1063 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
1064 		if (rc != 0)
1065 			goto fail_mac_fcntl_set;
1066 	}
1067 
1068 	port->flow_ctrl = fcntl;
1069 	port->flow_ctrl_autoneg = fc_conf->autoneg;
1070 
1071 	sfc_adapter_unlock(sa);
1072 
1073 	return 0;
1074 
1075 fail_mac_fcntl_set:
1076 	sfc_adapter_unlock(sa);
1077 fail_inval:
1078 	SFC_ASSERT(rc > 0);
1079 	return -rc;
1080 }
1081 
1082 static int
1083 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
1084 {
1085 	struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
1086 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1087 	boolean_t scatter_enabled;
1088 	const char *error;
1089 	unsigned int i;
1090 
1091 	for (i = 0; i < sas->rxq_count; i++) {
1092 		if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
1093 			continue;
1094 
1095 		scatter_enabled = (sas->rxq_info[i].type_flags &
1096 				   EFX_RXQ_FLAG_SCATTER);
1097 
1098 		if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
1099 					  encp->enc_rx_prefix_size,
1100 					  scatter_enabled,
1101 					  encp->enc_rx_scatter_max, &error)) {
1102 			sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
1103 				error);
1104 			return EINVAL;
1105 		}
1106 	}
1107 
1108 	return 0;
1109 }
1110 
1111 static int
1112 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
1113 {
1114 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1115 	size_t pdu = EFX_MAC_PDU(mtu);
1116 	size_t old_pdu;
1117 	int rc;
1118 
1119 	sfc_log_init(sa, "mtu=%u", mtu);
1120 
1121 	rc = EINVAL;
1122 	if (pdu < EFX_MAC_PDU_MIN) {
1123 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
1124 			(unsigned int)mtu, (unsigned int)pdu,
1125 			EFX_MAC_PDU_MIN);
1126 		goto fail_inval;
1127 	}
1128 	if (pdu > EFX_MAC_PDU_MAX) {
1129 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
1130 			(unsigned int)mtu, (unsigned int)pdu,
1131 			(unsigned int)EFX_MAC_PDU_MAX);
1132 		goto fail_inval;
1133 	}
1134 
1135 	sfc_adapter_lock(sa);
1136 
1137 	rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
1138 	if (rc != 0)
1139 		goto fail_check_scatter;
1140 
1141 	if (pdu != sa->port.pdu) {
1142 		if (sa->state == SFC_ETHDEV_STARTED) {
1143 			sfc_stop(sa);
1144 
1145 			old_pdu = sa->port.pdu;
1146 			sa->port.pdu = pdu;
1147 			rc = sfc_start(sa);
1148 			if (rc != 0)
1149 				goto fail_start;
1150 		} else {
1151 			sa->port.pdu = pdu;
1152 		}
1153 	}
1154 
1155 	sfc_adapter_unlock(sa);
1156 
1157 	sfc_log_init(sa, "done");
1158 	return 0;
1159 
1160 fail_start:
1161 	sa->port.pdu = old_pdu;
1162 	if (sfc_start(sa) != 0)
1163 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1164 			"PDU max size - port is stopped",
1165 			(unsigned int)pdu, (unsigned int)old_pdu);
1166 
1167 fail_check_scatter:
1168 	sfc_adapter_unlock(sa);
1169 
1170 fail_inval:
1171 	sfc_log_init(sa, "failed %d", rc);
1172 	SFC_ASSERT(rc > 0);
1173 	return -rc;
1174 }
1175 static int
1176 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1177 {
1178 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1179 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1180 	struct sfc_port *port = &sa->port;
1181 	struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1182 	int rc = 0;
1183 
1184 	sfc_adapter_lock(sa);
1185 
1186 	if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr))
1187 		goto unlock;
1188 
1189 	/*
1190 	 * Copy the address to the device private data so that
1191 	 * it could be recalled in the case of adapter restart.
1192 	 */
1193 	rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1194 
1195 	/*
1196 	 * Neither of the two following checks can return
1197 	 * an error. The new MAC address is preserved in
1198 	 * the device private data and can be activated
1199 	 * on the next port start if the user prevents
1200 	 * isolated mode from being enabled.
1201 	 */
1202 	if (sfc_sa2shared(sa)->isolated) {
1203 		sfc_warn(sa, "isolated mode is active on the port");
1204 		sfc_warn(sa, "will not set MAC address");
1205 		goto unlock;
1206 	}
1207 
1208 	if (sa->state != SFC_ETHDEV_STARTED) {
1209 		sfc_notice(sa, "the port is not started");
1210 		sfc_notice(sa, "the new MAC address will be set on port start");
1211 
1212 		goto unlock;
1213 	}
1214 
1215 	if (encp->enc_allow_set_mac_with_installed_filters) {
1216 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1217 		if (rc != 0) {
1218 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1219 			goto unlock;
1220 		}
1221 
1222 		/*
1223 		 * Changing the MAC address by means of MCDI request
1224 		 * has no effect on received traffic, therefore
1225 		 * we also need to update unicast filters
1226 		 */
1227 		rc = sfc_set_rx_mode_unchecked(sa);
1228 		if (rc != 0) {
1229 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
1230 			/* Rollback the old address */
1231 			(void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1232 			(void)sfc_set_rx_mode_unchecked(sa);
1233 		}
1234 	} else {
1235 		sfc_warn(sa, "cannot set MAC address with filters installed");
1236 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1237 		sfc_warn(sa, "(some traffic may be dropped)");
1238 
1239 		/*
1240 		 * Since setting MAC address with filters installed is not
1241 		 * allowed on the adapter, the new MAC address will be set
1242 		 * by means of adapter restart. sfc_start() shall retrieve
1243 		 * the new address from the device private data and set it.
1244 		 */
1245 		sfc_stop(sa);
1246 		rc = sfc_start(sa);
1247 		if (rc != 0)
1248 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1249 	}
1250 
1251 unlock:
1252 	if (rc != 0)
1253 		rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1254 
1255 	sfc_adapter_unlock(sa);
1256 
1257 	SFC_ASSERT(rc >= 0);
1258 	return -rc;
1259 }
1260 
1261 
1262 static int
1263 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1264 		struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1265 {
1266 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1267 	struct sfc_port *port = &sa->port;
1268 	uint8_t *mc_addrs = port->mcast_addrs;
1269 	int rc;
1270 	unsigned int i;
1271 
1272 	if (sfc_sa2shared(sa)->isolated) {
1273 		sfc_err(sa, "isolated mode is active on the port");
1274 		sfc_err(sa, "will not set multicast address list");
1275 		return -ENOTSUP;
1276 	}
1277 
1278 	if (mc_addrs == NULL)
1279 		return -ENOBUFS;
1280 
1281 	if (nb_mc_addr > port->max_mcast_addrs) {
1282 		sfc_err(sa, "too many multicast addresses: %u > %u",
1283 			 nb_mc_addr, port->max_mcast_addrs);
1284 		return -EINVAL;
1285 	}
1286 
1287 	for (i = 0; i < nb_mc_addr; ++i) {
1288 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1289 				 EFX_MAC_ADDR_LEN);
1290 		mc_addrs += EFX_MAC_ADDR_LEN;
1291 	}
1292 
1293 	port->nb_mcast_addrs = nb_mc_addr;
1294 
1295 	if (sa->state != SFC_ETHDEV_STARTED)
1296 		return 0;
1297 
1298 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1299 					port->nb_mcast_addrs);
1300 	if (rc != 0)
1301 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1302 
1303 	SFC_ASSERT(rc >= 0);
1304 	return -rc;
1305 }
1306 
1307 /*
1308  * The function is used by the secondary process as well. It must not
1309  * use any process-local pointers from the adapter data.
1310  */
1311 static void
1312 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1313 		      struct rte_eth_rxq_info *qinfo)
1314 {
1315 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1316 	sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1317 	struct sfc_rxq_info *rxq_info;
1318 
1319 	rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1320 
1321 	qinfo->mp = rxq_info->refill_mb_pool;
1322 	qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1323 	qinfo->conf.rx_drop_en = 1;
1324 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1325 	qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1326 	if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1327 		qinfo->conf.offloads |= RTE_ETH_RX_OFFLOAD_SCATTER;
1328 		qinfo->scattered_rx = 1;
1329 	}
1330 	qinfo->nb_desc = rxq_info->entries;
1331 }
1332 
1333 /*
1334  * The function is used by the secondary process as well. It must not
1335  * use any process-local pointers from the adapter data.
1336  */
1337 static void
1338 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1339 		      struct rte_eth_txq_info *qinfo)
1340 {
1341 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1342 	struct sfc_txq_info *txq_info;
1343 
1344 	SFC_ASSERT(ethdev_qid < sas->ethdev_txq_count);
1345 
1346 	txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1347 
1348 	memset(qinfo, 0, sizeof(*qinfo));
1349 
1350 	qinfo->conf.offloads = txq_info->offloads;
1351 	qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1352 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1353 	qinfo->nb_desc = txq_info->entries;
1354 }
1355 
1356 /*
1357  * The function is used by the secondary process as well. It must not
1358  * use any process-local pointers from the adapter data.
1359  */
1360 static uint32_t
1361 sfc_rx_queue_count(void *rx_queue)
1362 {
1363 	struct sfc_dp_rxq *dp_rxq = rx_queue;
1364 	const struct sfc_dp_rx *dp_rx;
1365 	struct sfc_rxq_info *rxq_info;
1366 
1367 	dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1368 	rxq_info = sfc_rxq_info_by_dp_rxq(dp_rxq);
1369 
1370 	if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1371 		return 0;
1372 
1373 	return dp_rx->qdesc_npending(dp_rxq);
1374 }
1375 
1376 /*
1377  * The function is used by the secondary process as well. It must not
1378  * use any process-local pointers from the adapter data.
1379  */
1380 static int
1381 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1382 {
1383 	struct sfc_dp_rxq *dp_rxq = queue;
1384 	const struct sfc_dp_rx *dp_rx;
1385 
1386 	dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1387 
1388 	return dp_rx->qdesc_status(dp_rxq, offset);
1389 }
1390 
1391 /*
1392  * The function is used by the secondary process as well. It must not
1393  * use any process-local pointers from the adapter data.
1394  */
1395 static int
1396 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1397 {
1398 	struct sfc_dp_txq *dp_txq = queue;
1399 	const struct sfc_dp_tx *dp_tx;
1400 
1401 	dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1402 
1403 	return dp_tx->qdesc_status(dp_txq, offset);
1404 }
1405 
1406 static int
1407 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1408 {
1409 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1410 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1411 	sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1412 	struct sfc_rxq_info *rxq_info;
1413 	sfc_sw_index_t sw_index;
1414 	int rc;
1415 
1416 	sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1417 
1418 	sfc_adapter_lock(sa);
1419 
1420 	rc = EINVAL;
1421 	if (sa->state != SFC_ETHDEV_STARTED)
1422 		goto fail_not_started;
1423 
1424 	rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1425 	if (rxq_info->state != SFC_RXQ_INITIALIZED)
1426 		goto fail_not_setup;
1427 
1428 	sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1429 	rc = sfc_rx_qstart(sa, sw_index);
1430 	if (rc != 0)
1431 		goto fail_rx_qstart;
1432 
1433 	rxq_info->deferred_started = B_TRUE;
1434 
1435 	sfc_adapter_unlock(sa);
1436 
1437 	return 0;
1438 
1439 fail_rx_qstart:
1440 fail_not_setup:
1441 fail_not_started:
1442 	sfc_adapter_unlock(sa);
1443 	SFC_ASSERT(rc > 0);
1444 	return -rc;
1445 }
1446 
1447 static int
1448 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1449 {
1450 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1451 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1452 	sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1453 	struct sfc_rxq_info *rxq_info;
1454 	sfc_sw_index_t sw_index;
1455 
1456 	sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1457 
1458 	sfc_adapter_lock(sa);
1459 
1460 	sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1461 	sfc_rx_qstop(sa, sw_index);
1462 
1463 	rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1464 	rxq_info->deferred_started = B_FALSE;
1465 
1466 	sfc_adapter_unlock(sa);
1467 
1468 	return 0;
1469 }
1470 
1471 static int
1472 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1473 {
1474 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1475 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1476 	struct sfc_txq_info *txq_info;
1477 	sfc_sw_index_t sw_index;
1478 	int rc;
1479 
1480 	sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1481 
1482 	sfc_adapter_lock(sa);
1483 
1484 	rc = EINVAL;
1485 	if (sa->state != SFC_ETHDEV_STARTED)
1486 		goto fail_not_started;
1487 
1488 	txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1489 	if (txq_info->state != SFC_TXQ_INITIALIZED)
1490 		goto fail_not_setup;
1491 
1492 	sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1493 	rc = sfc_tx_qstart(sa, sw_index);
1494 	if (rc != 0)
1495 		goto fail_tx_qstart;
1496 
1497 	txq_info->deferred_started = B_TRUE;
1498 
1499 	sfc_adapter_unlock(sa);
1500 	return 0;
1501 
1502 fail_tx_qstart:
1503 
1504 fail_not_setup:
1505 fail_not_started:
1506 	sfc_adapter_unlock(sa);
1507 	SFC_ASSERT(rc > 0);
1508 	return -rc;
1509 }
1510 
1511 static int
1512 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1513 {
1514 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1515 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1516 	struct sfc_txq_info *txq_info;
1517 	sfc_sw_index_t sw_index;
1518 
1519 	sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1520 
1521 	sfc_adapter_lock(sa);
1522 
1523 	sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1524 	sfc_tx_qstop(sa, sw_index);
1525 
1526 	txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1527 	txq_info->deferred_started = B_FALSE;
1528 
1529 	sfc_adapter_unlock(sa);
1530 	return 0;
1531 }
1532 
1533 static efx_tunnel_protocol_t
1534 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1535 {
1536 	switch (rte_type) {
1537 	case RTE_ETH_TUNNEL_TYPE_VXLAN:
1538 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1539 	case RTE_ETH_TUNNEL_TYPE_GENEVE:
1540 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1541 	default:
1542 		return EFX_TUNNEL_NPROTOS;
1543 	}
1544 }
1545 
1546 enum sfc_udp_tunnel_op_e {
1547 	SFC_UDP_TUNNEL_ADD_PORT,
1548 	SFC_UDP_TUNNEL_DEL_PORT,
1549 };
1550 
1551 static int
1552 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1553 		      struct rte_eth_udp_tunnel *tunnel_udp,
1554 		      enum sfc_udp_tunnel_op_e op)
1555 {
1556 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1557 	efx_tunnel_protocol_t tunnel_proto;
1558 	int rc;
1559 
1560 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1561 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1562 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1563 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1564 
1565 	tunnel_proto =
1566 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1567 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1568 		rc = ENOTSUP;
1569 		goto fail_bad_proto;
1570 	}
1571 
1572 	sfc_adapter_lock(sa);
1573 
1574 	switch (op) {
1575 	case SFC_UDP_TUNNEL_ADD_PORT:
1576 		rc = efx_tunnel_config_udp_add(sa->nic,
1577 					       tunnel_udp->udp_port,
1578 					       tunnel_proto);
1579 		break;
1580 	case SFC_UDP_TUNNEL_DEL_PORT:
1581 		rc = efx_tunnel_config_udp_remove(sa->nic,
1582 						  tunnel_udp->udp_port,
1583 						  tunnel_proto);
1584 		break;
1585 	default:
1586 		rc = EINVAL;
1587 		goto fail_bad_op;
1588 	}
1589 
1590 	if (rc != 0)
1591 		goto fail_op;
1592 
1593 	if (sa->state == SFC_ETHDEV_STARTED) {
1594 		rc = efx_tunnel_reconfigure(sa->nic);
1595 		if (rc == EAGAIN) {
1596 			/*
1597 			 * Configuration is accepted by FW and MC reboot
1598 			 * is initiated to apply the changes. MC reboot
1599 			 * will be handled in a usual way (MC reboot
1600 			 * event on management event queue and adapter
1601 			 * restart).
1602 			 */
1603 			rc = 0;
1604 		} else if (rc != 0) {
1605 			goto fail_reconfigure;
1606 		}
1607 	}
1608 
1609 	sfc_adapter_unlock(sa);
1610 	return 0;
1611 
1612 fail_reconfigure:
1613 	/* Remove/restore entry since the change makes the trouble */
1614 	switch (op) {
1615 	case SFC_UDP_TUNNEL_ADD_PORT:
1616 		(void)efx_tunnel_config_udp_remove(sa->nic,
1617 						   tunnel_udp->udp_port,
1618 						   tunnel_proto);
1619 		break;
1620 	case SFC_UDP_TUNNEL_DEL_PORT:
1621 		(void)efx_tunnel_config_udp_add(sa->nic,
1622 						tunnel_udp->udp_port,
1623 						tunnel_proto);
1624 		break;
1625 	}
1626 
1627 fail_op:
1628 fail_bad_op:
1629 	sfc_adapter_unlock(sa);
1630 
1631 fail_bad_proto:
1632 	SFC_ASSERT(rc > 0);
1633 	return -rc;
1634 }
1635 
1636 static int
1637 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1638 			    struct rte_eth_udp_tunnel *tunnel_udp)
1639 {
1640 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1641 }
1642 
1643 static int
1644 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1645 			    struct rte_eth_udp_tunnel *tunnel_udp)
1646 {
1647 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1648 }
1649 
1650 /*
1651  * The function is used by the secondary process as well. It must not
1652  * use any process-local pointers from the adapter data.
1653  */
1654 static int
1655 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1656 			  struct rte_eth_rss_conf *rss_conf)
1657 {
1658 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1659 	struct sfc_rss *rss = &sas->rss;
1660 
1661 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1662 		return -ENOTSUP;
1663 
1664 	/*
1665 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1666 	 * hence, conversion is done here to derive a correct set of RTE_ETH_RSS
1667 	 * flags which corresponds to the active EFX configuration stored
1668 	 * locally in 'sfc_adapter' and kept up-to-date
1669 	 */
1670 	rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1671 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1672 	if (rss_conf->rss_key != NULL)
1673 		rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1674 
1675 	return 0;
1676 }
1677 
1678 static int
1679 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1680 			struct rte_eth_rss_conf *rss_conf)
1681 {
1682 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1683 	struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1684 	unsigned int efx_hash_types;
1685 	unsigned int n_contexts;
1686 	unsigned int mode_i = 0;
1687 	unsigned int key_i = 0;
1688 	uint32_t contexts[2];
1689 	unsigned int i = 0;
1690 	int rc = 0;
1691 
1692 	if (sfc_sa2shared(sa)->isolated)
1693 		return -ENOTSUP;
1694 
1695 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1696 		sfc_err(sa, "RSS is not available");
1697 		return -ENOTSUP;
1698 	}
1699 
1700 	if (rss->channels == 0) {
1701 		sfc_err(sa, "RSS is not configured");
1702 		return -EINVAL;
1703 	}
1704 
1705 	if ((rss_conf->rss_key != NULL) &&
1706 	    (rss_conf->rss_key_len != sizeof(rss->key))) {
1707 		sfc_err(sa, "RSS key size is wrong (should be %zu)",
1708 			sizeof(rss->key));
1709 		return -EINVAL;
1710 	}
1711 
1712 	sfc_adapter_lock(sa);
1713 
1714 	rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1715 	if (rc != 0)
1716 		goto fail_rx_hf_rte_to_efx;
1717 
1718 	contexts[0] = EFX_RSS_CONTEXT_DEFAULT;
1719 	contexts[1] = rss->dummy_ctx.nic_handle;
1720 	n_contexts = (rss->dummy_ctx.nic_handle_refcnt == 0) ? 1 : 2;
1721 
1722 	for (mode_i = 0; mode_i < n_contexts; mode_i++) {
1723 		rc = efx_rx_scale_mode_set(sa->nic, contexts[mode_i],
1724 					   rss->hash_alg, efx_hash_types,
1725 					   B_TRUE);
1726 		if (rc != 0)
1727 			goto fail_scale_mode_set;
1728 	}
1729 
1730 	if (rss_conf->rss_key != NULL) {
1731 		if (sa->state == SFC_ETHDEV_STARTED) {
1732 			for (key_i = 0; key_i < n_contexts; key_i++) {
1733 				rc = efx_rx_scale_key_set(sa->nic,
1734 							  contexts[key_i],
1735 							  rss_conf->rss_key,
1736 							  sizeof(rss->key));
1737 				if (rc != 0)
1738 					goto fail_scale_key_set;
1739 			}
1740 		}
1741 
1742 		rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1743 	}
1744 
1745 	rss->hash_types = efx_hash_types;
1746 
1747 	sfc_adapter_unlock(sa);
1748 
1749 	return 0;
1750 
1751 fail_scale_key_set:
1752 	for (i = 0; i < key_i; i++) {
1753 		if (efx_rx_scale_key_set(sa->nic, contexts[i], rss->key,
1754 					 sizeof(rss->key)) != 0)
1755 			sfc_err(sa, "failed to restore RSS key");
1756 	}
1757 
1758 fail_scale_mode_set:
1759 	for (i = 0; i < mode_i; i++) {
1760 		if (efx_rx_scale_mode_set(sa->nic, contexts[i],
1761 					  EFX_RX_HASHALG_TOEPLITZ,
1762 					  rss->hash_types, B_TRUE) != 0)
1763 			sfc_err(sa, "failed to restore RSS mode");
1764 	}
1765 
1766 fail_rx_hf_rte_to_efx:
1767 	sfc_adapter_unlock(sa);
1768 	return -rc;
1769 }
1770 
1771 /*
1772  * The function is used by the secondary process as well. It must not
1773  * use any process-local pointers from the adapter data.
1774  */
1775 static int
1776 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1777 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1778 		       uint16_t reta_size)
1779 {
1780 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1781 	struct sfc_rss *rss = &sas->rss;
1782 	int entry;
1783 
1784 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1785 		return -ENOTSUP;
1786 
1787 	if (rss->channels == 0)
1788 		return -EINVAL;
1789 
1790 	if (reta_size != EFX_RSS_TBL_SIZE)
1791 		return -EINVAL;
1792 
1793 	for (entry = 0; entry < reta_size; entry++) {
1794 		int grp = entry / RTE_ETH_RETA_GROUP_SIZE;
1795 		int grp_idx = entry % RTE_ETH_RETA_GROUP_SIZE;
1796 
1797 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1798 			reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1799 	}
1800 
1801 	return 0;
1802 }
1803 
1804 static int
1805 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1806 			struct rte_eth_rss_reta_entry64 *reta_conf,
1807 			uint16_t reta_size)
1808 {
1809 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1810 	struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1811 	unsigned int *rss_tbl_new;
1812 	uint16_t entry;
1813 	int rc = 0;
1814 
1815 
1816 	if (sfc_sa2shared(sa)->isolated)
1817 		return -ENOTSUP;
1818 
1819 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1820 		sfc_err(sa, "RSS is not available");
1821 		return -ENOTSUP;
1822 	}
1823 
1824 	if (rss->channels == 0) {
1825 		sfc_err(sa, "RSS is not configured");
1826 		return -EINVAL;
1827 	}
1828 
1829 	if (reta_size != EFX_RSS_TBL_SIZE) {
1830 		sfc_err(sa, "RETA size is wrong (should be %u)",
1831 			EFX_RSS_TBL_SIZE);
1832 		return -EINVAL;
1833 	}
1834 
1835 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1836 	if (rss_tbl_new == NULL)
1837 		return -ENOMEM;
1838 
1839 	sfc_adapter_lock(sa);
1840 
1841 	rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1842 
1843 	for (entry = 0; entry < reta_size; entry++) {
1844 		int grp_idx = entry % RTE_ETH_RETA_GROUP_SIZE;
1845 		struct rte_eth_rss_reta_entry64 *grp;
1846 
1847 		grp = &reta_conf[entry / RTE_ETH_RETA_GROUP_SIZE];
1848 
1849 		if (grp->mask & (1ull << grp_idx)) {
1850 			if (grp->reta[grp_idx] >= rss->channels) {
1851 				rc = EINVAL;
1852 				goto bad_reta_entry;
1853 			}
1854 			rss_tbl_new[entry] = grp->reta[grp_idx];
1855 		}
1856 	}
1857 
1858 	if (sa->state == SFC_ETHDEV_STARTED) {
1859 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1860 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1861 		if (rc != 0)
1862 			goto fail_scale_tbl_set;
1863 	}
1864 
1865 	rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1866 
1867 fail_scale_tbl_set:
1868 bad_reta_entry:
1869 	sfc_adapter_unlock(sa);
1870 
1871 	rte_free(rss_tbl_new);
1872 
1873 	SFC_ASSERT(rc >= 0);
1874 	return -rc;
1875 }
1876 
1877 static int
1878 sfc_dev_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
1879 		     const struct rte_flow_ops **ops)
1880 {
1881 	*ops = &sfc_flow_ops;
1882 	return 0;
1883 }
1884 
1885 static int
1886 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1887 {
1888 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1889 
1890 	/*
1891 	 * If Rx datapath does not provide callback to check mempool,
1892 	 * all pools are supported.
1893 	 */
1894 	if (sap->dp_rx->pool_ops_supported == NULL)
1895 		return 1;
1896 
1897 	return sap->dp_rx->pool_ops_supported(pool);
1898 }
1899 
1900 static int
1901 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1902 {
1903 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1904 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1905 	sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1906 	struct sfc_rxq_info *rxq_info;
1907 
1908 	rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1909 
1910 	return sap->dp_rx->intr_enable(rxq_info->dp);
1911 }
1912 
1913 static int
1914 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1915 {
1916 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1917 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1918 	sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1919 	struct sfc_rxq_info *rxq_info;
1920 
1921 	rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1922 
1923 	return sap->dp_rx->intr_disable(rxq_info->dp);
1924 }
1925 
1926 struct sfc_mport_journal_ctx {
1927 	struct sfc_adapter		*sa;
1928 	uint16_t			switch_domain_id;
1929 	uint32_t			mcdi_handle;
1930 	bool				controllers_assigned;
1931 	efx_pcie_interface_t		*controllers;
1932 	size_t				nb_controllers;
1933 };
1934 
1935 static int
1936 sfc_journal_ctx_add_controller(struct sfc_mport_journal_ctx *ctx,
1937 			       efx_pcie_interface_t intf)
1938 {
1939 	efx_pcie_interface_t *new_controllers;
1940 	size_t i, target;
1941 	size_t new_size;
1942 
1943 	if (ctx->controllers == NULL) {
1944 		ctx->controllers = rte_malloc("sfc_controller_mapping",
1945 					      sizeof(ctx->controllers[0]), 0);
1946 		if (ctx->controllers == NULL)
1947 			return ENOMEM;
1948 
1949 		ctx->controllers[0] = intf;
1950 		ctx->nb_controllers = 1;
1951 
1952 		return 0;
1953 	}
1954 
1955 	for (i = 0; i < ctx->nb_controllers; i++) {
1956 		if (ctx->controllers[i] == intf)
1957 			return 0;
1958 		if (ctx->controllers[i] > intf)
1959 			break;
1960 	}
1961 	target = i;
1962 
1963 	ctx->nb_controllers += 1;
1964 	new_size = ctx->nb_controllers * sizeof(ctx->controllers[0]);
1965 
1966 	new_controllers = rte_realloc(ctx->controllers, new_size, 0);
1967 	if (new_controllers == NULL) {
1968 		rte_free(ctx->controllers);
1969 		return ENOMEM;
1970 	}
1971 	ctx->controllers = new_controllers;
1972 
1973 	for (i = target + 1; i < ctx->nb_controllers; i++)
1974 		ctx->controllers[i] = ctx->controllers[i - 1];
1975 
1976 	ctx->controllers[target] = intf;
1977 
1978 	return 0;
1979 }
1980 
1981 static efx_rc_t
1982 sfc_process_mport_journal_entry(struct sfc_mport_journal_ctx *ctx,
1983 				efx_mport_desc_t *mport)
1984 {
1985 	struct sfc_mae_switch_port_request req;
1986 	efx_mport_sel_t entity_selector;
1987 	efx_mport_sel_t ethdev_mport;
1988 	uint16_t switch_port_id;
1989 	efx_rc_t efx_rc;
1990 	int rc;
1991 
1992 	sfc_dbg(ctx->sa,
1993 		"processing mport id %u (controller %u pf %u vf %u)",
1994 		mport->emd_id.id, mport->emd_vnic.ev_intf,
1995 		mport->emd_vnic.ev_pf, mport->emd_vnic.ev_vf);
1996 	efx_mae_mport_invalid(&ethdev_mport);
1997 
1998 	if (!ctx->controllers_assigned) {
1999 		rc = sfc_journal_ctx_add_controller(ctx,
2000 						    mport->emd_vnic.ev_intf);
2001 		if (rc != 0)
2002 			return rc;
2003 	}
2004 
2005 	/* Build Mport selector */
2006 	efx_rc = efx_mae_mport_by_pcie_mh_function(mport->emd_vnic.ev_intf,
2007 						mport->emd_vnic.ev_pf,
2008 						mport->emd_vnic.ev_vf,
2009 						&entity_selector);
2010 	if (efx_rc != 0) {
2011 		sfc_err(ctx->sa, "failed to build entity mport selector for c%upf%uvf%u",
2012 			mport->emd_vnic.ev_intf,
2013 			mport->emd_vnic.ev_pf,
2014 			mport->emd_vnic.ev_vf);
2015 		return efx_rc;
2016 	}
2017 
2018 	rc = sfc_mae_switch_port_id_by_entity(ctx->switch_domain_id,
2019 					      &entity_selector,
2020 					      SFC_MAE_SWITCH_PORT_REPRESENTOR,
2021 					      &switch_port_id);
2022 	switch (rc) {
2023 	case 0:
2024 		/* Already registered */
2025 		break;
2026 	case ENOENT:
2027 		/*
2028 		 * No representor has been created for this entity.
2029 		 * Create a dummy switch registry entry with an invalid ethdev
2030 		 * mport selector. When a corresponding representor is created,
2031 		 * this entry will be updated.
2032 		 */
2033 		req.type = SFC_MAE_SWITCH_PORT_REPRESENTOR;
2034 		req.entity_mportp = &entity_selector;
2035 		req.ethdev_mportp = &ethdev_mport;
2036 		req.ethdev_port_id = RTE_MAX_ETHPORTS;
2037 		req.port_data.repr.intf = mport->emd_vnic.ev_intf;
2038 		req.port_data.repr.pf = mport->emd_vnic.ev_pf;
2039 		req.port_data.repr.vf = mport->emd_vnic.ev_vf;
2040 
2041 		rc = sfc_mae_assign_switch_port(ctx->switch_domain_id,
2042 						&req, &switch_port_id);
2043 		if (rc != 0) {
2044 			sfc_err(ctx->sa,
2045 				"failed to assign MAE switch port for c%upf%uvf%u: %s",
2046 				mport->emd_vnic.ev_intf,
2047 				mport->emd_vnic.ev_pf,
2048 				mport->emd_vnic.ev_vf,
2049 				rte_strerror(rc));
2050 			return rc;
2051 		}
2052 		break;
2053 	default:
2054 		sfc_err(ctx->sa, "failed to find MAE switch port for c%upf%uvf%u: %s",
2055 			mport->emd_vnic.ev_intf,
2056 			mport->emd_vnic.ev_pf,
2057 			mport->emd_vnic.ev_vf,
2058 			rte_strerror(rc));
2059 		return rc;
2060 	}
2061 
2062 	return 0;
2063 }
2064 
2065 static efx_rc_t
2066 sfc_process_mport_journal_cb(void *data, efx_mport_desc_t *mport,
2067 			     size_t mport_len)
2068 {
2069 	struct sfc_mport_journal_ctx *ctx = data;
2070 
2071 	if (ctx == NULL || ctx->sa == NULL) {
2072 		SFC_GENERIC_LOG(ERR, "received NULL context or SFC adapter");
2073 		return EINVAL;
2074 	}
2075 
2076 	if (mport_len != sizeof(*mport)) {
2077 		sfc_err(ctx->sa, "actual and expected mport buffer sizes differ");
2078 		return EINVAL;
2079 	}
2080 
2081 	SFC_ASSERT(sfc_adapter_is_locked(ctx->sa));
2082 
2083 	/*
2084 	 * If a zombie flag is set, it means the mport has been marked for
2085 	 * deletion and cannot be used for any new operations. The mport will
2086 	 * be destroyed completely once all references to it are released.
2087 	 */
2088 	if (mport->emd_zombie) {
2089 		sfc_dbg(ctx->sa, "mport is a zombie, skipping");
2090 		return 0;
2091 	}
2092 	if (mport->emd_type != EFX_MPORT_TYPE_VNIC) {
2093 		sfc_dbg(ctx->sa, "mport is not a VNIC, skipping");
2094 		return 0;
2095 	}
2096 	if (mport->emd_vnic.ev_client_type != EFX_MPORT_VNIC_CLIENT_FUNCTION) {
2097 		sfc_dbg(ctx->sa, "mport is not a function, skipping");
2098 		return 0;
2099 	}
2100 	if (mport->emd_vnic.ev_handle == ctx->mcdi_handle) {
2101 		sfc_dbg(ctx->sa, "mport is this driver instance, skipping");
2102 		return 0;
2103 	}
2104 
2105 	return sfc_process_mport_journal_entry(ctx, mport);
2106 }
2107 
2108 static int
2109 sfc_process_mport_journal(struct sfc_adapter *sa)
2110 {
2111 	struct sfc_mport_journal_ctx ctx;
2112 	const efx_pcie_interface_t *controllers;
2113 	size_t nb_controllers;
2114 	efx_rc_t efx_rc;
2115 	int rc;
2116 
2117 	memset(&ctx, 0, sizeof(ctx));
2118 	ctx.sa = sa;
2119 	ctx.switch_domain_id = sa->mae.switch_domain_id;
2120 
2121 	efx_rc = efx_mcdi_get_own_client_handle(sa->nic, &ctx.mcdi_handle);
2122 	if (efx_rc != 0) {
2123 		sfc_err(sa, "failed to get own MCDI handle");
2124 		SFC_ASSERT(efx_rc > 0);
2125 		return efx_rc;
2126 	}
2127 
2128 	rc = sfc_mae_switch_domain_controllers(ctx.switch_domain_id,
2129 					       &controllers, &nb_controllers);
2130 	if (rc != 0) {
2131 		sfc_err(sa, "failed to get controller mapping");
2132 		return rc;
2133 	}
2134 
2135 	ctx.controllers_assigned = controllers != NULL;
2136 	ctx.controllers = NULL;
2137 	ctx.nb_controllers = 0;
2138 
2139 	efx_rc = efx_mae_read_mport_journal(sa->nic,
2140 					    sfc_process_mport_journal_cb, &ctx);
2141 	if (efx_rc != 0) {
2142 		sfc_err(sa, "failed to process MAE mport journal");
2143 		SFC_ASSERT(efx_rc > 0);
2144 		return efx_rc;
2145 	}
2146 
2147 	if (controllers == NULL) {
2148 		rc = sfc_mae_switch_domain_map_controllers(ctx.switch_domain_id,
2149 							   ctx.controllers,
2150 							   ctx.nb_controllers);
2151 		if (rc != 0)
2152 			return rc;
2153 	}
2154 
2155 	return 0;
2156 }
2157 
2158 static void
2159 sfc_count_representors_cb(enum sfc_mae_switch_port_type type,
2160 			  const efx_mport_sel_t *ethdev_mportp __rte_unused,
2161 			  uint16_t ethdev_port_id __rte_unused,
2162 			  const efx_mport_sel_t *entity_mportp __rte_unused,
2163 			  uint16_t switch_port_id __rte_unused,
2164 			  union sfc_mae_switch_port_data *port_datap
2165 				__rte_unused,
2166 			  void *user_datap)
2167 {
2168 	int *counter = user_datap;
2169 
2170 	SFC_ASSERT(counter != NULL);
2171 
2172 	if (type == SFC_MAE_SWITCH_PORT_REPRESENTOR)
2173 		(*counter)++;
2174 }
2175 
2176 struct sfc_get_representors_ctx {
2177 	struct rte_eth_representor_info	*info;
2178 	struct sfc_adapter		*sa;
2179 	uint16_t			switch_domain_id;
2180 	const efx_pcie_interface_t	*controllers;
2181 	size_t				nb_controllers;
2182 };
2183 
2184 static void
2185 sfc_get_representors_cb(enum sfc_mae_switch_port_type type,
2186 			const efx_mport_sel_t *ethdev_mportp __rte_unused,
2187 			uint16_t ethdev_port_id __rte_unused,
2188 			const efx_mport_sel_t *entity_mportp __rte_unused,
2189 			uint16_t switch_port_id,
2190 			union sfc_mae_switch_port_data *port_datap,
2191 			void *user_datap)
2192 {
2193 	struct sfc_get_representors_ctx *ctx = user_datap;
2194 	struct rte_eth_representor_range *range;
2195 	int ret;
2196 	int rc;
2197 
2198 	SFC_ASSERT(ctx != NULL);
2199 	SFC_ASSERT(ctx->info != NULL);
2200 	SFC_ASSERT(ctx->sa != NULL);
2201 
2202 	if (type != SFC_MAE_SWITCH_PORT_REPRESENTOR) {
2203 		sfc_dbg(ctx->sa, "not a representor, skipping");
2204 		return;
2205 	}
2206 	if (ctx->info->nb_ranges >= ctx->info->nb_ranges_alloc) {
2207 		sfc_dbg(ctx->sa, "info structure is full already");
2208 		return;
2209 	}
2210 
2211 	range = &ctx->info->ranges[ctx->info->nb_ranges];
2212 	rc = sfc_mae_switch_controller_from_mapping(ctx->controllers,
2213 						    ctx->nb_controllers,
2214 						    port_datap->repr.intf,
2215 						    &range->controller);
2216 	if (rc != 0) {
2217 		sfc_err(ctx->sa, "invalid representor controller: %d",
2218 			port_datap->repr.intf);
2219 		range->controller = -1;
2220 	}
2221 	range->pf = port_datap->repr.pf;
2222 	range->id_base = switch_port_id;
2223 	range->id_end = switch_port_id;
2224 
2225 	if (port_datap->repr.vf != EFX_PCI_VF_INVALID) {
2226 		range->type = RTE_ETH_REPRESENTOR_VF;
2227 		range->vf = port_datap->repr.vf;
2228 		ret = snprintf(range->name, RTE_DEV_NAME_MAX_LEN,
2229 			       "c%dpf%dvf%d", range->controller, range->pf,
2230 			       range->vf);
2231 	} else {
2232 		range->type = RTE_ETH_REPRESENTOR_PF;
2233 		ret = snprintf(range->name, RTE_DEV_NAME_MAX_LEN,
2234 			 "c%dpf%d", range->controller, range->pf);
2235 	}
2236 	if (ret >= RTE_DEV_NAME_MAX_LEN) {
2237 		sfc_err(ctx->sa, "representor name has been truncated: %s",
2238 			range->name);
2239 	}
2240 
2241 	ctx->info->nb_ranges++;
2242 }
2243 
2244 static int
2245 sfc_representor_info_get(struct rte_eth_dev *dev,
2246 			 struct rte_eth_representor_info *info)
2247 {
2248 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2249 	struct sfc_get_representors_ctx get_repr_ctx;
2250 	const efx_nic_cfg_t *nic_cfg;
2251 	uint16_t switch_domain_id;
2252 	uint32_t nb_repr;
2253 	int controller;
2254 	int rc;
2255 
2256 	sfc_adapter_lock(sa);
2257 
2258 	if (sa->mae.status != SFC_MAE_STATUS_ADMIN) {
2259 		sfc_adapter_unlock(sa);
2260 		return -ENOTSUP;
2261 	}
2262 
2263 	rc = sfc_process_mport_journal(sa);
2264 	if (rc != 0) {
2265 		sfc_adapter_unlock(sa);
2266 		SFC_ASSERT(rc > 0);
2267 		return -rc;
2268 	}
2269 
2270 	switch_domain_id = sa->mae.switch_domain_id;
2271 
2272 	nb_repr = 0;
2273 	rc = sfc_mae_switch_ports_iterate(switch_domain_id,
2274 					  sfc_count_representors_cb,
2275 					  &nb_repr);
2276 	if (rc != 0) {
2277 		sfc_adapter_unlock(sa);
2278 		SFC_ASSERT(rc > 0);
2279 		return -rc;
2280 	}
2281 
2282 	if (info == NULL) {
2283 		sfc_adapter_unlock(sa);
2284 		return nb_repr;
2285 	}
2286 
2287 	rc = sfc_mae_switch_domain_controllers(switch_domain_id,
2288 					       &get_repr_ctx.controllers,
2289 					       &get_repr_ctx.nb_controllers);
2290 	if (rc != 0) {
2291 		sfc_adapter_unlock(sa);
2292 		SFC_ASSERT(rc > 0);
2293 		return -rc;
2294 	}
2295 
2296 	nic_cfg = efx_nic_cfg_get(sa->nic);
2297 
2298 	rc = sfc_mae_switch_domain_get_controller(switch_domain_id,
2299 						  nic_cfg->enc_intf,
2300 						  &controller);
2301 	if (rc != 0) {
2302 		sfc_err(sa, "invalid controller: %d", nic_cfg->enc_intf);
2303 		controller = -1;
2304 	}
2305 
2306 	info->controller = controller;
2307 	info->pf = nic_cfg->enc_pf;
2308 
2309 	get_repr_ctx.info = info;
2310 	get_repr_ctx.sa = sa;
2311 	get_repr_ctx.switch_domain_id = switch_domain_id;
2312 	rc = sfc_mae_switch_ports_iterate(switch_domain_id,
2313 					  sfc_get_representors_cb,
2314 					  &get_repr_ctx);
2315 	if (rc != 0) {
2316 		sfc_adapter_unlock(sa);
2317 		SFC_ASSERT(rc > 0);
2318 		return -rc;
2319 	}
2320 
2321 	sfc_adapter_unlock(sa);
2322 	return nb_repr;
2323 }
2324 
2325 static int
2326 sfc_rx_metadata_negotiate(struct rte_eth_dev *dev, uint64_t *features)
2327 {
2328 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2329 	uint64_t supported = 0;
2330 
2331 	sfc_adapter_lock(sa);
2332 
2333 	if ((sa->priv.dp_rx->features & SFC_DP_RX_FEAT_FLOW_FLAG) != 0)
2334 		supported |= RTE_ETH_RX_METADATA_USER_FLAG;
2335 
2336 	if ((sa->priv.dp_rx->features & SFC_DP_RX_FEAT_FLOW_MARK) != 0)
2337 		supported |= RTE_ETH_RX_METADATA_USER_MARK;
2338 
2339 	if (sfc_ft_is_supported(sa))
2340 		supported |= RTE_ETH_RX_METADATA_TUNNEL_ID;
2341 
2342 	sa->negotiated_rx_metadata = supported & *features;
2343 	*features = sa->negotiated_rx_metadata;
2344 
2345 	sfc_adapter_unlock(sa);
2346 
2347 	return 0;
2348 }
2349 
2350 static unsigned int
2351 sfc_fec_get_capa_speed_to_fec(uint32_t supported_caps,
2352 			      struct rte_eth_fec_capa *speed_fec_capa)
2353 {
2354 	unsigned int num = 0;
2355 	bool baser = false;
2356 	bool rs = false;
2357 
2358 	if (supported_caps & EFX_PHY_CAP_FEC_BIT(BASER_FEC))
2359 		baser = true;
2360 	if (supported_caps & EFX_PHY_CAP_FEC_BIT(RS_FEC))
2361 		rs = true;
2362 
2363 	/*
2364 	 * NOFEC and AUTO FEC modes are always supported.
2365 	 * FW does not provide information about the supported
2366 	 * FEC modes per the link speed.
2367 	 * Supported FEC depends on supported link speeds and
2368 	 * supported FEC modes by a device.
2369 	 */
2370 	if (supported_caps & (1u << EFX_PHY_CAP_10000FDX)) {
2371 		if (speed_fec_capa != NULL) {
2372 			speed_fec_capa[num].speed = RTE_ETH_SPEED_NUM_10G;
2373 			speed_fec_capa[num].capa =
2374 				RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC) |
2375 				RTE_ETH_FEC_MODE_CAPA_MASK(AUTO);
2376 			if (baser) {
2377 				speed_fec_capa[num].capa |=
2378 					RTE_ETH_FEC_MODE_CAPA_MASK(BASER);
2379 			}
2380 		}
2381 		num++;
2382 	}
2383 	if (supported_caps & (1u << EFX_PHY_CAP_25000FDX)) {
2384 		if (speed_fec_capa != NULL) {
2385 			speed_fec_capa[num].speed = RTE_ETH_SPEED_NUM_25G;
2386 			speed_fec_capa[num].capa =
2387 				RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC) |
2388 				RTE_ETH_FEC_MODE_CAPA_MASK(AUTO);
2389 			if (baser) {
2390 				speed_fec_capa[num].capa |=
2391 					RTE_ETH_FEC_MODE_CAPA_MASK(BASER);
2392 			}
2393 			if (rs) {
2394 				speed_fec_capa[num].capa |=
2395 					RTE_ETH_FEC_MODE_CAPA_MASK(RS);
2396 			}
2397 		}
2398 		num++;
2399 	}
2400 	if (supported_caps & (1u << EFX_PHY_CAP_40000FDX)) {
2401 		if (speed_fec_capa != NULL) {
2402 			speed_fec_capa[num].speed = RTE_ETH_SPEED_NUM_40G;
2403 			speed_fec_capa[num].capa =
2404 				RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC) |
2405 				RTE_ETH_FEC_MODE_CAPA_MASK(AUTO);
2406 			if (baser) {
2407 				speed_fec_capa[num].capa |=
2408 					RTE_ETH_FEC_MODE_CAPA_MASK(BASER);
2409 			}
2410 		}
2411 		num++;
2412 	}
2413 	if (supported_caps & (1u << EFX_PHY_CAP_50000FDX)) {
2414 		if (speed_fec_capa != NULL) {
2415 			speed_fec_capa[num].speed = RTE_ETH_SPEED_NUM_50G;
2416 			speed_fec_capa[num].capa =
2417 				RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC) |
2418 				RTE_ETH_FEC_MODE_CAPA_MASK(AUTO);
2419 			if (baser) {
2420 				speed_fec_capa[num].capa |=
2421 					RTE_ETH_FEC_MODE_CAPA_MASK(BASER);
2422 			}
2423 			if (rs) {
2424 				speed_fec_capa[num].capa |=
2425 					RTE_ETH_FEC_MODE_CAPA_MASK(RS);
2426 			}
2427 		}
2428 		num++;
2429 	}
2430 	if (supported_caps & (1u << EFX_PHY_CAP_100000FDX)) {
2431 		if (speed_fec_capa != NULL) {
2432 			speed_fec_capa[num].speed = RTE_ETH_SPEED_NUM_100G;
2433 			speed_fec_capa[num].capa =
2434 				RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC) |
2435 				RTE_ETH_FEC_MODE_CAPA_MASK(AUTO);
2436 			if (rs) {
2437 				speed_fec_capa[num].capa |=
2438 					RTE_ETH_FEC_MODE_CAPA_MASK(RS);
2439 			}
2440 		}
2441 		num++;
2442 	}
2443 
2444 	return num;
2445 }
2446 
2447 static int
2448 sfc_fec_get_capability(struct rte_eth_dev *dev,
2449 		       struct rte_eth_fec_capa *speed_fec_capa,
2450 		       unsigned int num)
2451 {
2452 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2453 	unsigned int num_entries;
2454 	uint32_t supported_caps;
2455 
2456 	sfc_adapter_lock(sa);
2457 
2458 	efx_phy_adv_cap_get(sa->nic, EFX_PHY_CAP_PERM, &supported_caps);
2459 
2460 	num_entries = sfc_fec_get_capa_speed_to_fec(supported_caps, NULL);
2461 	if (speed_fec_capa == NULL || num < num_entries)
2462 		goto adapter_unlock;
2463 
2464 	num_entries = sfc_fec_get_capa_speed_to_fec(supported_caps,
2465 						    speed_fec_capa);
2466 
2467 adapter_unlock:
2468 	sfc_adapter_unlock(sa);
2469 
2470 	return num_entries;
2471 }
2472 
2473 static uint32_t
2474 sfc_efx_caps_to_fec(uint32_t caps, bool is_25g)
2475 {
2476 	bool rs_req = caps & EFX_PHY_CAP_FEC_BIT(RS_FEC_REQUESTED);
2477 	bool rs = caps & EFX_PHY_CAP_FEC_BIT(RS_FEC);
2478 	bool baser_req;
2479 	bool baser;
2480 
2481 	if (is_25g) {
2482 		baser = caps & EFX_PHY_CAP_FEC_BIT(25G_BASER_FEC);
2483 		baser_req = caps & EFX_PHY_CAP_FEC_BIT(25G_BASER_FEC_REQUESTED);
2484 	} else {
2485 		baser = caps & EFX_PHY_CAP_FEC_BIT(BASER_FEC);
2486 		baser_req = caps & EFX_PHY_CAP_FEC_BIT(BASER_FEC_REQUESTED);
2487 	}
2488 
2489 	if (!baser && !rs)
2490 		return RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_NOFEC);
2491 
2492 	if (rs_req)
2493 		return RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_RS);
2494 
2495 	if (baser_req)
2496 		return RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_BASER);
2497 
2498 	return 0;
2499 }
2500 
2501 static int
2502 sfc_fec_get(struct rte_eth_dev *dev, uint32_t *fec_capa)
2503 {
2504 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2505 	struct sfc_port *port = &sa->port;
2506 	struct rte_eth_link current_link;
2507 	efx_phy_fec_type_t active_fec;
2508 	bool is_25g = false;
2509 	int rc = 0;
2510 
2511 	sfc_adapter_lock(sa);
2512 
2513 	sfc_dev_get_rte_link(dev, 1, &current_link);
2514 
2515 	if (current_link.link_status == RTE_ETH_LINK_DOWN) {
2516 		uint32_t speed = current_link.link_speed;
2517 
2518 		if (port->fec_auto) {
2519 			*fec_capa = RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_AUTO);
2520 			goto adapter_unlock;
2521 		}
2522 
2523 		is_25g = (speed == RTE_ETH_SPEED_NUM_25G ||
2524 			  speed == RTE_ETH_SPEED_NUM_50G);
2525 
2526 		*fec_capa = sfc_efx_caps_to_fec(port->fec_cfg, is_25g);
2527 		if (*fec_capa == 0)
2528 			rc = ENOTSUP;
2529 
2530 		goto adapter_unlock;
2531 	}
2532 
2533 	rc = efx_phy_fec_type_get(sa->nic, &active_fec);
2534 	if (rc != 0)
2535 		goto adapter_unlock;
2536 
2537 	switch (active_fec) {
2538 	case EFX_PHY_FEC_NONE:
2539 		*fec_capa = RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_NOFEC);
2540 		break;
2541 	case EFX_PHY_FEC_BASER:
2542 		*fec_capa = RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_BASER);
2543 		break;
2544 	case EFX_PHY_FEC_RS:
2545 		*fec_capa = RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_RS);
2546 		break;
2547 	default:
2548 		rc = ENOTSUP;
2549 		break;
2550 	}
2551 
2552 adapter_unlock:
2553 	sfc_adapter_unlock(sa);
2554 
2555 	if (rc != 0)
2556 		sfc_err(sa, "failed to get FEC mode");
2557 
2558 	SFC_ASSERT(rc >= 0);
2559 	return -rc;
2560 }
2561 
2562 static int
2563 sfc_fec_capa_check(struct rte_eth_dev *dev, uint32_t fec_capa,
2564 		   uint32_t supported_caps)
2565 {
2566 	struct rte_eth_fec_capa *speed_fec_capa;
2567 	struct rte_eth_link current_link;
2568 	bool is_supported = false;
2569 	unsigned int num_entries;
2570 	bool auto_fec = false;
2571 	unsigned int i;
2572 
2573 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2574 
2575 	if (sa->state != SFC_ETHDEV_STARTED)
2576 		return 0;
2577 
2578 	if (fec_capa & RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_AUTO)) {
2579 		auto_fec = true;
2580 		fec_capa &= ~RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_AUTO);
2581 	}
2582 
2583 	/*
2584 	 * If only the AUTO bit is set, the decision on which FEC
2585 	 * mode to use will be made by HW/FW or driver.
2586 	 */
2587 	if (auto_fec && fec_capa == 0)
2588 		return 0;
2589 
2590 	sfc_dev_get_rte_link(dev, 1, &current_link);
2591 
2592 	num_entries = sfc_fec_get_capa_speed_to_fec(supported_caps, NULL);
2593 	if (num_entries == 0)
2594 		return ENOTSUP;
2595 
2596 	speed_fec_capa = rte_calloc("fec_capa", num_entries,
2597 				    sizeof(*speed_fec_capa), 0);
2598 	num_entries = sfc_fec_get_capa_speed_to_fec(supported_caps,
2599 						    speed_fec_capa);
2600 
2601 	for (i = 0; i < num_entries; i++) {
2602 		if (speed_fec_capa[i].speed == current_link.link_speed) {
2603 			if ((fec_capa & speed_fec_capa[i].capa) != 0)
2604 				is_supported = true;
2605 
2606 			break;
2607 		}
2608 	}
2609 
2610 	rte_free(speed_fec_capa);
2611 
2612 	if (is_supported)
2613 		return 0;
2614 
2615 	return ENOTSUP;
2616 }
2617 
2618 static int
2619 sfc_fec_capa_to_efx(uint32_t supported_caps, uint32_t fec_capa,
2620 		    uint32_t *efx_fec_caps)
2621 {
2622 	bool fec_is_set = false;
2623 	bool auto_fec = false;
2624 	bool nofec = false;
2625 	uint32_t ret = 0;
2626 
2627 	if (efx_fec_caps == NULL)
2628 		return EINVAL;
2629 
2630 	if (fec_capa & RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_AUTO))
2631 		auto_fec = true;
2632 
2633 	if (fec_capa & RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_NOFEC))
2634 		nofec = true;
2635 
2636 	if (fec_capa == RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_AUTO)) {
2637 		ret |= (EFX_PHY_CAP_FEC_BIT(BASER_FEC) |
2638 			EFX_PHY_CAP_FEC_BIT(25G_BASER_FEC) |
2639 			EFX_PHY_CAP_FEC_BIT(RS_FEC)) & supported_caps;
2640 		goto done;
2641 	}
2642 
2643 	if (fec_capa & RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_RS)) {
2644 		fec_is_set = true;
2645 
2646 		if (supported_caps & EFX_PHY_CAP_FEC_BIT(RS_FEC)) {
2647 			ret |= EFX_PHY_CAP_FEC_BIT(RS_FEC) |
2648 			       EFX_PHY_CAP_FEC_BIT(RS_FEC_REQUESTED);
2649 		}
2650 	}
2651 	if (fec_capa & RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_BASER)) {
2652 		if (!auto_fec && fec_is_set)
2653 			return EINVAL;
2654 
2655 		if (supported_caps & EFX_PHY_CAP_FEC_BIT(BASER_FEC)) {
2656 			ret |= EFX_PHY_CAP_FEC_BIT(BASER_FEC) |
2657 			       EFX_PHY_CAP_FEC_BIT(BASER_FEC_REQUESTED);
2658 		}
2659 		if (supported_caps & EFX_PHY_CAP_FEC_BIT(25G_BASER_FEC)) {
2660 			ret |= EFX_PHY_CAP_FEC_BIT(25G_BASER_FEC) |
2661 			       EFX_PHY_CAP_FEC_BIT(25G_BASER_FEC_REQUESTED);
2662 		}
2663 	}
2664 
2665 	if (ret == 0 && !nofec)
2666 		return ENOTSUP;
2667 
2668 done:
2669 	*efx_fec_caps = ret;
2670 	return 0;
2671 }
2672 
2673 static int
2674 sfc_fec_set(struct rte_eth_dev *dev, uint32_t fec_capa)
2675 {
2676 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2677 	struct sfc_port *port = &sa->port;
2678 	uint32_t supported_caps;
2679 	uint32_t efx_fec_caps;
2680 	uint32_t updated_caps;
2681 	int rc = 0;
2682 
2683 	sfc_adapter_lock(sa);
2684 
2685 	efx_phy_adv_cap_get(sa->nic, EFX_PHY_CAP_PERM, &supported_caps);
2686 
2687 	rc = sfc_fec_capa_check(dev, fec_capa, supported_caps);
2688 	if (rc != 0)
2689 		goto adapter_unlock;
2690 
2691 	rc = sfc_fec_capa_to_efx(supported_caps, fec_capa, &efx_fec_caps);
2692 	if (rc != 0)
2693 		goto adapter_unlock;
2694 
2695 	if (sa->state == SFC_ETHDEV_STARTED) {
2696 		efx_phy_adv_cap_get(sa->nic, EFX_PHY_CAP_CURRENT,
2697 				    &updated_caps);
2698 		updated_caps = updated_caps & ~EFX_PHY_CAP_FEC_MASK;
2699 		updated_caps |= efx_fec_caps;
2700 
2701 		rc = efx_phy_adv_cap_set(sa->nic, updated_caps);
2702 		if (rc != 0)
2703 			goto adapter_unlock;
2704 	}
2705 
2706 	port->fec_cfg = efx_fec_caps;
2707 	/*
2708 	 * There is no chance to recognize AUTO mode from the
2709 	 * saved FEC capabilities as AUTO mode can have the same
2710 	 * set of bits as any other mode from the EFX point of view.
2711 	 * Save it in the proper variable.
2712 	 */
2713 	if (fec_capa & RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_AUTO))
2714 		port->fec_auto = true;
2715 	else
2716 		port->fec_auto = false;
2717 
2718 adapter_unlock:
2719 	sfc_adapter_unlock(sa);
2720 
2721 	SFC_ASSERT(rc >= 0);
2722 	return -rc;
2723 }
2724 
2725 static uint64_t
2726 sfc_get_restore_flags(__rte_unused struct rte_eth_dev *dev,
2727 		      __rte_unused enum rte_eth_dev_operation op)
2728 {
2729 	/* sfc PMD does not require any configuration restore */
2730 	return 0;
2731 }
2732 
2733 static const struct eth_dev_ops sfc_eth_dev_ops = {
2734 	.dev_configure			= sfc_dev_configure,
2735 	.dev_start			= sfc_dev_start,
2736 	.dev_stop			= sfc_dev_stop,
2737 	.dev_set_link_up		= sfc_dev_set_link_up,
2738 	.dev_set_link_down		= sfc_dev_set_link_down,
2739 	.dev_close			= sfc_dev_close,
2740 	.promiscuous_enable		= sfc_dev_promisc_enable,
2741 	.promiscuous_disable		= sfc_dev_promisc_disable,
2742 	.allmulticast_enable		= sfc_dev_allmulti_enable,
2743 	.allmulticast_disable		= sfc_dev_allmulti_disable,
2744 	.link_update			= sfc_dev_link_update,
2745 	.stats_get			= sfc_stats_get,
2746 	.stats_reset			= sfc_stats_reset,
2747 	.xstats_get			= sfc_xstats_get,
2748 	.xstats_reset			= sfc_stats_reset,
2749 	.xstats_get_names		= sfc_xstats_get_names,
2750 	.dev_infos_get			= sfc_dev_infos_get,
2751 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
2752 	.mtu_set			= sfc_dev_set_mtu,
2753 	.rx_queue_start			= sfc_rx_queue_start,
2754 	.rx_queue_stop			= sfc_rx_queue_stop,
2755 	.tx_queue_start			= sfc_tx_queue_start,
2756 	.tx_queue_stop			= sfc_tx_queue_stop,
2757 	.rx_queue_setup			= sfc_rx_queue_setup,
2758 	.rx_queue_release		= sfc_rx_queue_release,
2759 	.rx_queue_intr_enable		= sfc_rx_queue_intr_enable,
2760 	.rx_queue_intr_disable		= sfc_rx_queue_intr_disable,
2761 	.tx_queue_setup			= sfc_tx_queue_setup,
2762 	.tx_queue_release		= sfc_tx_queue_release,
2763 	.flow_ctrl_get			= sfc_flow_ctrl_get,
2764 	.flow_ctrl_set			= sfc_flow_ctrl_set,
2765 	.mac_addr_set			= sfc_mac_addr_set,
2766 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
2767 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
2768 	.reta_update			= sfc_dev_rss_reta_update,
2769 	.reta_query			= sfc_dev_rss_reta_query,
2770 	.rss_hash_update		= sfc_dev_rss_hash_update,
2771 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
2772 	.flow_ops_get			= sfc_dev_flow_ops_get,
2773 	.set_mc_addr_list		= sfc_set_mc_addr_list,
2774 	.rxq_info_get			= sfc_rx_queue_info_get,
2775 	.txq_info_get			= sfc_tx_queue_info_get,
2776 	.fw_version_get			= sfc_fw_version_get,
2777 	.xstats_get_by_id		= sfc_xstats_get_by_id,
2778 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
2779 	.pool_ops_supported		= sfc_pool_ops_supported,
2780 	.representor_info_get		= sfc_representor_info_get,
2781 	.rx_metadata_negotiate		= sfc_rx_metadata_negotiate,
2782 	.fec_get_capability		= sfc_fec_get_capability,
2783 	.fec_get			= sfc_fec_get,
2784 	.fec_set			= sfc_fec_set,
2785 	.get_restore_flags		= sfc_get_restore_flags,
2786 };
2787 
2788 struct sfc_ethdev_init_data {
2789 	uint16_t		nb_representors;
2790 };
2791 
2792 /**
2793  * Duplicate a string in potentially shared memory required for
2794  * multi-process support.
2795  *
2796  * strdup() allocates from process-local heap/memory.
2797  */
2798 static char *
2799 sfc_strdup(const char *str)
2800 {
2801 	size_t size;
2802 	char *copy;
2803 
2804 	if (str == NULL)
2805 		return NULL;
2806 
2807 	size = strlen(str) + 1;
2808 	copy = rte_malloc(__func__, size, 0);
2809 	if (copy != NULL)
2810 		rte_memcpy(copy, str, size);
2811 
2812 	return copy;
2813 }
2814 
2815 static int
2816 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
2817 {
2818 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2819 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2820 	const struct sfc_dp_rx *dp_rx;
2821 	const struct sfc_dp_tx *dp_tx;
2822 	const efx_nic_cfg_t *encp;
2823 	unsigned int avail_caps = 0;
2824 	const char *rx_name = NULL;
2825 	const char *tx_name = NULL;
2826 	int rc;
2827 
2828 	switch (sa->family) {
2829 	case EFX_FAMILY_HUNTINGTON:
2830 	case EFX_FAMILY_MEDFORD:
2831 	case EFX_FAMILY_MEDFORD2:
2832 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
2833 		avail_caps |= SFC_DP_HW_FW_CAP_RX_EFX;
2834 		avail_caps |= SFC_DP_HW_FW_CAP_TX_EFX;
2835 		break;
2836 	case EFX_FAMILY_RIVERHEAD:
2837 		avail_caps |= SFC_DP_HW_FW_CAP_EF100;
2838 		break;
2839 	default:
2840 		break;
2841 	}
2842 
2843 	encp = efx_nic_cfg_get(sa->nic);
2844 	if (encp->enc_rx_es_super_buffer_supported)
2845 		avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
2846 
2847 	rc = sfc_kvargs_process_opt(sa, SFC_KVARG_RX_DATAPATH,
2848 				    sfc_kvarg_string_handler, &rx_name);
2849 	if (rc != 0)
2850 		goto fail_kvarg_rx_datapath;
2851 
2852 	if (rx_name != NULL) {
2853 		dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
2854 		if (dp_rx == NULL) {
2855 			sfc_err(sa, "Rx datapath %s not found", rx_name);
2856 			rc = ENOENT;
2857 			goto fail_dp_rx;
2858 		}
2859 		if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
2860 			sfc_err(sa,
2861 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
2862 				rx_name);
2863 			rc = EINVAL;
2864 			goto fail_dp_rx_caps;
2865 		}
2866 	} else {
2867 		dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
2868 		if (dp_rx == NULL) {
2869 			sfc_err(sa, "Rx datapath by caps %#x not found",
2870 				avail_caps);
2871 			rc = ENOENT;
2872 			goto fail_dp_rx;
2873 		}
2874 	}
2875 
2876 	sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
2877 	if (sas->dp_rx_name == NULL) {
2878 		rc = ENOMEM;
2879 		goto fail_dp_rx_name;
2880 	}
2881 
2882 	if (strcmp(dp_rx->dp.name, SFC_KVARG_DATAPATH_EF10_ESSB) == 0) {
2883 		/* FLAG and MARK are always available from Rx prefix. */
2884 		sa->negotiated_rx_metadata |= RTE_ETH_RX_METADATA_USER_FLAG;
2885 		sa->negotiated_rx_metadata |= RTE_ETH_RX_METADATA_USER_MARK;
2886 	}
2887 
2888 	sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
2889 
2890 	rc = sfc_kvargs_process_opt(sa, SFC_KVARG_TX_DATAPATH,
2891 				    sfc_kvarg_string_handler, &tx_name);
2892 	if (rc != 0)
2893 		goto fail_kvarg_tx_datapath;
2894 
2895 	if (tx_name != NULL) {
2896 		dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
2897 		if (dp_tx == NULL) {
2898 			sfc_err(sa, "Tx datapath %s not found", tx_name);
2899 			rc = ENOENT;
2900 			goto fail_dp_tx;
2901 		}
2902 		if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
2903 			sfc_err(sa,
2904 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
2905 				tx_name);
2906 			rc = EINVAL;
2907 			goto fail_dp_tx_caps;
2908 		}
2909 	} else {
2910 		dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
2911 		if (dp_tx == NULL) {
2912 			sfc_err(sa, "Tx datapath by caps %#x not found",
2913 				avail_caps);
2914 			rc = ENOENT;
2915 			goto fail_dp_tx;
2916 		}
2917 	}
2918 
2919 	sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
2920 	if (sas->dp_tx_name == NULL) {
2921 		rc = ENOMEM;
2922 		goto fail_dp_tx_name;
2923 	}
2924 
2925 	sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
2926 
2927 	sa->priv.dp_rx = dp_rx;
2928 	sa->priv.dp_tx = dp_tx;
2929 
2930 	dev->rx_pkt_burst = dp_rx->pkt_burst;
2931 	dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2932 	dev->tx_pkt_burst = dp_tx->pkt_burst;
2933 
2934 	dev->rx_queue_count = sfc_rx_queue_count;
2935 	dev->rx_descriptor_status = sfc_rx_descriptor_status;
2936 	dev->tx_descriptor_status = sfc_tx_descriptor_status;
2937 	dev->dev_ops = &sfc_eth_dev_ops;
2938 
2939 	return 0;
2940 
2941 fail_dp_tx_name:
2942 fail_dp_tx_caps:
2943 fail_dp_tx:
2944 fail_kvarg_tx_datapath:
2945 	rte_free(sas->dp_rx_name);
2946 	sas->dp_rx_name = NULL;
2947 
2948 fail_dp_rx_name:
2949 fail_dp_rx_caps:
2950 fail_dp_rx:
2951 fail_kvarg_rx_datapath:
2952 	return rc;
2953 }
2954 
2955 static void
2956 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
2957 {
2958 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2959 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2960 
2961 	dev->dev_ops = NULL;
2962 	dev->tx_pkt_prepare = NULL;
2963 	dev->rx_pkt_burst = NULL;
2964 	dev->tx_pkt_burst = NULL;
2965 
2966 	rte_free(sas->dp_tx_name);
2967 	sas->dp_tx_name = NULL;
2968 	sa->priv.dp_tx = NULL;
2969 
2970 	rte_free(sas->dp_rx_name);
2971 	sas->dp_rx_name = NULL;
2972 	sa->priv.dp_rx = NULL;
2973 }
2974 
2975 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2976 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
2977 	.reta_query			= sfc_dev_rss_reta_query,
2978 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
2979 	.rxq_info_get			= sfc_rx_queue_info_get,
2980 	.txq_info_get			= sfc_tx_queue_info_get,
2981 };
2982 
2983 static int
2984 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2985 {
2986 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2987 	struct sfc_adapter_priv *sap;
2988 	const struct sfc_dp_rx *dp_rx;
2989 	const struct sfc_dp_tx *dp_tx;
2990 	int rc;
2991 
2992 	/*
2993 	 * Allocate process private data from heap, since it should not
2994 	 * be located in shared memory allocated using rte_malloc() API.
2995 	 */
2996 	sap = calloc(1, sizeof(*sap));
2997 	if (sap == NULL) {
2998 		rc = ENOMEM;
2999 		goto fail_alloc_priv;
3000 	}
3001 
3002 	sap->logtype_main = logtype_main;
3003 
3004 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
3005 	if (dp_rx == NULL) {
3006 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
3007 			"cannot find %s Rx datapath", sas->dp_rx_name);
3008 		rc = ENOENT;
3009 		goto fail_dp_rx;
3010 	}
3011 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
3012 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
3013 			"%s Rx datapath does not support multi-process",
3014 			sas->dp_rx_name);
3015 		rc = EINVAL;
3016 		goto fail_dp_rx_multi_process;
3017 	}
3018 
3019 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
3020 	if (dp_tx == NULL) {
3021 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
3022 			"cannot find %s Tx datapath", sas->dp_tx_name);
3023 		rc = ENOENT;
3024 		goto fail_dp_tx;
3025 	}
3026 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
3027 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
3028 			"%s Tx datapath does not support multi-process",
3029 			sas->dp_tx_name);
3030 		rc = EINVAL;
3031 		goto fail_dp_tx_multi_process;
3032 	}
3033 
3034 	sap->dp_rx = dp_rx;
3035 	sap->dp_tx = dp_tx;
3036 
3037 	dev->process_private = sap;
3038 	dev->rx_pkt_burst = dp_rx->pkt_burst;
3039 	dev->tx_pkt_prepare = dp_tx->pkt_prepare;
3040 	dev->tx_pkt_burst = dp_tx->pkt_burst;
3041 	dev->rx_queue_count = sfc_rx_queue_count;
3042 	dev->rx_descriptor_status = sfc_rx_descriptor_status;
3043 	dev->tx_descriptor_status = sfc_tx_descriptor_status;
3044 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
3045 
3046 	return 0;
3047 
3048 fail_dp_tx_multi_process:
3049 fail_dp_tx:
3050 fail_dp_rx_multi_process:
3051 fail_dp_rx:
3052 	free(sap);
3053 
3054 fail_alloc_priv:
3055 	return rc;
3056 }
3057 
3058 static void
3059 sfc_register_dp(void)
3060 {
3061 	/* Register once */
3062 	if (TAILQ_EMPTY(&sfc_dp_head)) {
3063 		/* Prefer EF10 datapath */
3064 		sfc_dp_register(&sfc_dp_head, &sfc_ef100_rx.dp);
3065 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
3066 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
3067 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
3068 
3069 		sfc_dp_register(&sfc_dp_head, &sfc_ef100_tx.dp);
3070 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
3071 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
3072 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
3073 	}
3074 }
3075 
3076 static int
3077 sfc_parse_switch_mode(struct sfc_adapter *sa, bool has_representors)
3078 {
3079 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
3080 	const char *switch_mode = NULL;
3081 	int rc;
3082 
3083 	sfc_log_init(sa, "entry");
3084 
3085 	rc = sfc_kvargs_process_opt(sa, SFC_KVARG_SWITCH_MODE,
3086 				    sfc_kvarg_string_handler, &switch_mode);
3087 	if (rc != 0)
3088 		goto fail_kvargs;
3089 
3090 	if (switch_mode == NULL) {
3091 		sa->switchdev = encp->enc_mae_admin &&
3092 				(!encp->enc_datapath_cap_evb ||
3093 				 has_representors);
3094 	} else if (strcasecmp(switch_mode, SFC_KVARG_SWITCH_MODE_LEGACY) == 0) {
3095 		sa->switchdev = false;
3096 	} else if (strcasecmp(switch_mode,
3097 			      SFC_KVARG_SWITCH_MODE_SWITCHDEV) == 0) {
3098 		sa->switchdev = true;
3099 	} else {
3100 		sfc_err(sa, "invalid switch mode device argument '%s'",
3101 			switch_mode);
3102 		rc = EINVAL;
3103 		goto fail_mode;
3104 	}
3105 
3106 	sfc_log_init(sa, "done");
3107 
3108 	return 0;
3109 
3110 fail_mode:
3111 fail_kvargs:
3112 	sfc_log_init(sa, "failed: %s", rte_strerror(rc));
3113 
3114 	return rc;
3115 }
3116 
3117 static int
3118 sfc_eth_dev_init(struct rte_eth_dev *dev, void *init_params)
3119 {
3120 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
3121 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
3122 	struct sfc_ethdev_init_data *init_data = init_params;
3123 	uint32_t logtype_main;
3124 	struct sfc_adapter *sa;
3125 	int rc;
3126 	const efx_nic_cfg_t *encp;
3127 	const struct rte_ether_addr *from;
3128 	int ret;
3129 
3130 	if (sfc_efx_dev_class_get(pci_dev->device.devargs) !=
3131 			SFC_EFX_DEV_CLASS_NET) {
3132 		SFC_GENERIC_LOG(DEBUG,
3133 			"Incompatible device class: skip probing, should be probed by other sfc driver.");
3134 		return 1;
3135 	}
3136 
3137 	rc = sfc_dp_mport_register();
3138 	if (rc != 0)
3139 		return rc;
3140 
3141 	sfc_register_dp();
3142 
3143 	logtype_main = sfc_register_logtype(&pci_dev->addr,
3144 					    SFC_LOGTYPE_MAIN_STR,
3145 					    RTE_LOG_NOTICE);
3146 
3147 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
3148 		return -sfc_eth_dev_secondary_init(dev, logtype_main);
3149 
3150 	/* Required for logging */
3151 	ret = snprintf(sas->log_prefix, sizeof(sas->log_prefix),
3152 			"PMD: sfc_efx " PCI_PRI_FMT " #%" PRIu16 ": ",
3153 			pci_dev->addr.domain, pci_dev->addr.bus,
3154 			pci_dev->addr.devid, pci_dev->addr.function,
3155 			dev->data->port_id);
3156 	if (ret < 0 || ret >= (int)sizeof(sas->log_prefix)) {
3157 		SFC_GENERIC_LOG(ERR,
3158 			"reserved log prefix is too short for " PCI_PRI_FMT,
3159 			pci_dev->addr.domain, pci_dev->addr.bus,
3160 			pci_dev->addr.devid, pci_dev->addr.function);
3161 		return -EINVAL;
3162 	}
3163 	sas->pci_addr = pci_dev->addr;
3164 	sas->port_id = dev->data->port_id;
3165 
3166 	/*
3167 	 * Allocate process private data from heap, since it should not
3168 	 * be located in shared memory allocated using rte_malloc() API.
3169 	 */
3170 	sa = calloc(1, sizeof(*sa));
3171 	if (sa == NULL) {
3172 		rc = ENOMEM;
3173 		goto fail_alloc_sa;
3174 	}
3175 
3176 	dev->process_private = sa;
3177 
3178 	/* Required for logging */
3179 	sa->priv.shared = sas;
3180 	sa->priv.logtype_main = logtype_main;
3181 
3182 	sa->eth_dev = dev;
3183 
3184 	/* Copy PCI device info to the dev->data */
3185 	rte_eth_copy_pci_info(dev, pci_dev);
3186 	dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE;
3187 
3188 	rc = sfc_kvargs_parse(sa);
3189 	if (rc != 0)
3190 		goto fail_kvargs_parse;
3191 
3192 	sfc_log_init(sa, "entry");
3193 
3194 	dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
3195 	if (dev->data->mac_addrs == NULL) {
3196 		rc = ENOMEM;
3197 		goto fail_mac_addrs;
3198 	}
3199 
3200 	sfc_adapter_lock_init(sa);
3201 	sfc_adapter_lock(sa);
3202 
3203 	sfc_log_init(sa, "probing");
3204 	rc = sfc_probe(sa);
3205 	if (rc != 0)
3206 		goto fail_probe;
3207 
3208 	/*
3209 	 * Selecting a default switch mode requires the NIC to be probed and
3210 	 * to have its capabilities filled in.
3211 	 */
3212 	rc = sfc_parse_switch_mode(sa, init_data->nb_representors > 0);
3213 	if (rc != 0)
3214 		goto fail_switch_mode;
3215 
3216 	sfc_log_init(sa, "set device ops");
3217 	rc = sfc_eth_dev_set_ops(dev);
3218 	if (rc != 0)
3219 		goto fail_set_ops;
3220 
3221 	sfc_log_init(sa, "attaching");
3222 	rc = sfc_attach(sa);
3223 	if (rc != 0)
3224 		goto fail_attach;
3225 
3226 	if (sa->switchdev && sa->mae.status != SFC_MAE_STATUS_ADMIN) {
3227 		sfc_err(sa,
3228 			"failed to enable switchdev mode without admin MAE privilege");
3229 		rc = ENOTSUP;
3230 		goto fail_switchdev_no_mae;
3231 	}
3232 
3233 	encp = efx_nic_cfg_get(sa->nic);
3234 
3235 	/*
3236 	 * The arguments are really reverse order in comparison to
3237 	 * Linux kernel. Copy from NIC config to Ethernet device data.
3238 	 */
3239 	from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
3240 	rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
3241 
3242 	/*
3243 	 * Setup the NIC DMA mapping handler. All internal mempools
3244 	 * MUST be created on attach before this point, and the
3245 	 * adapter MUST NOT create mempools with the adapter lock
3246 	 * held after this point.
3247 	 */
3248 	rc = sfc_nic_dma_attach(sa);
3249 	if (rc != 0)
3250 		goto fail_nic_dma_attach;
3251 
3252 	sfc_adapter_unlock(sa);
3253 
3254 	sfc_log_init(sa, "done");
3255 	return 0;
3256 
3257 fail_nic_dma_attach:
3258 fail_switchdev_no_mae:
3259 	sfc_detach(sa);
3260 
3261 fail_attach:
3262 	sfc_eth_dev_clear_ops(dev);
3263 
3264 fail_set_ops:
3265 fail_switch_mode:
3266 	sfc_unprobe(sa);
3267 
3268 fail_probe:
3269 	sfc_adapter_unlock(sa);
3270 	sfc_adapter_lock_fini(sa);
3271 	rte_free(dev->data->mac_addrs);
3272 	dev->data->mac_addrs = NULL;
3273 
3274 fail_mac_addrs:
3275 	sfc_kvargs_cleanup(sa);
3276 
3277 fail_kvargs_parse:
3278 	sfc_log_init(sa, "failed %d", rc);
3279 	dev->process_private = NULL;
3280 	free(sa);
3281 
3282 fail_alloc_sa:
3283 	SFC_ASSERT(rc > 0);
3284 	return -rc;
3285 }
3286 
3287 static int
3288 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
3289 {
3290 	sfc_dev_close(dev);
3291 
3292 	return 0;
3293 }
3294 
3295 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
3296 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
3297 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
3298 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
3299 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
3300 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
3301 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
3302 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
3303 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
3304 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD) },
3305 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD_VF) },
3306 	{ .vendor_id = 0 /* sentinel */ }
3307 };
3308 
3309 static int
3310 sfc_parse_rte_devargs(const char *args, struct rte_eth_devargs *devargs)
3311 {
3312 	struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
3313 	int rc;
3314 
3315 	if (args != NULL) {
3316 		rc = rte_eth_devargs_parse(args, &eth_da, 1);
3317 		if (rc < 0) {
3318 			SFC_GENERIC_LOG(ERR,
3319 					"Failed to parse generic devargs '%s'",
3320 					args);
3321 			return rc;
3322 		}
3323 	}
3324 
3325 	*devargs = eth_da;
3326 
3327 	return 0;
3328 }
3329 
3330 static int
3331 sfc_eth_dev_find_or_create(struct rte_pci_device *pci_dev,
3332 			   struct sfc_ethdev_init_data *init_data,
3333 			   struct rte_eth_dev **devp,
3334 			   bool *dev_created)
3335 {
3336 	struct rte_eth_dev *dev;
3337 	bool created = false;
3338 	int rc;
3339 
3340 	dev = rte_eth_dev_allocated(pci_dev->device.name);
3341 	if (dev == NULL) {
3342 		rc = rte_eth_dev_create(&pci_dev->device, pci_dev->device.name,
3343 					sizeof(struct sfc_adapter_shared),
3344 					eth_dev_pci_specific_init, pci_dev,
3345 					sfc_eth_dev_init, init_data);
3346 		if (rc != 0) {
3347 			SFC_GENERIC_LOG(ERR, "Failed to create sfc ethdev '%s'",
3348 					pci_dev->device.name);
3349 			return rc;
3350 		}
3351 
3352 		created = true;
3353 
3354 		dev = rte_eth_dev_allocated(pci_dev->device.name);
3355 		if (dev == NULL) {
3356 			SFC_GENERIC_LOG(ERR,
3357 				"Failed to find allocated sfc ethdev '%s'",
3358 				pci_dev->device.name);
3359 			return -ENODEV;
3360 		}
3361 	}
3362 
3363 	*devp = dev;
3364 	*dev_created = created;
3365 
3366 	return 0;
3367 }
3368 
3369 static int
3370 sfc_eth_dev_create_repr(struct sfc_adapter *sa,
3371 			efx_pcie_interface_t controller,
3372 			uint16_t port,
3373 			uint16_t repr_port,
3374 			enum rte_eth_representor_type type)
3375 {
3376 	struct sfc_repr_entity_info entity;
3377 	efx_mport_sel_t mport_sel;
3378 	int rc;
3379 
3380 	switch (type) {
3381 	case RTE_ETH_REPRESENTOR_NONE:
3382 		return 0;
3383 	case RTE_ETH_REPRESENTOR_VF:
3384 	case RTE_ETH_REPRESENTOR_PF:
3385 		break;
3386 	case RTE_ETH_REPRESENTOR_SF:
3387 		sfc_err(sa, "SF representors are not supported");
3388 		return ENOTSUP;
3389 	default:
3390 		sfc_err(sa, "unknown representor type: %d", type);
3391 		return ENOTSUP;
3392 	}
3393 
3394 	rc = efx_mae_mport_by_pcie_mh_function(controller,
3395 					       port,
3396 					       repr_port,
3397 					       &mport_sel);
3398 	if (rc != 0) {
3399 		sfc_err(sa,
3400 			"failed to get m-port selector for controller %u port %u repr_port %u: %s",
3401 			controller, port, repr_port, rte_strerror(-rc));
3402 		return rc;
3403 	}
3404 
3405 	memset(&entity, 0, sizeof(entity));
3406 	entity.type = type;
3407 	entity.intf = controller;
3408 	entity.pf = port;
3409 	entity.vf = repr_port;
3410 
3411 	rc = sfc_repr_create(sa->eth_dev, &entity, sa->mae.switch_domain_id,
3412 			     &mport_sel);
3413 	if (rc != 0) {
3414 		sfc_err(sa,
3415 			"failed to create representor for controller %u port %u repr_port %u: %s",
3416 			controller, port, repr_port, rte_strerror(-rc));
3417 		return rc;
3418 	}
3419 
3420 	return 0;
3421 }
3422 
3423 static int
3424 sfc_eth_dev_create_repr_port(struct sfc_adapter *sa,
3425 			     const struct rte_eth_devargs *eth_da,
3426 			     efx_pcie_interface_t controller,
3427 			     uint16_t port)
3428 {
3429 	int first_error = 0;
3430 	uint16_t i;
3431 	int rc;
3432 
3433 	if (eth_da->type == RTE_ETH_REPRESENTOR_PF) {
3434 		return sfc_eth_dev_create_repr(sa, controller, port,
3435 					       EFX_PCI_VF_INVALID,
3436 					       eth_da->type);
3437 	}
3438 
3439 	for (i = 0; i < eth_da->nb_representor_ports; i++) {
3440 		rc = sfc_eth_dev_create_repr(sa, controller, port,
3441 					     eth_da->representor_ports[i],
3442 					     eth_da->type);
3443 		if (rc != 0 && first_error == 0)
3444 			first_error = rc;
3445 	}
3446 
3447 	return first_error;
3448 }
3449 
3450 static int
3451 sfc_eth_dev_create_repr_controller(struct sfc_adapter *sa,
3452 				   const struct rte_eth_devargs *eth_da,
3453 				   efx_pcie_interface_t controller)
3454 {
3455 	const efx_nic_cfg_t *encp;
3456 	int first_error = 0;
3457 	uint16_t default_port;
3458 	uint16_t i;
3459 	int rc;
3460 
3461 	if (eth_da->nb_ports == 0) {
3462 		encp = efx_nic_cfg_get(sa->nic);
3463 		default_port = encp->enc_intf == controller ? encp->enc_pf : 0;
3464 		return sfc_eth_dev_create_repr_port(sa, eth_da, controller,
3465 						    default_port);
3466 	}
3467 
3468 	for (i = 0; i < eth_da->nb_ports; i++) {
3469 		rc = sfc_eth_dev_create_repr_port(sa, eth_da, controller,
3470 						  eth_da->ports[i]);
3471 		if (rc != 0 && first_error == 0)
3472 			first_error = rc;
3473 	}
3474 
3475 	return first_error;
3476 }
3477 
3478 static int
3479 sfc_eth_dev_create_representors(struct rte_eth_dev *dev,
3480 				const struct rte_eth_devargs *eth_da)
3481 {
3482 	efx_pcie_interface_t intf;
3483 	const efx_nic_cfg_t *encp;
3484 	struct sfc_adapter *sa;
3485 	uint16_t switch_domain_id;
3486 	uint16_t i;
3487 	int rc;
3488 
3489 	sa = sfc_adapter_by_eth_dev(dev);
3490 	switch_domain_id = sa->mae.switch_domain_id;
3491 
3492 	switch (eth_da->type) {
3493 	case RTE_ETH_REPRESENTOR_NONE:
3494 		return 0;
3495 	case RTE_ETH_REPRESENTOR_PF:
3496 	case RTE_ETH_REPRESENTOR_VF:
3497 		break;
3498 	case RTE_ETH_REPRESENTOR_SF:
3499 		sfc_err(sa, "SF representors are not supported");
3500 		return -ENOTSUP;
3501 	default:
3502 		sfc_err(sa, "unknown representor type: %d",
3503 			eth_da->type);
3504 		return -ENOTSUP;
3505 	}
3506 
3507 	if (!sa->switchdev) {
3508 		sfc_err(sa, "cannot create representors in non-switchdev mode");
3509 		return -EINVAL;
3510 	}
3511 
3512 	if (!sfc_repr_available(sfc_sa2shared(sa))) {
3513 		sfc_err(sa, "cannot create representors: unsupported");
3514 
3515 		return -ENOTSUP;
3516 	}
3517 
3518 	/*
3519 	 * This is needed to construct the DPDK controller -> EFX interface
3520 	 * mapping.
3521 	 */
3522 	sfc_adapter_lock(sa);
3523 	rc = sfc_process_mport_journal(sa);
3524 	sfc_adapter_unlock(sa);
3525 	if (rc != 0) {
3526 		SFC_ASSERT(rc > 0);
3527 		return -rc;
3528 	}
3529 
3530 	if (eth_da->nb_mh_controllers > 0) {
3531 		for (i = 0; i < eth_da->nb_mh_controllers; i++) {
3532 			rc = sfc_mae_switch_domain_get_intf(switch_domain_id,
3533 						eth_da->mh_controllers[i],
3534 						&intf);
3535 			if (rc != 0) {
3536 				sfc_err(sa, "failed to get representor");
3537 				continue;
3538 			}
3539 			sfc_eth_dev_create_repr_controller(sa, eth_da, intf);
3540 		}
3541 	} else {
3542 		encp = efx_nic_cfg_get(sa->nic);
3543 		sfc_eth_dev_create_repr_controller(sa, eth_da, encp->enc_intf);
3544 	}
3545 
3546 	return 0;
3547 }
3548 
3549 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
3550 	struct rte_pci_device *pci_dev)
3551 {
3552 	struct sfc_ethdev_init_data init_data;
3553 	struct rte_eth_devargs eth_da;
3554 	struct rte_eth_dev *dev;
3555 	bool dev_created;
3556 	int rc;
3557 
3558 	if (pci_dev->device.devargs != NULL) {
3559 		rc = sfc_parse_rte_devargs(pci_dev->device.devargs->args,
3560 					   &eth_da);
3561 		if (rc != 0)
3562 			return rc;
3563 	} else {
3564 		memset(&eth_da, 0, sizeof(eth_da));
3565 	}
3566 
3567 	/* If no VF representors specified, check for PF ones */
3568 	if (eth_da.nb_representor_ports > 0)
3569 		init_data.nb_representors = eth_da.nb_representor_ports;
3570 	else
3571 		init_data.nb_representors = eth_da.nb_ports;
3572 
3573 	if (init_data.nb_representors > 0 &&
3574 	    rte_eal_process_type() != RTE_PROC_PRIMARY) {
3575 		SFC_GENERIC_LOG(ERR,
3576 			"Create representors from secondary process not supported, dev '%s'",
3577 			pci_dev->device.name);
3578 		return -ENOTSUP;
3579 	}
3580 
3581 	/*
3582 	 * Driver supports RTE_PCI_DRV_PROBE_AGAIN. Hence create device only
3583 	 * if it does not already exist. Re-probing an existing device is
3584 	 * expected to allow additional representors to be configured.
3585 	 */
3586 	rc = sfc_eth_dev_find_or_create(pci_dev, &init_data, &dev,
3587 					&dev_created);
3588 	if (rc != 0)
3589 		return rc;
3590 
3591 	rc = sfc_eth_dev_create_representors(dev, &eth_da);
3592 	if (rc != 0) {
3593 		if (dev_created)
3594 			(void)rte_eth_dev_destroy(dev, sfc_eth_dev_uninit);
3595 
3596 		return rc;
3597 	}
3598 
3599 	return 0;
3600 }
3601 
3602 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
3603 {
3604 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
3605 }
3606 
3607 static struct rte_pci_driver sfc_efx_pmd = {
3608 	.id_table = pci_id_sfc_efx_map,
3609 	.drv_flags =
3610 		RTE_PCI_DRV_INTR_LSC |
3611 		RTE_PCI_DRV_NEED_MAPPING |
3612 		RTE_PCI_DRV_PROBE_AGAIN,
3613 	.probe = sfc_eth_dev_pci_probe,
3614 	.remove = sfc_eth_dev_pci_remove,
3615 };
3616 
3617 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
3618 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
3619 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
3620 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
3621 	SFC_KVARG_SWITCH_MODE "=" SFC_KVARG_VALUES_SWITCH_MODE " "
3622 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
3623 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
3624 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
3625 	SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
3626 	SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
3627 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
3628 RTE_LOG_REGISTER_SUFFIX(sfc_logtype_driver, driver, NOTICE);
3629