xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision 0f1dc8cb671203d52488fd66936f2fe6dcca03cc)
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 uint32_t sfc_logtype_driver;
41 
42 static struct sfc_dp_list sfc_dp_head =
43 	TAILQ_HEAD_INITIALIZER(sfc_dp_head);
44 
45 
46 static void sfc_eth_dev_clear_ops(struct rte_eth_dev *dev);
47 
48 
49 static int
50 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
51 {
52 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
53 	efx_nic_fw_info_t enfi;
54 	int ret;
55 	int rc;
56 
57 	rc = efx_nic_get_fw_version(sa->nic, &enfi);
58 	if (rc != 0)
59 		return -rc;
60 
61 	ret = snprintf(fw_version, fw_size,
62 		       "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
63 		       enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
64 		       enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
65 	if (ret < 0)
66 		return ret;
67 
68 	if (enfi.enfi_dpcpu_fw_ids_valid) {
69 		size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
70 		int ret_extra;
71 
72 		ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
73 				     fw_size - dpcpu_fw_ids_offset,
74 				     " rx%" PRIx16 " tx%" PRIx16,
75 				     enfi.enfi_rx_dpcpu_fw_id,
76 				     enfi.enfi_tx_dpcpu_fw_id);
77 		if (ret_extra < 0)
78 			return ret_extra;
79 
80 		ret += ret_extra;
81 	}
82 
83 	if (fw_size < (size_t)(++ret))
84 		return ret;
85 	else
86 		return 0;
87 }
88 
89 static int
90 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
91 {
92 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
93 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
94 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
95 	struct sfc_rss *rss = &sas->rss;
96 	struct sfc_mae *mae = &sa->mae;
97 
98 	sfc_log_init(sa, "entry");
99 
100 	dev_info->min_mtu = RTE_ETHER_MIN_MTU;
101 	dev_info->max_mtu = EFX_MAC_SDU_MAX;
102 
103 	dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
104 
105 	dev_info->max_vfs = sa->sriov.num_vfs;
106 
107 	/* Autonegotiation may be disabled */
108 	dev_info->speed_capa = RTE_ETH_LINK_SPEED_FIXED;
109 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_1000FDX))
110 		dev_info->speed_capa |= RTE_ETH_LINK_SPEED_1G;
111 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_10000FDX))
112 		dev_info->speed_capa |= RTE_ETH_LINK_SPEED_10G;
113 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_25000FDX))
114 		dev_info->speed_capa |= RTE_ETH_LINK_SPEED_25G;
115 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_40000FDX))
116 		dev_info->speed_capa |= RTE_ETH_LINK_SPEED_40G;
117 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_50000FDX))
118 		dev_info->speed_capa |= RTE_ETH_LINK_SPEED_50G;
119 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_100000FDX))
120 		dev_info->speed_capa |= RTE_ETH_LINK_SPEED_100G;
121 
122 	dev_info->max_rx_queues = sa->rxq_max;
123 	dev_info->max_tx_queues = sa->txq_max;
124 
125 	/* By default packets are dropped if no descriptors are available */
126 	dev_info->default_rxconf.rx_drop_en = 1;
127 
128 	dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
129 
130 	/*
131 	 * rx_offload_capa includes both device and queue offloads since
132 	 * the latter may be requested on a per device basis which makes
133 	 * sense when some offloads are needed to be set on all queues.
134 	 */
135 	dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
136 				    dev_info->rx_queue_offload_capa;
137 
138 	dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
139 
140 	/*
141 	 * tx_offload_capa includes both device and queue offloads since
142 	 * the latter may be requested on a per device basis which makes
143 	 * sense when some offloads are needed to be set on all queues.
144 	 */
145 	dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
146 				    dev_info->tx_queue_offload_capa;
147 
148 	if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
149 		uint64_t rte_hf = 0;
150 		unsigned int i;
151 
152 		for (i = 0; i < rss->hf_map_nb_entries; ++i)
153 			rte_hf |= rss->hf_map[i].rte;
154 
155 		dev_info->reta_size = EFX_RSS_TBL_SIZE;
156 		dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
157 		dev_info->flow_type_rss_offloads = rte_hf;
158 	}
159 
160 	/* Initialize to hardware limits */
161 	dev_info->rx_desc_lim.nb_max = sa->rxq_max_entries;
162 	dev_info->rx_desc_lim.nb_min = sa->rxq_min_entries;
163 	/* The RXQ hardware requires that the descriptor count is a power
164 	 * of 2, but rx_desc_lim cannot properly describe that constraint.
165 	 */
166 	dev_info->rx_desc_lim.nb_align = sa->rxq_min_entries;
167 
168 	/* Initialize to hardware limits */
169 	dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
170 	dev_info->tx_desc_lim.nb_min = sa->txq_min_entries;
171 	/*
172 	 * The TXQ hardware requires that the descriptor count is a power
173 	 * of 2, but tx_desc_lim cannot properly describe that constraint
174 	 */
175 	dev_info->tx_desc_lim.nb_align = sa->txq_min_entries;
176 
177 	if (sap->dp_rx->get_dev_info != NULL)
178 		sap->dp_rx->get_dev_info(dev_info);
179 	if (sap->dp_tx->get_dev_info != NULL)
180 		sap->dp_tx->get_dev_info(dev_info);
181 
182 	dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
183 			     RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
184 	dev_info->dev_capa &= ~RTE_ETH_DEV_CAPA_FLOW_RULE_KEEP;
185 
186 	if (mae->status == SFC_MAE_STATUS_SUPPORTED ||
187 	    mae->status == SFC_MAE_STATUS_ADMIN) {
188 		dev_info->switch_info.name = dev->device->driver->name;
189 		dev_info->switch_info.domain_id = mae->switch_domain_id;
190 		dev_info->switch_info.port_id = mae->switch_port_id;
191 	}
192 
193 	return 0;
194 }
195 
196 static const uint32_t *
197 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev, size_t *no_of_elements)
198 {
199 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
200 
201 	return sap->dp_rx->supported_ptypes_get(sap->shared->tunnel_encaps,
202 						no_of_elements);
203 }
204 
205 static int
206 sfc_dev_configure(struct rte_eth_dev *dev)
207 {
208 	struct rte_eth_dev_data *dev_data = dev->data;
209 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
210 	int rc;
211 
212 	sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
213 		     dev_data->nb_rx_queues, dev_data->nb_tx_queues);
214 
215 	sfc_adapter_lock(sa);
216 	switch (sa->state) {
217 	case SFC_ETHDEV_CONFIGURED:
218 		/* FALLTHROUGH */
219 	case SFC_ETHDEV_INITIALIZED:
220 		rc = sfc_configure(sa);
221 		break;
222 	default:
223 		sfc_err(sa, "unexpected adapter state %u to configure",
224 			sa->state);
225 		rc = EINVAL;
226 		break;
227 	}
228 	sfc_adapter_unlock(sa);
229 
230 	sfc_log_init(sa, "done %d", rc);
231 	SFC_ASSERT(rc >= 0);
232 	return -rc;
233 }
234 
235 static int
236 sfc_dev_start(struct rte_eth_dev *dev)
237 {
238 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
239 	int rc;
240 
241 	sfc_log_init(sa, "entry");
242 
243 	sfc_adapter_lock(sa);
244 	rc = sfc_start(sa);
245 	sfc_adapter_unlock(sa);
246 
247 	sfc_log_init(sa, "done %d", rc);
248 	SFC_ASSERT(rc >= 0);
249 	return -rc;
250 }
251 
252 static void
253 sfc_dev_get_rte_link(struct rte_eth_dev *dev, int wait_to_complete,
254 		     struct rte_eth_link *link)
255 {
256 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
257 
258 	SFC_ASSERT(link != NULL);
259 
260 	if (sa->state != SFC_ETHDEV_STARTED) {
261 		sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, link);
262 	} else if (wait_to_complete) {
263 		efx_link_mode_t link_mode;
264 
265 		if (efx_port_poll(sa->nic, &link_mode) != 0)
266 			link_mode = EFX_LINK_UNKNOWN;
267 		sfc_port_link_mode_to_info(link_mode, link);
268 	} else {
269 		sfc_ev_mgmt_qpoll(sa);
270 		rte_eth_linkstatus_get(dev, link);
271 	}
272 }
273 
274 static int
275 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
276 {
277 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
278 	struct rte_eth_link current_link;
279 	int ret;
280 
281 	sfc_log_init(sa, "entry");
282 
283 	sfc_dev_get_rte_link(dev, wait_to_complete, &current_link);
284 
285 	ret = rte_eth_linkstatus_set(dev, &current_link);
286 	if (ret == 0)
287 		sfc_notice(sa, "Link status is %s",
288 			   current_link.link_status ? "UP" : "DOWN");
289 
290 	return ret;
291 }
292 
293 static int
294 sfc_dev_stop(struct rte_eth_dev *dev)
295 {
296 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
297 
298 	sfc_log_init(sa, "entry");
299 
300 	sfc_adapter_lock(sa);
301 	sfc_stop(sa);
302 	sfc_adapter_unlock(sa);
303 
304 	sfc_log_init(sa, "done");
305 
306 	return 0;
307 }
308 
309 static int
310 sfc_dev_set_link_up(struct rte_eth_dev *dev)
311 {
312 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
313 	int rc;
314 
315 	sfc_log_init(sa, "entry");
316 
317 	sfc_adapter_lock(sa);
318 	rc = sfc_start(sa);
319 	sfc_adapter_unlock(sa);
320 
321 	SFC_ASSERT(rc >= 0);
322 	return -rc;
323 }
324 
325 static int
326 sfc_dev_set_link_down(struct rte_eth_dev *dev)
327 {
328 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
329 
330 	sfc_log_init(sa, "entry");
331 
332 	sfc_adapter_lock(sa);
333 	sfc_stop(sa);
334 	sfc_adapter_unlock(sa);
335 
336 	return 0;
337 }
338 
339 static void
340 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
341 {
342 	free(dev->process_private);
343 	rte_eth_dev_release_port(dev);
344 }
345 
346 static int
347 sfc_dev_close(struct rte_eth_dev *dev)
348 {
349 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
350 
351 	sfc_log_init(sa, "entry");
352 
353 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
354 		sfc_eth_dev_secondary_clear_ops(dev);
355 		return 0;
356 	}
357 
358 	sfc_pre_detach(sa);
359 
360 	sfc_adapter_lock(sa);
361 	switch (sa->state) {
362 	case SFC_ETHDEV_STARTED:
363 		sfc_stop(sa);
364 		SFC_ASSERT(sa->state == SFC_ETHDEV_CONFIGURED);
365 		/* FALLTHROUGH */
366 	case SFC_ETHDEV_CONFIGURED:
367 		sfc_close(sa);
368 		SFC_ASSERT(sa->state == SFC_ETHDEV_INITIALIZED);
369 		/* FALLTHROUGH */
370 	case SFC_ETHDEV_INITIALIZED:
371 		break;
372 	default:
373 		sfc_err(sa, "unexpected adapter state %u on close", sa->state);
374 		break;
375 	}
376 
377 	/*
378 	 * Cleanup all resources.
379 	 * Rollback primary process sfc_eth_dev_init() below.
380 	 */
381 
382 	sfc_eth_dev_clear_ops(dev);
383 
384 	sfc_nic_dma_detach(sa);
385 	sfc_detach(sa);
386 	sfc_unprobe(sa);
387 
388 	sfc_kvargs_cleanup(sa);
389 
390 	sfc_adapter_unlock(sa);
391 	sfc_adapter_lock_fini(sa);
392 
393 	sfc_log_init(sa, "done");
394 
395 	/* Required for logging, so cleanup last */
396 	sa->eth_dev = NULL;
397 
398 	free(sa);
399 
400 	return 0;
401 }
402 
403 static int
404 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
405 		   boolean_t enabled)
406 {
407 	struct sfc_port *port;
408 	boolean_t *toggle;
409 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
410 	boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
411 	const char *desc = (allmulti) ? "all-multi" : "promiscuous";
412 	int rc = 0;
413 
414 	sfc_adapter_lock(sa);
415 
416 	port = &sa->port;
417 	toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
418 
419 	if (*toggle != enabled) {
420 		*toggle = enabled;
421 
422 		if (sfc_sa2shared(sa)->isolated) {
423 			sfc_warn(sa, "isolated mode is active on the port");
424 			sfc_warn(sa, "the change is to be applied on the next "
425 				     "start provided that isolated mode is "
426 				     "disabled prior the next start");
427 		} else if ((sa->state == SFC_ETHDEV_STARTED) &&
428 			   ((rc = sfc_set_rx_mode(sa)) != 0)) {
429 			*toggle = !(enabled);
430 			sfc_warn(sa, "Failed to %s %s mode, rc = %d",
431 				 ((enabled) ? "enable" : "disable"), desc, rc);
432 
433 			/*
434 			 * For promiscuous and all-multicast filters a
435 			 * permission failure should be reported as an
436 			 * unsupported filter.
437 			 */
438 			if (rc == EPERM)
439 				rc = ENOTSUP;
440 		}
441 	}
442 
443 	sfc_adapter_unlock(sa);
444 	return rc;
445 }
446 
447 static int
448 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
449 {
450 	int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
451 
452 	SFC_ASSERT(rc >= 0);
453 	return -rc;
454 }
455 
456 static int
457 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
458 {
459 	int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
460 
461 	SFC_ASSERT(rc >= 0);
462 	return -rc;
463 }
464 
465 static int
466 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
467 {
468 	int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
469 
470 	SFC_ASSERT(rc >= 0);
471 	return -rc;
472 }
473 
474 static int
475 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
476 {
477 	int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
478 
479 	SFC_ASSERT(rc >= 0);
480 	return -rc;
481 }
482 
483 static int
484 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid,
485 		   uint16_t nb_rx_desc, unsigned int socket_id,
486 		   const struct rte_eth_rxconf *rx_conf,
487 		   struct rte_mempool *mb_pool)
488 {
489 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
490 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
491 	sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
492 	struct sfc_rxq_info *rxq_info;
493 	sfc_sw_index_t sw_index;
494 	int rc;
495 
496 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
497 		     ethdev_qid, nb_rx_desc, socket_id);
498 
499 	sfc_adapter_lock(sa);
500 
501 	sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
502 	rc = sfc_rx_qinit(sa, sw_index, nb_rx_desc, socket_id,
503 			  rx_conf, mb_pool);
504 	if (rc != 0)
505 		goto fail_rx_qinit;
506 
507 	rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
508 	dev->data->rx_queues[ethdev_qid] = rxq_info->dp;
509 
510 	sfc_adapter_unlock(sa);
511 
512 	return 0;
513 
514 fail_rx_qinit:
515 	sfc_adapter_unlock(sa);
516 	SFC_ASSERT(rc > 0);
517 	return -rc;
518 }
519 
520 static void
521 sfc_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
522 {
523 	struct sfc_dp_rxq *dp_rxq = dev->data->rx_queues[qid];
524 	struct sfc_rxq *rxq;
525 	struct sfc_adapter *sa;
526 	sfc_sw_index_t sw_index;
527 
528 	if (dp_rxq == NULL)
529 		return;
530 
531 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
532 	sa = rxq->evq->sa;
533 	sfc_adapter_lock(sa);
534 
535 	sw_index = dp_rxq->dpq.queue_id;
536 
537 	sfc_log_init(sa, "RxQ=%u", sw_index);
538 
539 	sfc_rx_qfini(sa, sw_index);
540 
541 	sfc_adapter_unlock(sa);
542 }
543 
544 static int
545 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid,
546 		   uint16_t nb_tx_desc, unsigned int socket_id,
547 		   const struct rte_eth_txconf *tx_conf)
548 {
549 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
550 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
551 	struct sfc_txq_info *txq_info;
552 	sfc_sw_index_t sw_index;
553 	int rc;
554 
555 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
556 		     ethdev_qid, nb_tx_desc, socket_id);
557 
558 	sfc_adapter_lock(sa);
559 
560 	sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
561 	rc = sfc_tx_qinit(sa, sw_index, nb_tx_desc, socket_id, tx_conf);
562 	if (rc != 0)
563 		goto fail_tx_qinit;
564 
565 	txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
566 	dev->data->tx_queues[ethdev_qid] = txq_info->dp;
567 
568 	sfc_adapter_unlock(sa);
569 	return 0;
570 
571 fail_tx_qinit:
572 	sfc_adapter_unlock(sa);
573 	SFC_ASSERT(rc > 0);
574 	return -rc;
575 }
576 
577 static void
578 sfc_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
579 {
580 	struct sfc_dp_txq *dp_txq = dev->data->tx_queues[qid];
581 	struct sfc_txq *txq;
582 	sfc_sw_index_t sw_index;
583 	struct sfc_adapter *sa;
584 
585 	if (dp_txq == NULL)
586 		return;
587 
588 	txq = sfc_txq_by_dp_txq(dp_txq);
589 	sw_index = dp_txq->dpq.queue_id;
590 
591 	SFC_ASSERT(txq->evq != NULL);
592 	sa = txq->evq->sa;
593 
594 	sfc_log_init(sa, "TxQ = %u", sw_index);
595 
596 	sfc_adapter_lock(sa);
597 
598 	sfc_tx_qfini(sa, sw_index);
599 
600 	sfc_adapter_unlock(sa);
601 }
602 
603 static void
604 sfc_stats_get_dp_rx(struct sfc_adapter *sa, uint64_t *pkts, uint64_t *bytes)
605 {
606 	struct sfc_adapter_shared *sas = sfc_sa2shared(sa);
607 	uint64_t pkts_sum = 0;
608 	uint64_t bytes_sum = 0;
609 	unsigned int i;
610 
611 	for (i = 0; i < sas->ethdev_rxq_count; ++i) {
612 		struct sfc_rxq_info *rxq_info;
613 
614 		rxq_info = sfc_rxq_info_by_ethdev_qid(sas, i);
615 		if (rxq_info->state & SFC_RXQ_INITIALIZED) {
616 			union sfc_pkts_bytes qstats;
617 
618 			sfc_pkts_bytes_get(&rxq_info->dp->dpq.stats, &qstats);
619 			pkts_sum += qstats.pkts -
620 					sa->sw_stats.reset_rx_pkts[i];
621 			bytes_sum += qstats.bytes -
622 					sa->sw_stats.reset_rx_bytes[i];
623 		}
624 	}
625 
626 	*pkts = pkts_sum;
627 	*bytes = bytes_sum;
628 }
629 
630 static void
631 sfc_stats_get_dp_tx(struct sfc_adapter *sa, uint64_t *pkts, uint64_t *bytes)
632 {
633 	struct sfc_adapter_shared *sas = sfc_sa2shared(sa);
634 	uint64_t pkts_sum = 0;
635 	uint64_t bytes_sum = 0;
636 	unsigned int i;
637 
638 	for (i = 0; i < sas->ethdev_txq_count; ++i) {
639 		struct sfc_txq_info *txq_info;
640 
641 		txq_info = sfc_txq_info_by_ethdev_qid(sas, i);
642 		if (txq_info->state & SFC_TXQ_INITIALIZED) {
643 			union sfc_pkts_bytes qstats;
644 
645 			sfc_pkts_bytes_get(&txq_info->dp->dpq.stats, &qstats);
646 			pkts_sum += qstats.pkts -
647 					sa->sw_stats.reset_tx_pkts[i];
648 			bytes_sum += qstats.bytes -
649 					sa->sw_stats.reset_tx_bytes[i];
650 		}
651 	}
652 
653 	*pkts = pkts_sum;
654 	*bytes = bytes_sum;
655 }
656 
657 /*
658  * Some statistics are computed as A - B where A and B each increase
659  * monotonically with some hardware counter(s) and the counters are read
660  * asynchronously.
661  *
662  * If packet X is counted in A, but not counted in B yet, computed value is
663  * greater than real.
664  *
665  * If packet X is not counted in A at the moment of reading the counter,
666  * but counted in B at the moment of reading the counter, computed value
667  * is less than real.
668  *
669  * However, counter which grows backward is worse evil than slightly wrong
670  * value. So, let's try to guarantee that it never happens except may be
671  * the case when the MAC stats are zeroed as a result of a NIC reset.
672  */
673 static void
674 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
675 {
676 	if ((int64_t)(newval - *stat) > 0 || newval == 0)
677 		*stat = newval;
678 }
679 
680 static int
681 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
682 {
683 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
684 	bool have_dp_rx_stats = sap->dp_rx->features & SFC_DP_RX_FEAT_STATS;
685 	bool have_dp_tx_stats = sap->dp_tx->features & SFC_DP_TX_FEAT_STATS;
686 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
687 	struct sfc_port *port = &sa->port;
688 	uint64_t *mac_stats;
689 	int ret;
690 
691 	sfc_adapter_lock(sa);
692 
693 	if (have_dp_rx_stats) {
694 		sfc_stats_get_dp_rx(sa, &stats->ipackets, &stats->ibytes);
695 		if (dev->data->dev_conf.rxmode.offloads &
696 		    RTE_ETH_RX_OFFLOAD_KEEP_CRC) {
697 			stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
698 		}
699 	}
700 	if (have_dp_tx_stats)
701 		sfc_stats_get_dp_tx(sa, &stats->opackets, &stats->obytes);
702 
703 	ret = sfc_port_update_mac_stats(sa, B_FALSE);
704 	if (ret != 0)
705 		goto unlock;
706 
707 	mac_stats = port->mac_stats_buf;
708 
709 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
710 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
711 		if (!have_dp_rx_stats) {
712 			stats->ipackets =
713 				mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
714 				mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
715 				mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
716 			stats->ibytes =
717 				mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
718 				mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
719 				mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
720 
721 			/* CRC is included in these stats, but shouldn't be */
722 			stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
723 		}
724 		if (!have_dp_tx_stats) {
725 			stats->opackets =
726 				mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
727 				mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
728 				mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
729 			stats->obytes =
730 				mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
731 				mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
732 				mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
733 
734 			/* CRC is included in these stats, but shouldn't be */
735 			stats->obytes -= stats->opackets * RTE_ETHER_CRC_LEN;
736 		}
737 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
738 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
739 	} else {
740 		if (!have_dp_tx_stats) {
741 			stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
742 			stats->obytes = mac_stats[EFX_MAC_TX_OCTETS] -
743 				mac_stats[EFX_MAC_TX_PKTS] * RTE_ETHER_CRC_LEN;
744 		}
745 
746 		/*
747 		 * Take into account stats which are whenever supported
748 		 * on EF10. If some stat is not supported by current
749 		 * firmware variant or HW revision, it is guaranteed
750 		 * to be zero in mac_stats.
751 		 */
752 		stats->imissed =
753 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
754 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
755 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
756 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
757 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
758 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
759 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
760 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
761 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
762 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
763 		stats->ierrors =
764 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
765 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
766 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
767 		/* no oerrors counters supported on EF10 */
768 
769 		if (!have_dp_rx_stats) {
770 			/* Exclude missed, errors and pauses from Rx packets */
771 			sfc_update_diff_stat(&port->ipackets,
772 				mac_stats[EFX_MAC_RX_PKTS] -
773 				mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
774 				stats->imissed - stats->ierrors);
775 			stats->ipackets = port->ipackets;
776 			stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS] -
777 				mac_stats[EFX_MAC_RX_PKTS] * RTE_ETHER_CRC_LEN;
778 		}
779 	}
780 
781 unlock:
782 	sfc_adapter_unlock(sa);
783 	SFC_ASSERT(ret >= 0);
784 	return -ret;
785 }
786 
787 static int
788 sfc_stats_reset(struct rte_eth_dev *dev)
789 {
790 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
791 	struct sfc_port *port = &sa->port;
792 	int rc;
793 
794 	sfc_adapter_lock(sa);
795 
796 	if (sa->state != SFC_ETHDEV_STARTED) {
797 		/*
798 		 * The operation cannot be done if port is not started; it
799 		 * will be scheduled to be done during the next port start
800 		 */
801 		port->mac_stats_reset_pending = B_TRUE;
802 		sfc_adapter_unlock(sa);
803 		return 0;
804 	}
805 
806 	rc = sfc_port_reset_mac_stats(sa);
807 	if (rc != 0)
808 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
809 
810 	sfc_sw_xstats_reset(sa);
811 
812 	sfc_adapter_unlock(sa);
813 
814 	SFC_ASSERT(rc >= 0);
815 	return -rc;
816 }
817 
818 static unsigned int
819 sfc_xstats_get_nb_supported(struct sfc_adapter *sa)
820 {
821 	struct sfc_port *port = &sa->port;
822 	unsigned int nb_supported;
823 
824 	sfc_adapter_lock(sa);
825 	nb_supported = port->mac_stats_nb_supported +
826 		       sfc_sw_xstats_get_nb_supported(sa);
827 	sfc_adapter_unlock(sa);
828 
829 	return nb_supported;
830 }
831 
832 static int
833 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
834 	       unsigned int xstats_count)
835 {
836 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
837 	unsigned int nb_written = 0;
838 	unsigned int nb_supported = 0;
839 	int rc;
840 
841 	if (unlikely(xstats == NULL))
842 		return sfc_xstats_get_nb_supported(sa);
843 
844 	rc = sfc_port_get_mac_stats(sa, xstats, xstats_count, &nb_written);
845 	if (rc < 0)
846 		return rc;
847 
848 	nb_supported = rc;
849 	sfc_sw_xstats_get_vals(sa, xstats, xstats_count, &nb_written,
850 			       &nb_supported);
851 
852 	return nb_supported;
853 }
854 
855 static int
856 sfc_xstats_get_names(struct rte_eth_dev *dev,
857 		     struct rte_eth_xstat_name *xstats_names,
858 		     unsigned int xstats_count)
859 {
860 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
861 	struct sfc_port *port = &sa->port;
862 	unsigned int i;
863 	unsigned int nstats = 0;
864 	unsigned int nb_written = 0;
865 	int ret;
866 
867 	if (unlikely(xstats_names == NULL))
868 		return sfc_xstats_get_nb_supported(sa);
869 
870 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
871 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
872 			if (nstats < xstats_count) {
873 				strlcpy(xstats_names[nstats].name,
874 					efx_mac_stat_name(sa->nic, i),
875 					sizeof(xstats_names[0].name));
876 				nb_written++;
877 			}
878 			nstats++;
879 		}
880 	}
881 
882 	ret = sfc_sw_xstats_get_names(sa, xstats_names, xstats_count,
883 				      &nb_written, &nstats);
884 	if (ret != 0) {
885 		SFC_ASSERT(ret < 0);
886 		return ret;
887 	}
888 
889 	return nstats;
890 }
891 
892 static int
893 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
894 		     uint64_t *values, unsigned int n)
895 {
896 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
897 	struct sfc_port *port = &sa->port;
898 	unsigned int nb_supported;
899 	unsigned int i;
900 	int rc;
901 
902 	if (unlikely(ids == NULL || values == NULL))
903 		return -EINVAL;
904 
905 	/*
906 	 * Values array could be filled in nonsequential order. Fill values with
907 	 * constant indicating invalid ID first.
908 	 */
909 	for (i = 0; i < n; i++)
910 		values[i] = SFC_XSTAT_ID_INVALID_VAL;
911 
912 	rc = sfc_port_get_mac_stats_by_id(sa, ids, values, n);
913 	if (rc != 0)
914 		return rc;
915 
916 	nb_supported = port->mac_stats_nb_supported;
917 	sfc_sw_xstats_get_vals_by_id(sa, ids, values, n, &nb_supported);
918 
919 	/* Return number of written stats before invalid ID is encountered. */
920 	for (i = 0; i < n; i++) {
921 		if (values[i] == SFC_XSTAT_ID_INVALID_VAL)
922 			return i;
923 	}
924 
925 	return n;
926 }
927 
928 static int
929 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
930 			   const uint64_t *ids,
931 			   struct rte_eth_xstat_name *xstats_names,
932 			   unsigned int size)
933 {
934 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
935 	struct sfc_port *port = &sa->port;
936 	unsigned int nb_supported;
937 	unsigned int i;
938 	int ret;
939 
940 	if (unlikely(xstats_names == NULL && ids != NULL) ||
941 	    unlikely(xstats_names != NULL && ids == NULL))
942 		return -EINVAL;
943 
944 	if (unlikely(xstats_names == NULL && ids == NULL))
945 		return sfc_xstats_get_nb_supported(sa);
946 
947 	/*
948 	 * Names array could be filled in nonsequential order. Fill names with
949 	 * string indicating invalid ID first.
950 	 */
951 	for (i = 0; i < size; i++)
952 		xstats_names[i].name[0] = SFC_XSTAT_ID_INVALID_NAME;
953 
954 	sfc_adapter_lock(sa);
955 
956 	SFC_ASSERT(port->mac_stats_nb_supported <=
957 		   RTE_DIM(port->mac_stats_by_id));
958 
959 	for (i = 0; i < size; i++) {
960 		if (ids[i] < port->mac_stats_nb_supported) {
961 			strlcpy(xstats_names[i].name,
962 				efx_mac_stat_name(sa->nic,
963 						 port->mac_stats_by_id[ids[i]]),
964 				sizeof(xstats_names[0].name));
965 		}
966 	}
967 
968 	nb_supported = port->mac_stats_nb_supported;
969 
970 	sfc_adapter_unlock(sa);
971 
972 	ret = sfc_sw_xstats_get_names_by_id(sa, ids, xstats_names, size,
973 					    &nb_supported);
974 	if (ret != 0) {
975 		SFC_ASSERT(ret < 0);
976 		return ret;
977 	}
978 
979 	/* Return number of written names before invalid ID is encountered. */
980 	for (i = 0; i < size; i++) {
981 		if (xstats_names[i].name[0] == SFC_XSTAT_ID_INVALID_NAME)
982 			return i;
983 	}
984 
985 	return size;
986 }
987 
988 static int
989 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
990 {
991 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
992 	unsigned int wanted_fc, link_fc;
993 
994 	memset(fc_conf, 0, sizeof(*fc_conf));
995 
996 	sfc_adapter_lock(sa);
997 
998 	if (sa->state == SFC_ETHDEV_STARTED)
999 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
1000 	else
1001 		link_fc = sa->port.flow_ctrl;
1002 
1003 	switch (link_fc) {
1004 	case 0:
1005 		fc_conf->mode = RTE_ETH_FC_NONE;
1006 		break;
1007 	case EFX_FCNTL_RESPOND:
1008 		fc_conf->mode = RTE_ETH_FC_RX_PAUSE;
1009 		break;
1010 	case EFX_FCNTL_GENERATE:
1011 		fc_conf->mode = RTE_ETH_FC_TX_PAUSE;
1012 		break;
1013 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
1014 		fc_conf->mode = RTE_ETH_FC_FULL;
1015 		break;
1016 	default:
1017 		sfc_err(sa, "%s: unexpected flow control value %#x",
1018 			__func__, link_fc);
1019 	}
1020 
1021 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
1022 
1023 	sfc_adapter_unlock(sa);
1024 
1025 	return 0;
1026 }
1027 
1028 static int
1029 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1030 {
1031 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1032 	struct sfc_port *port = &sa->port;
1033 	unsigned int fcntl;
1034 	int rc;
1035 
1036 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
1037 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
1038 	    fc_conf->mac_ctrl_frame_fwd != 0) {
1039 		sfc_err(sa, "unsupported flow control settings specified");
1040 		rc = EINVAL;
1041 		goto fail_inval;
1042 	}
1043 
1044 	switch (fc_conf->mode) {
1045 	case RTE_ETH_FC_NONE:
1046 		fcntl = 0;
1047 		break;
1048 	case RTE_ETH_FC_RX_PAUSE:
1049 		fcntl = EFX_FCNTL_RESPOND;
1050 		break;
1051 	case RTE_ETH_FC_TX_PAUSE:
1052 		fcntl = EFX_FCNTL_GENERATE;
1053 		break;
1054 	case RTE_ETH_FC_FULL:
1055 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
1056 		break;
1057 	default:
1058 		rc = EINVAL;
1059 		goto fail_inval;
1060 	}
1061 
1062 	sfc_adapter_lock(sa);
1063 
1064 	if (sa->state == SFC_ETHDEV_STARTED) {
1065 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
1066 		if (rc != 0)
1067 			goto fail_mac_fcntl_set;
1068 	}
1069 
1070 	port->flow_ctrl = fcntl;
1071 	port->flow_ctrl_autoneg = fc_conf->autoneg;
1072 
1073 	sfc_adapter_unlock(sa);
1074 
1075 	return 0;
1076 
1077 fail_mac_fcntl_set:
1078 	sfc_adapter_unlock(sa);
1079 fail_inval:
1080 	SFC_ASSERT(rc > 0);
1081 	return -rc;
1082 }
1083 
1084 static int
1085 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
1086 {
1087 	struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
1088 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1089 	boolean_t scatter_enabled;
1090 	const char *error;
1091 	unsigned int i;
1092 
1093 	for (i = 0; i < sas->rxq_count; i++) {
1094 		if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
1095 			continue;
1096 
1097 		scatter_enabled = (sas->rxq_info[i].type_flags &
1098 				   EFX_RXQ_FLAG_SCATTER);
1099 
1100 		if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
1101 					  encp->enc_rx_prefix_size,
1102 					  scatter_enabled,
1103 					  encp->enc_rx_scatter_max, &error)) {
1104 			sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
1105 				error);
1106 			return EINVAL;
1107 		}
1108 	}
1109 
1110 	return 0;
1111 }
1112 
1113 static int
1114 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
1115 {
1116 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1117 	size_t pdu = EFX_MAC_PDU(mtu);
1118 	size_t old_pdu;
1119 	int rc;
1120 
1121 	sfc_log_init(sa, "mtu=%u", mtu);
1122 
1123 	rc = EINVAL;
1124 	if (pdu < EFX_MAC_PDU_MIN) {
1125 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
1126 			(unsigned int)mtu, (unsigned int)pdu,
1127 			EFX_MAC_PDU_MIN);
1128 		goto fail_inval;
1129 	}
1130 	if (pdu > EFX_MAC_PDU_MAX) {
1131 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
1132 			(unsigned int)mtu, (unsigned int)pdu,
1133 			(unsigned int)EFX_MAC_PDU_MAX);
1134 		goto fail_inval;
1135 	}
1136 
1137 	sfc_adapter_lock(sa);
1138 
1139 	rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
1140 	if (rc != 0)
1141 		goto fail_check_scatter;
1142 
1143 	if (pdu != sa->port.pdu) {
1144 		if (sa->state == SFC_ETHDEV_STARTED) {
1145 			sfc_stop(sa);
1146 
1147 			old_pdu = sa->port.pdu;
1148 			sa->port.pdu = pdu;
1149 			rc = sfc_start(sa);
1150 			if (rc != 0)
1151 				goto fail_start;
1152 		} else {
1153 			sa->port.pdu = pdu;
1154 		}
1155 	}
1156 
1157 	sfc_adapter_unlock(sa);
1158 
1159 	sfc_log_init(sa, "done");
1160 	return 0;
1161 
1162 fail_start:
1163 	sa->port.pdu = old_pdu;
1164 	if (sfc_start(sa) != 0)
1165 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1166 			"PDU max size - port is stopped",
1167 			(unsigned int)pdu, (unsigned int)old_pdu);
1168 
1169 fail_check_scatter:
1170 	sfc_adapter_unlock(sa);
1171 
1172 fail_inval:
1173 	sfc_log_init(sa, "failed %d", rc);
1174 	SFC_ASSERT(rc > 0);
1175 	return -rc;
1176 }
1177 static int
1178 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1179 {
1180 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1181 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1182 	struct sfc_port *port = &sa->port;
1183 	struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1184 	int rc = 0;
1185 
1186 	sfc_adapter_lock(sa);
1187 
1188 	if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr))
1189 		goto unlock;
1190 
1191 	/*
1192 	 * Copy the address to the device private data so that
1193 	 * it could be recalled in the case of adapter restart.
1194 	 */
1195 	rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1196 
1197 	/*
1198 	 * Neither of the two following checks can return
1199 	 * an error. The new MAC address is preserved in
1200 	 * the device private data and can be activated
1201 	 * on the next port start if the user prevents
1202 	 * isolated mode from being enabled.
1203 	 */
1204 	if (sfc_sa2shared(sa)->isolated) {
1205 		sfc_warn(sa, "isolated mode is active on the port");
1206 		sfc_warn(sa, "will not set MAC address");
1207 		goto unlock;
1208 	}
1209 
1210 	if (sa->state != SFC_ETHDEV_STARTED) {
1211 		sfc_notice(sa, "the port is not started");
1212 		sfc_notice(sa, "the new MAC address will be set on port start");
1213 
1214 		goto unlock;
1215 	}
1216 
1217 	if (encp->enc_allow_set_mac_with_installed_filters) {
1218 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1219 		if (rc != 0) {
1220 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1221 			goto unlock;
1222 		}
1223 
1224 		/*
1225 		 * Changing the MAC address by means of MCDI request
1226 		 * has no effect on received traffic, therefore
1227 		 * we also need to update unicast filters
1228 		 */
1229 		rc = sfc_set_rx_mode_unchecked(sa);
1230 		if (rc != 0) {
1231 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
1232 			/* Rollback the old address */
1233 			(void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1234 			(void)sfc_set_rx_mode_unchecked(sa);
1235 		}
1236 	} else {
1237 		sfc_warn(sa, "cannot set MAC address with filters installed");
1238 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1239 		sfc_warn(sa, "(some traffic may be dropped)");
1240 
1241 		/*
1242 		 * Since setting MAC address with filters installed is not
1243 		 * allowed on the adapter, the new MAC address will be set
1244 		 * by means of adapter restart. sfc_start() shall retrieve
1245 		 * the new address from the device private data and set it.
1246 		 */
1247 		sfc_stop(sa);
1248 		rc = sfc_start(sa);
1249 		if (rc != 0)
1250 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1251 	}
1252 
1253 unlock:
1254 	if (rc != 0)
1255 		rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1256 
1257 	sfc_adapter_unlock(sa);
1258 
1259 	SFC_ASSERT(rc >= 0);
1260 	return -rc;
1261 }
1262 
1263 
1264 static int
1265 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1266 		struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1267 {
1268 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1269 	struct sfc_port *port = &sa->port;
1270 	uint8_t *mc_addrs = port->mcast_addrs;
1271 	int rc;
1272 	unsigned int i;
1273 
1274 	if (sfc_sa2shared(sa)->isolated) {
1275 		sfc_err(sa, "isolated mode is active on the port");
1276 		sfc_err(sa, "will not set multicast address list");
1277 		return -ENOTSUP;
1278 	}
1279 
1280 	if (mc_addrs == NULL)
1281 		return -ENOBUFS;
1282 
1283 	if (nb_mc_addr > port->max_mcast_addrs) {
1284 		sfc_err(sa, "too many multicast addresses: %u > %u",
1285 			 nb_mc_addr, port->max_mcast_addrs);
1286 		return -EINVAL;
1287 	}
1288 
1289 	for (i = 0; i < nb_mc_addr; ++i) {
1290 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1291 				 EFX_MAC_ADDR_LEN);
1292 		mc_addrs += EFX_MAC_ADDR_LEN;
1293 	}
1294 
1295 	port->nb_mcast_addrs = nb_mc_addr;
1296 
1297 	if (sa->state != SFC_ETHDEV_STARTED)
1298 		return 0;
1299 
1300 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1301 					port->nb_mcast_addrs);
1302 	if (rc != 0)
1303 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1304 
1305 	SFC_ASSERT(rc >= 0);
1306 	return -rc;
1307 }
1308 
1309 /*
1310  * The function is used by the secondary process as well. It must not
1311  * use any process-local pointers from the adapter data.
1312  */
1313 static void
1314 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1315 		      struct rte_eth_rxq_info *qinfo)
1316 {
1317 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1318 	sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1319 	struct sfc_rxq_info *rxq_info;
1320 
1321 	rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1322 
1323 	qinfo->mp = rxq_info->refill_mb_pool;
1324 	qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1325 	qinfo->conf.rx_drop_en = 1;
1326 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1327 	qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1328 	if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1329 		qinfo->conf.offloads |= RTE_ETH_RX_OFFLOAD_SCATTER;
1330 		qinfo->scattered_rx = 1;
1331 	}
1332 	qinfo->nb_desc = rxq_info->entries;
1333 }
1334 
1335 /*
1336  * The function is used by the secondary process as well. It must not
1337  * use any process-local pointers from the adapter data.
1338  */
1339 static void
1340 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1341 		      struct rte_eth_txq_info *qinfo)
1342 {
1343 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1344 	struct sfc_txq_info *txq_info;
1345 
1346 	SFC_ASSERT(ethdev_qid < sas->ethdev_txq_count);
1347 
1348 	txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1349 
1350 	memset(qinfo, 0, sizeof(*qinfo));
1351 
1352 	qinfo->conf.offloads = txq_info->offloads;
1353 	qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1354 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1355 	qinfo->nb_desc = txq_info->entries;
1356 }
1357 
1358 /*
1359  * The function is used by the secondary process as well. It must not
1360  * use any process-local pointers from the adapter data.
1361  */
1362 static uint32_t
1363 sfc_rx_queue_count(void *rx_queue)
1364 {
1365 	struct sfc_dp_rxq *dp_rxq = rx_queue;
1366 	const struct sfc_dp_rx *dp_rx;
1367 	struct sfc_rxq_info *rxq_info;
1368 
1369 	dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1370 	rxq_info = sfc_rxq_info_by_dp_rxq(dp_rxq);
1371 
1372 	if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1373 		return 0;
1374 
1375 	return dp_rx->qdesc_npending(dp_rxq);
1376 }
1377 
1378 /*
1379  * The function is used by the secondary process as well. It must not
1380  * use any process-local pointers from the adapter data.
1381  */
1382 static int
1383 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1384 {
1385 	struct sfc_dp_rxq *dp_rxq = queue;
1386 	const struct sfc_dp_rx *dp_rx;
1387 
1388 	dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1389 
1390 	return dp_rx->qdesc_status(dp_rxq, offset);
1391 }
1392 
1393 /*
1394  * The function is used by the secondary process as well. It must not
1395  * use any process-local pointers from the adapter data.
1396  */
1397 static int
1398 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1399 {
1400 	struct sfc_dp_txq *dp_txq = queue;
1401 	const struct sfc_dp_tx *dp_tx;
1402 
1403 	dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1404 
1405 	return dp_tx->qdesc_status(dp_txq, offset);
1406 }
1407 
1408 static int
1409 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1410 {
1411 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1412 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1413 	sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1414 	struct sfc_rxq_info *rxq_info;
1415 	sfc_sw_index_t sw_index;
1416 	int rc;
1417 
1418 	sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1419 
1420 	sfc_adapter_lock(sa);
1421 
1422 	rc = EINVAL;
1423 	if (sa->state != SFC_ETHDEV_STARTED)
1424 		goto fail_not_started;
1425 
1426 	rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1427 	if (rxq_info->state != SFC_RXQ_INITIALIZED)
1428 		goto fail_not_setup;
1429 
1430 	sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1431 	rc = sfc_rx_qstart(sa, sw_index);
1432 	if (rc != 0)
1433 		goto fail_rx_qstart;
1434 
1435 	rxq_info->deferred_started = B_TRUE;
1436 
1437 	sfc_adapter_unlock(sa);
1438 
1439 	return 0;
1440 
1441 fail_rx_qstart:
1442 fail_not_setup:
1443 fail_not_started:
1444 	sfc_adapter_unlock(sa);
1445 	SFC_ASSERT(rc > 0);
1446 	return -rc;
1447 }
1448 
1449 static int
1450 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1451 {
1452 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1453 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1454 	sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1455 	struct sfc_rxq_info *rxq_info;
1456 	sfc_sw_index_t sw_index;
1457 
1458 	sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1459 
1460 	sfc_adapter_lock(sa);
1461 
1462 	sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1463 	sfc_rx_qstop(sa, sw_index);
1464 
1465 	rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1466 	rxq_info->deferred_started = B_FALSE;
1467 
1468 	sfc_adapter_unlock(sa);
1469 
1470 	return 0;
1471 }
1472 
1473 static int
1474 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1475 {
1476 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1477 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1478 	struct sfc_txq_info *txq_info;
1479 	sfc_sw_index_t sw_index;
1480 	int rc;
1481 
1482 	sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1483 
1484 	sfc_adapter_lock(sa);
1485 
1486 	rc = EINVAL;
1487 	if (sa->state != SFC_ETHDEV_STARTED)
1488 		goto fail_not_started;
1489 
1490 	txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1491 	if (txq_info->state != SFC_TXQ_INITIALIZED)
1492 		goto fail_not_setup;
1493 
1494 	sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1495 	rc = sfc_tx_qstart(sa, sw_index);
1496 	if (rc != 0)
1497 		goto fail_tx_qstart;
1498 
1499 	txq_info->deferred_started = B_TRUE;
1500 
1501 	sfc_adapter_unlock(sa);
1502 	return 0;
1503 
1504 fail_tx_qstart:
1505 
1506 fail_not_setup:
1507 fail_not_started:
1508 	sfc_adapter_unlock(sa);
1509 	SFC_ASSERT(rc > 0);
1510 	return -rc;
1511 }
1512 
1513 static int
1514 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1515 {
1516 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1517 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1518 	struct sfc_txq_info *txq_info;
1519 	sfc_sw_index_t sw_index;
1520 
1521 	sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1522 
1523 	sfc_adapter_lock(sa);
1524 
1525 	sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1526 	sfc_tx_qstop(sa, sw_index);
1527 
1528 	txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1529 	txq_info->deferred_started = B_FALSE;
1530 
1531 	sfc_adapter_unlock(sa);
1532 	return 0;
1533 }
1534 
1535 static efx_tunnel_protocol_t
1536 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1537 {
1538 	switch (rte_type) {
1539 	case RTE_ETH_TUNNEL_TYPE_VXLAN:
1540 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1541 	case RTE_ETH_TUNNEL_TYPE_GENEVE:
1542 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1543 	default:
1544 		return EFX_TUNNEL_NPROTOS;
1545 	}
1546 }
1547 
1548 enum sfc_udp_tunnel_op_e {
1549 	SFC_UDP_TUNNEL_ADD_PORT,
1550 	SFC_UDP_TUNNEL_DEL_PORT,
1551 };
1552 
1553 static int
1554 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1555 		      struct rte_eth_udp_tunnel *tunnel_udp,
1556 		      enum sfc_udp_tunnel_op_e op)
1557 {
1558 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1559 	efx_tunnel_protocol_t tunnel_proto;
1560 	int rc;
1561 
1562 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1563 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1564 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1565 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1566 
1567 	tunnel_proto =
1568 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1569 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1570 		rc = ENOTSUP;
1571 		goto fail_bad_proto;
1572 	}
1573 
1574 	sfc_adapter_lock(sa);
1575 
1576 	switch (op) {
1577 	case SFC_UDP_TUNNEL_ADD_PORT:
1578 		rc = efx_tunnel_config_udp_add(sa->nic,
1579 					       tunnel_udp->udp_port,
1580 					       tunnel_proto);
1581 		break;
1582 	case SFC_UDP_TUNNEL_DEL_PORT:
1583 		rc = efx_tunnel_config_udp_remove(sa->nic,
1584 						  tunnel_udp->udp_port,
1585 						  tunnel_proto);
1586 		break;
1587 	default:
1588 		rc = EINVAL;
1589 		goto fail_bad_op;
1590 	}
1591 
1592 	if (rc != 0)
1593 		goto fail_op;
1594 
1595 	if (sa->state == SFC_ETHDEV_STARTED) {
1596 		rc = efx_tunnel_reconfigure(sa->nic);
1597 		if (rc == EAGAIN) {
1598 			/*
1599 			 * Configuration is accepted by FW and MC reboot
1600 			 * is initiated to apply the changes. MC reboot
1601 			 * will be handled in a usual way (MC reboot
1602 			 * event on management event queue and adapter
1603 			 * restart).
1604 			 */
1605 			rc = 0;
1606 		} else if (rc != 0) {
1607 			goto fail_reconfigure;
1608 		}
1609 	}
1610 
1611 	sfc_adapter_unlock(sa);
1612 	return 0;
1613 
1614 fail_reconfigure:
1615 	/* Remove/restore entry since the change makes the trouble */
1616 	switch (op) {
1617 	case SFC_UDP_TUNNEL_ADD_PORT:
1618 		(void)efx_tunnel_config_udp_remove(sa->nic,
1619 						   tunnel_udp->udp_port,
1620 						   tunnel_proto);
1621 		break;
1622 	case SFC_UDP_TUNNEL_DEL_PORT:
1623 		(void)efx_tunnel_config_udp_add(sa->nic,
1624 						tunnel_udp->udp_port,
1625 						tunnel_proto);
1626 		break;
1627 	}
1628 
1629 fail_op:
1630 fail_bad_op:
1631 	sfc_adapter_unlock(sa);
1632 
1633 fail_bad_proto:
1634 	SFC_ASSERT(rc > 0);
1635 	return -rc;
1636 }
1637 
1638 static int
1639 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1640 			    struct rte_eth_udp_tunnel *tunnel_udp)
1641 {
1642 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1643 }
1644 
1645 static int
1646 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1647 			    struct rte_eth_udp_tunnel *tunnel_udp)
1648 {
1649 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1650 }
1651 
1652 /*
1653  * The function is used by the secondary process as well. It must not
1654  * use any process-local pointers from the adapter data.
1655  */
1656 static int
1657 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1658 			  struct rte_eth_rss_conf *rss_conf)
1659 {
1660 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1661 	struct sfc_rss *rss = &sas->rss;
1662 
1663 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1664 		return -ENOTSUP;
1665 
1666 	/*
1667 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1668 	 * hence, conversion is done here to derive a correct set of RTE_ETH_RSS
1669 	 * flags which corresponds to the active EFX configuration stored
1670 	 * locally in 'sfc_adapter' and kept up-to-date
1671 	 */
1672 	rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1673 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1674 	if (rss_conf->rss_key != NULL)
1675 		rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1676 
1677 	return 0;
1678 }
1679 
1680 static int
1681 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1682 			struct rte_eth_rss_conf *rss_conf)
1683 {
1684 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1685 	struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1686 	unsigned int efx_hash_types;
1687 	unsigned int n_contexts;
1688 	unsigned int mode_i = 0;
1689 	unsigned int key_i = 0;
1690 	uint32_t contexts[2];
1691 	unsigned int i = 0;
1692 	int rc = 0;
1693 
1694 	if (sfc_sa2shared(sa)->isolated)
1695 		return -ENOTSUP;
1696 
1697 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1698 		sfc_err(sa, "RSS is not available");
1699 		return -ENOTSUP;
1700 	}
1701 
1702 	if (rss->channels == 0) {
1703 		sfc_err(sa, "RSS is not configured");
1704 		return -EINVAL;
1705 	}
1706 
1707 	if ((rss_conf->rss_key != NULL) &&
1708 	    (rss_conf->rss_key_len != sizeof(rss->key))) {
1709 		sfc_err(sa, "RSS key size is wrong (should be %zu)",
1710 			sizeof(rss->key));
1711 		return -EINVAL;
1712 	}
1713 
1714 	sfc_adapter_lock(sa);
1715 
1716 	rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1717 	if (rc != 0)
1718 		goto fail_rx_hf_rte_to_efx;
1719 
1720 	contexts[0] = EFX_RSS_CONTEXT_DEFAULT;
1721 	contexts[1] = rss->dummy_ctx.nic_handle;
1722 	n_contexts = (rss->dummy_ctx.nic_handle_refcnt == 0) ? 1 : 2;
1723 
1724 	for (mode_i = 0; mode_i < n_contexts; mode_i++) {
1725 		rc = efx_rx_scale_mode_set(sa->nic, contexts[mode_i],
1726 					   rss->hash_alg, efx_hash_types,
1727 					   B_TRUE);
1728 		if (rc != 0)
1729 			goto fail_scale_mode_set;
1730 	}
1731 
1732 	if (rss_conf->rss_key != NULL) {
1733 		if (sa->state == SFC_ETHDEV_STARTED) {
1734 			for (key_i = 0; key_i < n_contexts; key_i++) {
1735 				rc = efx_rx_scale_key_set(sa->nic,
1736 							  contexts[key_i],
1737 							  rss_conf->rss_key,
1738 							  sizeof(rss->key));
1739 				if (rc != 0)
1740 					goto fail_scale_key_set;
1741 			}
1742 		}
1743 
1744 		rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1745 	}
1746 
1747 	rss->hash_types = efx_hash_types;
1748 
1749 	sfc_adapter_unlock(sa);
1750 
1751 	return 0;
1752 
1753 fail_scale_key_set:
1754 	for (i = 0; i < key_i; i++) {
1755 		if (efx_rx_scale_key_set(sa->nic, contexts[i], rss->key,
1756 					 sizeof(rss->key)) != 0)
1757 			sfc_err(sa, "failed to restore RSS key");
1758 	}
1759 
1760 fail_scale_mode_set:
1761 	for (i = 0; i < mode_i; i++) {
1762 		if (efx_rx_scale_mode_set(sa->nic, contexts[i],
1763 					  EFX_RX_HASHALG_TOEPLITZ,
1764 					  rss->hash_types, B_TRUE) != 0)
1765 			sfc_err(sa, "failed to restore RSS mode");
1766 	}
1767 
1768 fail_rx_hf_rte_to_efx:
1769 	sfc_adapter_unlock(sa);
1770 	return -rc;
1771 }
1772 
1773 /*
1774  * The function is used by the secondary process as well. It must not
1775  * use any process-local pointers from the adapter data.
1776  */
1777 static int
1778 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1779 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1780 		       uint16_t reta_size)
1781 {
1782 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1783 	struct sfc_rss *rss = &sas->rss;
1784 	int entry;
1785 
1786 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1787 		return -ENOTSUP;
1788 
1789 	if (rss->channels == 0)
1790 		return -EINVAL;
1791 
1792 	if (reta_size != EFX_RSS_TBL_SIZE)
1793 		return -EINVAL;
1794 
1795 	for (entry = 0; entry < reta_size; entry++) {
1796 		int grp = entry / RTE_ETH_RETA_GROUP_SIZE;
1797 		int grp_idx = entry % RTE_ETH_RETA_GROUP_SIZE;
1798 
1799 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1800 			reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1801 	}
1802 
1803 	return 0;
1804 }
1805 
1806 static int
1807 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1808 			struct rte_eth_rss_reta_entry64 *reta_conf,
1809 			uint16_t reta_size)
1810 {
1811 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1812 	struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1813 	unsigned int *rss_tbl_new;
1814 	uint16_t entry;
1815 	int rc = 0;
1816 
1817 
1818 	if (sfc_sa2shared(sa)->isolated)
1819 		return -ENOTSUP;
1820 
1821 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1822 		sfc_err(sa, "RSS is not available");
1823 		return -ENOTSUP;
1824 	}
1825 
1826 	if (rss->channels == 0) {
1827 		sfc_err(sa, "RSS is not configured");
1828 		return -EINVAL;
1829 	}
1830 
1831 	if (reta_size != EFX_RSS_TBL_SIZE) {
1832 		sfc_err(sa, "RETA size is wrong (should be %u)",
1833 			EFX_RSS_TBL_SIZE);
1834 		return -EINVAL;
1835 	}
1836 
1837 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1838 	if (rss_tbl_new == NULL)
1839 		return -ENOMEM;
1840 
1841 	sfc_adapter_lock(sa);
1842 
1843 	rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1844 
1845 	for (entry = 0; entry < reta_size; entry++) {
1846 		int grp_idx = entry % RTE_ETH_RETA_GROUP_SIZE;
1847 		struct rte_eth_rss_reta_entry64 *grp;
1848 
1849 		grp = &reta_conf[entry / RTE_ETH_RETA_GROUP_SIZE];
1850 
1851 		if (grp->mask & (1ull << grp_idx)) {
1852 			if (grp->reta[grp_idx] >= rss->channels) {
1853 				rc = EINVAL;
1854 				goto bad_reta_entry;
1855 			}
1856 			rss_tbl_new[entry] = grp->reta[grp_idx];
1857 		}
1858 	}
1859 
1860 	if (sa->state == SFC_ETHDEV_STARTED) {
1861 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1862 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1863 		if (rc != 0)
1864 			goto fail_scale_tbl_set;
1865 	}
1866 
1867 	rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1868 
1869 fail_scale_tbl_set:
1870 bad_reta_entry:
1871 	sfc_adapter_unlock(sa);
1872 
1873 	rte_free(rss_tbl_new);
1874 
1875 	SFC_ASSERT(rc >= 0);
1876 	return -rc;
1877 }
1878 
1879 static int
1880 sfc_dev_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
1881 		     const struct rte_flow_ops **ops)
1882 {
1883 	*ops = &sfc_flow_ops;
1884 	return 0;
1885 }
1886 
1887 static int
1888 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1889 {
1890 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1891 
1892 	/*
1893 	 * If Rx datapath does not provide callback to check mempool,
1894 	 * all pools are supported.
1895 	 */
1896 	if (sap->dp_rx->pool_ops_supported == NULL)
1897 		return 1;
1898 
1899 	return sap->dp_rx->pool_ops_supported(pool);
1900 }
1901 
1902 static int
1903 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1904 {
1905 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1906 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1907 	sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1908 	struct sfc_rxq_info *rxq_info;
1909 
1910 	rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1911 
1912 	return sap->dp_rx->intr_enable(rxq_info->dp);
1913 }
1914 
1915 static int
1916 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1917 {
1918 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1919 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1920 	sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1921 	struct sfc_rxq_info *rxq_info;
1922 
1923 	rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1924 
1925 	return sap->dp_rx->intr_disable(rxq_info->dp);
1926 }
1927 
1928 struct sfc_mport_journal_ctx {
1929 	struct sfc_adapter		*sa;
1930 	uint16_t			switch_domain_id;
1931 	uint32_t			mcdi_handle;
1932 	bool				controllers_assigned;
1933 	efx_pcie_interface_t		*controllers;
1934 	size_t				nb_controllers;
1935 };
1936 
1937 static int
1938 sfc_journal_ctx_add_controller(struct sfc_mport_journal_ctx *ctx,
1939 			       efx_pcie_interface_t intf)
1940 {
1941 	efx_pcie_interface_t *new_controllers;
1942 	size_t i, target;
1943 	size_t new_size;
1944 
1945 	if (ctx->controllers == NULL) {
1946 		ctx->controllers = rte_malloc("sfc_controller_mapping",
1947 					      sizeof(ctx->controllers[0]), 0);
1948 		if (ctx->controllers == NULL)
1949 			return ENOMEM;
1950 
1951 		ctx->controllers[0] = intf;
1952 		ctx->nb_controllers = 1;
1953 
1954 		return 0;
1955 	}
1956 
1957 	for (i = 0; i < ctx->nb_controllers; i++) {
1958 		if (ctx->controllers[i] == intf)
1959 			return 0;
1960 		if (ctx->controllers[i] > intf)
1961 			break;
1962 	}
1963 	target = i;
1964 
1965 	ctx->nb_controllers += 1;
1966 	new_size = ctx->nb_controllers * sizeof(ctx->controllers[0]);
1967 
1968 	new_controllers = rte_realloc(ctx->controllers, new_size, 0);
1969 	if (new_controllers == NULL) {
1970 		rte_free(ctx->controllers);
1971 		return ENOMEM;
1972 	}
1973 	ctx->controllers = new_controllers;
1974 
1975 	for (i = target + 1; i < ctx->nb_controllers; i++)
1976 		ctx->controllers[i] = ctx->controllers[i - 1];
1977 
1978 	ctx->controllers[target] = intf;
1979 
1980 	return 0;
1981 }
1982 
1983 static efx_rc_t
1984 sfc_process_mport_journal_entry(struct sfc_mport_journal_ctx *ctx,
1985 				efx_mport_desc_t *mport)
1986 {
1987 	struct sfc_mae_switch_port_request req;
1988 	efx_mport_sel_t entity_selector;
1989 	efx_mport_sel_t ethdev_mport;
1990 	uint16_t switch_port_id;
1991 	efx_rc_t efx_rc;
1992 	int rc;
1993 
1994 	sfc_dbg(ctx->sa,
1995 		"processing mport id %u (controller %u pf %u vf %u)",
1996 		mport->emd_id.id, mport->emd_vnic.ev_intf,
1997 		mport->emd_vnic.ev_pf, mport->emd_vnic.ev_vf);
1998 	efx_mae_mport_invalid(&ethdev_mport);
1999 
2000 	if (!ctx->controllers_assigned) {
2001 		rc = sfc_journal_ctx_add_controller(ctx,
2002 						    mport->emd_vnic.ev_intf);
2003 		if (rc != 0)
2004 			return rc;
2005 	}
2006 
2007 	/* Build Mport selector */
2008 	efx_rc = efx_mae_mport_by_pcie_mh_function(mport->emd_vnic.ev_intf,
2009 						mport->emd_vnic.ev_pf,
2010 						mport->emd_vnic.ev_vf,
2011 						&entity_selector);
2012 	if (efx_rc != 0) {
2013 		sfc_err(ctx->sa, "failed to build entity mport selector for c%upf%uvf%u",
2014 			mport->emd_vnic.ev_intf,
2015 			mport->emd_vnic.ev_pf,
2016 			mport->emd_vnic.ev_vf);
2017 		return efx_rc;
2018 	}
2019 
2020 	rc = sfc_mae_switch_port_id_by_entity(ctx->switch_domain_id,
2021 					      &entity_selector,
2022 					      SFC_MAE_SWITCH_PORT_REPRESENTOR,
2023 					      &switch_port_id);
2024 	switch (rc) {
2025 	case 0:
2026 		/* Already registered */
2027 		break;
2028 	case ENOENT:
2029 		/*
2030 		 * No representor has been created for this entity.
2031 		 * Create a dummy switch registry entry with an invalid ethdev
2032 		 * mport selector. When a corresponding representor is created,
2033 		 * this entry will be updated.
2034 		 */
2035 		req.type = SFC_MAE_SWITCH_PORT_REPRESENTOR;
2036 		req.entity_mportp = &entity_selector;
2037 		req.ethdev_mportp = &ethdev_mport;
2038 		req.ethdev_port_id = RTE_MAX_ETHPORTS;
2039 		req.port_data.repr.intf = mport->emd_vnic.ev_intf;
2040 		req.port_data.repr.pf = mport->emd_vnic.ev_pf;
2041 		req.port_data.repr.vf = mport->emd_vnic.ev_vf;
2042 
2043 		rc = sfc_mae_assign_switch_port(ctx->switch_domain_id,
2044 						&req, &switch_port_id);
2045 		if (rc != 0) {
2046 			sfc_err(ctx->sa,
2047 				"failed to assign MAE switch port for c%upf%uvf%u: %s",
2048 				mport->emd_vnic.ev_intf,
2049 				mport->emd_vnic.ev_pf,
2050 				mport->emd_vnic.ev_vf,
2051 				rte_strerror(rc));
2052 			return rc;
2053 		}
2054 		break;
2055 	default:
2056 		sfc_err(ctx->sa, "failed to find MAE switch port for c%upf%uvf%u: %s",
2057 			mport->emd_vnic.ev_intf,
2058 			mport->emd_vnic.ev_pf,
2059 			mport->emd_vnic.ev_vf,
2060 			rte_strerror(rc));
2061 		return rc;
2062 	}
2063 
2064 	return 0;
2065 }
2066 
2067 static efx_rc_t
2068 sfc_process_mport_journal_cb(void *data, efx_mport_desc_t *mport,
2069 			     size_t mport_len)
2070 {
2071 	struct sfc_mport_journal_ctx *ctx = data;
2072 
2073 	if (ctx == NULL || ctx->sa == NULL) {
2074 		SFC_GENERIC_LOG(ERR, "received NULL context or SFC adapter");
2075 		return EINVAL;
2076 	}
2077 
2078 	if (mport_len != sizeof(*mport)) {
2079 		sfc_err(ctx->sa, "actual and expected mport buffer sizes differ");
2080 		return EINVAL;
2081 	}
2082 
2083 	SFC_ASSERT(sfc_adapter_is_locked(ctx->sa));
2084 
2085 	/*
2086 	 * If a zombie flag is set, it means the mport has been marked for
2087 	 * deletion and cannot be used for any new operations. The mport will
2088 	 * be destroyed completely once all references to it are released.
2089 	 */
2090 	if (mport->emd_zombie) {
2091 		sfc_dbg(ctx->sa, "mport is a zombie, skipping");
2092 		return 0;
2093 	}
2094 	if (mport->emd_type != EFX_MPORT_TYPE_VNIC) {
2095 		sfc_dbg(ctx->sa, "mport is not a VNIC, skipping");
2096 		return 0;
2097 	}
2098 	if (mport->emd_vnic.ev_client_type != EFX_MPORT_VNIC_CLIENT_FUNCTION) {
2099 		sfc_dbg(ctx->sa, "mport is not a function, skipping");
2100 		return 0;
2101 	}
2102 	if (mport->emd_vnic.ev_handle == ctx->mcdi_handle) {
2103 		sfc_dbg(ctx->sa, "mport is this driver instance, skipping");
2104 		return 0;
2105 	}
2106 
2107 	return sfc_process_mport_journal_entry(ctx, mport);
2108 }
2109 
2110 static int
2111 sfc_process_mport_journal(struct sfc_adapter *sa)
2112 {
2113 	struct sfc_mport_journal_ctx ctx;
2114 	const efx_pcie_interface_t *controllers;
2115 	size_t nb_controllers;
2116 	efx_rc_t efx_rc;
2117 	int rc;
2118 
2119 	memset(&ctx, 0, sizeof(ctx));
2120 	ctx.sa = sa;
2121 	ctx.switch_domain_id = sa->mae.switch_domain_id;
2122 
2123 	efx_rc = efx_mcdi_get_own_client_handle(sa->nic, &ctx.mcdi_handle);
2124 	if (efx_rc != 0) {
2125 		sfc_err(sa, "failed to get own MCDI handle");
2126 		SFC_ASSERT(efx_rc > 0);
2127 		return efx_rc;
2128 	}
2129 
2130 	rc = sfc_mae_switch_domain_controllers(ctx.switch_domain_id,
2131 					       &controllers, &nb_controllers);
2132 	if (rc != 0) {
2133 		sfc_err(sa, "failed to get controller mapping");
2134 		return rc;
2135 	}
2136 
2137 	ctx.controllers_assigned = controllers != NULL;
2138 	ctx.controllers = NULL;
2139 	ctx.nb_controllers = 0;
2140 
2141 	efx_rc = efx_mae_read_mport_journal(sa->nic,
2142 					    sfc_process_mport_journal_cb, &ctx);
2143 	if (efx_rc != 0) {
2144 		sfc_err(sa, "failed to process MAE mport journal");
2145 		SFC_ASSERT(efx_rc > 0);
2146 		return efx_rc;
2147 	}
2148 
2149 	if (controllers == NULL) {
2150 		rc = sfc_mae_switch_domain_map_controllers(ctx.switch_domain_id,
2151 							   ctx.controllers,
2152 							   ctx.nb_controllers);
2153 		if (rc != 0)
2154 			return rc;
2155 	}
2156 
2157 	return 0;
2158 }
2159 
2160 static void
2161 sfc_count_representors_cb(enum sfc_mae_switch_port_type type,
2162 			  const efx_mport_sel_t *ethdev_mportp __rte_unused,
2163 			  uint16_t ethdev_port_id __rte_unused,
2164 			  const efx_mport_sel_t *entity_mportp __rte_unused,
2165 			  uint16_t switch_port_id __rte_unused,
2166 			  union sfc_mae_switch_port_data *port_datap
2167 				__rte_unused,
2168 			  void *user_datap)
2169 {
2170 	int *counter = user_datap;
2171 
2172 	SFC_ASSERT(counter != NULL);
2173 
2174 	if (type == SFC_MAE_SWITCH_PORT_REPRESENTOR)
2175 		(*counter)++;
2176 }
2177 
2178 struct sfc_get_representors_ctx {
2179 	struct rte_eth_representor_info	*info;
2180 	struct sfc_adapter		*sa;
2181 	uint16_t			switch_domain_id;
2182 	const efx_pcie_interface_t	*controllers;
2183 	size_t				nb_controllers;
2184 };
2185 
2186 static void
2187 sfc_get_representors_cb(enum sfc_mae_switch_port_type type,
2188 			const efx_mport_sel_t *ethdev_mportp __rte_unused,
2189 			uint16_t ethdev_port_id __rte_unused,
2190 			const efx_mport_sel_t *entity_mportp __rte_unused,
2191 			uint16_t switch_port_id,
2192 			union sfc_mae_switch_port_data *port_datap,
2193 			void *user_datap)
2194 {
2195 	struct sfc_get_representors_ctx *ctx = user_datap;
2196 	struct rte_eth_representor_range *range;
2197 	int ret;
2198 	int rc;
2199 
2200 	SFC_ASSERT(ctx != NULL);
2201 	SFC_ASSERT(ctx->info != NULL);
2202 	SFC_ASSERT(ctx->sa != NULL);
2203 
2204 	if (type != SFC_MAE_SWITCH_PORT_REPRESENTOR) {
2205 		sfc_dbg(ctx->sa, "not a representor, skipping");
2206 		return;
2207 	}
2208 	if (ctx->info->nb_ranges >= ctx->info->nb_ranges_alloc) {
2209 		sfc_dbg(ctx->sa, "info structure is full already");
2210 		return;
2211 	}
2212 
2213 	range = &ctx->info->ranges[ctx->info->nb_ranges];
2214 	rc = sfc_mae_switch_controller_from_mapping(ctx->controllers,
2215 						    ctx->nb_controllers,
2216 						    port_datap->repr.intf,
2217 						    &range->controller);
2218 	if (rc != 0) {
2219 		sfc_err(ctx->sa, "invalid representor controller: %d",
2220 			port_datap->repr.intf);
2221 		range->controller = -1;
2222 	}
2223 	range->pf = port_datap->repr.pf;
2224 	range->id_base = switch_port_id;
2225 	range->id_end = switch_port_id;
2226 
2227 	if (port_datap->repr.vf != EFX_PCI_VF_INVALID) {
2228 		range->type = RTE_ETH_REPRESENTOR_VF;
2229 		range->vf = port_datap->repr.vf;
2230 		ret = snprintf(range->name, RTE_DEV_NAME_MAX_LEN,
2231 			       "c%dpf%dvf%d", range->controller, range->pf,
2232 			       range->vf);
2233 	} else {
2234 		range->type = RTE_ETH_REPRESENTOR_PF;
2235 		ret = snprintf(range->name, RTE_DEV_NAME_MAX_LEN,
2236 			 "c%dpf%d", range->controller, range->pf);
2237 	}
2238 	if (ret >= RTE_DEV_NAME_MAX_LEN) {
2239 		sfc_err(ctx->sa, "representor name has been truncated: %s",
2240 			range->name);
2241 	}
2242 
2243 	ctx->info->nb_ranges++;
2244 }
2245 
2246 static int
2247 sfc_representor_info_get(struct rte_eth_dev *dev,
2248 			 struct rte_eth_representor_info *info)
2249 {
2250 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2251 	struct sfc_get_representors_ctx get_repr_ctx;
2252 	const efx_nic_cfg_t *nic_cfg;
2253 	uint16_t switch_domain_id;
2254 	uint32_t nb_repr;
2255 	int controller;
2256 	int rc;
2257 
2258 	sfc_adapter_lock(sa);
2259 
2260 	if (sa->mae.status != SFC_MAE_STATUS_ADMIN) {
2261 		sfc_adapter_unlock(sa);
2262 		return -ENOTSUP;
2263 	}
2264 
2265 	rc = sfc_process_mport_journal(sa);
2266 	if (rc != 0) {
2267 		sfc_adapter_unlock(sa);
2268 		SFC_ASSERT(rc > 0);
2269 		return -rc;
2270 	}
2271 
2272 	switch_domain_id = sa->mae.switch_domain_id;
2273 
2274 	nb_repr = 0;
2275 	rc = sfc_mae_switch_ports_iterate(switch_domain_id,
2276 					  sfc_count_representors_cb,
2277 					  &nb_repr);
2278 	if (rc != 0) {
2279 		sfc_adapter_unlock(sa);
2280 		SFC_ASSERT(rc > 0);
2281 		return -rc;
2282 	}
2283 
2284 	if (info == NULL) {
2285 		sfc_adapter_unlock(sa);
2286 		return nb_repr;
2287 	}
2288 
2289 	rc = sfc_mae_switch_domain_controllers(switch_domain_id,
2290 					       &get_repr_ctx.controllers,
2291 					       &get_repr_ctx.nb_controllers);
2292 	if (rc != 0) {
2293 		sfc_adapter_unlock(sa);
2294 		SFC_ASSERT(rc > 0);
2295 		return -rc;
2296 	}
2297 
2298 	nic_cfg = efx_nic_cfg_get(sa->nic);
2299 
2300 	rc = sfc_mae_switch_domain_get_controller(switch_domain_id,
2301 						  nic_cfg->enc_intf,
2302 						  &controller);
2303 	if (rc != 0) {
2304 		sfc_err(sa, "invalid controller: %d", nic_cfg->enc_intf);
2305 		controller = -1;
2306 	}
2307 
2308 	info->controller = controller;
2309 	info->pf = nic_cfg->enc_pf;
2310 
2311 	get_repr_ctx.info = info;
2312 	get_repr_ctx.sa = sa;
2313 	get_repr_ctx.switch_domain_id = switch_domain_id;
2314 	rc = sfc_mae_switch_ports_iterate(switch_domain_id,
2315 					  sfc_get_representors_cb,
2316 					  &get_repr_ctx);
2317 	if (rc != 0) {
2318 		sfc_adapter_unlock(sa);
2319 		SFC_ASSERT(rc > 0);
2320 		return -rc;
2321 	}
2322 
2323 	sfc_adapter_unlock(sa);
2324 	return nb_repr;
2325 }
2326 
2327 static int
2328 sfc_rx_metadata_negotiate(struct rte_eth_dev *dev, uint64_t *features)
2329 {
2330 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2331 	uint64_t supported = 0;
2332 
2333 	sfc_adapter_lock(sa);
2334 
2335 	if ((sa->priv.dp_rx->features & SFC_DP_RX_FEAT_FLOW_FLAG) != 0)
2336 		supported |= RTE_ETH_RX_METADATA_USER_FLAG;
2337 
2338 	if ((sa->priv.dp_rx->features & SFC_DP_RX_FEAT_FLOW_MARK) != 0)
2339 		supported |= RTE_ETH_RX_METADATA_USER_MARK;
2340 
2341 	if (sfc_ft_is_supported(sa))
2342 		supported |= RTE_ETH_RX_METADATA_TUNNEL_ID;
2343 
2344 	sa->negotiated_rx_metadata = supported & *features;
2345 	*features = sa->negotiated_rx_metadata;
2346 
2347 	sfc_adapter_unlock(sa);
2348 
2349 	return 0;
2350 }
2351 
2352 static unsigned int
2353 sfc_fec_get_capa_speed_to_fec(uint32_t supported_caps,
2354 			      struct rte_eth_fec_capa *speed_fec_capa)
2355 {
2356 	unsigned int num = 0;
2357 	bool baser = false;
2358 	bool rs = false;
2359 
2360 	if (supported_caps & EFX_PHY_CAP_FEC_BIT(BASER_FEC))
2361 		baser = true;
2362 	if (supported_caps & EFX_PHY_CAP_FEC_BIT(RS_FEC))
2363 		rs = true;
2364 
2365 	/*
2366 	 * NOFEC and AUTO FEC modes are always supported.
2367 	 * FW does not provide information about the supported
2368 	 * FEC modes per the link speed.
2369 	 * Supported FEC depends on supported link speeds and
2370 	 * supported FEC modes by a device.
2371 	 */
2372 	if (supported_caps & (1u << EFX_PHY_CAP_10000FDX)) {
2373 		if (speed_fec_capa != NULL) {
2374 			speed_fec_capa[num].speed = RTE_ETH_SPEED_NUM_10G;
2375 			speed_fec_capa[num].capa =
2376 				RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC) |
2377 				RTE_ETH_FEC_MODE_CAPA_MASK(AUTO);
2378 			if (baser) {
2379 				speed_fec_capa[num].capa |=
2380 					RTE_ETH_FEC_MODE_CAPA_MASK(BASER);
2381 			}
2382 		}
2383 		num++;
2384 	}
2385 	if (supported_caps & (1u << EFX_PHY_CAP_25000FDX)) {
2386 		if (speed_fec_capa != NULL) {
2387 			speed_fec_capa[num].speed = RTE_ETH_SPEED_NUM_25G;
2388 			speed_fec_capa[num].capa =
2389 				RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC) |
2390 				RTE_ETH_FEC_MODE_CAPA_MASK(AUTO);
2391 			if (baser) {
2392 				speed_fec_capa[num].capa |=
2393 					RTE_ETH_FEC_MODE_CAPA_MASK(BASER);
2394 			}
2395 			if (rs) {
2396 				speed_fec_capa[num].capa |=
2397 					RTE_ETH_FEC_MODE_CAPA_MASK(RS);
2398 			}
2399 		}
2400 		num++;
2401 	}
2402 	if (supported_caps & (1u << EFX_PHY_CAP_40000FDX)) {
2403 		if (speed_fec_capa != NULL) {
2404 			speed_fec_capa[num].speed = RTE_ETH_SPEED_NUM_40G;
2405 			speed_fec_capa[num].capa =
2406 				RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC) |
2407 				RTE_ETH_FEC_MODE_CAPA_MASK(AUTO);
2408 			if (baser) {
2409 				speed_fec_capa[num].capa |=
2410 					RTE_ETH_FEC_MODE_CAPA_MASK(BASER);
2411 			}
2412 		}
2413 		num++;
2414 	}
2415 	if (supported_caps & (1u << EFX_PHY_CAP_50000FDX)) {
2416 		if (speed_fec_capa != NULL) {
2417 			speed_fec_capa[num].speed = RTE_ETH_SPEED_NUM_50G;
2418 			speed_fec_capa[num].capa =
2419 				RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC) |
2420 				RTE_ETH_FEC_MODE_CAPA_MASK(AUTO);
2421 			if (baser) {
2422 				speed_fec_capa[num].capa |=
2423 					RTE_ETH_FEC_MODE_CAPA_MASK(BASER);
2424 			}
2425 			if (rs) {
2426 				speed_fec_capa[num].capa |=
2427 					RTE_ETH_FEC_MODE_CAPA_MASK(RS);
2428 			}
2429 		}
2430 		num++;
2431 	}
2432 	if (supported_caps & (1u << EFX_PHY_CAP_100000FDX)) {
2433 		if (speed_fec_capa != NULL) {
2434 			speed_fec_capa[num].speed = RTE_ETH_SPEED_NUM_100G;
2435 			speed_fec_capa[num].capa =
2436 				RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC) |
2437 				RTE_ETH_FEC_MODE_CAPA_MASK(AUTO);
2438 			if (rs) {
2439 				speed_fec_capa[num].capa |=
2440 					RTE_ETH_FEC_MODE_CAPA_MASK(RS);
2441 			}
2442 		}
2443 		num++;
2444 	}
2445 
2446 	return num;
2447 }
2448 
2449 static int
2450 sfc_fec_get_capability(struct rte_eth_dev *dev,
2451 		       struct rte_eth_fec_capa *speed_fec_capa,
2452 		       unsigned int num)
2453 {
2454 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2455 	unsigned int num_entries;
2456 	uint32_t supported_caps;
2457 
2458 	sfc_adapter_lock(sa);
2459 
2460 	efx_phy_adv_cap_get(sa->nic, EFX_PHY_CAP_PERM, &supported_caps);
2461 
2462 	num_entries = sfc_fec_get_capa_speed_to_fec(supported_caps, NULL);
2463 	if (speed_fec_capa == NULL || num < num_entries)
2464 		goto adapter_unlock;
2465 
2466 	num_entries = sfc_fec_get_capa_speed_to_fec(supported_caps,
2467 						    speed_fec_capa);
2468 
2469 adapter_unlock:
2470 	sfc_adapter_unlock(sa);
2471 
2472 	return num_entries;
2473 }
2474 
2475 static uint32_t
2476 sfc_efx_caps_to_fec(uint32_t caps, bool is_25g)
2477 {
2478 	bool rs_req = caps & EFX_PHY_CAP_FEC_BIT(RS_FEC_REQUESTED);
2479 	bool rs = caps & EFX_PHY_CAP_FEC_BIT(RS_FEC);
2480 	bool baser_req;
2481 	bool baser;
2482 
2483 	if (is_25g) {
2484 		baser = caps & EFX_PHY_CAP_FEC_BIT(25G_BASER_FEC);
2485 		baser_req = caps & EFX_PHY_CAP_FEC_BIT(25G_BASER_FEC_REQUESTED);
2486 	} else {
2487 		baser = caps & EFX_PHY_CAP_FEC_BIT(BASER_FEC);
2488 		baser_req = caps & EFX_PHY_CAP_FEC_BIT(BASER_FEC_REQUESTED);
2489 	}
2490 
2491 	if (!baser && !rs)
2492 		return RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_NOFEC);
2493 
2494 	if (rs_req)
2495 		return RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_RS);
2496 
2497 	if (baser_req)
2498 		return RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_BASER);
2499 
2500 	return 0;
2501 }
2502 
2503 static int
2504 sfc_fec_get(struct rte_eth_dev *dev, uint32_t *fec_capa)
2505 {
2506 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2507 	struct sfc_port *port = &sa->port;
2508 	struct rte_eth_link current_link;
2509 	efx_phy_fec_type_t active_fec;
2510 	bool is_25g = false;
2511 	int rc = 0;
2512 
2513 	sfc_adapter_lock(sa);
2514 
2515 	sfc_dev_get_rte_link(dev, 1, &current_link);
2516 
2517 	if (current_link.link_status == RTE_ETH_LINK_DOWN) {
2518 		uint32_t speed = current_link.link_speed;
2519 
2520 		if (port->fec_auto) {
2521 			*fec_capa = RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_AUTO);
2522 			goto adapter_unlock;
2523 		}
2524 
2525 		is_25g = (speed == RTE_ETH_SPEED_NUM_25G ||
2526 			  speed == RTE_ETH_SPEED_NUM_50G);
2527 
2528 		*fec_capa = sfc_efx_caps_to_fec(port->fec_cfg, is_25g);
2529 		if (*fec_capa == 0)
2530 			rc = ENOTSUP;
2531 
2532 		goto adapter_unlock;
2533 	}
2534 
2535 	rc = efx_phy_fec_type_get(sa->nic, &active_fec);
2536 	if (rc != 0)
2537 		goto adapter_unlock;
2538 
2539 	switch (active_fec) {
2540 	case EFX_PHY_FEC_NONE:
2541 		*fec_capa = RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_NOFEC);
2542 		break;
2543 	case EFX_PHY_FEC_BASER:
2544 		*fec_capa = RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_BASER);
2545 		break;
2546 	case EFX_PHY_FEC_RS:
2547 		*fec_capa = RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_RS);
2548 		break;
2549 	default:
2550 		rc = ENOTSUP;
2551 		break;
2552 	}
2553 
2554 adapter_unlock:
2555 	sfc_adapter_unlock(sa);
2556 
2557 	if (rc != 0)
2558 		sfc_err(sa, "failed to get FEC mode");
2559 
2560 	SFC_ASSERT(rc >= 0);
2561 	return -rc;
2562 }
2563 
2564 static int
2565 sfc_fec_capa_check(struct rte_eth_dev *dev, uint32_t fec_capa,
2566 		   uint32_t supported_caps)
2567 {
2568 	struct rte_eth_fec_capa *speed_fec_capa;
2569 	struct rte_eth_link current_link;
2570 	bool is_supported = false;
2571 	unsigned int num_entries;
2572 	bool auto_fec = false;
2573 	unsigned int i;
2574 
2575 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2576 
2577 	if (sa->state != SFC_ETHDEV_STARTED)
2578 		return 0;
2579 
2580 	if (fec_capa & RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_AUTO)) {
2581 		auto_fec = true;
2582 		fec_capa &= ~RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_AUTO);
2583 	}
2584 
2585 	/*
2586 	 * If only the AUTO bit is set, the decision on which FEC
2587 	 * mode to use will be made by HW/FW or driver.
2588 	 */
2589 	if (auto_fec && fec_capa == 0)
2590 		return 0;
2591 
2592 	sfc_dev_get_rte_link(dev, 1, &current_link);
2593 
2594 	num_entries = sfc_fec_get_capa_speed_to_fec(supported_caps, NULL);
2595 	if (num_entries == 0)
2596 		return ENOTSUP;
2597 
2598 	speed_fec_capa = rte_calloc("fec_capa", num_entries,
2599 				    sizeof(*speed_fec_capa), 0);
2600 	num_entries = sfc_fec_get_capa_speed_to_fec(supported_caps,
2601 						    speed_fec_capa);
2602 
2603 	for (i = 0; i < num_entries; i++) {
2604 		if (speed_fec_capa[i].speed == current_link.link_speed) {
2605 			if ((fec_capa & speed_fec_capa[i].capa) != 0)
2606 				is_supported = true;
2607 
2608 			break;
2609 		}
2610 	}
2611 
2612 	rte_free(speed_fec_capa);
2613 
2614 	if (is_supported)
2615 		return 0;
2616 
2617 	return ENOTSUP;
2618 }
2619 
2620 static int
2621 sfc_fec_capa_to_efx(uint32_t supported_caps, uint32_t fec_capa,
2622 		    uint32_t *efx_fec_caps)
2623 {
2624 	bool fec_is_set = false;
2625 	bool auto_fec = false;
2626 	bool nofec = false;
2627 	uint32_t ret = 0;
2628 
2629 	if (efx_fec_caps == NULL)
2630 		return EINVAL;
2631 
2632 	if (fec_capa & RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_AUTO))
2633 		auto_fec = true;
2634 
2635 	if (fec_capa & RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_NOFEC))
2636 		nofec = true;
2637 
2638 	if (fec_capa == RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_AUTO)) {
2639 		ret |= (EFX_PHY_CAP_FEC_BIT(BASER_FEC) |
2640 			EFX_PHY_CAP_FEC_BIT(25G_BASER_FEC) |
2641 			EFX_PHY_CAP_FEC_BIT(RS_FEC)) & supported_caps;
2642 		goto done;
2643 	}
2644 
2645 	if (fec_capa & RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_RS)) {
2646 		fec_is_set = true;
2647 
2648 		if (supported_caps & EFX_PHY_CAP_FEC_BIT(RS_FEC)) {
2649 			ret |= EFX_PHY_CAP_FEC_BIT(RS_FEC) |
2650 			       EFX_PHY_CAP_FEC_BIT(RS_FEC_REQUESTED);
2651 		}
2652 	}
2653 	if (fec_capa & RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_BASER)) {
2654 		if (!auto_fec && fec_is_set)
2655 			return EINVAL;
2656 
2657 		if (supported_caps & EFX_PHY_CAP_FEC_BIT(BASER_FEC)) {
2658 			ret |= EFX_PHY_CAP_FEC_BIT(BASER_FEC) |
2659 			       EFX_PHY_CAP_FEC_BIT(BASER_FEC_REQUESTED);
2660 		}
2661 		if (supported_caps & EFX_PHY_CAP_FEC_BIT(25G_BASER_FEC)) {
2662 			ret |= EFX_PHY_CAP_FEC_BIT(25G_BASER_FEC) |
2663 			       EFX_PHY_CAP_FEC_BIT(25G_BASER_FEC_REQUESTED);
2664 		}
2665 	}
2666 
2667 	if (ret == 0 && !nofec)
2668 		return ENOTSUP;
2669 
2670 done:
2671 	*efx_fec_caps = ret;
2672 	return 0;
2673 }
2674 
2675 static int
2676 sfc_fec_set(struct rte_eth_dev *dev, uint32_t fec_capa)
2677 {
2678 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2679 	struct sfc_port *port = &sa->port;
2680 	uint32_t supported_caps;
2681 	uint32_t efx_fec_caps;
2682 	uint32_t updated_caps;
2683 	int rc = 0;
2684 
2685 	sfc_adapter_lock(sa);
2686 
2687 	efx_phy_adv_cap_get(sa->nic, EFX_PHY_CAP_PERM, &supported_caps);
2688 
2689 	rc = sfc_fec_capa_check(dev, fec_capa, supported_caps);
2690 	if (rc != 0)
2691 		goto adapter_unlock;
2692 
2693 	rc = sfc_fec_capa_to_efx(supported_caps, fec_capa, &efx_fec_caps);
2694 	if (rc != 0)
2695 		goto adapter_unlock;
2696 
2697 	if (sa->state == SFC_ETHDEV_STARTED) {
2698 		efx_phy_adv_cap_get(sa->nic, EFX_PHY_CAP_CURRENT,
2699 				    &updated_caps);
2700 		updated_caps = updated_caps & ~EFX_PHY_CAP_FEC_MASK;
2701 		updated_caps |= efx_fec_caps;
2702 
2703 		rc = efx_phy_adv_cap_set(sa->nic, updated_caps);
2704 		if (rc != 0)
2705 			goto adapter_unlock;
2706 	}
2707 
2708 	port->fec_cfg = efx_fec_caps;
2709 	/*
2710 	 * There is no chance to recognize AUTO mode from the
2711 	 * saved FEC capabilities as AUTO mode can have the same
2712 	 * set of bits as any other mode from the EFX point of view.
2713 	 * Save it in the proper variable.
2714 	 */
2715 	if (fec_capa & RTE_ETH_FEC_MODE_TO_CAPA(RTE_ETH_FEC_AUTO))
2716 		port->fec_auto = true;
2717 	else
2718 		port->fec_auto = false;
2719 
2720 adapter_unlock:
2721 	sfc_adapter_unlock(sa);
2722 
2723 	SFC_ASSERT(rc >= 0);
2724 	return -rc;
2725 }
2726 
2727 static const struct eth_dev_ops sfc_eth_dev_ops = {
2728 	.dev_configure			= sfc_dev_configure,
2729 	.dev_start			= sfc_dev_start,
2730 	.dev_stop			= sfc_dev_stop,
2731 	.dev_set_link_up		= sfc_dev_set_link_up,
2732 	.dev_set_link_down		= sfc_dev_set_link_down,
2733 	.dev_close			= sfc_dev_close,
2734 	.promiscuous_enable		= sfc_dev_promisc_enable,
2735 	.promiscuous_disable		= sfc_dev_promisc_disable,
2736 	.allmulticast_enable		= sfc_dev_allmulti_enable,
2737 	.allmulticast_disable		= sfc_dev_allmulti_disable,
2738 	.link_update			= sfc_dev_link_update,
2739 	.stats_get			= sfc_stats_get,
2740 	.stats_reset			= sfc_stats_reset,
2741 	.xstats_get			= sfc_xstats_get,
2742 	.xstats_reset			= sfc_stats_reset,
2743 	.xstats_get_names		= sfc_xstats_get_names,
2744 	.dev_infos_get			= sfc_dev_infos_get,
2745 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
2746 	.mtu_set			= sfc_dev_set_mtu,
2747 	.rx_queue_start			= sfc_rx_queue_start,
2748 	.rx_queue_stop			= sfc_rx_queue_stop,
2749 	.tx_queue_start			= sfc_tx_queue_start,
2750 	.tx_queue_stop			= sfc_tx_queue_stop,
2751 	.rx_queue_setup			= sfc_rx_queue_setup,
2752 	.rx_queue_release		= sfc_rx_queue_release,
2753 	.rx_queue_intr_enable		= sfc_rx_queue_intr_enable,
2754 	.rx_queue_intr_disable		= sfc_rx_queue_intr_disable,
2755 	.tx_queue_setup			= sfc_tx_queue_setup,
2756 	.tx_queue_release		= sfc_tx_queue_release,
2757 	.flow_ctrl_get			= sfc_flow_ctrl_get,
2758 	.flow_ctrl_set			= sfc_flow_ctrl_set,
2759 	.mac_addr_set			= sfc_mac_addr_set,
2760 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
2761 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
2762 	.reta_update			= sfc_dev_rss_reta_update,
2763 	.reta_query			= sfc_dev_rss_reta_query,
2764 	.rss_hash_update		= sfc_dev_rss_hash_update,
2765 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
2766 	.flow_ops_get			= sfc_dev_flow_ops_get,
2767 	.set_mc_addr_list		= sfc_set_mc_addr_list,
2768 	.rxq_info_get			= sfc_rx_queue_info_get,
2769 	.txq_info_get			= sfc_tx_queue_info_get,
2770 	.fw_version_get			= sfc_fw_version_get,
2771 	.xstats_get_by_id		= sfc_xstats_get_by_id,
2772 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
2773 	.pool_ops_supported		= sfc_pool_ops_supported,
2774 	.representor_info_get		= sfc_representor_info_get,
2775 	.rx_metadata_negotiate		= sfc_rx_metadata_negotiate,
2776 	.fec_get_capability		= sfc_fec_get_capability,
2777 	.fec_get			= sfc_fec_get,
2778 	.fec_set			= sfc_fec_set,
2779 };
2780 
2781 struct sfc_ethdev_init_data {
2782 	uint16_t		nb_representors;
2783 };
2784 
2785 /**
2786  * Duplicate a string in potentially shared memory required for
2787  * multi-process support.
2788  *
2789  * strdup() allocates from process-local heap/memory.
2790  */
2791 static char *
2792 sfc_strdup(const char *str)
2793 {
2794 	size_t size;
2795 	char *copy;
2796 
2797 	if (str == NULL)
2798 		return NULL;
2799 
2800 	size = strlen(str) + 1;
2801 	copy = rte_malloc(__func__, size, 0);
2802 	if (copy != NULL)
2803 		rte_memcpy(copy, str, size);
2804 
2805 	return copy;
2806 }
2807 
2808 static int
2809 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
2810 {
2811 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2812 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2813 	const struct sfc_dp_rx *dp_rx;
2814 	const struct sfc_dp_tx *dp_tx;
2815 	const efx_nic_cfg_t *encp;
2816 	unsigned int avail_caps = 0;
2817 	const char *rx_name = NULL;
2818 	const char *tx_name = NULL;
2819 	int rc;
2820 
2821 	switch (sa->family) {
2822 	case EFX_FAMILY_HUNTINGTON:
2823 	case EFX_FAMILY_MEDFORD:
2824 	case EFX_FAMILY_MEDFORD2:
2825 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
2826 		avail_caps |= SFC_DP_HW_FW_CAP_RX_EFX;
2827 		avail_caps |= SFC_DP_HW_FW_CAP_TX_EFX;
2828 		break;
2829 	case EFX_FAMILY_RIVERHEAD:
2830 		avail_caps |= SFC_DP_HW_FW_CAP_EF100;
2831 		break;
2832 	default:
2833 		break;
2834 	}
2835 
2836 	encp = efx_nic_cfg_get(sa->nic);
2837 	if (encp->enc_rx_es_super_buffer_supported)
2838 		avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
2839 
2840 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
2841 				sfc_kvarg_string_handler, &rx_name);
2842 	if (rc != 0)
2843 		goto fail_kvarg_rx_datapath;
2844 
2845 	if (rx_name != NULL) {
2846 		dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
2847 		if (dp_rx == NULL) {
2848 			sfc_err(sa, "Rx datapath %s not found", rx_name);
2849 			rc = ENOENT;
2850 			goto fail_dp_rx;
2851 		}
2852 		if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
2853 			sfc_err(sa,
2854 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
2855 				rx_name);
2856 			rc = EINVAL;
2857 			goto fail_dp_rx_caps;
2858 		}
2859 	} else {
2860 		dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
2861 		if (dp_rx == NULL) {
2862 			sfc_err(sa, "Rx datapath by caps %#x not found",
2863 				avail_caps);
2864 			rc = ENOENT;
2865 			goto fail_dp_rx;
2866 		}
2867 	}
2868 
2869 	sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
2870 	if (sas->dp_rx_name == NULL) {
2871 		rc = ENOMEM;
2872 		goto fail_dp_rx_name;
2873 	}
2874 
2875 	if (strcmp(dp_rx->dp.name, SFC_KVARG_DATAPATH_EF10_ESSB) == 0) {
2876 		/* FLAG and MARK are always available from Rx prefix. */
2877 		sa->negotiated_rx_metadata |= RTE_ETH_RX_METADATA_USER_FLAG;
2878 		sa->negotiated_rx_metadata |= RTE_ETH_RX_METADATA_USER_MARK;
2879 	}
2880 
2881 	sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
2882 
2883 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
2884 				sfc_kvarg_string_handler, &tx_name);
2885 	if (rc != 0)
2886 		goto fail_kvarg_tx_datapath;
2887 
2888 	if (tx_name != NULL) {
2889 		dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
2890 		if (dp_tx == NULL) {
2891 			sfc_err(sa, "Tx datapath %s not found", tx_name);
2892 			rc = ENOENT;
2893 			goto fail_dp_tx;
2894 		}
2895 		if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
2896 			sfc_err(sa,
2897 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
2898 				tx_name);
2899 			rc = EINVAL;
2900 			goto fail_dp_tx_caps;
2901 		}
2902 	} else {
2903 		dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
2904 		if (dp_tx == NULL) {
2905 			sfc_err(sa, "Tx datapath by caps %#x not found",
2906 				avail_caps);
2907 			rc = ENOENT;
2908 			goto fail_dp_tx;
2909 		}
2910 	}
2911 
2912 	sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
2913 	if (sas->dp_tx_name == NULL) {
2914 		rc = ENOMEM;
2915 		goto fail_dp_tx_name;
2916 	}
2917 
2918 	sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
2919 
2920 	sa->priv.dp_rx = dp_rx;
2921 	sa->priv.dp_tx = dp_tx;
2922 
2923 	dev->rx_pkt_burst = dp_rx->pkt_burst;
2924 	dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2925 	dev->tx_pkt_burst = dp_tx->pkt_burst;
2926 
2927 	dev->rx_queue_count = sfc_rx_queue_count;
2928 	dev->rx_descriptor_status = sfc_rx_descriptor_status;
2929 	dev->tx_descriptor_status = sfc_tx_descriptor_status;
2930 	dev->dev_ops = &sfc_eth_dev_ops;
2931 
2932 	return 0;
2933 
2934 fail_dp_tx_name:
2935 fail_dp_tx_caps:
2936 fail_dp_tx:
2937 fail_kvarg_tx_datapath:
2938 	rte_free(sas->dp_rx_name);
2939 	sas->dp_rx_name = NULL;
2940 
2941 fail_dp_rx_name:
2942 fail_dp_rx_caps:
2943 fail_dp_rx:
2944 fail_kvarg_rx_datapath:
2945 	return rc;
2946 }
2947 
2948 static void
2949 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
2950 {
2951 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2952 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2953 
2954 	dev->dev_ops = NULL;
2955 	dev->tx_pkt_prepare = NULL;
2956 	dev->rx_pkt_burst = NULL;
2957 	dev->tx_pkt_burst = NULL;
2958 
2959 	rte_free(sas->dp_tx_name);
2960 	sas->dp_tx_name = NULL;
2961 	sa->priv.dp_tx = NULL;
2962 
2963 	rte_free(sas->dp_rx_name);
2964 	sas->dp_rx_name = NULL;
2965 	sa->priv.dp_rx = NULL;
2966 }
2967 
2968 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2969 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
2970 	.reta_query			= sfc_dev_rss_reta_query,
2971 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
2972 	.rxq_info_get			= sfc_rx_queue_info_get,
2973 	.txq_info_get			= sfc_tx_queue_info_get,
2974 };
2975 
2976 static int
2977 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2978 {
2979 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2980 	struct sfc_adapter_priv *sap;
2981 	const struct sfc_dp_rx *dp_rx;
2982 	const struct sfc_dp_tx *dp_tx;
2983 	int rc;
2984 
2985 	/*
2986 	 * Allocate process private data from heap, since it should not
2987 	 * be located in shared memory allocated using rte_malloc() API.
2988 	 */
2989 	sap = calloc(1, sizeof(*sap));
2990 	if (sap == NULL) {
2991 		rc = ENOMEM;
2992 		goto fail_alloc_priv;
2993 	}
2994 
2995 	sap->logtype_main = logtype_main;
2996 
2997 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2998 	if (dp_rx == NULL) {
2999 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
3000 			"cannot find %s Rx datapath", sas->dp_rx_name);
3001 		rc = ENOENT;
3002 		goto fail_dp_rx;
3003 	}
3004 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
3005 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
3006 			"%s Rx datapath does not support multi-process",
3007 			sas->dp_rx_name);
3008 		rc = EINVAL;
3009 		goto fail_dp_rx_multi_process;
3010 	}
3011 
3012 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
3013 	if (dp_tx == NULL) {
3014 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
3015 			"cannot find %s Tx datapath", sas->dp_tx_name);
3016 		rc = ENOENT;
3017 		goto fail_dp_tx;
3018 	}
3019 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
3020 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
3021 			"%s Tx datapath does not support multi-process",
3022 			sas->dp_tx_name);
3023 		rc = EINVAL;
3024 		goto fail_dp_tx_multi_process;
3025 	}
3026 
3027 	sap->dp_rx = dp_rx;
3028 	sap->dp_tx = dp_tx;
3029 
3030 	dev->process_private = sap;
3031 	dev->rx_pkt_burst = dp_rx->pkt_burst;
3032 	dev->tx_pkt_prepare = dp_tx->pkt_prepare;
3033 	dev->tx_pkt_burst = dp_tx->pkt_burst;
3034 	dev->rx_queue_count = sfc_rx_queue_count;
3035 	dev->rx_descriptor_status = sfc_rx_descriptor_status;
3036 	dev->tx_descriptor_status = sfc_tx_descriptor_status;
3037 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
3038 
3039 	return 0;
3040 
3041 fail_dp_tx_multi_process:
3042 fail_dp_tx:
3043 fail_dp_rx_multi_process:
3044 fail_dp_rx:
3045 	free(sap);
3046 
3047 fail_alloc_priv:
3048 	return rc;
3049 }
3050 
3051 static void
3052 sfc_register_dp(void)
3053 {
3054 	/* Register once */
3055 	if (TAILQ_EMPTY(&sfc_dp_head)) {
3056 		/* Prefer EF10 datapath */
3057 		sfc_dp_register(&sfc_dp_head, &sfc_ef100_rx.dp);
3058 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
3059 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
3060 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
3061 
3062 		sfc_dp_register(&sfc_dp_head, &sfc_ef100_tx.dp);
3063 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
3064 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
3065 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
3066 	}
3067 }
3068 
3069 static int
3070 sfc_parse_switch_mode(struct sfc_adapter *sa, bool has_representors)
3071 {
3072 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
3073 	const char *switch_mode = NULL;
3074 	int rc;
3075 
3076 	sfc_log_init(sa, "entry");
3077 
3078 	rc = sfc_kvargs_process(sa, SFC_KVARG_SWITCH_MODE,
3079 				sfc_kvarg_string_handler, &switch_mode);
3080 	if (rc != 0)
3081 		goto fail_kvargs;
3082 
3083 	if (switch_mode == NULL) {
3084 		sa->switchdev = encp->enc_mae_admin &&
3085 				(!encp->enc_datapath_cap_evb ||
3086 				 has_representors);
3087 	} else if (strcasecmp(switch_mode, SFC_KVARG_SWITCH_MODE_LEGACY) == 0) {
3088 		sa->switchdev = false;
3089 	} else if (strcasecmp(switch_mode,
3090 			      SFC_KVARG_SWITCH_MODE_SWITCHDEV) == 0) {
3091 		sa->switchdev = true;
3092 	} else {
3093 		sfc_err(sa, "invalid switch mode device argument '%s'",
3094 			switch_mode);
3095 		rc = EINVAL;
3096 		goto fail_mode;
3097 	}
3098 
3099 	sfc_log_init(sa, "done");
3100 
3101 	return 0;
3102 
3103 fail_mode:
3104 fail_kvargs:
3105 	sfc_log_init(sa, "failed: %s", rte_strerror(rc));
3106 
3107 	return rc;
3108 }
3109 
3110 static int
3111 sfc_eth_dev_init(struct rte_eth_dev *dev, void *init_params)
3112 {
3113 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
3114 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
3115 	struct sfc_ethdev_init_data *init_data = init_params;
3116 	uint32_t logtype_main;
3117 	struct sfc_adapter *sa;
3118 	int rc;
3119 	const efx_nic_cfg_t *encp;
3120 	const struct rte_ether_addr *from;
3121 	int ret;
3122 
3123 	if (sfc_efx_dev_class_get(pci_dev->device.devargs) !=
3124 			SFC_EFX_DEV_CLASS_NET) {
3125 		SFC_GENERIC_LOG(DEBUG,
3126 			"Incompatible device class: skip probing, should be probed by other sfc driver.");
3127 		return 1;
3128 	}
3129 
3130 	rc = sfc_dp_mport_register();
3131 	if (rc != 0)
3132 		return rc;
3133 
3134 	sfc_register_dp();
3135 
3136 	logtype_main = sfc_register_logtype(&pci_dev->addr,
3137 					    SFC_LOGTYPE_MAIN_STR,
3138 					    RTE_LOG_NOTICE);
3139 
3140 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
3141 		return -sfc_eth_dev_secondary_init(dev, logtype_main);
3142 
3143 	/* Required for logging */
3144 	ret = snprintf(sas->log_prefix, sizeof(sas->log_prefix),
3145 			"PMD: sfc_efx " PCI_PRI_FMT " #%" PRIu16 ": ",
3146 			pci_dev->addr.domain, pci_dev->addr.bus,
3147 			pci_dev->addr.devid, pci_dev->addr.function,
3148 			dev->data->port_id);
3149 	if (ret < 0 || ret >= (int)sizeof(sas->log_prefix)) {
3150 		SFC_GENERIC_LOG(ERR,
3151 			"reserved log prefix is too short for " PCI_PRI_FMT,
3152 			pci_dev->addr.domain, pci_dev->addr.bus,
3153 			pci_dev->addr.devid, pci_dev->addr.function);
3154 		return -EINVAL;
3155 	}
3156 	sas->pci_addr = pci_dev->addr;
3157 	sas->port_id = dev->data->port_id;
3158 
3159 	/*
3160 	 * Allocate process private data from heap, since it should not
3161 	 * be located in shared memory allocated using rte_malloc() API.
3162 	 */
3163 	sa = calloc(1, sizeof(*sa));
3164 	if (sa == NULL) {
3165 		rc = ENOMEM;
3166 		goto fail_alloc_sa;
3167 	}
3168 
3169 	dev->process_private = sa;
3170 
3171 	/* Required for logging */
3172 	sa->priv.shared = sas;
3173 	sa->priv.logtype_main = logtype_main;
3174 
3175 	sa->eth_dev = dev;
3176 
3177 	/* Copy PCI device info to the dev->data */
3178 	rte_eth_copy_pci_info(dev, pci_dev);
3179 	dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE;
3180 
3181 	rc = sfc_kvargs_parse(sa);
3182 	if (rc != 0)
3183 		goto fail_kvargs_parse;
3184 
3185 	sfc_log_init(sa, "entry");
3186 
3187 	dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
3188 	if (dev->data->mac_addrs == NULL) {
3189 		rc = ENOMEM;
3190 		goto fail_mac_addrs;
3191 	}
3192 
3193 	sfc_adapter_lock_init(sa);
3194 	sfc_adapter_lock(sa);
3195 
3196 	sfc_log_init(sa, "probing");
3197 	rc = sfc_probe(sa);
3198 	if (rc != 0)
3199 		goto fail_probe;
3200 
3201 	/*
3202 	 * Selecting a default switch mode requires the NIC to be probed and
3203 	 * to have its capabilities filled in.
3204 	 */
3205 	rc = sfc_parse_switch_mode(sa, init_data->nb_representors > 0);
3206 	if (rc != 0)
3207 		goto fail_switch_mode;
3208 
3209 	sfc_log_init(sa, "set device ops");
3210 	rc = sfc_eth_dev_set_ops(dev);
3211 	if (rc != 0)
3212 		goto fail_set_ops;
3213 
3214 	sfc_log_init(sa, "attaching");
3215 	rc = sfc_attach(sa);
3216 	if (rc != 0)
3217 		goto fail_attach;
3218 
3219 	if (sa->switchdev && sa->mae.status != SFC_MAE_STATUS_ADMIN) {
3220 		sfc_err(sa,
3221 			"failed to enable switchdev mode without admin MAE privilege");
3222 		rc = ENOTSUP;
3223 		goto fail_switchdev_no_mae;
3224 	}
3225 
3226 	encp = efx_nic_cfg_get(sa->nic);
3227 
3228 	/*
3229 	 * The arguments are really reverse order in comparison to
3230 	 * Linux kernel. Copy from NIC config to Ethernet device data.
3231 	 */
3232 	from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
3233 	rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
3234 
3235 	/*
3236 	 * Setup the NIC DMA mapping handler. All internal mempools
3237 	 * MUST be created on attach before this point, and the
3238 	 * adapter MUST NOT create mempools with the adapter lock
3239 	 * held after this point.
3240 	 */
3241 	rc = sfc_nic_dma_attach(sa);
3242 	if (rc != 0)
3243 		goto fail_nic_dma_attach;
3244 
3245 	sfc_adapter_unlock(sa);
3246 
3247 	sfc_log_init(sa, "done");
3248 	return 0;
3249 
3250 fail_nic_dma_attach:
3251 fail_switchdev_no_mae:
3252 	sfc_detach(sa);
3253 
3254 fail_attach:
3255 	sfc_eth_dev_clear_ops(dev);
3256 
3257 fail_set_ops:
3258 fail_switch_mode:
3259 	sfc_unprobe(sa);
3260 
3261 fail_probe:
3262 	sfc_adapter_unlock(sa);
3263 	sfc_adapter_lock_fini(sa);
3264 	rte_free(dev->data->mac_addrs);
3265 	dev->data->mac_addrs = NULL;
3266 
3267 fail_mac_addrs:
3268 	sfc_kvargs_cleanup(sa);
3269 
3270 fail_kvargs_parse:
3271 	sfc_log_init(sa, "failed %d", rc);
3272 	dev->process_private = NULL;
3273 	free(sa);
3274 
3275 fail_alloc_sa:
3276 	SFC_ASSERT(rc > 0);
3277 	return -rc;
3278 }
3279 
3280 static int
3281 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
3282 {
3283 	sfc_dev_close(dev);
3284 
3285 	return 0;
3286 }
3287 
3288 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
3289 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
3290 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
3291 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
3292 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
3293 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
3294 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
3295 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
3296 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
3297 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD) },
3298 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD_VF) },
3299 	{ .vendor_id = 0 /* sentinel */ }
3300 };
3301 
3302 static int
3303 sfc_parse_rte_devargs(const char *args, struct rte_eth_devargs *devargs)
3304 {
3305 	struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
3306 	int rc;
3307 
3308 	if (args != NULL) {
3309 		rc = rte_eth_devargs_parse(args, &eth_da, 1);
3310 		if (rc < 0) {
3311 			SFC_GENERIC_LOG(ERR,
3312 					"Failed to parse generic devargs '%s'",
3313 					args);
3314 			return rc;
3315 		}
3316 	}
3317 
3318 	*devargs = eth_da;
3319 
3320 	return 0;
3321 }
3322 
3323 static int
3324 sfc_eth_dev_find_or_create(struct rte_pci_device *pci_dev,
3325 			   struct sfc_ethdev_init_data *init_data,
3326 			   struct rte_eth_dev **devp,
3327 			   bool *dev_created)
3328 {
3329 	struct rte_eth_dev *dev;
3330 	bool created = false;
3331 	int rc;
3332 
3333 	dev = rte_eth_dev_allocated(pci_dev->device.name);
3334 	if (dev == NULL) {
3335 		rc = rte_eth_dev_create(&pci_dev->device, pci_dev->device.name,
3336 					sizeof(struct sfc_adapter_shared),
3337 					eth_dev_pci_specific_init, pci_dev,
3338 					sfc_eth_dev_init, init_data);
3339 		if (rc != 0) {
3340 			SFC_GENERIC_LOG(ERR, "Failed to create sfc ethdev '%s'",
3341 					pci_dev->device.name);
3342 			return rc;
3343 		}
3344 
3345 		created = true;
3346 
3347 		dev = rte_eth_dev_allocated(pci_dev->device.name);
3348 		if (dev == NULL) {
3349 			SFC_GENERIC_LOG(ERR,
3350 				"Failed to find allocated sfc ethdev '%s'",
3351 				pci_dev->device.name);
3352 			return -ENODEV;
3353 		}
3354 	}
3355 
3356 	*devp = dev;
3357 	*dev_created = created;
3358 
3359 	return 0;
3360 }
3361 
3362 static int
3363 sfc_eth_dev_create_repr(struct sfc_adapter *sa,
3364 			efx_pcie_interface_t controller,
3365 			uint16_t port,
3366 			uint16_t repr_port,
3367 			enum rte_eth_representor_type type)
3368 {
3369 	struct sfc_repr_entity_info entity;
3370 	efx_mport_sel_t mport_sel;
3371 	int rc;
3372 
3373 	switch (type) {
3374 	case RTE_ETH_REPRESENTOR_NONE:
3375 		return 0;
3376 	case RTE_ETH_REPRESENTOR_VF:
3377 	case RTE_ETH_REPRESENTOR_PF:
3378 		break;
3379 	case RTE_ETH_REPRESENTOR_SF:
3380 		sfc_err(sa, "SF representors are not supported");
3381 		return ENOTSUP;
3382 	default:
3383 		sfc_err(sa, "unknown representor type: %d", type);
3384 		return ENOTSUP;
3385 	}
3386 
3387 	rc = efx_mae_mport_by_pcie_mh_function(controller,
3388 					       port,
3389 					       repr_port,
3390 					       &mport_sel);
3391 	if (rc != 0) {
3392 		sfc_err(sa,
3393 			"failed to get m-port selector for controller %u port %u repr_port %u: %s",
3394 			controller, port, repr_port, rte_strerror(-rc));
3395 		return rc;
3396 	}
3397 
3398 	memset(&entity, 0, sizeof(entity));
3399 	entity.type = type;
3400 	entity.intf = controller;
3401 	entity.pf = port;
3402 	entity.vf = repr_port;
3403 
3404 	rc = sfc_repr_create(sa->eth_dev, &entity, sa->mae.switch_domain_id,
3405 			     &mport_sel);
3406 	if (rc != 0) {
3407 		sfc_err(sa,
3408 			"failed to create representor for controller %u port %u repr_port %u: %s",
3409 			controller, port, repr_port, rte_strerror(-rc));
3410 		return rc;
3411 	}
3412 
3413 	return 0;
3414 }
3415 
3416 static int
3417 sfc_eth_dev_create_repr_port(struct sfc_adapter *sa,
3418 			     const struct rte_eth_devargs *eth_da,
3419 			     efx_pcie_interface_t controller,
3420 			     uint16_t port)
3421 {
3422 	int first_error = 0;
3423 	uint16_t i;
3424 	int rc;
3425 
3426 	if (eth_da->type == RTE_ETH_REPRESENTOR_PF) {
3427 		return sfc_eth_dev_create_repr(sa, controller, port,
3428 					       EFX_PCI_VF_INVALID,
3429 					       eth_da->type);
3430 	}
3431 
3432 	for (i = 0; i < eth_da->nb_representor_ports; i++) {
3433 		rc = sfc_eth_dev_create_repr(sa, controller, port,
3434 					     eth_da->representor_ports[i],
3435 					     eth_da->type);
3436 		if (rc != 0 && first_error == 0)
3437 			first_error = rc;
3438 	}
3439 
3440 	return first_error;
3441 }
3442 
3443 static int
3444 sfc_eth_dev_create_repr_controller(struct sfc_adapter *sa,
3445 				   const struct rte_eth_devargs *eth_da,
3446 				   efx_pcie_interface_t controller)
3447 {
3448 	const efx_nic_cfg_t *encp;
3449 	int first_error = 0;
3450 	uint16_t default_port;
3451 	uint16_t i;
3452 	int rc;
3453 
3454 	if (eth_da->nb_ports == 0) {
3455 		encp = efx_nic_cfg_get(sa->nic);
3456 		default_port = encp->enc_intf == controller ? encp->enc_pf : 0;
3457 		return sfc_eth_dev_create_repr_port(sa, eth_da, controller,
3458 						    default_port);
3459 	}
3460 
3461 	for (i = 0; i < eth_da->nb_ports; i++) {
3462 		rc = sfc_eth_dev_create_repr_port(sa, eth_da, controller,
3463 						  eth_da->ports[i]);
3464 		if (rc != 0 && first_error == 0)
3465 			first_error = rc;
3466 	}
3467 
3468 	return first_error;
3469 }
3470 
3471 static int
3472 sfc_eth_dev_create_representors(struct rte_eth_dev *dev,
3473 				const struct rte_eth_devargs *eth_da)
3474 {
3475 	efx_pcie_interface_t intf;
3476 	const efx_nic_cfg_t *encp;
3477 	struct sfc_adapter *sa;
3478 	uint16_t switch_domain_id;
3479 	uint16_t i;
3480 	int rc;
3481 
3482 	sa = sfc_adapter_by_eth_dev(dev);
3483 	switch_domain_id = sa->mae.switch_domain_id;
3484 
3485 	switch (eth_da->type) {
3486 	case RTE_ETH_REPRESENTOR_NONE:
3487 		return 0;
3488 	case RTE_ETH_REPRESENTOR_PF:
3489 	case RTE_ETH_REPRESENTOR_VF:
3490 		break;
3491 	case RTE_ETH_REPRESENTOR_SF:
3492 		sfc_err(sa, "SF representors are not supported");
3493 		return -ENOTSUP;
3494 	default:
3495 		sfc_err(sa, "unknown representor type: %d",
3496 			eth_da->type);
3497 		return -ENOTSUP;
3498 	}
3499 
3500 	if (!sa->switchdev) {
3501 		sfc_err(sa, "cannot create representors in non-switchdev mode");
3502 		return -EINVAL;
3503 	}
3504 
3505 	if (!sfc_repr_available(sfc_sa2shared(sa))) {
3506 		sfc_err(sa, "cannot create representors: unsupported");
3507 
3508 		return -ENOTSUP;
3509 	}
3510 
3511 	/*
3512 	 * This is needed to construct the DPDK controller -> EFX interface
3513 	 * mapping.
3514 	 */
3515 	sfc_adapter_lock(sa);
3516 	rc = sfc_process_mport_journal(sa);
3517 	sfc_adapter_unlock(sa);
3518 	if (rc != 0) {
3519 		SFC_ASSERT(rc > 0);
3520 		return -rc;
3521 	}
3522 
3523 	if (eth_da->nb_mh_controllers > 0) {
3524 		for (i = 0; i < eth_da->nb_mh_controllers; i++) {
3525 			rc = sfc_mae_switch_domain_get_intf(switch_domain_id,
3526 						eth_da->mh_controllers[i],
3527 						&intf);
3528 			if (rc != 0) {
3529 				sfc_err(sa, "failed to get representor");
3530 				continue;
3531 			}
3532 			sfc_eth_dev_create_repr_controller(sa, eth_da, intf);
3533 		}
3534 	} else {
3535 		encp = efx_nic_cfg_get(sa->nic);
3536 		sfc_eth_dev_create_repr_controller(sa, eth_da, encp->enc_intf);
3537 	}
3538 
3539 	return 0;
3540 }
3541 
3542 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
3543 	struct rte_pci_device *pci_dev)
3544 {
3545 	struct sfc_ethdev_init_data init_data;
3546 	struct rte_eth_devargs eth_da;
3547 	struct rte_eth_dev *dev;
3548 	bool dev_created;
3549 	int rc;
3550 
3551 	if (pci_dev->device.devargs != NULL) {
3552 		rc = sfc_parse_rte_devargs(pci_dev->device.devargs->args,
3553 					   &eth_da);
3554 		if (rc != 0)
3555 			return rc;
3556 	} else {
3557 		memset(&eth_da, 0, sizeof(eth_da));
3558 	}
3559 
3560 	/* If no VF representors specified, check for PF ones */
3561 	if (eth_da.nb_representor_ports > 0)
3562 		init_data.nb_representors = eth_da.nb_representor_ports;
3563 	else
3564 		init_data.nb_representors = eth_da.nb_ports;
3565 
3566 	if (init_data.nb_representors > 0 &&
3567 	    rte_eal_process_type() != RTE_PROC_PRIMARY) {
3568 		SFC_GENERIC_LOG(ERR,
3569 			"Create representors from secondary process not supported, dev '%s'",
3570 			pci_dev->device.name);
3571 		return -ENOTSUP;
3572 	}
3573 
3574 	/*
3575 	 * Driver supports RTE_PCI_DRV_PROBE_AGAIN. Hence create device only
3576 	 * if it does not already exist. Re-probing an existing device is
3577 	 * expected to allow additional representors to be configured.
3578 	 */
3579 	rc = sfc_eth_dev_find_or_create(pci_dev, &init_data, &dev,
3580 					&dev_created);
3581 	if (rc != 0)
3582 		return rc;
3583 
3584 	rc = sfc_eth_dev_create_representors(dev, &eth_da);
3585 	if (rc != 0) {
3586 		if (dev_created)
3587 			(void)rte_eth_dev_destroy(dev, sfc_eth_dev_uninit);
3588 
3589 		return rc;
3590 	}
3591 
3592 	return 0;
3593 }
3594 
3595 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
3596 {
3597 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
3598 }
3599 
3600 static struct rte_pci_driver sfc_efx_pmd = {
3601 	.id_table = pci_id_sfc_efx_map,
3602 	.drv_flags =
3603 		RTE_PCI_DRV_INTR_LSC |
3604 		RTE_PCI_DRV_NEED_MAPPING |
3605 		RTE_PCI_DRV_PROBE_AGAIN,
3606 	.probe = sfc_eth_dev_pci_probe,
3607 	.remove = sfc_eth_dev_pci_remove,
3608 };
3609 
3610 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
3611 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
3612 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
3613 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
3614 	SFC_KVARG_SWITCH_MODE "=" SFC_KVARG_VALUES_SWITCH_MODE " "
3615 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
3616 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
3617 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
3618 	SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
3619 	SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
3620 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
3621 
3622 RTE_INIT(sfc_driver_register_logtype)
3623 {
3624 	int ret;
3625 
3626 	ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
3627 						   RTE_LOG_NOTICE);
3628 	sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_EAL : ret;
3629 }
3630