xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision 3f95dfb9c4ed6a026112330af78f8e3bae7156fc)
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 <rte_dev.h>
11 #include <ethdev_driver.h>
12 #include <ethdev_pci.h>
13 #include <rte_pci.h>
14 #include <rte_bus_pci.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_dp.h"
30 #include "sfc_dp_rx.h"
31 #include "sfc_sw_stats.h"
32 
33 #define SFC_XSTAT_ID_INVALID_VAL  UINT64_MAX
34 #define SFC_XSTAT_ID_INVALID_NAME '\0'
35 
36 uint32_t sfc_logtype_driver;
37 
38 static struct sfc_dp_list sfc_dp_head =
39 	TAILQ_HEAD_INITIALIZER(sfc_dp_head);
40 
41 
42 static void sfc_eth_dev_clear_ops(struct rte_eth_dev *dev);
43 
44 
45 static int
46 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
47 {
48 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
49 	efx_nic_fw_info_t enfi;
50 	int ret;
51 	int rc;
52 
53 	rc = efx_nic_get_fw_version(sa->nic, &enfi);
54 	if (rc != 0)
55 		return -rc;
56 
57 	ret = snprintf(fw_version, fw_size,
58 		       "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
59 		       enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
60 		       enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
61 	if (ret < 0)
62 		return ret;
63 
64 	if (enfi.enfi_dpcpu_fw_ids_valid) {
65 		size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
66 		int ret_extra;
67 
68 		ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
69 				     fw_size - dpcpu_fw_ids_offset,
70 				     " rx%" PRIx16 " tx%" PRIx16,
71 				     enfi.enfi_rx_dpcpu_fw_id,
72 				     enfi.enfi_tx_dpcpu_fw_id);
73 		if (ret_extra < 0)
74 			return ret_extra;
75 
76 		ret += ret_extra;
77 	}
78 
79 	if (fw_size < (size_t)(++ret))
80 		return ret;
81 	else
82 		return 0;
83 }
84 
85 static int
86 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
87 {
88 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
89 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
90 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
91 	struct sfc_rss *rss = &sas->rss;
92 	struct sfc_mae *mae = &sa->mae;
93 	uint64_t txq_offloads_def = 0;
94 
95 	sfc_log_init(sa, "entry");
96 
97 	dev_info->min_mtu = RTE_ETHER_MIN_MTU;
98 	dev_info->max_mtu = EFX_MAC_SDU_MAX;
99 
100 	dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
101 
102 	dev_info->max_vfs = sa->sriov.num_vfs;
103 
104 	/* Autonegotiation may be disabled */
105 	dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
106 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_1000FDX))
107 		dev_info->speed_capa |= ETH_LINK_SPEED_1G;
108 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_10000FDX))
109 		dev_info->speed_capa |= ETH_LINK_SPEED_10G;
110 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_25000FDX))
111 		dev_info->speed_capa |= ETH_LINK_SPEED_25G;
112 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_40000FDX))
113 		dev_info->speed_capa |= ETH_LINK_SPEED_40G;
114 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_50000FDX))
115 		dev_info->speed_capa |= ETH_LINK_SPEED_50G;
116 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_100000FDX))
117 		dev_info->speed_capa |= ETH_LINK_SPEED_100G;
118 
119 	dev_info->max_rx_queues = sa->rxq_max;
120 	dev_info->max_tx_queues = sa->txq_max;
121 
122 	/* By default packets are dropped if no descriptors are available */
123 	dev_info->default_rxconf.rx_drop_en = 1;
124 
125 	dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
126 
127 	/*
128 	 * rx_offload_capa includes both device and queue offloads since
129 	 * the latter may be requested on a per device basis which makes
130 	 * sense when some offloads are needed to be set on all queues.
131 	 */
132 	dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
133 				    dev_info->rx_queue_offload_capa;
134 
135 	dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
136 
137 	/*
138 	 * tx_offload_capa includes both device and queue offloads since
139 	 * the latter may be requested on a per device basis which makes
140 	 * sense when some offloads are needed to be set on all queues.
141 	 */
142 	dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
143 				    dev_info->tx_queue_offload_capa;
144 
145 	if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
146 		txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
147 
148 	dev_info->default_txconf.offloads |= txq_offloads_def;
149 
150 	if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
151 		uint64_t rte_hf = 0;
152 		unsigned int i;
153 
154 		for (i = 0; i < rss->hf_map_nb_entries; ++i)
155 			rte_hf |= rss->hf_map[i].rte;
156 
157 		dev_info->reta_size = EFX_RSS_TBL_SIZE;
158 		dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
159 		dev_info->flow_type_rss_offloads = rte_hf;
160 	}
161 
162 	/* Initialize to hardware limits */
163 	dev_info->rx_desc_lim.nb_max = sa->rxq_max_entries;
164 	dev_info->rx_desc_lim.nb_min = sa->rxq_min_entries;
165 	/* The RXQ hardware requires that the descriptor count is a power
166 	 * of 2, but rx_desc_lim cannot properly describe that constraint.
167 	 */
168 	dev_info->rx_desc_lim.nb_align = sa->rxq_min_entries;
169 
170 	/* Initialize to hardware limits */
171 	dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
172 	dev_info->tx_desc_lim.nb_min = sa->txq_min_entries;
173 	/*
174 	 * The TXQ hardware requires that the descriptor count is a power
175 	 * of 2, but tx_desc_lim cannot properly describe that constraint
176 	 */
177 	dev_info->tx_desc_lim.nb_align = sa->txq_min_entries;
178 
179 	if (sap->dp_rx->get_dev_info != NULL)
180 		sap->dp_rx->get_dev_info(dev_info);
181 	if (sap->dp_tx->get_dev_info != NULL)
182 		sap->dp_tx->get_dev_info(dev_info);
183 
184 	dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
185 			     RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
186 
187 	if (mae->status == SFC_MAE_STATUS_SUPPORTED) {
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)
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 }
203 
204 static int
205 sfc_dev_configure(struct rte_eth_dev *dev)
206 {
207 	struct rte_eth_dev_data *dev_data = dev->data;
208 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
209 	int rc;
210 
211 	sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
212 		     dev_data->nb_rx_queues, dev_data->nb_tx_queues);
213 
214 	sfc_adapter_lock(sa);
215 	switch (sa->state) {
216 	case SFC_ADAPTER_CONFIGURED:
217 		/* FALLTHROUGH */
218 	case SFC_ADAPTER_INITIALIZED:
219 		rc = sfc_configure(sa);
220 		break;
221 	default:
222 		sfc_err(sa, "unexpected adapter state %u to configure",
223 			sa->state);
224 		rc = EINVAL;
225 		break;
226 	}
227 	sfc_adapter_unlock(sa);
228 
229 	sfc_log_init(sa, "done %d", rc);
230 	SFC_ASSERT(rc >= 0);
231 	return -rc;
232 }
233 
234 static int
235 sfc_dev_start(struct rte_eth_dev *dev)
236 {
237 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
238 	int rc;
239 
240 	sfc_log_init(sa, "entry");
241 
242 	sfc_adapter_lock(sa);
243 	rc = sfc_start(sa);
244 	sfc_adapter_unlock(sa);
245 
246 	sfc_log_init(sa, "done %d", rc);
247 	SFC_ASSERT(rc >= 0);
248 	return -rc;
249 }
250 
251 static int
252 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
253 {
254 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
255 	struct rte_eth_link current_link;
256 	int ret;
257 
258 	sfc_log_init(sa, "entry");
259 
260 	if (sa->state != SFC_ADAPTER_STARTED) {
261 		sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_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, &current_link);
268 
269 	} else {
270 		sfc_ev_mgmt_qpoll(sa);
271 		rte_eth_linkstatus_get(dev, &current_link);
272 	}
273 
274 	ret = rte_eth_linkstatus_set(dev, &current_link);
275 	if (ret == 0)
276 		sfc_notice(sa, "Link status is %s",
277 			   current_link.link_status ? "UP" : "DOWN");
278 
279 	return ret;
280 }
281 
282 static int
283 sfc_dev_stop(struct rte_eth_dev *dev)
284 {
285 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
286 
287 	sfc_log_init(sa, "entry");
288 
289 	sfc_adapter_lock(sa);
290 	sfc_stop(sa);
291 	sfc_adapter_unlock(sa);
292 
293 	sfc_log_init(sa, "done");
294 
295 	return 0;
296 }
297 
298 static int
299 sfc_dev_set_link_up(struct rte_eth_dev *dev)
300 {
301 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
302 	int rc;
303 
304 	sfc_log_init(sa, "entry");
305 
306 	sfc_adapter_lock(sa);
307 	rc = sfc_start(sa);
308 	sfc_adapter_unlock(sa);
309 
310 	SFC_ASSERT(rc >= 0);
311 	return -rc;
312 }
313 
314 static int
315 sfc_dev_set_link_down(struct rte_eth_dev *dev)
316 {
317 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
318 
319 	sfc_log_init(sa, "entry");
320 
321 	sfc_adapter_lock(sa);
322 	sfc_stop(sa);
323 	sfc_adapter_unlock(sa);
324 
325 	return 0;
326 }
327 
328 static void
329 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
330 {
331 	free(dev->process_private);
332 	rte_eth_dev_release_port(dev);
333 }
334 
335 static int
336 sfc_dev_close(struct rte_eth_dev *dev)
337 {
338 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
339 
340 	sfc_log_init(sa, "entry");
341 
342 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
343 		sfc_eth_dev_secondary_clear_ops(dev);
344 		return 0;
345 	}
346 
347 	sfc_adapter_lock(sa);
348 	switch (sa->state) {
349 	case SFC_ADAPTER_STARTED:
350 		sfc_stop(sa);
351 		SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
352 		/* FALLTHROUGH */
353 	case SFC_ADAPTER_CONFIGURED:
354 		sfc_close(sa);
355 		SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
356 		/* FALLTHROUGH */
357 	case SFC_ADAPTER_INITIALIZED:
358 		break;
359 	default:
360 		sfc_err(sa, "unexpected adapter state %u on close", sa->state);
361 		break;
362 	}
363 
364 	/*
365 	 * Cleanup all resources.
366 	 * Rollback primary process sfc_eth_dev_init() below.
367 	 */
368 
369 	sfc_eth_dev_clear_ops(dev);
370 
371 	sfc_detach(sa);
372 	sfc_unprobe(sa);
373 
374 	sfc_kvargs_cleanup(sa);
375 
376 	sfc_adapter_unlock(sa);
377 	sfc_adapter_lock_fini(sa);
378 
379 	sfc_log_init(sa, "done");
380 
381 	/* Required for logging, so cleanup last */
382 	sa->eth_dev = NULL;
383 
384 	free(sa);
385 
386 	return 0;
387 }
388 
389 static int
390 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
391 		   boolean_t enabled)
392 {
393 	struct sfc_port *port;
394 	boolean_t *toggle;
395 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
396 	boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
397 	const char *desc = (allmulti) ? "all-multi" : "promiscuous";
398 	int rc = 0;
399 
400 	sfc_adapter_lock(sa);
401 
402 	port = &sa->port;
403 	toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
404 
405 	if (*toggle != enabled) {
406 		*toggle = enabled;
407 
408 		if (sfc_sa2shared(sa)->isolated) {
409 			sfc_warn(sa, "isolated mode is active on the port");
410 			sfc_warn(sa, "the change is to be applied on the next "
411 				     "start provided that isolated mode is "
412 				     "disabled prior the next start");
413 		} else if ((sa->state == SFC_ADAPTER_STARTED) &&
414 			   ((rc = sfc_set_rx_mode(sa)) != 0)) {
415 			*toggle = !(enabled);
416 			sfc_warn(sa, "Failed to %s %s mode, rc = %d",
417 				 ((enabled) ? "enable" : "disable"), desc, rc);
418 
419 			/*
420 			 * For promiscuous and all-multicast filters a
421 			 * permission failure should be reported as an
422 			 * unsupported filter.
423 			 */
424 			if (rc == EPERM)
425 				rc = ENOTSUP;
426 		}
427 	}
428 
429 	sfc_adapter_unlock(sa);
430 	return rc;
431 }
432 
433 static int
434 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
435 {
436 	int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
437 
438 	SFC_ASSERT(rc >= 0);
439 	return -rc;
440 }
441 
442 static int
443 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
444 {
445 	int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
446 
447 	SFC_ASSERT(rc >= 0);
448 	return -rc;
449 }
450 
451 static int
452 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
453 {
454 	int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
455 
456 	SFC_ASSERT(rc >= 0);
457 	return -rc;
458 }
459 
460 static int
461 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
462 {
463 	int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
464 
465 	SFC_ASSERT(rc >= 0);
466 	return -rc;
467 }
468 
469 static int
470 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid,
471 		   uint16_t nb_rx_desc, unsigned int socket_id,
472 		   const struct rte_eth_rxconf *rx_conf,
473 		   struct rte_mempool *mb_pool)
474 {
475 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
476 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
477 	sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
478 	struct sfc_rxq_info *rxq_info;
479 	sfc_sw_index_t sw_index;
480 	int rc;
481 
482 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
483 		     ethdev_qid, nb_rx_desc, socket_id);
484 
485 	sfc_adapter_lock(sa);
486 
487 	sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
488 	rc = sfc_rx_qinit(sa, sw_index, nb_rx_desc, socket_id,
489 			  rx_conf, mb_pool);
490 	if (rc != 0)
491 		goto fail_rx_qinit;
492 
493 	rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
494 	dev->data->rx_queues[ethdev_qid] = rxq_info->dp;
495 
496 	sfc_adapter_unlock(sa);
497 
498 	return 0;
499 
500 fail_rx_qinit:
501 	sfc_adapter_unlock(sa);
502 	SFC_ASSERT(rc > 0);
503 	return -rc;
504 }
505 
506 static void
507 sfc_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
508 {
509 	struct sfc_dp_rxq *dp_rxq = dev->data->rx_queues[qid];
510 	struct sfc_rxq *rxq;
511 	struct sfc_adapter *sa;
512 	sfc_sw_index_t sw_index;
513 
514 	if (dp_rxq == NULL)
515 		return;
516 
517 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
518 	sa = rxq->evq->sa;
519 	sfc_adapter_lock(sa);
520 
521 	sw_index = dp_rxq->dpq.queue_id;
522 
523 	sfc_log_init(sa, "RxQ=%u", sw_index);
524 
525 	sfc_rx_qfini(sa, sw_index);
526 
527 	sfc_adapter_unlock(sa);
528 }
529 
530 static int
531 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid,
532 		   uint16_t nb_tx_desc, unsigned int socket_id,
533 		   const struct rte_eth_txconf *tx_conf)
534 {
535 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
536 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
537 	struct sfc_txq_info *txq_info;
538 	sfc_sw_index_t sw_index;
539 	int rc;
540 
541 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
542 		     ethdev_qid, nb_tx_desc, socket_id);
543 
544 	sfc_adapter_lock(sa);
545 
546 	sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
547 	rc = sfc_tx_qinit(sa, sw_index, nb_tx_desc, socket_id, tx_conf);
548 	if (rc != 0)
549 		goto fail_tx_qinit;
550 
551 	txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
552 	dev->data->tx_queues[ethdev_qid] = txq_info->dp;
553 
554 	sfc_adapter_unlock(sa);
555 	return 0;
556 
557 fail_tx_qinit:
558 	sfc_adapter_unlock(sa);
559 	SFC_ASSERT(rc > 0);
560 	return -rc;
561 }
562 
563 static void
564 sfc_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
565 {
566 	struct sfc_dp_txq *dp_txq = dev->data->tx_queues[qid];
567 	struct sfc_txq *txq;
568 	sfc_sw_index_t sw_index;
569 	struct sfc_adapter *sa;
570 
571 	if (dp_txq == NULL)
572 		return;
573 
574 	txq = sfc_txq_by_dp_txq(dp_txq);
575 	sw_index = dp_txq->dpq.queue_id;
576 
577 	SFC_ASSERT(txq->evq != NULL);
578 	sa = txq->evq->sa;
579 
580 	sfc_log_init(sa, "TxQ = %u", sw_index);
581 
582 	sfc_adapter_lock(sa);
583 
584 	sfc_tx_qfini(sa, sw_index);
585 
586 	sfc_adapter_unlock(sa);
587 }
588 
589 static void
590 sfc_stats_get_dp_rx(struct sfc_adapter *sa, uint64_t *pkts, uint64_t *bytes)
591 {
592 	struct sfc_adapter_shared *sas = sfc_sa2shared(sa);
593 	uint64_t pkts_sum = 0;
594 	uint64_t bytes_sum = 0;
595 	unsigned int i;
596 
597 	for (i = 0; i < sas->ethdev_rxq_count; ++i) {
598 		struct sfc_rxq_info *rxq_info;
599 
600 		rxq_info = sfc_rxq_info_by_ethdev_qid(sas, i);
601 		if (rxq_info->state & SFC_RXQ_INITIALIZED) {
602 			union sfc_pkts_bytes qstats;
603 
604 			sfc_pkts_bytes_get(&rxq_info->dp->dpq.stats, &qstats);
605 			pkts_sum += qstats.pkts -
606 					sa->sw_stats.reset_rx_pkts[i];
607 			bytes_sum += qstats.bytes -
608 					sa->sw_stats.reset_rx_bytes[i];
609 		}
610 	}
611 
612 	*pkts = pkts_sum;
613 	*bytes = bytes_sum;
614 }
615 
616 static void
617 sfc_stats_get_dp_tx(struct sfc_adapter *sa, uint64_t *pkts, uint64_t *bytes)
618 {
619 	struct sfc_adapter_shared *sas = sfc_sa2shared(sa);
620 	uint64_t pkts_sum = 0;
621 	uint64_t bytes_sum = 0;
622 	unsigned int i;
623 
624 	for (i = 0; i < sas->ethdev_txq_count; ++i) {
625 		struct sfc_txq_info *txq_info;
626 
627 		txq_info = sfc_txq_info_by_ethdev_qid(sas, i);
628 		if (txq_info->state & SFC_TXQ_INITIALIZED) {
629 			union sfc_pkts_bytes qstats;
630 
631 			sfc_pkts_bytes_get(&txq_info->dp->dpq.stats, &qstats);
632 			pkts_sum += qstats.pkts -
633 					sa->sw_stats.reset_tx_pkts[i];
634 			bytes_sum += qstats.bytes -
635 					sa->sw_stats.reset_tx_bytes[i];
636 		}
637 	}
638 
639 	*pkts = pkts_sum;
640 	*bytes = bytes_sum;
641 }
642 
643 /*
644  * Some statistics are computed as A - B where A and B each increase
645  * monotonically with some hardware counter(s) and the counters are read
646  * asynchronously.
647  *
648  * If packet X is counted in A, but not counted in B yet, computed value is
649  * greater than real.
650  *
651  * If packet X is not counted in A at the moment of reading the counter,
652  * but counted in B at the moment of reading the counter, computed value
653  * is less than real.
654  *
655  * However, counter which grows backward is worse evil than slightly wrong
656  * value. So, let's try to guarantee that it never happens except may be
657  * the case when the MAC stats are zeroed as a result of a NIC reset.
658  */
659 static void
660 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
661 {
662 	if ((int64_t)(newval - *stat) > 0 || newval == 0)
663 		*stat = newval;
664 }
665 
666 static int
667 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
668 {
669 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
670 	bool have_dp_rx_stats = sap->dp_rx->features & SFC_DP_RX_FEAT_STATS;
671 	bool have_dp_tx_stats = sap->dp_tx->features & SFC_DP_TX_FEAT_STATS;
672 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
673 	struct sfc_port *port = &sa->port;
674 	uint64_t *mac_stats;
675 	int ret;
676 
677 	sfc_adapter_lock(sa);
678 
679 	if (have_dp_rx_stats)
680 		sfc_stats_get_dp_rx(sa, &stats->ipackets, &stats->ibytes);
681 	if (have_dp_tx_stats)
682 		sfc_stats_get_dp_tx(sa, &stats->opackets, &stats->obytes);
683 
684 	ret = sfc_port_update_mac_stats(sa, B_FALSE);
685 	if (ret != 0)
686 		goto unlock;
687 
688 	mac_stats = port->mac_stats_buf;
689 
690 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
691 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
692 		if (!have_dp_rx_stats) {
693 			stats->ipackets =
694 				mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
695 				mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
696 				mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
697 			stats->ibytes =
698 				mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
699 				mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
700 				mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
701 
702 			/* CRC is included in these stats, but shouldn't be */
703 			stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
704 		}
705 		if (!have_dp_tx_stats) {
706 			stats->opackets =
707 				mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
708 				mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
709 				mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
710 			stats->obytes =
711 				mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
712 				mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
713 				mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
714 
715 			/* CRC is included in these stats, but shouldn't be */
716 			stats->obytes -= stats->opackets * RTE_ETHER_CRC_LEN;
717 		}
718 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
719 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
720 	} else {
721 		if (!have_dp_tx_stats) {
722 			stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
723 			stats->obytes = mac_stats[EFX_MAC_TX_OCTETS] -
724 				mac_stats[EFX_MAC_TX_PKTS] * RTE_ETHER_CRC_LEN;
725 		}
726 
727 		/*
728 		 * Take into account stats which are whenever supported
729 		 * on EF10. If some stat is not supported by current
730 		 * firmware variant or HW revision, it is guaranteed
731 		 * to be zero in mac_stats.
732 		 */
733 		stats->imissed =
734 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
735 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
736 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
737 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
738 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
739 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
740 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
741 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
742 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
743 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
744 		stats->ierrors =
745 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
746 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
747 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
748 		/* no oerrors counters supported on EF10 */
749 
750 		if (!have_dp_rx_stats) {
751 			/* Exclude missed, errors and pauses from Rx packets */
752 			sfc_update_diff_stat(&port->ipackets,
753 				mac_stats[EFX_MAC_RX_PKTS] -
754 				mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
755 				stats->imissed - stats->ierrors);
756 			stats->ipackets = port->ipackets;
757 			stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS] -
758 				mac_stats[EFX_MAC_RX_PKTS] * RTE_ETHER_CRC_LEN;
759 		}
760 	}
761 
762 unlock:
763 	sfc_adapter_unlock(sa);
764 	SFC_ASSERT(ret >= 0);
765 	return -ret;
766 }
767 
768 static int
769 sfc_stats_reset(struct rte_eth_dev *dev)
770 {
771 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
772 	struct sfc_port *port = &sa->port;
773 	int rc;
774 
775 	sfc_adapter_lock(sa);
776 
777 	if (sa->state != SFC_ADAPTER_STARTED) {
778 		/*
779 		 * The operation cannot be done if port is not started; it
780 		 * will be scheduled to be done during the next port start
781 		 */
782 		port->mac_stats_reset_pending = B_TRUE;
783 		sfc_adapter_unlock(sa);
784 		return 0;
785 	}
786 
787 	rc = sfc_port_reset_mac_stats(sa);
788 	if (rc != 0)
789 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
790 
791 	sfc_sw_xstats_reset(sa);
792 
793 	sfc_adapter_unlock(sa);
794 
795 	SFC_ASSERT(rc >= 0);
796 	return -rc;
797 }
798 
799 static unsigned int
800 sfc_xstats_get_nb_supported(struct sfc_adapter *sa)
801 {
802 	struct sfc_port *port = &sa->port;
803 	unsigned int nb_supported;
804 
805 	sfc_adapter_lock(sa);
806 	nb_supported = port->mac_stats_nb_supported +
807 		       sfc_sw_xstats_get_nb_supported(sa);
808 	sfc_adapter_unlock(sa);
809 
810 	return nb_supported;
811 }
812 
813 static int
814 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
815 	       unsigned int xstats_count)
816 {
817 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
818 	unsigned int nb_written = 0;
819 	unsigned int nb_supported = 0;
820 	int rc;
821 
822 	if (unlikely(xstats == NULL))
823 		return sfc_xstats_get_nb_supported(sa);
824 
825 	rc = sfc_port_get_mac_stats(sa, xstats, xstats_count, &nb_written);
826 	if (rc < 0)
827 		return rc;
828 
829 	nb_supported = rc;
830 	sfc_sw_xstats_get_vals(sa, xstats, xstats_count, &nb_written,
831 			       &nb_supported);
832 
833 	return nb_supported;
834 }
835 
836 static int
837 sfc_xstats_get_names(struct rte_eth_dev *dev,
838 		     struct rte_eth_xstat_name *xstats_names,
839 		     unsigned int xstats_count)
840 {
841 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
842 	struct sfc_port *port = &sa->port;
843 	unsigned int i;
844 	unsigned int nstats = 0;
845 	unsigned int nb_written = 0;
846 	int ret;
847 
848 	if (unlikely(xstats_names == NULL))
849 		return sfc_xstats_get_nb_supported(sa);
850 
851 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
852 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
853 			if (nstats < xstats_count) {
854 				strlcpy(xstats_names[nstats].name,
855 					efx_mac_stat_name(sa->nic, i),
856 					sizeof(xstats_names[0].name));
857 				nb_written++;
858 			}
859 			nstats++;
860 		}
861 	}
862 
863 	ret = sfc_sw_xstats_get_names(sa, xstats_names, xstats_count,
864 				      &nb_written, &nstats);
865 	if (ret != 0) {
866 		SFC_ASSERT(ret < 0);
867 		return ret;
868 	}
869 
870 	return nstats;
871 }
872 
873 static int
874 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
875 		     uint64_t *values, unsigned int n)
876 {
877 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
878 	struct sfc_port *port = &sa->port;
879 	unsigned int nb_supported;
880 	unsigned int i;
881 	int rc;
882 
883 	if (unlikely(ids == NULL || values == NULL))
884 		return -EINVAL;
885 
886 	/*
887 	 * Values array could be filled in nonsequential order. Fill values with
888 	 * constant indicating invalid ID first.
889 	 */
890 	for (i = 0; i < n; i++)
891 		values[i] = SFC_XSTAT_ID_INVALID_VAL;
892 
893 	rc = sfc_port_get_mac_stats_by_id(sa, ids, values, n);
894 	if (rc != 0)
895 		return rc;
896 
897 	nb_supported = port->mac_stats_nb_supported;
898 	sfc_sw_xstats_get_vals_by_id(sa, ids, values, n, &nb_supported);
899 
900 	/* Return number of written stats before invalid ID is encountered. */
901 	for (i = 0; i < n; i++) {
902 		if (values[i] == SFC_XSTAT_ID_INVALID_VAL)
903 			return i;
904 	}
905 
906 	return n;
907 }
908 
909 static int
910 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
911 			   const uint64_t *ids,
912 			   struct rte_eth_xstat_name *xstats_names,
913 			   unsigned int size)
914 {
915 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
916 	struct sfc_port *port = &sa->port;
917 	unsigned int nb_supported;
918 	unsigned int i;
919 	int ret;
920 
921 	if (unlikely(xstats_names == NULL && ids != NULL) ||
922 	    unlikely(xstats_names != NULL && ids == NULL))
923 		return -EINVAL;
924 
925 	if (unlikely(xstats_names == NULL && ids == NULL))
926 		return sfc_xstats_get_nb_supported(sa);
927 
928 	/*
929 	 * Names array could be filled in nonsequential order. Fill names with
930 	 * string indicating invalid ID first.
931 	 */
932 	for (i = 0; i < size; i++)
933 		xstats_names[i].name[0] = SFC_XSTAT_ID_INVALID_NAME;
934 
935 	sfc_adapter_lock(sa);
936 
937 	SFC_ASSERT(port->mac_stats_nb_supported <=
938 		   RTE_DIM(port->mac_stats_by_id));
939 
940 	for (i = 0; i < size; i++) {
941 		if (ids[i] < port->mac_stats_nb_supported) {
942 			strlcpy(xstats_names[i].name,
943 				efx_mac_stat_name(sa->nic,
944 						 port->mac_stats_by_id[ids[i]]),
945 				sizeof(xstats_names[0].name));
946 		}
947 	}
948 
949 	nb_supported = port->mac_stats_nb_supported;
950 
951 	sfc_adapter_unlock(sa);
952 
953 	ret = sfc_sw_xstats_get_names_by_id(sa, ids, xstats_names, size,
954 					    &nb_supported);
955 	if (ret != 0) {
956 		SFC_ASSERT(ret < 0);
957 		return ret;
958 	}
959 
960 	/* Return number of written names before invalid ID is encountered. */
961 	for (i = 0; i < size; i++) {
962 		if (xstats_names[i].name[0] == SFC_XSTAT_ID_INVALID_NAME)
963 			return i;
964 	}
965 
966 	return size;
967 }
968 
969 static int
970 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
971 {
972 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
973 	unsigned int wanted_fc, link_fc;
974 
975 	memset(fc_conf, 0, sizeof(*fc_conf));
976 
977 	sfc_adapter_lock(sa);
978 
979 	if (sa->state == SFC_ADAPTER_STARTED)
980 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
981 	else
982 		link_fc = sa->port.flow_ctrl;
983 
984 	switch (link_fc) {
985 	case 0:
986 		fc_conf->mode = RTE_FC_NONE;
987 		break;
988 	case EFX_FCNTL_RESPOND:
989 		fc_conf->mode = RTE_FC_RX_PAUSE;
990 		break;
991 	case EFX_FCNTL_GENERATE:
992 		fc_conf->mode = RTE_FC_TX_PAUSE;
993 		break;
994 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
995 		fc_conf->mode = RTE_FC_FULL;
996 		break;
997 	default:
998 		sfc_err(sa, "%s: unexpected flow control value %#x",
999 			__func__, link_fc);
1000 	}
1001 
1002 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
1003 
1004 	sfc_adapter_unlock(sa);
1005 
1006 	return 0;
1007 }
1008 
1009 static int
1010 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1011 {
1012 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1013 	struct sfc_port *port = &sa->port;
1014 	unsigned int fcntl;
1015 	int rc;
1016 
1017 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
1018 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
1019 	    fc_conf->mac_ctrl_frame_fwd != 0) {
1020 		sfc_err(sa, "unsupported flow control settings specified");
1021 		rc = EINVAL;
1022 		goto fail_inval;
1023 	}
1024 
1025 	switch (fc_conf->mode) {
1026 	case RTE_FC_NONE:
1027 		fcntl = 0;
1028 		break;
1029 	case RTE_FC_RX_PAUSE:
1030 		fcntl = EFX_FCNTL_RESPOND;
1031 		break;
1032 	case RTE_FC_TX_PAUSE:
1033 		fcntl = EFX_FCNTL_GENERATE;
1034 		break;
1035 	case RTE_FC_FULL:
1036 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
1037 		break;
1038 	default:
1039 		rc = EINVAL;
1040 		goto fail_inval;
1041 	}
1042 
1043 	sfc_adapter_lock(sa);
1044 
1045 	if (sa->state == SFC_ADAPTER_STARTED) {
1046 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
1047 		if (rc != 0)
1048 			goto fail_mac_fcntl_set;
1049 	}
1050 
1051 	port->flow_ctrl = fcntl;
1052 	port->flow_ctrl_autoneg = fc_conf->autoneg;
1053 
1054 	sfc_adapter_unlock(sa);
1055 
1056 	return 0;
1057 
1058 fail_mac_fcntl_set:
1059 	sfc_adapter_unlock(sa);
1060 fail_inval:
1061 	SFC_ASSERT(rc > 0);
1062 	return -rc;
1063 }
1064 
1065 static int
1066 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
1067 {
1068 	struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
1069 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1070 	boolean_t scatter_enabled;
1071 	const char *error;
1072 	unsigned int i;
1073 
1074 	for (i = 0; i < sas->rxq_count; i++) {
1075 		if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
1076 			continue;
1077 
1078 		scatter_enabled = (sas->rxq_info[i].type_flags &
1079 				   EFX_RXQ_FLAG_SCATTER);
1080 
1081 		if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
1082 					  encp->enc_rx_prefix_size,
1083 					  scatter_enabled,
1084 					  encp->enc_rx_scatter_max, &error)) {
1085 			sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
1086 				error);
1087 			return EINVAL;
1088 		}
1089 	}
1090 
1091 	return 0;
1092 }
1093 
1094 static int
1095 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
1096 {
1097 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1098 	size_t pdu = EFX_MAC_PDU(mtu);
1099 	size_t old_pdu;
1100 	int rc;
1101 
1102 	sfc_log_init(sa, "mtu=%u", mtu);
1103 
1104 	rc = EINVAL;
1105 	if (pdu < EFX_MAC_PDU_MIN) {
1106 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
1107 			(unsigned int)mtu, (unsigned int)pdu,
1108 			EFX_MAC_PDU_MIN);
1109 		goto fail_inval;
1110 	}
1111 	if (pdu > EFX_MAC_PDU_MAX) {
1112 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
1113 			(unsigned int)mtu, (unsigned int)pdu,
1114 			(unsigned int)EFX_MAC_PDU_MAX);
1115 		goto fail_inval;
1116 	}
1117 
1118 	sfc_adapter_lock(sa);
1119 
1120 	rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
1121 	if (rc != 0)
1122 		goto fail_check_scatter;
1123 
1124 	if (pdu != sa->port.pdu) {
1125 		if (sa->state == SFC_ADAPTER_STARTED) {
1126 			sfc_stop(sa);
1127 
1128 			old_pdu = sa->port.pdu;
1129 			sa->port.pdu = pdu;
1130 			rc = sfc_start(sa);
1131 			if (rc != 0)
1132 				goto fail_start;
1133 		} else {
1134 			sa->port.pdu = pdu;
1135 		}
1136 	}
1137 
1138 	/*
1139 	 * The driver does not use it, but other PMDs update jumbo frame
1140 	 * flag and max_rx_pkt_len when MTU is set.
1141 	 */
1142 	if (mtu > RTE_ETHER_MTU) {
1143 		struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1144 		rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1145 	}
1146 
1147 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
1148 
1149 	sfc_adapter_unlock(sa);
1150 
1151 	sfc_log_init(sa, "done");
1152 	return 0;
1153 
1154 fail_start:
1155 	sa->port.pdu = old_pdu;
1156 	if (sfc_start(sa) != 0)
1157 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1158 			"PDU max size - port is stopped",
1159 			(unsigned int)pdu, (unsigned int)old_pdu);
1160 
1161 fail_check_scatter:
1162 	sfc_adapter_unlock(sa);
1163 
1164 fail_inval:
1165 	sfc_log_init(sa, "failed %d", rc);
1166 	SFC_ASSERT(rc > 0);
1167 	return -rc;
1168 }
1169 static int
1170 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1171 {
1172 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1173 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1174 	struct sfc_port *port = &sa->port;
1175 	struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1176 	int rc = 0;
1177 
1178 	sfc_adapter_lock(sa);
1179 
1180 	if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr))
1181 		goto unlock;
1182 
1183 	/*
1184 	 * Copy the address to the device private data so that
1185 	 * it could be recalled in the case of adapter restart.
1186 	 */
1187 	rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1188 
1189 	/*
1190 	 * Neither of the two following checks can return
1191 	 * an error. The new MAC address is preserved in
1192 	 * the device private data and can be activated
1193 	 * on the next port start if the user prevents
1194 	 * isolated mode from being enabled.
1195 	 */
1196 	if (sfc_sa2shared(sa)->isolated) {
1197 		sfc_warn(sa, "isolated mode is active on the port");
1198 		sfc_warn(sa, "will not set MAC address");
1199 		goto unlock;
1200 	}
1201 
1202 	if (sa->state != SFC_ADAPTER_STARTED) {
1203 		sfc_notice(sa, "the port is not started");
1204 		sfc_notice(sa, "the new MAC address will be set on port start");
1205 
1206 		goto unlock;
1207 	}
1208 
1209 	if (encp->enc_allow_set_mac_with_installed_filters) {
1210 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1211 		if (rc != 0) {
1212 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1213 			goto unlock;
1214 		}
1215 
1216 		/*
1217 		 * Changing the MAC address by means of MCDI request
1218 		 * has no effect on received traffic, therefore
1219 		 * we also need to update unicast filters
1220 		 */
1221 		rc = sfc_set_rx_mode_unchecked(sa);
1222 		if (rc != 0) {
1223 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
1224 			/* Rollback the old address */
1225 			(void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1226 			(void)sfc_set_rx_mode_unchecked(sa);
1227 		}
1228 	} else {
1229 		sfc_warn(sa, "cannot set MAC address with filters installed");
1230 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1231 		sfc_warn(sa, "(some traffic may be dropped)");
1232 
1233 		/*
1234 		 * Since setting MAC address with filters installed is not
1235 		 * allowed on the adapter, the new MAC address will be set
1236 		 * by means of adapter restart. sfc_start() shall retrieve
1237 		 * the new address from the device private data and set it.
1238 		 */
1239 		sfc_stop(sa);
1240 		rc = sfc_start(sa);
1241 		if (rc != 0)
1242 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1243 	}
1244 
1245 unlock:
1246 	if (rc != 0)
1247 		rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1248 
1249 	sfc_adapter_unlock(sa);
1250 
1251 	SFC_ASSERT(rc >= 0);
1252 	return -rc;
1253 }
1254 
1255 
1256 static int
1257 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1258 		struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1259 {
1260 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1261 	struct sfc_port *port = &sa->port;
1262 	uint8_t *mc_addrs = port->mcast_addrs;
1263 	int rc;
1264 	unsigned int i;
1265 
1266 	if (sfc_sa2shared(sa)->isolated) {
1267 		sfc_err(sa, "isolated mode is active on the port");
1268 		sfc_err(sa, "will not set multicast address list");
1269 		return -ENOTSUP;
1270 	}
1271 
1272 	if (mc_addrs == NULL)
1273 		return -ENOBUFS;
1274 
1275 	if (nb_mc_addr > port->max_mcast_addrs) {
1276 		sfc_err(sa, "too many multicast addresses: %u > %u",
1277 			 nb_mc_addr, port->max_mcast_addrs);
1278 		return -EINVAL;
1279 	}
1280 
1281 	for (i = 0; i < nb_mc_addr; ++i) {
1282 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1283 				 EFX_MAC_ADDR_LEN);
1284 		mc_addrs += EFX_MAC_ADDR_LEN;
1285 	}
1286 
1287 	port->nb_mcast_addrs = nb_mc_addr;
1288 
1289 	if (sa->state != SFC_ADAPTER_STARTED)
1290 		return 0;
1291 
1292 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1293 					port->nb_mcast_addrs);
1294 	if (rc != 0)
1295 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1296 
1297 	SFC_ASSERT(rc >= 0);
1298 	return -rc;
1299 }
1300 
1301 /*
1302  * The function is used by the secondary process as well. It must not
1303  * use any process-local pointers from the adapter data.
1304  */
1305 static void
1306 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1307 		      struct rte_eth_rxq_info *qinfo)
1308 {
1309 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1310 	sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1311 	struct sfc_rxq_info *rxq_info;
1312 
1313 	rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1314 
1315 	qinfo->mp = rxq_info->refill_mb_pool;
1316 	qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1317 	qinfo->conf.rx_drop_en = 1;
1318 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1319 	qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1320 	if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1321 		qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1322 		qinfo->scattered_rx = 1;
1323 	}
1324 	qinfo->nb_desc = rxq_info->entries;
1325 }
1326 
1327 /*
1328  * The function is used by the secondary process as well. It must not
1329  * use any process-local pointers from the adapter data.
1330  */
1331 static void
1332 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1333 		      struct rte_eth_txq_info *qinfo)
1334 {
1335 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1336 	struct sfc_txq_info *txq_info;
1337 
1338 	SFC_ASSERT(ethdev_qid < sas->ethdev_txq_count);
1339 
1340 	txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1341 
1342 	memset(qinfo, 0, sizeof(*qinfo));
1343 
1344 	qinfo->conf.offloads = txq_info->offloads;
1345 	qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1346 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1347 	qinfo->nb_desc = txq_info->entries;
1348 }
1349 
1350 /*
1351  * The function is used by the secondary process as well. It must not
1352  * use any process-local pointers from the adapter data.
1353  */
1354 static uint32_t
1355 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1356 {
1357 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1358 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1359 	sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1360 	struct sfc_rxq_info *rxq_info;
1361 
1362 	rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1363 
1364 	if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1365 		return 0;
1366 
1367 	return sap->dp_rx->qdesc_npending(rxq_info->dp);
1368 }
1369 
1370 /*
1371  * The function is used by the secondary process as well. It must not
1372  * use any process-local pointers from the adapter data.
1373  */
1374 static int
1375 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1376 {
1377 	struct sfc_dp_rxq *dp_rxq = queue;
1378 	const struct sfc_dp_rx *dp_rx;
1379 
1380 	dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1381 
1382 	return dp_rx->qdesc_status(dp_rxq, offset);
1383 }
1384 
1385 /*
1386  * The function is used by the secondary process as well. It must not
1387  * use any process-local pointers from the adapter data.
1388  */
1389 static int
1390 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1391 {
1392 	struct sfc_dp_txq *dp_txq = queue;
1393 	const struct sfc_dp_tx *dp_tx;
1394 
1395 	dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1396 
1397 	return dp_tx->qdesc_status(dp_txq, offset);
1398 }
1399 
1400 static int
1401 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1402 {
1403 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1404 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1405 	sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1406 	struct sfc_rxq_info *rxq_info;
1407 	sfc_sw_index_t sw_index;
1408 	int rc;
1409 
1410 	sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1411 
1412 	sfc_adapter_lock(sa);
1413 
1414 	rc = EINVAL;
1415 	if (sa->state != SFC_ADAPTER_STARTED)
1416 		goto fail_not_started;
1417 
1418 	rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1419 	if (rxq_info->state != SFC_RXQ_INITIALIZED)
1420 		goto fail_not_setup;
1421 
1422 	sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1423 	rc = sfc_rx_qstart(sa, sw_index);
1424 	if (rc != 0)
1425 		goto fail_rx_qstart;
1426 
1427 	rxq_info->deferred_started = B_TRUE;
1428 
1429 	sfc_adapter_unlock(sa);
1430 
1431 	return 0;
1432 
1433 fail_rx_qstart:
1434 fail_not_setup:
1435 fail_not_started:
1436 	sfc_adapter_unlock(sa);
1437 	SFC_ASSERT(rc > 0);
1438 	return -rc;
1439 }
1440 
1441 static int
1442 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1443 {
1444 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1445 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1446 	sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1447 	struct sfc_rxq_info *rxq_info;
1448 	sfc_sw_index_t sw_index;
1449 
1450 	sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1451 
1452 	sfc_adapter_lock(sa);
1453 
1454 	sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1455 	sfc_rx_qstop(sa, sw_index);
1456 
1457 	rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1458 	rxq_info->deferred_started = B_FALSE;
1459 
1460 	sfc_adapter_unlock(sa);
1461 
1462 	return 0;
1463 }
1464 
1465 static int
1466 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1467 {
1468 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1469 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1470 	struct sfc_txq_info *txq_info;
1471 	sfc_sw_index_t sw_index;
1472 	int rc;
1473 
1474 	sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1475 
1476 	sfc_adapter_lock(sa);
1477 
1478 	rc = EINVAL;
1479 	if (sa->state != SFC_ADAPTER_STARTED)
1480 		goto fail_not_started;
1481 
1482 	txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1483 	if (txq_info->state != SFC_TXQ_INITIALIZED)
1484 		goto fail_not_setup;
1485 
1486 	sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1487 	rc = sfc_tx_qstart(sa, sw_index);
1488 	if (rc != 0)
1489 		goto fail_tx_qstart;
1490 
1491 	txq_info->deferred_started = B_TRUE;
1492 
1493 	sfc_adapter_unlock(sa);
1494 	return 0;
1495 
1496 fail_tx_qstart:
1497 
1498 fail_not_setup:
1499 fail_not_started:
1500 	sfc_adapter_unlock(sa);
1501 	SFC_ASSERT(rc > 0);
1502 	return -rc;
1503 }
1504 
1505 static int
1506 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1507 {
1508 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1509 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1510 	struct sfc_txq_info *txq_info;
1511 	sfc_sw_index_t sw_index;
1512 
1513 	sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1514 
1515 	sfc_adapter_lock(sa);
1516 
1517 	sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1518 	sfc_tx_qstop(sa, sw_index);
1519 
1520 	txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1521 	txq_info->deferred_started = B_FALSE;
1522 
1523 	sfc_adapter_unlock(sa);
1524 	return 0;
1525 }
1526 
1527 static efx_tunnel_protocol_t
1528 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1529 {
1530 	switch (rte_type) {
1531 	case RTE_TUNNEL_TYPE_VXLAN:
1532 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1533 	case RTE_TUNNEL_TYPE_GENEVE:
1534 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1535 	default:
1536 		return EFX_TUNNEL_NPROTOS;
1537 	}
1538 }
1539 
1540 enum sfc_udp_tunnel_op_e {
1541 	SFC_UDP_TUNNEL_ADD_PORT,
1542 	SFC_UDP_TUNNEL_DEL_PORT,
1543 };
1544 
1545 static int
1546 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1547 		      struct rte_eth_udp_tunnel *tunnel_udp,
1548 		      enum sfc_udp_tunnel_op_e op)
1549 {
1550 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1551 	efx_tunnel_protocol_t tunnel_proto;
1552 	int rc;
1553 
1554 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1555 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1556 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1557 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1558 
1559 	tunnel_proto =
1560 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1561 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1562 		rc = ENOTSUP;
1563 		goto fail_bad_proto;
1564 	}
1565 
1566 	sfc_adapter_lock(sa);
1567 
1568 	switch (op) {
1569 	case SFC_UDP_TUNNEL_ADD_PORT:
1570 		rc = efx_tunnel_config_udp_add(sa->nic,
1571 					       tunnel_udp->udp_port,
1572 					       tunnel_proto);
1573 		break;
1574 	case SFC_UDP_TUNNEL_DEL_PORT:
1575 		rc = efx_tunnel_config_udp_remove(sa->nic,
1576 						  tunnel_udp->udp_port,
1577 						  tunnel_proto);
1578 		break;
1579 	default:
1580 		rc = EINVAL;
1581 		goto fail_bad_op;
1582 	}
1583 
1584 	if (rc != 0)
1585 		goto fail_op;
1586 
1587 	if (sa->state == SFC_ADAPTER_STARTED) {
1588 		rc = efx_tunnel_reconfigure(sa->nic);
1589 		if (rc == EAGAIN) {
1590 			/*
1591 			 * Configuration is accepted by FW and MC reboot
1592 			 * is initiated to apply the changes. MC reboot
1593 			 * will be handled in a usual way (MC reboot
1594 			 * event on management event queue and adapter
1595 			 * restart).
1596 			 */
1597 			rc = 0;
1598 		} else if (rc != 0) {
1599 			goto fail_reconfigure;
1600 		}
1601 	}
1602 
1603 	sfc_adapter_unlock(sa);
1604 	return 0;
1605 
1606 fail_reconfigure:
1607 	/* Remove/restore entry since the change makes the trouble */
1608 	switch (op) {
1609 	case SFC_UDP_TUNNEL_ADD_PORT:
1610 		(void)efx_tunnel_config_udp_remove(sa->nic,
1611 						   tunnel_udp->udp_port,
1612 						   tunnel_proto);
1613 		break;
1614 	case SFC_UDP_TUNNEL_DEL_PORT:
1615 		(void)efx_tunnel_config_udp_add(sa->nic,
1616 						tunnel_udp->udp_port,
1617 						tunnel_proto);
1618 		break;
1619 	}
1620 
1621 fail_op:
1622 fail_bad_op:
1623 	sfc_adapter_unlock(sa);
1624 
1625 fail_bad_proto:
1626 	SFC_ASSERT(rc > 0);
1627 	return -rc;
1628 }
1629 
1630 static int
1631 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1632 			    struct rte_eth_udp_tunnel *tunnel_udp)
1633 {
1634 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1635 }
1636 
1637 static int
1638 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1639 			    struct rte_eth_udp_tunnel *tunnel_udp)
1640 {
1641 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1642 }
1643 
1644 /*
1645  * The function is used by the secondary process as well. It must not
1646  * use any process-local pointers from the adapter data.
1647  */
1648 static int
1649 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1650 			  struct rte_eth_rss_conf *rss_conf)
1651 {
1652 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1653 	struct sfc_rss *rss = &sas->rss;
1654 
1655 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1656 		return -ENOTSUP;
1657 
1658 	/*
1659 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1660 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1661 	 * flags which corresponds to the active EFX configuration stored
1662 	 * locally in 'sfc_adapter' and kept up-to-date
1663 	 */
1664 	rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1665 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1666 	if (rss_conf->rss_key != NULL)
1667 		rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1668 
1669 	return 0;
1670 }
1671 
1672 static int
1673 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1674 			struct rte_eth_rss_conf *rss_conf)
1675 {
1676 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1677 	struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1678 	unsigned int efx_hash_types;
1679 	uint32_t contexts[] = {EFX_RSS_CONTEXT_DEFAULT, rss->dummy_rss_context};
1680 	unsigned int n_contexts;
1681 	unsigned int mode_i = 0;
1682 	unsigned int key_i = 0;
1683 	unsigned int i = 0;
1684 	int rc = 0;
1685 
1686 	n_contexts = rss->dummy_rss_context == EFX_RSS_CONTEXT_DEFAULT ? 1 : 2;
1687 
1688 	if (sfc_sa2shared(sa)->isolated)
1689 		return -ENOTSUP;
1690 
1691 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1692 		sfc_err(sa, "RSS is not available");
1693 		return -ENOTSUP;
1694 	}
1695 
1696 	if (rss->channels == 0) {
1697 		sfc_err(sa, "RSS is not configured");
1698 		return -EINVAL;
1699 	}
1700 
1701 	if ((rss_conf->rss_key != NULL) &&
1702 	    (rss_conf->rss_key_len != sizeof(rss->key))) {
1703 		sfc_err(sa, "RSS key size is wrong (should be %zu)",
1704 			sizeof(rss->key));
1705 		return -EINVAL;
1706 	}
1707 
1708 	sfc_adapter_lock(sa);
1709 
1710 	rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1711 	if (rc != 0)
1712 		goto fail_rx_hf_rte_to_efx;
1713 
1714 	for (mode_i = 0; mode_i < n_contexts; mode_i++) {
1715 		rc = efx_rx_scale_mode_set(sa->nic, contexts[mode_i],
1716 					   rss->hash_alg, efx_hash_types,
1717 					   B_TRUE);
1718 		if (rc != 0)
1719 			goto fail_scale_mode_set;
1720 	}
1721 
1722 	if (rss_conf->rss_key != NULL) {
1723 		if (sa->state == SFC_ADAPTER_STARTED) {
1724 			for (key_i = 0; key_i < n_contexts; key_i++) {
1725 				rc = efx_rx_scale_key_set(sa->nic,
1726 							  contexts[key_i],
1727 							  rss_conf->rss_key,
1728 							  sizeof(rss->key));
1729 				if (rc != 0)
1730 					goto fail_scale_key_set;
1731 			}
1732 		}
1733 
1734 		rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1735 	}
1736 
1737 	rss->hash_types = efx_hash_types;
1738 
1739 	sfc_adapter_unlock(sa);
1740 
1741 	return 0;
1742 
1743 fail_scale_key_set:
1744 	for (i = 0; i < key_i; i++) {
1745 		if (efx_rx_scale_key_set(sa->nic, contexts[i], rss->key,
1746 					 sizeof(rss->key)) != 0)
1747 			sfc_err(sa, "failed to restore RSS key");
1748 	}
1749 
1750 fail_scale_mode_set:
1751 	for (i = 0; i < mode_i; i++) {
1752 		if (efx_rx_scale_mode_set(sa->nic, contexts[i],
1753 					  EFX_RX_HASHALG_TOEPLITZ,
1754 					  rss->hash_types, B_TRUE) != 0)
1755 			sfc_err(sa, "failed to restore RSS mode");
1756 	}
1757 
1758 fail_rx_hf_rte_to_efx:
1759 	sfc_adapter_unlock(sa);
1760 	return -rc;
1761 }
1762 
1763 /*
1764  * The function is used by the secondary process as well. It must not
1765  * use any process-local pointers from the adapter data.
1766  */
1767 static int
1768 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1769 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1770 		       uint16_t reta_size)
1771 {
1772 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1773 	struct sfc_rss *rss = &sas->rss;
1774 	int entry;
1775 
1776 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1777 		return -ENOTSUP;
1778 
1779 	if (rss->channels == 0)
1780 		return -EINVAL;
1781 
1782 	if (reta_size != EFX_RSS_TBL_SIZE)
1783 		return -EINVAL;
1784 
1785 	for (entry = 0; entry < reta_size; entry++) {
1786 		int grp = entry / RTE_RETA_GROUP_SIZE;
1787 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1788 
1789 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1790 			reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1791 	}
1792 
1793 	return 0;
1794 }
1795 
1796 static int
1797 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1798 			struct rte_eth_rss_reta_entry64 *reta_conf,
1799 			uint16_t reta_size)
1800 {
1801 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1802 	struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1803 	unsigned int *rss_tbl_new;
1804 	uint16_t entry;
1805 	int rc = 0;
1806 
1807 
1808 	if (sfc_sa2shared(sa)->isolated)
1809 		return -ENOTSUP;
1810 
1811 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1812 		sfc_err(sa, "RSS is not available");
1813 		return -ENOTSUP;
1814 	}
1815 
1816 	if (rss->channels == 0) {
1817 		sfc_err(sa, "RSS is not configured");
1818 		return -EINVAL;
1819 	}
1820 
1821 	if (reta_size != EFX_RSS_TBL_SIZE) {
1822 		sfc_err(sa, "RETA size is wrong (should be %u)",
1823 			EFX_RSS_TBL_SIZE);
1824 		return -EINVAL;
1825 	}
1826 
1827 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1828 	if (rss_tbl_new == NULL)
1829 		return -ENOMEM;
1830 
1831 	sfc_adapter_lock(sa);
1832 
1833 	rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1834 
1835 	for (entry = 0; entry < reta_size; entry++) {
1836 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1837 		struct rte_eth_rss_reta_entry64 *grp;
1838 
1839 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1840 
1841 		if (grp->mask & (1ull << grp_idx)) {
1842 			if (grp->reta[grp_idx] >= rss->channels) {
1843 				rc = EINVAL;
1844 				goto bad_reta_entry;
1845 			}
1846 			rss_tbl_new[entry] = grp->reta[grp_idx];
1847 		}
1848 	}
1849 
1850 	if (sa->state == SFC_ADAPTER_STARTED) {
1851 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1852 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1853 		if (rc != 0)
1854 			goto fail_scale_tbl_set;
1855 	}
1856 
1857 	rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1858 
1859 fail_scale_tbl_set:
1860 bad_reta_entry:
1861 	sfc_adapter_unlock(sa);
1862 
1863 	rte_free(rss_tbl_new);
1864 
1865 	SFC_ASSERT(rc >= 0);
1866 	return -rc;
1867 }
1868 
1869 static int
1870 sfc_dev_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
1871 		     const struct rte_flow_ops **ops)
1872 {
1873 	*ops = &sfc_flow_ops;
1874 	return 0;
1875 }
1876 
1877 static int
1878 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1879 {
1880 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1881 
1882 	/*
1883 	 * If Rx datapath does not provide callback to check mempool,
1884 	 * all pools are supported.
1885 	 */
1886 	if (sap->dp_rx->pool_ops_supported == NULL)
1887 		return 1;
1888 
1889 	return sap->dp_rx->pool_ops_supported(pool);
1890 }
1891 
1892 static int
1893 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1894 {
1895 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1896 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1897 	sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1898 	struct sfc_rxq_info *rxq_info;
1899 
1900 	rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1901 
1902 	return sap->dp_rx->intr_enable(rxq_info->dp);
1903 }
1904 
1905 static int
1906 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1907 {
1908 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1909 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1910 	sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1911 	struct sfc_rxq_info *rxq_info;
1912 
1913 	rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1914 
1915 	return sap->dp_rx->intr_disable(rxq_info->dp);
1916 }
1917 
1918 static const struct eth_dev_ops sfc_eth_dev_ops = {
1919 	.dev_configure			= sfc_dev_configure,
1920 	.dev_start			= sfc_dev_start,
1921 	.dev_stop			= sfc_dev_stop,
1922 	.dev_set_link_up		= sfc_dev_set_link_up,
1923 	.dev_set_link_down		= sfc_dev_set_link_down,
1924 	.dev_close			= sfc_dev_close,
1925 	.promiscuous_enable		= sfc_dev_promisc_enable,
1926 	.promiscuous_disable		= sfc_dev_promisc_disable,
1927 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1928 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1929 	.link_update			= sfc_dev_link_update,
1930 	.stats_get			= sfc_stats_get,
1931 	.stats_reset			= sfc_stats_reset,
1932 	.xstats_get			= sfc_xstats_get,
1933 	.xstats_reset			= sfc_stats_reset,
1934 	.xstats_get_names		= sfc_xstats_get_names,
1935 	.dev_infos_get			= sfc_dev_infos_get,
1936 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1937 	.mtu_set			= sfc_dev_set_mtu,
1938 	.rx_queue_start			= sfc_rx_queue_start,
1939 	.rx_queue_stop			= sfc_rx_queue_stop,
1940 	.tx_queue_start			= sfc_tx_queue_start,
1941 	.tx_queue_stop			= sfc_tx_queue_stop,
1942 	.rx_queue_setup			= sfc_rx_queue_setup,
1943 	.rx_queue_release		= sfc_rx_queue_release,
1944 	.rx_queue_intr_enable		= sfc_rx_queue_intr_enable,
1945 	.rx_queue_intr_disable		= sfc_rx_queue_intr_disable,
1946 	.tx_queue_setup			= sfc_tx_queue_setup,
1947 	.tx_queue_release		= sfc_tx_queue_release,
1948 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1949 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1950 	.mac_addr_set			= sfc_mac_addr_set,
1951 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1952 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
1953 	.reta_update			= sfc_dev_rss_reta_update,
1954 	.reta_query			= sfc_dev_rss_reta_query,
1955 	.rss_hash_update		= sfc_dev_rss_hash_update,
1956 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1957 	.flow_ops_get			= sfc_dev_flow_ops_get,
1958 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1959 	.rxq_info_get			= sfc_rx_queue_info_get,
1960 	.txq_info_get			= sfc_tx_queue_info_get,
1961 	.fw_version_get			= sfc_fw_version_get,
1962 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1963 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1964 	.pool_ops_supported		= sfc_pool_ops_supported,
1965 };
1966 
1967 /**
1968  * Duplicate a string in potentially shared memory required for
1969  * multi-process support.
1970  *
1971  * strdup() allocates from process-local heap/memory.
1972  */
1973 static char *
1974 sfc_strdup(const char *str)
1975 {
1976 	size_t size;
1977 	char *copy;
1978 
1979 	if (str == NULL)
1980 		return NULL;
1981 
1982 	size = strlen(str) + 1;
1983 	copy = rte_malloc(__func__, size, 0);
1984 	if (copy != NULL)
1985 		rte_memcpy(copy, str, size);
1986 
1987 	return copy;
1988 }
1989 
1990 static int
1991 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1992 {
1993 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1994 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1995 	const struct sfc_dp_rx *dp_rx;
1996 	const struct sfc_dp_tx *dp_tx;
1997 	const efx_nic_cfg_t *encp;
1998 	unsigned int avail_caps = 0;
1999 	const char *rx_name = NULL;
2000 	const char *tx_name = NULL;
2001 	int rc;
2002 
2003 	switch (sa->family) {
2004 	case EFX_FAMILY_HUNTINGTON:
2005 	case EFX_FAMILY_MEDFORD:
2006 	case EFX_FAMILY_MEDFORD2:
2007 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
2008 		avail_caps |= SFC_DP_HW_FW_CAP_RX_EFX;
2009 		avail_caps |= SFC_DP_HW_FW_CAP_TX_EFX;
2010 		break;
2011 	case EFX_FAMILY_RIVERHEAD:
2012 		avail_caps |= SFC_DP_HW_FW_CAP_EF100;
2013 		break;
2014 	default:
2015 		break;
2016 	}
2017 
2018 	encp = efx_nic_cfg_get(sa->nic);
2019 	if (encp->enc_rx_es_super_buffer_supported)
2020 		avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
2021 
2022 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
2023 				sfc_kvarg_string_handler, &rx_name);
2024 	if (rc != 0)
2025 		goto fail_kvarg_rx_datapath;
2026 
2027 	if (rx_name != NULL) {
2028 		dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
2029 		if (dp_rx == NULL) {
2030 			sfc_err(sa, "Rx datapath %s not found", rx_name);
2031 			rc = ENOENT;
2032 			goto fail_dp_rx;
2033 		}
2034 		if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
2035 			sfc_err(sa,
2036 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
2037 				rx_name);
2038 			rc = EINVAL;
2039 			goto fail_dp_rx_caps;
2040 		}
2041 	} else {
2042 		dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
2043 		if (dp_rx == NULL) {
2044 			sfc_err(sa, "Rx datapath by caps %#x not found",
2045 				avail_caps);
2046 			rc = ENOENT;
2047 			goto fail_dp_rx;
2048 		}
2049 	}
2050 
2051 	sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
2052 	if (sas->dp_rx_name == NULL) {
2053 		rc = ENOMEM;
2054 		goto fail_dp_rx_name;
2055 	}
2056 
2057 	sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
2058 
2059 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
2060 				sfc_kvarg_string_handler, &tx_name);
2061 	if (rc != 0)
2062 		goto fail_kvarg_tx_datapath;
2063 
2064 	if (tx_name != NULL) {
2065 		dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
2066 		if (dp_tx == NULL) {
2067 			sfc_err(sa, "Tx datapath %s not found", tx_name);
2068 			rc = ENOENT;
2069 			goto fail_dp_tx;
2070 		}
2071 		if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
2072 			sfc_err(sa,
2073 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
2074 				tx_name);
2075 			rc = EINVAL;
2076 			goto fail_dp_tx_caps;
2077 		}
2078 	} else {
2079 		dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
2080 		if (dp_tx == NULL) {
2081 			sfc_err(sa, "Tx datapath by caps %#x not found",
2082 				avail_caps);
2083 			rc = ENOENT;
2084 			goto fail_dp_tx;
2085 		}
2086 	}
2087 
2088 	sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
2089 	if (sas->dp_tx_name == NULL) {
2090 		rc = ENOMEM;
2091 		goto fail_dp_tx_name;
2092 	}
2093 
2094 	sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
2095 
2096 	sa->priv.dp_rx = dp_rx;
2097 	sa->priv.dp_tx = dp_tx;
2098 
2099 	dev->rx_pkt_burst = dp_rx->pkt_burst;
2100 	dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2101 	dev->tx_pkt_burst = dp_tx->pkt_burst;
2102 
2103 	dev->rx_queue_count = sfc_rx_queue_count;
2104 	dev->rx_descriptor_status = sfc_rx_descriptor_status;
2105 	dev->tx_descriptor_status = sfc_tx_descriptor_status;
2106 	dev->dev_ops = &sfc_eth_dev_ops;
2107 
2108 	return 0;
2109 
2110 fail_dp_tx_name:
2111 fail_dp_tx_caps:
2112 fail_dp_tx:
2113 fail_kvarg_tx_datapath:
2114 	rte_free(sas->dp_rx_name);
2115 	sas->dp_rx_name = NULL;
2116 
2117 fail_dp_rx_name:
2118 fail_dp_rx_caps:
2119 fail_dp_rx:
2120 fail_kvarg_rx_datapath:
2121 	return rc;
2122 }
2123 
2124 static void
2125 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
2126 {
2127 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2128 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2129 
2130 	dev->dev_ops = NULL;
2131 	dev->tx_pkt_prepare = NULL;
2132 	dev->rx_pkt_burst = NULL;
2133 	dev->tx_pkt_burst = NULL;
2134 
2135 	rte_free(sas->dp_tx_name);
2136 	sas->dp_tx_name = NULL;
2137 	sa->priv.dp_tx = NULL;
2138 
2139 	rte_free(sas->dp_rx_name);
2140 	sas->dp_rx_name = NULL;
2141 	sa->priv.dp_rx = NULL;
2142 }
2143 
2144 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2145 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
2146 	.reta_query			= sfc_dev_rss_reta_query,
2147 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
2148 	.rxq_info_get			= sfc_rx_queue_info_get,
2149 	.txq_info_get			= sfc_tx_queue_info_get,
2150 };
2151 
2152 static int
2153 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2154 {
2155 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2156 	struct sfc_adapter_priv *sap;
2157 	const struct sfc_dp_rx *dp_rx;
2158 	const struct sfc_dp_tx *dp_tx;
2159 	int rc;
2160 
2161 	/*
2162 	 * Allocate process private data from heap, since it should not
2163 	 * be located in shared memory allocated using rte_malloc() API.
2164 	 */
2165 	sap = calloc(1, sizeof(*sap));
2166 	if (sap == NULL) {
2167 		rc = ENOMEM;
2168 		goto fail_alloc_priv;
2169 	}
2170 
2171 	sap->logtype_main = logtype_main;
2172 
2173 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2174 	if (dp_rx == NULL) {
2175 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2176 			"cannot find %s Rx datapath", sas->dp_rx_name);
2177 		rc = ENOENT;
2178 		goto fail_dp_rx;
2179 	}
2180 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2181 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2182 			"%s Rx datapath does not support multi-process",
2183 			sas->dp_rx_name);
2184 		rc = EINVAL;
2185 		goto fail_dp_rx_multi_process;
2186 	}
2187 
2188 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2189 	if (dp_tx == NULL) {
2190 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2191 			"cannot find %s Tx datapath", sas->dp_tx_name);
2192 		rc = ENOENT;
2193 		goto fail_dp_tx;
2194 	}
2195 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2196 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2197 			"%s Tx datapath does not support multi-process",
2198 			sas->dp_tx_name);
2199 		rc = EINVAL;
2200 		goto fail_dp_tx_multi_process;
2201 	}
2202 
2203 	sap->dp_rx = dp_rx;
2204 	sap->dp_tx = dp_tx;
2205 
2206 	dev->process_private = sap;
2207 	dev->rx_pkt_burst = dp_rx->pkt_burst;
2208 	dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2209 	dev->tx_pkt_burst = dp_tx->pkt_burst;
2210 	dev->rx_queue_count = sfc_rx_queue_count;
2211 	dev->rx_descriptor_status = sfc_rx_descriptor_status;
2212 	dev->tx_descriptor_status = sfc_tx_descriptor_status;
2213 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
2214 
2215 	return 0;
2216 
2217 fail_dp_tx_multi_process:
2218 fail_dp_tx:
2219 fail_dp_rx_multi_process:
2220 fail_dp_rx:
2221 	free(sap);
2222 
2223 fail_alloc_priv:
2224 	return rc;
2225 }
2226 
2227 static void
2228 sfc_register_dp(void)
2229 {
2230 	/* Register once */
2231 	if (TAILQ_EMPTY(&sfc_dp_head)) {
2232 		/* Prefer EF10 datapath */
2233 		sfc_dp_register(&sfc_dp_head, &sfc_ef100_rx.dp);
2234 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2235 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2236 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2237 
2238 		sfc_dp_register(&sfc_dp_head, &sfc_ef100_tx.dp);
2239 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2240 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2241 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2242 	}
2243 }
2244 
2245 static int
2246 sfc_parse_switch_mode(struct sfc_adapter *sa)
2247 {
2248 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
2249 	const char *switch_mode = NULL;
2250 	int rc;
2251 
2252 	sfc_log_init(sa, "entry");
2253 
2254 	rc = sfc_kvargs_process(sa, SFC_KVARG_SWITCH_MODE,
2255 				sfc_kvarg_string_handler, &switch_mode);
2256 	if (rc != 0)
2257 		goto fail_kvargs;
2258 
2259 	if (switch_mode == NULL) {
2260 		sa->switchdev = encp->enc_mae_supported &&
2261 				!encp->enc_datapath_cap_evb;
2262 	} else if (strcasecmp(switch_mode, SFC_KVARG_SWITCH_MODE_LEGACY) == 0) {
2263 		sa->switchdev = false;
2264 	} else if (strcasecmp(switch_mode,
2265 			      SFC_KVARG_SWITCH_MODE_SWITCHDEV) == 0) {
2266 		sa->switchdev = true;
2267 	} else {
2268 		sfc_err(sa, "invalid switch mode device argument '%s'",
2269 			switch_mode);
2270 		rc = EINVAL;
2271 		goto fail_mode;
2272 	}
2273 
2274 	sfc_log_init(sa, "done");
2275 
2276 	return 0;
2277 
2278 fail_mode:
2279 fail_kvargs:
2280 	sfc_log_init(sa, "failed: %s", rte_strerror(rc));
2281 
2282 	return rc;
2283 }
2284 
2285 static int
2286 sfc_eth_dev_init(struct rte_eth_dev *dev)
2287 {
2288 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2289 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2290 	uint32_t logtype_main;
2291 	struct sfc_adapter *sa;
2292 	int rc;
2293 	const efx_nic_cfg_t *encp;
2294 	const struct rte_ether_addr *from;
2295 	int ret;
2296 
2297 	if (sfc_efx_dev_class_get(pci_dev->device.devargs) !=
2298 			SFC_EFX_DEV_CLASS_NET) {
2299 		SFC_GENERIC_LOG(DEBUG,
2300 			"Incompatible device class: skip probing, should be probed by other sfc driver.");
2301 		return 1;
2302 	}
2303 
2304 	rc = sfc_dp_mport_register();
2305 	if (rc != 0)
2306 		return rc;
2307 
2308 	sfc_register_dp();
2309 
2310 	logtype_main = sfc_register_logtype(&pci_dev->addr,
2311 					    SFC_LOGTYPE_MAIN_STR,
2312 					    RTE_LOG_NOTICE);
2313 
2314 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2315 		return -sfc_eth_dev_secondary_init(dev, logtype_main);
2316 
2317 	/* Required for logging */
2318 	ret = snprintf(sas->log_prefix, sizeof(sas->log_prefix),
2319 			"PMD: sfc_efx " PCI_PRI_FMT " #%" PRIu16 ": ",
2320 			pci_dev->addr.domain, pci_dev->addr.bus,
2321 			pci_dev->addr.devid, pci_dev->addr.function,
2322 			dev->data->port_id);
2323 	if (ret < 0 || ret >= (int)sizeof(sas->log_prefix)) {
2324 		SFC_GENERIC_LOG(ERR,
2325 			"reserved log prefix is too short for " PCI_PRI_FMT,
2326 			pci_dev->addr.domain, pci_dev->addr.bus,
2327 			pci_dev->addr.devid, pci_dev->addr.function);
2328 		return -EINVAL;
2329 	}
2330 	sas->pci_addr = pci_dev->addr;
2331 	sas->port_id = dev->data->port_id;
2332 
2333 	/*
2334 	 * Allocate process private data from heap, since it should not
2335 	 * be located in shared memory allocated using rte_malloc() API.
2336 	 */
2337 	sa = calloc(1, sizeof(*sa));
2338 	if (sa == NULL) {
2339 		rc = ENOMEM;
2340 		goto fail_alloc_sa;
2341 	}
2342 
2343 	dev->process_private = sa;
2344 
2345 	/* Required for logging */
2346 	sa->priv.shared = sas;
2347 	sa->priv.logtype_main = logtype_main;
2348 
2349 	sa->eth_dev = dev;
2350 
2351 	/* Copy PCI device info to the dev->data */
2352 	rte_eth_copy_pci_info(dev, pci_dev);
2353 	dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
2354 	dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE;
2355 
2356 	rc = sfc_kvargs_parse(sa);
2357 	if (rc != 0)
2358 		goto fail_kvargs_parse;
2359 
2360 	sfc_log_init(sa, "entry");
2361 
2362 	dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2363 	if (dev->data->mac_addrs == NULL) {
2364 		rc = ENOMEM;
2365 		goto fail_mac_addrs;
2366 	}
2367 
2368 	sfc_adapter_lock_init(sa);
2369 	sfc_adapter_lock(sa);
2370 
2371 	sfc_log_init(sa, "probing");
2372 	rc = sfc_probe(sa);
2373 	if (rc != 0)
2374 		goto fail_probe;
2375 
2376 	/*
2377 	 * Selecting a default switch mode requires the NIC to be probed and
2378 	 * to have its capabilities filled in.
2379 	 */
2380 	rc = sfc_parse_switch_mode(sa);
2381 	if (rc != 0)
2382 		goto fail_switch_mode;
2383 
2384 	sfc_log_init(sa, "set device ops");
2385 	rc = sfc_eth_dev_set_ops(dev);
2386 	if (rc != 0)
2387 		goto fail_set_ops;
2388 
2389 	sfc_log_init(sa, "attaching");
2390 	rc = sfc_attach(sa);
2391 	if (rc != 0)
2392 		goto fail_attach;
2393 
2394 	if (sa->switchdev && sa->mae.status != SFC_MAE_STATUS_SUPPORTED) {
2395 		sfc_err(sa,
2396 			"failed to enable switchdev mode without MAE support");
2397 		rc = ENOTSUP;
2398 		goto fail_switchdev_no_mae;
2399 	}
2400 
2401 	encp = efx_nic_cfg_get(sa->nic);
2402 
2403 	/*
2404 	 * The arguments are really reverse order in comparison to
2405 	 * Linux kernel. Copy from NIC config to Ethernet device data.
2406 	 */
2407 	from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2408 	rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2409 
2410 	sfc_adapter_unlock(sa);
2411 
2412 	sfc_log_init(sa, "done");
2413 	return 0;
2414 
2415 fail_switchdev_no_mae:
2416 	sfc_detach(sa);
2417 
2418 fail_attach:
2419 	sfc_eth_dev_clear_ops(dev);
2420 
2421 fail_set_ops:
2422 fail_switch_mode:
2423 	sfc_unprobe(sa);
2424 
2425 fail_probe:
2426 	sfc_adapter_unlock(sa);
2427 	sfc_adapter_lock_fini(sa);
2428 	rte_free(dev->data->mac_addrs);
2429 	dev->data->mac_addrs = NULL;
2430 
2431 fail_mac_addrs:
2432 	sfc_kvargs_cleanup(sa);
2433 
2434 fail_kvargs_parse:
2435 	sfc_log_init(sa, "failed %d", rc);
2436 	dev->process_private = NULL;
2437 	free(sa);
2438 
2439 fail_alloc_sa:
2440 	SFC_ASSERT(rc > 0);
2441 	return -rc;
2442 }
2443 
2444 static int
2445 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2446 {
2447 	sfc_dev_close(dev);
2448 
2449 	return 0;
2450 }
2451 
2452 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2453 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2454 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2455 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2456 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2457 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2458 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2459 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2460 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2461 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD) },
2462 	{ .vendor_id = 0 /* sentinel */ }
2463 };
2464 
2465 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2466 	struct rte_pci_device *pci_dev)
2467 {
2468 	return rte_eth_dev_pci_generic_probe(pci_dev,
2469 		sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
2470 }
2471 
2472 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2473 {
2474 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2475 }
2476 
2477 static struct rte_pci_driver sfc_efx_pmd = {
2478 	.id_table = pci_id_sfc_efx_map,
2479 	.drv_flags =
2480 		RTE_PCI_DRV_INTR_LSC |
2481 		RTE_PCI_DRV_NEED_MAPPING,
2482 	.probe = sfc_eth_dev_pci_probe,
2483 	.remove = sfc_eth_dev_pci_remove,
2484 };
2485 
2486 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2487 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2488 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2489 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2490 	SFC_KVARG_SWITCH_MODE "=" SFC_KVARG_VALUES_SWITCH_MODE " "
2491 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2492 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2493 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2494 	SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2495 	SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2496 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2497 
2498 RTE_INIT(sfc_driver_register_logtype)
2499 {
2500 	int ret;
2501 
2502 	ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2503 						   RTE_LOG_NOTICE);
2504 	sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2505 }
2506