xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision a720e6740afcd6d21955d188c6e6a7136fcc7d6c)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  *
3  * Copyright (c) 2016-2018 Solarflare Communications Inc.
4  * All rights reserved.
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 <rte_ethdev_driver.h>
12 #include <rte_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 
18 #include "efx.h"
19 
20 #include "sfc.h"
21 #include "sfc_debug.h"
22 #include "sfc_log.h"
23 #include "sfc_kvargs.h"
24 #include "sfc_ev.h"
25 #include "sfc_rx.h"
26 #include "sfc_tx.h"
27 #include "sfc_flow.h"
28 #include "sfc_dp.h"
29 #include "sfc_dp_rx.h"
30 
31 uint32_t sfc_logtype_driver;
32 
33 static struct sfc_dp_list sfc_dp_head =
34 	TAILQ_HEAD_INITIALIZER(sfc_dp_head);
35 
36 static int
37 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
38 {
39 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
40 	efx_nic_fw_info_t enfi;
41 	int ret;
42 	int rc;
43 
44 	/*
45 	 * Return value of the callback is likely supposed to be
46 	 * equal to or greater than 0, nevertheless, if an error
47 	 * occurs, it will be desirable to pass it to the caller
48 	 */
49 	if ((fw_version == NULL) || (fw_size == 0))
50 		return -EINVAL;
51 
52 	rc = efx_nic_get_fw_version(sa->nic, &enfi);
53 	if (rc != 0)
54 		return -rc;
55 
56 	ret = snprintf(fw_version, fw_size,
57 		       "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
58 		       enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
59 		       enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
60 	if (ret < 0)
61 		return ret;
62 
63 	if (enfi.enfi_dpcpu_fw_ids_valid) {
64 		size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
65 		int ret_extra;
66 
67 		ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
68 				     fw_size - dpcpu_fw_ids_offset,
69 				     " rx%" PRIx16 " tx%" PRIx16,
70 				     enfi.enfi_rx_dpcpu_fw_id,
71 				     enfi.enfi_tx_dpcpu_fw_id);
72 		if (ret_extra < 0)
73 			return ret_extra;
74 
75 		ret += ret_extra;
76 	}
77 
78 	if (fw_size < (size_t)(++ret))
79 		return ret;
80 	else
81 		return 0;
82 }
83 
84 static void
85 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
86 {
87 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
88 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
89 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
90 	struct sfc_rss *rss = &sas->rss;
91 	uint64_t txq_offloads_def = 0;
92 
93 	sfc_log_init(sa, "entry");
94 
95 	dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
96 
97 	/* Autonegotiation may be disabled */
98 	dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
99 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_1000FDX))
100 		dev_info->speed_capa |= ETH_LINK_SPEED_1G;
101 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_10000FDX))
102 		dev_info->speed_capa |= ETH_LINK_SPEED_10G;
103 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_25000FDX))
104 		dev_info->speed_capa |= ETH_LINK_SPEED_25G;
105 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_40000FDX))
106 		dev_info->speed_capa |= ETH_LINK_SPEED_40G;
107 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_50000FDX))
108 		dev_info->speed_capa |= ETH_LINK_SPEED_50G;
109 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_100000FDX))
110 		dev_info->speed_capa |= ETH_LINK_SPEED_100G;
111 
112 	dev_info->max_rx_queues = sa->rxq_max;
113 	dev_info->max_tx_queues = sa->txq_max;
114 
115 	/* By default packets are dropped if no descriptors are available */
116 	dev_info->default_rxconf.rx_drop_en = 1;
117 
118 	dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
119 
120 	/*
121 	 * rx_offload_capa includes both device and queue offloads since
122 	 * the latter may be requested on a per device basis which makes
123 	 * sense when some offloads are needed to be set on all queues.
124 	 */
125 	dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
126 				    dev_info->rx_queue_offload_capa;
127 
128 	dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
129 
130 	/*
131 	 * tx_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->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
136 				    dev_info->tx_queue_offload_capa;
137 
138 	if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
139 		txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
140 
141 	dev_info->default_txconf.offloads |= txq_offloads_def;
142 
143 	if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
144 		uint64_t rte_hf = 0;
145 		unsigned int i;
146 
147 		for (i = 0; i < rss->hf_map_nb_entries; ++i)
148 			rte_hf |= rss->hf_map[i].rte;
149 
150 		dev_info->reta_size = EFX_RSS_TBL_SIZE;
151 		dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
152 		dev_info->flow_type_rss_offloads = rte_hf;
153 	}
154 
155 	/* Initialize to hardware limits */
156 	dev_info->rx_desc_lim.nb_max = sa->rxq_max_entries;
157 	dev_info->rx_desc_lim.nb_min = sa->rxq_min_entries;
158 	/* The RXQ hardware requires that the descriptor count is a power
159 	 * of 2, but rx_desc_lim cannot properly describe that constraint.
160 	 */
161 	dev_info->rx_desc_lim.nb_align = sa->rxq_min_entries;
162 
163 	/* Initialize to hardware limits */
164 	dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
165 	dev_info->tx_desc_lim.nb_min = sa->txq_min_entries;
166 	/*
167 	 * The TXQ hardware requires that the descriptor count is a power
168 	 * of 2, but tx_desc_lim cannot properly describe that constraint
169 	 */
170 	dev_info->tx_desc_lim.nb_align = sa->txq_min_entries;
171 
172 	if (sap->dp_rx->get_dev_info != NULL)
173 		sap->dp_rx->get_dev_info(dev_info);
174 	if (sap->dp_tx->get_dev_info != NULL)
175 		sap->dp_tx->get_dev_info(dev_info);
176 
177 	dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
178 			     RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
179 }
180 
181 static const uint32_t *
182 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
183 {
184 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
185 
186 	return sap->dp_rx->supported_ptypes_get(sap->shared->tunnel_encaps);
187 }
188 
189 static int
190 sfc_dev_configure(struct rte_eth_dev *dev)
191 {
192 	struct rte_eth_dev_data *dev_data = dev->data;
193 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
194 	int rc;
195 
196 	sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
197 		     dev_data->nb_rx_queues, dev_data->nb_tx_queues);
198 
199 	sfc_adapter_lock(sa);
200 	switch (sa->state) {
201 	case SFC_ADAPTER_CONFIGURED:
202 		/* FALLTHROUGH */
203 	case SFC_ADAPTER_INITIALIZED:
204 		rc = sfc_configure(sa);
205 		break;
206 	default:
207 		sfc_err(sa, "unexpected adapter state %u to configure",
208 			sa->state);
209 		rc = EINVAL;
210 		break;
211 	}
212 	sfc_adapter_unlock(sa);
213 
214 	sfc_log_init(sa, "done %d", rc);
215 	SFC_ASSERT(rc >= 0);
216 	return -rc;
217 }
218 
219 static int
220 sfc_dev_start(struct rte_eth_dev *dev)
221 {
222 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
223 	int rc;
224 
225 	sfc_log_init(sa, "entry");
226 
227 	sfc_adapter_lock(sa);
228 	rc = sfc_start(sa);
229 	sfc_adapter_unlock(sa);
230 
231 	sfc_log_init(sa, "done %d", rc);
232 	SFC_ASSERT(rc >= 0);
233 	return -rc;
234 }
235 
236 static int
237 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
238 {
239 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
240 	struct rte_eth_link current_link;
241 	int ret;
242 
243 	sfc_log_init(sa, "entry");
244 
245 	if (sa->state != SFC_ADAPTER_STARTED) {
246 		sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
247 	} else if (wait_to_complete) {
248 		efx_link_mode_t link_mode;
249 
250 		if (efx_port_poll(sa->nic, &link_mode) != 0)
251 			link_mode = EFX_LINK_UNKNOWN;
252 		sfc_port_link_mode_to_info(link_mode, &current_link);
253 
254 	} else {
255 		sfc_ev_mgmt_qpoll(sa);
256 		rte_eth_linkstatus_get(dev, &current_link);
257 	}
258 
259 	ret = rte_eth_linkstatus_set(dev, &current_link);
260 	if (ret == 0)
261 		sfc_notice(sa, "Link status is %s",
262 			   current_link.link_status ? "UP" : "DOWN");
263 
264 	return ret;
265 }
266 
267 static void
268 sfc_dev_stop(struct rte_eth_dev *dev)
269 {
270 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
271 
272 	sfc_log_init(sa, "entry");
273 
274 	sfc_adapter_lock(sa);
275 	sfc_stop(sa);
276 	sfc_adapter_unlock(sa);
277 
278 	sfc_log_init(sa, "done");
279 }
280 
281 static int
282 sfc_dev_set_link_up(struct rte_eth_dev *dev)
283 {
284 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
285 	int rc;
286 
287 	sfc_log_init(sa, "entry");
288 
289 	sfc_adapter_lock(sa);
290 	rc = sfc_start(sa);
291 	sfc_adapter_unlock(sa);
292 
293 	SFC_ASSERT(rc >= 0);
294 	return -rc;
295 }
296 
297 static int
298 sfc_dev_set_link_down(struct rte_eth_dev *dev)
299 {
300 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
301 
302 	sfc_log_init(sa, "entry");
303 
304 	sfc_adapter_lock(sa);
305 	sfc_stop(sa);
306 	sfc_adapter_unlock(sa);
307 
308 	return 0;
309 }
310 
311 static void
312 sfc_dev_close(struct rte_eth_dev *dev)
313 {
314 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
315 
316 	sfc_log_init(sa, "entry");
317 
318 	sfc_adapter_lock(sa);
319 	switch (sa->state) {
320 	case SFC_ADAPTER_STARTED:
321 		sfc_stop(sa);
322 		SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
323 		/* FALLTHROUGH */
324 	case SFC_ADAPTER_CONFIGURED:
325 		sfc_close(sa);
326 		SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
327 		/* FALLTHROUGH */
328 	case SFC_ADAPTER_INITIALIZED:
329 		break;
330 	default:
331 		sfc_err(sa, "unexpected adapter state %u on close", sa->state);
332 		break;
333 	}
334 	sfc_adapter_unlock(sa);
335 
336 	sfc_log_init(sa, "done");
337 }
338 
339 static void
340 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
341 		   boolean_t enabled)
342 {
343 	struct sfc_port *port;
344 	boolean_t *toggle;
345 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
346 	boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
347 	const char *desc = (allmulti) ? "all-multi" : "promiscuous";
348 
349 	sfc_adapter_lock(sa);
350 
351 	port = &sa->port;
352 	toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
353 
354 	if (*toggle != enabled) {
355 		*toggle = enabled;
356 
357 		if (sfc_sa2shared(sa)->isolated) {
358 			sfc_warn(sa, "isolated mode is active on the port");
359 			sfc_warn(sa, "the change is to be applied on the next "
360 				     "start provided that isolated mode is "
361 				     "disabled prior the next start");
362 		} else if ((sa->state == SFC_ADAPTER_STARTED) &&
363 			   (sfc_set_rx_mode(sa) != 0)) {
364 			*toggle = !(enabled);
365 			sfc_warn(sa, "Failed to %s %s mode",
366 				 ((enabled) ? "enable" : "disable"), desc);
367 		}
368 	}
369 
370 	sfc_adapter_unlock(sa);
371 }
372 
373 static void
374 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
375 {
376 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
377 }
378 
379 static void
380 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
381 {
382 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
383 }
384 
385 static void
386 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
387 {
388 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
389 }
390 
391 static void
392 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
393 {
394 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
395 }
396 
397 static int
398 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
399 		   uint16_t nb_rx_desc, unsigned int socket_id,
400 		   const struct rte_eth_rxconf *rx_conf,
401 		   struct rte_mempool *mb_pool)
402 {
403 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
404 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
405 	int rc;
406 
407 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
408 		     rx_queue_id, nb_rx_desc, socket_id);
409 
410 	sfc_adapter_lock(sa);
411 
412 	rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
413 			  rx_conf, mb_pool);
414 	if (rc != 0)
415 		goto fail_rx_qinit;
416 
417 	dev->data->rx_queues[rx_queue_id] = sas->rxq_info[rx_queue_id].dp;
418 
419 	sfc_adapter_unlock(sa);
420 
421 	return 0;
422 
423 fail_rx_qinit:
424 	sfc_adapter_unlock(sa);
425 	SFC_ASSERT(rc > 0);
426 	return -rc;
427 }
428 
429 static void
430 sfc_rx_queue_release(void *queue)
431 {
432 	struct sfc_dp_rxq *dp_rxq = queue;
433 	struct sfc_rxq *rxq;
434 	struct sfc_adapter *sa;
435 	unsigned int sw_index;
436 
437 	if (dp_rxq == NULL)
438 		return;
439 
440 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
441 	sa = rxq->evq->sa;
442 	sfc_adapter_lock(sa);
443 
444 	sw_index = dp_rxq->dpq.queue_id;
445 
446 	sfc_log_init(sa, "RxQ=%u", sw_index);
447 
448 	sfc_rx_qfini(sa, sw_index);
449 
450 	sfc_adapter_unlock(sa);
451 }
452 
453 static int
454 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
455 		   uint16_t nb_tx_desc, unsigned int socket_id,
456 		   const struct rte_eth_txconf *tx_conf)
457 {
458 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
459 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
460 	int rc;
461 
462 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
463 		     tx_queue_id, nb_tx_desc, socket_id);
464 
465 	sfc_adapter_lock(sa);
466 
467 	rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
468 	if (rc != 0)
469 		goto fail_tx_qinit;
470 
471 	dev->data->tx_queues[tx_queue_id] = sas->txq_info[tx_queue_id].dp;
472 
473 	sfc_adapter_unlock(sa);
474 	return 0;
475 
476 fail_tx_qinit:
477 	sfc_adapter_unlock(sa);
478 	SFC_ASSERT(rc > 0);
479 	return -rc;
480 }
481 
482 static void
483 sfc_tx_queue_release(void *queue)
484 {
485 	struct sfc_dp_txq *dp_txq = queue;
486 	struct sfc_txq *txq;
487 	unsigned int sw_index;
488 	struct sfc_adapter *sa;
489 
490 	if (dp_txq == NULL)
491 		return;
492 
493 	txq = sfc_txq_by_dp_txq(dp_txq);
494 	sw_index = dp_txq->dpq.queue_id;
495 
496 	SFC_ASSERT(txq->evq != NULL);
497 	sa = txq->evq->sa;
498 
499 	sfc_log_init(sa, "TxQ = %u", sw_index);
500 
501 	sfc_adapter_lock(sa);
502 
503 	sfc_tx_qfini(sa, sw_index);
504 
505 	sfc_adapter_unlock(sa);
506 }
507 
508 /*
509  * Some statistics are computed as A - B where A and B each increase
510  * monotonically with some hardware counter(s) and the counters are read
511  * asynchronously.
512  *
513  * If packet X is counted in A, but not counted in B yet, computed value is
514  * greater than real.
515  *
516  * If packet X is not counted in A at the moment of reading the counter,
517  * but counted in B at the moment of reading the counter, computed value
518  * is less than real.
519  *
520  * However, counter which grows backward is worse evil than slightly wrong
521  * value. So, let's try to guarantee that it never happens except may be
522  * the case when the MAC stats are zeroed as a result of a NIC reset.
523  */
524 static void
525 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
526 {
527 	if ((int64_t)(newval - *stat) > 0 || newval == 0)
528 		*stat = newval;
529 }
530 
531 static int
532 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
533 {
534 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
535 	struct sfc_port *port = &sa->port;
536 	uint64_t *mac_stats;
537 	int ret;
538 
539 	rte_spinlock_lock(&port->mac_stats_lock);
540 
541 	ret = sfc_port_update_mac_stats(sa);
542 	if (ret != 0)
543 		goto unlock;
544 
545 	mac_stats = port->mac_stats_buf;
546 
547 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
548 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
549 		stats->ipackets =
550 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
551 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
552 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
553 		stats->opackets =
554 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
555 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
556 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
557 		stats->ibytes =
558 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
559 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
560 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
561 		stats->obytes =
562 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
563 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
564 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
565 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
566 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
567 	} else {
568 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
569 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
570 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
571 		/*
572 		 * Take into account stats which are whenever supported
573 		 * on EF10. If some stat is not supported by current
574 		 * firmware variant or HW revision, it is guaranteed
575 		 * to be zero in mac_stats.
576 		 */
577 		stats->imissed =
578 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
579 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
580 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
581 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
582 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
583 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
584 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
585 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
586 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
587 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
588 		stats->ierrors =
589 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
590 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
591 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
592 		/* no oerrors counters supported on EF10 */
593 
594 		/* Exclude missed, errors and pauses from Rx packets */
595 		sfc_update_diff_stat(&port->ipackets,
596 			mac_stats[EFX_MAC_RX_PKTS] -
597 			mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
598 			stats->imissed - stats->ierrors);
599 		stats->ipackets = port->ipackets;
600 	}
601 
602 unlock:
603 	rte_spinlock_unlock(&port->mac_stats_lock);
604 	SFC_ASSERT(ret >= 0);
605 	return -ret;
606 }
607 
608 static void
609 sfc_stats_reset(struct rte_eth_dev *dev)
610 {
611 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
612 	struct sfc_port *port = &sa->port;
613 	int rc;
614 
615 	if (sa->state != SFC_ADAPTER_STARTED) {
616 		/*
617 		 * The operation cannot be done if port is not started; it
618 		 * will be scheduled to be done during the next port start
619 		 */
620 		port->mac_stats_reset_pending = B_TRUE;
621 		return;
622 	}
623 
624 	rc = sfc_port_reset_mac_stats(sa);
625 	if (rc != 0)
626 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
627 }
628 
629 static int
630 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
631 	       unsigned int xstats_count)
632 {
633 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
634 	struct sfc_port *port = &sa->port;
635 	uint64_t *mac_stats;
636 	int rc;
637 	unsigned int i;
638 	int nstats = 0;
639 
640 	rte_spinlock_lock(&port->mac_stats_lock);
641 
642 	rc = sfc_port_update_mac_stats(sa);
643 	if (rc != 0) {
644 		SFC_ASSERT(rc > 0);
645 		nstats = -rc;
646 		goto unlock;
647 	}
648 
649 	mac_stats = port->mac_stats_buf;
650 
651 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
652 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
653 			if (xstats != NULL && nstats < (int)xstats_count) {
654 				xstats[nstats].id = nstats;
655 				xstats[nstats].value = mac_stats[i];
656 			}
657 			nstats++;
658 		}
659 	}
660 
661 unlock:
662 	rte_spinlock_unlock(&port->mac_stats_lock);
663 
664 	return nstats;
665 }
666 
667 static int
668 sfc_xstats_get_names(struct rte_eth_dev *dev,
669 		     struct rte_eth_xstat_name *xstats_names,
670 		     unsigned int xstats_count)
671 {
672 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
673 	struct sfc_port *port = &sa->port;
674 	unsigned int i;
675 	unsigned int nstats = 0;
676 
677 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
678 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
679 			if (xstats_names != NULL && nstats < xstats_count)
680 				strlcpy(xstats_names[nstats].name,
681 					efx_mac_stat_name(sa->nic, i),
682 					sizeof(xstats_names[0].name));
683 			nstats++;
684 		}
685 	}
686 
687 	return nstats;
688 }
689 
690 static int
691 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
692 		     uint64_t *values, unsigned int n)
693 {
694 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
695 	struct sfc_port *port = &sa->port;
696 	uint64_t *mac_stats;
697 	unsigned int nb_supported = 0;
698 	unsigned int nb_written = 0;
699 	unsigned int i;
700 	int ret;
701 	int rc;
702 
703 	if (unlikely(values == NULL) ||
704 	    unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
705 		return port->mac_stats_nb_supported;
706 
707 	rte_spinlock_lock(&port->mac_stats_lock);
708 
709 	rc = sfc_port_update_mac_stats(sa);
710 	if (rc != 0) {
711 		SFC_ASSERT(rc > 0);
712 		ret = -rc;
713 		goto unlock;
714 	}
715 
716 	mac_stats = port->mac_stats_buf;
717 
718 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
719 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
720 			continue;
721 
722 		if ((ids == NULL) || (ids[nb_written] == nb_supported))
723 			values[nb_written++] = mac_stats[i];
724 
725 		++nb_supported;
726 	}
727 
728 	ret = nb_written;
729 
730 unlock:
731 	rte_spinlock_unlock(&port->mac_stats_lock);
732 
733 	return ret;
734 }
735 
736 static int
737 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
738 			   struct rte_eth_xstat_name *xstats_names,
739 			   const uint64_t *ids, unsigned int size)
740 {
741 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
742 	struct sfc_port *port = &sa->port;
743 	unsigned int nb_supported = 0;
744 	unsigned int nb_written = 0;
745 	unsigned int i;
746 
747 	if (unlikely(xstats_names == NULL) ||
748 	    unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
749 		return port->mac_stats_nb_supported;
750 
751 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
752 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
753 			continue;
754 
755 		if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
756 			char *name = xstats_names[nb_written++].name;
757 
758 			strlcpy(name, efx_mac_stat_name(sa->nic, i),
759 				sizeof(xstats_names[0].name));
760 		}
761 
762 		++nb_supported;
763 	}
764 
765 	return nb_written;
766 }
767 
768 static int
769 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
770 {
771 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
772 	unsigned int wanted_fc, link_fc;
773 
774 	memset(fc_conf, 0, sizeof(*fc_conf));
775 
776 	sfc_adapter_lock(sa);
777 
778 	if (sa->state == SFC_ADAPTER_STARTED)
779 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
780 	else
781 		link_fc = sa->port.flow_ctrl;
782 
783 	switch (link_fc) {
784 	case 0:
785 		fc_conf->mode = RTE_FC_NONE;
786 		break;
787 	case EFX_FCNTL_RESPOND:
788 		fc_conf->mode = RTE_FC_RX_PAUSE;
789 		break;
790 	case EFX_FCNTL_GENERATE:
791 		fc_conf->mode = RTE_FC_TX_PAUSE;
792 		break;
793 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
794 		fc_conf->mode = RTE_FC_FULL;
795 		break;
796 	default:
797 		sfc_err(sa, "%s: unexpected flow control value %#x",
798 			__func__, link_fc);
799 	}
800 
801 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
802 
803 	sfc_adapter_unlock(sa);
804 
805 	return 0;
806 }
807 
808 static int
809 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
810 {
811 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
812 	struct sfc_port *port = &sa->port;
813 	unsigned int fcntl;
814 	int rc;
815 
816 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
817 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
818 	    fc_conf->mac_ctrl_frame_fwd != 0) {
819 		sfc_err(sa, "unsupported flow control settings specified");
820 		rc = EINVAL;
821 		goto fail_inval;
822 	}
823 
824 	switch (fc_conf->mode) {
825 	case RTE_FC_NONE:
826 		fcntl = 0;
827 		break;
828 	case RTE_FC_RX_PAUSE:
829 		fcntl = EFX_FCNTL_RESPOND;
830 		break;
831 	case RTE_FC_TX_PAUSE:
832 		fcntl = EFX_FCNTL_GENERATE;
833 		break;
834 	case RTE_FC_FULL:
835 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
836 		break;
837 	default:
838 		rc = EINVAL;
839 		goto fail_inval;
840 	}
841 
842 	sfc_adapter_lock(sa);
843 
844 	if (sa->state == SFC_ADAPTER_STARTED) {
845 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
846 		if (rc != 0)
847 			goto fail_mac_fcntl_set;
848 	}
849 
850 	port->flow_ctrl = fcntl;
851 	port->flow_ctrl_autoneg = fc_conf->autoneg;
852 
853 	sfc_adapter_unlock(sa);
854 
855 	return 0;
856 
857 fail_mac_fcntl_set:
858 	sfc_adapter_unlock(sa);
859 fail_inval:
860 	SFC_ASSERT(rc > 0);
861 	return -rc;
862 }
863 
864 static int
865 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
866 {
867 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
868 	size_t pdu = EFX_MAC_PDU(mtu);
869 	size_t old_pdu;
870 	int rc;
871 
872 	sfc_log_init(sa, "mtu=%u", mtu);
873 
874 	rc = EINVAL;
875 	if (pdu < EFX_MAC_PDU_MIN) {
876 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
877 			(unsigned int)mtu, (unsigned int)pdu,
878 			EFX_MAC_PDU_MIN);
879 		goto fail_inval;
880 	}
881 	if (pdu > EFX_MAC_PDU_MAX) {
882 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
883 			(unsigned int)mtu, (unsigned int)pdu,
884 			EFX_MAC_PDU_MAX);
885 		goto fail_inval;
886 	}
887 
888 	sfc_adapter_lock(sa);
889 
890 	if (pdu != sa->port.pdu) {
891 		if (sa->state == SFC_ADAPTER_STARTED) {
892 			sfc_stop(sa);
893 
894 			old_pdu = sa->port.pdu;
895 			sa->port.pdu = pdu;
896 			rc = sfc_start(sa);
897 			if (rc != 0)
898 				goto fail_start;
899 		} else {
900 			sa->port.pdu = pdu;
901 		}
902 	}
903 
904 	/*
905 	 * The driver does not use it, but other PMDs update jumbo frame
906 	 * flag and max_rx_pkt_len when MTU is set.
907 	 */
908 	if (mtu > ETHER_MAX_LEN) {
909 		struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
910 		rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
911 	}
912 
913 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
914 
915 	sfc_adapter_unlock(sa);
916 
917 	sfc_log_init(sa, "done");
918 	return 0;
919 
920 fail_start:
921 	sa->port.pdu = old_pdu;
922 	if (sfc_start(sa) != 0)
923 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
924 			"PDU max size - port is stopped",
925 			(unsigned int)pdu, (unsigned int)old_pdu);
926 	sfc_adapter_unlock(sa);
927 
928 fail_inval:
929 	sfc_log_init(sa, "failed %d", rc);
930 	SFC_ASSERT(rc > 0);
931 	return -rc;
932 }
933 static int
934 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
935 {
936 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
937 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
938 	struct sfc_port *port = &sa->port;
939 	struct ether_addr *old_addr = &dev->data->mac_addrs[0];
940 	int rc = 0;
941 
942 	sfc_adapter_lock(sa);
943 
944 	/*
945 	 * Copy the address to the device private data so that
946 	 * it could be recalled in the case of adapter restart.
947 	 */
948 	ether_addr_copy(mac_addr, &port->default_mac_addr);
949 
950 	/*
951 	 * Neither of the two following checks can return
952 	 * an error. The new MAC address is preserved in
953 	 * the device private data and can be activated
954 	 * on the next port start if the user prevents
955 	 * isolated mode from being enabled.
956 	 */
957 	if (sfc_sa2shared(sa)->isolated) {
958 		sfc_warn(sa, "isolated mode is active on the port");
959 		sfc_warn(sa, "will not set MAC address");
960 		goto unlock;
961 	}
962 
963 	if (sa->state != SFC_ADAPTER_STARTED) {
964 		sfc_notice(sa, "the port is not started");
965 		sfc_notice(sa, "the new MAC address will be set on port start");
966 
967 		goto unlock;
968 	}
969 
970 	if (encp->enc_allow_set_mac_with_installed_filters) {
971 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
972 		if (rc != 0) {
973 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
974 			goto unlock;
975 		}
976 
977 		/*
978 		 * Changing the MAC address by means of MCDI request
979 		 * has no effect on received traffic, therefore
980 		 * we also need to update unicast filters
981 		 */
982 		rc = sfc_set_rx_mode(sa);
983 		if (rc != 0) {
984 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
985 			/* Rollback the old address */
986 			(void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
987 			(void)sfc_set_rx_mode(sa);
988 		}
989 	} else {
990 		sfc_warn(sa, "cannot set MAC address with filters installed");
991 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
992 		sfc_warn(sa, "(some traffic may be dropped)");
993 
994 		/*
995 		 * Since setting MAC address with filters installed is not
996 		 * allowed on the adapter, the new MAC address will be set
997 		 * by means of adapter restart. sfc_start() shall retrieve
998 		 * the new address from the device private data and set it.
999 		 */
1000 		sfc_stop(sa);
1001 		rc = sfc_start(sa);
1002 		if (rc != 0)
1003 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1004 	}
1005 
1006 unlock:
1007 	if (rc != 0)
1008 		ether_addr_copy(old_addr, &port->default_mac_addr);
1009 
1010 	sfc_adapter_unlock(sa);
1011 
1012 	SFC_ASSERT(rc >= 0);
1013 	return -rc;
1014 }
1015 
1016 
1017 static int
1018 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
1019 		     uint32_t nb_mc_addr)
1020 {
1021 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1022 	struct sfc_port *port = &sa->port;
1023 	uint8_t *mc_addrs = port->mcast_addrs;
1024 	int rc;
1025 	unsigned int i;
1026 
1027 	if (sfc_sa2shared(sa)->isolated) {
1028 		sfc_err(sa, "isolated mode is active on the port");
1029 		sfc_err(sa, "will not set multicast address list");
1030 		return -ENOTSUP;
1031 	}
1032 
1033 	if (mc_addrs == NULL)
1034 		return -ENOBUFS;
1035 
1036 	if (nb_mc_addr > port->max_mcast_addrs) {
1037 		sfc_err(sa, "too many multicast addresses: %u > %u",
1038 			 nb_mc_addr, port->max_mcast_addrs);
1039 		return -EINVAL;
1040 	}
1041 
1042 	for (i = 0; i < nb_mc_addr; ++i) {
1043 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1044 				 EFX_MAC_ADDR_LEN);
1045 		mc_addrs += EFX_MAC_ADDR_LEN;
1046 	}
1047 
1048 	port->nb_mcast_addrs = nb_mc_addr;
1049 
1050 	if (sa->state != SFC_ADAPTER_STARTED)
1051 		return 0;
1052 
1053 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1054 					port->nb_mcast_addrs);
1055 	if (rc != 0)
1056 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1057 
1058 	SFC_ASSERT(rc >= 0);
1059 	return -rc;
1060 }
1061 
1062 /*
1063  * The function is used by the secondary process as well. It must not
1064  * use any process-local pointers from the adapter data.
1065  */
1066 static void
1067 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1068 		      struct rte_eth_rxq_info *qinfo)
1069 {
1070 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1071 	struct sfc_rxq_info *rxq_info;
1072 
1073 	SFC_ASSERT(rx_queue_id < sas->rxq_count);
1074 
1075 	rxq_info = &sas->rxq_info[rx_queue_id];
1076 
1077 	qinfo->mp = rxq_info->refill_mb_pool;
1078 	qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1079 	qinfo->conf.rx_drop_en = 1;
1080 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1081 	qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1082 	if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1083 		qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1084 		qinfo->scattered_rx = 1;
1085 	}
1086 	qinfo->nb_desc = rxq_info->entries;
1087 }
1088 
1089 /*
1090  * The function is used by the secondary process as well. It must not
1091  * use any process-local pointers from the adapter data.
1092  */
1093 static void
1094 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1095 		      struct rte_eth_txq_info *qinfo)
1096 {
1097 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1098 	struct sfc_txq_info *txq_info;
1099 
1100 	SFC_ASSERT(tx_queue_id < sas->txq_count);
1101 
1102 	txq_info = &sas->txq_info[tx_queue_id];
1103 
1104 	memset(qinfo, 0, sizeof(*qinfo));
1105 
1106 	qinfo->conf.offloads = txq_info->offloads;
1107 	qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1108 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1109 	qinfo->nb_desc = txq_info->entries;
1110 }
1111 
1112 /*
1113  * The function is used by the secondary process as well. It must not
1114  * use any process-local pointers from the adapter data.
1115  */
1116 static uint32_t
1117 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1118 {
1119 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1120 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1121 	struct sfc_rxq_info *rxq_info;
1122 
1123 	SFC_ASSERT(rx_queue_id < sas->rxq_count);
1124 	rxq_info = &sas->rxq_info[rx_queue_id];
1125 
1126 	if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1127 		return 0;
1128 
1129 	return sap->dp_rx->qdesc_npending(rxq_info->dp);
1130 }
1131 
1132 /*
1133  * The function is used by the secondary process as well. It must not
1134  * use any process-local pointers from the adapter data.
1135  */
1136 static int
1137 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1138 {
1139 	struct sfc_dp_rxq *dp_rxq = queue;
1140 	const struct sfc_dp_rx *dp_rx;
1141 
1142 	dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1143 
1144 	return offset < dp_rx->qdesc_npending(dp_rxq);
1145 }
1146 
1147 /*
1148  * The function is used by the secondary process as well. It must not
1149  * use any process-local pointers from the adapter data.
1150  */
1151 static int
1152 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1153 {
1154 	struct sfc_dp_rxq *dp_rxq = queue;
1155 	const struct sfc_dp_rx *dp_rx;
1156 
1157 	dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1158 
1159 	return dp_rx->qdesc_status(dp_rxq, offset);
1160 }
1161 
1162 /*
1163  * The function is used by the secondary process as well. It must not
1164  * use any process-local pointers from the adapter data.
1165  */
1166 static int
1167 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1168 {
1169 	struct sfc_dp_txq *dp_txq = queue;
1170 	const struct sfc_dp_tx *dp_tx;
1171 
1172 	dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1173 
1174 	return dp_tx->qdesc_status(dp_txq, offset);
1175 }
1176 
1177 static int
1178 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1179 {
1180 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1181 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1182 	int rc;
1183 
1184 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1185 
1186 	sfc_adapter_lock(sa);
1187 
1188 	rc = EINVAL;
1189 	if (sa->state != SFC_ADAPTER_STARTED)
1190 		goto fail_not_started;
1191 
1192 	if (sas->rxq_info[rx_queue_id].state != SFC_RXQ_INITIALIZED)
1193 		goto fail_not_setup;
1194 
1195 	rc = sfc_rx_qstart(sa, rx_queue_id);
1196 	if (rc != 0)
1197 		goto fail_rx_qstart;
1198 
1199 	sas->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1200 
1201 	sfc_adapter_unlock(sa);
1202 
1203 	return 0;
1204 
1205 fail_rx_qstart:
1206 fail_not_setup:
1207 fail_not_started:
1208 	sfc_adapter_unlock(sa);
1209 	SFC_ASSERT(rc > 0);
1210 	return -rc;
1211 }
1212 
1213 static int
1214 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1215 {
1216 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1217 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1218 
1219 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1220 
1221 	sfc_adapter_lock(sa);
1222 	sfc_rx_qstop(sa, rx_queue_id);
1223 
1224 	sas->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1225 
1226 	sfc_adapter_unlock(sa);
1227 
1228 	return 0;
1229 }
1230 
1231 static int
1232 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1233 {
1234 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1235 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1236 	int rc;
1237 
1238 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1239 
1240 	sfc_adapter_lock(sa);
1241 
1242 	rc = EINVAL;
1243 	if (sa->state != SFC_ADAPTER_STARTED)
1244 		goto fail_not_started;
1245 
1246 	if (sas->txq_info[tx_queue_id].state != SFC_TXQ_INITIALIZED)
1247 		goto fail_not_setup;
1248 
1249 	rc = sfc_tx_qstart(sa, tx_queue_id);
1250 	if (rc != 0)
1251 		goto fail_tx_qstart;
1252 
1253 	sas->txq_info[tx_queue_id].deferred_started = B_TRUE;
1254 
1255 	sfc_adapter_unlock(sa);
1256 	return 0;
1257 
1258 fail_tx_qstart:
1259 
1260 fail_not_setup:
1261 fail_not_started:
1262 	sfc_adapter_unlock(sa);
1263 	SFC_ASSERT(rc > 0);
1264 	return -rc;
1265 }
1266 
1267 static int
1268 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1269 {
1270 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1271 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1272 
1273 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1274 
1275 	sfc_adapter_lock(sa);
1276 
1277 	sfc_tx_qstop(sa, tx_queue_id);
1278 
1279 	sas->txq_info[tx_queue_id].deferred_started = B_FALSE;
1280 
1281 	sfc_adapter_unlock(sa);
1282 	return 0;
1283 }
1284 
1285 static efx_tunnel_protocol_t
1286 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1287 {
1288 	switch (rte_type) {
1289 	case RTE_TUNNEL_TYPE_VXLAN:
1290 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1291 	case RTE_TUNNEL_TYPE_GENEVE:
1292 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1293 	default:
1294 		return EFX_TUNNEL_NPROTOS;
1295 	}
1296 }
1297 
1298 enum sfc_udp_tunnel_op_e {
1299 	SFC_UDP_TUNNEL_ADD_PORT,
1300 	SFC_UDP_TUNNEL_DEL_PORT,
1301 };
1302 
1303 static int
1304 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1305 		      struct rte_eth_udp_tunnel *tunnel_udp,
1306 		      enum sfc_udp_tunnel_op_e op)
1307 {
1308 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1309 	efx_tunnel_protocol_t tunnel_proto;
1310 	int rc;
1311 
1312 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1313 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1314 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1315 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1316 
1317 	tunnel_proto =
1318 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1319 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1320 		rc = ENOTSUP;
1321 		goto fail_bad_proto;
1322 	}
1323 
1324 	sfc_adapter_lock(sa);
1325 
1326 	switch (op) {
1327 	case SFC_UDP_TUNNEL_ADD_PORT:
1328 		rc = efx_tunnel_config_udp_add(sa->nic,
1329 					       tunnel_udp->udp_port,
1330 					       tunnel_proto);
1331 		break;
1332 	case SFC_UDP_TUNNEL_DEL_PORT:
1333 		rc = efx_tunnel_config_udp_remove(sa->nic,
1334 						  tunnel_udp->udp_port,
1335 						  tunnel_proto);
1336 		break;
1337 	default:
1338 		rc = EINVAL;
1339 		goto fail_bad_op;
1340 	}
1341 
1342 	if (rc != 0)
1343 		goto fail_op;
1344 
1345 	if (sa->state == SFC_ADAPTER_STARTED) {
1346 		rc = efx_tunnel_reconfigure(sa->nic);
1347 		if (rc == EAGAIN) {
1348 			/*
1349 			 * Configuration is accepted by FW and MC reboot
1350 			 * is initiated to apply the changes. MC reboot
1351 			 * will be handled in a usual way (MC reboot
1352 			 * event on management event queue and adapter
1353 			 * restart).
1354 			 */
1355 			rc = 0;
1356 		} else if (rc != 0) {
1357 			goto fail_reconfigure;
1358 		}
1359 	}
1360 
1361 	sfc_adapter_unlock(sa);
1362 	return 0;
1363 
1364 fail_reconfigure:
1365 	/* Remove/restore entry since the change makes the trouble */
1366 	switch (op) {
1367 	case SFC_UDP_TUNNEL_ADD_PORT:
1368 		(void)efx_tunnel_config_udp_remove(sa->nic,
1369 						   tunnel_udp->udp_port,
1370 						   tunnel_proto);
1371 		break;
1372 	case SFC_UDP_TUNNEL_DEL_PORT:
1373 		(void)efx_tunnel_config_udp_add(sa->nic,
1374 						tunnel_udp->udp_port,
1375 						tunnel_proto);
1376 		break;
1377 	}
1378 
1379 fail_op:
1380 fail_bad_op:
1381 	sfc_adapter_unlock(sa);
1382 
1383 fail_bad_proto:
1384 	SFC_ASSERT(rc > 0);
1385 	return -rc;
1386 }
1387 
1388 static int
1389 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1390 			    struct rte_eth_udp_tunnel *tunnel_udp)
1391 {
1392 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1393 }
1394 
1395 static int
1396 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1397 			    struct rte_eth_udp_tunnel *tunnel_udp)
1398 {
1399 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1400 }
1401 
1402 /*
1403  * The function is used by the secondary process as well. It must not
1404  * use any process-local pointers from the adapter data.
1405  */
1406 static int
1407 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1408 			  struct rte_eth_rss_conf *rss_conf)
1409 {
1410 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1411 	struct sfc_rss *rss = &sas->rss;
1412 
1413 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1414 		return -ENOTSUP;
1415 
1416 	/*
1417 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1418 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1419 	 * flags which corresponds to the active EFX configuration stored
1420 	 * locally in 'sfc_adapter' and kept up-to-date
1421 	 */
1422 	rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1423 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1424 	if (rss_conf->rss_key != NULL)
1425 		rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1426 
1427 	return 0;
1428 }
1429 
1430 static int
1431 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1432 			struct rte_eth_rss_conf *rss_conf)
1433 {
1434 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1435 	struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1436 	unsigned int efx_hash_types;
1437 	int rc = 0;
1438 
1439 	if (sfc_sa2shared(sa)->isolated)
1440 		return -ENOTSUP;
1441 
1442 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1443 		sfc_err(sa, "RSS is not available");
1444 		return -ENOTSUP;
1445 	}
1446 
1447 	if (rss->channels == 0) {
1448 		sfc_err(sa, "RSS is not configured");
1449 		return -EINVAL;
1450 	}
1451 
1452 	if ((rss_conf->rss_key != NULL) &&
1453 	    (rss_conf->rss_key_len != sizeof(rss->key))) {
1454 		sfc_err(sa, "RSS key size is wrong (should be %lu)",
1455 			sizeof(rss->key));
1456 		return -EINVAL;
1457 	}
1458 
1459 	sfc_adapter_lock(sa);
1460 
1461 	rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1462 	if (rc != 0)
1463 		goto fail_rx_hf_rte_to_efx;
1464 
1465 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1466 				   rss->hash_alg, efx_hash_types, B_TRUE);
1467 	if (rc != 0)
1468 		goto fail_scale_mode_set;
1469 
1470 	if (rss_conf->rss_key != NULL) {
1471 		if (sa->state == SFC_ADAPTER_STARTED) {
1472 			rc = efx_rx_scale_key_set(sa->nic,
1473 						  EFX_RSS_CONTEXT_DEFAULT,
1474 						  rss_conf->rss_key,
1475 						  sizeof(rss->key));
1476 			if (rc != 0)
1477 				goto fail_scale_key_set;
1478 		}
1479 
1480 		rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1481 	}
1482 
1483 	rss->hash_types = efx_hash_types;
1484 
1485 	sfc_adapter_unlock(sa);
1486 
1487 	return 0;
1488 
1489 fail_scale_key_set:
1490 	if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1491 				  EFX_RX_HASHALG_TOEPLITZ,
1492 				  rss->hash_types, B_TRUE) != 0)
1493 		sfc_err(sa, "failed to restore RSS mode");
1494 
1495 fail_scale_mode_set:
1496 fail_rx_hf_rte_to_efx:
1497 	sfc_adapter_unlock(sa);
1498 	return -rc;
1499 }
1500 
1501 /*
1502  * The function is used by the secondary process as well. It must not
1503  * use any process-local pointers from the adapter data.
1504  */
1505 static int
1506 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1507 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1508 		       uint16_t reta_size)
1509 {
1510 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1511 	struct sfc_rss *rss = &sas->rss;
1512 	int entry;
1513 
1514 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1515 		return -ENOTSUP;
1516 
1517 	if (rss->channels == 0)
1518 		return -EINVAL;
1519 
1520 	if (reta_size != EFX_RSS_TBL_SIZE)
1521 		return -EINVAL;
1522 
1523 	for (entry = 0; entry < reta_size; entry++) {
1524 		int grp = entry / RTE_RETA_GROUP_SIZE;
1525 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1526 
1527 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1528 			reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1529 	}
1530 
1531 	return 0;
1532 }
1533 
1534 static int
1535 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1536 			struct rte_eth_rss_reta_entry64 *reta_conf,
1537 			uint16_t reta_size)
1538 {
1539 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1540 	struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1541 	unsigned int *rss_tbl_new;
1542 	uint16_t entry;
1543 	int rc = 0;
1544 
1545 
1546 	if (sfc_sa2shared(sa)->isolated)
1547 		return -ENOTSUP;
1548 
1549 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1550 		sfc_err(sa, "RSS is not available");
1551 		return -ENOTSUP;
1552 	}
1553 
1554 	if (rss->channels == 0) {
1555 		sfc_err(sa, "RSS is not configured");
1556 		return -EINVAL;
1557 	}
1558 
1559 	if (reta_size != EFX_RSS_TBL_SIZE) {
1560 		sfc_err(sa, "RETA size is wrong (should be %u)",
1561 			EFX_RSS_TBL_SIZE);
1562 		return -EINVAL;
1563 	}
1564 
1565 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1566 	if (rss_tbl_new == NULL)
1567 		return -ENOMEM;
1568 
1569 	sfc_adapter_lock(sa);
1570 
1571 	rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1572 
1573 	for (entry = 0; entry < reta_size; entry++) {
1574 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1575 		struct rte_eth_rss_reta_entry64 *grp;
1576 
1577 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1578 
1579 		if (grp->mask & (1ull << grp_idx)) {
1580 			if (grp->reta[grp_idx] >= rss->channels) {
1581 				rc = EINVAL;
1582 				goto bad_reta_entry;
1583 			}
1584 			rss_tbl_new[entry] = grp->reta[grp_idx];
1585 		}
1586 	}
1587 
1588 	if (sa->state == SFC_ADAPTER_STARTED) {
1589 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1590 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1591 		if (rc != 0)
1592 			goto fail_scale_tbl_set;
1593 	}
1594 
1595 	rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1596 
1597 fail_scale_tbl_set:
1598 bad_reta_entry:
1599 	sfc_adapter_unlock(sa);
1600 
1601 	rte_free(rss_tbl_new);
1602 
1603 	SFC_ASSERT(rc >= 0);
1604 	return -rc;
1605 }
1606 
1607 static int
1608 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1609 		    enum rte_filter_op filter_op,
1610 		    void *arg)
1611 {
1612 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1613 	int rc = ENOTSUP;
1614 
1615 	sfc_log_init(sa, "entry");
1616 
1617 	switch (filter_type) {
1618 	case RTE_ETH_FILTER_NONE:
1619 		sfc_err(sa, "Global filters configuration not supported");
1620 		break;
1621 	case RTE_ETH_FILTER_MACVLAN:
1622 		sfc_err(sa, "MACVLAN filters not supported");
1623 		break;
1624 	case RTE_ETH_FILTER_ETHERTYPE:
1625 		sfc_err(sa, "EtherType filters not supported");
1626 		break;
1627 	case RTE_ETH_FILTER_FLEXIBLE:
1628 		sfc_err(sa, "Flexible filters not supported");
1629 		break;
1630 	case RTE_ETH_FILTER_SYN:
1631 		sfc_err(sa, "SYN filters not supported");
1632 		break;
1633 	case RTE_ETH_FILTER_NTUPLE:
1634 		sfc_err(sa, "NTUPLE filters not supported");
1635 		break;
1636 	case RTE_ETH_FILTER_TUNNEL:
1637 		sfc_err(sa, "Tunnel filters not supported");
1638 		break;
1639 	case RTE_ETH_FILTER_FDIR:
1640 		sfc_err(sa, "Flow Director filters not supported");
1641 		break;
1642 	case RTE_ETH_FILTER_HASH:
1643 		sfc_err(sa, "Hash filters not supported");
1644 		break;
1645 	case RTE_ETH_FILTER_GENERIC:
1646 		if (filter_op != RTE_ETH_FILTER_GET) {
1647 			rc = EINVAL;
1648 		} else {
1649 			*(const void **)arg = &sfc_flow_ops;
1650 			rc = 0;
1651 		}
1652 		break;
1653 	default:
1654 		sfc_err(sa, "Unknown filter type %u", filter_type);
1655 		break;
1656 	}
1657 
1658 	sfc_log_init(sa, "exit: %d", -rc);
1659 	SFC_ASSERT(rc >= 0);
1660 	return -rc;
1661 }
1662 
1663 static int
1664 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1665 {
1666 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1667 
1668 	/*
1669 	 * If Rx datapath does not provide callback to check mempool,
1670 	 * all pools are supported.
1671 	 */
1672 	if (sap->dp_rx->pool_ops_supported == NULL)
1673 		return 1;
1674 
1675 	return sap->dp_rx->pool_ops_supported(pool);
1676 }
1677 
1678 static const struct eth_dev_ops sfc_eth_dev_ops = {
1679 	.dev_configure			= sfc_dev_configure,
1680 	.dev_start			= sfc_dev_start,
1681 	.dev_stop			= sfc_dev_stop,
1682 	.dev_set_link_up		= sfc_dev_set_link_up,
1683 	.dev_set_link_down		= sfc_dev_set_link_down,
1684 	.dev_close			= sfc_dev_close,
1685 	.promiscuous_enable		= sfc_dev_promisc_enable,
1686 	.promiscuous_disable		= sfc_dev_promisc_disable,
1687 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1688 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1689 	.link_update			= sfc_dev_link_update,
1690 	.stats_get			= sfc_stats_get,
1691 	.stats_reset			= sfc_stats_reset,
1692 	.xstats_get			= sfc_xstats_get,
1693 	.xstats_reset			= sfc_stats_reset,
1694 	.xstats_get_names		= sfc_xstats_get_names,
1695 	.dev_infos_get			= sfc_dev_infos_get,
1696 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1697 	.mtu_set			= sfc_dev_set_mtu,
1698 	.rx_queue_start			= sfc_rx_queue_start,
1699 	.rx_queue_stop			= sfc_rx_queue_stop,
1700 	.tx_queue_start			= sfc_tx_queue_start,
1701 	.tx_queue_stop			= sfc_tx_queue_stop,
1702 	.rx_queue_setup			= sfc_rx_queue_setup,
1703 	.rx_queue_release		= sfc_rx_queue_release,
1704 	.rx_queue_count			= sfc_rx_queue_count,
1705 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1706 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1707 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1708 	.tx_queue_setup			= sfc_tx_queue_setup,
1709 	.tx_queue_release		= sfc_tx_queue_release,
1710 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1711 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1712 	.mac_addr_set			= sfc_mac_addr_set,
1713 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1714 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
1715 	.reta_update			= sfc_dev_rss_reta_update,
1716 	.reta_query			= sfc_dev_rss_reta_query,
1717 	.rss_hash_update		= sfc_dev_rss_hash_update,
1718 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1719 	.filter_ctrl			= sfc_dev_filter_ctrl,
1720 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1721 	.rxq_info_get			= sfc_rx_queue_info_get,
1722 	.txq_info_get			= sfc_tx_queue_info_get,
1723 	.fw_version_get			= sfc_fw_version_get,
1724 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1725 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1726 	.pool_ops_supported		= sfc_pool_ops_supported,
1727 };
1728 
1729 /**
1730  * Duplicate a string in potentially shared memory required for
1731  * multi-process support.
1732  *
1733  * strdup() allocates from process-local heap/memory.
1734  */
1735 static char *
1736 sfc_strdup(const char *str)
1737 {
1738 	size_t size;
1739 	char *copy;
1740 
1741 	if (str == NULL)
1742 		return NULL;
1743 
1744 	size = strlen(str) + 1;
1745 	copy = rte_malloc(__func__, size, 0);
1746 	if (copy != NULL)
1747 		rte_memcpy(copy, str, size);
1748 
1749 	return copy;
1750 }
1751 
1752 static int
1753 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1754 {
1755 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1756 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1757 	const struct sfc_dp_rx *dp_rx;
1758 	const struct sfc_dp_tx *dp_tx;
1759 	const efx_nic_cfg_t *encp;
1760 	unsigned int avail_caps = 0;
1761 	const char *rx_name = NULL;
1762 	const char *tx_name = NULL;
1763 	int rc;
1764 
1765 	switch (sa->family) {
1766 	case EFX_FAMILY_HUNTINGTON:
1767 	case EFX_FAMILY_MEDFORD:
1768 	case EFX_FAMILY_MEDFORD2:
1769 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1770 		break;
1771 	default:
1772 		break;
1773 	}
1774 
1775 	encp = efx_nic_cfg_get(sa->nic);
1776 	if (encp->enc_rx_es_super_buffer_supported)
1777 		avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1778 
1779 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1780 				sfc_kvarg_string_handler, &rx_name);
1781 	if (rc != 0)
1782 		goto fail_kvarg_rx_datapath;
1783 
1784 	if (rx_name != NULL) {
1785 		dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1786 		if (dp_rx == NULL) {
1787 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1788 			rc = ENOENT;
1789 			goto fail_dp_rx;
1790 		}
1791 		if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1792 			sfc_err(sa,
1793 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1794 				rx_name);
1795 			rc = EINVAL;
1796 			goto fail_dp_rx_caps;
1797 		}
1798 	} else {
1799 		dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1800 		if (dp_rx == NULL) {
1801 			sfc_err(sa, "Rx datapath by caps %#x not found",
1802 				avail_caps);
1803 			rc = ENOENT;
1804 			goto fail_dp_rx;
1805 		}
1806 	}
1807 
1808 	sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1809 	if (sas->dp_rx_name == NULL) {
1810 		rc = ENOMEM;
1811 		goto fail_dp_rx_name;
1812 	}
1813 
1814 	sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
1815 
1816 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1817 				sfc_kvarg_string_handler, &tx_name);
1818 	if (rc != 0)
1819 		goto fail_kvarg_tx_datapath;
1820 
1821 	if (tx_name != NULL) {
1822 		dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1823 		if (dp_tx == NULL) {
1824 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1825 			rc = ENOENT;
1826 			goto fail_dp_tx;
1827 		}
1828 		if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
1829 			sfc_err(sa,
1830 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1831 				tx_name);
1832 			rc = EINVAL;
1833 			goto fail_dp_tx_caps;
1834 		}
1835 	} else {
1836 		dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1837 		if (dp_tx == NULL) {
1838 			sfc_err(sa, "Tx datapath by caps %#x not found",
1839 				avail_caps);
1840 			rc = ENOENT;
1841 			goto fail_dp_tx;
1842 		}
1843 	}
1844 
1845 	sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
1846 	if (sas->dp_tx_name == NULL) {
1847 		rc = ENOMEM;
1848 		goto fail_dp_tx_name;
1849 	}
1850 
1851 	sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
1852 
1853 	sa->priv.dp_rx = dp_rx;
1854 	sa->priv.dp_tx = dp_tx;
1855 
1856 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1857 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1858 
1859 	dev->dev_ops = &sfc_eth_dev_ops;
1860 
1861 	return 0;
1862 
1863 fail_dp_tx_name:
1864 fail_dp_tx_caps:
1865 fail_dp_tx:
1866 fail_kvarg_tx_datapath:
1867 	rte_free(sas->dp_rx_name);
1868 	sas->dp_rx_name = NULL;
1869 
1870 fail_dp_rx_name:
1871 fail_dp_rx_caps:
1872 fail_dp_rx:
1873 fail_kvarg_rx_datapath:
1874 	return rc;
1875 }
1876 
1877 static void
1878 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1879 {
1880 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1881 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1882 
1883 	dev->dev_ops = NULL;
1884 	dev->rx_pkt_burst = NULL;
1885 	dev->tx_pkt_burst = NULL;
1886 
1887 	rte_free(sas->dp_tx_name);
1888 	sas->dp_tx_name = NULL;
1889 	sa->priv.dp_tx = NULL;
1890 
1891 	rte_free(sas->dp_rx_name);
1892 	sas->dp_rx_name = NULL;
1893 	sa->priv.dp_rx = NULL;
1894 }
1895 
1896 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1897 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1898 	.rx_queue_count			= sfc_rx_queue_count,
1899 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1900 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1901 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1902 	.reta_query			= sfc_dev_rss_reta_query,
1903 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1904 	.rxq_info_get			= sfc_rx_queue_info_get,
1905 	.txq_info_get			= sfc_tx_queue_info_get,
1906 };
1907 
1908 static int
1909 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
1910 {
1911 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1912 	struct sfc_adapter_priv *sap;
1913 	const struct sfc_dp_rx *dp_rx;
1914 	const struct sfc_dp_tx *dp_tx;
1915 	int rc;
1916 
1917 	/*
1918 	 * Allocate process private data from heap, since it should not
1919 	 * be located in shared memory allocated using rte_malloc() API.
1920 	 */
1921 	sap = calloc(1, sizeof(*sap));
1922 	if (sap == NULL) {
1923 		rc = ENOMEM;
1924 		goto fail_alloc_priv;
1925 	}
1926 
1927 	sap->logtype_main = logtype_main;
1928 
1929 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
1930 	if (dp_rx == NULL) {
1931 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1932 			"cannot find %s Rx datapath", sas->dp_rx_name);
1933 		rc = ENOENT;
1934 		goto fail_dp_rx;
1935 	}
1936 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1937 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1938 			"%s Rx datapath does not support multi-process",
1939 			sas->dp_rx_name);
1940 		rc = EINVAL;
1941 		goto fail_dp_rx_multi_process;
1942 	}
1943 
1944 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
1945 	if (dp_tx == NULL) {
1946 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1947 			"cannot find %s Tx datapath", sas->dp_tx_name);
1948 		rc = ENOENT;
1949 		goto fail_dp_tx;
1950 	}
1951 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1952 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1953 			"%s Tx datapath does not support multi-process",
1954 			sas->dp_tx_name);
1955 		rc = EINVAL;
1956 		goto fail_dp_tx_multi_process;
1957 	}
1958 
1959 	sap->dp_rx = dp_rx;
1960 	sap->dp_tx = dp_tx;
1961 
1962 	dev->process_private = sap;
1963 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1964 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1965 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
1966 
1967 	return 0;
1968 
1969 fail_dp_tx_multi_process:
1970 fail_dp_tx:
1971 fail_dp_rx_multi_process:
1972 fail_dp_rx:
1973 	free(sap);
1974 
1975 fail_alloc_priv:
1976 	return rc;
1977 }
1978 
1979 static void
1980 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1981 {
1982 	free(dev->process_private);
1983 	dev->process_private = NULL;
1984 	dev->dev_ops = NULL;
1985 	dev->tx_pkt_burst = NULL;
1986 	dev->rx_pkt_burst = NULL;
1987 }
1988 
1989 static void
1990 sfc_register_dp(void)
1991 {
1992 	/* Register once */
1993 	if (TAILQ_EMPTY(&sfc_dp_head)) {
1994 		/* Prefer EF10 datapath */
1995 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
1996 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1997 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1998 
1999 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2000 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2001 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2002 	}
2003 }
2004 
2005 static int
2006 sfc_eth_dev_init(struct rte_eth_dev *dev)
2007 {
2008 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2009 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2010 	uint32_t logtype_main;
2011 	struct sfc_adapter *sa;
2012 	int rc;
2013 	const efx_nic_cfg_t *encp;
2014 	const struct ether_addr *from;
2015 
2016 	sfc_register_dp();
2017 
2018 	logtype_main = sfc_register_logtype(&pci_dev->addr,
2019 					    SFC_LOGTYPE_MAIN_STR,
2020 					    RTE_LOG_NOTICE);
2021 
2022 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2023 		return -sfc_eth_dev_secondary_init(dev, logtype_main);
2024 
2025 	/* Required for logging */
2026 	sas->pci_addr = pci_dev->addr;
2027 	sas->port_id = dev->data->port_id;
2028 
2029 	/*
2030 	 * Allocate process private data from heap, since it should not
2031 	 * be located in shared memory allocated using rte_malloc() API.
2032 	 */
2033 	sa = calloc(1, sizeof(*sa));
2034 	if (sa == NULL) {
2035 		rc = ENOMEM;
2036 		goto fail_alloc_sa;
2037 	}
2038 
2039 	dev->process_private = sa;
2040 
2041 	/* Required for logging */
2042 	sa->priv.shared = sas;
2043 	sa->priv.logtype_main = logtype_main;
2044 
2045 	sa->eth_dev = dev;
2046 
2047 	/* Copy PCI device info to the dev->data */
2048 	rte_eth_copy_pci_info(dev, pci_dev);
2049 
2050 	rc = sfc_kvargs_parse(sa);
2051 	if (rc != 0)
2052 		goto fail_kvargs_parse;
2053 
2054 	sfc_log_init(sa, "entry");
2055 
2056 	dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
2057 	if (dev->data->mac_addrs == NULL) {
2058 		rc = ENOMEM;
2059 		goto fail_mac_addrs;
2060 	}
2061 
2062 	sfc_adapter_lock_init(sa);
2063 	sfc_adapter_lock(sa);
2064 
2065 	sfc_log_init(sa, "probing");
2066 	rc = sfc_probe(sa);
2067 	if (rc != 0)
2068 		goto fail_probe;
2069 
2070 	sfc_log_init(sa, "set device ops");
2071 	rc = sfc_eth_dev_set_ops(dev);
2072 	if (rc != 0)
2073 		goto fail_set_ops;
2074 
2075 	sfc_log_init(sa, "attaching");
2076 	rc = sfc_attach(sa);
2077 	if (rc != 0)
2078 		goto fail_attach;
2079 
2080 	encp = efx_nic_cfg_get(sa->nic);
2081 
2082 	/*
2083 	 * The arguments are really reverse order in comparison to
2084 	 * Linux kernel. Copy from NIC config to Ethernet device data.
2085 	 */
2086 	from = (const struct ether_addr *)(encp->enc_mac_addr);
2087 	ether_addr_copy(from, &dev->data->mac_addrs[0]);
2088 
2089 	sfc_adapter_unlock(sa);
2090 
2091 	sfc_log_init(sa, "done");
2092 	return 0;
2093 
2094 fail_attach:
2095 	sfc_eth_dev_clear_ops(dev);
2096 
2097 fail_set_ops:
2098 	sfc_unprobe(sa);
2099 
2100 fail_probe:
2101 	sfc_adapter_unlock(sa);
2102 	sfc_adapter_lock_fini(sa);
2103 	rte_free(dev->data->mac_addrs);
2104 	dev->data->mac_addrs = NULL;
2105 
2106 fail_mac_addrs:
2107 	sfc_kvargs_cleanup(sa);
2108 
2109 fail_kvargs_parse:
2110 	sfc_log_init(sa, "failed %d", rc);
2111 	dev->process_private = NULL;
2112 	free(sa);
2113 
2114 fail_alloc_sa:
2115 	SFC_ASSERT(rc > 0);
2116 	return -rc;
2117 }
2118 
2119 static int
2120 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2121 {
2122 	struct sfc_adapter *sa;
2123 
2124 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2125 		sfc_eth_dev_secondary_clear_ops(dev);
2126 		return 0;
2127 	}
2128 
2129 	sa = sfc_adapter_by_eth_dev(dev);
2130 	sfc_log_init(sa, "entry");
2131 
2132 	sfc_adapter_lock(sa);
2133 
2134 	sfc_eth_dev_clear_ops(dev);
2135 
2136 	sfc_detach(sa);
2137 	sfc_unprobe(sa);
2138 
2139 	sfc_kvargs_cleanup(sa);
2140 
2141 	sfc_adapter_unlock(sa);
2142 	sfc_adapter_lock_fini(sa);
2143 
2144 	sfc_log_init(sa, "done");
2145 
2146 	/* Required for logging, so cleanup last */
2147 	sa->eth_dev = NULL;
2148 
2149 	dev->process_private = NULL;
2150 	free(sa);
2151 
2152 	return 0;
2153 }
2154 
2155 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2156 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2157 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2158 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2159 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2160 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2161 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2162 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2163 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2164 	{ .vendor_id = 0 /* sentinel */ }
2165 };
2166 
2167 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2168 	struct rte_pci_device *pci_dev)
2169 {
2170 	return rte_eth_dev_pci_generic_probe(pci_dev,
2171 		sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
2172 }
2173 
2174 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2175 {
2176 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2177 }
2178 
2179 static struct rte_pci_driver sfc_efx_pmd = {
2180 	.id_table = pci_id_sfc_efx_map,
2181 	.drv_flags =
2182 		RTE_PCI_DRV_INTR_LSC |
2183 		RTE_PCI_DRV_NEED_MAPPING,
2184 	.probe = sfc_eth_dev_pci_probe,
2185 	.remove = sfc_eth_dev_pci_remove,
2186 };
2187 
2188 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2189 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2190 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2191 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2192 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2193 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2194 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2195 	SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2196 	SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2197 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2198 
2199 RTE_INIT(sfc_driver_register_logtype)
2200 {
2201 	int ret;
2202 
2203 	ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2204 						   RTE_LOG_NOTICE);
2205 	sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2206 }
2207