xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision d5b0924ba6baae2cb6ac7c880db0977d4befedc6)
1 /*-
2  *   BSD LICENSE
3  *
4  * Copyright (c) 2016-2017 Solarflare Communications Inc.
5  * All rights reserved.
6  *
7  * This software was jointly developed between OKTET Labs (under contract
8  * for Solarflare) and Solarflare Communications, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions are met:
12  *
13  * 1. Redistributions of source code must retain the above copyright notice,
14  *    this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright notice,
16  *    this list of conditions and the following disclaimer in the documentation
17  *    and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
20  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
21  * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
23  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
24  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
25  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
26  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
27  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
28  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
29  * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 #include <rte_dev.h>
33 #include <rte_ethdev.h>
34 #include <rte_ethdev_pci.h>
35 #include <rte_pci.h>
36 #include <rte_errno.h>
37 
38 #include "efx.h"
39 
40 #include "sfc.h"
41 #include "sfc_debug.h"
42 #include "sfc_log.h"
43 #include "sfc_kvargs.h"
44 #include "sfc_ev.h"
45 #include "sfc_rx.h"
46 #include "sfc_tx.h"
47 #include "sfc_flow.h"
48 #include "sfc_dp.h"
49 #include "sfc_dp_rx.h"
50 
51 static struct sfc_dp_list sfc_dp_head =
52 	TAILQ_HEAD_INITIALIZER(sfc_dp_head);
53 
54 static int
55 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
56 {
57 	struct sfc_adapter *sa = dev->data->dev_private;
58 	efx_nic_fw_info_t enfi;
59 	int ret;
60 	int rc;
61 
62 	/*
63 	 * Return value of the callback is likely supposed to be
64 	 * equal to or greater than 0, nevertheless, if an error
65 	 * occurs, it will be desirable to pass it to the caller
66 	 */
67 	if ((fw_version == NULL) || (fw_size == 0))
68 		return -EINVAL;
69 
70 	rc = efx_nic_get_fw_version(sa->nic, &enfi);
71 	if (rc != 0)
72 		return -rc;
73 
74 	ret = snprintf(fw_version, fw_size,
75 		       "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
76 		       enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
77 		       enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
78 	if (ret < 0)
79 		return ret;
80 
81 	if (enfi.enfi_dpcpu_fw_ids_valid) {
82 		size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
83 		int ret_extra;
84 
85 		ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
86 				     fw_size - dpcpu_fw_ids_offset,
87 				     " rx%" PRIx16 " tx%" PRIx16,
88 				     enfi.enfi_rx_dpcpu_fw_id,
89 				     enfi.enfi_tx_dpcpu_fw_id);
90 		if (ret_extra < 0)
91 			return ret_extra;
92 
93 		ret += ret_extra;
94 	}
95 
96 	if (fw_size < (size_t)(++ret))
97 		return ret;
98 	else
99 		return 0;
100 }
101 
102 static void
103 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
104 {
105 	struct sfc_adapter *sa = dev->data->dev_private;
106 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
107 
108 	sfc_log_init(sa, "entry");
109 
110 	dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
111 	dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
112 
113 	/* Autonegotiation may be disabled */
114 	dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
115 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
116 		dev_info->speed_capa |= ETH_LINK_SPEED_1G;
117 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
118 		dev_info->speed_capa |= ETH_LINK_SPEED_10G;
119 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
120 		dev_info->speed_capa |= ETH_LINK_SPEED_40G;
121 
122 	dev_info->max_rx_queues = sa->rxq_max;
123 	dev_info->max_tx_queues = sa->txq_max;
124 
125 	/* By default packets are dropped if no descriptors are available */
126 	dev_info->default_rxconf.rx_drop_en = 1;
127 
128 	dev_info->rx_offload_capa =
129 		DEV_RX_OFFLOAD_IPV4_CKSUM |
130 		DEV_RX_OFFLOAD_UDP_CKSUM |
131 		DEV_RX_OFFLOAD_TCP_CKSUM;
132 
133 	dev_info->tx_offload_capa =
134 		DEV_TX_OFFLOAD_IPV4_CKSUM |
135 		DEV_TX_OFFLOAD_UDP_CKSUM |
136 		DEV_TX_OFFLOAD_TCP_CKSUM;
137 
138 	dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
139 	if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
140 	    !encp->enc_hw_tx_insert_vlan_enabled)
141 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
142 	else
143 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_VLAN_INSERT;
144 
145 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
146 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
147 
148 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_POOL)
149 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTMEMP;
150 
151 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_REFCNT)
152 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOREFCOUNT;
153 
154 #if EFSYS_OPT_RX_SCALE
155 	if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
156 		dev_info->reta_size = EFX_RSS_TBL_SIZE;
157 		dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
158 		dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
159 	}
160 #endif
161 
162 	if (sa->tso)
163 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
164 
165 	dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
166 	dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
167 	/* The RXQ hardware requires that the descriptor count is a power
168 	 * of 2, but rx_desc_lim cannot properly describe that constraint.
169 	 */
170 	dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
171 
172 	dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
173 	dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
174 	/*
175 	 * The TXQ hardware requires that the descriptor count is a power
176 	 * of 2, but tx_desc_lim cannot properly describe that constraint
177 	 */
178 	dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
179 }
180 
181 static const uint32_t *
182 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
183 {
184 	struct sfc_adapter *sa = dev->data->dev_private;
185 
186 	return sa->dp_rx->supported_ptypes_get();
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 = dev_data->dev_private;
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 = dev->data->dev_private;
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 = dev->data->dev_private;
240 	struct rte_eth_link *dev_link = &dev->data->dev_link;
241 	struct rte_eth_link old_link;
242 	struct rte_eth_link current_link;
243 
244 	sfc_log_init(sa, "entry");
245 
246 retry:
247 	EFX_STATIC_ASSERT(sizeof(*dev_link) == sizeof(rte_atomic64_t));
248 	*(int64_t *)&old_link = rte_atomic64_read((rte_atomic64_t *)dev_link);
249 
250 	if (sa->state != SFC_ADAPTER_STARTED) {
251 		sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
252 		if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
253 					 *(uint64_t *)&old_link,
254 					 *(uint64_t *)&current_link))
255 			goto retry;
256 	} else if (wait_to_complete) {
257 		efx_link_mode_t link_mode;
258 
259 		if (efx_port_poll(sa->nic, &link_mode) != 0)
260 			link_mode = EFX_LINK_UNKNOWN;
261 		sfc_port_link_mode_to_info(link_mode, &current_link);
262 
263 		if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
264 					 *(uint64_t *)&old_link,
265 					 *(uint64_t *)&current_link))
266 			goto retry;
267 	} else {
268 		sfc_ev_mgmt_qpoll(sa);
269 		*(int64_t *)&current_link =
270 			rte_atomic64_read((rte_atomic64_t *)dev_link);
271 	}
272 
273 	if (old_link.link_status != current_link.link_status)
274 		sfc_info(sa, "Link status is %s",
275 			 current_link.link_status ? "UP" : "DOWN");
276 
277 	return old_link.link_status == current_link.link_status ? 0 : -1;
278 }
279 
280 static void
281 sfc_dev_stop(struct rte_eth_dev *dev)
282 {
283 	struct sfc_adapter *sa = dev->data->dev_private;
284 
285 	sfc_log_init(sa, "entry");
286 
287 	sfc_adapter_lock(sa);
288 	sfc_stop(sa);
289 	sfc_adapter_unlock(sa);
290 
291 	sfc_log_init(sa, "done");
292 }
293 
294 static int
295 sfc_dev_set_link_up(struct rte_eth_dev *dev)
296 {
297 	struct sfc_adapter *sa = dev->data->dev_private;
298 	int rc;
299 
300 	sfc_log_init(sa, "entry");
301 
302 	sfc_adapter_lock(sa);
303 	rc = sfc_start(sa);
304 	sfc_adapter_unlock(sa);
305 
306 	SFC_ASSERT(rc >= 0);
307 	return -rc;
308 }
309 
310 static int
311 sfc_dev_set_link_down(struct rte_eth_dev *dev)
312 {
313 	struct sfc_adapter *sa = dev->data->dev_private;
314 
315 	sfc_log_init(sa, "entry");
316 
317 	sfc_adapter_lock(sa);
318 	sfc_stop(sa);
319 	sfc_adapter_unlock(sa);
320 
321 	return 0;
322 }
323 
324 static void
325 sfc_dev_close(struct rte_eth_dev *dev)
326 {
327 	struct sfc_adapter *sa = dev->data->dev_private;
328 
329 	sfc_log_init(sa, "entry");
330 
331 	sfc_adapter_lock(sa);
332 	switch (sa->state) {
333 	case SFC_ADAPTER_STARTED:
334 		sfc_stop(sa);
335 		SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
336 		/* FALLTHROUGH */
337 	case SFC_ADAPTER_CONFIGURED:
338 		sfc_close(sa);
339 		SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
340 		/* FALLTHROUGH */
341 	case SFC_ADAPTER_INITIALIZED:
342 		break;
343 	default:
344 		sfc_err(sa, "unexpected adapter state %u on close", sa->state);
345 		break;
346 	}
347 	sfc_adapter_unlock(sa);
348 
349 	sfc_log_init(sa, "done");
350 }
351 
352 static void
353 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
354 		   boolean_t enabled)
355 {
356 	struct sfc_port *port;
357 	boolean_t *toggle;
358 	struct sfc_adapter *sa = dev->data->dev_private;
359 	boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
360 	const char *desc = (allmulti) ? "all-multi" : "promiscuous";
361 
362 	sfc_adapter_lock(sa);
363 
364 	port = &sa->port;
365 	toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
366 
367 	if (*toggle != enabled) {
368 		*toggle = enabled;
369 
370 		if (port->isolated) {
371 			sfc_warn(sa, "isolated mode is active on the port");
372 			sfc_warn(sa, "the change is to be applied on the next "
373 				     "start provided that isolated mode is "
374 				     "disabled prior the next start");
375 		} else if ((sa->state == SFC_ADAPTER_STARTED) &&
376 			   (sfc_set_rx_mode(sa) != 0)) {
377 			*toggle = !(enabled);
378 			sfc_warn(sa, "Failed to %s %s mode",
379 				 ((enabled) ? "enable" : "disable"), desc);
380 		}
381 	}
382 
383 	sfc_adapter_unlock(sa);
384 }
385 
386 static void
387 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
388 {
389 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
390 }
391 
392 static void
393 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
394 {
395 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
396 }
397 
398 static void
399 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
400 {
401 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
402 }
403 
404 static void
405 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
406 {
407 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
408 }
409 
410 static int
411 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
412 		   uint16_t nb_rx_desc, unsigned int socket_id,
413 		   const struct rte_eth_rxconf *rx_conf,
414 		   struct rte_mempool *mb_pool)
415 {
416 	struct sfc_adapter *sa = dev->data->dev_private;
417 	int rc;
418 
419 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
420 		     rx_queue_id, nb_rx_desc, socket_id);
421 
422 	sfc_adapter_lock(sa);
423 
424 	rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
425 			  rx_conf, mb_pool);
426 	if (rc != 0)
427 		goto fail_rx_qinit;
428 
429 	dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
430 
431 	sfc_adapter_unlock(sa);
432 
433 	return 0;
434 
435 fail_rx_qinit:
436 	sfc_adapter_unlock(sa);
437 	SFC_ASSERT(rc > 0);
438 	return -rc;
439 }
440 
441 static void
442 sfc_rx_queue_release(void *queue)
443 {
444 	struct sfc_dp_rxq *dp_rxq = queue;
445 	struct sfc_rxq *rxq;
446 	struct sfc_adapter *sa;
447 	unsigned int sw_index;
448 
449 	if (dp_rxq == NULL)
450 		return;
451 
452 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
453 	sa = rxq->evq->sa;
454 	sfc_adapter_lock(sa);
455 
456 	sw_index = sfc_rxq_sw_index(rxq);
457 
458 	sfc_log_init(sa, "RxQ=%u", sw_index);
459 
460 	sa->eth_dev->data->rx_queues[sw_index] = NULL;
461 
462 	sfc_rx_qfini(sa, sw_index);
463 
464 	sfc_adapter_unlock(sa);
465 }
466 
467 static int
468 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
469 		   uint16_t nb_tx_desc, unsigned int socket_id,
470 		   const struct rte_eth_txconf *tx_conf)
471 {
472 	struct sfc_adapter *sa = dev->data->dev_private;
473 	int rc;
474 
475 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
476 		     tx_queue_id, nb_tx_desc, socket_id);
477 
478 	sfc_adapter_lock(sa);
479 
480 	rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
481 	if (rc != 0)
482 		goto fail_tx_qinit;
483 
484 	dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
485 
486 	sfc_adapter_unlock(sa);
487 	return 0;
488 
489 fail_tx_qinit:
490 	sfc_adapter_unlock(sa);
491 	SFC_ASSERT(rc > 0);
492 	return -rc;
493 }
494 
495 static void
496 sfc_tx_queue_release(void *queue)
497 {
498 	struct sfc_dp_txq *dp_txq = queue;
499 	struct sfc_txq *txq;
500 	unsigned int sw_index;
501 	struct sfc_adapter *sa;
502 
503 	if (dp_txq == NULL)
504 		return;
505 
506 	txq = sfc_txq_by_dp_txq(dp_txq);
507 	sw_index = sfc_txq_sw_index(txq);
508 
509 	SFC_ASSERT(txq->evq != NULL);
510 	sa = txq->evq->sa;
511 
512 	sfc_log_init(sa, "TxQ = %u", sw_index);
513 
514 	sfc_adapter_lock(sa);
515 
516 	SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
517 	sa->eth_dev->data->tx_queues[sw_index] = NULL;
518 
519 	sfc_tx_qfini(sa, sw_index);
520 
521 	sfc_adapter_unlock(sa);
522 }
523 
524 static int
525 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
526 {
527 	struct sfc_adapter *sa = dev->data->dev_private;
528 	struct sfc_port *port = &sa->port;
529 	uint64_t *mac_stats;
530 	int ret;
531 
532 	rte_spinlock_lock(&port->mac_stats_lock);
533 
534 	ret = sfc_port_update_mac_stats(sa);
535 	if (ret != 0)
536 		goto unlock;
537 
538 	mac_stats = port->mac_stats_buf;
539 
540 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
541 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
542 		stats->ipackets =
543 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
544 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
545 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
546 		stats->opackets =
547 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
548 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
549 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
550 		stats->ibytes =
551 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
552 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
553 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
554 		stats->obytes =
555 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
556 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
557 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
558 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
559 		stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
560 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
561 	} else {
562 		stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
563 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
564 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
565 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
566 		/*
567 		 * Take into account stats which are whenever supported
568 		 * on EF10. If some stat is not supported by current
569 		 * firmware variant or HW revision, it is guaranteed
570 		 * to be zero in mac_stats.
571 		 */
572 		stats->imissed =
573 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
574 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
575 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
576 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
577 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
578 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
579 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
580 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
581 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
582 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
583 		stats->ierrors =
584 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
585 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
586 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
587 		/* no oerrors counters supported on EF10 */
588 	}
589 
590 unlock:
591 	rte_spinlock_unlock(&port->mac_stats_lock);
592 	SFC_ASSERT(ret >= 0);
593 	return -ret;
594 }
595 
596 static void
597 sfc_stats_reset(struct rte_eth_dev *dev)
598 {
599 	struct sfc_adapter *sa = dev->data->dev_private;
600 	struct sfc_port *port = &sa->port;
601 	int rc;
602 
603 	if (sa->state != SFC_ADAPTER_STARTED) {
604 		/*
605 		 * The operation cannot be done if port is not started; it
606 		 * will be scheduled to be done during the next port start
607 		 */
608 		port->mac_stats_reset_pending = B_TRUE;
609 		return;
610 	}
611 
612 	rc = sfc_port_reset_mac_stats(sa);
613 	if (rc != 0)
614 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
615 }
616 
617 static int
618 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
619 	       unsigned int xstats_count)
620 {
621 	struct sfc_adapter *sa = dev->data->dev_private;
622 	struct sfc_port *port = &sa->port;
623 	uint64_t *mac_stats;
624 	int rc;
625 	unsigned int i;
626 	int nstats = 0;
627 
628 	rte_spinlock_lock(&port->mac_stats_lock);
629 
630 	rc = sfc_port_update_mac_stats(sa);
631 	if (rc != 0) {
632 		SFC_ASSERT(rc > 0);
633 		nstats = -rc;
634 		goto unlock;
635 	}
636 
637 	mac_stats = port->mac_stats_buf;
638 
639 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
640 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
641 			if (xstats != NULL && nstats < (int)xstats_count) {
642 				xstats[nstats].id = nstats;
643 				xstats[nstats].value = mac_stats[i];
644 			}
645 			nstats++;
646 		}
647 	}
648 
649 unlock:
650 	rte_spinlock_unlock(&port->mac_stats_lock);
651 
652 	return nstats;
653 }
654 
655 static int
656 sfc_xstats_get_names(struct rte_eth_dev *dev,
657 		     struct rte_eth_xstat_name *xstats_names,
658 		     unsigned int xstats_count)
659 {
660 	struct sfc_adapter *sa = dev->data->dev_private;
661 	struct sfc_port *port = &sa->port;
662 	unsigned int i;
663 	unsigned int nstats = 0;
664 
665 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
666 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
667 			if (xstats_names != NULL && nstats < xstats_count)
668 				strncpy(xstats_names[nstats].name,
669 					efx_mac_stat_name(sa->nic, i),
670 					sizeof(xstats_names[0].name));
671 			nstats++;
672 		}
673 	}
674 
675 	return nstats;
676 }
677 
678 static int
679 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
680 		     uint64_t *values, unsigned int n)
681 {
682 	struct sfc_adapter *sa = dev->data->dev_private;
683 	struct sfc_port *port = &sa->port;
684 	uint64_t *mac_stats;
685 	unsigned int nb_supported = 0;
686 	unsigned int nb_written = 0;
687 	unsigned int i;
688 	int ret;
689 	int rc;
690 
691 	if (unlikely(values == NULL) ||
692 	    unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
693 		return port->mac_stats_nb_supported;
694 
695 	rte_spinlock_lock(&port->mac_stats_lock);
696 
697 	rc = sfc_port_update_mac_stats(sa);
698 	if (rc != 0) {
699 		SFC_ASSERT(rc > 0);
700 		ret = -rc;
701 		goto unlock;
702 	}
703 
704 	mac_stats = port->mac_stats_buf;
705 
706 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
707 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
708 			continue;
709 
710 		if ((ids == NULL) || (ids[nb_written] == nb_supported))
711 			values[nb_written++] = mac_stats[i];
712 
713 		++nb_supported;
714 	}
715 
716 	ret = nb_written;
717 
718 unlock:
719 	rte_spinlock_unlock(&port->mac_stats_lock);
720 
721 	return ret;
722 }
723 
724 static int
725 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
726 			   struct rte_eth_xstat_name *xstats_names,
727 			   const uint64_t *ids, unsigned int size)
728 {
729 	struct sfc_adapter *sa = dev->data->dev_private;
730 	struct sfc_port *port = &sa->port;
731 	unsigned int nb_supported = 0;
732 	unsigned int nb_written = 0;
733 	unsigned int i;
734 
735 	if (unlikely(xstats_names == NULL) ||
736 	    unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
737 		return port->mac_stats_nb_supported;
738 
739 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
740 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
741 			continue;
742 
743 		if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
744 			char *name = xstats_names[nb_written++].name;
745 
746 			strncpy(name, efx_mac_stat_name(sa->nic, i),
747 				sizeof(xstats_names[0].name));
748 			name[sizeof(xstats_names[0].name) - 1] = '\0';
749 		}
750 
751 		++nb_supported;
752 	}
753 
754 	return nb_written;
755 }
756 
757 static int
758 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
759 {
760 	struct sfc_adapter *sa = dev->data->dev_private;
761 	unsigned int wanted_fc, link_fc;
762 
763 	memset(fc_conf, 0, sizeof(*fc_conf));
764 
765 	sfc_adapter_lock(sa);
766 
767 	if (sa->state == SFC_ADAPTER_STARTED)
768 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
769 	else
770 		link_fc = sa->port.flow_ctrl;
771 
772 	switch (link_fc) {
773 	case 0:
774 		fc_conf->mode = RTE_FC_NONE;
775 		break;
776 	case EFX_FCNTL_RESPOND:
777 		fc_conf->mode = RTE_FC_RX_PAUSE;
778 		break;
779 	case EFX_FCNTL_GENERATE:
780 		fc_conf->mode = RTE_FC_TX_PAUSE;
781 		break;
782 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
783 		fc_conf->mode = RTE_FC_FULL;
784 		break;
785 	default:
786 		sfc_err(sa, "%s: unexpected flow control value %#x",
787 			__func__, link_fc);
788 	}
789 
790 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
791 
792 	sfc_adapter_unlock(sa);
793 
794 	return 0;
795 }
796 
797 static int
798 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
799 {
800 	struct sfc_adapter *sa = dev->data->dev_private;
801 	struct sfc_port *port = &sa->port;
802 	unsigned int fcntl;
803 	int rc;
804 
805 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
806 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
807 	    fc_conf->mac_ctrl_frame_fwd != 0) {
808 		sfc_err(sa, "unsupported flow control settings specified");
809 		rc = EINVAL;
810 		goto fail_inval;
811 	}
812 
813 	switch (fc_conf->mode) {
814 	case RTE_FC_NONE:
815 		fcntl = 0;
816 		break;
817 	case RTE_FC_RX_PAUSE:
818 		fcntl = EFX_FCNTL_RESPOND;
819 		break;
820 	case RTE_FC_TX_PAUSE:
821 		fcntl = EFX_FCNTL_GENERATE;
822 		break;
823 	case RTE_FC_FULL:
824 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
825 		break;
826 	default:
827 		rc = EINVAL;
828 		goto fail_inval;
829 	}
830 
831 	sfc_adapter_lock(sa);
832 
833 	if (sa->state == SFC_ADAPTER_STARTED) {
834 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
835 		if (rc != 0)
836 			goto fail_mac_fcntl_set;
837 	}
838 
839 	port->flow_ctrl = fcntl;
840 	port->flow_ctrl_autoneg = fc_conf->autoneg;
841 
842 	sfc_adapter_unlock(sa);
843 
844 	return 0;
845 
846 fail_mac_fcntl_set:
847 	sfc_adapter_unlock(sa);
848 fail_inval:
849 	SFC_ASSERT(rc > 0);
850 	return -rc;
851 }
852 
853 static int
854 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
855 {
856 	struct sfc_adapter *sa = dev->data->dev_private;
857 	size_t pdu = EFX_MAC_PDU(mtu);
858 	size_t old_pdu;
859 	int rc;
860 
861 	sfc_log_init(sa, "mtu=%u", mtu);
862 
863 	rc = EINVAL;
864 	if (pdu < EFX_MAC_PDU_MIN) {
865 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
866 			(unsigned int)mtu, (unsigned int)pdu,
867 			EFX_MAC_PDU_MIN);
868 		goto fail_inval;
869 	}
870 	if (pdu > EFX_MAC_PDU_MAX) {
871 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
872 			(unsigned int)mtu, (unsigned int)pdu,
873 			EFX_MAC_PDU_MAX);
874 		goto fail_inval;
875 	}
876 
877 	sfc_adapter_lock(sa);
878 
879 	if (pdu != sa->port.pdu) {
880 		if (sa->state == SFC_ADAPTER_STARTED) {
881 			sfc_stop(sa);
882 
883 			old_pdu = sa->port.pdu;
884 			sa->port.pdu = pdu;
885 			rc = sfc_start(sa);
886 			if (rc != 0)
887 				goto fail_start;
888 		} else {
889 			sa->port.pdu = pdu;
890 		}
891 	}
892 
893 	/*
894 	 * The driver does not use it, but other PMDs update jumbo_frame
895 	 * flag and max_rx_pkt_len when MTU is set.
896 	 */
897 	dev->data->dev_conf.rxmode.jumbo_frame = (mtu > ETHER_MAX_LEN);
898 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
899 
900 	sfc_adapter_unlock(sa);
901 
902 	sfc_log_init(sa, "done");
903 	return 0;
904 
905 fail_start:
906 	sa->port.pdu = old_pdu;
907 	if (sfc_start(sa) != 0)
908 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
909 			"PDU max size - port is stopped",
910 			(unsigned int)pdu, (unsigned int)old_pdu);
911 	sfc_adapter_unlock(sa);
912 
913 fail_inval:
914 	sfc_log_init(sa, "failed %d", rc);
915 	SFC_ASSERT(rc > 0);
916 	return -rc;
917 }
918 static void
919 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
920 {
921 	struct sfc_adapter *sa = dev->data->dev_private;
922 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
923 	struct sfc_port *port = &sa->port;
924 	int rc;
925 
926 	sfc_adapter_lock(sa);
927 
928 	if (port->isolated) {
929 		sfc_err(sa, "isolated mode is active on the port");
930 		sfc_err(sa, "will not set MAC address");
931 		goto unlock;
932 	}
933 
934 	if (sa->state != SFC_ADAPTER_STARTED) {
935 		sfc_info(sa, "the port is not started");
936 		sfc_info(sa, "the new MAC address will be set on port start");
937 
938 		goto unlock;
939 	}
940 
941 	if (encp->enc_allow_set_mac_with_installed_filters) {
942 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
943 		if (rc != 0) {
944 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
945 			goto unlock;
946 		}
947 
948 		/*
949 		 * Changing the MAC address by means of MCDI request
950 		 * has no effect on received traffic, therefore
951 		 * we also need to update unicast filters
952 		 */
953 		rc = sfc_set_rx_mode(sa);
954 		if (rc != 0)
955 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
956 	} else {
957 		sfc_warn(sa, "cannot set MAC address with filters installed");
958 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
959 		sfc_warn(sa, "(some traffic may be dropped)");
960 
961 		/*
962 		 * Since setting MAC address with filters installed is not
963 		 * allowed on the adapter, one needs to simply restart adapter
964 		 * so that the new MAC address will be taken from an outer
965 		 * storage and set flawlessly by means of sfc_start() call
966 		 */
967 		sfc_stop(sa);
968 		rc = sfc_start(sa);
969 		if (rc != 0)
970 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
971 	}
972 
973 unlock:
974 	sfc_adapter_unlock(sa);
975 }
976 
977 
978 static int
979 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
980 		     uint32_t nb_mc_addr)
981 {
982 	struct sfc_adapter *sa = dev->data->dev_private;
983 	struct sfc_port *port = &sa->port;
984 	uint8_t *mc_addrs = port->mcast_addrs;
985 	int rc;
986 	unsigned int i;
987 
988 	if (port->isolated) {
989 		sfc_err(sa, "isolated mode is active on the port");
990 		sfc_err(sa, "will not set multicast address list");
991 		return -ENOTSUP;
992 	}
993 
994 	if (mc_addrs == NULL)
995 		return -ENOBUFS;
996 
997 	if (nb_mc_addr > port->max_mcast_addrs) {
998 		sfc_err(sa, "too many multicast addresses: %u > %u",
999 			 nb_mc_addr, port->max_mcast_addrs);
1000 		return -EINVAL;
1001 	}
1002 
1003 	for (i = 0; i < nb_mc_addr; ++i) {
1004 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1005 				 EFX_MAC_ADDR_LEN);
1006 		mc_addrs += EFX_MAC_ADDR_LEN;
1007 	}
1008 
1009 	port->nb_mcast_addrs = nb_mc_addr;
1010 
1011 	if (sa->state != SFC_ADAPTER_STARTED)
1012 		return 0;
1013 
1014 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1015 					port->nb_mcast_addrs);
1016 	if (rc != 0)
1017 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1018 
1019 	SFC_ASSERT(rc > 0);
1020 	return -rc;
1021 }
1022 
1023 /*
1024  * The function is used by the secondary process as well. It must not
1025  * use any process-local pointers from the adapter data.
1026  */
1027 static void
1028 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1029 		      struct rte_eth_rxq_info *qinfo)
1030 {
1031 	struct sfc_adapter *sa = dev->data->dev_private;
1032 	struct sfc_rxq_info *rxq_info;
1033 	struct sfc_rxq *rxq;
1034 
1035 	sfc_adapter_lock(sa);
1036 
1037 	SFC_ASSERT(rx_queue_id < sa->rxq_count);
1038 
1039 	rxq_info = &sa->rxq_info[rx_queue_id];
1040 	rxq = rxq_info->rxq;
1041 	SFC_ASSERT(rxq != NULL);
1042 
1043 	qinfo->mp = rxq->refill_mb_pool;
1044 	qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1045 	qinfo->conf.rx_drop_en = 1;
1046 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1047 	qinfo->scattered_rx = (rxq_info->type == EFX_RXQ_TYPE_SCATTER);
1048 	qinfo->nb_desc = rxq_info->entries;
1049 
1050 	sfc_adapter_unlock(sa);
1051 }
1052 
1053 /*
1054  * The function is used by the secondary process as well. It must not
1055  * use any process-local pointers from the adapter data.
1056  */
1057 static void
1058 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1059 		      struct rte_eth_txq_info *qinfo)
1060 {
1061 	struct sfc_adapter *sa = dev->data->dev_private;
1062 	struct sfc_txq_info *txq_info;
1063 
1064 	sfc_adapter_lock(sa);
1065 
1066 	SFC_ASSERT(tx_queue_id < sa->txq_count);
1067 
1068 	txq_info = &sa->txq_info[tx_queue_id];
1069 	SFC_ASSERT(txq_info->txq != NULL);
1070 
1071 	memset(qinfo, 0, sizeof(*qinfo));
1072 
1073 	qinfo->conf.txq_flags = txq_info->txq->flags;
1074 	qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1075 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1076 	qinfo->nb_desc = txq_info->entries;
1077 
1078 	sfc_adapter_unlock(sa);
1079 }
1080 
1081 static uint32_t
1082 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1083 {
1084 	struct sfc_adapter *sa = dev->data->dev_private;
1085 
1086 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1087 
1088 	return sfc_rx_qdesc_npending(sa, rx_queue_id);
1089 }
1090 
1091 static int
1092 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1093 {
1094 	struct sfc_dp_rxq *dp_rxq = queue;
1095 
1096 	return sfc_rx_qdesc_done(dp_rxq, offset);
1097 }
1098 
1099 static int
1100 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1101 {
1102 	struct sfc_dp_rxq *dp_rxq = queue;
1103 	struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1104 
1105 	return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1106 }
1107 
1108 static int
1109 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1110 {
1111 	struct sfc_dp_txq *dp_txq = queue;
1112 	struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1113 
1114 	return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1115 }
1116 
1117 static int
1118 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1119 {
1120 	struct sfc_adapter *sa = dev->data->dev_private;
1121 	int rc;
1122 
1123 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1124 
1125 	sfc_adapter_lock(sa);
1126 
1127 	rc = EINVAL;
1128 	if (sa->state != SFC_ADAPTER_STARTED)
1129 		goto fail_not_started;
1130 
1131 	rc = sfc_rx_qstart(sa, rx_queue_id);
1132 	if (rc != 0)
1133 		goto fail_rx_qstart;
1134 
1135 	sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1136 
1137 	sfc_adapter_unlock(sa);
1138 
1139 	return 0;
1140 
1141 fail_rx_qstart:
1142 fail_not_started:
1143 	sfc_adapter_unlock(sa);
1144 	SFC_ASSERT(rc > 0);
1145 	return -rc;
1146 }
1147 
1148 static int
1149 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1150 {
1151 	struct sfc_adapter *sa = dev->data->dev_private;
1152 
1153 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1154 
1155 	sfc_adapter_lock(sa);
1156 	sfc_rx_qstop(sa, rx_queue_id);
1157 
1158 	sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1159 
1160 	sfc_adapter_unlock(sa);
1161 
1162 	return 0;
1163 }
1164 
1165 static int
1166 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1167 {
1168 	struct sfc_adapter *sa = dev->data->dev_private;
1169 	int rc;
1170 
1171 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1172 
1173 	sfc_adapter_lock(sa);
1174 
1175 	rc = EINVAL;
1176 	if (sa->state != SFC_ADAPTER_STARTED)
1177 		goto fail_not_started;
1178 
1179 	rc = sfc_tx_qstart(sa, tx_queue_id);
1180 	if (rc != 0)
1181 		goto fail_tx_qstart;
1182 
1183 	sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1184 
1185 	sfc_adapter_unlock(sa);
1186 	return 0;
1187 
1188 fail_tx_qstart:
1189 
1190 fail_not_started:
1191 	sfc_adapter_unlock(sa);
1192 	SFC_ASSERT(rc > 0);
1193 	return -rc;
1194 }
1195 
1196 static int
1197 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1198 {
1199 	struct sfc_adapter *sa = dev->data->dev_private;
1200 
1201 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1202 
1203 	sfc_adapter_lock(sa);
1204 
1205 	sfc_tx_qstop(sa, tx_queue_id);
1206 
1207 	sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1208 
1209 	sfc_adapter_unlock(sa);
1210 	return 0;
1211 }
1212 
1213 #if EFSYS_OPT_RX_SCALE
1214 static int
1215 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1216 			  struct rte_eth_rss_conf *rss_conf)
1217 {
1218 	struct sfc_adapter *sa = dev->data->dev_private;
1219 	struct sfc_port *port = &sa->port;
1220 
1221 	if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1222 		return -ENOTSUP;
1223 
1224 	if (sa->rss_channels == 0)
1225 		return -EINVAL;
1226 
1227 	sfc_adapter_lock(sa);
1228 
1229 	/*
1230 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1231 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1232 	 * flags which corresponds to the active EFX configuration stored
1233 	 * locally in 'sfc_adapter' and kept up-to-date
1234 	 */
1235 	rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1236 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1237 	if (rss_conf->rss_key != NULL)
1238 		rte_memcpy(rss_conf->rss_key, sa->rss_key, EFX_RSS_KEY_SIZE);
1239 
1240 	sfc_adapter_unlock(sa);
1241 
1242 	return 0;
1243 }
1244 
1245 static int
1246 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1247 			struct rte_eth_rss_conf *rss_conf)
1248 {
1249 	struct sfc_adapter *sa = dev->data->dev_private;
1250 	struct sfc_port *port = &sa->port;
1251 	unsigned int efx_hash_types;
1252 	int rc = 0;
1253 
1254 	if (port->isolated)
1255 		return -ENOTSUP;
1256 
1257 	if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1258 		sfc_err(sa, "RSS is not available");
1259 		return -ENOTSUP;
1260 	}
1261 
1262 	if (sa->rss_channels == 0) {
1263 		sfc_err(sa, "RSS is not configured");
1264 		return -EINVAL;
1265 	}
1266 
1267 	if ((rss_conf->rss_key != NULL) &&
1268 	    (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1269 		sfc_err(sa, "RSS key size is wrong (should be %lu)",
1270 			sizeof(sa->rss_key));
1271 		return -EINVAL;
1272 	}
1273 
1274 	if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1275 		sfc_err(sa, "unsupported hash functions requested");
1276 		return -EINVAL;
1277 	}
1278 
1279 	sfc_adapter_lock(sa);
1280 
1281 	efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1282 
1283 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1284 				   EFX_RX_HASHALG_TOEPLITZ,
1285 				   efx_hash_types, B_TRUE);
1286 	if (rc != 0)
1287 		goto fail_scale_mode_set;
1288 
1289 	if (rss_conf->rss_key != NULL) {
1290 		if (sa->state == SFC_ADAPTER_STARTED) {
1291 			rc = efx_rx_scale_key_set(sa->nic,
1292 						  EFX_RSS_CONTEXT_DEFAULT,
1293 						  rss_conf->rss_key,
1294 						  sizeof(sa->rss_key));
1295 			if (rc != 0)
1296 				goto fail_scale_key_set;
1297 		}
1298 
1299 		rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1300 	}
1301 
1302 	sa->rss_hash_types = efx_hash_types;
1303 
1304 	sfc_adapter_unlock(sa);
1305 
1306 	return 0;
1307 
1308 fail_scale_key_set:
1309 	if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1310 				  EFX_RX_HASHALG_TOEPLITZ,
1311 				  sa->rss_hash_types, B_TRUE) != 0)
1312 		sfc_err(sa, "failed to restore RSS mode");
1313 
1314 fail_scale_mode_set:
1315 	sfc_adapter_unlock(sa);
1316 	return -rc;
1317 }
1318 
1319 static int
1320 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1321 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1322 		       uint16_t reta_size)
1323 {
1324 	struct sfc_adapter *sa = dev->data->dev_private;
1325 	struct sfc_port *port = &sa->port;
1326 	int entry;
1327 
1328 	if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1329 		return -ENOTSUP;
1330 
1331 	if (sa->rss_channels == 0)
1332 		return -EINVAL;
1333 
1334 	if (reta_size != EFX_RSS_TBL_SIZE)
1335 		return -EINVAL;
1336 
1337 	sfc_adapter_lock(sa);
1338 
1339 	for (entry = 0; entry < reta_size; entry++) {
1340 		int grp = entry / RTE_RETA_GROUP_SIZE;
1341 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1342 
1343 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1344 			reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1345 	}
1346 
1347 	sfc_adapter_unlock(sa);
1348 
1349 	return 0;
1350 }
1351 
1352 static int
1353 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1354 			struct rte_eth_rss_reta_entry64 *reta_conf,
1355 			uint16_t reta_size)
1356 {
1357 	struct sfc_adapter *sa = dev->data->dev_private;
1358 	struct sfc_port *port = &sa->port;
1359 	unsigned int *rss_tbl_new;
1360 	uint16_t entry;
1361 	int rc = 0;
1362 
1363 
1364 	if (port->isolated)
1365 		return -ENOTSUP;
1366 
1367 	if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1368 		sfc_err(sa, "RSS is not available");
1369 		return -ENOTSUP;
1370 	}
1371 
1372 	if (sa->rss_channels == 0) {
1373 		sfc_err(sa, "RSS is not configured");
1374 		return -EINVAL;
1375 	}
1376 
1377 	if (reta_size != EFX_RSS_TBL_SIZE) {
1378 		sfc_err(sa, "RETA size is wrong (should be %u)",
1379 			EFX_RSS_TBL_SIZE);
1380 		return -EINVAL;
1381 	}
1382 
1383 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1384 	if (rss_tbl_new == NULL)
1385 		return -ENOMEM;
1386 
1387 	sfc_adapter_lock(sa);
1388 
1389 	rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1390 
1391 	for (entry = 0; entry < reta_size; entry++) {
1392 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1393 		struct rte_eth_rss_reta_entry64 *grp;
1394 
1395 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1396 
1397 		if (grp->mask & (1ull << grp_idx)) {
1398 			if (grp->reta[grp_idx] >= sa->rss_channels) {
1399 				rc = EINVAL;
1400 				goto bad_reta_entry;
1401 			}
1402 			rss_tbl_new[entry] = grp->reta[grp_idx];
1403 		}
1404 	}
1405 
1406 	if (sa->state == SFC_ADAPTER_STARTED) {
1407 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1408 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1409 		if (rc != 0)
1410 			goto fail_scale_tbl_set;
1411 	}
1412 
1413 	rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1414 
1415 fail_scale_tbl_set:
1416 bad_reta_entry:
1417 	sfc_adapter_unlock(sa);
1418 
1419 	rte_free(rss_tbl_new);
1420 
1421 	SFC_ASSERT(rc >= 0);
1422 	return -rc;
1423 }
1424 #endif
1425 
1426 static int
1427 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1428 		    enum rte_filter_op filter_op,
1429 		    void *arg)
1430 {
1431 	struct sfc_adapter *sa = dev->data->dev_private;
1432 	int rc = ENOTSUP;
1433 
1434 	sfc_log_init(sa, "entry");
1435 
1436 	switch (filter_type) {
1437 	case RTE_ETH_FILTER_NONE:
1438 		sfc_err(sa, "Global filters configuration not supported");
1439 		break;
1440 	case RTE_ETH_FILTER_MACVLAN:
1441 		sfc_err(sa, "MACVLAN filters not supported");
1442 		break;
1443 	case RTE_ETH_FILTER_ETHERTYPE:
1444 		sfc_err(sa, "EtherType filters not supported");
1445 		break;
1446 	case RTE_ETH_FILTER_FLEXIBLE:
1447 		sfc_err(sa, "Flexible filters not supported");
1448 		break;
1449 	case RTE_ETH_FILTER_SYN:
1450 		sfc_err(sa, "SYN filters not supported");
1451 		break;
1452 	case RTE_ETH_FILTER_NTUPLE:
1453 		sfc_err(sa, "NTUPLE filters not supported");
1454 		break;
1455 	case RTE_ETH_FILTER_TUNNEL:
1456 		sfc_err(sa, "Tunnel filters not supported");
1457 		break;
1458 	case RTE_ETH_FILTER_FDIR:
1459 		sfc_err(sa, "Flow Director filters not supported");
1460 		break;
1461 	case RTE_ETH_FILTER_HASH:
1462 		sfc_err(sa, "Hash filters not supported");
1463 		break;
1464 	case RTE_ETH_FILTER_GENERIC:
1465 		if (filter_op != RTE_ETH_FILTER_GET) {
1466 			rc = EINVAL;
1467 		} else {
1468 			*(const void **)arg = &sfc_flow_ops;
1469 			rc = 0;
1470 		}
1471 		break;
1472 	default:
1473 		sfc_err(sa, "Unknown filter type %u", filter_type);
1474 		break;
1475 	}
1476 
1477 	sfc_log_init(sa, "exit: %d", -rc);
1478 	SFC_ASSERT(rc >= 0);
1479 	return -rc;
1480 }
1481 
1482 static const struct eth_dev_ops sfc_eth_dev_ops = {
1483 	.dev_configure			= sfc_dev_configure,
1484 	.dev_start			= sfc_dev_start,
1485 	.dev_stop			= sfc_dev_stop,
1486 	.dev_set_link_up		= sfc_dev_set_link_up,
1487 	.dev_set_link_down		= sfc_dev_set_link_down,
1488 	.dev_close			= sfc_dev_close,
1489 	.promiscuous_enable		= sfc_dev_promisc_enable,
1490 	.promiscuous_disable		= sfc_dev_promisc_disable,
1491 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1492 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1493 	.link_update			= sfc_dev_link_update,
1494 	.stats_get			= sfc_stats_get,
1495 	.stats_reset			= sfc_stats_reset,
1496 	.xstats_get			= sfc_xstats_get,
1497 	.xstats_reset			= sfc_stats_reset,
1498 	.xstats_get_names		= sfc_xstats_get_names,
1499 	.dev_infos_get			= sfc_dev_infos_get,
1500 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1501 	.mtu_set			= sfc_dev_set_mtu,
1502 	.rx_queue_start			= sfc_rx_queue_start,
1503 	.rx_queue_stop			= sfc_rx_queue_stop,
1504 	.tx_queue_start			= sfc_tx_queue_start,
1505 	.tx_queue_stop			= sfc_tx_queue_stop,
1506 	.rx_queue_setup			= sfc_rx_queue_setup,
1507 	.rx_queue_release		= sfc_rx_queue_release,
1508 	.rx_queue_count			= sfc_rx_queue_count,
1509 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1510 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1511 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1512 	.tx_queue_setup			= sfc_tx_queue_setup,
1513 	.tx_queue_release		= sfc_tx_queue_release,
1514 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1515 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1516 	.mac_addr_set			= sfc_mac_addr_set,
1517 #if EFSYS_OPT_RX_SCALE
1518 	.reta_update			= sfc_dev_rss_reta_update,
1519 	.reta_query			= sfc_dev_rss_reta_query,
1520 	.rss_hash_update		= sfc_dev_rss_hash_update,
1521 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1522 #endif
1523 	.filter_ctrl			= sfc_dev_filter_ctrl,
1524 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1525 	.rxq_info_get			= sfc_rx_queue_info_get,
1526 	.txq_info_get			= sfc_tx_queue_info_get,
1527 	.fw_version_get			= sfc_fw_version_get,
1528 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1529 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1530 };
1531 
1532 /**
1533  * Duplicate a string in potentially shared memory required for
1534  * multi-process support.
1535  *
1536  * strdup() allocates from process-local heap/memory.
1537  */
1538 static char *
1539 sfc_strdup(const char *str)
1540 {
1541 	size_t size;
1542 	char *copy;
1543 
1544 	if (str == NULL)
1545 		return NULL;
1546 
1547 	size = strlen(str) + 1;
1548 	copy = rte_malloc(__func__, size, 0);
1549 	if (copy != NULL)
1550 		rte_memcpy(copy, str, size);
1551 
1552 	return copy;
1553 }
1554 
1555 static int
1556 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1557 {
1558 	struct sfc_adapter *sa = dev->data->dev_private;
1559 	unsigned int avail_caps = 0;
1560 	const char *rx_name = NULL;
1561 	const char *tx_name = NULL;
1562 	int rc;
1563 
1564 	switch (sa->family) {
1565 	case EFX_FAMILY_HUNTINGTON:
1566 	case EFX_FAMILY_MEDFORD:
1567 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1568 		break;
1569 	default:
1570 		break;
1571 	}
1572 
1573 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1574 				sfc_kvarg_string_handler, &rx_name);
1575 	if (rc != 0)
1576 		goto fail_kvarg_rx_datapath;
1577 
1578 	if (rx_name != NULL) {
1579 		sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1580 		if (sa->dp_rx == NULL) {
1581 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1582 			rc = ENOENT;
1583 			goto fail_dp_rx;
1584 		}
1585 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1586 			sfc_err(sa,
1587 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1588 				rx_name);
1589 			rc = EINVAL;
1590 			goto fail_dp_rx_caps;
1591 		}
1592 	} else {
1593 		sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1594 		if (sa->dp_rx == NULL) {
1595 			sfc_err(sa, "Rx datapath by caps %#x not found",
1596 				avail_caps);
1597 			rc = ENOENT;
1598 			goto fail_dp_rx;
1599 		}
1600 	}
1601 
1602 	sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1603 	if (sa->dp_rx_name == NULL) {
1604 		rc = ENOMEM;
1605 		goto fail_dp_rx_name;
1606 	}
1607 
1608 	sfc_info(sa, "use %s Rx datapath", sa->dp_rx_name);
1609 
1610 	dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1611 
1612 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1613 				sfc_kvarg_string_handler, &tx_name);
1614 	if (rc != 0)
1615 		goto fail_kvarg_tx_datapath;
1616 
1617 	if (tx_name != NULL) {
1618 		sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1619 		if (sa->dp_tx == NULL) {
1620 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1621 			rc = ENOENT;
1622 			goto fail_dp_tx;
1623 		}
1624 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1625 			sfc_err(sa,
1626 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1627 				tx_name);
1628 			rc = EINVAL;
1629 			goto fail_dp_tx_caps;
1630 		}
1631 	} else {
1632 		sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1633 		if (sa->dp_tx == NULL) {
1634 			sfc_err(sa, "Tx datapath by caps %#x not found",
1635 				avail_caps);
1636 			rc = ENOENT;
1637 			goto fail_dp_tx;
1638 		}
1639 	}
1640 
1641 	sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1642 	if (sa->dp_tx_name == NULL) {
1643 		rc = ENOMEM;
1644 		goto fail_dp_tx_name;
1645 	}
1646 
1647 	sfc_info(sa, "use %s Tx datapath", sa->dp_tx_name);
1648 
1649 	dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1650 
1651 	dev->dev_ops = &sfc_eth_dev_ops;
1652 
1653 	return 0;
1654 
1655 fail_dp_tx_name:
1656 fail_dp_tx_caps:
1657 	sa->dp_tx = NULL;
1658 
1659 fail_dp_tx:
1660 fail_kvarg_tx_datapath:
1661 	rte_free(sa->dp_rx_name);
1662 	sa->dp_rx_name = NULL;
1663 
1664 fail_dp_rx_name:
1665 fail_dp_rx_caps:
1666 	sa->dp_rx = NULL;
1667 
1668 fail_dp_rx:
1669 fail_kvarg_rx_datapath:
1670 	return rc;
1671 }
1672 
1673 static void
1674 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1675 {
1676 	struct sfc_adapter *sa = dev->data->dev_private;
1677 
1678 	dev->dev_ops = NULL;
1679 	dev->rx_pkt_burst = NULL;
1680 	dev->tx_pkt_burst = NULL;
1681 
1682 	rte_free(sa->dp_tx_name);
1683 	sa->dp_tx_name = NULL;
1684 	sa->dp_tx = NULL;
1685 
1686 	rte_free(sa->dp_rx_name);
1687 	sa->dp_rx_name = NULL;
1688 	sa->dp_rx = NULL;
1689 }
1690 
1691 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1692 	.rxq_info_get			= sfc_rx_queue_info_get,
1693 	.txq_info_get			= sfc_tx_queue_info_get,
1694 };
1695 
1696 static int
1697 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1698 {
1699 	/*
1700 	 * Device private data has really many process-local pointers.
1701 	 * Below code should be extremely careful to use data located
1702 	 * in shared memory only.
1703 	 */
1704 	struct sfc_adapter *sa = dev->data->dev_private;
1705 	const struct sfc_dp_rx *dp_rx;
1706 	const struct sfc_dp_tx *dp_tx;
1707 	int rc;
1708 
1709 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1710 	if (dp_rx == NULL) {
1711 		sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1712 		rc = ENOENT;
1713 		goto fail_dp_rx;
1714 	}
1715 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1716 		sfc_err(sa, "%s Rx datapath does not support multi-process",
1717 			sa->dp_tx_name);
1718 		rc = EINVAL;
1719 		goto fail_dp_rx_multi_process;
1720 	}
1721 
1722 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1723 	if (dp_tx == NULL) {
1724 		sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1725 		rc = ENOENT;
1726 		goto fail_dp_tx;
1727 	}
1728 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1729 		sfc_err(sa, "%s Tx datapath does not support multi-process",
1730 			sa->dp_tx_name);
1731 		rc = EINVAL;
1732 		goto fail_dp_tx_multi_process;
1733 	}
1734 
1735 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1736 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1737 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
1738 
1739 	return 0;
1740 
1741 fail_dp_tx_multi_process:
1742 fail_dp_tx:
1743 fail_dp_rx_multi_process:
1744 fail_dp_rx:
1745 	return rc;
1746 }
1747 
1748 static void
1749 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1750 {
1751 	dev->dev_ops = NULL;
1752 	dev->tx_pkt_burst = NULL;
1753 	dev->rx_pkt_burst = NULL;
1754 }
1755 
1756 static void
1757 sfc_register_dp(void)
1758 {
1759 	/* Register once */
1760 	if (TAILQ_EMPTY(&sfc_dp_head)) {
1761 		/* Prefer EF10 datapath */
1762 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1763 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1764 
1765 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1766 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1767 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1768 	}
1769 }
1770 
1771 static int
1772 sfc_eth_dev_init(struct rte_eth_dev *dev)
1773 {
1774 	struct sfc_adapter *sa = dev->data->dev_private;
1775 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1776 	int rc;
1777 	const efx_nic_cfg_t *encp;
1778 	const struct ether_addr *from;
1779 
1780 	sfc_register_dp();
1781 
1782 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1783 		return -sfc_eth_dev_secondary_set_ops(dev);
1784 
1785 	/* Required for logging */
1786 	sa->pci_addr = pci_dev->addr;
1787 	sa->port_id = dev->data->port_id;
1788 
1789 	sa->eth_dev = dev;
1790 
1791 	/* Copy PCI device info to the dev->data */
1792 	rte_eth_copy_pci_info(dev, pci_dev);
1793 
1794 	dev->data->dev_flags |= RTE_ETH_DEV_DETACHABLE;
1795 
1796 	rc = sfc_kvargs_parse(sa);
1797 	if (rc != 0)
1798 		goto fail_kvargs_parse;
1799 
1800 	rc = sfc_kvargs_process(sa, SFC_KVARG_DEBUG_INIT,
1801 				sfc_kvarg_bool_handler, &sa->debug_init);
1802 	if (rc != 0)
1803 		goto fail_kvarg_debug_init;
1804 
1805 	sfc_log_init(sa, "entry");
1806 
1807 	dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1808 	if (dev->data->mac_addrs == NULL) {
1809 		rc = ENOMEM;
1810 		goto fail_mac_addrs;
1811 	}
1812 
1813 	sfc_adapter_lock_init(sa);
1814 	sfc_adapter_lock(sa);
1815 
1816 	sfc_log_init(sa, "probing");
1817 	rc = sfc_probe(sa);
1818 	if (rc != 0)
1819 		goto fail_probe;
1820 
1821 	sfc_log_init(sa, "set device ops");
1822 	rc = sfc_eth_dev_set_ops(dev);
1823 	if (rc != 0)
1824 		goto fail_set_ops;
1825 
1826 	sfc_log_init(sa, "attaching");
1827 	rc = sfc_attach(sa);
1828 	if (rc != 0)
1829 		goto fail_attach;
1830 
1831 	encp = efx_nic_cfg_get(sa->nic);
1832 
1833 	/*
1834 	 * The arguments are really reverse order in comparison to
1835 	 * Linux kernel. Copy from NIC config to Ethernet device data.
1836 	 */
1837 	from = (const struct ether_addr *)(encp->enc_mac_addr);
1838 	ether_addr_copy(from, &dev->data->mac_addrs[0]);
1839 
1840 	sfc_adapter_unlock(sa);
1841 
1842 	sfc_log_init(sa, "done");
1843 	return 0;
1844 
1845 fail_attach:
1846 	sfc_eth_dev_clear_ops(dev);
1847 
1848 fail_set_ops:
1849 	sfc_unprobe(sa);
1850 
1851 fail_probe:
1852 	sfc_adapter_unlock(sa);
1853 	sfc_adapter_lock_fini(sa);
1854 	rte_free(dev->data->mac_addrs);
1855 	dev->data->mac_addrs = NULL;
1856 
1857 fail_mac_addrs:
1858 fail_kvarg_debug_init:
1859 	sfc_kvargs_cleanup(sa);
1860 
1861 fail_kvargs_parse:
1862 	sfc_log_init(sa, "failed %d", rc);
1863 	SFC_ASSERT(rc > 0);
1864 	return -rc;
1865 }
1866 
1867 static int
1868 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
1869 {
1870 	struct sfc_adapter *sa;
1871 
1872 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1873 		sfc_eth_dev_secondary_clear_ops(dev);
1874 		return 0;
1875 	}
1876 
1877 	sa = dev->data->dev_private;
1878 	sfc_log_init(sa, "entry");
1879 
1880 	sfc_adapter_lock(sa);
1881 
1882 	sfc_eth_dev_clear_ops(dev);
1883 
1884 	sfc_detach(sa);
1885 	sfc_unprobe(sa);
1886 
1887 	rte_free(dev->data->mac_addrs);
1888 	dev->data->mac_addrs = NULL;
1889 
1890 	sfc_kvargs_cleanup(sa);
1891 
1892 	sfc_adapter_unlock(sa);
1893 	sfc_adapter_lock_fini(sa);
1894 
1895 	sfc_log_init(sa, "done");
1896 
1897 	/* Required for logging, so cleanup last */
1898 	sa->eth_dev = NULL;
1899 	return 0;
1900 }
1901 
1902 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
1903 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
1904 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
1905 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
1906 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
1907 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
1908 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
1909 	{ .vendor_id = 0 /* sentinel */ }
1910 };
1911 
1912 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1913 	struct rte_pci_device *pci_dev)
1914 {
1915 	return rte_eth_dev_pci_generic_probe(pci_dev,
1916 		sizeof(struct sfc_adapter), sfc_eth_dev_init);
1917 }
1918 
1919 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
1920 {
1921 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
1922 }
1923 
1924 static struct rte_pci_driver sfc_efx_pmd = {
1925 	.id_table = pci_id_sfc_efx_map,
1926 	.drv_flags =
1927 		RTE_PCI_DRV_INTR_LSC |
1928 		RTE_PCI_DRV_NEED_MAPPING,
1929 	.probe = sfc_eth_dev_pci_probe,
1930 	.remove = sfc_eth_dev_pci_remove,
1931 };
1932 
1933 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
1934 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
1935 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
1936 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
1937 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
1938 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
1939 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
1940 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long> "
1941 	SFC_KVARG_MCDI_LOGGING "=" SFC_KVARG_VALUES_BOOL " "
1942 	SFC_KVARG_DEBUG_INIT "=" SFC_KVARG_VALUES_BOOL);
1943