xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision 3c335b7f5e4edeb3ad19d996e4651f7bb5882a7e)
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_bus_pci.h>
37 #include <rte_errno.h>
38 
39 #include "efx.h"
40 
41 #include "sfc.h"
42 #include "sfc_debug.h"
43 #include "sfc_log.h"
44 #include "sfc_kvargs.h"
45 #include "sfc_ev.h"
46 #include "sfc_rx.h"
47 #include "sfc_tx.h"
48 #include "sfc_flow.h"
49 #include "sfc_dp.h"
50 #include "sfc_dp_rx.h"
51 
52 static struct sfc_dp_list sfc_dp_head =
53 	TAILQ_HEAD_INITIALIZER(sfc_dp_head);
54 
55 static int
56 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
57 {
58 	struct sfc_adapter *sa = dev->data->dev_private;
59 	efx_nic_fw_info_t enfi;
60 	int ret;
61 	int rc;
62 
63 	/*
64 	 * Return value of the callback is likely supposed to be
65 	 * equal to or greater than 0, nevertheless, if an error
66 	 * occurs, it will be desirable to pass it to the caller
67 	 */
68 	if ((fw_version == NULL) || (fw_size == 0))
69 		return -EINVAL;
70 
71 	rc = efx_nic_get_fw_version(sa->nic, &enfi);
72 	if (rc != 0)
73 		return -rc;
74 
75 	ret = snprintf(fw_version, fw_size,
76 		       "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
77 		       enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
78 		       enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
79 	if (ret < 0)
80 		return ret;
81 
82 	if (enfi.enfi_dpcpu_fw_ids_valid) {
83 		size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
84 		int ret_extra;
85 
86 		ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
87 				     fw_size - dpcpu_fw_ids_offset,
88 				     " rx%" PRIx16 " tx%" PRIx16,
89 				     enfi.enfi_rx_dpcpu_fw_id,
90 				     enfi.enfi_tx_dpcpu_fw_id);
91 		if (ret_extra < 0)
92 			return ret_extra;
93 
94 		ret += ret_extra;
95 	}
96 
97 	if (fw_size < (size_t)(++ret))
98 		return ret;
99 	else
100 		return 0;
101 }
102 
103 static void
104 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
105 {
106 	struct sfc_adapter *sa = dev->data->dev_private;
107 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
108 
109 	sfc_log_init(sa, "entry");
110 
111 	dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
112 	dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
113 
114 	/* Autonegotiation may be disabled */
115 	dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
116 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
117 		dev_info->speed_capa |= ETH_LINK_SPEED_1G;
118 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
119 		dev_info->speed_capa |= ETH_LINK_SPEED_10G;
120 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
121 		dev_info->speed_capa |= ETH_LINK_SPEED_40G;
122 
123 	dev_info->max_rx_queues = sa->rxq_max;
124 	dev_info->max_tx_queues = sa->txq_max;
125 
126 	/* By default packets are dropped if no descriptors are available */
127 	dev_info->default_rxconf.rx_drop_en = 1;
128 
129 	dev_info->rx_offload_capa =
130 		DEV_RX_OFFLOAD_IPV4_CKSUM |
131 		DEV_RX_OFFLOAD_UDP_CKSUM |
132 		DEV_RX_OFFLOAD_TCP_CKSUM;
133 
134 	if ((encp->enc_tunnel_encapsulations_supported != 0) &&
135 	    (sa->dp_rx->features & SFC_DP_RX_FEAT_TUNNELS))
136 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM;
137 
138 	dev_info->tx_offload_capa =
139 		DEV_TX_OFFLOAD_IPV4_CKSUM |
140 		DEV_TX_OFFLOAD_UDP_CKSUM |
141 		DEV_TX_OFFLOAD_TCP_CKSUM;
142 
143 	if (encp->enc_tunnel_encapsulations_supported != 0)
144 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM;
145 
146 	dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
147 	if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
148 	    !encp->enc_hw_tx_insert_vlan_enabled)
149 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
150 	else
151 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_VLAN_INSERT;
152 
153 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
154 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
155 
156 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_POOL)
157 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTMEMP;
158 
159 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_REFCNT)
160 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOREFCOUNT;
161 
162 #if EFSYS_OPT_RX_SCALE
163 	if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
164 		dev_info->reta_size = EFX_RSS_TBL_SIZE;
165 		dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
166 		dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
167 	}
168 #endif
169 
170 	if (sa->tso)
171 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
172 
173 	/* Initialize to hardware limits */
174 	dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
175 	dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
176 	/* The RXQ hardware requires that the descriptor count is a power
177 	 * of 2, but rx_desc_lim cannot properly describe that constraint.
178 	 */
179 	dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
180 
181 	dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
182 	dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
183 	/*
184 	 * The TXQ hardware requires that the descriptor count is a power
185 	 * of 2, but tx_desc_lim cannot properly describe that constraint
186 	 */
187 	dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
188 
189 	if (sa->dp_rx->get_dev_info != NULL)
190 		sa->dp_rx->get_dev_info(dev_info);
191 }
192 
193 static const uint32_t *
194 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
195 {
196 	struct sfc_adapter *sa = dev->data->dev_private;
197 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
198 	uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
199 
200 	return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
201 }
202 
203 static int
204 sfc_dev_configure(struct rte_eth_dev *dev)
205 {
206 	struct rte_eth_dev_data *dev_data = dev->data;
207 	struct sfc_adapter *sa = dev_data->dev_private;
208 	int rc;
209 
210 	sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
211 		     dev_data->nb_rx_queues, dev_data->nb_tx_queues);
212 
213 	sfc_adapter_lock(sa);
214 	switch (sa->state) {
215 	case SFC_ADAPTER_CONFIGURED:
216 		/* FALLTHROUGH */
217 	case SFC_ADAPTER_INITIALIZED:
218 		rc = sfc_configure(sa);
219 		break;
220 	default:
221 		sfc_err(sa, "unexpected adapter state %u to configure",
222 			sa->state);
223 		rc = EINVAL;
224 		break;
225 	}
226 	sfc_adapter_unlock(sa);
227 
228 	sfc_log_init(sa, "done %d", rc);
229 	SFC_ASSERT(rc >= 0);
230 	return -rc;
231 }
232 
233 static int
234 sfc_dev_start(struct rte_eth_dev *dev)
235 {
236 	struct sfc_adapter *sa = dev->data->dev_private;
237 	int rc;
238 
239 	sfc_log_init(sa, "entry");
240 
241 	sfc_adapter_lock(sa);
242 	rc = sfc_start(sa);
243 	sfc_adapter_unlock(sa);
244 
245 	sfc_log_init(sa, "done %d", rc);
246 	SFC_ASSERT(rc >= 0);
247 	return -rc;
248 }
249 
250 static int
251 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
252 {
253 	struct sfc_adapter *sa = dev->data->dev_private;
254 	struct rte_eth_link *dev_link = &dev->data->dev_link;
255 	struct rte_eth_link old_link;
256 	struct rte_eth_link current_link;
257 
258 	sfc_log_init(sa, "entry");
259 
260 retry:
261 	EFX_STATIC_ASSERT(sizeof(*dev_link) == sizeof(rte_atomic64_t));
262 	*(int64_t *)&old_link = rte_atomic64_read((rte_atomic64_t *)dev_link);
263 
264 	if (sa->state != SFC_ADAPTER_STARTED) {
265 		sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
266 		if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
267 					 *(uint64_t *)&old_link,
268 					 *(uint64_t *)&current_link))
269 			goto retry;
270 	} else if (wait_to_complete) {
271 		efx_link_mode_t link_mode;
272 
273 		if (efx_port_poll(sa->nic, &link_mode) != 0)
274 			link_mode = EFX_LINK_UNKNOWN;
275 		sfc_port_link_mode_to_info(link_mode, &current_link);
276 
277 		if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
278 					 *(uint64_t *)&old_link,
279 					 *(uint64_t *)&current_link))
280 			goto retry;
281 	} else {
282 		sfc_ev_mgmt_qpoll(sa);
283 		*(int64_t *)&current_link =
284 			rte_atomic64_read((rte_atomic64_t *)dev_link);
285 	}
286 
287 	if (old_link.link_status != current_link.link_status)
288 		sfc_info(sa, "Link status is %s",
289 			 current_link.link_status ? "UP" : "DOWN");
290 
291 	return old_link.link_status == current_link.link_status ? 0 : -1;
292 }
293 
294 static void
295 sfc_dev_stop(struct rte_eth_dev *dev)
296 {
297 	struct sfc_adapter *sa = dev->data->dev_private;
298 
299 	sfc_log_init(sa, "entry");
300 
301 	sfc_adapter_lock(sa);
302 	sfc_stop(sa);
303 	sfc_adapter_unlock(sa);
304 
305 	sfc_log_init(sa, "done");
306 }
307 
308 static int
309 sfc_dev_set_link_up(struct rte_eth_dev *dev)
310 {
311 	struct sfc_adapter *sa = dev->data->dev_private;
312 	int rc;
313 
314 	sfc_log_init(sa, "entry");
315 
316 	sfc_adapter_lock(sa);
317 	rc = sfc_start(sa);
318 	sfc_adapter_unlock(sa);
319 
320 	SFC_ASSERT(rc >= 0);
321 	return -rc;
322 }
323 
324 static int
325 sfc_dev_set_link_down(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 	sfc_stop(sa);
333 	sfc_adapter_unlock(sa);
334 
335 	return 0;
336 }
337 
338 static void
339 sfc_dev_close(struct rte_eth_dev *dev)
340 {
341 	struct sfc_adapter *sa = dev->data->dev_private;
342 
343 	sfc_log_init(sa, "entry");
344 
345 	sfc_adapter_lock(sa);
346 	switch (sa->state) {
347 	case SFC_ADAPTER_STARTED:
348 		sfc_stop(sa);
349 		SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
350 		/* FALLTHROUGH */
351 	case SFC_ADAPTER_CONFIGURED:
352 		sfc_close(sa);
353 		SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
354 		/* FALLTHROUGH */
355 	case SFC_ADAPTER_INITIALIZED:
356 		break;
357 	default:
358 		sfc_err(sa, "unexpected adapter state %u on close", sa->state);
359 		break;
360 	}
361 	sfc_adapter_unlock(sa);
362 
363 	sfc_log_init(sa, "done");
364 }
365 
366 static void
367 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
368 		   boolean_t enabled)
369 {
370 	struct sfc_port *port;
371 	boolean_t *toggle;
372 	struct sfc_adapter *sa = dev->data->dev_private;
373 	boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
374 	const char *desc = (allmulti) ? "all-multi" : "promiscuous";
375 
376 	sfc_adapter_lock(sa);
377 
378 	port = &sa->port;
379 	toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
380 
381 	if (*toggle != enabled) {
382 		*toggle = enabled;
383 
384 		if (port->isolated) {
385 			sfc_warn(sa, "isolated mode is active on the port");
386 			sfc_warn(sa, "the change is to be applied on the next "
387 				     "start provided that isolated mode is "
388 				     "disabled prior the next start");
389 		} else if ((sa->state == SFC_ADAPTER_STARTED) &&
390 			   (sfc_set_rx_mode(sa) != 0)) {
391 			*toggle = !(enabled);
392 			sfc_warn(sa, "Failed to %s %s mode",
393 				 ((enabled) ? "enable" : "disable"), desc);
394 		}
395 	}
396 
397 	sfc_adapter_unlock(sa);
398 }
399 
400 static void
401 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
402 {
403 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
404 }
405 
406 static void
407 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
408 {
409 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
410 }
411 
412 static void
413 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
414 {
415 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
416 }
417 
418 static void
419 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
420 {
421 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
422 }
423 
424 static int
425 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
426 		   uint16_t nb_rx_desc, unsigned int socket_id,
427 		   const struct rte_eth_rxconf *rx_conf,
428 		   struct rte_mempool *mb_pool)
429 {
430 	struct sfc_adapter *sa = dev->data->dev_private;
431 	int rc;
432 
433 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
434 		     rx_queue_id, nb_rx_desc, socket_id);
435 
436 	sfc_adapter_lock(sa);
437 
438 	rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
439 			  rx_conf, mb_pool);
440 	if (rc != 0)
441 		goto fail_rx_qinit;
442 
443 	dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
444 
445 	sfc_adapter_unlock(sa);
446 
447 	return 0;
448 
449 fail_rx_qinit:
450 	sfc_adapter_unlock(sa);
451 	SFC_ASSERT(rc > 0);
452 	return -rc;
453 }
454 
455 static void
456 sfc_rx_queue_release(void *queue)
457 {
458 	struct sfc_dp_rxq *dp_rxq = queue;
459 	struct sfc_rxq *rxq;
460 	struct sfc_adapter *sa;
461 	unsigned int sw_index;
462 
463 	if (dp_rxq == NULL)
464 		return;
465 
466 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
467 	sa = rxq->evq->sa;
468 	sfc_adapter_lock(sa);
469 
470 	sw_index = sfc_rxq_sw_index(rxq);
471 
472 	sfc_log_init(sa, "RxQ=%u", sw_index);
473 
474 	sa->eth_dev->data->rx_queues[sw_index] = NULL;
475 
476 	sfc_rx_qfini(sa, sw_index);
477 
478 	sfc_adapter_unlock(sa);
479 }
480 
481 static int
482 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
483 		   uint16_t nb_tx_desc, unsigned int socket_id,
484 		   const struct rte_eth_txconf *tx_conf)
485 {
486 	struct sfc_adapter *sa = dev->data->dev_private;
487 	int rc;
488 
489 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
490 		     tx_queue_id, nb_tx_desc, socket_id);
491 
492 	sfc_adapter_lock(sa);
493 
494 	rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
495 	if (rc != 0)
496 		goto fail_tx_qinit;
497 
498 	dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
499 
500 	sfc_adapter_unlock(sa);
501 	return 0;
502 
503 fail_tx_qinit:
504 	sfc_adapter_unlock(sa);
505 	SFC_ASSERT(rc > 0);
506 	return -rc;
507 }
508 
509 static void
510 sfc_tx_queue_release(void *queue)
511 {
512 	struct sfc_dp_txq *dp_txq = queue;
513 	struct sfc_txq *txq;
514 	unsigned int sw_index;
515 	struct sfc_adapter *sa;
516 
517 	if (dp_txq == NULL)
518 		return;
519 
520 	txq = sfc_txq_by_dp_txq(dp_txq);
521 	sw_index = sfc_txq_sw_index(txq);
522 
523 	SFC_ASSERT(txq->evq != NULL);
524 	sa = txq->evq->sa;
525 
526 	sfc_log_init(sa, "TxQ = %u", sw_index);
527 
528 	sfc_adapter_lock(sa);
529 
530 	SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
531 	sa->eth_dev->data->tx_queues[sw_index] = NULL;
532 
533 	sfc_tx_qfini(sa, sw_index);
534 
535 	sfc_adapter_unlock(sa);
536 }
537 
538 static int
539 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
540 {
541 	struct sfc_adapter *sa = dev->data->dev_private;
542 	struct sfc_port *port = &sa->port;
543 	uint64_t *mac_stats;
544 	int ret;
545 
546 	rte_spinlock_lock(&port->mac_stats_lock);
547 
548 	ret = sfc_port_update_mac_stats(sa);
549 	if (ret != 0)
550 		goto unlock;
551 
552 	mac_stats = port->mac_stats_buf;
553 
554 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
555 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
556 		stats->ipackets =
557 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
558 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
559 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
560 		stats->opackets =
561 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
562 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
563 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
564 		stats->ibytes =
565 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
566 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
567 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
568 		stats->obytes =
569 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
570 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
571 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
572 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
573 		stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
574 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
575 	} else {
576 		stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
577 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
578 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
579 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
580 		/*
581 		 * Take into account stats which are whenever supported
582 		 * on EF10. If some stat is not supported by current
583 		 * firmware variant or HW revision, it is guaranteed
584 		 * to be zero in mac_stats.
585 		 */
586 		stats->imissed =
587 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
588 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
589 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
590 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
591 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
592 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
593 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
594 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
595 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
596 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
597 		stats->ierrors =
598 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
599 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
600 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
601 		/* no oerrors counters supported on EF10 */
602 	}
603 
604 unlock:
605 	rte_spinlock_unlock(&port->mac_stats_lock);
606 	SFC_ASSERT(ret >= 0);
607 	return -ret;
608 }
609 
610 static void
611 sfc_stats_reset(struct rte_eth_dev *dev)
612 {
613 	struct sfc_adapter *sa = dev->data->dev_private;
614 	struct sfc_port *port = &sa->port;
615 	int rc;
616 
617 	if (sa->state != SFC_ADAPTER_STARTED) {
618 		/*
619 		 * The operation cannot be done if port is not started; it
620 		 * will be scheduled to be done during the next port start
621 		 */
622 		port->mac_stats_reset_pending = B_TRUE;
623 		return;
624 	}
625 
626 	rc = sfc_port_reset_mac_stats(sa);
627 	if (rc != 0)
628 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
629 }
630 
631 static int
632 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
633 	       unsigned int xstats_count)
634 {
635 	struct sfc_adapter *sa = dev->data->dev_private;
636 	struct sfc_port *port = &sa->port;
637 	uint64_t *mac_stats;
638 	int rc;
639 	unsigned int i;
640 	int nstats = 0;
641 
642 	rte_spinlock_lock(&port->mac_stats_lock);
643 
644 	rc = sfc_port_update_mac_stats(sa);
645 	if (rc != 0) {
646 		SFC_ASSERT(rc > 0);
647 		nstats = -rc;
648 		goto unlock;
649 	}
650 
651 	mac_stats = port->mac_stats_buf;
652 
653 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
654 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
655 			if (xstats != NULL && nstats < (int)xstats_count) {
656 				xstats[nstats].id = nstats;
657 				xstats[nstats].value = mac_stats[i];
658 			}
659 			nstats++;
660 		}
661 	}
662 
663 unlock:
664 	rte_spinlock_unlock(&port->mac_stats_lock);
665 
666 	return nstats;
667 }
668 
669 static int
670 sfc_xstats_get_names(struct rte_eth_dev *dev,
671 		     struct rte_eth_xstat_name *xstats_names,
672 		     unsigned int xstats_count)
673 {
674 	struct sfc_adapter *sa = dev->data->dev_private;
675 	struct sfc_port *port = &sa->port;
676 	unsigned int i;
677 	unsigned int nstats = 0;
678 
679 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
680 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
681 			if (xstats_names != NULL && nstats < xstats_count)
682 				strncpy(xstats_names[nstats].name,
683 					efx_mac_stat_name(sa->nic, i),
684 					sizeof(xstats_names[0].name));
685 			nstats++;
686 		}
687 	}
688 
689 	return nstats;
690 }
691 
692 static int
693 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
694 		     uint64_t *values, unsigned int n)
695 {
696 	struct sfc_adapter *sa = dev->data->dev_private;
697 	struct sfc_port *port = &sa->port;
698 	uint64_t *mac_stats;
699 	unsigned int nb_supported = 0;
700 	unsigned int nb_written = 0;
701 	unsigned int i;
702 	int ret;
703 	int rc;
704 
705 	if (unlikely(values == NULL) ||
706 	    unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
707 		return port->mac_stats_nb_supported;
708 
709 	rte_spinlock_lock(&port->mac_stats_lock);
710 
711 	rc = sfc_port_update_mac_stats(sa);
712 	if (rc != 0) {
713 		SFC_ASSERT(rc > 0);
714 		ret = -rc;
715 		goto unlock;
716 	}
717 
718 	mac_stats = port->mac_stats_buf;
719 
720 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
721 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
722 			continue;
723 
724 		if ((ids == NULL) || (ids[nb_written] == nb_supported))
725 			values[nb_written++] = mac_stats[i];
726 
727 		++nb_supported;
728 	}
729 
730 	ret = nb_written;
731 
732 unlock:
733 	rte_spinlock_unlock(&port->mac_stats_lock);
734 
735 	return ret;
736 }
737 
738 static int
739 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
740 			   struct rte_eth_xstat_name *xstats_names,
741 			   const uint64_t *ids, unsigned int size)
742 {
743 	struct sfc_adapter *sa = dev->data->dev_private;
744 	struct sfc_port *port = &sa->port;
745 	unsigned int nb_supported = 0;
746 	unsigned int nb_written = 0;
747 	unsigned int i;
748 
749 	if (unlikely(xstats_names == NULL) ||
750 	    unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
751 		return port->mac_stats_nb_supported;
752 
753 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
754 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
755 			continue;
756 
757 		if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
758 			char *name = xstats_names[nb_written++].name;
759 
760 			strncpy(name, efx_mac_stat_name(sa->nic, i),
761 				sizeof(xstats_names[0].name));
762 			name[sizeof(xstats_names[0].name) - 1] = '\0';
763 		}
764 
765 		++nb_supported;
766 	}
767 
768 	return nb_written;
769 }
770 
771 static int
772 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
773 {
774 	struct sfc_adapter *sa = dev->data->dev_private;
775 	unsigned int wanted_fc, link_fc;
776 
777 	memset(fc_conf, 0, sizeof(*fc_conf));
778 
779 	sfc_adapter_lock(sa);
780 
781 	if (sa->state == SFC_ADAPTER_STARTED)
782 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
783 	else
784 		link_fc = sa->port.flow_ctrl;
785 
786 	switch (link_fc) {
787 	case 0:
788 		fc_conf->mode = RTE_FC_NONE;
789 		break;
790 	case EFX_FCNTL_RESPOND:
791 		fc_conf->mode = RTE_FC_RX_PAUSE;
792 		break;
793 	case EFX_FCNTL_GENERATE:
794 		fc_conf->mode = RTE_FC_TX_PAUSE;
795 		break;
796 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
797 		fc_conf->mode = RTE_FC_FULL;
798 		break;
799 	default:
800 		sfc_err(sa, "%s: unexpected flow control value %#x",
801 			__func__, link_fc);
802 	}
803 
804 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
805 
806 	sfc_adapter_unlock(sa);
807 
808 	return 0;
809 }
810 
811 static int
812 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
813 {
814 	struct sfc_adapter *sa = dev->data->dev_private;
815 	struct sfc_port *port = &sa->port;
816 	unsigned int fcntl;
817 	int rc;
818 
819 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
820 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
821 	    fc_conf->mac_ctrl_frame_fwd != 0) {
822 		sfc_err(sa, "unsupported flow control settings specified");
823 		rc = EINVAL;
824 		goto fail_inval;
825 	}
826 
827 	switch (fc_conf->mode) {
828 	case RTE_FC_NONE:
829 		fcntl = 0;
830 		break;
831 	case RTE_FC_RX_PAUSE:
832 		fcntl = EFX_FCNTL_RESPOND;
833 		break;
834 	case RTE_FC_TX_PAUSE:
835 		fcntl = EFX_FCNTL_GENERATE;
836 		break;
837 	case RTE_FC_FULL:
838 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
839 		break;
840 	default:
841 		rc = EINVAL;
842 		goto fail_inval;
843 	}
844 
845 	sfc_adapter_lock(sa);
846 
847 	if (sa->state == SFC_ADAPTER_STARTED) {
848 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
849 		if (rc != 0)
850 			goto fail_mac_fcntl_set;
851 	}
852 
853 	port->flow_ctrl = fcntl;
854 	port->flow_ctrl_autoneg = fc_conf->autoneg;
855 
856 	sfc_adapter_unlock(sa);
857 
858 	return 0;
859 
860 fail_mac_fcntl_set:
861 	sfc_adapter_unlock(sa);
862 fail_inval:
863 	SFC_ASSERT(rc > 0);
864 	return -rc;
865 }
866 
867 static int
868 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
869 {
870 	struct sfc_adapter *sa = dev->data->dev_private;
871 	size_t pdu = EFX_MAC_PDU(mtu);
872 	size_t old_pdu;
873 	int rc;
874 
875 	sfc_log_init(sa, "mtu=%u", mtu);
876 
877 	rc = EINVAL;
878 	if (pdu < EFX_MAC_PDU_MIN) {
879 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
880 			(unsigned int)mtu, (unsigned int)pdu,
881 			EFX_MAC_PDU_MIN);
882 		goto fail_inval;
883 	}
884 	if (pdu > EFX_MAC_PDU_MAX) {
885 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
886 			(unsigned int)mtu, (unsigned int)pdu,
887 			EFX_MAC_PDU_MAX);
888 		goto fail_inval;
889 	}
890 
891 	sfc_adapter_lock(sa);
892 
893 	if (pdu != sa->port.pdu) {
894 		if (sa->state == SFC_ADAPTER_STARTED) {
895 			sfc_stop(sa);
896 
897 			old_pdu = sa->port.pdu;
898 			sa->port.pdu = pdu;
899 			rc = sfc_start(sa);
900 			if (rc != 0)
901 				goto fail_start;
902 		} else {
903 			sa->port.pdu = pdu;
904 		}
905 	}
906 
907 	/*
908 	 * The driver does not use it, but other PMDs update jumbo_frame
909 	 * flag and max_rx_pkt_len when MTU is set.
910 	 */
911 	dev->data->dev_conf.rxmode.jumbo_frame = (mtu > ETHER_MAX_LEN);
912 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
913 
914 	sfc_adapter_unlock(sa);
915 
916 	sfc_log_init(sa, "done");
917 	return 0;
918 
919 fail_start:
920 	sa->port.pdu = old_pdu;
921 	if (sfc_start(sa) != 0)
922 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
923 			"PDU max size - port is stopped",
924 			(unsigned int)pdu, (unsigned int)old_pdu);
925 	sfc_adapter_unlock(sa);
926 
927 fail_inval:
928 	sfc_log_init(sa, "failed %d", rc);
929 	SFC_ASSERT(rc > 0);
930 	return -rc;
931 }
932 static void
933 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
934 {
935 	struct sfc_adapter *sa = dev->data->dev_private;
936 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
937 	struct sfc_port *port = &sa->port;
938 	int rc;
939 
940 	sfc_adapter_lock(sa);
941 
942 	/*
943 	 * Copy the address to the device private data so that
944 	 * it could be recalled in the case of adapter restart.
945 	 */
946 	ether_addr_copy(mac_addr, &port->default_mac_addr);
947 
948 	if (port->isolated) {
949 		sfc_err(sa, "isolated mode is active on the port");
950 		sfc_err(sa, "will not set MAC address");
951 		goto unlock;
952 	}
953 
954 	if (sa->state != SFC_ADAPTER_STARTED) {
955 		sfc_info(sa, "the port is not started");
956 		sfc_info(sa, "the new MAC address will be set on port start");
957 
958 		goto unlock;
959 	}
960 
961 	if (encp->enc_allow_set_mac_with_installed_filters) {
962 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
963 		if (rc != 0) {
964 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
965 			goto unlock;
966 		}
967 
968 		/*
969 		 * Changing the MAC address by means of MCDI request
970 		 * has no effect on received traffic, therefore
971 		 * we also need to update unicast filters
972 		 */
973 		rc = sfc_set_rx_mode(sa);
974 		if (rc != 0)
975 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
976 	} else {
977 		sfc_warn(sa, "cannot set MAC address with filters installed");
978 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
979 		sfc_warn(sa, "(some traffic may be dropped)");
980 
981 		/*
982 		 * Since setting MAC address with filters installed is not
983 		 * allowed on the adapter, the new MAC address will be set
984 		 * by means of adapter restart. sfc_start() shall retrieve
985 		 * the new address from the device private data and set it.
986 		 */
987 		sfc_stop(sa);
988 		rc = sfc_start(sa);
989 		if (rc != 0)
990 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
991 	}
992 
993 unlock:
994 	/*
995 	 * In the case of failure sa->port->default_mac_addr does not
996 	 * need rollback since no error code is returned, and the upper
997 	 * API will anyway update the external MAC address storage.
998 	 * To be consistent with that new value it is better to keep
999 	 * the device private value the same.
1000 	 */
1001 	sfc_adapter_unlock(sa);
1002 }
1003 
1004 
1005 static int
1006 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
1007 		     uint32_t nb_mc_addr)
1008 {
1009 	struct sfc_adapter *sa = dev->data->dev_private;
1010 	struct sfc_port *port = &sa->port;
1011 	uint8_t *mc_addrs = port->mcast_addrs;
1012 	int rc;
1013 	unsigned int i;
1014 
1015 	if (port->isolated) {
1016 		sfc_err(sa, "isolated mode is active on the port");
1017 		sfc_err(sa, "will not set multicast address list");
1018 		return -ENOTSUP;
1019 	}
1020 
1021 	if (mc_addrs == NULL)
1022 		return -ENOBUFS;
1023 
1024 	if (nb_mc_addr > port->max_mcast_addrs) {
1025 		sfc_err(sa, "too many multicast addresses: %u > %u",
1026 			 nb_mc_addr, port->max_mcast_addrs);
1027 		return -EINVAL;
1028 	}
1029 
1030 	for (i = 0; i < nb_mc_addr; ++i) {
1031 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1032 				 EFX_MAC_ADDR_LEN);
1033 		mc_addrs += EFX_MAC_ADDR_LEN;
1034 	}
1035 
1036 	port->nb_mcast_addrs = nb_mc_addr;
1037 
1038 	if (sa->state != SFC_ADAPTER_STARTED)
1039 		return 0;
1040 
1041 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1042 					port->nb_mcast_addrs);
1043 	if (rc != 0)
1044 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1045 
1046 	SFC_ASSERT(rc > 0);
1047 	return -rc;
1048 }
1049 
1050 /*
1051  * The function is used by the secondary process as well. It must not
1052  * use any process-local pointers from the adapter data.
1053  */
1054 static void
1055 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1056 		      struct rte_eth_rxq_info *qinfo)
1057 {
1058 	struct sfc_adapter *sa = dev->data->dev_private;
1059 	struct sfc_rxq_info *rxq_info;
1060 	struct sfc_rxq *rxq;
1061 
1062 	sfc_adapter_lock(sa);
1063 
1064 	SFC_ASSERT(rx_queue_id < sa->rxq_count);
1065 
1066 	rxq_info = &sa->rxq_info[rx_queue_id];
1067 	rxq = rxq_info->rxq;
1068 	SFC_ASSERT(rxq != NULL);
1069 
1070 	qinfo->mp = rxq->refill_mb_pool;
1071 	qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1072 	qinfo->conf.rx_drop_en = 1;
1073 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1074 	qinfo->scattered_rx =
1075 		((rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) != 0);
1076 	qinfo->nb_desc = rxq_info->entries;
1077 
1078 	sfc_adapter_unlock(sa);
1079 }
1080 
1081 /*
1082  * The function is used by the secondary process as well. It must not
1083  * use any process-local pointers from the adapter data.
1084  */
1085 static void
1086 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1087 		      struct rte_eth_txq_info *qinfo)
1088 {
1089 	struct sfc_adapter *sa = dev->data->dev_private;
1090 	struct sfc_txq_info *txq_info;
1091 
1092 	sfc_adapter_lock(sa);
1093 
1094 	SFC_ASSERT(tx_queue_id < sa->txq_count);
1095 
1096 	txq_info = &sa->txq_info[tx_queue_id];
1097 	SFC_ASSERT(txq_info->txq != NULL);
1098 
1099 	memset(qinfo, 0, sizeof(*qinfo));
1100 
1101 	qinfo->conf.txq_flags = txq_info->txq->flags;
1102 	qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1103 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1104 	qinfo->nb_desc = txq_info->entries;
1105 
1106 	sfc_adapter_unlock(sa);
1107 }
1108 
1109 static uint32_t
1110 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1111 {
1112 	struct sfc_adapter *sa = dev->data->dev_private;
1113 
1114 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1115 
1116 	return sfc_rx_qdesc_npending(sa, rx_queue_id);
1117 }
1118 
1119 static int
1120 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1121 {
1122 	struct sfc_dp_rxq *dp_rxq = queue;
1123 
1124 	return sfc_rx_qdesc_done(dp_rxq, offset);
1125 }
1126 
1127 static int
1128 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1129 {
1130 	struct sfc_dp_rxq *dp_rxq = queue;
1131 	struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1132 
1133 	return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1134 }
1135 
1136 static int
1137 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1138 {
1139 	struct sfc_dp_txq *dp_txq = queue;
1140 	struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1141 
1142 	return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1143 }
1144 
1145 static int
1146 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1147 {
1148 	struct sfc_adapter *sa = dev->data->dev_private;
1149 	int rc;
1150 
1151 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1152 
1153 	sfc_adapter_lock(sa);
1154 
1155 	rc = EINVAL;
1156 	if (sa->state != SFC_ADAPTER_STARTED)
1157 		goto fail_not_started;
1158 
1159 	rc = sfc_rx_qstart(sa, rx_queue_id);
1160 	if (rc != 0)
1161 		goto fail_rx_qstart;
1162 
1163 	sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1164 
1165 	sfc_adapter_unlock(sa);
1166 
1167 	return 0;
1168 
1169 fail_rx_qstart:
1170 fail_not_started:
1171 	sfc_adapter_unlock(sa);
1172 	SFC_ASSERT(rc > 0);
1173 	return -rc;
1174 }
1175 
1176 static int
1177 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1178 {
1179 	struct sfc_adapter *sa = dev->data->dev_private;
1180 
1181 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1182 
1183 	sfc_adapter_lock(sa);
1184 	sfc_rx_qstop(sa, rx_queue_id);
1185 
1186 	sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1187 
1188 	sfc_adapter_unlock(sa);
1189 
1190 	return 0;
1191 }
1192 
1193 static int
1194 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1195 {
1196 	struct sfc_adapter *sa = dev->data->dev_private;
1197 	int rc;
1198 
1199 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1200 
1201 	sfc_adapter_lock(sa);
1202 
1203 	rc = EINVAL;
1204 	if (sa->state != SFC_ADAPTER_STARTED)
1205 		goto fail_not_started;
1206 
1207 	rc = sfc_tx_qstart(sa, tx_queue_id);
1208 	if (rc != 0)
1209 		goto fail_tx_qstart;
1210 
1211 	sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1212 
1213 	sfc_adapter_unlock(sa);
1214 	return 0;
1215 
1216 fail_tx_qstart:
1217 
1218 fail_not_started:
1219 	sfc_adapter_unlock(sa);
1220 	SFC_ASSERT(rc > 0);
1221 	return -rc;
1222 }
1223 
1224 static int
1225 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1226 {
1227 	struct sfc_adapter *sa = dev->data->dev_private;
1228 
1229 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1230 
1231 	sfc_adapter_lock(sa);
1232 
1233 	sfc_tx_qstop(sa, tx_queue_id);
1234 
1235 	sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1236 
1237 	sfc_adapter_unlock(sa);
1238 	return 0;
1239 }
1240 
1241 static efx_tunnel_protocol_t
1242 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1243 {
1244 	switch (rte_type) {
1245 	case RTE_TUNNEL_TYPE_VXLAN:
1246 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1247 	case RTE_TUNNEL_TYPE_GENEVE:
1248 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1249 	default:
1250 		return EFX_TUNNEL_NPROTOS;
1251 	}
1252 }
1253 
1254 enum sfc_udp_tunnel_op_e {
1255 	SFC_UDP_TUNNEL_ADD_PORT,
1256 	SFC_UDP_TUNNEL_DEL_PORT,
1257 };
1258 
1259 static int
1260 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1261 		      struct rte_eth_udp_tunnel *tunnel_udp,
1262 		      enum sfc_udp_tunnel_op_e op)
1263 {
1264 	struct sfc_adapter *sa = dev->data->dev_private;
1265 	efx_tunnel_protocol_t tunnel_proto;
1266 	int rc;
1267 
1268 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1269 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1270 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1271 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1272 
1273 	tunnel_proto =
1274 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1275 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1276 		rc = ENOTSUP;
1277 		goto fail_bad_proto;
1278 	}
1279 
1280 	sfc_adapter_lock(sa);
1281 
1282 	switch (op) {
1283 	case SFC_UDP_TUNNEL_ADD_PORT:
1284 		rc = efx_tunnel_config_udp_add(sa->nic,
1285 					       tunnel_udp->udp_port,
1286 					       tunnel_proto);
1287 		break;
1288 	case SFC_UDP_TUNNEL_DEL_PORT:
1289 		rc = efx_tunnel_config_udp_remove(sa->nic,
1290 						  tunnel_udp->udp_port,
1291 						  tunnel_proto);
1292 		break;
1293 	default:
1294 		rc = EINVAL;
1295 		goto fail_bad_op;
1296 	}
1297 
1298 	if (rc != 0)
1299 		goto fail_op;
1300 
1301 	if (sa->state == SFC_ADAPTER_STARTED) {
1302 		rc = efx_tunnel_reconfigure(sa->nic);
1303 		if (rc == EAGAIN) {
1304 			/*
1305 			 * Configuration is accepted by FW and MC reboot
1306 			 * is initiated to apply the changes. MC reboot
1307 			 * will be handled in a usual way (MC reboot
1308 			 * event on management event queue and adapter
1309 			 * restart).
1310 			 */
1311 			rc = 0;
1312 		} else if (rc != 0) {
1313 			goto fail_reconfigure;
1314 		}
1315 	}
1316 
1317 	sfc_adapter_unlock(sa);
1318 	return 0;
1319 
1320 fail_reconfigure:
1321 	/* Remove/restore entry since the change makes the trouble */
1322 	switch (op) {
1323 	case SFC_UDP_TUNNEL_ADD_PORT:
1324 		(void)efx_tunnel_config_udp_remove(sa->nic,
1325 						   tunnel_udp->udp_port,
1326 						   tunnel_proto);
1327 		break;
1328 	case SFC_UDP_TUNNEL_DEL_PORT:
1329 		(void)efx_tunnel_config_udp_add(sa->nic,
1330 						tunnel_udp->udp_port,
1331 						tunnel_proto);
1332 		break;
1333 	}
1334 
1335 fail_op:
1336 fail_bad_op:
1337 	sfc_adapter_unlock(sa);
1338 
1339 fail_bad_proto:
1340 	SFC_ASSERT(rc > 0);
1341 	return -rc;
1342 }
1343 
1344 static int
1345 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1346 			    struct rte_eth_udp_tunnel *tunnel_udp)
1347 {
1348 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1349 }
1350 
1351 static int
1352 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1353 			    struct rte_eth_udp_tunnel *tunnel_udp)
1354 {
1355 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1356 }
1357 
1358 #if EFSYS_OPT_RX_SCALE
1359 static int
1360 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1361 			  struct rte_eth_rss_conf *rss_conf)
1362 {
1363 	struct sfc_adapter *sa = dev->data->dev_private;
1364 	struct sfc_port *port = &sa->port;
1365 
1366 	if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1367 		return -ENOTSUP;
1368 
1369 	if (sa->rss_channels == 0)
1370 		return -EINVAL;
1371 
1372 	sfc_adapter_lock(sa);
1373 
1374 	/*
1375 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1376 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1377 	 * flags which corresponds to the active EFX configuration stored
1378 	 * locally in 'sfc_adapter' and kept up-to-date
1379 	 */
1380 	rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1381 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1382 	if (rss_conf->rss_key != NULL)
1383 		rte_memcpy(rss_conf->rss_key, sa->rss_key, EFX_RSS_KEY_SIZE);
1384 
1385 	sfc_adapter_unlock(sa);
1386 
1387 	return 0;
1388 }
1389 
1390 static int
1391 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1392 			struct rte_eth_rss_conf *rss_conf)
1393 {
1394 	struct sfc_adapter *sa = dev->data->dev_private;
1395 	struct sfc_port *port = &sa->port;
1396 	unsigned int efx_hash_types;
1397 	int rc = 0;
1398 
1399 	if (port->isolated)
1400 		return -ENOTSUP;
1401 
1402 	if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1403 		sfc_err(sa, "RSS is not available");
1404 		return -ENOTSUP;
1405 	}
1406 
1407 	if (sa->rss_channels == 0) {
1408 		sfc_err(sa, "RSS is not configured");
1409 		return -EINVAL;
1410 	}
1411 
1412 	if ((rss_conf->rss_key != NULL) &&
1413 	    (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1414 		sfc_err(sa, "RSS key size is wrong (should be %lu)",
1415 			sizeof(sa->rss_key));
1416 		return -EINVAL;
1417 	}
1418 
1419 	if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1420 		sfc_err(sa, "unsupported hash functions requested");
1421 		return -EINVAL;
1422 	}
1423 
1424 	sfc_adapter_lock(sa);
1425 
1426 	efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1427 
1428 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1429 				   EFX_RX_HASHALG_TOEPLITZ,
1430 				   efx_hash_types, B_TRUE);
1431 	if (rc != 0)
1432 		goto fail_scale_mode_set;
1433 
1434 	if (rss_conf->rss_key != NULL) {
1435 		if (sa->state == SFC_ADAPTER_STARTED) {
1436 			rc = efx_rx_scale_key_set(sa->nic,
1437 						  EFX_RSS_CONTEXT_DEFAULT,
1438 						  rss_conf->rss_key,
1439 						  sizeof(sa->rss_key));
1440 			if (rc != 0)
1441 				goto fail_scale_key_set;
1442 		}
1443 
1444 		rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1445 	}
1446 
1447 	sa->rss_hash_types = efx_hash_types;
1448 
1449 	sfc_adapter_unlock(sa);
1450 
1451 	return 0;
1452 
1453 fail_scale_key_set:
1454 	if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1455 				  EFX_RX_HASHALG_TOEPLITZ,
1456 				  sa->rss_hash_types, B_TRUE) != 0)
1457 		sfc_err(sa, "failed to restore RSS mode");
1458 
1459 fail_scale_mode_set:
1460 	sfc_adapter_unlock(sa);
1461 	return -rc;
1462 }
1463 
1464 static int
1465 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1466 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1467 		       uint16_t reta_size)
1468 {
1469 	struct sfc_adapter *sa = dev->data->dev_private;
1470 	struct sfc_port *port = &sa->port;
1471 	int entry;
1472 
1473 	if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1474 		return -ENOTSUP;
1475 
1476 	if (sa->rss_channels == 0)
1477 		return -EINVAL;
1478 
1479 	if (reta_size != EFX_RSS_TBL_SIZE)
1480 		return -EINVAL;
1481 
1482 	sfc_adapter_lock(sa);
1483 
1484 	for (entry = 0; entry < reta_size; entry++) {
1485 		int grp = entry / RTE_RETA_GROUP_SIZE;
1486 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1487 
1488 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1489 			reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1490 	}
1491 
1492 	sfc_adapter_unlock(sa);
1493 
1494 	return 0;
1495 }
1496 
1497 static int
1498 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1499 			struct rte_eth_rss_reta_entry64 *reta_conf,
1500 			uint16_t reta_size)
1501 {
1502 	struct sfc_adapter *sa = dev->data->dev_private;
1503 	struct sfc_port *port = &sa->port;
1504 	unsigned int *rss_tbl_new;
1505 	uint16_t entry;
1506 	int rc = 0;
1507 
1508 
1509 	if (port->isolated)
1510 		return -ENOTSUP;
1511 
1512 	if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1513 		sfc_err(sa, "RSS is not available");
1514 		return -ENOTSUP;
1515 	}
1516 
1517 	if (sa->rss_channels == 0) {
1518 		sfc_err(sa, "RSS is not configured");
1519 		return -EINVAL;
1520 	}
1521 
1522 	if (reta_size != EFX_RSS_TBL_SIZE) {
1523 		sfc_err(sa, "RETA size is wrong (should be %u)",
1524 			EFX_RSS_TBL_SIZE);
1525 		return -EINVAL;
1526 	}
1527 
1528 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1529 	if (rss_tbl_new == NULL)
1530 		return -ENOMEM;
1531 
1532 	sfc_adapter_lock(sa);
1533 
1534 	rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1535 
1536 	for (entry = 0; entry < reta_size; entry++) {
1537 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1538 		struct rte_eth_rss_reta_entry64 *grp;
1539 
1540 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1541 
1542 		if (grp->mask & (1ull << grp_idx)) {
1543 			if (grp->reta[grp_idx] >= sa->rss_channels) {
1544 				rc = EINVAL;
1545 				goto bad_reta_entry;
1546 			}
1547 			rss_tbl_new[entry] = grp->reta[grp_idx];
1548 		}
1549 	}
1550 
1551 	if (sa->state == SFC_ADAPTER_STARTED) {
1552 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1553 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1554 		if (rc != 0)
1555 			goto fail_scale_tbl_set;
1556 	}
1557 
1558 	rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1559 
1560 fail_scale_tbl_set:
1561 bad_reta_entry:
1562 	sfc_adapter_unlock(sa);
1563 
1564 	rte_free(rss_tbl_new);
1565 
1566 	SFC_ASSERT(rc >= 0);
1567 	return -rc;
1568 }
1569 #endif
1570 
1571 static int
1572 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1573 		    enum rte_filter_op filter_op,
1574 		    void *arg)
1575 {
1576 	struct sfc_adapter *sa = dev->data->dev_private;
1577 	int rc = ENOTSUP;
1578 
1579 	sfc_log_init(sa, "entry");
1580 
1581 	switch (filter_type) {
1582 	case RTE_ETH_FILTER_NONE:
1583 		sfc_err(sa, "Global filters configuration not supported");
1584 		break;
1585 	case RTE_ETH_FILTER_MACVLAN:
1586 		sfc_err(sa, "MACVLAN filters not supported");
1587 		break;
1588 	case RTE_ETH_FILTER_ETHERTYPE:
1589 		sfc_err(sa, "EtherType filters not supported");
1590 		break;
1591 	case RTE_ETH_FILTER_FLEXIBLE:
1592 		sfc_err(sa, "Flexible filters not supported");
1593 		break;
1594 	case RTE_ETH_FILTER_SYN:
1595 		sfc_err(sa, "SYN filters not supported");
1596 		break;
1597 	case RTE_ETH_FILTER_NTUPLE:
1598 		sfc_err(sa, "NTUPLE filters not supported");
1599 		break;
1600 	case RTE_ETH_FILTER_TUNNEL:
1601 		sfc_err(sa, "Tunnel filters not supported");
1602 		break;
1603 	case RTE_ETH_FILTER_FDIR:
1604 		sfc_err(sa, "Flow Director filters not supported");
1605 		break;
1606 	case RTE_ETH_FILTER_HASH:
1607 		sfc_err(sa, "Hash filters not supported");
1608 		break;
1609 	case RTE_ETH_FILTER_GENERIC:
1610 		if (filter_op != RTE_ETH_FILTER_GET) {
1611 			rc = EINVAL;
1612 		} else {
1613 			*(const void **)arg = &sfc_flow_ops;
1614 			rc = 0;
1615 		}
1616 		break;
1617 	default:
1618 		sfc_err(sa, "Unknown filter type %u", filter_type);
1619 		break;
1620 	}
1621 
1622 	sfc_log_init(sa, "exit: %d", -rc);
1623 	SFC_ASSERT(rc >= 0);
1624 	return -rc;
1625 }
1626 
1627 static const struct eth_dev_ops sfc_eth_dev_ops = {
1628 	.dev_configure			= sfc_dev_configure,
1629 	.dev_start			= sfc_dev_start,
1630 	.dev_stop			= sfc_dev_stop,
1631 	.dev_set_link_up		= sfc_dev_set_link_up,
1632 	.dev_set_link_down		= sfc_dev_set_link_down,
1633 	.dev_close			= sfc_dev_close,
1634 	.promiscuous_enable		= sfc_dev_promisc_enable,
1635 	.promiscuous_disable		= sfc_dev_promisc_disable,
1636 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1637 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1638 	.link_update			= sfc_dev_link_update,
1639 	.stats_get			= sfc_stats_get,
1640 	.stats_reset			= sfc_stats_reset,
1641 	.xstats_get			= sfc_xstats_get,
1642 	.xstats_reset			= sfc_stats_reset,
1643 	.xstats_get_names		= sfc_xstats_get_names,
1644 	.dev_infos_get			= sfc_dev_infos_get,
1645 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1646 	.mtu_set			= sfc_dev_set_mtu,
1647 	.rx_queue_start			= sfc_rx_queue_start,
1648 	.rx_queue_stop			= sfc_rx_queue_stop,
1649 	.tx_queue_start			= sfc_tx_queue_start,
1650 	.tx_queue_stop			= sfc_tx_queue_stop,
1651 	.rx_queue_setup			= sfc_rx_queue_setup,
1652 	.rx_queue_release		= sfc_rx_queue_release,
1653 	.rx_queue_count			= sfc_rx_queue_count,
1654 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1655 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1656 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1657 	.tx_queue_setup			= sfc_tx_queue_setup,
1658 	.tx_queue_release		= sfc_tx_queue_release,
1659 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1660 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1661 	.mac_addr_set			= sfc_mac_addr_set,
1662 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1663 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
1664 #if EFSYS_OPT_RX_SCALE
1665 	.reta_update			= sfc_dev_rss_reta_update,
1666 	.reta_query			= sfc_dev_rss_reta_query,
1667 	.rss_hash_update		= sfc_dev_rss_hash_update,
1668 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1669 #endif
1670 	.filter_ctrl			= sfc_dev_filter_ctrl,
1671 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1672 	.rxq_info_get			= sfc_rx_queue_info_get,
1673 	.txq_info_get			= sfc_tx_queue_info_get,
1674 	.fw_version_get			= sfc_fw_version_get,
1675 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1676 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1677 };
1678 
1679 /**
1680  * Duplicate a string in potentially shared memory required for
1681  * multi-process support.
1682  *
1683  * strdup() allocates from process-local heap/memory.
1684  */
1685 static char *
1686 sfc_strdup(const char *str)
1687 {
1688 	size_t size;
1689 	char *copy;
1690 
1691 	if (str == NULL)
1692 		return NULL;
1693 
1694 	size = strlen(str) + 1;
1695 	copy = rte_malloc(__func__, size, 0);
1696 	if (copy != NULL)
1697 		rte_memcpy(copy, str, size);
1698 
1699 	return copy;
1700 }
1701 
1702 static int
1703 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1704 {
1705 	struct sfc_adapter *sa = dev->data->dev_private;
1706 	unsigned int avail_caps = 0;
1707 	const char *rx_name = NULL;
1708 	const char *tx_name = NULL;
1709 	int rc;
1710 
1711 	switch (sa->family) {
1712 	case EFX_FAMILY_HUNTINGTON:
1713 	case EFX_FAMILY_MEDFORD:
1714 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1715 		break;
1716 	default:
1717 		break;
1718 	}
1719 
1720 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1721 				sfc_kvarg_string_handler, &rx_name);
1722 	if (rc != 0)
1723 		goto fail_kvarg_rx_datapath;
1724 
1725 	if (rx_name != NULL) {
1726 		sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1727 		if (sa->dp_rx == NULL) {
1728 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1729 			rc = ENOENT;
1730 			goto fail_dp_rx;
1731 		}
1732 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1733 			sfc_err(sa,
1734 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1735 				rx_name);
1736 			rc = EINVAL;
1737 			goto fail_dp_rx_caps;
1738 		}
1739 	} else {
1740 		sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1741 		if (sa->dp_rx == NULL) {
1742 			sfc_err(sa, "Rx datapath by caps %#x not found",
1743 				avail_caps);
1744 			rc = ENOENT;
1745 			goto fail_dp_rx;
1746 		}
1747 	}
1748 
1749 	sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1750 	if (sa->dp_rx_name == NULL) {
1751 		rc = ENOMEM;
1752 		goto fail_dp_rx_name;
1753 	}
1754 
1755 	sfc_info(sa, "use %s Rx datapath", sa->dp_rx_name);
1756 
1757 	dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1758 
1759 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1760 				sfc_kvarg_string_handler, &tx_name);
1761 	if (rc != 0)
1762 		goto fail_kvarg_tx_datapath;
1763 
1764 	if (tx_name != NULL) {
1765 		sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1766 		if (sa->dp_tx == NULL) {
1767 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1768 			rc = ENOENT;
1769 			goto fail_dp_tx;
1770 		}
1771 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1772 			sfc_err(sa,
1773 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1774 				tx_name);
1775 			rc = EINVAL;
1776 			goto fail_dp_tx_caps;
1777 		}
1778 	} else {
1779 		sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1780 		if (sa->dp_tx == NULL) {
1781 			sfc_err(sa, "Tx datapath by caps %#x not found",
1782 				avail_caps);
1783 			rc = ENOENT;
1784 			goto fail_dp_tx;
1785 		}
1786 	}
1787 
1788 	sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1789 	if (sa->dp_tx_name == NULL) {
1790 		rc = ENOMEM;
1791 		goto fail_dp_tx_name;
1792 	}
1793 
1794 	sfc_info(sa, "use %s Tx datapath", sa->dp_tx_name);
1795 
1796 	dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1797 
1798 	dev->dev_ops = &sfc_eth_dev_ops;
1799 
1800 	return 0;
1801 
1802 fail_dp_tx_name:
1803 fail_dp_tx_caps:
1804 	sa->dp_tx = NULL;
1805 
1806 fail_dp_tx:
1807 fail_kvarg_tx_datapath:
1808 	rte_free(sa->dp_rx_name);
1809 	sa->dp_rx_name = NULL;
1810 
1811 fail_dp_rx_name:
1812 fail_dp_rx_caps:
1813 	sa->dp_rx = NULL;
1814 
1815 fail_dp_rx:
1816 fail_kvarg_rx_datapath:
1817 	return rc;
1818 }
1819 
1820 static void
1821 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1822 {
1823 	struct sfc_adapter *sa = dev->data->dev_private;
1824 
1825 	dev->dev_ops = NULL;
1826 	dev->rx_pkt_burst = NULL;
1827 	dev->tx_pkt_burst = NULL;
1828 
1829 	rte_free(sa->dp_tx_name);
1830 	sa->dp_tx_name = NULL;
1831 	sa->dp_tx = NULL;
1832 
1833 	rte_free(sa->dp_rx_name);
1834 	sa->dp_rx_name = NULL;
1835 	sa->dp_rx = NULL;
1836 }
1837 
1838 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1839 	.rxq_info_get			= sfc_rx_queue_info_get,
1840 	.txq_info_get			= sfc_tx_queue_info_get,
1841 };
1842 
1843 static int
1844 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1845 {
1846 	/*
1847 	 * Device private data has really many process-local pointers.
1848 	 * Below code should be extremely careful to use data located
1849 	 * in shared memory only.
1850 	 */
1851 	struct sfc_adapter *sa = dev->data->dev_private;
1852 	const struct sfc_dp_rx *dp_rx;
1853 	const struct sfc_dp_tx *dp_tx;
1854 	int rc;
1855 
1856 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1857 	if (dp_rx == NULL) {
1858 		sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1859 		rc = ENOENT;
1860 		goto fail_dp_rx;
1861 	}
1862 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1863 		sfc_err(sa, "%s Rx datapath does not support multi-process",
1864 			sa->dp_tx_name);
1865 		rc = EINVAL;
1866 		goto fail_dp_rx_multi_process;
1867 	}
1868 
1869 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1870 	if (dp_tx == NULL) {
1871 		sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1872 		rc = ENOENT;
1873 		goto fail_dp_tx;
1874 	}
1875 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1876 		sfc_err(sa, "%s Tx datapath does not support multi-process",
1877 			sa->dp_tx_name);
1878 		rc = EINVAL;
1879 		goto fail_dp_tx_multi_process;
1880 	}
1881 
1882 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1883 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1884 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
1885 
1886 	return 0;
1887 
1888 fail_dp_tx_multi_process:
1889 fail_dp_tx:
1890 fail_dp_rx_multi_process:
1891 fail_dp_rx:
1892 	return rc;
1893 }
1894 
1895 static void
1896 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1897 {
1898 	dev->dev_ops = NULL;
1899 	dev->tx_pkt_burst = NULL;
1900 	dev->rx_pkt_burst = NULL;
1901 }
1902 
1903 static void
1904 sfc_register_dp(void)
1905 {
1906 	/* Register once */
1907 	if (TAILQ_EMPTY(&sfc_dp_head)) {
1908 		/* Prefer EF10 datapath */
1909 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1910 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1911 
1912 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1913 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1914 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1915 	}
1916 }
1917 
1918 static int
1919 sfc_eth_dev_init(struct rte_eth_dev *dev)
1920 {
1921 	struct sfc_adapter *sa = dev->data->dev_private;
1922 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1923 	int rc;
1924 	const efx_nic_cfg_t *encp;
1925 	const struct ether_addr *from;
1926 
1927 	sfc_register_dp();
1928 
1929 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1930 		return -sfc_eth_dev_secondary_set_ops(dev);
1931 
1932 	/* Required for logging */
1933 	sa->pci_addr = pci_dev->addr;
1934 	sa->port_id = dev->data->port_id;
1935 
1936 	sa->eth_dev = dev;
1937 
1938 	/* Copy PCI device info to the dev->data */
1939 	rte_eth_copy_pci_info(dev, pci_dev);
1940 
1941 	rc = sfc_kvargs_parse(sa);
1942 	if (rc != 0)
1943 		goto fail_kvargs_parse;
1944 
1945 	rc = sfc_kvargs_process(sa, SFC_KVARG_DEBUG_INIT,
1946 				sfc_kvarg_bool_handler, &sa->debug_init);
1947 	if (rc != 0)
1948 		goto fail_kvarg_debug_init;
1949 
1950 	sfc_log_init(sa, "entry");
1951 
1952 	dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1953 	if (dev->data->mac_addrs == NULL) {
1954 		rc = ENOMEM;
1955 		goto fail_mac_addrs;
1956 	}
1957 
1958 	sfc_adapter_lock_init(sa);
1959 	sfc_adapter_lock(sa);
1960 
1961 	sfc_log_init(sa, "probing");
1962 	rc = sfc_probe(sa);
1963 	if (rc != 0)
1964 		goto fail_probe;
1965 
1966 	sfc_log_init(sa, "set device ops");
1967 	rc = sfc_eth_dev_set_ops(dev);
1968 	if (rc != 0)
1969 		goto fail_set_ops;
1970 
1971 	sfc_log_init(sa, "attaching");
1972 	rc = sfc_attach(sa);
1973 	if (rc != 0)
1974 		goto fail_attach;
1975 
1976 	encp = efx_nic_cfg_get(sa->nic);
1977 
1978 	/*
1979 	 * The arguments are really reverse order in comparison to
1980 	 * Linux kernel. Copy from NIC config to Ethernet device data.
1981 	 */
1982 	from = (const struct ether_addr *)(encp->enc_mac_addr);
1983 	ether_addr_copy(from, &dev->data->mac_addrs[0]);
1984 
1985 	sfc_adapter_unlock(sa);
1986 
1987 	sfc_log_init(sa, "done");
1988 	return 0;
1989 
1990 fail_attach:
1991 	sfc_eth_dev_clear_ops(dev);
1992 
1993 fail_set_ops:
1994 	sfc_unprobe(sa);
1995 
1996 fail_probe:
1997 	sfc_adapter_unlock(sa);
1998 	sfc_adapter_lock_fini(sa);
1999 	rte_free(dev->data->mac_addrs);
2000 	dev->data->mac_addrs = NULL;
2001 
2002 fail_mac_addrs:
2003 fail_kvarg_debug_init:
2004 	sfc_kvargs_cleanup(sa);
2005 
2006 fail_kvargs_parse:
2007 	sfc_log_init(sa, "failed %d", rc);
2008 	SFC_ASSERT(rc > 0);
2009 	return -rc;
2010 }
2011 
2012 static int
2013 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2014 {
2015 	struct sfc_adapter *sa;
2016 
2017 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2018 		sfc_eth_dev_secondary_clear_ops(dev);
2019 		return 0;
2020 	}
2021 
2022 	sa = dev->data->dev_private;
2023 	sfc_log_init(sa, "entry");
2024 
2025 	sfc_adapter_lock(sa);
2026 
2027 	sfc_eth_dev_clear_ops(dev);
2028 
2029 	sfc_detach(sa);
2030 	sfc_unprobe(sa);
2031 
2032 	rte_free(dev->data->mac_addrs);
2033 	dev->data->mac_addrs = NULL;
2034 
2035 	sfc_kvargs_cleanup(sa);
2036 
2037 	sfc_adapter_unlock(sa);
2038 	sfc_adapter_lock_fini(sa);
2039 
2040 	sfc_log_init(sa, "done");
2041 
2042 	/* Required for logging, so cleanup last */
2043 	sa->eth_dev = NULL;
2044 	return 0;
2045 }
2046 
2047 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2048 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2049 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2050 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2051 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2052 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2053 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2054 	{ .vendor_id = 0 /* sentinel */ }
2055 };
2056 
2057 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2058 	struct rte_pci_device *pci_dev)
2059 {
2060 	return rte_eth_dev_pci_generic_probe(pci_dev,
2061 		sizeof(struct sfc_adapter), sfc_eth_dev_init);
2062 }
2063 
2064 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2065 {
2066 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2067 }
2068 
2069 static struct rte_pci_driver sfc_efx_pmd = {
2070 	.id_table = pci_id_sfc_efx_map,
2071 	.drv_flags =
2072 		RTE_PCI_DRV_INTR_LSC |
2073 		RTE_PCI_DRV_NEED_MAPPING,
2074 	.probe = sfc_eth_dev_pci_probe,
2075 	.remove = sfc_eth_dev_pci_remove,
2076 };
2077 
2078 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2079 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2080 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2081 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2082 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2083 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2084 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2085 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long> "
2086 	SFC_KVARG_MCDI_LOGGING "=" SFC_KVARG_VALUES_BOOL " "
2087 	SFC_KVARG_DEBUG_INIT "=" SFC_KVARG_VALUES_BOOL);
2088